/*
- * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2013 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <sys/vm.h>
-#include <sys/time.h>
+#include <sys/time.h>
#include <kern/clock.h>
#include <libkern/OSAtomic.h>
#include <nfs/nfsproto.h>
#include <nfs/nfs.h>
#include <nfs/nfsnode.h>
+#include <nfs/nfs_gss.h>
#include <nfs/nfsmount.h>
#include <nfs/nfs_lock.h>
#include <nfs/xdr_subs.h>
#include <net/if.h>
#include <netinet/in.h>
#include <netinet/in_var.h>
+
#include <vm/vm_kern.h>
+#include <vm/vm_pageout.h>
#include <kern/task.h>
#include <kern/sched_prim.h>
-#include <sys/kdebug.h>
-
-#define FSDBG(A, B, C, D, E) \
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_NONE, \
- (int)(B), (int)(C), (int)(D), (int)(E), 0)
-#define FSDBG_TOP(A, B, C, D, E) \
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_START, \
- (int)(B), (int)(C), (int)(D), (int)(E), 0)
-#define FSDBG_BOT(A, B, C, D, E) \
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_END, \
- (int)(B), (int)(C), (int)(D), (int)(E), 0)
-
-static int nfsspec_read(struct vnop_read_args *);
-static int nfsspec_write(struct vnop_write_args *);
-static int nfsfifo_read(struct vnop_read_args *);
-static int nfsfifo_write(struct vnop_write_args *);
-static int nfsspec_close(struct vnop_close_args *);
-static int nfsfifo_close(struct vnop_close_args *);
-static int nfs_ioctl(struct vnop_ioctl_args *);
-static int nfs_select(struct vnop_select_args *);
-static int nfs_setattrrpc(vnode_t,struct vnode_attr *,kauth_cred_t,proc_t);
-static int nfs_lookup(struct vnop_lookup_args *);
-static int nfs_create(struct vnop_create_args *);
-static int nfs_mknod(struct vnop_mknod_args *);
-static int nfs_open(struct vnop_open_args *);
-static int nfs_close(struct vnop_close_args *);
-static int nfs_access(struct vnop_access_args *);
-static int nfs_vnop_getattr(struct vnop_getattr_args *);
-static int nfs_setattr(struct vnop_setattr_args *);
-static int nfs_read(struct vnop_read_args *);
-static int nfs_mmap(struct vnop_mmap_args *);
-static int nfs_fsync(struct vnop_fsync_args *);
-static int nfs_remove(struct vnop_remove_args *);
-static int nfs_link(struct vnop_link_args *);
-static int nfs_rename(struct vnop_rename_args *);
-static int nfs_mkdir(struct vnop_mkdir_args *);
-static int nfs_rmdir(struct vnop_rmdir_args *);
-static int nfs_symlink(struct vnop_symlink_args *);
-static int nfs_readdir(struct vnop_readdir_args *);
-static int nfs_lookitup(vnode_t,char *,int,kauth_cred_t,proc_t,struct nfsnode **);
-static int nfs_sillyrename(vnode_t,vnode_t,struct componentname *,kauth_cred_t,proc_t);
-static int nfs_readlink(struct vnop_readlink_args *);
-static int nfs_pathconf(struct vnop_pathconf_args *);
-static int nfs_advlock(struct vnop_advlock_args *);
-static int nfs_pagein(struct vnop_pagein_args *);
-static int nfs_pageout(struct vnop_pageout_args *);
-static int nfs_blktooff(struct vnop_blktooff_args *);
-static int nfs_offtoblk(struct vnop_offtoblk_args *);
-static int nfs_blockmap(struct vnop_blockmap_args *);
+#define NFS_VNOP_DBG(...) NFS_DBG(NFS_FAC_VNOP, 7, ## __VA_ARGS__)
/*
- * Global vfs data structures for nfs
+ * NFS vnode ops
*/
+int nfs_vnop_lookup(struct vnop_lookup_args *);
+int nfsspec_vnop_read(struct vnop_read_args *);
+int nfsspec_vnop_write(struct vnop_write_args *);
+int nfsspec_vnop_close(struct vnop_close_args *);
+#if FIFO
+int nfsfifo_vnop_read(struct vnop_read_args *);
+int nfsfifo_vnop_write(struct vnop_write_args *);
+int nfsfifo_vnop_close(struct vnop_close_args *);
+#endif
+int nfs_vnop_ioctl(struct vnop_ioctl_args *);
+int nfs_vnop_select(struct vnop_select_args *);
+int nfs_vnop_setattr(struct vnop_setattr_args *);
+int nfs_vnop_fsync(struct vnop_fsync_args *);
+int nfs_vnop_rename(struct vnop_rename_args *);
+int nfs_vnop_readdir(struct vnop_readdir_args *);
+int nfs_vnop_readlink(struct vnop_readlink_args *);
+int nfs_vnop_pathconf(struct vnop_pathconf_args *);
+int nfs_vnop_pagein(struct vnop_pagein_args *);
+int nfs_vnop_pageout(struct vnop_pageout_args *);
+int nfs_vnop_blktooff(struct vnop_blktooff_args *);
+int nfs_vnop_offtoblk(struct vnop_offtoblk_args *);
+int nfs_vnop_blockmap(struct vnop_blockmap_args *);
+int nfs_vnop_monitor(struct vnop_monitor_args *);
+
+int nfs3_vnop_create(struct vnop_create_args *);
+int nfs3_vnop_mknod(struct vnop_mknod_args *);
+int nfs3_vnop_getattr(struct vnop_getattr_args *);
+int nfs3_vnop_link(struct vnop_link_args *);
+int nfs3_vnop_mkdir(struct vnop_mkdir_args *);
+int nfs3_vnop_rmdir(struct vnop_rmdir_args *);
+int nfs3_vnop_symlink(struct vnop_symlink_args *);
+
vnop_t **nfsv2_vnodeop_p;
static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
{ &vnop_default_desc, (vnop_t *)vn_default_error },
- { &vnop_lookup_desc, (vnop_t *)nfs_lookup }, /* lookup */
- { &vnop_create_desc, (vnop_t *)nfs_create }, /* create */
- { &vnop_mknod_desc, (vnop_t *)nfs_mknod }, /* mknod */
- { &vnop_open_desc, (vnop_t *)nfs_open }, /* open */
- { &vnop_close_desc, (vnop_t *)nfs_close }, /* close */
- { &vnop_access_desc, (vnop_t *)nfs_access }, /* access */
- { &vnop_getattr_desc, (vnop_t *)nfs_vnop_getattr }, /* getattr */
- { &vnop_setattr_desc, (vnop_t *)nfs_setattr }, /* setattr */
- { &vnop_read_desc, (vnop_t *)nfs_read }, /* read */
- { &vnop_write_desc, (vnop_t *)nfs_write }, /* write */
- { &vnop_ioctl_desc, (vnop_t *)nfs_ioctl }, /* ioctl */
- { &vnop_select_desc, (vnop_t *)nfs_select }, /* select */
- { &vnop_revoke_desc, (vnop_t *)nfs_revoke }, /* revoke */
- { &vnop_mmap_desc, (vnop_t *)nfs_mmap }, /* mmap */
- { &vnop_fsync_desc, (vnop_t *)nfs_fsync }, /* fsync */
- { &vnop_remove_desc, (vnop_t *)nfs_remove }, /* remove */
- { &vnop_link_desc, (vnop_t *)nfs_link }, /* link */
- { &vnop_rename_desc, (vnop_t *)nfs_rename }, /* rename */
- { &vnop_mkdir_desc, (vnop_t *)nfs_mkdir }, /* mkdir */
- { &vnop_rmdir_desc, (vnop_t *)nfs_rmdir }, /* rmdir */
- { &vnop_symlink_desc, (vnop_t *)nfs_symlink }, /* symlink */
- { &vnop_readdir_desc, (vnop_t *)nfs_readdir }, /* readdir */
- { &vnop_readlink_desc, (vnop_t *)nfs_readlink }, /* readlink */
- { &vnop_inactive_desc, (vnop_t *)nfs_inactive }, /* inactive */
- { &vnop_reclaim_desc, (vnop_t *)nfs_reclaim }, /* reclaim */
+ { &vnop_lookup_desc, (vnop_t *)nfs_vnop_lookup }, /* lookup */
+ { &vnop_create_desc, (vnop_t *)nfs3_vnop_create }, /* create */
+ { &vnop_mknod_desc, (vnop_t *)nfs3_vnop_mknod }, /* mknod */
+ { &vnop_open_desc, (vnop_t *)nfs_vnop_open }, /* open */
+ { &vnop_close_desc, (vnop_t *)nfs_vnop_close }, /* close */
+ { &vnop_access_desc, (vnop_t *)nfs_vnop_access }, /* access */
+ { &vnop_getattr_desc, (vnop_t *)nfs3_vnop_getattr }, /* getattr */
+ { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
+ { &vnop_read_desc, (vnop_t *)nfs_vnop_read }, /* read */
+ { &vnop_write_desc, (vnop_t *)nfs_vnop_write }, /* write */
+ { &vnop_ioctl_desc, (vnop_t *)nfs_vnop_ioctl }, /* ioctl */
+ { &vnop_select_desc, (vnop_t *)nfs_vnop_select }, /* select */
+ { &vnop_revoke_desc, (vnop_t *)nfs_vnop_revoke }, /* revoke */
+ { &vnop_mmap_desc, (vnop_t *)nfs_vnop_mmap }, /* mmap */
+ { &vnop_mnomap_desc, (vnop_t *)nfs_vnop_mnomap }, /* mnomap */
+ { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
+ { &vnop_remove_desc, (vnop_t *)nfs_vnop_remove }, /* remove */
+ { &vnop_link_desc, (vnop_t *)nfs3_vnop_link }, /* link */
+ { &vnop_rename_desc, (vnop_t *)nfs_vnop_rename }, /* rename */
+ { &vnop_mkdir_desc, (vnop_t *)nfs3_vnop_mkdir }, /* mkdir */
+ { &vnop_rmdir_desc, (vnop_t *)nfs3_vnop_rmdir }, /* rmdir */
+ { &vnop_symlink_desc, (vnop_t *)nfs3_vnop_symlink }, /* symlink */
+ { &vnop_readdir_desc, (vnop_t *)nfs_vnop_readdir }, /* readdir */
+ { &vnop_readlink_desc, (vnop_t *)nfs_vnop_readlink }, /* readlink */
+ { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
+ { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
{ &vnop_strategy_desc, (vnop_t *)err_strategy }, /* strategy */
- { &vnop_pathconf_desc, (vnop_t *)nfs_pathconf }, /* pathconf */
- { &vnop_advlock_desc, (vnop_t *)nfs_advlock }, /* advlock */
+ { &vnop_pathconf_desc, (vnop_t *)nfs_vnop_pathconf }, /* pathconf */
+ { &vnop_advlock_desc, (vnop_t *)nfs_vnop_advlock }, /* advlock */
{ &vnop_bwrite_desc, (vnop_t *)err_bwrite }, /* bwrite */
- { &vnop_pagein_desc, (vnop_t *)nfs_pagein }, /* Pagein */
- { &vnop_pageout_desc, (vnop_t *)nfs_pageout }, /* Pageout */
+ { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
+ { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
{ &vnop_copyfile_desc, (vnop_t *)err_copyfile }, /* Copyfile */
- { &vnop_blktooff_desc, (vnop_t *)nfs_blktooff }, /* blktooff */
- { &vnop_offtoblk_desc, (vnop_t *)nfs_offtoblk }, /* offtoblk */
- { &vnop_blockmap_desc, (vnop_t *)nfs_blockmap }, /* blockmap */
+ { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
+ { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
+ { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
+ { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
{ NULL, NULL }
};
struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
{ &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
-#ifdef __FreeBSD__
-VNODEOP_SET(nfsv2_vnodeop_opv_desc);
+
+vnop_t **nfsv4_vnodeop_p;
+static struct vnodeopv_entry_desc nfsv4_vnodeop_entries[] = {
+ { &vnop_default_desc, (vnop_t *)vn_default_error },
+ { &vnop_lookup_desc, (vnop_t *)nfs_vnop_lookup }, /* lookup */
+ { &vnop_create_desc, (vnop_t *)nfs4_vnop_create }, /* create */
+ { &vnop_mknod_desc, (vnop_t *)nfs4_vnop_mknod }, /* mknod */
+ { &vnop_open_desc, (vnop_t *)nfs_vnop_open }, /* open */
+ { &vnop_close_desc, (vnop_t *)nfs_vnop_close }, /* close */
+ { &vnop_access_desc, (vnop_t *)nfs_vnop_access }, /* access */
+ { &vnop_getattr_desc, (vnop_t *)nfs4_vnop_getattr }, /* getattr */
+ { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
+ { &vnop_read_desc, (vnop_t *)nfs_vnop_read }, /* read */
+ { &vnop_write_desc, (vnop_t *)nfs_vnop_write }, /* write */
+ { &vnop_ioctl_desc, (vnop_t *)nfs_vnop_ioctl }, /* ioctl */
+ { &vnop_select_desc, (vnop_t *)nfs_vnop_select }, /* select */
+ { &vnop_revoke_desc, (vnop_t *)nfs_vnop_revoke }, /* revoke */
+ { &vnop_mmap_desc, (vnop_t *)nfs_vnop_mmap }, /* mmap */
+ { &vnop_mnomap_desc, (vnop_t *)nfs_vnop_mnomap }, /* mnomap */
+ { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
+ { &vnop_remove_desc, (vnop_t *)nfs_vnop_remove }, /* remove */
+ { &vnop_link_desc, (vnop_t *)nfs4_vnop_link }, /* link */
+ { &vnop_rename_desc, (vnop_t *)nfs_vnop_rename }, /* rename */
+ { &vnop_mkdir_desc, (vnop_t *)nfs4_vnop_mkdir }, /* mkdir */
+ { &vnop_rmdir_desc, (vnop_t *)nfs4_vnop_rmdir }, /* rmdir */
+ { &vnop_symlink_desc, (vnop_t *)nfs4_vnop_symlink }, /* symlink */
+ { &vnop_readdir_desc, (vnop_t *)nfs_vnop_readdir }, /* readdir */
+ { &vnop_readlink_desc, (vnop_t *)nfs_vnop_readlink }, /* readlink */
+ { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
+ { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
+ { &vnop_strategy_desc, (vnop_t *)err_strategy }, /* strategy */
+ { &vnop_pathconf_desc, (vnop_t *)nfs_vnop_pathconf }, /* pathconf */
+ { &vnop_advlock_desc, (vnop_t *)nfs_vnop_advlock }, /* advlock */
+ { &vnop_bwrite_desc, (vnop_t *)err_bwrite }, /* bwrite */
+ { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
+ { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
+ { &vnop_copyfile_desc, (vnop_t *)err_copyfile }, /* Copyfile */
+ { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
+ { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
+ { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
+ { &vnop_getxattr_desc, (vnop_t *)nfs4_vnop_getxattr }, /* getxattr */
+ { &vnop_setxattr_desc, (vnop_t *)nfs4_vnop_setxattr }, /* setxattr */
+ { &vnop_removexattr_desc, (vnop_t *)nfs4_vnop_removexattr },/* removexattr */
+ { &vnop_listxattr_desc, (vnop_t *)nfs4_vnop_listxattr },/* listxattr */
+#if NAMEDSTREAMS
+ { &vnop_getnamedstream_desc, (vnop_t *)nfs4_vnop_getnamedstream }, /* getnamedstream */
+ { &vnop_makenamedstream_desc, (vnop_t *)nfs4_vnop_makenamedstream }, /* makenamedstream */
+ { &vnop_removenamedstream_desc, (vnop_t *)nfs4_vnop_removenamedstream },/* removenamedstream */
#endif
+ { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
+ { NULL, NULL }
+};
+struct vnodeopv_desc nfsv4_vnodeop_opv_desc =
+ { &nfsv4_vnodeop_p, nfsv4_vnodeop_entries };
/*
* Special device vnode ops
{ &vnop_create_desc, (vnop_t *)spec_create }, /* create */
{ &vnop_mknod_desc, (vnop_t *)spec_mknod }, /* mknod */
{ &vnop_open_desc, (vnop_t *)spec_open }, /* open */
- { &vnop_close_desc, (vnop_t *)nfsspec_close }, /* close */
- { &vnop_getattr_desc, (vnop_t *)nfs_vnop_getattr }, /* getattr */
- { &vnop_setattr_desc, (vnop_t *)nfs_setattr }, /* setattr */
- { &vnop_read_desc, (vnop_t *)nfsspec_read }, /* read */
- { &vnop_write_desc, (vnop_t *)nfsspec_write }, /* write */
+ { &vnop_close_desc, (vnop_t *)nfsspec_vnop_close }, /* close */
+ { &vnop_getattr_desc, (vnop_t *)nfs3_vnop_getattr }, /* getattr */
+ { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
+ { &vnop_read_desc, (vnop_t *)nfsspec_vnop_read }, /* read */
+ { &vnop_write_desc, (vnop_t *)nfsspec_vnop_write }, /* write */
{ &vnop_ioctl_desc, (vnop_t *)spec_ioctl }, /* ioctl */
{ &vnop_select_desc, (vnop_t *)spec_select }, /* select */
{ &vnop_revoke_desc, (vnop_t *)spec_revoke }, /* revoke */
{ &vnop_mmap_desc, (vnop_t *)spec_mmap }, /* mmap */
- { &vnop_fsync_desc, (vnop_t *)nfs_fsync }, /* fsync */
+ { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
{ &vnop_remove_desc, (vnop_t *)spec_remove }, /* remove */
{ &vnop_link_desc, (vnop_t *)spec_link }, /* link */
{ &vnop_rename_desc, (vnop_t *)spec_rename }, /* rename */
{ &vnop_symlink_desc, (vnop_t *)spec_symlink }, /* symlink */
{ &vnop_readdir_desc, (vnop_t *)spec_readdir }, /* readdir */
{ &vnop_readlink_desc, (vnop_t *)spec_readlink }, /* readlink */
- { &vnop_inactive_desc, (vnop_t *)nfs_inactive }, /* inactive */
- { &vnop_reclaim_desc, (vnop_t *)nfs_reclaim }, /* reclaim */
+ { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
+ { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
{ &vnop_strategy_desc, (vnop_t *)spec_strategy }, /* strategy */
{ &vnop_pathconf_desc, (vnop_t *)spec_pathconf }, /* pathconf */
{ &vnop_advlock_desc, (vnop_t *)spec_advlock }, /* advlock */
{ &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
- { &vnop_pagein_desc, (vnop_t *)nfs_pagein }, /* Pagein */
- { &vnop_pageout_desc, (vnop_t *)nfs_pageout }, /* Pageout */
- { &vnop_blktooff_desc, (vnop_t *)nfs_blktooff }, /* blktooff */
- { &vnop_offtoblk_desc, (vnop_t *)nfs_offtoblk }, /* offtoblk */
- { &vnop_blockmap_desc, (vnop_t *)nfs_blockmap }, /* blockmap */
+ { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
+ { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
+ { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
+ { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
+ { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
+ { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
{ NULL, NULL }
};
struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
{ &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
-#ifdef __FreeBSD__
-VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
+vnop_t **spec_nfsv4nodeop_p;
+static struct vnodeopv_entry_desc spec_nfsv4nodeop_entries[] = {
+ { &vnop_default_desc, (vnop_t *)vn_default_error },
+ { &vnop_lookup_desc, (vnop_t *)spec_lookup }, /* lookup */
+ { &vnop_create_desc, (vnop_t *)spec_create }, /* create */
+ { &vnop_mknod_desc, (vnop_t *)spec_mknod }, /* mknod */
+ { &vnop_open_desc, (vnop_t *)spec_open }, /* open */
+ { &vnop_close_desc, (vnop_t *)nfsspec_vnop_close }, /* close */
+ { &vnop_getattr_desc, (vnop_t *)nfs4_vnop_getattr }, /* getattr */
+ { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
+ { &vnop_read_desc, (vnop_t *)nfsspec_vnop_read }, /* read */
+ { &vnop_write_desc, (vnop_t *)nfsspec_vnop_write }, /* write */
+ { &vnop_ioctl_desc, (vnop_t *)spec_ioctl }, /* ioctl */
+ { &vnop_select_desc, (vnop_t *)spec_select }, /* select */
+ { &vnop_revoke_desc, (vnop_t *)spec_revoke }, /* revoke */
+ { &vnop_mmap_desc, (vnop_t *)spec_mmap }, /* mmap */
+ { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
+ { &vnop_remove_desc, (vnop_t *)spec_remove }, /* remove */
+ { &vnop_link_desc, (vnop_t *)spec_link }, /* link */
+ { &vnop_rename_desc, (vnop_t *)spec_rename }, /* rename */
+ { &vnop_mkdir_desc, (vnop_t *)spec_mkdir }, /* mkdir */
+ { &vnop_rmdir_desc, (vnop_t *)spec_rmdir }, /* rmdir */
+ { &vnop_symlink_desc, (vnop_t *)spec_symlink }, /* symlink */
+ { &vnop_readdir_desc, (vnop_t *)spec_readdir }, /* readdir */
+ { &vnop_readlink_desc, (vnop_t *)spec_readlink }, /* readlink */
+ { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
+ { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
+ { &vnop_strategy_desc, (vnop_t *)spec_strategy }, /* strategy */
+ { &vnop_pathconf_desc, (vnop_t *)spec_pathconf }, /* pathconf */
+ { &vnop_advlock_desc, (vnop_t *)spec_advlock }, /* advlock */
+ { &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
+ { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
+ { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
+ { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
+ { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
+ { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
+ { &vnop_getxattr_desc, (vnop_t *)nfs4_vnop_getxattr }, /* getxattr */
+ { &vnop_setxattr_desc, (vnop_t *)nfs4_vnop_setxattr }, /* setxattr */
+ { &vnop_removexattr_desc, (vnop_t *)nfs4_vnop_removexattr },/* removexattr */
+ { &vnop_listxattr_desc, (vnop_t *)nfs4_vnop_listxattr },/* listxattr */
+#if NAMEDSTREAMS
+ { &vnop_getnamedstream_desc, (vnop_t *)nfs4_vnop_getnamedstream }, /* getnamedstream */
+ { &vnop_makenamedstream_desc, (vnop_t *)nfs4_vnop_makenamedstream }, /* makenamedstream */
+ { &vnop_removenamedstream_desc, (vnop_t *)nfs4_vnop_removenamedstream },/* removenamedstream */
#endif
+ { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
+ { NULL, NULL }
+};
+struct vnodeopv_desc spec_nfsv4nodeop_opv_desc =
+ { &spec_nfsv4nodeop_p, spec_nfsv4nodeop_entries };
+#if FIFO
vnop_t **fifo_nfsv2nodeop_p;
static struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
{ &vnop_default_desc, (vnop_t *)vn_default_error },
{ &vnop_create_desc, (vnop_t *)fifo_create }, /* create */
{ &vnop_mknod_desc, (vnop_t *)fifo_mknod }, /* mknod */
{ &vnop_open_desc, (vnop_t *)fifo_open }, /* open */
- { &vnop_close_desc, (vnop_t *)nfsfifo_close }, /* close */
- { &vnop_getattr_desc, (vnop_t *)nfs_vnop_getattr }, /* getattr */
- { &vnop_setattr_desc, (vnop_t *)nfs_setattr }, /* setattr */
- { &vnop_read_desc, (vnop_t *)nfsfifo_read }, /* read */
- { &vnop_write_desc, (vnop_t *)nfsfifo_write }, /* write */
+ { &vnop_close_desc, (vnop_t *)nfsfifo_vnop_close }, /* close */
+ { &vnop_getattr_desc, (vnop_t *)nfs3_vnop_getattr }, /* getattr */
+ { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
+ { &vnop_read_desc, (vnop_t *)nfsfifo_vnop_read }, /* read */
+ { &vnop_write_desc, (vnop_t *)nfsfifo_vnop_write }, /* write */
{ &vnop_ioctl_desc, (vnop_t *)fifo_ioctl }, /* ioctl */
{ &vnop_select_desc, (vnop_t *)fifo_select }, /* select */
{ &vnop_revoke_desc, (vnop_t *)fifo_revoke }, /* revoke */
{ &vnop_mmap_desc, (vnop_t *)fifo_mmap }, /* mmap */
- { &vnop_fsync_desc, (vnop_t *)nfs_fsync }, /* fsync */
+ { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
{ &vnop_remove_desc, (vnop_t *)fifo_remove }, /* remove */
{ &vnop_link_desc, (vnop_t *)fifo_link }, /* link */
{ &vnop_rename_desc, (vnop_t *)fifo_rename }, /* rename */
{ &vnop_symlink_desc, (vnop_t *)fifo_symlink }, /* symlink */
{ &vnop_readdir_desc, (vnop_t *)fifo_readdir }, /* readdir */
{ &vnop_readlink_desc, (vnop_t *)fifo_readlink }, /* readlink */
- { &vnop_inactive_desc, (vnop_t *)nfs_inactive }, /* inactive */
- { &vnop_reclaim_desc, (vnop_t *)nfs_reclaim }, /* reclaim */
+ { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
+ { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
{ &vnop_strategy_desc, (vnop_t *)fifo_strategy }, /* strategy */
{ &vnop_pathconf_desc, (vnop_t *)fifo_pathconf }, /* pathconf */
{ &vnop_advlock_desc, (vnop_t *)fifo_advlock }, /* advlock */
{ &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
- { &vnop_pagein_desc, (vnop_t *)nfs_pagein }, /* Pagein */
- { &vnop_pageout_desc, (vnop_t *)nfs_pageout }, /* Pageout */
- { &vnop_blktooff_desc, (vnop_t *)nfs_blktooff }, /* blktooff */
- { &vnop_offtoblk_desc, (vnop_t *)nfs_offtoblk }, /* offtoblk */
- { &vnop_blockmap_desc, (vnop_t *)nfs_blockmap }, /* blockmap */
+ { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
+ { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
+ { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
+ { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
+ { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
+ { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
{ NULL, NULL }
};
struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
{ &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
-#ifdef __FreeBSD__
-VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
+
+vnop_t **fifo_nfsv4nodeop_p;
+static struct vnodeopv_entry_desc fifo_nfsv4nodeop_entries[] = {
+ { &vnop_default_desc, (vnop_t *)vn_default_error },
+ { &vnop_lookup_desc, (vnop_t *)fifo_lookup }, /* lookup */
+ { &vnop_create_desc, (vnop_t *)fifo_create }, /* create */
+ { &vnop_mknod_desc, (vnop_t *)fifo_mknod }, /* mknod */
+ { &vnop_open_desc, (vnop_t *)fifo_open }, /* open */
+ { &vnop_close_desc, (vnop_t *)nfsfifo_vnop_close }, /* close */
+ { &vnop_getattr_desc, (vnop_t *)nfs4_vnop_getattr }, /* getattr */
+ { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
+ { &vnop_read_desc, (vnop_t *)nfsfifo_vnop_read }, /* read */
+ { &vnop_write_desc, (vnop_t *)nfsfifo_vnop_write }, /* write */
+ { &vnop_ioctl_desc, (vnop_t *)fifo_ioctl }, /* ioctl */
+ { &vnop_select_desc, (vnop_t *)fifo_select }, /* select */
+ { &vnop_revoke_desc, (vnop_t *)fifo_revoke }, /* revoke */
+ { &vnop_mmap_desc, (vnop_t *)fifo_mmap }, /* mmap */
+ { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
+ { &vnop_remove_desc, (vnop_t *)fifo_remove }, /* remove */
+ { &vnop_link_desc, (vnop_t *)fifo_link }, /* link */
+ { &vnop_rename_desc, (vnop_t *)fifo_rename }, /* rename */
+ { &vnop_mkdir_desc, (vnop_t *)fifo_mkdir }, /* mkdir */
+ { &vnop_rmdir_desc, (vnop_t *)fifo_rmdir }, /* rmdir */
+ { &vnop_symlink_desc, (vnop_t *)fifo_symlink }, /* symlink */
+ { &vnop_readdir_desc, (vnop_t *)fifo_readdir }, /* readdir */
+ { &vnop_readlink_desc, (vnop_t *)fifo_readlink }, /* readlink */
+ { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
+ { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
+ { &vnop_strategy_desc, (vnop_t *)fifo_strategy }, /* strategy */
+ { &vnop_pathconf_desc, (vnop_t *)fifo_pathconf }, /* pathconf */
+ { &vnop_advlock_desc, (vnop_t *)fifo_advlock }, /* advlock */
+ { &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
+ { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
+ { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
+ { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
+ { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
+ { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
+ { &vnop_getxattr_desc, (vnop_t *)nfs4_vnop_getxattr }, /* getxattr */
+ { &vnop_setxattr_desc, (vnop_t *)nfs4_vnop_setxattr }, /* setxattr */
+ { &vnop_removexattr_desc, (vnop_t *)nfs4_vnop_removexattr },/* removexattr */
+ { &vnop_listxattr_desc, (vnop_t *)nfs4_vnop_listxattr },/* listxattr */
+#if NAMEDSTREAMS
+ { &vnop_getnamedstream_desc, (vnop_t *)nfs4_vnop_getnamedstream }, /* getnamedstream */
+ { &vnop_makenamedstream_desc, (vnop_t *)nfs4_vnop_makenamedstream }, /* makenamedstream */
+ { &vnop_removenamedstream_desc, (vnop_t *)nfs4_vnop_removenamedstream },/* removenamedstream */
#endif
+ { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
+ { NULL, NULL }
+};
+struct vnodeopv_desc fifo_nfsv4nodeop_opv_desc =
+ { &fifo_nfsv4nodeop_p, fifo_nfsv4nodeop_entries };
+#endif /* FIFO */
+
-static int nfs_mknodrpc(vnode_t dvp, vnode_t *vpp,
- struct componentname *cnp,
- struct vnode_attr *vap,
- kauth_cred_t cred, proc_t p);
-static int nfs_removerpc(vnode_t dvp, char *name, int namelen,
- kauth_cred_t cred, proc_t proc);
-static int nfs_renamerpc(vnode_t fdvp, char *fnameptr,
- int fnamelen, vnode_t tdvp,
- char *tnameptr, int tnamelen,
- kauth_cred_t cred, proc_t proc);
+int nfs_sillyrename(nfsnode_t,nfsnode_t,struct componentname *,vfs_context_t);
/*
- * Global variables
- */
-extern u_long nfs_xdrneg1;
-extern u_long nfs_true, nfs_false;
-extern struct nfsstats nfsstats;
-extern nfstype nfsv3_type[9];
-proc_t nfs_iodwant[NFS_MAXASYNCDAEMON];
-struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
-
-lck_grp_t *nfs_iod_lck_grp;
-lck_grp_attr_t *nfs_iod_lck_grp_attr;
-lck_attr_t *nfs_iod_lck_attr;
-lck_mtx_t *nfs_iod_mutex;
-
-int nfs_numasync = 0;
-int nfs_ioddelwri = 0;
-
-#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
-
-static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
-/* SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
- &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
-*/
-#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
- | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
- | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
-
-
-/*
- * the following are needed only by nfs_pageout to know how to handle errors
- * see nfs_pageout comments on explanation of actions.
- * the errors here are copied from errno.h and errors returned by servers
- * are expected to match the same numbers here. If not, our actions maybe
- * erroneous.
+ * Find the slot in the access cache for this UID.
+ * If adding and no existing slot is found, reuse slots in FIFO order.
+ * The index of the next slot to use is kept in the last entry of the n_access array.
*/
-enum actiontype {NOACTION, DUMP, DUMPANDLOG, RETRY, RETRYWITHSLEEP, SEVER};
-
-static int errorcount[ELAST+1]; /* better be zeros when initialized */
-
-static const short errortooutcome[ELAST+1] = {
- NOACTION,
- DUMP, /* EPERM 1 Operation not permitted */
- DUMP, /* ENOENT 2 No such file or directory */
- DUMPANDLOG, /* ESRCH 3 No such process */
- RETRY, /* EINTR 4 Interrupted system call */
- DUMP, /* EIO 5 Input/output error */
- DUMP, /* ENXIO 6 Device not configured */
- DUMPANDLOG, /* E2BIG 7 Argument list too long */
- DUMPANDLOG, /* ENOEXEC 8 Exec format error */
- DUMPANDLOG, /* EBADF 9 Bad file descriptor */
- DUMPANDLOG, /* ECHILD 10 No child processes */
- DUMPANDLOG, /* EDEADLK 11 Resource deadlock avoided - was EAGAIN */
- RETRY, /* ENOMEM 12 Cannot allocate memory */
- DUMP, /* EACCES 13 Permission denied */
- DUMPANDLOG, /* EFAULT 14 Bad address */
- DUMPANDLOG, /* ENOTBLK 15 POSIX - Block device required */
- RETRY, /* EBUSY 16 Device busy */
- DUMP, /* EEXIST 17 File exists */
- DUMP, /* EXDEV 18 Cross-device link */
- DUMP, /* ENODEV 19 Operation not supported by device */
- DUMP, /* ENOTDIR 20 Not a directory */
- DUMP, /* EISDIR 21 Is a directory */
- DUMP, /* EINVAL 22 Invalid argument */
- DUMPANDLOG, /* ENFILE 23 Too many open files in system */
- DUMPANDLOG, /* EMFILE 24 Too many open files */
- DUMPANDLOG, /* ENOTTY 25 Inappropriate ioctl for device */
- DUMPANDLOG, /* ETXTBSY 26 Text file busy - POSIX */
- DUMP, /* EFBIG 27 File too large */
- DUMP, /* ENOSPC 28 No space left on device */
- DUMPANDLOG, /* ESPIPE 29 Illegal seek */
- DUMP, /* EROFS 30 Read-only file system */
- DUMP, /* EMLINK 31 Too many links */
- RETRY, /* EPIPE 32 Broken pipe */
- /* math software */
- DUMPANDLOG, /* EDOM 33 Numerical argument out of domain */
- DUMPANDLOG, /* ERANGE 34 Result too large */
- RETRY, /* EAGAIN/EWOULDBLOCK 35 Resource temporarily unavailable */
- DUMPANDLOG, /* EINPROGRESS 36 Operation now in progress */
- DUMPANDLOG, /* EALREADY 37 Operation already in progress */
- /* ipc/network software -- argument errors */
- DUMPANDLOG, /* ENOTSOC 38 Socket operation on non-socket */
- DUMPANDLOG, /* EDESTADDRREQ 39 Destination address required */
- DUMPANDLOG, /* EMSGSIZE 40 Message too long */
- DUMPANDLOG, /* EPROTOTYPE 41 Protocol wrong type for socket */
- DUMPANDLOG, /* ENOPROTOOPT 42 Protocol not available */
- DUMPANDLOG, /* EPROTONOSUPPORT 43 Protocol not supported */
- DUMPANDLOG, /* ESOCKTNOSUPPORT 44 Socket type not supported */
- DUMPANDLOG, /* ENOTSUP 45 Operation not supported */
- DUMPANDLOG, /* EPFNOSUPPORT 46 Protocol family not supported */
- DUMPANDLOG, /* EAFNOSUPPORT 47 Address family not supported by protocol family */
- DUMPANDLOG, /* EADDRINUSE 48 Address already in use */
- DUMPANDLOG, /* EADDRNOTAVAIL 49 Can't assign requested address */
- /* ipc/network software -- operational errors */
- RETRY, /* ENETDOWN 50 Network is down */
- RETRY, /* ENETUNREACH 51 Network is unreachable */
- RETRY, /* ENETRESET 52 Network dropped connection on reset */
- RETRY, /* ECONNABORTED 53 Software caused connection abort */
- RETRY, /* ECONNRESET 54 Connection reset by peer */
- RETRY, /* ENOBUFS 55 No buffer space available */
- RETRY, /* EISCONN 56 Socket is already connected */
- RETRY, /* ENOTCONN 57 Socket is not connected */
- RETRY, /* ESHUTDOWN 58 Can't send after socket shutdown */
- RETRY, /* ETOOMANYREFS 59 Too many references: can't splice */
- RETRY, /* ETIMEDOUT 60 Operation timed out */
- RETRY, /* ECONNREFUSED 61 Connection refused */
-
- DUMPANDLOG, /* ELOOP 62 Too many levels of symbolic links */
- DUMP, /* ENAMETOOLONG 63 File name too long */
- RETRY, /* EHOSTDOWN 64 Host is down */
- RETRY, /* EHOSTUNREACH 65 No route to host */
- DUMP, /* ENOTEMPTY 66 Directory not empty */
- /* quotas & mush */
- DUMPANDLOG, /* PROCLIM 67 Too many processes */
- DUMPANDLOG, /* EUSERS 68 Too many users */
- DUMPANDLOG, /* EDQUOT 69 Disc quota exceeded */
- /* Network File System */
- DUMP, /* ESTALE 70 Stale NFS file handle */
- DUMP, /* EREMOTE 71 Too many levels of remote in path */
- DUMPANDLOG, /* EBADRPC 72 RPC struct is bad */
- DUMPANDLOG, /* ERPCMISMATCH 73 RPC version wrong */
- DUMPANDLOG, /* EPROGUNAVAIL 74 RPC prog. not avail */
- DUMPANDLOG, /* EPROGMISMATCH 75 Program version wrong */
- DUMPANDLOG, /* EPROCUNAVAIL 76 Bad procedure for program */
-
- DUMPANDLOG, /* ENOLCK 77 No locks available */
- DUMPANDLOG, /* ENOSYS 78 Function not implemented */
- DUMPANDLOG, /* EFTYPE 79 Inappropriate file type or format */
- DUMPANDLOG, /* EAUTH 80 Authentication error */
- DUMPANDLOG, /* ENEEDAUTH 81 Need authenticator */
- /* Intelligent device errors */
- DUMPANDLOG, /* EPWROFF 82 Device power is off */
- DUMPANDLOG, /* EDEVERR 83 Device error, e.g. paper out */
- DUMPANDLOG, /* EOVERFLOW 84 Value too large to be stored in data type */
- /* Program loading errors */
- DUMPANDLOG, /* EBADEXEC 85 Bad executable */
- DUMPANDLOG, /* EBADARCH 86 Bad CPU type in executable */
- DUMPANDLOG, /* ESHLIBVERS 87 Shared library version mismatch */
- DUMPANDLOG, /* EBADMACHO 88 Malformed Macho file */
-};
-
-
-static short
-nfs_pageouterrorhandler(int error)
+int
+nfs_node_access_slot(nfsnode_t np, uid_t uid, int add)
{
- if (error > ELAST)
- return(DUMP);
- else
- return(errortooutcome[error]);
+ int slot;
+
+ for (slot=0; slot < NFS_ACCESS_CACHE_SIZE; slot++)
+ if (np->n_accessuid[slot] == uid)
+ break;
+ if (slot == NFS_ACCESS_CACHE_SIZE) {
+ if (!add)
+ return (-1);
+ slot = np->n_access[NFS_ACCESS_CACHE_SIZE];
+ np->n_access[NFS_ACCESS_CACHE_SIZE] = (slot + 1) % NFS_ACCESS_CACHE_SIZE;
+ }
+ return (slot);
}
-static int
-nfs3_access_otw(vnode_t vp,
- int wmode,
- proc_t p,
- kauth_cred_t cred)
+int
+nfs3_access_rpc(nfsnode_t np, u_int32_t *access, vfs_context_t ctx)
{
- const int v3 = 1;
- u_long *tl;
- int error = 0, attrflag;
-
- mbuf_t mreq, mrep, md, mb, mb2;
- caddr_t bpos, dpos, cp2;
- register long t1, t2;
- register caddr_t cp;
- u_int32_t rmode;
- struct nfsnode *np = VTONFS(vp);
+ int error = 0, lockerror = ENOENT, status, slot;
+ uint32_t access_result = 0;
u_int64_t xid;
+ struct nfsm_chain nmreq, nmrep;
struct timeval now;
+ uid_t uid;
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(NFS_VER3) + NFSX_UNSIGNED);
+ nfsm_chain_add_fh(error, &nmreq, NFS_VER3, np->n_fhp, np->n_fhsize);
+ nfsm_chain_add_32(error, &nmreq, *access);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request(np, NULL, &nmreq, NFSPROC_ACCESS, ctx, NULL, &nmrep, &xid, &status);
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
+ if (!error)
+ error = status;
+ nfsm_chain_get_32(error, &nmrep, access_result);
+ nfsmout_if(error);
- nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED);
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_ACCESS]);
- nfsm_fhtom(vp, v3);
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = txdr_unsigned(wmode);
- nfsm_request(vp, NFSPROC_ACCESS, p, cred, &xid);
- if (mrep) {
- nfsm_postop_attr_update(vp, 1, attrflag, &xid);
- }
- if (!error) {
- nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
- rmode = fxdr_unsigned(u_int32_t, *tl);
- np->n_mode = rmode;
- np->n_modeuid = kauth_cred_getuid(cred);
- microuptime(&now);
- np->n_modestamp = now.tv_sec;
- }
- nfsm_reqdone;
- return error;
+ uid = kauth_cred_getuid(vfs_context_ucred(ctx));
+ slot = nfs_node_access_slot(np, uid, 1);
+ np->n_accessuid[slot] = uid;
+ microuptime(&now);
+ np->n_accessstamp[slot] = now.tv_sec;
+ np->n_access[slot] = access_result;
+
+ /*
+ * If we asked for DELETE but didn't get it, the server
+ * may simply not support returning that bit (possible
+ * on UNIX systems). So, we'll assume that it is OK,
+ * and just let any subsequent delete action fail if it
+ * really isn't deletable.
+ */
+ if ((*access & NFS_ACCESS_DELETE) &&
+ !(np->n_access[slot] & NFS_ACCESS_DELETE))
+ np->n_access[slot] |= NFS_ACCESS_DELETE;
+ /* ".zfs" subdirectories may erroneously give a denied answer for add/remove */
+ if (nfs_access_dotzfs && (np->n_flag & NISDOTZFSCHILD))
+ np->n_access[slot] |= (NFS_ACCESS_MODIFY|NFS_ACCESS_EXTEND|NFS_ACCESS_DELETE);
+ /* pass back the access returned with this request */
+ *access = np->n_access[slot];
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
}
/*
- * nfs access vnode op.
- * For nfs version 2, just return ok. File accesses may fail later.
- * For nfs version 3, use the access rpc to check accessibility. If file modes
- * are changed on the server, accesses might still fail later.
+ * NFS access vnode op.
+ * For NFS version 2, just return ok. File accesses may fail later.
+ * For NFS version 3+, use the access RPC to check accessibility. If file
+ * permissions are changed on the server, accesses might still fail later.
*/
-static int
-nfs_access(ap)
+int
+nfs_vnop_access(
struct vnop_access_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
- int a_mode;
+ int a_action;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t vp = ap->a_vp;
- int error = 0, dorpc;
- u_long mode, wmode;
- int v3 = NFS_ISV3(vp);
- struct nfsnode *np = VTONFS(vp);
+ int error = 0, slot, dorpc;
+ u_int32_t access, waccess;
+ nfsnode_t np = VTONFS(vp);
+ struct nfsmount *nmp;
+ int nfsvers;
struct timeval now;
- kauth_cred_t cred;
+ uid_t uid;
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+
+ if (nfsvers == NFS_VER2) {
+ if ((ap->a_action & KAUTH_VNODE_WRITE_RIGHTS) &&
+ vfs_isrdonly(vnode_mount(vp)))
+ return (EROFS);
+ return (0);
+ }
/*
- * For nfs v3, do an access rpc, otherwise you are stuck emulating
+ * For NFS v3, do an access rpc, otherwise you are stuck emulating
* ufs_access() locally using the vattr. This may not be correct,
* since the server may apply other access criteria such as
* client uid-->server uid mapping that we do not know about, but
* this is better than just returning anything that is lying about
* in the cache.
*/
- if (v3) {
- /*
- * Convert KAUTH primitives to NFS access rights.
- */
- mode = 0;
- if (vnode_isdir(vp)) {
- /* directory */
- if (ap->a_action &
- (KAUTH_VNODE_LIST_DIRECTORY |
- KAUTH_VNODE_READ_EXTATTRIBUTES))
- mode |= NFSV3ACCESS_READ;
- if (ap->a_action & KAUTH_VNODE_SEARCH)
- mode |= NFSV3ACCESS_LOOKUP;
- if (ap->a_action &
- (KAUTH_VNODE_ADD_FILE |
- KAUTH_VNODE_ADD_SUBDIRECTORY))
- mode |= NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND;
- if (ap->a_action & KAUTH_VNODE_DELETE_CHILD)
- mode |= NFSV3ACCESS_MODIFY;
- } else {
- /* file */
- if (ap->a_action &
- (KAUTH_VNODE_READ_DATA |
- KAUTH_VNODE_READ_EXTATTRIBUTES))
- mode |= NFSV3ACCESS_READ;
- if (ap->a_action & KAUTH_VNODE_WRITE_DATA)
- mode |= NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND;
- if (ap->a_action & KAUTH_VNODE_APPEND_DATA)
- mode |= NFSV3ACCESS_EXTEND;
- if (ap->a_action & KAUTH_VNODE_EXECUTE)
- mode |= NFSV3ACCESS_EXECUTE;
- }
- /* common */
- if (ap->a_action & KAUTH_VNODE_DELETE)
- mode |= NFSV3ACCESS_DELETE;
- if (ap->a_action &
- (KAUTH_VNODE_WRITE_ATTRIBUTES |
- KAUTH_VNODE_WRITE_EXTATTRIBUTES |
- KAUTH_VNODE_WRITE_SECURITY))
- mode |= NFSV3ACCESS_MODIFY;
- /* XXX this is pretty dubious */
- if (ap->a_action & KAUTH_VNODE_CHANGE_OWNER)
- mode |= NFSV3ACCESS_MODIFY;
-
- /* if caching, always ask for every right */
- if (nfsaccess_cache_timeout > 0) {
- wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
- NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
- NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
- } else
- wmode = mode;
-
- cred = vfs_context_ucred(ap->a_context);
- /*
- * Does our cached result allow us to give a definite yes to
- * this request?
- */
- dorpc = 1;
- if (NMODEVALID(np)) {
- microuptime(&now);
- if ((now.tv_sec < (np->n_modestamp + nfsaccess_cache_timeout)) &&
- (kauth_cred_getuid(cred) == np->n_modeuid) &&
- ((np->n_mode & mode) == mode)) {
- /* OSAddAtomic(1, (SInt32*)&nfsstats.accesscache_hits); */
- dorpc = 0;
- }
- }
- if (dorpc) {
- /* Either a no, or a don't know. Go to the wire. */
- /* OSAddAtomic(1, (SInt32*)&nfsstats.accesscache_misses); */
- error = nfs3_access_otw(vp, wmode, vfs_context_proc(ap->a_context), cred);
- }
- if (!error) {
- /*
- * If we asked for DELETE but didn't get it, the server
- * may simply not support returning that bit (possible
- * on UNIX systems). So, we'll assume that it is OK,
- * and just let any subsequent delete action fail if it
- * really isn't deletable.
- */
- if ((mode & NFSV3ACCESS_DELETE) &&
- !(np->n_mode & NFSV3ACCESS_DELETE))
- np->n_mode |= NFSV3ACCESS_DELETE;
- if ((np->n_mode & mode) != mode)
- error = EACCES;
- }
+ /*
+ * In addition if the kernel is checking for access, KAUTH_VNODE_ACCESS
+ * not set, just return. At this moment do not know what the state of
+ * the server is and what ever we get back be it either yea or nay is
+ * going to be stale. Finder (Desktop services/FileURL) might hang when
+ * going over the wire when just asking getattrlist for the roots FSID
+ * since we are going to be called to see if we're authorized for
+ * search. Since we are returning without checking the cache and/or
+ * going over the wire, it makes no sense to update the cache.
+ *
+ * N.B. This is also the strategy that SMB is using.
+ */
+ if (!(ap->a_action & KAUTH_VNODE_ACCESS))
+ return (0);
+
+ /*
+ * Convert KAUTH primitives to NFS access rights.
+ */
+ access = 0;
+ if (vnode_isdir(vp)) {
+ /* directory */
+ if (ap->a_action &
+ (KAUTH_VNODE_LIST_DIRECTORY |
+ KAUTH_VNODE_READ_EXTATTRIBUTES))
+ access |= NFS_ACCESS_READ;
+ if (ap->a_action & KAUTH_VNODE_SEARCH)
+ access |= NFS_ACCESS_LOOKUP;
+ if (ap->a_action &
+ (KAUTH_VNODE_ADD_FILE |
+ KAUTH_VNODE_ADD_SUBDIRECTORY))
+ access |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
+ if (ap->a_action & KAUTH_VNODE_DELETE_CHILD)
+ access |= NFS_ACCESS_MODIFY;
} else {
- /* v2 */
- if ((ap->a_action & KAUTH_VNODE_WRITE_RIGHTS) && vfs_isrdonly(vnode_mount(vp))) {
- error = EROFS;
- } else {
- error = 0;
+ /* file */
+ if (ap->a_action &
+ (KAUTH_VNODE_READ_DATA |
+ KAUTH_VNODE_READ_EXTATTRIBUTES))
+ access |= NFS_ACCESS_READ;
+ if (ap->a_action & KAUTH_VNODE_WRITE_DATA)
+ access |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
+ if (ap->a_action & KAUTH_VNODE_APPEND_DATA)
+ access |= NFS_ACCESS_EXTEND;
+ if (ap->a_action & KAUTH_VNODE_EXECUTE)
+ access |= NFS_ACCESS_EXECUTE;
+ }
+ /* common */
+ if (ap->a_action & KAUTH_VNODE_DELETE)
+ access |= NFS_ACCESS_DELETE;
+ if (ap->a_action &
+ (KAUTH_VNODE_WRITE_ATTRIBUTES |
+ KAUTH_VNODE_WRITE_EXTATTRIBUTES |
+ KAUTH_VNODE_WRITE_SECURITY))
+ access |= NFS_ACCESS_MODIFY;
+ /* XXX this is pretty dubious */
+ if (ap->a_action & KAUTH_VNODE_CHANGE_OWNER)
+ access |= NFS_ACCESS_MODIFY;
+
+ /* if caching, always ask for every right */
+ if (nfs_access_cache_timeout > 0) {
+ waccess = NFS_ACCESS_READ | NFS_ACCESS_MODIFY |
+ NFS_ACCESS_EXTEND | NFS_ACCESS_EXECUTE |
+ NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
+ } else {
+ waccess = access;
+ }
+
+ if ((error = nfs_node_lock(np)))
+ return (error);
+
+ /*
+ * Does our cached result allow us to give a definite yes to
+ * this request?
+ */
+ uid = kauth_cred_getuid(vfs_context_ucred(ctx));
+ slot = nfs_node_access_slot(np, uid, 0);
+ dorpc = 1;
+ if (access == 0) {
+ /* not asking for any rights understood by NFS, so don't bother doing an RPC */
+ /* OSAddAtomic(1, &nfsstats.accesscache_hits); */
+ dorpc = 0;
+ waccess = 0;
+ } else if (NACCESSVALID(np, slot)) {
+ microuptime(&now);
+ if ((now.tv_sec < (np->n_accessstamp[slot] + nfs_access_cache_timeout)) &&
+ ((np->n_access[slot] & access) == access)) {
+ /* OSAddAtomic(1, &nfsstats.accesscache_hits); */
+ dorpc = 0;
+ waccess = np->n_access[slot];
}
}
+ nfs_node_unlock(np);
+ if (dorpc) {
+ /* Either a no, or a don't know. Go to the wire. */
+ /* OSAddAtomic(1, &nfsstats.accesscache_misses); */
+ error = nmp->nm_funcs->nf_access_rpc(np, &waccess, ctx);
+ }
+ if (!error && ((waccess & access) != access))
+ error = EACCES;
return (error);
}
+
/*
- * nfs open vnode op
- * Check to see if the type is ok
- * and that deletion is not in progress.
- * For paged in text files, you will need to flush the page cache
- * if consistency is lost.
+ * NFS open vnode op
+ *
+ * Perform various update/invalidation checks and then add the
+ * open to the node. Regular files will have an open file structure
+ * on the node and, for NFSv4, perform an OPEN request on the server.
*/
-/* ARGSUSED */
-
-static int
-nfs_open(ap)
+int
+nfs_vnop_open(
struct vnop_open_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
int a_mode;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t vp = ap->a_vp;
- struct nfsnode *np = VTONFS(vp);
- struct nfs_vattr nvattr;
- kauth_cred_t cred;
- proc_t p;
+ nfsnode_t np = VTONFS(vp);
+ struct nfsmount *nmp;
+ int error, accessMode, denyMode, opened = 0;
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
enum vtype vtype;
- int error;
+
+ if (!(ap->a_mode & (FREAD|FWRITE)))
+ return (EINVAL);
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+ if (np->n_flag & NREVOKE)
+ return (EIO);
vtype = vnode_vtype(vp);
- if (vtype != VREG && vtype != VDIR && vtype != VLNK) {
+ if ((vtype != VREG) && (vtype != VDIR) && (vtype != VLNK))
return (EACCES);
- }
-
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ /* First, check if we need to update/invalidate */
+ if (ISSET(np->n_flag, NUPDATESIZE))
+ nfs_data_update_size(np, 0);
+ if ((error = nfs_node_lock(np)))
+ return (error);
if (np->n_flag & NNEEDINVALIDATE) {
np->n_flag &= ~NNEEDINVALIDATE;
- nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, cred, p, 1);
+ if (vtype == VDIR)
+ nfs_invaldir(np);
+ nfs_node_unlock(np);
+ nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1);
+ if ((error = nfs_node_lock(np)))
+ return (error);
}
+ if (vtype == VREG)
+ np->n_lastrahead = -1;
if (np->n_flag & NMODIFIED) {
- if ((error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) == EINTR)
- return (error);
- NATTRINVALIDATE(np);
if (vtype == VDIR)
- np->n_direofoffset = 0;
- error = nfs_getattr(vp, &nvattr, cred, p);
- if (error)
+ nfs_invaldir(np);
+ nfs_node_unlock(np);
+ if ((error = nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1)))
return (error);
- if (vtype == VDIR) {
- /* if directory changed, purge any name cache entries */
- if (nfstimespeccmp(&np->n_ncmtime, &nvattr.nva_mtime, !=))
- cache_purge(vp);
- np->n_ncmtime = nvattr.nva_mtime;
- }
- np->n_mtime = nvattr.nva_mtime;
} else {
- error = nfs_getattr(vp, &nvattr, cred, p);
- if (error)
- return (error);
- if (nfstimespeccmp(&np->n_mtime, &nvattr.nva_mtime, !=)) {
- if (vtype == VDIR) {
- np->n_direofoffset = 0;
- nfs_invaldir(vp);
- /* purge name cache entries */
- if (nfstimespeccmp(&np->n_ncmtime, &nvattr.nva_mtime, !=))
- cache_purge(vp);
+ nfs_node_unlock(np);
+ }
+
+ /* nfs_getattr() will check changed and purge caches */
+ if ((error = nfs_getattr(np, NULL, ctx, NGA_UNCACHED)))
+ return (error);
+
+ if (vtype != VREG) {
+ /* Just mark that it was opened */
+ lck_mtx_lock(&np->n_openlock);
+ np->n_openrefcnt++;
+ lck_mtx_unlock(&np->n_openlock);
+ return (0);
+ }
+
+ /* mode contains some combination of: FREAD, FWRITE, O_SHLOCK, O_EXLOCK */
+ accessMode = 0;
+ if (ap->a_mode & FREAD)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_READ;
+ if (ap->a_mode & FWRITE)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_WRITE;
+ if (ap->a_mode & O_EXLOCK)
+ denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+ else if (ap->a_mode & O_SHLOCK)
+ denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+ else
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ // XXX don't do deny modes just yet (and never do it for !v4)
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+
+ noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 1);
+ if (!noop)
+ return (ENOMEM);
+
+restart:
+ error = nfs_mount_state_in_use_start(nmp, vfs_context_thread(ctx));
+ if (error) {
+ nfs_open_owner_rele(noop);
+ return (error);
+ }
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ nfs_mount_state_in_use_end(nmp, 0);
+ nfs_open_owner_rele(noop);
+ return (error);
+ }
+
+ error = nfs_open_file_find(np, noop, &nofp, accessMode, denyMode, 1);
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
+ NP(np, "nfs_vnop_open: LOST %d", kauth_cred_getuid(nofp->nof_owner->noo_cred));
+ error = EIO;
+ }
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ error = nfs4_reopen(nofp, vfs_context_thread(ctx));
+ nofp = NULL;
+ if (!error)
+ goto restart;
+ }
+ if (!error)
+ error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
+ if (error) {
+ nofp = NULL;
+ goto out;
+ }
+
+ if (nmp->nm_vers < NFS_VER4) {
+ /*
+ * NFS v2/v3 opens are always allowed - so just add it.
+ */
+ nfs_open_file_add_open(nofp, accessMode, denyMode, 0);
+ goto out;
+ }
+
+ /*
+ * If we just created the file and the modes match, then we simply use
+ * the open performed in the create. Otherwise, send the request.
+ */
+ if ((nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
+ (nofp->nof_creator == current_thread()) &&
+ (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) &&
+ (denyMode == NFS_OPEN_SHARE_DENY_NONE)) {
+ nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
+ nofp->nof_creator = NULL;
+ } else {
+ if (!opened)
+ error = nfs4_open(np, nofp, accessMode, denyMode, ctx);
+ if ((error == EACCES) && (nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
+ (nofp->nof_creator == current_thread())) {
+ /*
+ * Ugh. This can happen if we just created the file with read-only
+ * perms and we're trying to open it for real with different modes
+ * (e.g. write-only or with a deny mode) and the server decides to
+ * not allow the second open because of the read-only perms.
+ * The best we can do is to just use the create's open.
+ * We may have access we don't need or we may not have a requested
+ * deny mode. We may log complaints later, but we'll try to avoid it.
+ */
+ if (denyMode != NFS_OPEN_SHARE_DENY_NONE)
+ NP(np, "nfs_vnop_open: deny mode foregone on create, %d", kauth_cred_getuid(nofp->nof_owner->noo_cred));
+ nofp->nof_creator = NULL;
+ error = 0;
+ }
+ if (error)
+ goto out;
+ opened = 1;
+ /*
+ * If we had just created the file, we already had it open.
+ * If the actual open mode is less than what we grabbed at
+ * create time, then we'll downgrade the open here.
+ */
+ if ((nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
+ (nofp->nof_creator == current_thread())) {
+ error = nfs_close(np, nofp, NFS_OPEN_SHARE_ACCESS_BOTH, NFS_OPEN_SHARE_DENY_NONE, ctx);
+ if (error)
+ NP(np, "nfs_vnop_open: create close error %d, %d", error, kauth_cred_getuid(nofp->nof_owner->noo_cred));
+ if (!nfs_mount_state_error_should_restart(error)) {
+ error = 0;
+ nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
}
- if ((error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1)) == EINTR)
- return (error);
- if (vtype == VDIR)
- np->n_ncmtime = nvattr.nva_mtime;
- np->n_mtime = nvattr.nva_mtime;
}
}
- NATTRINVALIDATE(np); /* For Open/Close consistency */
- return (0);
+
+out:
+ if (nofp)
+ nfs_open_file_clear_busy(nofp);
+ if (nfs_mount_state_in_use_end(nmp, error)) {
+ nofp = NULL;
+ goto restart;
+ }
+ if (error)
+ NP(np, "nfs_vnop_open: error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
+ if (noop)
+ nfs_open_owner_rele(noop);
+ if (!error && vtype == VREG && (ap->a_mode & FWRITE)) {
+ lck_mtx_lock(&nmp->nm_lock);
+ nmp->nm_state &= ~NFSSTA_SQUISHY;
+ nmp->nm_curdeadtimeout = nmp->nm_deadtimeout;
+ if (nmp->nm_curdeadtimeout <= 0)
+ nmp->nm_deadto_start = 0;
+ nmp->nm_writers++;
+ lck_mtx_unlock(&nmp->nm_lock);
+ }
+
+ return (error);
+}
+
+static uint32_t
+nfs_no_of_open_file_writers(nfsnode_t np)
+{
+ uint32_t writers = 0;
+ struct nfs_open_file *nofp;
+
+ TAILQ_FOREACH(nofp, &np->n_opens, nof_link) {
+ writers += nofp->nof_w + nofp->nof_rw + nofp->nof_w_dw + nofp->nof_rw_dw +
+ nofp->nof_w_drw + nofp->nof_rw_drw + nofp->nof_d_w_dw +
+ nofp->nof_d_rw_dw + nofp->nof_d_w_drw + nofp->nof_d_rw_drw +
+ nofp->nof_d_w + nofp->nof_d_rw;
+ }
+
+ return (writers);
}
/*
- * nfs close vnode op
+ * NFS close vnode op
+ *
* What an NFS client should do upon close after writing is a debatable issue.
* Most NFS clients push delayed writes to the server upon close, basically for
* two reasons:
*
* The current code does the following:
* for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
- * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
- * them.
+ * for NFS Version 3 - flush dirty buffers to the server but don't invalidate them.
+ * for NFS Version 4 - basically the same as NFSv3
*/
-/* ARGSUSED */
-static int
-nfs_close(ap)
+int
+nfs_vnop_close(
struct vnop_close_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
int a_fflag;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t vp = ap->a_vp;
- struct nfsnode *np = VTONFS(vp);
+ nfsnode_t np = VTONFS(vp);
struct nfsmount *nmp;
- kauth_cred_t cred;
- proc_t p;
- int error = 0;
+ int error = 0, error1, nfsvers;
+ int fflag = ap->a_fflag;
+ enum vtype vtype;
+ int accessMode, denyMode;
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ vtype = vnode_vtype(vp);
- if (vnode_vtype(vp) == VREG) {
-#if DIAGNOSTIC
- register struct sillyrename *sp = np->n_sillyrename;
- if (sp)
- kprintf("nfs_close: %s, dvp=%x, vp=%x, ap=%x, np=%x, sp=%x\n",
- &sp->s_name[0], (unsigned)(sp->s_dvp), (unsigned)vp,
- (unsigned)ap, (unsigned)np, (unsigned)sp);
-#endif
- nmp = VFSTONFS(vnode_mount(vp));
- if (!nmp)
- return (ENXIO);
- if (np->n_flag & NNEEDINVALIDATE) {
+ /* First, check if we need to update/flush/invalidate */
+ if (ISSET(np->n_flag, NUPDATESIZE))
+ nfs_data_update_size(np, 0);
+ nfs_node_lock_force(np);
+ if (np->n_flag & NNEEDINVALIDATE) {
np->n_flag &= ~NNEEDINVALIDATE;
- nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, cred, p, 1);
- }
- if (np->n_flag & NMODIFIED) {
- if (NFS_ISV3(vp)) {
- error = nfs_flush(vp, MNT_WAIT, cred, p, 0);
- /*
- * We cannot clear the NMODIFIED bit in np->n_flag due to
- * potential races with other processes
- * NMODIFIED is a hint
- */
- /* np->n_flag &= ~NMODIFIED; */
- } else {
- error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1);
- }
+ nfs_node_unlock(np);
+ nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1);
+ nfs_node_lock_force(np);
+ }
+ if ((vtype == VREG) && (np->n_flag & NMODIFIED) && (fflag & FWRITE)) {
+ /* we're closing an open for write and the file is modified, so flush it */
+ nfs_node_unlock(np);
+ if (nfsvers != NFS_VER2)
+ error = nfs_flush(np, MNT_WAIT, vfs_context_thread(ctx), 0);
+ else
+ error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
+ nfs_node_lock_force(np);
NATTRINVALIDATE(np);
- }
- if (np->n_flag & NWRITEERR) {
+ }
+ if (np->n_flag & NWRITEERR) {
np->n_flag &= ~NWRITEERR;
error = np->n_error;
- }
}
+ nfs_node_unlock(np);
+
+ if (vtype != VREG) {
+ /* Just mark that it was closed */
+ lck_mtx_lock(&np->n_openlock);
+ if (np->n_openrefcnt == 0) {
+ if (fflag & (FREAD|FWRITE)) {
+ NP(np, "nfs_vnop_close: open reference underrun");
+ error = EINVAL;
+ }
+ } else if (fflag & (FREAD|FWRITE)) {
+ np->n_openrefcnt--;
+ } else {
+ /* No FREAD/FWRITE set - probably the final close */
+ np->n_openrefcnt = 0;
+ }
+ lck_mtx_unlock(&np->n_openlock);
+ return (error);
+ }
+ error1 = error;
+
+ /* fflag should contain some combination of: FREAD, FWRITE, FHASLOCK */
+ accessMode = 0;
+ if (fflag & FREAD)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_READ;
+ if (fflag & FWRITE)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_WRITE;
+// XXX It would be nice if we still had the O_EXLOCK/O_SHLOCK flags that were on the open
+// if (fflag & O_EXLOCK)
+// denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+// else if (fflag & O_SHLOCK)
+// denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+// else
+// denyMode = NFS_OPEN_SHARE_DENY_NONE;
+#if 0 // Not yet
+ if (fflag & FHASLOCK) {
+ /* XXX assume FHASLOCK is for the deny mode and not flock */
+ /* FHASLOCK flock will be unlocked in the close path, but the flag is not cleared. */
+ if (nofp->nof_deny & NFS_OPEN_SHARE_DENY_READ)
+ denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+ else if (nofp->nof_deny & NFS_OPEN_SHARE_DENY_WRITE)
+ denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+ else
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ } else {
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ }
+#else
+ // XXX don't do deny modes just yet (and never do it for !v4)
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+#endif
+
+ if (!accessMode) {
+ /*
+ * No mode given to close?
+ * Guess this is the final close.
+ * We should unlock all locks and close all opens.
+ */
+ uint32_t writers;
+ mount_t mp = vnode_mount(vp);
+ int force = (!mp || (mp->mnt_kern_flag & MNTK_FRCUNMOUNT));
+
+ writers = nfs_no_of_open_file_writers(np);
+ nfs_release_open_state_for_node(np, force);
+ if (writers) {
+ lck_mtx_lock(&nmp->nm_lock);
+ if (writers > nmp->nm_writers) {
+ NP(np, "nfs_vnop_close: number of write opens for mount underrun. Node has %d"
+ " opens for write. Mount has total of %d opens for write\n",
+ writers, nmp->nm_writers);
+ nmp->nm_writers = 0;
+ } else {
+ nmp->nm_writers -= writers;
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+ }
+
+ return (error);
+ } else if (fflag & FWRITE) {
+ lck_mtx_lock(&nmp->nm_lock);
+ if (nmp->nm_writers == 0) {
+ NP(np, "nfs_vnop_close: removing open writer from mount, but mount has no files open for writing");
+ } else {
+ nmp->nm_writers--;
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+ }
+
+
+ noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 0);
+ if (!noop) {
+ // printf("nfs_vnop_close: can't get open owner!\n");
+ return (EIO);
+ }
+
+restart:
+ error = nfs_mount_state_in_use_start(nmp, NULL);
+ if (error) {
+ nfs_open_owner_rele(noop);
+ return (error);
+ }
+
+ error = nfs_open_file_find(np, noop, &nofp, 0, 0, 0);
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ error = nfs4_reopen(nofp, NULL);
+ nofp = NULL;
+ if (!error)
+ goto restart;
+ }
+ if (error) {
+ NP(np, "nfs_vnop_close: no open file for owner, error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
+ error = EBADF;
+ goto out;
+ }
+ error = nfs_open_file_set_busy(nofp, NULL);
+ if (error) {
+ nofp = NULL;
+ goto out;
+ }
+
+ error = nfs_close(np, nofp, accessMode, denyMode, ctx);
+ if (error)
+ NP(np, "nfs_vnop_close: close error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
+
+out:
+ if (nofp)
+ nfs_open_file_clear_busy(nofp);
+ if (nfs_mount_state_in_use_end(nmp, error)) {
+ nofp = NULL;
+ goto restart;
+ }
+ if (!error)
+ error = error1;
+ if (error)
+ NP(np, "nfs_vnop_close: error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
+ if (noop)
+ nfs_open_owner_rele(noop);
return (error);
}
-
+/*
+ * nfs_close(): common function that does all the heavy lifting of file closure
+ *
+ * Takes an open file structure and a set of access/deny modes and figures out how
+ * to update the open file structure (and the state on the server) appropriately.
+ */
int
-nfs_getattr_no_vnode(
- mount_t mp,
- u_char *fhp,
- int fhsize,
- kauth_cred_t cred,
- proc_t p,
- struct nfs_vattr *nvap,
- u_int64_t *xidp)
+nfs_close(
+ nfsnode_t np,
+ struct nfs_open_file *nofp,
+ uint32_t accessMode,
+ uint32_t denyMode,
+ vfs_context_t ctx)
{
- mbuf_t mreq, mrep, md, mb, mb2;
- caddr_t bpos, dpos;
- int t2;
- u_long *tl;
- caddr_t cp;
- struct nfsmount *nmp = VFSTONFS(mp);
- int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
- int hsiz;
- int error = 0;
+ struct nfs_lock_owner *nlop;
+ int error = 0, changed = 0, delegated = 0, closed = 0, downgrade = 0;
+ uint32_t newAccessMode, newDenyMode;
+
+ /* warn if modes don't match current state */
+ if (((accessMode & nofp->nof_access) != accessMode) || ((denyMode & nofp->nof_deny) != denyMode))
+ NP(np, "nfs_close: mode mismatch %d %d, current %d %d, %d",
+ accessMode, denyMode, nofp->nof_access, nofp->nof_deny,
+ kauth_cred_getuid(nofp->nof_owner->noo_cred));
- // XXX fix this to use macros once the macros get cleaned up
- //nfsm_reqhead(NFSX_FH(v3));
- hsiz = NFSX_FH(v3);
- mb = NULL;
- if (hsiz >= nfs_mbuf_minclsize)
- error = mbuf_mclget(MBUF_WAITOK, MBUF_TYPE_DATA, &mb);
- else
- error = mbuf_get(MBUF_WAITOK, MBUF_TYPE_DATA, &mb);
- if (error)
- return (error);
- bpos = mbuf_data(mb);
- mreq = mb;
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_GETATTR]);
- //nfsm_fhtom(vp, v3);
- if (v3) {
- t2 = nfsm_rndup(fhsize) + NFSX_UNSIGNED;
- if (t2 <= mbuf_trailingspace(mb)) {
- nfsm_build(tl, u_long *, t2);
- *tl++ = txdr_unsigned(fhsize);
- *(tl + ((t2>>2) - 2)) = 0;
- bcopy((caddr_t)fhp,(caddr_t)tl, fhsize);
- } else if ((t2 = nfsm_strtmbuf(&mb, &bpos, (caddr_t)fhp, fhsize))) {
- error = t2;
- mbuf_freem(mreq);
- goto nfsmout;
+ /*
+ * If we're closing a write-only open, we may not have a write-only count
+ * if we also grabbed read access. So, check the read-write count.
+ */
+ if (denyMode == NFS_OPEN_SHARE_DENY_NONE) {
+ if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ (nofp->nof_w == 0) && (nofp->nof_d_w == 0) &&
+ (nofp->nof_rw || nofp->nof_d_rw))
+ accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
+ } else if (denyMode == NFS_OPEN_SHARE_DENY_WRITE) {
+ if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ (nofp->nof_w_dw == 0) && (nofp->nof_d_w_dw == 0) &&
+ (nofp->nof_rw_dw || nofp->nof_d_rw_dw))
+ accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
+ } else { /* NFS_OPEN_SHARE_DENY_BOTH */
+ if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ (nofp->nof_w_drw == 0) && (nofp->nof_d_w_drw == 0) &&
+ (nofp->nof_rw_drw || nofp->nof_d_rw_drw))
+ accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
+ }
+
+ nfs_open_file_remove_open_find(nofp, accessMode, denyMode, &newAccessMode, &newDenyMode, &delegated);
+ if ((newAccessMode != nofp->nof_access) || (newDenyMode != nofp->nof_deny))
+ changed = 1;
+ else
+ changed = 0;
+
+ if (NFSTONMP(np)->nm_vers < NFS_VER4) /* NFS v2/v3 closes simply need to remove the open. */
+ goto v3close;
+
+ if ((newAccessMode == 0) || (nofp->nof_opencnt == 1)) {
+ /*
+ * No more access after this close, so clean up and close it.
+ * Don't send a close RPC if we're closing a delegated open.
+ */
+ nfs_wait_bufs(np);
+ closed = 1;
+ if (!delegated && !(nofp->nof_flags & NFS_OPEN_FILE_LOST))
+ error = nfs4_close_rpc(np, nofp, vfs_context_thread(ctx), vfs_context_ucred(ctx), 0);
+ if (error == NFSERR_LOCKS_HELD) {
+ /*
+ * Hmm... the server says we have locks we need to release first
+ * Find the lock owner and try to unlock everything.
+ */
+ nlop = nfs_lock_owner_find(np, vfs_context_proc(ctx), 0);
+ if (nlop) {
+ nfs4_unlock_rpc(np, nlop, F_WRLCK, 0, UINT64_MAX,
+ 0, vfs_context_thread(ctx), vfs_context_ucred(ctx));
+ nfs_lock_owner_rele(nlop);
}
- } else {
- nfsm_build(cp, caddr_t, NFSX_V2FH);
- bcopy((caddr_t)fhp, cp, NFSX_V2FH);
- }
- //nfsm_request(vp, NFSPROC_GETATTR, p, cred, xidp);
- if ((error = nfs_request(NULL, mp, mreq, NFSPROC_GETATTR, p, cred, &mrep, &md, &dpos, xidp))) {
- if (error & NFSERR_RETERR)
- error &= ~NFSERR_RETERR;
- else
- goto nfsmout;
+ error = nfs4_close_rpc(np, nofp, vfs_context_thread(ctx), vfs_context_ucred(ctx), 0);
}
- if (!error) {
- //nfsm_loadattr(vp, nvap, xidp);
- error = nfs_parsefattr(&md, &dpos, v3, nvap);
- if (error) {
- mbuf_freem(mrep);
- goto nfsmout;
+ } else if (changed) {
+ /*
+ * File is still open but with less access, so downgrade the open.
+ * Don't send a downgrade RPC if we're closing a delegated open.
+ */
+ if (!delegated && !(nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
+ downgrade = 1;
+ /*
+ * If we have delegated opens, we should probably claim them before sending
+ * the downgrade because the server may not know the open we are downgrading to.
+ */
+ if (nofp->nof_d_rw_drw || nofp->nof_d_w_drw || nofp->nof_d_r_drw ||
+ nofp->nof_d_rw_dw || nofp->nof_d_w_dw || nofp->nof_d_r_dw ||
+ nofp->nof_d_rw || nofp->nof_d_w || nofp->nof_d_r)
+ nfs4_claim_delegated_state_for_open_file(nofp, 0);
+ /* need to remove the open before sending the downgrade */
+ nfs_open_file_remove_open(nofp, accessMode, denyMode);
+ error = nfs4_open_downgrade_rpc(np, nofp, ctx);
+ if (error) /* Hmm.. that didn't work. Add the open back in. */
+ nfs_open_file_add_open(nofp, accessMode, denyMode, delegated);
}
}
- nfsm_reqdone;
+
+ if (error) {
+ NP(np, "nfs_close: error %d, %d", error, kauth_cred_getuid(nofp->nof_owner->noo_cred));
+ return (error);
+ }
+
+v3close:
+ if (!downgrade)
+ nfs_open_file_remove_open(nofp, accessMode, denyMode);
+
+ if (closed) {
+ lck_mtx_lock(&nofp->nof_lock);
+ if (nofp->nof_r || nofp->nof_d_r || nofp->nof_w || nofp->nof_d_w || nofp->nof_d_rw ||
+ (nofp->nof_rw && !((nofp->nof_flags & NFS_OPEN_FILE_CREATE) && !nofp->nof_creator && (nofp->nof_rw == 1))) ||
+ nofp->nof_r_dw || nofp->nof_d_r_dw || nofp->nof_w_dw || nofp->nof_d_w_dw ||
+ nofp->nof_rw_dw || nofp->nof_d_rw_dw || nofp->nof_r_drw || nofp->nof_d_r_drw ||
+ nofp->nof_w_drw || nofp->nof_d_w_drw || nofp->nof_rw_drw || nofp->nof_d_rw_drw)
+ NP(np, "nfs_close: unexpected count: %u.%u %u.%u %u.%u dw %u.%u %u.%u %u.%u drw %u.%u %u.%u %u.%u flags 0x%x, %d",
+ nofp->nof_r, nofp->nof_d_r, nofp->nof_w, nofp->nof_d_w,
+ nofp->nof_rw, nofp->nof_d_rw, nofp->nof_r_dw, nofp->nof_d_r_dw,
+ nofp->nof_w_dw, nofp->nof_d_w_dw, nofp->nof_rw_dw, nofp->nof_d_rw_dw,
+ nofp->nof_r_drw, nofp->nof_d_r_drw, nofp->nof_w_drw, nofp->nof_d_w_drw,
+ nofp->nof_rw_drw, nofp->nof_d_rw_drw, nofp->nof_flags,
+ kauth_cred_getuid(nofp->nof_owner->noo_cred));
+ /* clear out all open info, just to be safe */
+ nofp->nof_access = nofp->nof_deny = 0;
+ nofp->nof_mmap_access = nofp->nof_mmap_deny = 0;
+ nofp->nof_r = nofp->nof_d_r = 0;
+ nofp->nof_w = nofp->nof_d_w = 0;
+ nofp->nof_rw = nofp->nof_d_rw = 0;
+ nofp->nof_r_dw = nofp->nof_d_r_dw = 0;
+ nofp->nof_w_dw = nofp->nof_d_w_dw = 0;
+ nofp->nof_rw_dw = nofp->nof_d_rw_dw = 0;
+ nofp->nof_r_drw = nofp->nof_d_r_drw = 0;
+ nofp->nof_w_drw = nofp->nof_d_w_drw = 0;
+ nofp->nof_rw_drw = nofp->nof_d_rw_drw = 0;
+ nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
+ lck_mtx_unlock(&nofp->nof_lock);
+ /* XXX we may potentially want to clean up idle/unused open file structures */
+ }
+ if (nofp->nof_flags & NFS_OPEN_FILE_LOST) {
+ error = EIO;
+ NP(np, "nfs_close: LOST%s, %d", !nofp->nof_opencnt ? " (last)" : "",
+ kauth_cred_getuid(nofp->nof_owner->noo_cred));
+ }
+
return (error);
}
-/*
- * nfs getattr call from vfs.
- */
+
+
+
int
-nfs_getattr(
- vnode_t vp,
+nfs3_getattr_rpc(
+ nfsnode_t np,
+ mount_t mp,
+ u_char *fhp,
+ size_t fhsize,
+ int flags,
+ vfs_context_t ctx,
struct nfs_vattr *nvap,
- kauth_cred_t cred,
- proc_t p)
+ u_int64_t *xidp)
{
- struct nfsnode *np = VTONFS(vp);
- caddr_t cp;
- u_long *tl;
- int t1, t2;
- caddr_t bpos, dpos;
- int error = 0;
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3;
+ struct nfsmount *nmp = mp ? VFSTONFS(mp) : NFSTONMP(np);
+ int error = 0, status, nfsvers, rpcflags = 0;
+ struct nfsm_chain nmreq, nmrep;
+
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+
+ if (flags & NGA_MONITOR) /* vnode monitor requests should be soft */
+ rpcflags = R_RECOVER;
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(nfsvers));
+ if (nfsvers != NFS_VER2)
+ nfsm_chain_add_32(error, &nmreq, fhsize);
+ nfsm_chain_add_opaque(error, &nmreq, fhp, fhsize);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request2(np, mp, &nmreq, NFSPROC_GETATTR,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx),
+ NULL, rpcflags, &nmrep, xidp, &status);
+ if (!error)
+ error = status;
+ nfsmout_if(error);
+ error = nfs_parsefattr(&nmrep, nfsvers, nvap);
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
+
+
+int
+nfs_getattr(nfsnode_t np, struct nfs_vattr *nvap, vfs_context_t ctx, int flags)
+{
+ struct nfsmount *nmp;
+ int error = 0, nfsvers, inprogset = 0, wanted = 0, avoidfloods;
+ struct nfs_vattr nvattr;
+ struct timespec ts = { 2, 0 };
u_int64_t xid;
- int avoidfloods;
FSDBG_TOP(513, np->n_size, np, np->n_vattr.nva_size, np->n_flag);
- /*
- * Update local times for special files.
- */
- if (np->n_flag & (NACC | NUPD))
+ if (!(nmp = NFSTONMP(np)))
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+
+ if (!nvap)
+ nvap = &nvattr;
+ NVATTR_INIT(nvap);
+
+ /* Update local times for special files. */
+ if (np->n_flag & (NACC | NUPD)) {
+ nfs_node_lock_force(np);
np->n_flag |= NCHG;
- /*
- * First look in the cache.
- */
- if ((error = nfs_getattrcache(vp, nvap)) == 0) {
- FSDBG_BOT(513, np->n_size, 0, np->n_vattr.nva_size, np->n_flag);
- return (0);
- }
- if (error != ENOENT) {
- FSDBG_BOT(513, np->n_size, error, np->n_vattr.nva_size,
- np->n_flag);
- return (error);
+ nfs_node_unlock(np);
}
+ /* Update size, if necessary */
+ if (ISSET(np->n_flag, NUPDATESIZE))
+ nfs_data_update_size(np, 0);
- if (!VFSTONFS(vnode_mount(vp))) {
- FSDBG_BOT(513, np->n_size, ENXIO, np->n_vattr.nva_size, np->n_flag);
- return (ENXIO);
+ error = nfs_node_lock(np);
+ nfsmout_if(error);
+ if (!(flags & (NGA_UNCACHED|NGA_MONITOR)) || ((nfsvers >= NFS_VER4) && (np->n_openflags & N_DELEG_MASK))) {
+ /*
+ * Use the cache or wait for any getattr in progress if:
+ * - it's a cached request, or
+ * - we have a delegation
+ */
+ while (1) {
+ error = nfs_getattrcache(np, nvap, flags);
+ if (!error || (error != ENOENT)) {
+ nfs_node_unlock(np);
+ goto nfsmout;
+ }
+ error = 0;
+ if (!ISSET(np->n_flag, NGETATTRINPROG))
+ break;
+ if (flags & NGA_MONITOR) {
+ /* no need to wait if a request is pending */
+ error = EINPROGRESS;
+ nfs_node_unlock(np);
+ goto nfsmout;
+ }
+ SET(np->n_flag, NGETATTRWANT);
+ msleep(np, &np->n_lock, PZERO-1, "nfsgetattrwant", &ts);
+ if ((error = nfs_sigintr(NFSTONMP(np), NULL, vfs_context_thread(ctx), 0))) {
+ nfs_node_unlock(np);
+ goto nfsmout;
+ }
+ }
+ SET(np->n_flag, NGETATTRINPROG);
+ inprogset = 1;
+ } else if (!ISSET(np->n_flag, NGETATTRINPROG)) {
+ SET(np->n_flag, NGETATTRINPROG);
+ inprogset = 1;
+ } else if (flags & NGA_MONITOR) {
+ /* no need to make a request if one is pending */
+ error = EINPROGRESS;
}
- v3 = NFS_ISV3(vp);
- error = 0;
+ nfs_node_unlock(np);
+
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ error = ENXIO;
+ if (error)
+ goto nfsmout;
/*
- * Try to get both the attributes and access info by making an
- * ACCESS call and seeing if it returns updated attributes.
+ * We might want to try to get both the attributes and access info by
+ * making an ACCESS call and seeing if it returns updated attributes.
* But don't bother if we aren't caching access info or if the
* attributes returned wouldn't be cached.
*/
- if (v3 && (nfsaccess_cache_timeout > 0) &&
- (nfs_attrcachetimeout(vp) > 0)) {
- /* OSAddAtomic(1, (SInt32*)&nfsstats.accesscache_misses); */
- if ((error = nfs3_access_otw(vp, NFSV3ACCESS_ALL, p, cred)))
- return (error);
- if ((error = nfs_getattrcache(vp, nvap)) == 0)
- return (0);
- if (error != ENOENT)
- return (error);
- error = 0;
+ if (!(flags & NGA_ACL) && (nfsvers != NFS_VER2) && nfs_access_for_getattr && (nfs_access_cache_timeout > 0)) {
+ if (nfs_attrcachetimeout(np) > 0) {
+ /* OSAddAtomic(1, &nfsstats.accesscache_misses); */
+ u_int32_t access = NFS_ACCESS_ALL;
+ error = nmp->nm_funcs->nf_access_rpc(np, &access, ctx);
+ if (error)
+ goto nfsmout;
+ nfs_node_lock_force(np);
+ error = nfs_getattrcache(np, nvap, flags);
+ nfs_node_unlock(np);
+ if (!error || (error != ENOENT))
+ goto nfsmout;
+ /* Well, that didn't work... just do a getattr... */
+ error = 0;
+ }
}
+
avoidfloods = 0;
tryagain:
- nfsm_reqhead(NFSX_FH(v3));
- if (error) {
- FSDBG_BOT(513, np->n_size, error, np->n_vattr.nva_size, np->n_flag);
- return (error);
+ error = nmp->nm_funcs->nf_getattr_rpc(np, NULL, np->n_fhp, np->n_fhsize, flags, ctx, nvap, &xid);
+ if (!error) {
+ nfs_node_lock_force(np);
+ error = nfs_loadattrcache(np, nvap, &xid, 0);
+ nfs_node_unlock(np);
+ }
+ nfsmout_if(error);
+ if (!xid) { /* out-of-order rpc - attributes were dropped */
+ FSDBG(513, -1, np, np->n_xid >> 32, np->n_xid);
+ if (avoidfloods++ < 20)
+ goto tryagain;
+ /* avoidfloods>1 is bizarre. at 20 pull the plug */
+ /* just return the last attributes we got */
+ }
+nfsmout:
+ nfs_node_lock_force(np);
+ if (inprogset) {
+ wanted = ISSET(np->n_flag, NGETATTRWANT);
+ CLR(np->n_flag, (NGETATTRINPROG | NGETATTRWANT));
}
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_GETATTR]);
- nfsm_fhtom(vp, v3);
- nfsm_request(vp, NFSPROC_GETATTR, p, cred, &xid);
if (!error) {
- nfsm_loadattr(vp, v3, nvap, &xid);
- if (!xid) { /* out-of-order rpc - attributes were dropped */
- mbuf_freem(mrep);
- mrep = NULL;
- FSDBG(513, -1, np, np->n_xid << 32, np->n_xid);
- if (avoidfloods++ < 100)
- goto tryagain;
- /*
- * avoidfloods>1 is bizarre. at 100 pull the plug
- */
- panic("nfs_getattr: getattr flood\n");
+ /* check if the node changed on us */
+ vnode_t vp = NFSTOV(np);
+ enum vtype vtype = vnode_vtype(vp);
+ if ((vtype == VDIR) && NFS_CHANGED_NC(nfsvers, np, nvap)) {
+ FSDBG(513, -1, np, 0, np);
+ np->n_flag &= ~NNEGNCENTRIES;
+ cache_purge(vp);
+ np->n_ncgen++;
+ NFS_CHANGED_UPDATE_NC(nfsvers, np, nvap);
+ NFS_VNOP_DBG("Purge directory 0x%llx\n",
+ (uint64_t)VM_KERNEL_ADDRPERM(vp));
}
- if (nfstimespeccmp(&np->n_mtime, &nvap->nva_mtime, !=)) {
- enum vtype vtype = vnode_vtype(vp);
- FSDBG(513, -1, np, -1, vp);
+ if (NFS_CHANGED(nfsvers, np, nvap)) {
+ FSDBG(513, -1, np, -1, np);
if (vtype == VDIR) {
- nfs_invaldir(vp);
- /* purge name cache entries */
- if (nfstimespeccmp(&np->n_ncmtime, &nvap->nva_mtime, !=))
- cache_purge(vp);
+ NFS_VNOP_DBG("Invalidate directory 0x%llx\n",
+ (uint64_t)VM_KERNEL_ADDRPERM(vp));
+ nfs_invaldir(np);
}
- error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1);
+ nfs_node_unlock(np);
+ if (wanted)
+ wakeup(np);
+ error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
FSDBG(513, -1, np, -2, error);
if (!error) {
- if (vtype == VDIR)
- np->n_ncmtime = nvap->nva_mtime;
- np->n_mtime = nvap->nva_mtime;
+ nfs_node_lock_force(np);
+ NFS_CHANGED_UPDATE(nfsvers, np, nvap);
+ nfs_node_unlock(np);
}
+ } else {
+ nfs_node_unlock(np);
+ if (wanted)
+ wakeup(np);
}
+ } else {
+ nfs_node_unlock(np);
+ if (wanted)
+ wakeup(np);
}
- nfsm_reqdone;
- FSDBG_BOT(513, np->n_size, -1, np->n_vattr.nva_size, error);
+ if (nvap == &nvattr) {
+ NVATTR_CLEANUP(nvap);
+ } else if (!(flags & NGA_ACL)) {
+ /* make sure we don't return an ACL if it wasn't asked for */
+ NFS_BITMAP_CLR(nvap->nva_bitmap, NFS_FATTR_ACL);
+ if (nvap->nva_acl) {
+ kauth_acl_free(nvap->nva_acl);
+ nvap->nva_acl = NULL;
+ }
+ }
+ FSDBG_BOT(513, np->n_size, error, np->n_vattr.nva_size, np->n_flag);
return (error);
}
+/*
+ * NFS getattr call from vfs.
+ */
+
+/*
+ * The attributes we support over the wire.
+ * We also get fsid but the vfs layer gets it out of the mount
+ * structure after this calling us so there's no need to return it,
+ * and Finder expects to call getattrlist just looking for the FSID
+ * with out hanging on a non responsive server.
+ */
+#define NFS3_SUPPORTED_VATTRS \
+ (VNODE_ATTR_va_rdev | \
+ VNODE_ATTR_va_nlink | \
+ VNODE_ATTR_va_data_size | \
+ VNODE_ATTR_va_data_alloc | \
+ VNODE_ATTR_va_uid | \
+ VNODE_ATTR_va_gid | \
+ VNODE_ATTR_va_mode | \
+ VNODE_ATTR_va_modify_time | \
+ VNODE_ATTR_va_change_time | \
+ VNODE_ATTR_va_access_time | \
+ VNODE_ATTR_va_fileid | \
+ VNODE_ATTR_va_type)
-static int
-nfs_vnop_getattr(
+int
+nfs3_vnop_getattr(
struct vnop_getattr_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
int error;
struct nfs_vattr nva;
struct vnode_attr *vap = ap->a_vap;
+ dev_t rdev;
+
+ /*
+ * Lets don't go over the wire if we don't support any of the attributes.
+ * Just fall through at the VFS layer and let it cons up what it needs.
+ */
+ /* Return the io size no matter what, since we don't go over the wire for this */
+ VATTR_RETURN(vap, va_iosize, nfs_iosize);
+ if ((vap->va_active & NFS3_SUPPORTED_VATTRS) == 0)
+ return (0);
- error = nfs_getattr(ap->a_vp, &nva,
- vfs_context_ucred(ap->a_context),
- vfs_context_proc(ap->a_context));
+ if (VATTR_IS_ACTIVE(ap->a_vap, va_name))
+ NFS_VNOP_DBG("Getting attrs for 0x%llx, vname is %s\n",
+ (uint64_t)VM_KERNEL_ADDRPERM(ap->a_vp),
+ ap->a_vp->v_name ? ap->a_vp->v_name : "empty");
+ error = nfs_getattr(VTONFS(ap->a_vp), &nva, ap->a_context, NGA_CACHED);
if (error)
return (error);
/* copy nva to *a_vap */
- VATTR_RETURN(vap, va_type, nva.nva_type);
- VATTR_RETURN(vap, va_mode, nva.nva_mode);
- VATTR_RETURN(vap, va_rdev, nva.nva_rdev);
- VATTR_RETURN(vap, va_uid, nva.nva_uid);
- VATTR_RETURN(vap, va_gid, nva.nva_gid);
- VATTR_RETURN(vap, va_nlink, nva.nva_nlink);
- VATTR_RETURN(vap, va_fileid, nva.nva_fileid);
- VATTR_RETURN(vap, va_data_size, nva.nva_size);
- VATTR_RETURN(vap, va_data_alloc, nva.nva_bytes);
- VATTR_RETURN(vap, va_iosize, nva.nva_blocksize); /* should this just be f_iosize? */
- VATTR_RETURN(vap, va_fsid, nva.nva_fsid);
- vap->va_access_time.tv_sec = nva.nva_atime.tv_sec;
- vap->va_access_time.tv_nsec = nva.nva_atime.tv_nsec;
- VATTR_SET_SUPPORTED(vap, va_access_time);
- vap->va_modify_time.tv_sec = nva.nva_mtime.tv_sec;
- vap->va_modify_time.tv_nsec = nva.nva_mtime.tv_nsec;
- VATTR_SET_SUPPORTED(vap, va_modify_time);
- vap->va_change_time.tv_sec = nva.nva_ctime.tv_sec;
- vap->va_change_time.tv_nsec = nva.nva_ctime.tv_nsec;
- VATTR_SET_SUPPORTED(vap, va_change_time);
+ VATTR_RETURN(vap, va_type, nva.nva_type);
+ VATTR_RETURN(vap, va_mode, nva.nva_mode);
+ rdev = makedev(nva.nva_rawdev.specdata1, nva.nva_rawdev.specdata2);
+ VATTR_RETURN(vap, va_rdev, rdev);
+ VATTR_RETURN(vap, va_uid, nva.nva_uid);
+ VATTR_RETURN(vap, va_gid, nva.nva_gid);
+ VATTR_RETURN(vap, va_nlink, nva.nva_nlink);
+ VATTR_RETURN(vap, va_fileid, nva.nva_fileid);
+ VATTR_RETURN(vap, va_data_size, nva.nva_size);
+ VATTR_RETURN(vap, va_data_alloc, nva.nva_bytes);
+ vap->va_access_time.tv_sec = nva.nva_timesec[NFSTIME_ACCESS];
+ vap->va_access_time.tv_nsec = nva.nva_timensec[NFSTIME_ACCESS];
+ VATTR_SET_SUPPORTED(vap, va_access_time);
+ vap->va_modify_time.tv_sec = nva.nva_timesec[NFSTIME_MODIFY];
+ vap->va_modify_time.tv_nsec = nva.nva_timensec[NFSTIME_MODIFY];
+ VATTR_SET_SUPPORTED(vap, va_modify_time);
+ vap->va_change_time.tv_sec = nva.nva_timesec[NFSTIME_CHANGE];
+ vap->va_change_time.tv_nsec = nva.nva_timensec[NFSTIME_CHANGE];
+ VATTR_SET_SUPPORTED(vap, va_change_time);
+ // VATTR_RETURN(vap, va_encoding, 0xffff /* kTextEncodingUnknown */);
return (error);
}
/*
- * nfs setattr call.
+ * NFS setattr call.
*/
-static int
-nfs_setattr(ap)
+int
+nfs_vnop_setattr(
struct vnop_setattr_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
struct vnode_attr *a_vap;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t vp = ap->a_vp;
- struct nfsnode *np = VTONFS(vp);
+ nfsnode_t np = VTONFS(vp);
+ struct nfsmount *nmp;
struct vnode_attr *vap = ap->a_vap;
int error = 0;
- u_quad_t tsize;
- kauth_cred_t cred;
- proc_t p;
+ int biosize, nfsvers, namedattrs;
+ u_quad_t origsize, vapsize;
+ struct nfs_dulookup dul;
+ nfsnode_t dnp = NULL;
+ vnode_t dvp = NULL;
+ const char *vname = NULL;
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
+ biosize = nmp->nm_biosize;
-#ifndef nolint
- tsize = (u_quad_t)0;
-#endif
-
- /* Setting of flags is not supported. */
- if (VATTR_IS_ACTIVE(vap, va_flags))
- return (ENOTSUP);
-
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
-
- VATTR_SET_SUPPORTED(vap, va_mode);
- VATTR_SET_SUPPORTED(vap, va_uid);
- VATTR_SET_SUPPORTED(vap, va_gid);
- VATTR_SET_SUPPORTED(vap, va_data_size);
- VATTR_SET_SUPPORTED(vap, va_access_time);
- VATTR_SET_SUPPORTED(vap, va_modify_time);
-
- /* Disallow write attempts if the filesystem is mounted read-only. */
- if ((VATTR_IS_ACTIVE(vap, va_flags) || VATTR_IS_ACTIVE(vap, va_mode) ||
- VATTR_IS_ACTIVE(vap, va_uid) || VATTR_IS_ACTIVE(vap, va_gid) ||
- VATTR_IS_ACTIVE(vap, va_access_time) ||
- VATTR_IS_ACTIVE(vap, va_modify_time)) &&
- vnode_vfsisrdonly(vp))
- return (EROFS);
+ /* Disallow write attempts if the filesystem is mounted read-only. */
+ if (vnode_vfsisrdonly(vp))
+ return (EROFS);
+ origsize = np->n_size;
if (VATTR_IS_ACTIVE(vap, va_data_size)) {
- switch (vnode_vtype(vp)) {
- case VDIR:
- return (EISDIR);
- case VCHR:
- case VBLK:
- case VSOCK:
- case VFIFO:
+ switch (vnode_vtype(vp)) {
+ case VDIR:
+ return (EISDIR);
+ case VCHR:
+ case VBLK:
+ case VSOCK:
+ case VFIFO:
if (!VATTR_IS_ACTIVE(vap, va_modify_time) &&
!VATTR_IS_ACTIVE(vap, va_access_time) &&
!VATTR_IS_ACTIVE(vap, va_mode) &&
!VATTR_IS_ACTIVE(vap, va_uid) &&
- !VATTR_IS_ACTIVE(vap, va_gid))
+ !VATTR_IS_ACTIVE(vap, va_gid)) {
return (0);
+ }
VATTR_CLEAR_ACTIVE(vap, va_data_size);
- break;
- default:
+ break;
+ default:
/*
* Disallow write attempts if the filesystem is
* mounted read-only.
return (EROFS);
FSDBG_TOP(512, np->n_size, vap->va_data_size,
np->n_vattr.nva_size, np->n_flag);
- if (np->n_flag & NMODIFIED) {
- if (vap->va_data_size == 0)
- error = nfs_vinvalbuf(vp, 0, cred, p, 1);
- else
- error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1);
- if (error) {
- printf("nfs_setattr: nfs_vinvalbuf %d\n", error);
- FSDBG_BOT(512, np->n_size, vap->va_data_size,
- np->n_vattr.nva_size, -1);
- return (error);
+ /* clear NNEEDINVALIDATE, if set */
+ if ((error = nfs_node_lock(np)))
+ return (error);
+ if (np->n_flag & NNEEDINVALIDATE)
+ np->n_flag &= ~NNEEDINVALIDATE;
+ nfs_node_unlock(np);
+ /* flush everything */
+ error = nfs_vinvalbuf(vp, (vap->va_data_size ? V_SAVE : 0) , ctx, 1);
+ if (error) {
+ NP(np, "nfs_setattr: nfs_vinvalbuf %d", error);
+ FSDBG_BOT(512, np->n_size, vap->va_data_size, np->n_vattr.nva_size, -1);
+ return (error);
+ }
+ if (nfsvers >= NFS_VER4) {
+ /* setting file size requires having the file open for write access */
+ if (np->n_flag & NREVOKE)
+ return (EIO);
+ noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 1);
+ if (!noop)
+ return (ENOMEM);
+restart:
+ error = nfs_mount_state_in_use_start(nmp, vfs_context_thread(ctx));
+ if (error)
+ return (error);
+ if (np->n_flag & NREVOKE) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ return (EIO);
+ }
+ error = nfs_open_file_find(np, noop, &nofp, 0, 0, 1);
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_LOST))
+ error = EIO;
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ error = nfs4_reopen(nofp, vfs_context_thread(ctx));
+ nofp = NULL;
+ if (!error)
+ goto restart;
+ }
+ if (!error)
+ error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
+ if (error) {
+ nfs_open_owner_rele(noop);
+ return (error);
+ }
+ if (!(nofp->nof_access & NFS_OPEN_SHARE_ACCESS_WRITE)) {
+ /* we don't have the file open for write access, so open it */
+ error = nfs4_open(np, nofp, NFS_OPEN_SHARE_ACCESS_WRITE, NFS_OPEN_SHARE_DENY_NONE, ctx);
+ if (!error)
+ nofp->nof_flags |= NFS_OPEN_FILE_SETATTR;
+ if (nfs_mount_state_error_should_restart(error)) {
+ nfs_open_file_clear_busy(nofp);
+ nofp = NULL;
+ if (nfs_mount_state_in_use_end(nmp, error))
+ goto restart;
+ }
}
- } else if (np->n_size > vap->va_data_size) { /* shrinking? */
+ }
+ nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
+ if (np->n_size > vap->va_data_size) { /* shrinking? */
daddr64_t obn, bn;
- int biosize, neweofoff, mustwrite;
+ int neweofoff, mustwrite;
struct nfsbuf *bp;
- biosize = vfs_statfs(vnode_mount(vp))->f_iosize;
obn = (np->n_size - 1) / biosize;
- bn = vap->va_data_size / biosize;
+ bn = vap->va_data_size / biosize;
for ( ; obn >= bn; obn--) {
- if (!nfs_buf_is_incore(vp, obn))
+ if (!nfs_buf_is_incore(np, obn))
continue;
- error = nfs_buf_get(vp, obn, biosize, 0, NBLK_READ, &bp);
+ error = nfs_buf_get(np, obn, biosize, NULL, NBLK_READ, &bp);
if (error)
continue;
if (obn != bn) {
mustwrite = 0;
neweofoff = vap->va_data_size - NBOFF(bp);
/* check for any dirty data before the new EOF */
- if (bp->nb_dirtyend && bp->nb_dirtyoff < neweofoff) {
+ if ((bp->nb_dirtyend > 0) && (bp->nb_dirtyoff < neweofoff)) {
/* clip dirty range to EOF */
- if (bp->nb_dirtyend > neweofoff)
+ if (bp->nb_dirtyend > neweofoff) {
bp->nb_dirtyend = neweofoff;
- mustwrite++;
+ if (bp->nb_dirtyoff >= bp->nb_dirtyend)
+ bp->nb_dirtyoff = bp->nb_dirtyend = 0;
+ }
+ if ((bp->nb_dirtyend > 0) && (bp->nb_dirtyoff < neweofoff))
+ mustwrite++;
}
bp->nb_dirty &= (1 << round_page_32(neweofoff)/PAGE_SIZE) - 1;
if (bp->nb_dirty)
/* (NB_NOCACHE indicates buffer should be discarded) */
CLR(bp->nb_flags, (NB_DONE | NB_ERROR | NB_INVAL | NB_ASYNC | NB_READ));
SET(bp->nb_flags, NB_STABLE | NB_NOCACHE);
- if (bp->nb_wcred == NOCRED) {
+ if (!IS_VALID_CRED(bp->nb_wcred)) {
+ kauth_cred_t cred = vfs_context_ucred(ctx);
kauth_cred_ref(cred);
bp->nb_wcred = cred;
}
// Note: bp has been released
if (error) {
FSDBG(512, bp, 0xd00dee, 0xbad, error);
+ nfs_node_lock_force(np);
np->n_error = error;
np->n_flag |= NWRITEERR;
/*
* we may no longer know the correct size)
*/
NATTRINVALIDATE(np);
- nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, cred, p, 1);
+ nfs_node_unlock(np);
+ nfs_data_unlock(np);
+ nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1);
+ nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
error = 0;
}
}
}
- tsize = np->n_size;
+ if (vap->va_data_size != np->n_size)
+ ubc_setsize(vp, (off_t)vap->va_data_size); /* XXX error? */
+ origsize = np->n_size;
np->n_size = np->n_vattr.nva_size = vap->va_data_size;
- ubc_setsize(vp, (off_t)vap->va_data_size); /* XXX error? */
- }
- } else if ((VATTR_IS_ACTIVE(vap, va_modify_time) ||
- VATTR_IS_ACTIVE(vap, va_access_time)) &&
- (np->n_flag & NMODIFIED) && (vnode_vtype(vp) == VREG)) {
- error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1);
- if (error == EINTR)
+ nfs_node_lock_force(np);
+ CLR(np->n_flag, NUPDATESIZE);
+ nfs_node_unlock(np);
+ FSDBG(512, np, np->n_size, np->n_vattr.nva_size, 0xf00d0001);
+ }
+ } else if (VATTR_IS_ACTIVE(vap, va_modify_time) ||
+ VATTR_IS_ACTIVE(vap, va_access_time) ||
+ (vap->va_vaflags & VA_UTIMES_NULL)) {
+ if ((error = nfs_node_lock(np)))
return (error);
+ if ((np->n_flag & NMODIFIED) && (vnode_vtype(vp) == VREG)) {
+ nfs_node_unlock(np);
+ error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
+ if (error == EINTR)
+ return (error);
+ } else {
+ nfs_node_unlock(np);
+ }
+ }
+ if ((VATTR_IS_ACTIVE(vap, va_mode) || VATTR_IS_ACTIVE(vap, va_uid) || VATTR_IS_ACTIVE(vap, va_gid) ||
+ VATTR_IS_ACTIVE(vap, va_acl) || VATTR_IS_ACTIVE(vap, va_uuuid) || VATTR_IS_ACTIVE(vap, va_guuid)) &&
+ !(error = nfs_node_lock(np))) {
+ NACCESSINVALIDATE(np);
+ nfs_node_unlock(np);
+ if (!namedattrs) {
+ dvp = vnode_getparent(vp);
+ vname = vnode_getname(vp);
+ dnp = (dvp && vname) ? VTONFS(dvp) : NULL;
+ if (dnp) {
+ error = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
+ if (error) {
+ dnp = NULL;
+ error = 0;
+ }
+ }
+ if (dnp) {
+ nfs_dulookup_init(&dul, dnp, vname, strlen(vname), ctx);
+ nfs_dulookup_start(&dul, dnp, ctx);
+ }
+ }
}
- if (VATTR_IS_ACTIVE(vap, va_mode)) {
- NMODEINVALIDATE(np);
+
+ if (!error)
+ error = nmp->nm_funcs->nf_setattr_rpc(np, vap, ctx);
+
+ if (VATTR_IS_ACTIVE(vap, va_mode) || VATTR_IS_ACTIVE(vap, va_uid) || VATTR_IS_ACTIVE(vap, va_gid) ||
+ VATTR_IS_ACTIVE(vap, va_acl) || VATTR_IS_ACTIVE(vap, va_uuuid) || VATTR_IS_ACTIVE(vap, va_guuid)) {
+ if (!namedattrs) {
+ if (dnp) {
+ nfs_dulookup_finish(&dul, dnp, ctx);
+ nfs_node_clear_busy(dnp);
+ }
+ if (dvp != NULLVP)
+ vnode_put(dvp);
+ if (vname != NULL)
+ vnode_putname(vname);
+ }
}
- error = nfs_setattrrpc(vp, vap, cred, p);
- FSDBG_BOT(512, np->n_size, vap->va_data_size, np->n_vattr.nva_size, error);
- if (error && VATTR_IS_ACTIVE(vap, va_data_size)) {
- /* make every effort to resync file size w/ server... */
- int err; /* preserve "error" for return */
- np->n_size = np->n_vattr.nva_size = tsize;
- ubc_setsize(vp, (off_t)np->n_size); /* XXX check error */
- vap->va_data_size = tsize;
- err = nfs_setattrrpc(vp, vap, cred, p);
- printf("nfs_setattr: nfs_setattrrpc %d %d\n", error, err);
+ FSDBG_BOT(512, np->n_size, vap->va_data_size, np->n_vattr.nva_size, error);
+ if (VATTR_IS_ACTIVE(vap, va_data_size)) {
+ if (error && (origsize != np->n_size) &&
+ ((nfsvers < NFS_VER4) || !nfs_mount_state_error_should_restart(error))) {
+ /* make every effort to resync file size w/ server... */
+ /* (don't bother if we'll be restarting the operation) */
+ int err; /* preserve "error" for return */
+ np->n_size = np->n_vattr.nva_size = origsize;
+ nfs_node_lock_force(np);
+ CLR(np->n_flag, NUPDATESIZE);
+ nfs_node_unlock(np);
+ FSDBG(512, np, np->n_size, np->n_vattr.nva_size, 0xf00d0002);
+ ubc_setsize(vp, (off_t)np->n_size); /* XXX check error */
+ vapsize = vap->va_data_size;
+ vap->va_data_size = origsize;
+ err = nmp->nm_funcs->nf_setattr_rpc(np, vap, ctx);
+ if (err)
+ NP(np, "nfs_vnop_setattr: nfs%d_setattr_rpc %d %d", nfsvers, error, err);
+ vap->va_data_size = vapsize;
+ }
+ nfs_node_lock_force(np);
+ /*
+ * The size was just set. If the size is already marked for update, don't
+ * trust the newsize (it may have been set while the setattr was in progress).
+ * Clear the update flag and make sure we fetch new attributes so we are sure
+ * we have the latest size.
+ */
+ if (ISSET(np->n_flag, NUPDATESIZE)) {
+ CLR(np->n_flag, NUPDATESIZE);
+ NATTRINVALIDATE(np);
+ nfs_node_unlock(np);
+ nfs_getattr(np, NULL, ctx, NGA_UNCACHED);
+ } else {
+ nfs_node_unlock(np);
+ }
+ nfs_data_unlock(np);
+ if (nfsvers >= NFS_VER4) {
+ if (nofp) {
+ /* don't close our setattr open if we'll be restarting... */
+ if (!nfs_mount_state_error_should_restart(error) &&
+ (nofp->nof_flags & NFS_OPEN_FILE_SETATTR)) {
+ int err = nfs_close(np, nofp, NFS_OPEN_SHARE_ACCESS_WRITE, NFS_OPEN_SHARE_DENY_NONE, ctx);
+ if (err)
+ NP(np, "nfs_vnop_setattr: close error: %d", err);
+ nofp->nof_flags &= ~NFS_OPEN_FILE_SETATTR;
+ }
+ nfs_open_file_clear_busy(nofp);
+ nofp = NULL;
+ }
+ if (nfs_mount_state_in_use_end(nmp, error))
+ goto restart;
+ nfs_open_owner_rele(noop);
+ }
}
return (error);
}
/*
- * Do an nfs setattr rpc.
+ * Do an NFS setattr RPC.
*/
-static int
-nfs_setattrrpc(vp, vap, cred, procp)
- vnode_t vp;
- struct vnode_attr *vap;
- kauth_cred_t cred;
- proc_t procp;
+int
+nfs3_setattr_rpc(
+ nfsnode_t np,
+ struct vnode_attr *vap,
+ vfs_context_t ctx)
{
- register struct nfsv2_sattr *sp;
- register caddr_t cp;
- register long t1, t2;
- caddr_t bpos, dpos, cp2;
- u_long *tl;
- int error = 0, wccpostattr = 0;
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3;
- u_int64_t xid;
- struct timeval now;
+ struct nfsmount *nmp = NFSTONMP(np);
+ int error = 0, lockerror = ENOENT, status, wccpostattr = 0, nfsvers;
+ u_int64_t xid, nextxid;
+ struct nfsm_chain nmreq, nmrep;
- if (!VFSTONFS(vnode_mount(vp)))
+ if (!nmp)
return (ENXIO);
- v3 = NFS_ISV3(vp);
+ nfsvers = nmp->nm_vers;
- nfsm_reqhead(NFSX_FH(v3) + NFSX_SATTR(v3));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_SETATTR]);
- nfsm_fhtom(vp, v3);
- if (v3) {
+ VATTR_SET_SUPPORTED(vap, va_mode);
+ VATTR_SET_SUPPORTED(vap, va_uid);
+ VATTR_SET_SUPPORTED(vap, va_gid);
+ VATTR_SET_SUPPORTED(vap, va_data_size);
+ VATTR_SET_SUPPORTED(vap, va_access_time);
+ VATTR_SET_SUPPORTED(vap, va_modify_time);
+
+ if (VATTR_IS_ACTIVE(vap, va_flags)) {
+ if (vap->va_flags) { /* we don't support setting flags */
+ if (vap->va_active & ~VNODE_ATTR_va_flags)
+ return (EINVAL); /* return EINVAL if other attributes also set */
+ else
+ return (ENOTSUP); /* return ENOTSUP for chflags(2) */
+ }
+ /* no flags set, so we'll just ignore it */
+ if (!(vap->va_active & ~VNODE_ATTR_va_flags))
+ return (0); /* no (other) attributes to set, so nothing to do */
+ }
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + NFSX_SATTR(nfsvers));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
+ if (nfsvers == NFS_VER3) {
if (VATTR_IS_ACTIVE(vap, va_mode)) {
- nfsm_build(tl, u_long *, 2 * NFSX_UNSIGNED);
- *tl++ = nfs_true;
- *tl = txdr_unsigned(vap->va_mode);
+ nfsm_chain_add_32(error, &nmreq, TRUE);
+ nfsm_chain_add_32(error, &nmreq, vap->va_mode);
} else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = nfs_false;
+ nfsm_chain_add_32(error, &nmreq, FALSE);
}
if (VATTR_IS_ACTIVE(vap, va_uid)) {
- nfsm_build(tl, u_long *, 2 * NFSX_UNSIGNED);
- *tl++ = nfs_true;
- *tl = txdr_unsigned(vap->va_uid);
+ nfsm_chain_add_32(error, &nmreq, TRUE);
+ nfsm_chain_add_32(error, &nmreq, vap->va_uid);
} else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = nfs_false;
+ nfsm_chain_add_32(error, &nmreq, FALSE);
}
if (VATTR_IS_ACTIVE(vap, va_gid)) {
- nfsm_build(tl, u_long *, 2 * NFSX_UNSIGNED);
- *tl++ = nfs_true;
- *tl = txdr_unsigned(vap->va_gid);
+ nfsm_chain_add_32(error, &nmreq, TRUE);
+ nfsm_chain_add_32(error, &nmreq, vap->va_gid);
} else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = nfs_false;
+ nfsm_chain_add_32(error, &nmreq, FALSE);
}
if (VATTR_IS_ACTIVE(vap, va_data_size)) {
- nfsm_build(tl, u_long *, 3 * NFSX_UNSIGNED);
- *tl++ = nfs_true;
- txdr_hyper(&vap->va_data_size, tl);
+ nfsm_chain_add_32(error, &nmreq, TRUE);
+ nfsm_chain_add_64(error, &nmreq, vap->va_data_size);
} else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = nfs_false;
+ nfsm_chain_add_32(error, &nmreq, FALSE);
}
- microtime(&now);
- if (VATTR_IS_ACTIVE(vap, va_access_time)) {
- if (vap->va_access_time.tv_sec != now.tv_sec) {
- nfsm_build(tl, u_long *, 3 * NFSX_UNSIGNED);
- *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
- txdr_nfsv3time(&vap->va_access_time, tl);
+ if (vap->va_vaflags & VA_UTIMES_NULL) {
+ nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_SERVER);
+ nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_SERVER);
+ } else {
+ if (VATTR_IS_ACTIVE(vap, va_access_time)) {
+ nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_CLIENT);
+ nfsm_chain_add_32(error, &nmreq, vap->va_access_time.tv_sec);
+ nfsm_chain_add_32(error, &nmreq, vap->va_access_time.tv_nsec);
} else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
+ nfsm_chain_add_32(error, &nmreq, NFS_TIME_DONT_CHANGE);
}
- } else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
- }
- if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
- if (vap->va_modify_time.tv_sec != now.tv_sec) {
- nfsm_build(tl, u_long *, 3 * NFSX_UNSIGNED);
- *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
- txdr_nfsv3time(&vap->va_modify_time, tl);
+ if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
+ nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_CLIENT);
+ nfsm_chain_add_32(error, &nmreq, vap->va_modify_time.tv_sec);
+ nfsm_chain_add_32(error, &nmreq, vap->va_modify_time.tv_nsec);
} else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
+ nfsm_chain_add_32(error, &nmreq, NFS_TIME_DONT_CHANGE);
}
- } else {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
}
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl = nfs_false;
+ nfsm_chain_add_32(error, &nmreq, FALSE);
} else {
- struct timespec neg1time = { -1, -1 };
- nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
- if (VATTR_IS_ACTIVE(vap, va_mode))
- sp->sa_mode = vtonfsv2_mode(vnode_vtype(vp), vap->va_mode);
- else
- sp->sa_mode = nfs_xdrneg1;
- if (VATTR_IS_ACTIVE(vap, va_uid))
- sp->sa_uid = txdr_unsigned(vap->va_uid);
- else
- sp->sa_uid = nfs_xdrneg1;
- if (VATTR_IS_ACTIVE(vap, va_gid))
- sp->sa_gid = txdr_unsigned(vap->va_gid);
- else
- sp->sa_gid = nfs_xdrneg1;
- if (VATTR_IS_ACTIVE(vap, va_data_size))
- sp->sa_size = txdr_unsigned(vap->va_data_size);
- else
- sp->sa_size = nfs_xdrneg1;
+ nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_mode) ?
+ vtonfsv2_mode(vnode_vtype(NFSTOV(np)), vap->va_mode) : -1);
+ nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_uid) ?
+ vap->va_uid : (uint32_t)-1);
+ nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_gid) ?
+ vap->va_gid : (uint32_t)-1);
+ nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_data_size) ?
+ vap->va_data_size : (uint32_t)-1);
if (VATTR_IS_ACTIVE(vap, va_access_time)) {
- txdr_nfsv2time(&vap->va_access_time, &sp->sa_atime);
+ nfsm_chain_add_32(error, &nmreq, vap->va_access_time.tv_sec);
+ nfsm_chain_add_32(error, &nmreq, (vap->va_access_time.tv_nsec != -1) ?
+ ((uint32_t)vap->va_access_time.tv_nsec / 1000) : 0xffffffff);
} else {
- txdr_nfsv2time(&neg1time, &sp->sa_atime);
+ nfsm_chain_add_32(error, &nmreq, -1);
+ nfsm_chain_add_32(error, &nmreq, -1);
}
if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
- txdr_nfsv2time(&vap->va_modify_time, &sp->sa_mtime);
+ nfsm_chain_add_32(error, &nmreq, vap->va_modify_time.tv_sec);
+ nfsm_chain_add_32(error, &nmreq, (vap->va_modify_time.tv_nsec != -1) ?
+ ((uint32_t)vap->va_modify_time.tv_nsec / 1000) : 0xffffffff);
} else {
- txdr_nfsv2time(&neg1time, &sp->sa_mtime);
+ nfsm_chain_add_32(error, &nmreq, -1);
+ nfsm_chain_add_32(error, &nmreq, -1);
}
}
- nfsm_request(vp, NFSPROC_SETATTR, procp, cred, &xid);
- if (v3) {
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request(np, NULL, &nmreq, NFSPROC_SETATTR, ctx, NULL, &nmrep, &xid, &status);
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ if (nfsvers == NFS_VER3) {
struct timespec premtime = { 0, 0 };
- if (mrep) {
- nfsm_wcc_data(vp, &premtime, wccpostattr, &xid);
- }
+ nfsm_chain_get_wcc_data(error, &nmrep, np, &premtime, &wccpostattr, &xid);
+ nfsmout_if(error);
/* if file hadn't changed, update cached mtime */
- if (nfstimespeccmp(&VTONFS(vp)->n_mtime, &premtime, ==)) {
- VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.nva_mtime;
- }
+ if (nfstimespeccmp(&np->n_mtime, &premtime, ==))
+ NFS_CHANGED_UPDATE(nfsvers, np, &np->n_vattr);
/* if directory hadn't changed, update namecache mtime */
- if ((vnode_vtype(vp) == VDIR) &&
- nfstimespeccmp(&VTONFS(vp)->n_ncmtime, &premtime, ==)) {
- VTONFS(vp)->n_ncmtime = VTONFS(vp)->n_vattr.nva_mtime;
- }
+ if ((vnode_vtype(NFSTOV(np)) == VDIR) &&
+ nfstimespeccmp(&np->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, np, &np->n_vattr);
if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(vp));
+ NATTRINVALIDATE(np);
+ error = status;
} else {
- if (mrep) {
- nfsm_loadattr(vp, v3, NULL, &xid);
- }
+ if (!error)
+ error = status;
+ nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid);
}
- nfsm_reqdone;
+ /*
+ * We just changed the attributes and we want to make sure that we
+ * see the latest attributes. Get the next XID. If it's not the
+ * next XID after the SETATTR XID, then it's possible that another
+ * RPC was in flight at the same time and it might put stale attributes
+ * in the cache. In that case, we invalidate the attributes and set
+ * the attribute cache XID to guarantee that newer attributes will
+ * get loaded next.
+ */
+ nextxid = 0;
+ nfs_get_xid(&nextxid);
+ if (nextxid != (xid + 1)) {
+ np->n_xid = nextxid;
+ NATTRINVALIDATE(np);
+ }
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
return (error);
}
/*
- * nfs lookup call, one step at a time...
+ * NFS lookup call, one step at a time...
* First look in cache
- * If not found, unlock the directory nfsnode and do the rpc
+ * If not found, unlock the directory nfsnode and do the RPC
*/
-static int
-nfs_lookup(ap)
+int
+nfs_vnop_lookup(
struct vnop_lookup_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_dvp;
vnode_t *a_vpp;
struct componentname *a_cnp;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
struct componentname *cnp = ap->a_cnp;
vnode_t dvp = ap->a_dvp;
vnode_t *vpp = ap->a_vpp;
int flags = cnp->cn_flags;
vnode_t newvp;
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- caddr_t bpos, dpos, cp2;
- mbuf_t mreq, mrep, md, mb, mb2;
- long len;
- u_char *fhp;
- struct nfsnode *dnp, *np;
- int wantparent, error, attrflag, dattrflag, fhsize, fhisdvp;
- int v3 = NFS_ISV3(dvp);
- u_int64_t xid, dxid;
+ nfsnode_t dnp, np;
+ struct nfsmount *nmp;
+ mount_t mp;
+ int nfsvers, error, busyerror = ENOENT, isdot, isdotdot, negnamecache;
+ u_int64_t xid;
struct nfs_vattr nvattr;
- kauth_cred_t cred;
- proc_t p;
int ngflags;
+ struct vnop_access_args naa;
+ fhandle_t fh;
+ struct nfsreq rq, *req = &rq;
*vpp = NULLVP;
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
-
- wantparent = flags & (LOCKPARENT|WANTPARENT);
dnp = VTONFS(dvp);
+ NVATTR_INIT(&nvattr);
- error = nfs_getattr(dvp, &nvattr, cred, p);
- if (error)
+ mp = vnode_mount(dvp);
+ nmp = VFSTONFS(mp);
+ if (!nmp) {
+ error = ENXIO;
goto error_return;
- if (nfstimespeccmp(&dnp->n_ncmtime, &nvattr.nva_mtime, !=)) {
- /*
- * This directory has changed on us.
- * Purge any name cache entries.
- */
- cache_purge(dvp);
- dnp->n_ncmtime = nvattr.nva_mtime;
}
+ nfsvers = nmp->nm_vers;
+ negnamecache = !NMFLAG(nmp, NONEGNAMECACHE);
+
+ if ((error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx))))
+ goto error_return;
+ /* nfs_getattr() will check changed and purge caches */
+ if ((error = nfs_getattr(dnp, NULL, ctx, NGA_CACHED)))
+ goto error_return;
error = cache_lookup(dvp, vpp, cnp);
switch (error) {
case ENOENT:
- /* negative cache entry same as cache miss */
- error = 0;
- /* FALLTHROUGH */
+ /* negative cache entry */
+ goto error_return;
case 0:
/* cache miss */
- break;
+ if ((nfsvers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) {
+ /* if rdirplus, try dir buf cache lookup */
+ error = nfs_dir_buf_cache_lookup(dnp, &np, cnp, ctx, 0);
+ if (!error && np) {
+ /* dir buf cache hit */
+ *vpp = NFSTOV(np);
+ error = -1;
+ }
+ }
+ if (error != -1) /* cache miss */
+ break;
+ /* FALLTHROUGH */
case -1:
/* cache hit, not really an error */
- {
- struct vnop_access_args naa;
+ OSAddAtomic64(1, &nfsstats.lookupcache_hits);
- OSAddAtomic(1, (SInt32*)&nfsstats.lookupcache_hits);
+ nfs_node_clear_busy(dnp);
+ busyerror = ENOENT;
/* check for directory access */
+ naa.a_desc = &vnop_access_desc;
naa.a_vp = dvp;
naa.a_action = KAUTH_VNODE_SEARCH;
- naa.a_context = ap->a_context;
+ naa.a_context = ctx;
/* compute actual success/failure based on accessibility */
- error = nfs_access(&naa);
- }
+ error = nfs_vnop_access(&naa);
/* FALLTHROUGH */
default:
/* unexpected error from cache_lookup */
goto error_return;
}
-
- /* check for lookup of "." */
- if ((cnp->cn_nameptr[0] == '.') && (cnp->cn_namelen == 1)) {
- /* skip lookup, we know who we are */
- fhisdvp = 1;
- fhp = NULL;
- fhsize = 0;
- mrep = NULL;
+
+ /* skip lookup, if we know who we are: "." or ".." */
+ isdot = isdotdot = 0;
+ if (cnp->cn_nameptr[0] == '.') {
+ if (cnp->cn_namelen == 1)
+ isdot = 1;
+ if ((cnp->cn_namelen == 2) && (cnp->cn_nameptr[1] == '.'))
+ isdotdot = 1;
+ }
+ if (isdotdot || isdot) {
+ fh.fh_len = 0;
goto found;
}
+ if ((nfsvers >= NFS_VER4) && (dnp->n_vattr.nva_flags & NFS_FFLAG_TRIGGER)) {
+ /* we should never be looking things up in a trigger directory, return nothing */
+ error = ENOENT;
+ goto error_return;
+ }
/* do we know this name is too long? */
- if (v3) {
- /* For NFSv3: need uniform pathconf info to test pc_namemax */
- struct nfsmount *nmp = VFSTONFS(vnode_mount(dvp));
- if (!nmp) {
- error = ENXIO;
- goto error_return;
- }
- if (((nmp->nm_state & (NFSSTA_GOTFSINFO|NFSSTA_GOTPATHCONF)) ==
- (NFSSTA_GOTFSINFO|NFSSTA_GOTPATHCONF)) &&
- (nmp->nm_fsinfo.fsproperties & NFSV3FSINFO_HOMOGENEOUS) &&
- (cnp->cn_namelen > (long)nmp->nm_fsinfo.namemax)) {
- error = ENAMETOOLONG;
- goto error_return;
- }
- } else if (cnp->cn_namelen > NFS_MAXNAMLEN) {
+ nmp = VTONMP(dvp);
+ if (!nmp) {
+ error = ENXIO;
+ goto error_return;
+ }
+ if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME) &&
+ (cnp->cn_namelen > (int)nmp->nm_fsattr.nfsa_maxname)) {
error = ENAMETOOLONG;
goto error_return;
}
error = 0;
newvp = NULLVP;
- OSAddAtomic(1, (SInt32*)&nfsstats.lookupcache_misses);
- len = cnp->cn_namelen;
- nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
- if (error)
- goto error_return;
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_LOOKUP]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN, v3);
- /* nfsm_request for NFSv2 causes you to goto to nfsmout upon errors */
- nfsm_request(dvp, NFSPROC_LOOKUP, p, cred, &xid);
+ OSAddAtomic64(1, &nfsstats.lookupcache_misses);
- if (error) {
- if (mrep) {
- nfsm_postop_attr_update(dvp, v3, dattrflag, &xid);
- mbuf_freem(mrep);
- }
- goto nfsmout;
- }
+ error = nmp->nm_funcs->nf_lookup_rpc_async(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &req);
+ nfsmout_if(error);
+ error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, req, &xid, &fh, &nvattr);
+ nfsmout_if(error);
- /* get the filehandle */
- nfsm_getfh(fhp, fhsize, v3);
/* is the file handle the same as this directory's file handle? */
- fhisdvp = NFS_CMPFH(dnp, fhp, fhsize);
-
- /* get attributes */
- if (v3) {
- dxid = xid;
- nfsm_postop_attr_get(v3, attrflag, &nvattr);
- nfsm_postop_attr_update(dvp, v3, dattrflag, &dxid);
- if (!attrflag && (!fhisdvp || !dattrflag)) {
- /* We need valid attributes in order */
- /* to call nfs_nget/vnode_create(). */
- error = nfs_getattr_no_vnode(vnode_mount(dvp),
- fhp, fhsize, cred, p, &nvattr, &xid);
- if (error) {
- mbuf_freem(mrep);
+ isdot = NFS_CMPFH(dnp, fh.fh_data, fh.fh_len);
+
+found:
+ if (flags & ISLASTCN) {
+ switch (cnp->cn_nameiop) {
+ case DELETE:
+ cnp->cn_flags &= ~MAKEENTRY;
+ break;
+ case RENAME:
+ cnp->cn_flags &= ~MAKEENTRY;
+ if (isdot) {
+ error = EISDIR;
goto error_return;
}
+ break;
}
- } else {
- nfsm_attr_get(v3, &nvattr);
}
-found:
-
- /*
- * Handle RENAME case...
- */
- if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
- if (fhisdvp) {
- mbuf_freem(mrep);
- error = EISDIR;
- goto error_return;
- }
- error = nfs_nget(vnode_mount(dvp), dvp, cnp, fhp, fhsize,
- &nvattr, &xid, 0, &np);
- if (error) {
- mbuf_freem(mrep);
+ if (isdotdot) {
+ newvp = vnode_getparent(dvp);
+ if (!newvp) {
+ error = ENOENT;
goto error_return;
}
- *vpp = NFSTOV(np);
- mbuf_freem(mrep);
-
- goto error_return;
- }
-
- if ((cnp->cn_flags & MAKEENTRY) &&
- (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)))
- ngflags = NG_MAKEENTRY;
- else
- ngflags = 0;
-
- if (fhisdvp) {
+ } else if (isdot) {
error = vnode_get(dvp);
- if (error) {
- mbuf_freem(mrep);
+ if (error)
goto error_return;
- }
newvp = dvp;
- /* test fhp to see if we have valid attributes in nvattr */
- if (fhp && (dnp->n_xid <= xid)) {
- error = nfs_loadattrcache(dnp, &nvattr, &xid, 0);
- if (error) {
- vnode_put(dvp);
- mbuf_freem(mrep);
- goto error_return;
- }
- }
+ nfs_node_lock_force(dnp);
+ if (fh.fh_len && (dnp->n_xid <= xid))
+ nfs_loadattrcache(dnp, &nvattr, &xid, 0);
+ nfs_node_unlock(dnp);
} else {
- error = nfs_nget(vnode_mount(dvp), dvp, cnp, fhp, fhsize,
- &nvattr, &xid, ngflags, &np);
- if (error) {
- mbuf_freem(mrep);
+ ngflags = (cnp->cn_flags & MAKEENTRY) ? NG_MAKEENTRY : 0;
+ error = nfs_nget(mp, dnp, cnp, fh.fh_data, fh.fh_len, &nvattr, &xid, rq.r_auth, ngflags, &np);
+ if (error)
goto error_return;
- }
newvp = NFSTOV(np);
+ nfs_node_unlock(np);
}
*vpp = newvp;
-// if (error == 0 && *vpp != NULL && *vpp != dvp)
-// nfs_unlock(VTONFS(*vpp));
- nfsm_reqdone;
+nfsmout:
if (error) {
- if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
- (flags & ISLASTCN) && error == ENOENT) {
+ if (((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME)) &&
+ (flags & ISLASTCN) && (error == ENOENT)) {
if (vnode_mount(dvp) && vnode_vfsisrdonly(dvp))
error = EROFS;
else
error = EJUSTRETURN;
}
}
+ if ((error == ENOENT) && (cnp->cn_flags & MAKEENTRY) &&
+ (cnp->cn_nameiop != CREATE) && negnamecache) {
+ /* add a negative entry in the name cache */
+ nfs_node_lock_force(dnp);
+ cache_enter(dvp, NULL, cnp);
+ dnp->n_flag |= NNEGNCENTRIES;
+ nfs_node_unlock(dnp);
+ }
error_return:
+ NVATTR_CLEANUP(&nvattr);
+ if (!busyerror)
+ nfs_node_clear_busy(dnp);
if (error && *vpp) {
vnode_put(*vpp);
*vpp = NULLVP;
}
/*
- * nfs read call.
- * Just call nfs_bioread() to do the work.
- */
-static int
-nfs_read(ap)
- struct vnop_read_args /* {
- struct vnodeop_desc *a_desc;
- vnode_t a_vp;
- struct uio *a_uio;
- int a_ioflag;
- vfs_context_t a_context;
- } */ *ap;
-{
- if (vnode_vtype(ap->a_vp) != VREG)
- return (EPERM);
- return (nfs_bioread(ap->a_vp, ap->a_uio, ap->a_ioflag,
- vfs_context_ucred(ap->a_context),
- vfs_context_proc(ap->a_context)));
-}
-
-
-/*
- * nfs readlink call
+ * NFS readlink call
*/
-static int
-nfs_readlink(ap)
+int
+nfs_vnop_readlink(
struct vnop_readlink_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
struct uio *a_uio;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
+ nfsnode_t np = VTONFS(ap->a_vp);
+ struct nfsmount *nmp;
+ int error = 0, nfsvers;
+ uint32_t buflen;
+ uio_t uio = ap->a_uio;
+ struct nfsbuf *bp = NULL;
+
if (vnode_vtype(ap->a_vp) != VLNK)
return (EPERM);
- return (nfs_bioread(ap->a_vp, ap->a_uio, 0,
- vfs_context_ucred(ap->a_context),
- vfs_context_proc(ap->a_context)));
+
+ if (uio_resid(uio) == 0)
+ return (0);
+ if (uio_offset(uio) < 0)
+ return (EINVAL);
+
+ nmp = VTONMP(ap->a_vp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+
+ /* nfs_getattr() will check changed and purge caches */
+ if ((error = nfs_getattr(np, NULL, ctx, NGA_CACHED))) {
+ FSDBG(531, np, 0xd1e0001, 0, error);
+ return (error);
+ }
+
+ OSAddAtomic64(1, &nfsstats.biocache_readlinks);
+ error = nfs_buf_get(np, 0, NFS_MAXPATHLEN, vfs_context_thread(ctx), NBLK_READ, &bp);
+ if (error) {
+ FSDBG(531, np, 0xd1e0002, 0, error);
+ return (error);
+ }
+ if (!ISSET(bp->nb_flags, NB_CACHE)) {
+ OSAddAtomic64(1, &nfsstats.readlink_bios);
+ buflen = bp->nb_bufsize;
+ error = nmp->nm_funcs->nf_readlink_rpc(np, bp->nb_data, &buflen, ctx);
+ if (error) {
+ SET(bp->nb_flags, NB_ERROR);
+ bp->nb_error = error;
+ } else {
+ bp->nb_validoff = 0;
+ bp->nb_validend = buflen;
+ }
+ }
+ if (!error && (bp->nb_validend > 0))
+ error = uiomove(bp->nb_data, bp->nb_validend, uio);
+ FSDBG(531, np, bp->nb_validend, 0, error);
+ nfs_buf_release(bp, 1);
+ return (error);
}
/*
- * Do a readlink rpc.
- * Called by nfs_doio() from below the buffer cache.
+ * Do a readlink RPC.
*/
int
-nfs_readlinkrpc(
- vnode_t vp,
- struct uio *uiop,
- kauth_cred_t cred,
- proc_t p)
+nfs3_readlink_rpc(nfsnode_t np, char *buf, uint32_t *buflenp, vfs_context_t ctx)
{
- register u_long *tl;
- register caddr_t cp;
- register long t1, t2;
- caddr_t bpos, dpos, cp2;
- int error = 0, len, attrflag;
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3;
+ struct nfsmount *nmp;
+ int error = 0, lockerror = ENOENT, nfsvers, status;
+ uint32_t len;
u_int64_t xid;
+ struct nfsm_chain nmreq, nmrep;
- if (!VFSTONFS(vnode_mount(vp)))
+ nmp = NFSTONMP(np);
+ if (!nmp)
return (ENXIO);
- v3 = NFS_ISV3(vp);
-
- nfsm_reqhead(NFSX_FH(v3));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_READLINK]);
- nfsm_fhtom(vp, v3);
- nfsm_request(vp, NFSPROC_READLINK, p, cred, &xid);
- if (v3 && mrep)
- nfsm_postop_attr_update(vp, v3, attrflag, &xid);
- if (!error) {
- nfsm_strsiz(len, NFS_MAXPATHLEN, v3);
- if (len >= NFS_MAXPATHLEN) {
- struct nfsnode *np = VTONFS(vp);
-#if DIAGNOSTIC
- if (!np)
- panic("nfs_readlinkrpc: null np");
-#endif
- if (np->n_size && np->n_size < NFS_MAXPATHLEN)
- len = np->n_size;
- }
- nfsm_mtouio(uiop, len);
- }
- nfsm_reqdone;
+ nfsvers = nmp->nm_vers;
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(nfsvers));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request(np, NULL, &nmreq, NFSPROC_READLINK, ctx, NULL, &nmrep, &xid, &status);
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
+ if (!error)
+ error = status;
+ nfsm_chain_get_32(error, &nmrep, len);
+ nfsmout_if(error);
+ if ((nfsvers == NFS_VER2) && (len > *buflenp)) {
+ error = EBADRPC;
+ goto nfsmout;
+ }
+ if (len >= *buflenp) {
+ if (np->n_size && (np->n_size < *buflenp))
+ len = np->n_size;
+ else
+ len = *buflenp - 1;
+ }
+ nfsm_chain_get_opaque(error, &nmrep, len, buf);
+ if (!error)
+ *buflenp = len;
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
return (error);
}
/*
- * nfs read rpc call
+ * NFS read RPC call
* Ditto above
*/
int
-nfs_readrpc(
- vnode_t vp,
- struct uio *uiop,
- kauth_cred_t cred,
- proc_t p)
+nfs_read_rpc(nfsnode_t np, uio_t uio, vfs_context_t ctx)
{
- register u_long *tl;
- register caddr_t cp;
- register long t1, t2;
- caddr_t bpos, dpos, cp2;
- mbuf_t mreq, mrep, md, mb, mb2;
struct nfsmount *nmp;
- int error = 0, len, retlen, tsiz, eof = 0, attrflag;
- int v3, nmrsize;
- u_int64_t xid;
-
- FSDBG_TOP(536, vp, uiop->uio_offset, uio_uio_resid(uiop), 0);
- nmp = VFSTONFS(vnode_mount(vp));
+ int error = 0, nfsvers, eof = 0;
+ size_t nmrsize, len, retlen;
+ user_ssize_t tsiz;
+ off_t txoffset;
+ struct nfsreq rq, *req = &rq;
+ uint32_t stategenid = 0, restart = 0;
+
+ FSDBG_TOP(536, np, uio_offset(uio), uio_resid(uio), 0);
+ nmp = NFSTONMP(np);
if (!nmp)
return (ENXIO);
- v3 = NFS_ISV3(vp);
+ nfsvers = nmp->nm_vers;
nmrsize = nmp->nm_rsize;
- // LP64todo - fix this
- tsiz = uio_uio_resid(uiop);
- if (((u_int64_t)uiop->uio_offset + (unsigned int)tsiz > 0xffffffff) && !v3) {
- FSDBG_BOT(536, vp, uiop->uio_offset, uio_uio_resid(uiop), EFBIG);
+ txoffset = uio_offset(uio);
+ tsiz = uio_resid(uio);
+ if ((nfsvers == NFS_VER2) && ((uint64_t)(txoffset + tsiz) > 0xffffffffULL)) {
+ FSDBG_BOT(536, np, uio_offset(uio), uio_resid(uio), EFBIG);
return (EFBIG);
}
+
while (tsiz > 0) {
- len = (tsiz > nmrsize) ? nmrsize : tsiz;
- nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED * 3);
- if (error)
+ len = retlen = (tsiz > (user_ssize_t)nmrsize) ? nmrsize : (size_t)tsiz;
+ FSDBG(536, np, txoffset, len, 0);
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
break;
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_READ]);
- nfsm_fhtom(vp, v3);
- nfsm_build(tl, u_long *, NFSX_UNSIGNED * 3);
- if (v3) {
- txdr_hyper(&uiop->uio_offset, tl);
- *(tl + 2) = txdr_unsigned(len);
- } else {
- *tl++ = txdr_unsigned(uiop->uio_offset);
- *tl++ = txdr_unsigned(len);
- *tl = 0;
- }
- FSDBG(536, vp, uiop->uio_offset, len, 0);
- nfsm_request(vp, NFSPROC_READ, p, cred, &xid);
- if (v3) {
- if (mrep) {
- nfsm_postop_attr_update(vp, v3, attrflag, &xid);
- }
- if (error) {
- mbuf_freem(mrep);
- goto nfsmout;
+ }
+ if (nmp->nm_vers >= NFS_VER4)
+ stategenid = nmp->nm_stategenid;
+ error = nmp->nm_funcs->nf_read_rpc_async(np, txoffset, len,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, &req);
+ if (!error)
+ error = nmp->nm_funcs->nf_read_rpc_async_finish(np, req, uio, &retlen, &eof);
+ if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error) &&
+ (++restart <= nfs_mount_state_max_restarts(nmp))) { /* guard against no progress */
+ lck_mtx_lock(&nmp->nm_lock);
+ if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
+ NP(np, "nfs_read_rpc: error %d, initiating recovery", error);
+ nfs_need_recover(nmp, error);
}
- nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED);
- eof = fxdr_unsigned(int, *(tl + 1));
- } else {
- if (mrep) {
- nfsm_loadattr(vp, v3, NULL, &xid);
+ lck_mtx_unlock(&nmp->nm_lock);
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ } else {
+ if (error == NFSERR_GRACE)
+ tsleep(&nmp->nm_state, (PZERO-1), "nfsgrace", 2*hz);
+ if (!(error = nfs_mount_state_wait_for_recovery(nmp)))
+ continue;
}
}
- if (mrep) {
- nfsm_strsiz(retlen, nmrsize, 0);
- nfsm_mtouio(uiop, retlen);
- mbuf_freem(mrep);
- } else {
- retlen = 0;
- }
+ if (error)
+ break;
+ txoffset += retlen;
tsiz -= retlen;
- if (v3) {
- if (eof || retlen == 0)
+ if (nfsvers != NFS_VER2) {
+ if (eof || (retlen == 0))
tsiz = 0;
} else if (retlen < len)
tsiz = 0;
}
-nfsmout:
- FSDBG_BOT(536, vp, eof, uio_uio_resid(uiop), error);
+
+ FSDBG_BOT(536, np, eof, uio_resid(uio), error);
return (error);
}
-/*
- * nfs write call
- */
int
-nfs_writerpc(
- vnode_t vp,
- struct uio *uiop,
+nfs3_read_rpc_async(
+ nfsnode_t np,
+ off_t offset,
+ size_t len,
+ thread_t thd,
kauth_cred_t cred,
- proc_t p,
- int *iomode,
- uint64_t *wverfp)
+ struct nfsreq_cbinfo *cb,
+ struct nfsreq **reqp)
{
- register u_long *tl;
- register caddr_t cp;
- register int t1, t2, backup;
- caddr_t bpos, dpos, cp2;
- mbuf_t mreq, mrep, md, mb, mb2;
struct nfsmount *nmp;
- int error = 0, len, tsiz, updatemtime = 0, wccpostattr = 0, rlen, commit;
- int v3, committed = NFSV3WRITE_FILESYNC;
- u_int64_t xid, wverf;
- mount_t mp;
+ int error = 0, nfsvers;
+ struct nfsm_chain nmreq;
-#if DIAGNOSTIC
- if (uiop->uio_iovcnt != 1)
- panic("nfs_writerpc: iovcnt > 1");
-#endif
- FSDBG_TOP(537, vp, uiop->uio_offset, uio_uio_resid(uiop), *iomode);
- nmp = VFSTONFS(vnode_mount(vp));
+ nmp = NFSTONMP(np);
if (!nmp)
return (ENXIO);
- v3 = NFS_ISV3(vp);
- // LP64todo - fix this
- tsiz = uio_uio_resid(uiop);
- if (((u_int64_t)uiop->uio_offset + (unsigned int)tsiz > 0xffffffff) && !v3) {
- FSDBG_BOT(537, vp, uiop->uio_offset, uio_uio_resid(uiop), EFBIG);
- return (EFBIG);
+ nfsvers = nmp->nm_vers;
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(nfsvers) + 3 * NFSX_UNSIGNED);
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
+ if (nfsvers == NFS_VER3) {
+ nfsm_chain_add_64(error, &nmreq, offset);
+ nfsm_chain_add_32(error, &nmreq, len);
+ } else {
+ nfsm_chain_add_32(error, &nmreq, offset);
+ nfsm_chain_add_32(error, &nmreq, len);
+ nfsm_chain_add_32(error, &nmreq, 0);
}
- while (tsiz > 0) {
- nmp = VFSTONFS(vnode_mount(vp));
- if (!nmp) {
- error = ENXIO;
- break;
- }
- len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
- nfsm_reqhead(NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
- if (error)
- break;
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_WRITE]);
- nfsm_fhtom(vp, v3);
- if (v3) {
- nfsm_build(tl, u_long *, 5 * NFSX_UNSIGNED);
- txdr_hyper(&uiop->uio_offset, tl);
- tl += 2;
- *tl++ = txdr_unsigned(len);
- *tl++ = txdr_unsigned(*iomode);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request_async(np, NULL, &nmreq, NFSPROC_READ, thd, cred, NULL, 0, cb, reqp);
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ return (error);
+}
+
+int
+nfs3_read_rpc_async_finish(
+ nfsnode_t np,
+ struct nfsreq *req,
+ uio_t uio,
+ size_t *lenp,
+ int *eofp)
+{
+ int error = 0, lockerror, nfsvers, status, eof = 0;
+ size_t retlen = 0;
+ uint64_t xid;
+ struct nfsmount *nmp;
+ struct nfsm_chain nmrep;
+
+ nmp = NFSTONMP(np);
+ if (!nmp) {
+ nfs_request_async_cancel(req);
+ return (ENXIO);
+ }
+ nfsvers = nmp->nm_vers;
+
+ nfsm_chain_null(&nmrep);
+
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ if (error == EINPROGRESS) /* async request restarted */
+ return (error);
+
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
+ if (!error)
+ error = status;
+ if (nfsvers == NFS_VER3) {
+ nfsm_chain_adv(error, &nmrep, NFSX_UNSIGNED);
+ nfsm_chain_get_32(error, &nmrep, eof);
+ } else {
+ nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid);
+ }
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfsm_chain_get_32(error, &nmrep, retlen);
+ if ((nfsvers == NFS_VER2) && (retlen > *lenp))
+ error = EBADRPC;
+ nfsmout_if(error);
+ error = nfsm_chain_get_uio(&nmrep, MIN(retlen, *lenp), uio);
+ if (eofp) {
+ if (nfsvers == NFS_VER3) {
+ if (!eof && !retlen)
+ eof = 1;
+ } else if (retlen < *lenp) {
+ eof = 1;
+ }
+ *eofp = eof;
+ }
+ *lenp = MIN(retlen, *lenp);
+nfsmout:
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
+
+/*
+ * NFS write call
+ */
+int
+nfs_vnop_write(
+ struct vnop_write_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ struct uio *a_uio;
+ int a_ioflag;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ vfs_context_t ctx = ap->a_context;
+ uio_t uio = ap->a_uio;
+ vnode_t vp = ap->a_vp;
+ nfsnode_t np = VTONFS(vp);
+ int ioflag = ap->a_ioflag;
+ struct nfsbuf *bp;
+ struct nfsmount *nmp = VTONMP(vp);
+ daddr64_t lbn;
+ int biosize;
+ int n, on, error = 0;
+ off_t boff, start, end;
+ uio_t auio;
+ char auio_buf [ UIO_SIZEOF(1) ];
+ thread_t thd;
+ kauth_cred_t cred;
+
+ FSDBG_TOP(515, np, uio_offset(uio), uio_resid(uio), ioflag);
+
+ if (vnode_vtype(vp) != VREG) {
+ FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), EIO);
+ return (EIO);
+ }
+
+ thd = vfs_context_thread(ctx);
+ cred = vfs_context_ucred(ctx);
+
+ nfs_data_lock(np, NFS_DATA_LOCK_SHARED);
+
+ if ((error = nfs_node_lock(np))) {
+ nfs_data_unlock(np);
+ FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), error);
+ return (error);
+ }
+ np->n_wrbusy++;
+
+ if (np->n_flag & NWRITEERR) {
+ error = np->n_error;
+ np->n_flag &= ~NWRITEERR;
+ }
+ if (np->n_flag & NNEEDINVALIDATE) {
+ np->n_flag &= ~NNEEDINVALIDATE;
+ nfs_node_unlock(np);
+ nfs_data_unlock(np);
+ nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1);
+ nfs_data_lock(np, NFS_DATA_LOCK_SHARED);
+ } else {
+ nfs_node_unlock(np);
+ }
+ if (error)
+ goto out;
+
+ biosize = nmp->nm_biosize;
+
+ if (ioflag & (IO_APPEND | IO_SYNC)) {
+ nfs_node_lock_force(np);
+ if (np->n_flag & NMODIFIED) {
+ NATTRINVALIDATE(np);
+ nfs_node_unlock(np);
+ nfs_data_unlock(np);
+ error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
+ nfs_data_lock(np, NFS_DATA_LOCK_SHARED);
+ if (error) {
+ FSDBG(515, np, uio_offset(uio), 0x10bad01, error);
+ goto out;
+ }
} else {
- nfsm_build(tl, u_long *, 4 * NFSX_UNSIGNED);
- *++tl = txdr_unsigned(uiop->uio_offset);
- tl += 2;
- }
- *tl = txdr_unsigned(len);
- FSDBG(537, vp, uiop->uio_offset, len, 0);
- nfsm_uiotom(uiop, len);
- nfsm_request(vp, NFSPROC_WRITE, p, cred, &xid);
- nmp = VFSTONFS(vnode_mount(vp));
- if (!nmp)
- error = ENXIO;
- if (v3) {
- if (mrep) {
- struct timespec premtime;
- nfsm_wcc_data(vp, &premtime, wccpostattr, &xid);
- if (nfstimespeccmp(&VTONFS(vp)->n_mtime, &premtime, ==))
- updatemtime = 1;
+ nfs_node_unlock(np);
+ }
+ if (ioflag & IO_APPEND) {
+ nfs_data_unlock(np);
+ /* nfs_getattr() will check changed and purge caches */
+ error = nfs_getattr(np, NULL, ctx, NGA_UNCACHED);
+ /* we'll be extending the file, so take the data lock exclusive */
+ nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
+ if (error) {
+ FSDBG(515, np, uio_offset(uio), 0x10bad02, error);
+ goto out;
}
- if (!error) {
- nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED +
- NFSX_V3WRITEVERF);
- rlen = fxdr_unsigned(int, *tl++);
- if (rlen <= 0) {
- error = NFSERR_IO;
- break;
- } else if (rlen < len) {
- backup = len - rlen;
- uio_iov_base_add(uiop, -backup);
- uio_iov_len_add(uiop, backup);
- uiop->uio_offset -= backup;
- uio_uio_resid_add(uiop, backup);
- len = rlen;
- }
- commit = fxdr_unsigned(int, *tl++);
+ uio_setoffset(uio, np->n_size);
+ }
+ }
+ if (uio_offset(uio) < 0) {
+ error = EINVAL;
+ FSDBG_BOT(515, np, uio_offset(uio), 0xbad0ff, error);
+ goto out;
+ }
+ if (uio_resid(uio) == 0)
+ goto out;
+
+ if (((uio_offset(uio) + uio_resid(uio)) > (off_t)np->n_size) && !(ioflag & IO_APPEND)) {
+ /* it looks like we'll be extending the file, so take the data lock exclusive */
+ nfs_data_unlock(np);
+ nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
+ }
+
+ do {
+ OSAddAtomic64(1, &nfsstats.biocache_writes);
+ lbn = uio_offset(uio) / biosize;
+ on = uio_offset(uio) % biosize;
+ n = biosize - on;
+ if (uio_resid(uio) < n)
+ n = uio_resid(uio);
+again:
+ /*
+ * Get a cache block for writing. The range to be written is
+ * (off..off+n) within the block. We ensure that the block
+ * either has no dirty region or that the given range is
+ * contiguous with the existing dirty region.
+ */
+ error = nfs_buf_get(np, lbn, biosize, thd, NBLK_WRITE, &bp);
+ if (error)
+ goto out;
+ /* map the block because we know we're going to write to it */
+ NFS_BUF_MAP(bp);
+
+ if (ioflag & IO_NOCACHE)
+ SET(bp->nb_flags, NB_NOCACHE);
+
+ if (!IS_VALID_CRED(bp->nb_wcred)) {
+ kauth_cred_ref(cred);
+ bp->nb_wcred = cred;
+ }
+ /*
+ * If there's already a dirty range AND dirty pages in this block we
+ * need to send a commit AND write the dirty pages before continuing.
+ *
+ * If there's already a dirty range OR dirty pages in this block
+ * and the new write range is not contiguous with the existing range,
+ * then force the buffer to be written out now.
+ * (We used to just extend the dirty range to cover the valid,
+ * but unwritten, data in between also. But writing ranges
+ * of data that weren't actually written by an application
+ * risks overwriting some other client's data with stale data
+ * that's just masquerading as new written data.)
+ */
+ if (bp->nb_dirtyend > 0) {
+ if (on > bp->nb_dirtyend || (on + n) < bp->nb_dirtyoff || bp->nb_dirty) {
+ FSDBG(515, np, uio_offset(uio), bp, 0xd15c001);
+ /* write/commit buffer "synchronously" */
+ /* (NB_STABLE indicates that data writes should be FILESYNC) */
+ CLR(bp->nb_flags, (NB_DONE | NB_ERROR | NB_INVAL));
+ SET(bp->nb_flags, (NB_ASYNC | NB_STABLE));
+ error = nfs_buf_write(bp);
+ if (error)
+ goto out;
+ goto again;
+ }
+ } else if (bp->nb_dirty) {
+ int firstpg, lastpg;
+ u_int32_t pagemask;
+ /* calculate write range pagemask */
+ firstpg = on/PAGE_SIZE;
+ lastpg = (on+n-1)/PAGE_SIZE;
+ pagemask = ((1 << (lastpg+1)) - 1) & ~((1 << firstpg) - 1);
+ /* check if there are dirty pages outside the write range */
+ if (bp->nb_dirty & ~pagemask) {
+ FSDBG(515, np, uio_offset(uio), bp, 0xd15c002);
+ /* write/commit buffer "synchronously" */
+ /* (NB_STABLE indicates that data writes should be FILESYNC) */
+ CLR(bp->nb_flags, (NB_DONE | NB_ERROR | NB_INVAL));
+ SET(bp->nb_flags, (NB_ASYNC | NB_STABLE));
+ error = nfs_buf_write(bp);
+ if (error)
+ goto out;
+ goto again;
+ }
+ /* if the first or last pages are already dirty */
+ /* make sure that the dirty range encompasses those pages */
+ if (NBPGDIRTY(bp,firstpg) || NBPGDIRTY(bp,lastpg)) {
+ FSDBG(515, np, uio_offset(uio), bp, 0xd15c003);
+ bp->nb_dirtyoff = min(on, firstpg * PAGE_SIZE);
+ if (NBPGDIRTY(bp,lastpg)) {
+ bp->nb_dirtyend = (lastpg+1) * PAGE_SIZE;
+ /* clip to EOF */
+ if (NBOFF(bp) + bp->nb_dirtyend > (off_t)np->n_size) {
+ bp->nb_dirtyend = np->n_size - NBOFF(bp);
+ if (bp->nb_dirtyoff >= bp->nb_dirtyend)
+ bp->nb_dirtyoff = bp->nb_dirtyend = 0;
+ }
+ } else
+ bp->nb_dirtyend = on+n;
+ }
+ }
+
+ /*
+ * Are we extending the size of the file with this write?
+ * If so, update file size now that we have the block.
+ * If there was a partial buf at the old eof, validate
+ * and zero the new bytes.
+ */
+ if ((uio_offset(uio) + n) > (off_t)np->n_size) {
+ struct nfsbuf *eofbp = NULL;
+ daddr64_t eofbn = np->n_size / biosize;
+ int eofoff = np->n_size % biosize;
+ int neweofoff = (uio_offset(uio) + n) % biosize;
+
+ FSDBG(515, 0xb1ffa000, uio_offset(uio) + n, eofoff, neweofoff);
+
+ if (eofoff && (eofbn < lbn) &&
+ ((error = nfs_buf_get(np, eofbn, biosize, thd, NBLK_WRITE|NBLK_ONLYVALID, &eofbp))))
+ goto out;
+
+ /* if we're extending within the same last block */
+ /* and the block is flagged as being cached... */
+ if ((lbn == eofbn) && ISSET(bp->nb_flags, NB_CACHE)) {
+ /* ...check that all pages in buffer are valid */
+ int endpg = ((neweofoff ? neweofoff : biosize) - 1)/PAGE_SIZE;
+ u_int32_t pagemask;
+ /* pagemask only has to extend to last page being written to */
+ pagemask = (1 << (endpg+1)) - 1;
+ FSDBG(515, 0xb1ffa001, bp->nb_valid, pagemask, 0);
+ if ((bp->nb_valid & pagemask) != pagemask) {
+ /* zerofill any hole */
+ if (on > bp->nb_validend) {
+ int i;
+ for (i=bp->nb_validend/PAGE_SIZE; i <= (on - 1)/PAGE_SIZE; i++)
+ NBPGVALID_SET(bp, i);
+ NFS_BUF_MAP(bp);
+ FSDBG(516, bp, bp->nb_validend, on - bp->nb_validend, 0xf01e);
+ bzero((char *)bp->nb_data + bp->nb_validend,
+ on - bp->nb_validend);
+ }
+ /* zerofill any trailing data in the last page */
+ if (neweofoff) {
+ NFS_BUF_MAP(bp);
+ FSDBG(516, bp, neweofoff, PAGE_SIZE - (neweofoff & PAGE_MASK), 0xe0f);
+ bzero((char *)bp->nb_data + neweofoff,
+ PAGE_SIZE - (neweofoff & PAGE_MASK));
+ }
+ }
+ }
+ np->n_size = uio_offset(uio) + n;
+ nfs_node_lock_force(np);
+ CLR(np->n_flag, NUPDATESIZE);
+ np->n_flag |= NMODIFIED;
+ nfs_node_unlock(np);
+ FSDBG(516, np, np->n_size, np->n_vattr.nva_size, 0xf00d0001);
+ ubc_setsize(vp, (off_t)np->n_size); /* XXX errors */
+ if (eofbp) {
/*
- * Return the lowest committment level
- * obtained by any of the RPCs.
+ * We may need to zero any previously invalid data
+ * after the old EOF in the previous EOF buffer.
+ *
+ * For the old last page, don't zero bytes if there
+ * are invalid bytes in that page (i.e. the page isn't
+ * currently valid).
+ * For pages after the old last page, zero them and
+ * mark them as valid.
*/
- if (committed == NFSV3WRITE_FILESYNC)
- committed = commit;
- else if (committed == NFSV3WRITE_DATASYNC &&
- commit == NFSV3WRITE_UNSTABLE)
- committed = commit;
- fxdr_hyper(tl, &wverf);
- if (wverfp)
- *wverfp = wverf;
- if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
- nmp->nm_verf = wverf;
- nmp->nm_state |= NFSSTA_HASWRITEVERF;
- } else if (wverf != nmp->nm_verf) {
- nmp->nm_verf = wverf;
+ char *d;
+ int i;
+ if (ioflag & IO_NOCACHE)
+ SET(eofbp->nb_flags, NB_NOCACHE);
+ NFS_BUF_MAP(eofbp);
+ FSDBG(516, eofbp, eofoff, biosize - eofoff, 0xe0fff01e);
+ d = eofbp->nb_data;
+ i = eofoff/PAGE_SIZE;
+ while (eofoff < biosize) {
+ int poff = eofoff & PAGE_MASK;
+ if (!poff || NBPGVALID(eofbp,i)) {
+ bzero(d + eofoff, PAGE_SIZE - poff);
+ NBPGVALID_SET(eofbp, i);
+ }
+ if (bp->nb_validend == eofoff)
+ bp->nb_validend += PAGE_SIZE - poff;
+ eofoff += PAGE_SIZE - poff;
+ i++;
}
+ nfs_buf_release(eofbp, 1);
}
- } else {
- if (mrep) {
- nfsm_loadattr(vp, v3, NULL, &xid);
+ }
+ /*
+ * If dirtyend exceeds file size, chop it down. This should
+ * not occur unless there is a race.
+ */
+ if (NBOFF(bp) + bp->nb_dirtyend > (off_t)np->n_size) {
+ bp->nb_dirtyend = np->n_size - NBOFF(bp);
+ if (bp->nb_dirtyoff >= bp->nb_dirtyend)
+ bp->nb_dirtyoff = bp->nb_dirtyend = 0;
+ }
+ /*
+ * UBC doesn't handle partial pages, so we need to make sure
+ * that any pages left in the page cache are completely valid.
+ *
+ * Writes that are smaller than a block are delayed if they
+ * don't extend to the end of the block.
+ *
+ * If the block isn't (completely) cached, we may need to read
+ * in some parts of pages that aren't covered by the write.
+ * If the write offset (on) isn't page aligned, we'll need to
+ * read the start of the first page being written to. Likewise,
+ * if the offset of the end of the write (on+n) isn't page aligned,
+ * we'll need to read the end of the last page being written to.
+ *
+ * Notes:
+ * We don't want to read anything we're just going to write over.
+ * We don't want to read anything we're just going drop when the
+ * I/O is complete (i.e. don't do reads for NOCACHE requests).
+ * We don't want to issue multiple I/Os if we don't have to
+ * (because they're synchronous rpcs).
+ * We don't want to read anything we already have modified in the
+ * page cache.
+ */
+ if (!ISSET(bp->nb_flags, NB_CACHE) && (n < biosize)) {
+ int firstpg, lastpg, dirtypg;
+ int firstpgoff, lastpgoff;
+ start = end = -1;
+ firstpg = on/PAGE_SIZE;
+ firstpgoff = on & PAGE_MASK;
+ lastpg = (on+n-1)/PAGE_SIZE;
+ lastpgoff = (on+n) & PAGE_MASK;
+ if (firstpgoff && !NBPGVALID(bp,firstpg)) {
+ /* need to read start of first page */
+ start = firstpg * PAGE_SIZE;
+ end = start + firstpgoff;
+ }
+ if (lastpgoff && !NBPGVALID(bp,lastpg)) {
+ /* need to read end of last page */
+ if (start < 0)
+ start = (lastpg * PAGE_SIZE) + lastpgoff;
+ end = (lastpg + 1) * PAGE_SIZE;
+ }
+ if (ISSET(bp->nb_flags, NB_NOCACHE)) {
+ /*
+ * For nocache writes, if there is any partial page at the
+ * start or end of the write range, then we do the write
+ * synchronously to make sure that we can drop the data
+ * from the cache as soon as the WRITE finishes. Normally,
+ * we would do an unstable write and not drop the data until
+ * it was committed. But doing that here would risk allowing
+ * invalid data to be read from the cache between the WRITE
+ * and the COMMIT.
+ * (NB_STABLE indicates that data writes should be FILESYNC)
+ */
+ if (end > start)
+ SET(bp->nb_flags, NB_STABLE);
+ goto skipread;
+ }
+ if (end > start) {
+ /* need to read the data in range: start...end-1 */
+
+ /* first, check for dirty pages in between */
+ /* if there are, we'll have to do two reads because */
+ /* we don't want to overwrite the dirty pages. */
+ for (dirtypg=start/PAGE_SIZE; dirtypg <= (end-1)/PAGE_SIZE; dirtypg++)
+ if (NBPGDIRTY(bp,dirtypg))
+ break;
+
+ /* if start is at beginning of page, try */
+ /* to get any preceeding pages as well. */
+ if (!(start & PAGE_MASK)) {
+ /* stop at next dirty/valid page or start of block */
+ for (; start > 0; start-=PAGE_SIZE)
+ if (NBPGVALID(bp,((start-1)/PAGE_SIZE)))
+ break;
+ }
+
+ NFS_BUF_MAP(bp);
+ /* setup uio for read(s) */
+ boff = NBOFF(bp);
+ auio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_READ,
+ &auio_buf, sizeof(auio_buf));
+
+ if (dirtypg <= (end-1)/PAGE_SIZE) {
+ /* there's a dirty page in the way, so just do two reads */
+ /* we'll read the preceding data here */
+ uio_reset(auio, boff + start, UIO_SYSSPACE, UIO_READ);
+ uio_addiov(auio, CAST_USER_ADDR_T(bp->nb_data + start), on - start);
+ error = nfs_read_rpc(np, auio, ctx);
+ if (error) {
+ /* couldn't read the data, so treat buffer as synchronous NOCACHE */
+ SET(bp->nb_flags, (NB_NOCACHE|NB_STABLE));
+ goto skipread;
+ }
+ if (uio_resid(auio) > 0) {
+ FSDBG(516, bp, (caddr_t)uio_curriovbase(auio) - bp->nb_data, uio_resid(auio), 0xd00dee01);
+ bzero(CAST_DOWN(caddr_t, uio_curriovbase(auio)), uio_resid(auio));
+ }
+ if (!error) {
+ /* update validoff/validend if necessary */
+ if ((bp->nb_validoff < 0) || (bp->nb_validoff > start))
+ bp->nb_validoff = start;
+ if ((bp->nb_validend < 0) || (bp->nb_validend < on))
+ bp->nb_validend = on;
+ if ((off_t)np->n_size > boff + bp->nb_validend)
+ bp->nb_validend = min(np->n_size - (boff + start), biosize);
+ /* validate any pages before the write offset */
+ for (; start < on/PAGE_SIZE; start+=PAGE_SIZE)
+ NBPGVALID_SET(bp, start/PAGE_SIZE);
+ }
+ /* adjust start to read any trailing data */
+ start = on+n;
+ }
+
+ /* if end is at end of page, try to */
+ /* get any following pages as well. */
+ if (!(end & PAGE_MASK)) {
+ /* stop at next valid page or end of block */
+ for (; end < biosize; end+=PAGE_SIZE)
+ if (NBPGVALID(bp,end/PAGE_SIZE))
+ break;
+ }
+
+ if (((boff+start) >= (off_t)np->n_size) ||
+ ((start >= on) && ((boff + on + n) >= (off_t)np->n_size))) {
+ /*
+ * Either this entire read is beyond the current EOF
+ * or the range that we won't be modifying (on+n...end)
+ * is all beyond the current EOF.
+ * No need to make a trip across the network to
+ * read nothing. So, just zero the buffer instead.
+ */
+ FSDBG(516, bp, start, end - start, 0xd00dee00);
+ bzero(bp->nb_data + start, end - start);
+ error = 0;
+ } else {
+ /* now we'll read the (rest of the) data */
+ uio_reset(auio, boff + start, UIO_SYSSPACE, UIO_READ);
+ uio_addiov(auio, CAST_USER_ADDR_T(bp->nb_data + start), end - start);
+ error = nfs_read_rpc(np, auio, ctx);
+ if (error) {
+ /* couldn't read the data, so treat buffer as synchronous NOCACHE */
+ SET(bp->nb_flags, (NB_NOCACHE|NB_STABLE));
+ goto skipread;
+ }
+ if (uio_resid(auio) > 0) {
+ FSDBG(516, bp, (caddr_t)uio_curriovbase(auio) - bp->nb_data, uio_resid(auio), 0xd00dee02);
+ bzero(CAST_DOWN(caddr_t, uio_curriovbase(auio)), uio_resid(auio));
+ }
+ }
+ if (!error) {
+ /* update validoff/validend if necessary */
+ if ((bp->nb_validoff < 0) || (bp->nb_validoff > start))
+ bp->nb_validoff = start;
+ if ((bp->nb_validend < 0) || (bp->nb_validend < end))
+ bp->nb_validend = end;
+ if ((off_t)np->n_size > boff + bp->nb_validend)
+ bp->nb_validend = min(np->n_size - (boff + start), biosize);
+ /* validate any pages before the write offset's page */
+ for (; start < (off_t)trunc_page_32(on); start+=PAGE_SIZE)
+ NBPGVALID_SET(bp, start/PAGE_SIZE);
+ /* validate any pages after the range of pages being written to */
+ for (; (end - 1) > (off_t)round_page_32(on+n-1); end-=PAGE_SIZE)
+ NBPGVALID_SET(bp, (end-1)/PAGE_SIZE);
+ }
+ /* Note: pages being written to will be validated when written */
}
}
+skipread:
+
+ if (ISSET(bp->nb_flags, NB_ERROR)) {
+ error = bp->nb_error;
+ nfs_buf_release(bp, 1);
+ goto out;
+ }
+
+ nfs_node_lock_force(np);
+ np->n_flag |= NMODIFIED;
+ nfs_node_unlock(np);
+
+ NFS_BUF_MAP(bp);
+ error = uiomove((char *)bp->nb_data + on, n, uio);
+ if (error) {
+ SET(bp->nb_flags, NB_ERROR);
+ nfs_buf_release(bp, 1);
+ goto out;
+ }
+
+ /* validate any pages written to */
+ start = on & ~PAGE_MASK;
+ for (; start < on+n; start += PAGE_SIZE) {
+ NBPGVALID_SET(bp, start/PAGE_SIZE);
+ /*
+ * This may seem a little weird, but we don't actually set the
+ * dirty bits for writes. This is because we keep the dirty range
+ * in the nb_dirtyoff/nb_dirtyend fields. Also, particularly for
+ * delayed writes, when we give the pages back to the VM we don't
+ * want to keep them marked dirty, because when we later write the
+ * buffer we won't be able to tell which pages were written dirty
+ * and which pages were mmapped and dirtied.
+ */
+ }
+ if (bp->nb_dirtyend > 0) {
+ bp->nb_dirtyoff = min(on, bp->nb_dirtyoff);
+ bp->nb_dirtyend = max((on + n), bp->nb_dirtyend);
+ } else {
+ bp->nb_dirtyoff = on;
+ bp->nb_dirtyend = on + n;
+ }
+ if (bp->nb_validend <= 0 || bp->nb_validend < bp->nb_dirtyoff ||
+ bp->nb_validoff > bp->nb_dirtyend) {
+ bp->nb_validoff = bp->nb_dirtyoff;
+ bp->nb_validend = bp->nb_dirtyend;
+ } else {
+ bp->nb_validoff = min(bp->nb_validoff, bp->nb_dirtyoff);
+ bp->nb_validend = max(bp->nb_validend, bp->nb_dirtyend);
+ }
+ if (!ISSET(bp->nb_flags, NB_CACHE))
+ nfs_buf_normalize_valid_range(np, bp);
- if (updatemtime)
- VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.nva_mtime;
- mbuf_freem(mrep);
/*
- * we seem to have a case where we end up looping on shutdown
- * and taking down nfs servers. For V3, error cases, there is
- * no way to terminate loop, if the len was 0, meaning,
- * nmp->nm_wsize was trashed. FreeBSD has this fix in it.
- * Let's try it.
+ * Since this block is being modified, it must be written
+ * again and not just committed.
*/
+ if (ISSET(bp->nb_flags, NB_NEEDCOMMIT)) {
+ nfs_node_lock_force(np);
+ if (ISSET(bp->nb_flags, NB_NEEDCOMMIT)) {
+ np->n_needcommitcnt--;
+ CHECK_NEEDCOMMITCNT(np);
+ }
+ CLR(bp->nb_flags, NB_NEEDCOMMIT);
+ nfs_node_unlock(np);
+ }
+
+ if (ioflag & IO_SYNC) {
+ error = nfs_buf_write(bp);
+ if (error)
+ goto out;
+ } else if (((n + on) == biosize) || (ioflag & IO_APPEND) ||
+ (ioflag & IO_NOCACHE) || ISSET(bp->nb_flags, NB_NOCACHE)) {
+ SET(bp->nb_flags, NB_ASYNC);
+ error = nfs_buf_write(bp);
+ if (error)
+ goto out;
+ } else {
+ /* If the block wasn't already delayed: charge for the write */
+ if (!ISSET(bp->nb_flags, NB_DELWRI)) {
+ proc_t p = vfs_context_proc(ctx);
+ if (p && p->p_stats)
+ OSIncrementAtomicLong(&p->p_stats->p_ru.ru_oublock);
+ }
+ nfs_buf_write_delayed(bp);
+ }
+ if (np->n_needcommitcnt >= NFS_A_LOT_OF_NEEDCOMMITS)
+ nfs_flushcommits(np, 1);
+
+ } while (uio_resid(uio) > 0 && n > 0);
+
+out:
+ nfs_node_lock_force(np);
+ np->n_wrbusy--;
+ nfs_node_unlock(np);
+ nfs_data_unlock(np);
+ FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), error);
+ return (error);
+}
+
+
+/*
+ * NFS write call
+ */
+int
+nfs_write_rpc(
+ nfsnode_t np,
+ uio_t uio,
+ vfs_context_t ctx,
+ int *iomodep,
+ uint64_t *wverfp)
+{
+ return nfs_write_rpc2(np, uio, vfs_context_thread(ctx), vfs_context_ucred(ctx), iomodep, wverfp);
+}
+
+int
+nfs_write_rpc2(
+ nfsnode_t np,
+ uio_t uio,
+ thread_t thd,
+ kauth_cred_t cred,
+ int *iomodep,
+ uint64_t *wverfp)
+{
+ struct nfsmount *nmp;
+ int error = 0, nfsvers;
+ int wverfset, commit, committed;
+ uint64_t wverf = 0, wverf2;
+ size_t nmwsize, totalsize, tsiz, len, rlen;
+ struct nfsreq rq, *req = &rq;
+ uint32_t stategenid = 0, vrestart = 0, restart = 0;
+ uio_t uio_save = NULL;
+
+#if DIAGNOSTIC
+ /* XXX limitation based on need to back up uio on short write */
+ if (uio_iovcnt(uio) != 1)
+ panic("nfs3_write_rpc: iovcnt > 1");
+#endif
+ FSDBG_TOP(537, np, uio_offset(uio), uio_resid(uio), *iomodep);
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ nmwsize = nmp->nm_wsize;
+
+ wverfset = 0;
+ committed = NFS_WRITE_FILESYNC;
+
+ totalsize = tsiz = uio_resid(uio);
+ if ((nfsvers == NFS_VER2) && ((uint64_t)(uio_offset(uio) + tsiz) > 0xffffffffULL)) {
+ FSDBG_BOT(537, np, uio_offset(uio), uio_resid(uio), EFBIG);
+ return (EFBIG);
+ }
+
+ uio_save = uio_duplicate(uio);
+ if (uio_save == NULL) {
+ return (EIO);
+ }
+
+ while (tsiz > 0) {
+ len = (tsiz > nmwsize) ? nmwsize : tsiz;
+ FSDBG(537, np, uio_offset(uio), len, 0);
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ break;
+ }
+ if (nmp->nm_vers >= NFS_VER4)
+ stategenid = nmp->nm_stategenid;
+ error = nmp->nm_funcs->nf_write_rpc_async(np, uio, len, thd, cred, *iomodep, NULL, &req);
+ if (!error)
+ error = nmp->nm_funcs->nf_write_rpc_async_finish(np, req, &commit, &rlen, &wverf2);
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ error = ENXIO;
+ if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error) &&
+ (++restart <= nfs_mount_state_max_restarts(nmp))) { /* guard against no progress */
+ lck_mtx_lock(&nmp->nm_lock);
+ if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
+ NP(np, "nfs_write_rpc: error %d, initiating recovery", error);
+ nfs_need_recover(nmp, error);
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ } else {
+ if (error == NFSERR_GRACE)
+ tsleep(&nmp->nm_state, (PZERO-1), "nfsgrace", 2*hz);
+ if (!(error = nfs_mount_state_wait_for_recovery(nmp)))
+ continue;
+ }
+ }
if (error)
break;
+ if (nfsvers == NFS_VER2) {
+ tsiz -= len;
+ continue;
+ }
+
+ /* check for a short write */
+ if (rlen < len) {
+ /* Reset the uio to reflect the actual transfer */
+ *uio = *uio_save;
+ uio_update(uio, totalsize - (tsiz - rlen));
+ len = rlen;
+ }
+
+ /* return lowest commit level returned */
+ if (commit < committed)
+ committed = commit;
+
tsiz -= len;
+
+ /* check write verifier */
+ if (!wverfset) {
+ wverf = wverf2;
+ wverfset = 1;
+ } else if (wverf != wverf2) {
+ /* verifier changed, so we need to restart all the writes */
+ if (++vrestart > 100) {
+ /* give up after too many restarts */
+ error = EIO;
+ break;
+ }
+ *uio = *uio_save; // Reset the uio back to the start
+ committed = NFS_WRITE_FILESYNC;
+ wverfset = 0;
+ tsiz = totalsize;
+ }
}
-nfsmout:
- if ((mp = vnode_mount(vp)) && (vfs_flags(mp) & MNT_ASYNC))
- committed = NFSV3WRITE_FILESYNC;
- *iomode = committed;
+ if (uio_save)
+ uio_free(uio_save);
+ if (wverfset && wverfp)
+ *wverfp = wverf;
+ *iomodep = committed;
if (error)
- uio_uio_resid_set(uiop, tsiz);
- FSDBG_BOT(537, vp, committed, uio_uio_resid(uiop), error);
+ uio_setresid(uio, tsiz);
+ FSDBG_BOT(537, np, committed, uio_resid(uio), error);
+ return (error);
+}
+
+int
+nfs3_write_rpc_async(
+ nfsnode_t np,
+ uio_t uio,
+ size_t len,
+ thread_t thd,
+ kauth_cred_t cred,
+ int iomode,
+ struct nfsreq_cbinfo *cb,
+ struct nfsreq **reqp)
+{
+ struct nfsmount *nmp;
+ mount_t mp;
+ int error = 0, nfsvers;
+ struct nfsm_chain nmreq;
+
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+
+ /* for async mounts, don't bother sending sync write requests */
+ if ((iomode != NFS_WRITE_UNSTABLE) && nfs_allow_async &&
+ ((mp = NFSTOMP(np))) && (vfs_flags(mp) & MNT_ASYNC))
+ iomode = NFS_WRITE_UNSTABLE;
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
+ if (nfsvers == NFS_VER3) {
+ nfsm_chain_add_64(error, &nmreq, uio_offset(uio));
+ nfsm_chain_add_32(error, &nmreq, len);
+ nfsm_chain_add_32(error, &nmreq, iomode);
+ } else {
+ nfsm_chain_add_32(error, &nmreq, 0);
+ nfsm_chain_add_32(error, &nmreq, uio_offset(uio));
+ nfsm_chain_add_32(error, &nmreq, 0);
+ }
+ nfsm_chain_add_32(error, &nmreq, len);
+ nfsmout_if(error);
+ error = nfsm_chain_add_uio(&nmreq, uio, len);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request_async(np, NULL, &nmreq, NFSPROC_WRITE, thd, cred, NULL, 0, cb, reqp);
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ return (error);
+}
+
+int
+nfs3_write_rpc_async_finish(
+ nfsnode_t np,
+ struct nfsreq *req,
+ int *iomodep,
+ size_t *rlenp,
+ uint64_t *wverfp)
+{
+ struct nfsmount *nmp;
+ int error = 0, lockerror = ENOENT, nfsvers, status;
+ int updatemtime = 0, wccpostattr = 0, rlen, committed = NFS_WRITE_FILESYNC;
+ u_int64_t xid, wverf;
+ mount_t mp;
+ struct nfsm_chain nmrep;
+
+ nmp = NFSTONMP(np);
+ if (!nmp) {
+ nfs_request_async_cancel(req);
+ return (ENXIO);
+ }
+ nfsvers = nmp->nm_vers;
+
+ nfsm_chain_null(&nmrep);
+
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ if (error == EINPROGRESS) /* async request restarted */
+ return (error);
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ error = ENXIO;
+ if (!error && (lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ if (nfsvers == NFS_VER3) {
+ struct timespec premtime = { 0, 0 };
+ nfsm_chain_get_wcc_data(error, &nmrep, np, &premtime, &wccpostattr, &xid);
+ if (nfstimespeccmp(&np->n_mtime, &premtime, ==))
+ updatemtime = 1;
+ if (!error)
+ error = status;
+ nfsm_chain_get_32(error, &nmrep, rlen);
+ nfsmout_if(error);
+ *rlenp = rlen;
+ if (rlen <= 0)
+ error = NFSERR_IO;
+ nfsm_chain_get_32(error, &nmrep, committed);
+ nfsm_chain_get_64(error, &nmrep, wverf);
+ nfsmout_if(error);
+ if (wverfp)
+ *wverfp = wverf;
+ lck_mtx_lock(&nmp->nm_lock);
+ if (!(nmp->nm_state & NFSSTA_HASWRITEVERF)) {
+ nmp->nm_verf = wverf;
+ nmp->nm_state |= NFSSTA_HASWRITEVERF;
+ } else if (nmp->nm_verf != wverf) {
+ nmp->nm_verf = wverf;
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+ } else {
+ if (!error)
+ error = status;
+ nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid);
+ nfsmout_if(error);
+ }
+ if (updatemtime)
+ NFS_CHANGED_UPDATE(nfsvers, np, &np->n_vattr);
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfsm_chain_cleanup(&nmrep);
+ if ((committed != NFS_WRITE_FILESYNC) && nfs_allow_async &&
+ ((mp = NFSTOMP(np))) && (vfs_flags(mp) & MNT_ASYNC))
+ committed = NFS_WRITE_FILESYNC;
+ *iomodep = committed;
return (error);
}
/*
- * nfs mknod rpc
- * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
+ * NFS mknod vnode op
+ *
+ * For NFS v2 this is a kludge. Use a create RPC but with the IFMT bits of the
* mode set to specify the file type and the size field for rdev.
*/
-static int
-nfs_mknodrpc(
- vnode_t dvp,
- vnode_t *vpp,
- struct componentname *cnp,
- struct vnode_attr *vap,
- kauth_cred_t cred,
- proc_t p)
+int
+nfs3_vnop_mknod(
+ struct vnop_mknod_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_dvp;
+ vnode_t *a_vpp;
+ struct componentname *a_cnp;
+ struct vnode_attr *a_vap;
+ vfs_context_t a_context;
+ } */ *ap)
{
- register struct nfsv2_sattr *sp;
- register u_long *tl;
- register caddr_t cp;
- register long t1, t2;
- vnode_t newvp = (vnode_t)0;
- struct nfsnode *np = (struct nfsnode *)0;
+ vnode_t dvp = ap->a_dvp;
+ vnode_t *vpp = ap->a_vpp;
+ struct componentname *cnp = ap->a_cnp;
+ struct vnode_attr *vap = ap->a_vap;
+ vfs_context_t ctx = ap->a_context;
+ vnode_t newvp = NULL;
+ nfsnode_t np = NULL;
+ struct nfsmount *nmp;
+ nfsnode_t dnp = VTONFS(dvp);
struct nfs_vattr nvattr;
- char *cp2;
- caddr_t bpos, dpos;
- int error = 0, wccpostattr = 0, gotvp = 0;
+ fhandle_t fh;
+ int error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0;
struct timespec premtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- u_long rdev;
- u_int64_t xid;
- int v3 = NFS_ISV3(dvp);
- int gotuid, gotgid;
+ u_int32_t rdev;
+ u_int64_t xid = 0, dxid;
+ int nfsvers, gotuid, gotgid;
+ struct nfsm_chain nmreq, nmrep;
+ struct nfsreq rq, *req = &rq;
+
+ nmp = VTONMP(dvp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
if (!VATTR_IS_ACTIVE(vap, va_type))
return (EINVAL);
if (vap->va_type == VCHR || vap->va_type == VBLK) {
if (!VATTR_IS_ACTIVE(vap, va_rdev))
return (EINVAL);
- rdev = txdr_unsigned(vap->va_rdev);
+ rdev = vap->va_rdev;
} else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
rdev = 0xffffffff;
else {
return (ENOTSUP);
}
- nfsm_reqhead(NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
- nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
- if (error)
- return (error);
+ if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN))
+ return (ENAMETOOLONG);
+
+ nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
VATTR_SET_SUPPORTED(vap, va_mode);
VATTR_SET_SUPPORTED(vap, va_uid);
gotuid = VATTR_IS_ACTIVE(vap, va_uid);
gotgid = VATTR_IS_ACTIVE(vap, va_gid);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_MKNOD]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN, v3);
- if (v3) {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- *tl++ = vtonfsv3_type(vap->va_type);
- nfsm_v3sattr(vap);
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + 4 * NFSX_UNSIGNED +
+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(nfsvers));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
+ if (nfsvers == NFS_VER3) {
+ nfsm_chain_add_32(error, &nmreq, vtonfs_type(vap->va_type, nfsvers));
+ nfsm_chain_add_v3sattr(error, &nmreq, vap);
if (vap->va_type == VCHR || vap->va_type == VBLK) {
- nfsm_build(tl, u_long *, 2 * NFSX_UNSIGNED);
- *tl++ = txdr_unsigned(major(vap->va_rdev));
- *tl = txdr_unsigned(minor(vap->va_rdev));
+ nfsm_chain_add_32(error, &nmreq, major(vap->va_rdev));
+ nfsm_chain_add_32(error, &nmreq, minor(vap->va_rdev));
}
} else {
- struct timespec neg1time = { -1, -1 };
- nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
- sp->sa_mode = vtonfsv2_mode(vap->va_type,
- (VATTR_IS_ACTIVE(vap, va_mode) ? vap->va_mode : 0600));
- sp->sa_uid = gotuid ? (u_long)txdr_unsigned(vap->va_uid) : nfs_xdrneg1;
- sp->sa_gid = gotgid ? (u_long)txdr_unsigned(vap->va_gid) : nfs_xdrneg1;
- sp->sa_size = rdev;
- if (VATTR_IS_ACTIVE(vap, va_access_time)) {
- txdr_nfsv2time(&vap->va_access_time, &sp->sa_atime);
- } else {
- txdr_nfsv2time(&neg1time, &sp->sa_atime);
- }
- if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
- txdr_nfsv2time(&vap->va_modify_time, &sp->sa_mtime);
- } else {
- txdr_nfsv2time(&neg1time, &sp->sa_mtime);
- }
+ nfsm_chain_add_v2sattr(error, &nmreq, vap, rdev);
}
- nfsm_request(dvp, NFSPROC_MKNOD, p, cred, &xid);
+ nfsm_chain_build_done(error, &nmreq);
+ if (!error)
+ error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
+ nfsmout_if(error);
+
+ error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_MKNOD,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
+ if (!error)
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
/* XXX no EEXIST kludge here? */
- if (!error) {
- nfsm_mtofh(dvp, cnp, newvp, v3, &xid, gotvp);
- if (!gotvp) {
- error = nfs_lookitup(dvp, cnp->cn_nameptr,
- cnp->cn_namelen, cred, p, &np);
- if (!error)
- newvp = NFSTOV(np);
+ dxid = xid;
+ if (!error && !status) {
+ if (dnp->n_flag & NNEGNCENTRIES) {
+ dnp->n_flag &= ~NNEGNCENTRIES;
+ cache_purge_negatives(dvp);
}
+ error = nfsm_chain_get_fh_attr(&nmrep, dnp, ctx, nfsvers, &xid, &fh, &nvattr);
+ }
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
+ if (!error)
+ error = status;
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+
+ if (!lockerror) {
+ dnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
+ nfs_node_unlock(dnp);
+ /* nfs_getattr() will check changed and purge caches */
+ nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
}
- if (v3 && mrep)
- nfsm_wcc_data(dvp, &premtime, wccpostattr, &xid);
+
+ if (!error && fh.fh_len)
+ error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh.fh_data, fh.fh_len, &nvattr, &xid, rq.r_auth, NG_MAKEENTRY, &np);
+ if (!error && !np)
+ error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
+ if (!error && np)
+ newvp = NFSTOV(np);
+ if (!busyerror)
+ nfs_node_clear_busy(dnp);
+
if (!error && (gotuid || gotgid) &&
- (!newvp || nfs_getattrcache(newvp, &nvattr) ||
+ (!newvp || nfs_getattrcache(np, &nvattr, 0) ||
(gotuid && (nvattr.nva_uid != vap->va_uid)) ||
(gotgid && (nvattr.nva_gid != vap->va_gid)))) {
/* clear ID bits if server didn't use them (or we can't tell) */
VATTR_CLEAR_SUPPORTED(vap, va_uid);
VATTR_CLEAR_SUPPORTED(vap, va_gid);
}
- nfsm_reqdone;
if (error) {
- if (newvp)
+ if (newvp) {
+ nfs_node_unlock(np);
vnode_put(newvp);
+ }
} else {
*vpp = newvp;
+ nfs_node_unlock(np);
}
- VTONFS(dvp)->n_flag |= NMODIFIED;
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(dvp)->n_ncmtime, &premtime, ==))
- VTONFS(dvp)->n_ncmtime = VTONFS(dvp)->n_vattr.nva_mtime;
- if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(dvp));
- return (error);
-}
-
-/*
- * nfs mknod vop
- * just call nfs_mknodrpc() to do the work.
- */
-/* ARGSUSED */
-static int
-nfs_mknod(ap)
- struct vnop_mknod_args /* {
- struct vnodeop_desc *a_desc;
- vnode_t a_dvp;
- vnode_t *a_vpp;
- struct componentname *a_cnp;
- struct vnode_attr *a_vap;
- vfs_context_t a_context;
- } */ *ap;
-{
- int error;
-
- error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap,
- vfs_context_ucred(ap->a_context),
- vfs_context_proc(ap->a_context));
-
return (error);
}
-static u_long create_verf;
+static uint32_t create_verf;
/*
- * nfs file create call
+ * NFS file create call
*/
-static int
-nfs_create(ap)
+int
+nfs3_vnop_create(
struct vnop_create_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_dvp;
struct componentname *a_cnp;
struct vnode_attr *a_vap;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t dvp = ap->a_dvp;
struct vnode_attr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfs_vattr nvattr;
- struct nfsv2_sattr *sp;
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- struct nfsnode *np = (struct nfsnode *)0;
- vnode_t newvp = (vnode_t)0;
- caddr_t bpos, dpos, cp2;
- int error = 0, wccpostattr = 0, gotvp = 0, fmode = 0;
+ fhandle_t fh;
+ nfsnode_t np = NULL;
+ struct nfsmount *nmp;
+ nfsnode_t dnp = VTONFS(dvp);
+ vnode_t newvp = NULL;
+ int error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0, fmode = 0;
struct timespec premtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3 = NFS_ISV3(dvp);
- int gotuid, gotgid;
- u_int64_t xid;
- kauth_cred_t cred;
- proc_t p;
+ int nfsvers, gotuid, gotgid;
+ u_int64_t xid, dxid;
+ uint32_t val;
+ struct nfsm_chain nmreq, nmrep;
+ struct nfsreq rq, *req = &rq;
+ struct nfs_dulookup dul;
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ nmp = VTONMP(dvp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
- if (!VATTR_IS_ACTIVE(vap, va_type))
- return (EINVAL);
+ if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN))
+ return (ENAMETOOLONG);
- /*
- * Oops, not for me..
- */
- if (vap->va_type == VSOCK)
- return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap, cred, p));
+ nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
VATTR_SET_SUPPORTED(vap, va_mode);
VATTR_SET_SUPPORTED(vap, va_uid);
gotuid = VATTR_IS_ACTIVE(vap, va_uid);
gotgid = VATTR_IS_ACTIVE(vap, va_gid);
- if (vap->va_vaflags & VA_EXCLUSIVE)
+ if (vap->va_vaflags & VA_EXCLUSIVE) {
fmode |= O_EXCL;
-again:
- nfsm_reqhead(NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
- nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_CREATE]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN, v3);
- if (v3) {
- nfsm_build(tl, u_long *, NFSX_UNSIGNED);
- if (fmode & O_EXCL) {
- *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
- nfsm_build(tl, u_long *, NFSX_V3CREATEVERF);
- if (!TAILQ_EMPTY(&in_ifaddrhead))
- *tl++ = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr;
- else
- *tl++ = create_verf;
- *tl = ++create_verf;
- } else {
- *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
- nfsm_v3sattr(vap);
- }
- } else {
- struct timespec neg1time = { -1, -1 };
- nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
- sp->sa_mode = vtonfsv2_mode(vap->va_type,
- (VATTR_IS_ACTIVE(vap, va_mode) ? vap->va_mode : 0600));
- sp->sa_uid = gotuid ? (u_long)txdr_unsigned(vap->va_uid) : nfs_xdrneg1;
- sp->sa_gid = gotgid ? (u_long)txdr_unsigned(vap->va_gid) : nfs_xdrneg1;
- sp->sa_size = 0;
- if (VATTR_IS_ACTIVE(vap, va_access_time)) {
- txdr_nfsv2time(&vap->va_access_time, &sp->sa_atime);
- } else {
- txdr_nfsv2time(&neg1time, &sp->sa_atime);
- }
- if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
- txdr_nfsv2time(&vap->va_modify_time, &sp->sa_mtime);
+ if (!VATTR_IS_ACTIVE(vap, va_access_time) || !VATTR_IS_ACTIVE(vap, va_modify_time))
+ vap->va_vaflags |= VA_UTIMES_NULL;
+ }
+
+again:
+ error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
+ nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + 2 * NFSX_UNSIGNED +
+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(nfsvers));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
+ if (nfsvers == NFS_VER3) {
+ if (fmode & O_EXCL) {
+ nfsm_chain_add_32(error, &nmreq, NFS_CREATE_EXCLUSIVE);
+ lck_rw_lock_shared(in_ifaddr_rwlock);
+ if (!TAILQ_EMPTY(&in_ifaddrhead))
+ val = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr;
+ else
+ val = create_verf;
+ lck_rw_done(in_ifaddr_rwlock);
+ nfsm_chain_add_32(error, &nmreq, val);
+ ++create_verf;
+ nfsm_chain_add_32(error, &nmreq, create_verf);
} else {
- txdr_nfsv2time(&neg1time, &sp->sa_mtime);
+ nfsm_chain_add_32(error, &nmreq, NFS_CREATE_UNCHECKED);
+ nfsm_chain_add_v3sattr(error, &nmreq, vap);
}
+ } else {
+ nfsm_chain_add_v2sattr(error, &nmreq, vap, 0);
}
- nfsm_request(dvp, NFSPROC_CREATE, p, cred, &xid);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+
+ error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_CREATE,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
if (!error) {
- nfsm_mtofh(dvp, cnp, newvp, v3, &xid, gotvp);
- if (!gotvp) {
- error = nfs_lookitup(dvp, cnp->cn_nameptr,
- cnp->cn_namelen, cred, p, &np);
- if (!error)
- newvp = NFSTOV(np);
+ nfs_dulookup_start(&dul, dnp, ctx);
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ }
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+ dxid = xid;
+ if (!error && !status) {
+ if (dnp->n_flag & NNEGNCENTRIES) {
+ dnp->n_flag &= ~NNEGNCENTRIES;
+ cache_purge_negatives(dvp);
}
+ error = nfsm_chain_get_fh_attr(&nmrep, dnp, ctx, nfsvers, &xid, &fh, &nvattr);
}
- if (v3 && mrep)
- nfsm_wcc_data(dvp, &premtime, wccpostattr, &xid);
- nfsm_reqdone;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
+ if (!error)
+ error = status;
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+
+ if (!lockerror) {
+ dnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
+ nfs_node_unlock(dnp);
+ /* nfs_getattr() will check changed and purge caches */
+ nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
+ }
+
+ if (!error && fh.fh_len)
+ error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh.fh_data, fh.fh_len, &nvattr, &xid, rq.r_auth, NG_MAKEENTRY, &np);
+ if (!error && !np)
+ error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
+ if (!error && np)
+ newvp = NFSTOV(np);
+
+ nfs_dulookup_finish(&dul, dnp, ctx);
+ if (!busyerror)
+ nfs_node_clear_busy(dnp);
+
if (error) {
- if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
+ if ((nfsvers == NFS_VER3) && (fmode & O_EXCL) && (error == NFSERR_NOTSUPP)) {
fmode &= ~O_EXCL;
goto again;
}
- if (newvp)
+ if (newvp) {
+ nfs_node_unlock(np);
vnode_put(newvp);
- } else if (v3 && (fmode & O_EXCL)) {
- error = nfs_setattrrpc(newvp, vap, cred, p);
+ }
+ } else if ((nfsvers == NFS_VER3) && (fmode & O_EXCL)) {
+ nfs_node_unlock(np);
+ error = nfs3_setattr_rpc(np, vap, ctx);
if (error && (gotuid || gotgid)) {
/* it's possible the server didn't like our attempt to set IDs. */
/* so, let's try it again without those */
VATTR_CLEAR_ACTIVE(vap, va_uid);
VATTR_CLEAR_ACTIVE(vap, va_gid);
- error = nfs_setattrrpc(newvp, vap, cred, p);
+ error = nfs3_setattr_rpc(np, vap, ctx);
}
if (error)
vnode_put(newvp);
+ else
+ nfs_node_lock_force(np);
}
- if (!error) {
+ if (!error)
*ap->a_vpp = newvp;
- }
- VTONFS(dvp)->n_flag |= NMODIFIED;
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(dvp)->n_ncmtime, &premtime, ==))
- VTONFS(dvp)->n_ncmtime = VTONFS(dvp)->n_vattr.nva_mtime;
- if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(dvp));
if (!error && (gotuid || gotgid) &&
- (!newvp || nfs_getattrcache(newvp, &nvattr) ||
+ (!newvp || nfs_getattrcache(np, &nvattr, 0) ||
(gotuid && (nvattr.nva_uid != vap->va_uid)) ||
(gotgid && (nvattr.nva_gid != vap->va_gid)))) {
/* clear ID bits if server didn't use them (or we can't tell) */
VATTR_CLEAR_SUPPORTED(vap, va_uid);
VATTR_CLEAR_SUPPORTED(vap, va_gid);
}
+ if (!error)
+ nfs_node_unlock(np);
return (error);
}
/*
- * nfs file remove call
- * To try and make nfs semantics closer to ufs semantics, a file that has
+ * NFS file remove call
+ * To try and make NFS semantics closer to UFS semantics, a file that has
* other processes using the vnode is renamed instead of removed and then
* removed later on the last close.
* - If vnode_isinuse()
* If a rename is not already in the works
* call nfs_sillyrename() to set it up
* else
- * do the remove rpc
+ * do the remove RPC
*/
-static int
-nfs_remove(ap)
+int
+nfs_vnop_remove(
struct vnop_remove_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_dvp;
struct componentname *a_cnp;
int a_flags;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t vp = ap->a_vp;
vnode_t dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
- struct nfsnode *np = VTONFS(vp);
- int error = 0, gofree = 0;
+ nfsnode_t dnp = VTONFS(dvp);
+ nfsnode_t np = VTONFS(vp);
+ int error = 0, nfsvers, namedattrs, inuse, gotattr = 0, flushed = 0, setsize = 0;
struct nfs_vattr nvattr;
- kauth_cred_t cred;
- proc_t p;
+ struct nfsmount *nmp;
+ struct nfs_dulookup dul;
+
+ /* XXX prevent removing a sillyrenamed file? */
+
+ nmp = NFSTONMP(dnp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
+
+again_relock:
+ error = nfs_node_set_busy2(dnp, np, vfs_context_thread(ctx));
+ if (error)
+ return (error);
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ /* lock the node while we remove the file */
+ lck_mtx_lock(nfs_node_hash_mutex);
+ while (np->n_hflag & NHLOCKED) {
+ np->n_hflag |= NHLOCKWANT;
+ msleep(np, nfs_node_hash_mutex, PINOD, "nfs_remove", NULL);
+ }
+ np->n_hflag |= NHLOCKED;
+ lck_mtx_unlock(nfs_node_hash_mutex);
- gofree = vnode_isinuse(vp, 0) ? 0 : 1;
- if ((ap->a_flags & VNODE_REMOVE_NODELETEBUSY) && !gofree) {
+ if (!namedattrs)
+ nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
+again:
+ inuse = vnode_isinuse(vp, 0);
+ if ((ap->a_flags & VNODE_REMOVE_NODELETEBUSY) && inuse) {
/* Caller requested Carbon delete semantics, but file is busy */
- return (EBUSY);
+ error = EBUSY;
+ goto out;
}
- if (gofree || (np->n_sillyrename &&
- nfs_getattr(vp, &nvattr, cred, p) == 0 &&
- nvattr.nva_nlink > 1)) {
+ if (inuse && !gotattr) {
+ if (nfs_getattr(np, &nvattr, ctx, NGA_CACHED))
+ nvattr.nva_nlink = 1;
+ gotattr = 1;
+ goto again;
+ }
+ if (!inuse || (np->n_sillyrename && (nvattr.nva_nlink > 1))) {
+
+ if (!inuse && !flushed) { /* flush all the buffers first */
+ /* unlock the node */
+ lck_mtx_lock(nfs_node_hash_mutex);
+ np->n_hflag &= ~NHLOCKED;
+ if (np->n_hflag & NHLOCKWANT) {
+ np->n_hflag &= ~NHLOCKWANT;
+ wakeup(np);
+ }
+ lck_mtx_unlock(nfs_node_hash_mutex);
+ nfs_node_clear_busy2(dnp, np);
+ error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
+ FSDBG(260, np, np->n_size, np->n_vattr.nva_size, 0xf00d0011);
+ flushed = 1;
+ if (error == EINTR) {
+ nfs_node_lock_force(np);
+ NATTRINVALIDATE(np);
+ nfs_node_unlock(np);
+ return (error);
+ }
+ if (!namedattrs)
+ nfs_dulookup_finish(&dul, dnp, ctx);
+ goto again_relock;
+ }
+
+ if ((nmp->nm_vers >= NFS_VER4) && (np->n_openflags & N_DELEG_MASK))
+ nfs4_delegation_return(np, 0, vfs_context_thread(ctx), vfs_context_ucred(ctx));
+
/*
* Purge the name cache so that the chance of a lookup for
* the name succeeding while the remove is in progress is
* minimized.
*/
- cache_purge(vp);
- /*
- * throw away biocache buffers, mainly to avoid
- * unnecessary delayed writes later.
- */
- error = nfs_vinvalbuf(vp, 0, cred, p, 1);
- np->n_size = 0;
- ubc_setsize(vp, (off_t)0); /* XXX check error */
+ nfs_name_cache_purge(dnp, np, cnp, ctx);
+
+ if (!namedattrs)
+ nfs_dulookup_start(&dul, dnp, ctx);
+
/* Do the rpc */
- if (error != EINTR)
- error = nfs_removerpc(dvp, cnp->cn_nameptr,
- cnp->cn_namelen, cred, p);
+ error = nmp->nm_funcs->nf_remove_rpc(dnp, cnp->cn_nameptr, cnp->cn_namelen,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx));
+
/*
* Kludge City: If the first reply to the remove rpc is lost..
* the reply to the retransmitted request will be ENOENT
*/
if (error == ENOENT)
error = 0;
- if (!error) {
+
+ if (!error && !inuse && !np->n_sillyrename) {
/*
+ * removal succeeded, it's not in use, and not silly renamed so
* remove nfsnode from hash now so we can't accidentally find it
* again if another object gets created with the same filehandle
* before this vnode gets reclaimed
*/
lck_mtx_lock(nfs_node_hash_mutex);
- LIST_REMOVE(np, n_hash);
- np->n_flag &= ~NHASHED;
+ if (np->n_hflag & NHHASHED) {
+ LIST_REMOVE(np, n_hash);
+ np->n_hflag &= ~NHHASHED;
+ FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
+ }
lck_mtx_unlock(nfs_node_hash_mutex);
- }
- if (!error && !np->n_sillyrename) {
- /* clear flags now: won't get nfs_inactive for recycled vnode */
+ /* clear flags now: won't get nfs_vnop_inactive for recycled vnode */
/* clear all flags other than these */
- np->n_flag &= (NMODIFIED | NFLUSHINPROG | NFLUSHWANT | NHASHED);
+ nfs_node_lock_force(np);
+ np->n_flag &= (NMODIFIED);
+ NATTRINVALIDATE(np);
+ nfs_node_unlock(np);
vnode_recycle(vp);
+ setsize = 1;
+ } else {
+ nfs_node_lock_force(np);
+ NATTRINVALIDATE(np);
+ nfs_node_unlock(np);
}
} else if (!np->n_sillyrename) {
- error = nfs_sillyrename(dvp, vp, cnp, cred, p);
+ if (!namedattrs)
+ nfs_dulookup_start(&dul, dnp, ctx);
+ error = nfs_sillyrename(dnp, np, cnp, ctx);
+ nfs_node_lock_force(np);
+ NATTRINVALIDATE(np);
+ nfs_node_unlock(np);
+ } else {
+ nfs_node_lock_force(np);
+ NATTRINVALIDATE(np);
+ nfs_node_unlock(np);
+ if (!namedattrs)
+ nfs_dulookup_start(&dul, dnp, ctx);
}
- NATTRINVALIDATE(np);
+ /* nfs_getattr() will check changed and purge caches */
+ nfs_getattr(dnp, NULL, ctx, NGA_CACHED);
+ if (!namedattrs)
+ nfs_dulookup_finish(&dul, dnp, ctx);
+out:
+ /* unlock the node */
+ lck_mtx_lock(nfs_node_hash_mutex);
+ np->n_hflag &= ~NHLOCKED;
+ if (np->n_hflag & NHLOCKWANT) {
+ np->n_hflag &= ~NHLOCKWANT;
+ wakeup(np);
+ }
+ lck_mtx_unlock(nfs_node_hash_mutex);
+ nfs_node_clear_busy2(dnp, np);
+ if (setsize)
+ ubc_setsize(vp, 0);
return (error);
}
/*
- * nfs file remove rpc called from nfs_inactive
+ * NFS silly-renamed file removal function called from nfs_vnop_inactive
*/
int
-nfs_removeit(struct sillyrename *sp)
+nfs_removeit(struct nfs_sillyrename *nsp)
{
- return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, NULL));
+ struct nfsmount *nmp = NFSTONMP(nsp->nsr_dnp);
+ if (!nmp)
+ return (ENXIO);
+ return nmp->nm_funcs->nf_remove_rpc(nsp->nsr_dnp, nsp->nsr_name, nsp->nsr_namlen, NULL, nsp->nsr_cred);
}
/*
- * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
+ * NFS remove rpc, called from nfs_remove() and nfs_removeit().
*/
-static int
-nfs_removerpc(dvp, name, namelen, cred, proc)
- vnode_t dvp;
- char *name;
- int namelen;
- kauth_cred_t cred;
- proc_t proc;
+int
+nfs3_remove_rpc(
+ nfsnode_t dnp,
+ char *name,
+ int namelen,
+ thread_t thd,
+ kauth_cred_t cred)
{
- register u_long *tl;
- register caddr_t cp;
- register long t1, t2;
- caddr_t bpos, dpos, cp2;
- int error = 0, wccpostattr = 0;
+ int error = 0, lockerror = ENOENT, status, wccpostattr = 0;
struct timespec premtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3;
+ struct nfsmount *nmp;
+ int nfsvers;
u_int64_t xid;
+ struct nfsm_chain nmreq, nmrep;
- if (!VFSTONFS(vnode_mount(dvp)))
+ nmp = NFSTONMP(dnp);
+ if (!nmp)
return (ENXIO);
- v3 = NFS_ISV3(dvp);
-
- nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_REMOVE]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(name, namelen, NFS_MAXNAMLEN, v3);
- nfsm_request(dvp, NFSPROC_REMOVE, proc, cred, &xid);
- if (v3 && mrep)
- nfsm_wcc_data(dvp, &premtime, wccpostattr, &xid);
- nfsm_reqdone;
- VTONFS(dvp)->n_flag |= NMODIFIED;
+ nfsvers = nmp->nm_vers;
+ if ((nfsvers == NFS_VER2) && (namelen > NFS_MAXNAMLEN))
+ return (ENAMETOOLONG);
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + NFSX_UNSIGNED + nfsm_rndup(namelen));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, name, namelen, nmp);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+
+ error = nfs_request2(dnp, NULL, &nmreq, NFSPROC_REMOVE, thd, cred, NULL, 0, &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &xid);
+ nfsmout_if(error);
+ dnp->n_flag |= NMODIFIED;
/* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(dvp)->n_ncmtime, &premtime, ==))
- VTONFS(dvp)->n_ncmtime = VTONFS(dvp)->n_vattr.nva_mtime;
+ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(dvp));
+ NATTRINVALIDATE(dnp);
+ if (!error)
+ error = status;
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(dnp);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
return (error);
}
/*
- * nfs file rename call
+ * NFS file rename call
*/
-static int
-nfs_rename(ap)
+int
+nfs_vnop_rename(
struct vnop_rename_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_fdvp;
vnode_t a_tvp;
struct componentname *a_tcnp;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
- vnode_t fvp = ap->a_fvp;
- vnode_t tvp = ap->a_tvp;
+ vfs_context_t ctx = ap->a_context;
vnode_t fdvp = ap->a_fdvp;
+ vnode_t fvp = ap->a_fvp;
vnode_t tdvp = ap->a_tdvp;
+ vnode_t tvp = ap->a_tvp;
+ nfsnode_t fdnp, fnp, tdnp, tnp;
struct componentname *tcnp = ap->a_tcnp;
struct componentname *fcnp = ap->a_fcnp;
- int error, inuse=0;
+ int error, nfsvers, inuse=0, tvprecycle=0, locked=0;
mount_t fmp, tdmp, tmp;
- struct nfsnode *tnp;
- kauth_cred_t cred;
- proc_t p;
-
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ struct nfs_vattr nvattr;
+ struct nfsmount *nmp;
+ fdnp = VTONFS(fdvp);
+ fnp = VTONFS(fvp);
+ tdnp = VTONFS(tdvp);
tnp = tvp ? VTONFS(tvp) : NULL;
+ nmp = NFSTONMP(fdnp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+
+ error = nfs_node_set_busy4(fdnp, fnp, tdnp, tnp, vfs_context_thread(ctx));
+ if (error)
+ return (error);
+
+ if (tvp && (tvp != fvp)) {
+ /* lock the node while we rename over the existing file */
+ lck_mtx_lock(nfs_node_hash_mutex);
+ while (tnp->n_hflag & NHLOCKED) {
+ tnp->n_hflag |= NHLOCKWANT;
+ msleep(tnp, nfs_node_hash_mutex, PINOD, "nfs_rename", NULL);
+ }
+ tnp->n_hflag |= NHLOCKED;
+ lck_mtx_unlock(nfs_node_hash_mutex);
+ locked = 1;
+ }
+
/* Check for cross-device rename */
fmp = vnode_mount(fvp);
tmp = tvp ? vnode_mount(tvp) : NULL;
goto out;
}
+ /* XXX prevent renaming from/over a sillyrenamed file? */
+
/*
* If the tvp exists and is in use, sillyrename it before doing the
* rename of the new file over it.
* Don't sillyrename if source and target are same vnode (hard
* links or case-variants)
*/
- if (tvp && tvp != fvp) {
+ if (tvp && (tvp != fvp))
inuse = vnode_isinuse(tvp, 0);
- }
- if (inuse && !tnp->n_sillyrename && vnode_vtype(tvp) != VDIR) {
- if ((error = nfs_sillyrename(tdvp, tvp, tcnp, cred, p))) {
+ if (inuse && !tnp->n_sillyrename && (vnode_vtype(tvp) != VDIR)) {
+ error = nfs_sillyrename(tdnp, tnp, tcnp, ctx);
+ if (error) {
/* sillyrename failed. Instead of pressing on, return error */
goto out; /* should not be ENOENT. */
} else {
/* sillyrename succeeded.*/
tvp = NULL;
}
+ } else if (tvp && (nmp->nm_vers >= NFS_VER4) && (tnp->n_openflags & N_DELEG_MASK)) {
+ nfs4_delegation_return(tnp, 0, vfs_context_thread(ctx), vfs_context_ucred(ctx));
}
- error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
- tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, cred, p);
+ error = nmp->nm_funcs->nf_rename_rpc(fdnp, fcnp->cn_nameptr, fcnp->cn_namelen,
+ tdnp, tcnp->cn_nameptr, tcnp->cn_namelen, ctx);
/*
* Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
if (error == ENOENT)
error = 0;
- if (!error && tvp && tvp != fvp && !tnp->n_sillyrename) {
- /*
- * remove nfsnode from hash now so we can't accidentally find it
- * again if another object gets created with the same filehandle
- * before this vnode gets reclaimed
- */
+ if (tvp && (tvp != fvp) && !tnp->n_sillyrename) {
+ nfs_node_lock_force(tnp);
+ tvprecycle = (!error && !vnode_isinuse(tvp, 0) &&
+ (nfs_getattrcache(tnp, &nvattr, 0) || (nvattr.nva_nlink == 1)));
+ nfs_node_unlock(tnp);
lck_mtx_lock(nfs_node_hash_mutex);
- LIST_REMOVE(tnp, n_hash);
- tnp->n_flag &= ~NHASHED;
+ if (tvprecycle && (tnp->n_hflag & NHHASHED)) {
+ /*
+ * remove nfsnode from hash now so we can't accidentally find it
+ * again if another object gets created with the same filehandle
+ * before this vnode gets reclaimed
+ */
+ LIST_REMOVE(tnp, n_hash);
+ tnp->n_hflag &= ~NHHASHED;
+ FSDBG(266, 0, tnp, tnp->n_flag, 0xb1eb1e);
+ }
lck_mtx_unlock(nfs_node_hash_mutex);
}
-
+
/* purge the old name cache entries and enter the new one */
- cache_purge(fvp);
+ nfs_name_cache_purge(fdnp, fnp, fcnp, ctx);
if (tvp) {
- cache_purge(tvp);
- if (!error && !tnp->n_sillyrename) {
- /* clear flags now: won't get nfs_inactive for recycled vnode */
+ nfs_name_cache_purge(tdnp, tnp, tcnp, ctx);
+ if (tvprecycle) {
+ /* clear flags now: won't get nfs_vnop_inactive for recycled vnode */
/* clear all flags other than these */
- tnp->n_flag &= (NMODIFIED | NFLUSHINPROG | NFLUSHWANT | NHASHED);
+ nfs_node_lock_force(tnp);
+ tnp->n_flag &= (NMODIFIED);
+ nfs_node_unlock(tnp);
vnode_recycle(tvp);
}
}
- if (!error)
+ if (!error) {
+ nfs_node_lock_force(tdnp);
+ if (tdnp->n_flag & NNEGNCENTRIES) {
+ tdnp->n_flag &= ~NNEGNCENTRIES;
+ cache_purge_negatives(tdvp);
+ }
+ nfs_node_unlock(tdnp);
+ nfs_node_lock_force(fnp);
cache_enter(tdvp, fvp, tcnp);
-
+ if (tdvp != fdvp) { /* update parent pointer */
+ if (fnp->n_parent && !vnode_get(fnp->n_parent)) {
+ /* remove ref from old parent */
+ vnode_rele(fnp->n_parent);
+ vnode_put(fnp->n_parent);
+ }
+ fnp->n_parent = tdvp;
+ if (tdvp && !vnode_get(tdvp)) {
+ /* add ref to new parent */
+ vnode_ref(tdvp);
+ vnode_put(tdvp);
+ } else {
+ fnp->n_parent = NULL;
+ }
+ }
+ nfs_node_unlock(fnp);
+ }
out:
- /*
- * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
- */
- if (error == ENOENT)
- error = 0;
+ /* nfs_getattr() will check changed and purge caches */
+ nfs_getattr(fdnp, NULL, ctx, NGA_CACHED);
+ nfs_getattr(tdnp, NULL, ctx, NGA_CACHED);
+ if (locked) {
+ /* unlock node */
+ lck_mtx_lock(nfs_node_hash_mutex);
+ tnp->n_hflag &= ~NHLOCKED;
+ if (tnp->n_hflag & NHLOCKWANT) {
+ tnp->n_hflag &= ~NHLOCKWANT;
+ wakeup(tnp);
+ }
+ lck_mtx_unlock(nfs_node_hash_mutex);
+ }
+ nfs_node_clear_busy4(fdnp, fnp, tdnp, tnp);
return (error);
}
/*
- * Do an nfs rename rpc. Called from nfs_rename() and nfs_sillyrename().
+ * Do an NFS rename rpc. Called from nfs_vnop_rename() and nfs_sillyrename().
*/
-static int
-nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
- vnode_t fdvp;
- char *fnameptr;
- int fnamelen;
- vnode_t tdvp;
- char *tnameptr;
- int tnamelen;
- kauth_cred_t cred;
- proc_t proc;
+int
+nfs3_rename_rpc(
+ nfsnode_t fdnp,
+ char *fnameptr,
+ int fnamelen,
+ nfsnode_t tdnp,
+ char *tnameptr,
+ int tnamelen,
+ vfs_context_t ctx)
{
- register u_long *tl;
- register caddr_t cp;
- register long t1, t2;
- caddr_t bpos, dpos, cp2;
- int error = 0, fwccpostattr = 0, twccpostattr = 0;
+ int error = 0, lockerror = ENOENT, status, fwccpostattr = 0, twccpostattr = 0;
struct timespec fpremtime = { 0, 0 }, tpremtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3;
- u_int64_t xid;
+ struct nfsmount *nmp;
+ int nfsvers;
+ u_int64_t xid, txid;
+ struct nfsm_chain nmreq, nmrep;
- if (!VFSTONFS(vnode_mount(fdvp)))
+ nmp = NFSTONMP(fdnp);
+ if (!nmp)
return (ENXIO);
- v3 = NFS_ISV3(fdvp);
-
- nfsm_reqhead((NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
- nfsm_rndup(tnamelen));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_RENAME]);
- nfsm_fhtom(fdvp, v3);
- nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN, v3);
- nfsm_fhtom(tdvp, v3);
- nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN, v3);
- nfsm_request(fdvp, NFSPROC_RENAME, proc, cred, &xid);
- if (v3 && mrep) {
- u_int64_t txid = xid;
-
- nfsm_wcc_data(fdvp, &fpremtime, fwccpostattr, &xid);
- nfsm_wcc_data(tdvp, &tpremtime, twccpostattr, &txid);
- }
- nfsm_reqdone;
- VTONFS(fdvp)->n_flag |= NMODIFIED;
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(fdvp)->n_ncmtime, &fpremtime, ==))
- VTONFS(fdvp)->n_ncmtime = VTONFS(fdvp)->n_vattr.nva_mtime;
- if (!fwccpostattr)
- NATTRINVALIDATE(VTONFS(fdvp));
- VTONFS(tdvp)->n_flag |= NMODIFIED;
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(tdvp)->n_ncmtime, &tpremtime, ==))
- VTONFS(tdvp)->n_ncmtime = VTONFS(tdvp)->n_vattr.nva_mtime;
- if (!twccpostattr)
- NATTRINVALIDATE(VTONFS(tdvp));
+ nfsvers = nmp->nm_vers;
+ if ((nfsvers == NFS_VER2) &&
+ ((fnamelen > NFS_MAXNAMLEN) || (tnamelen > NFS_MAXNAMLEN)))
+ return (ENAMETOOLONG);
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ (NFSX_FH(nfsvers) + NFSX_UNSIGNED) * 2 +
+ nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, fdnp->n_fhp, fdnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, fnameptr, fnamelen, nmp);
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, tdnp->n_fhp, tdnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, tnameptr, tnamelen, nmp);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+
+ error = nfs_request(fdnp, NULL, &nmreq, NFSPROC_RENAME, ctx, NULL, &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock2(fdnp, tdnp)))
+ error = lockerror;
+ if (nfsvers == NFS_VER3) {
+ txid = xid;
+ nfsm_chain_get_wcc_data(error, &nmrep, fdnp, &fpremtime, &fwccpostattr, &xid);
+ nfsm_chain_get_wcc_data(error, &nmrep, tdnp, &tpremtime, &twccpostattr, &txid);
+ }
+ if (!error)
+ error = status;
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ if (!lockerror) {
+ fdnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&fdnp->n_ncmtime, &fpremtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, fdnp, &fdnp->n_vattr);
+ if (!fwccpostattr)
+ NATTRINVALIDATE(fdnp);
+ tdnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&tdnp->n_ncmtime, &tpremtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, tdnp, &tdnp->n_vattr);
+ if (!twccpostattr)
+ NATTRINVALIDATE(tdnp);
+ nfs_node_unlock2(fdnp, tdnp);
+ }
return (error);
}
/*
- * nfs hard link create call
+ * NFS hard link create call
*/
-static int
-nfs_link(ap)
+int
+nfs3_vnop_link(
struct vnop_link_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
vnode_t a_tdvp;
struct componentname *a_cnp;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t vp = ap->a_vp;
vnode_t tdvp = ap->a_tdvp;
struct componentname *cnp = ap->a_cnp;
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- caddr_t bpos, dpos, cp2;
- int error = 0, wccpostattr = 0, attrflag = 0;
+ int error = 0, lockerror = ENOENT, status, wccpostattr = 0, attrflag = 0;
struct timespec premtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3;
- u_int64_t xid;
- kauth_cred_t cred;
- proc_t p;
+ struct nfsmount *nmp;
+ nfsnode_t np = VTONFS(vp);
+ nfsnode_t tdnp = VTONFS(tdvp);
+ int nfsvers;
+ u_int64_t xid, txid;
+ struct nfsm_chain nmreq, nmrep;
- if (vnode_mount(vp) != vnode_mount(tdvp)) {
+ if (vnode_mount(vp) != vnode_mount(tdvp))
return (EXDEV);
- }
-
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
- v3 = NFS_ISV3(vp);
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN))
+ return (ENAMETOOLONG);
/*
* Push all writes to the server, so that the attribute cache
* doesn't get "out of sync" with the server.
* XXX There should be a better way!
*/
- nfs_flush(vp, MNT_WAIT, cred, p, 0);
+ nfs_flush(np, MNT_WAIT, vfs_context_thread(ctx), V_IGNORE_WRITEERR);
- nfsm_reqhead(NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
+ error = nfs_node_set_busy2(tdnp, np, vfs_context_thread(ctx));
if (error)
return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_LINK]);
- nfsm_fhtom(vp, v3);
- nfsm_fhtom(tdvp, v3);
- nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN, v3);
- nfsm_request(vp, NFSPROC_LINK, p, cred, &xid);
- if (v3 && mrep) {
- u_int64_t txid = xid;
-
- nfsm_postop_attr_update(vp, v3, attrflag, &xid);
- nfsm_wcc_data(tdvp, &premtime, wccpostattr, &txid);
- }
- nfsm_reqdone;
-
- VTONFS(tdvp)->n_flag |= NMODIFIED;
- if (!attrflag)
- NATTRINVALIDATE(VTONFS(vp));
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(tdvp)->n_ncmtime, &premtime, ==))
- VTONFS(tdvp)->n_ncmtime = VTONFS(tdvp)->n_vattr.nva_mtime;
- if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(tdvp));
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, tdnp->n_fhp, tdnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request(np, NULL, &nmreq, NFSPROC_LINK, ctx, NULL, &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock2(tdnp, np))) {
+ error = lockerror;
+ goto nfsmout;
+ }
+ if (nfsvers == NFS_VER3) {
+ txid = xid;
+ nfsm_chain_postop_attr_update_flag(error, &nmrep, np, attrflag, &xid);
+ nfsm_chain_get_wcc_data(error, &nmrep, tdnp, &premtime, &wccpostattr, &txid);
+ }
+ if (!error)
+ error = status;
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ if (!lockerror) {
+ if (!attrflag)
+ NATTRINVALIDATE(np);
+ tdnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&tdnp->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, tdnp, &tdnp->n_vattr);
+ if (!wccpostattr)
+ NATTRINVALIDATE(tdnp);
+ if (!error && (tdnp->n_flag & NNEGNCENTRIES)) {
+ tdnp->n_flag &= ~NNEGNCENTRIES;
+ cache_purge_negatives(tdvp);
+ }
+ nfs_node_unlock2(tdnp, np);
+ }
+ nfs_node_clear_busy2(tdnp, np);
/*
* Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
*/
}
/*
- * nfs symbolic link create call
+ * NFS symbolic link create call
*/
-static int
-nfs_symlink(ap)
+int
+nfs3_vnop_symlink(
struct vnop_symlink_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_dvp;
struct vnode_attr *a_vap;
char *a_target;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t dvp = ap->a_dvp;
struct vnode_attr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfs_vattr nvattr;
- struct nfsv2_sattr *sp;
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- caddr_t bpos, dpos, cp2;
- int slen, error = 0, wccpostattr = 0, gotvp = 0;
+ fhandle_t fh;
+ int slen, error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0;
struct timespec premtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- vnode_t newvp = (vnode_t)0;
- int v3 = NFS_ISV3(dvp);
- int gotuid, gotgid;
- u_int64_t xid;
- kauth_cred_t cred;
- proc_t p;
- struct nfsnode *np = NULL;
+ vnode_t newvp = NULL;
+ int nfsvers, gotuid, gotgid;
+ u_int64_t xid = 0, dxid;
+ nfsnode_t np = NULL;
+ nfsnode_t dnp = VTONFS(dvp);
+ struct nfsmount *nmp;
+ struct nfsm_chain nmreq, nmrep;
+ struct nfsreq rq, *req = &rq;
+ struct nfs_dulookup dul;
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ nmp = VTONMP(dvp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
slen = strlen(ap->a_target);
- nfsm_reqhead(NFSX_FH(v3) + 2*NFSX_UNSIGNED +
- nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
- if (error)
- return (error);
+ if ((nfsvers == NFS_VER2) &&
+ ((cnp->cn_namelen > NFS_MAXNAMLEN) || (slen > NFS_MAXPATHLEN)))
+ return (ENAMETOOLONG);
+
+ nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
VATTR_SET_SUPPORTED(vap, va_mode);
VATTR_SET_SUPPORTED(vap, va_uid);
gotuid = VATTR_IS_ACTIVE(vap, va_uid);
gotgid = VATTR_IS_ACTIVE(vap, va_gid);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_SYMLINK]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN, v3);
- if (v3) {
- nfsm_v3sattr(vap);
- }
- nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN, v3);
- if (!v3) {
- struct timespec neg1time = { -1, -1 };
- nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
- sp->sa_mode = vtonfsv2_mode(VLNK,
- (VATTR_IS_ACTIVE(vap, va_mode) ? vap->va_mode : 0600));
- sp->sa_uid = gotuid ? (u_long)txdr_unsigned(vap->va_uid) : nfs_xdrneg1;
- sp->sa_gid = gotgid ? (u_long)txdr_unsigned(vap->va_gid) : nfs_xdrneg1;
- sp->sa_size = nfs_xdrneg1;
- if (VATTR_IS_ACTIVE(vap, va_access_time)) {
- txdr_nfsv2time(&vap->va_access_time, &sp->sa_atime);
- } else {
- txdr_nfsv2time(&neg1time, &sp->sa_atime);
- }
- if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
- txdr_nfsv2time(&vap->va_modify_time, &sp->sa_mtime);
- } else {
- txdr_nfsv2time(&neg1time, &sp->sa_mtime);
+ error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
+ nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + 2 * NFSX_UNSIGNED +
+ nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(nfsvers));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_add_v3sattr(error, &nmreq, vap);
+ nfsm_chain_add_name(error, &nmreq, ap->a_target, slen, nmp);
+ if (nfsvers == NFS_VER2)
+ nfsm_chain_add_v2sattr(error, &nmreq, vap, -1);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+
+ error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_SYMLINK,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
+ if (!error) {
+ nfs_dulookup_start(&dul, dnp, ctx);
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ }
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+ dxid = xid;
+ if (!error && !status) {
+ if (dnp->n_flag & NNEGNCENTRIES) {
+ dnp->n_flag &= ~NNEGNCENTRIES;
+ cache_purge_negatives(dvp);
}
+ if (nfsvers == NFS_VER3)
+ error = nfsm_chain_get_fh_attr(&nmrep, dnp, ctx, nfsvers, &xid, &fh, &nvattr);
+ else
+ fh.fh_len = 0;
}
- nfsm_request(dvp, NFSPROC_SYMLINK, p, cred, &xid);
- if (v3 && mrep) {
- u_int64_t dxid = xid;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
+ if (!error)
+ error = status;
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
- if (!error)
- nfsm_mtofh(dvp, cnp, newvp, v3, &xid, gotvp);
- nfsm_wcc_data(dvp, &premtime, wccpostattr, &dxid);
+ if (!lockerror) {
+ dnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
+ nfs_node_unlock(dnp);
+ /* nfs_getattr() will check changed and purge caches */
+ nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
}
- nfsm_reqdone;
- VTONFS(dvp)->n_flag |= NMODIFIED;
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(dvp)->n_ncmtime, &premtime, ==))
- VTONFS(dvp)->n_ncmtime = VTONFS(dvp)->n_vattr.nva_mtime;
- if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(dvp));
+ if (!error && fh.fh_len)
+ error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh.fh_data, fh.fh_len, &nvattr, &xid, rq.r_auth, NG_MAKEENTRY, &np);
+ if (!error && np)
+ newvp = NFSTOV(np);
+
+ nfs_dulookup_finish(&dul, dnp, ctx);
/*
* Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
* if we can succeed in looking up the symlink.
*/
- if ((error == EEXIST) || (!error && !gotvp)) {
+ if ((error == EEXIST) || (!error && !newvp)) {
if (newvp) {
+ nfs_node_unlock(np);
vnode_put(newvp);
newvp = NULL;
}
- error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, cred, p, &np);
+ error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
if (!error) {
newvp = NFSTOV(np);
if (vnode_vtype(newvp) != VLNK)
error = EEXIST;
}
}
+ if (!busyerror)
+ nfs_node_clear_busy(dnp);
if (!error && (gotuid || gotgid) &&
- (!newvp || nfs_getattrcache(newvp, &nvattr) ||
+ (!newvp || nfs_getattrcache(np, &nvattr, 0) ||
(gotuid && (nvattr.nva_uid != vap->va_uid)) ||
(gotgid && (nvattr.nva_gid != vap->va_gid)))) {
/* clear ID bits if server didn't use them (or we can't tell) */
VATTR_CLEAR_SUPPORTED(vap, va_gid);
}
if (error) {
- if (newvp)
+ if (newvp) {
+ nfs_node_unlock(np);
vnode_put(newvp);
+ }
} else {
+ nfs_node_unlock(np);
*ap->a_vpp = newvp;
}
return (error);
}
/*
- * nfs make dir call
+ * NFS make dir call
*/
-static int
-nfs_mkdir(ap)
+int
+nfs3_vnop_mkdir(
struct vnop_mkdir_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_dvp;
struct componentname *a_cnp;
struct vnode_attr *a_vap;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t dvp = ap->a_dvp;
struct vnode_attr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfs_vattr nvattr;
- struct nfsv2_sattr *sp;
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- int len;
- struct nfsnode *np = (struct nfsnode *)0;
- vnode_t newvp = (vnode_t)0;
- caddr_t bpos, dpos, cp2;
- int error = 0, wccpostattr = 0;
+ nfsnode_t np = NULL;
+ struct nfsmount *nmp;
+ nfsnode_t dnp = VTONFS(dvp);
+ vnode_t newvp = NULL;
+ int error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0;
struct timespec premtime = { 0, 0 };
- int gotvp = 0;
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3 = NFS_ISV3(dvp);
- int gotuid, gotgid;
- u_int64_t xid, dxid;
- kauth_cred_t cred;
- proc_t p;
-
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ int nfsvers, gotuid, gotgid;
+ u_int64_t xid= 0, dxid;
+ fhandle_t fh;
+ struct nfsm_chain nmreq, nmrep;
+ struct nfsreq rq, *req = &rq;
+ struct nfs_dulookup dul;
+
+ nmp = VTONMP(dvp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN))
+ return (ENAMETOOLONG);
- len = cnp->cn_namelen;
- nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
- if (error)
- return (error);
+ nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
VATTR_SET_SUPPORTED(vap, va_mode);
VATTR_SET_SUPPORTED(vap, va_uid);
gotuid = VATTR_IS_ACTIVE(vap, va_uid);
gotgid = VATTR_IS_ACTIVE(vap, va_gid);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_MKDIR]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN, v3);
- if (v3) {
- nfsm_v3sattr(vap);
- } else {
- struct timespec neg1time = { -1, -1 };
- nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
- sp->sa_mode = vtonfsv2_mode(VDIR,
- (VATTR_IS_ACTIVE(vap, va_mode) ? vap->va_mode : 0600));
- sp->sa_uid = gotuid ? (u_long)txdr_unsigned(vap->va_uid) : nfs_xdrneg1;
- sp->sa_gid = gotgid ? (u_long)txdr_unsigned(vap->va_gid) : nfs_xdrneg1;
- sp->sa_size = nfs_xdrneg1;
- if (VATTR_IS_ACTIVE(vap, va_access_time)) {
- txdr_nfsv2time(&vap->va_access_time, &sp->sa_atime);
- } else {
- txdr_nfsv2time(&neg1time, &sp->sa_atime);
- }
- if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
- txdr_nfsv2time(&vap->va_modify_time, &sp->sa_mtime);
- } else {
- txdr_nfsv2time(&neg1time, &sp->sa_mtime);
- }
+ error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
+ nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + NFSX_UNSIGNED +
+ nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(nfsvers));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_add_v3sattr(error, &nmreq, vap);
+ else
+ nfsm_chain_add_v2sattr(error, &nmreq, vap, -1);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+
+ error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_MKDIR,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
+ if (!error) {
+ nfs_dulookup_start(&dul, dnp, ctx);
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
}
- nfsm_request(dvp, NFSPROC_MKDIR, p, cred, &xid);
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
dxid = xid;
+ if (!error && !status) {
+ if (dnp->n_flag & NNEGNCENTRIES) {
+ dnp->n_flag &= ~NNEGNCENTRIES;
+ cache_purge_negatives(dvp);
+ }
+ error = nfsm_chain_get_fh_attr(&nmrep, dnp, ctx, nfsvers, &xid, &fh, &nvattr);
+ }
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
if (!error)
- nfsm_mtofh(dvp, cnp, newvp, v3, &xid, gotvp);
- if (v3 && mrep)
- nfsm_wcc_data(dvp, &premtime, wccpostattr, &dxid);
- nfsm_reqdone;
- VTONFS(dvp)->n_flag |= NMODIFIED;
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(dvp)->n_ncmtime, &premtime, ==))
- VTONFS(dvp)->n_ncmtime = VTONFS(dvp)->n_vattr.nva_mtime;
- if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(dvp));
+ error = status;
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+
+ if (!lockerror) {
+ dnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
+ nfs_node_unlock(dnp);
+ /* nfs_getattr() will check changed and purge caches */
+ nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
+ }
+
+ if (!error && fh.fh_len)
+ error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh.fh_data, fh.fh_len, &nvattr, &xid, rq.r_auth, NG_MAKEENTRY, &np);
+ if (!error && np)
+ newvp = NFSTOV(np);
+
+ nfs_dulookup_finish(&dul, dnp, ctx);
+
/*
* Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
* if we can succeed in looking up the directory.
*/
- if (error == EEXIST || (!error && !gotvp)) {
+ if ((error == EEXIST) || (!error && !newvp)) {
if (newvp) {
+ nfs_node_unlock(np);
vnode_put(newvp);
newvp = NULL;
}
- error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cred, p, &np);
+ error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
if (!error) {
newvp = NFSTOV(np);
if (vnode_vtype(newvp) != VDIR)
error = EEXIST;
}
}
+ if (!busyerror)
+ nfs_node_clear_busy(dnp);
if (!error && (gotuid || gotgid) &&
- (!newvp || nfs_getattrcache(newvp, &nvattr) ||
+ (!newvp || nfs_getattrcache(np, &nvattr, 0) ||
(gotuid && (nvattr.nva_uid != vap->va_uid)) ||
(gotgid && (nvattr.nva_gid != vap->va_gid)))) {
/* clear ID bits if server didn't use them (or we can't tell) */
VATTR_CLEAR_SUPPORTED(vap, va_gid);
}
if (error) {
- if (newvp)
+ if (newvp) {
+ nfs_node_unlock(np);
vnode_put(newvp);
+ }
} else {
+ nfs_node_unlock(np);
*ap->a_vpp = newvp;
}
return (error);
}
/*
- * nfs remove directory call
+ * NFS remove directory call
*/
-static int
-nfs_rmdir(ap)
+int
+nfs3_vnop_rmdir(
struct vnop_rmdir_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_dvp;
vnode_t a_vp;
struct componentname *a_cnp;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
vnode_t vp = ap->a_vp;
vnode_t dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- caddr_t bpos, dpos, cp2;
- int error = 0, wccpostattr = 0;
+ int error = 0, lockerror = ENOENT, status, wccpostattr = 0;
struct timespec premtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- int v3 = NFS_ISV3(dvp);
+ struct nfsmount *nmp;
+ nfsnode_t np = VTONFS(vp);
+ nfsnode_t dnp = VTONFS(dvp);
+ int nfsvers;
u_int64_t xid;
- kauth_cred_t cred;
- proc_t p;
+ struct nfsm_chain nmreq, nmrep;
+ struct nfsreq rq, *req = &rq;
+ struct nfs_dulookup dul;
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN))
+ return (ENAMETOOLONG);
+
+ if ((error = nfs_node_set_busy2(dnp, np, vfs_context_thread(ctx))))
+ return (error);
+
+ nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+
+ error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_RMDIR,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
+ if (!error) {
+ nfs_dulookup_start(&dul, dnp, ctx);
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ }
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &xid);
+ if (!error)
+ error = status;
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ if (!lockerror) {
+ dnp->n_flag |= NMODIFIED;
+ /* if directory hadn't changed, update namecache mtime */
+ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==))
+ NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
+ nfs_node_unlock(dnp);
+ nfs_name_cache_purge(dnp, np, cnp, ctx);
+ /* nfs_getattr() will check changed and purge caches */
+ nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
+ }
+ nfs_dulookup_finish(&dul, dnp, ctx);
+ nfs_node_clear_busy2(dnp, np);
- nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_RMDIR]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN, v3);
- nfsm_request(dvp, NFSPROC_RMDIR, p, cred, &xid);
- if (v3 && mrep)
- nfsm_wcc_data(dvp, &premtime, wccpostattr, &xid);
- nfsm_reqdone;
- VTONFS(dvp)->n_flag |= NMODIFIED;
- /* if directory hadn't changed, update namecache mtime */
- if (nfstimespeccmp(&VTONFS(dvp)->n_ncmtime, &premtime, ==))
- VTONFS(dvp)->n_ncmtime = VTONFS(dvp)->n_vattr.nva_mtime;
- if (!wccpostattr)
- NATTRINVALIDATE(VTONFS(dvp));
- cache_purge(vp);
/*
* Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
*/
* before this vnode gets reclaimed
*/
lck_mtx_lock(nfs_node_hash_mutex);
- LIST_REMOVE(VTONFS(vp), n_hash);
- VTONFS(vp)->n_flag &= ~NHASHED;
+ if (np->n_hflag & NHHASHED) {
+ LIST_REMOVE(np, n_hash);
+ np->n_hflag &= ~NHHASHED;
+ FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
+ }
lck_mtx_unlock(nfs_node_hash_mutex);
}
return (error);
}
/*
- * nfs readdir call
+ * NFS readdir call
+ *
+ * The incoming "offset" is a directory cookie indicating where in the
+ * directory entries should be read from. A zero cookie means start at
+ * the beginning of the directory. Any other cookie will be a cookie
+ * returned from the server.
+ *
+ * Using that cookie, determine which buffer (and where in that buffer)
+ * to start returning entries from. Buffer logical block numbers are
+ * the cookies they start at. If a buffer is found that is not full,
+ * call into the bio/RPC code to fill it. The RPC code will probably
+ * fill several buffers (dropping the first, requiring a re-get).
+ *
+ * When done copying entries to the buffer, set the offset to the current
+ * entry's cookie and enter that cookie in the cookie cache.
+ *
+ * Note: because the getdirentries(2) API returns a long-typed offset,
+ * the incoming offset is a potentially truncated cookie (ptc).
+ * The cookie matching code is aware of this and will fall back to
+ * matching only 32 bits of the cookie.
*/
-static int
-nfs_readdir(ap)
+int
+nfs_vnop_readdir(
struct vnop_readdir_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
struct uio *a_uio;
+ int a_flags;
int *a_eofflag;
- int *a_ncookies;
- u_long **a_cookies;
+ int *a_numdirent;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
- vnode_t vp = ap->a_vp;
- struct nfsnode *np = VTONFS(vp);
- struct uio *uio = ap->a_uio;
- int tresid, error;
- struct nfs_vattr nvattr;
- kauth_cred_t cred;
- proc_t p;
+ vfs_context_t ctx = ap->a_context;
+ vnode_t dvp = ap->a_vp;
+ nfsnode_t dnp = VTONFS(dvp);
+ struct nfsmount *nmp;
+ uio_t uio = ap->a_uio;
+ int error, nfsvers, extended, numdirent, bigcookies, ptc, done;
+ uint16_t i, iptc, rlen, nlen;
+ uint64_t cookie, nextcookie, lbn = 0;
+ struct nfsbuf *bp = NULL;
+ struct nfs_dir_buf_header *ndbhp;
+ struct direntry *dp, *dpptc;
+ struct dirent dent;
+ char *cp = NULL;
+ thread_t thd;
+
+ nmp = VTONMP(dvp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ bigcookies = (nmp->nm_state & NFSSTA_BIGCOOKIES);
+ extended = (ap->a_flags & VNODE_READDIR_EXTENDED);
- if (vnode_vtype(vp) != VDIR)
+ if (vnode_vtype(dvp) != VDIR)
return (EPERM);
- cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ if (ap->a_eofflag)
+ *ap->a_eofflag = 0;
+
+ if (uio_resid(uio) == 0)
+ return (0);
+
+ if ((nfsvers >= NFS_VER4) && (dnp->n_vattr.nva_flags & NFS_FFLAG_TRIGGER)) {
+ /* trigger directories should never be read, return nothing */
+ return (0);
+ }
+
+ thd = vfs_context_thread(ctx);
+ numdirent = done = 0;
+ nextcookie = uio_offset(uio);
+ ptc = bigcookies && NFS_DIR_COOKIE_POTENTIALLY_TRUNCATED(nextcookie);
+
+ if ((error = nfs_node_lock(dnp)))
+ goto out;
+
+ if (dnp->n_flag & NNEEDINVALIDATE) {
+ dnp->n_flag &= ~NNEEDINVALIDATE;
+ nfs_invaldir(dnp);
+ nfs_node_unlock(dnp);
+ error = nfs_vinvalbuf(dvp, 0, ctx, 1);
+ if (!error)
+ error = nfs_node_lock(dnp);
+ if (error)
+ goto out;
+ }
/*
- * First, check for hit on the EOF offset cache
+ * check for need to invalidate when (re)starting at beginning
*/
- if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
- (np->n_flag & NMODIFIED) == 0) {
- if (!nfs_getattr(vp, &nvattr, cred, p)) {
- if (nfstimespeccmp(&np->n_mtime, &nvattr.nva_mtime, ==)) {
- OSAddAtomic(1, (SInt32*)&nfsstats.direofcache_hits);
- return (0);
+ if (!nextcookie) {
+ if (dnp->n_flag & NMODIFIED) {
+ nfs_invaldir(dnp);
+ nfs_node_unlock(dnp);
+ if ((error = nfs_vinvalbuf(dvp, 0, ctx, 1)))
+ goto out;
+ } else {
+ nfs_node_unlock(dnp);
+ }
+ /* nfs_getattr() will check changed and purge caches */
+ if ((error = nfs_getattr(dnp, NULL, ctx, NGA_UNCACHED)))
+ goto out;
+ } else {
+ nfs_node_unlock(dnp);
+ }
+
+ error = nfs_dir_cookie_to_lbn(dnp, nextcookie, &ptc, &lbn);
+ if (error) {
+ if (error < 0) { /* just hit EOF cookie */
+ done = 1;
+ error = 0;
+ }
+ if (ap->a_eofflag)
+ *ap->a_eofflag = 1;
+ }
+
+ while (!error && !done) {
+ OSAddAtomic64(1, &nfsstats.biocache_readdirs);
+ cookie = nextcookie;
+getbuffer:
+ error = nfs_buf_get(dnp, lbn, NFS_DIRBLKSIZ, thd, NBLK_READ, &bp);
+ if (error)
+ goto out;
+ ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
+ if (!ISSET(bp->nb_flags, NB_CACHE) || !ISSET(ndbhp->ndbh_flags, NDB_FULL)) {
+ if (!ISSET(bp->nb_flags, NB_CACHE)) { /* initialize the buffer */
+ ndbhp->ndbh_flags = 0;
+ ndbhp->ndbh_count = 0;
+ ndbhp->ndbh_entry_end = sizeof(*ndbhp);
+ ndbhp->ndbh_ncgen = dnp->n_ncgen;
+ }
+ error = nfs_buf_readdir(bp, ctx);
+ if (error == NFSERR_DIRBUFDROPPED)
+ goto getbuffer;
+ if (error)
+ nfs_buf_release(bp, 1);
+ if (error && (error != ENXIO) && (error != ETIMEDOUT) && (error != EINTR) && (error != ERESTART)) {
+ if (!nfs_node_lock(dnp)) {
+ nfs_invaldir(dnp);
+ nfs_node_unlock(dnp);
+ }
+ nfs_vinvalbuf(dvp, 0, ctx, 1);
+ if (error == NFSERR_BAD_COOKIE)
+ error = ENOENT;
+ }
+ if (error)
+ goto out;
+ }
+
+ /* find next entry to return */
+ dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
+ i = 0;
+ if ((lbn != cookie) && !(ptc && NFS_DIR_COOKIE_SAME32(lbn, cookie))) {
+ dpptc = NULL;
+ iptc = 0;
+ for (; (i < ndbhp->ndbh_count) && (cookie != dp->d_seekoff); i++) {
+ if (ptc && !dpptc && NFS_DIR_COOKIE_SAME32(cookie, dp->d_seekoff)) {
+ iptc = i;
+ dpptc = dp;
+ }
+ nextcookie = dp->d_seekoff;
+ dp = NFS_DIRENTRY_NEXT(dp);
+ }
+ if ((i == ndbhp->ndbh_count) && dpptc) {
+ i = iptc;
+ dp = dpptc;
+ }
+ if (i < ndbhp->ndbh_count) {
+ nextcookie = dp->d_seekoff;
+ dp = NFS_DIRENTRY_NEXT(dp);
+ i++;
+ }
+ }
+ ptc = 0; /* only have to deal with ptc on first cookie */
+
+ /* return as many entries as we can */
+ for (; i < ndbhp->ndbh_count; i++) {
+ if (extended) {
+ rlen = dp->d_reclen;
+ cp = (char*)dp;
+ } else {
+ if (!cp) {
+ cp = (char*)&dent;
+ bzero(cp, sizeof(dent));
+ }
+ if (dp->d_namlen > (sizeof(dent.d_name) - 1))
+ nlen = sizeof(dent.d_name) - 1;
+ else
+ nlen = dp->d_namlen;
+ rlen = NFS_DIRENT_LEN(nlen);
+ dent.d_reclen = rlen;
+ dent.d_ino = dp->d_ino;
+ dent.d_type = dp->d_type;
+ dent.d_namlen = nlen;
+ strlcpy(dent.d_name, dp->d_name, nlen + 1);
+ }
+ /* check that the record fits */
+ if (rlen > uio_resid(uio)) {
+ done = 1;
+ break;
}
- if (nfstimespeccmp(&np->n_ncmtime, &nvattr.nva_mtime, !=)) {
- /* directory changed, purge any name cache entries */
- cache_purge(vp);
+ if ((error = uiomove(cp, rlen, uio)))
+ break;
+ numdirent++;
+ nextcookie = dp->d_seekoff;
+ dp = NFS_DIRENTRY_NEXT(dp);
+ }
+
+ if (i == ndbhp->ndbh_count) {
+ /* hit end of buffer, move to next buffer */
+ lbn = nextcookie;
+ /* if we also hit EOF, we're done */
+ if (ISSET(ndbhp->ndbh_flags, NDB_EOF)) {
+ done = 1;
+ if (ap->a_eofflag)
+ *ap->a_eofflag = 1;
}
}
+ if (!error)
+ uio_setoffset(uio, nextcookie);
+ if (!error && !done && (nextcookie == cookie)) {
+ printf("nfs readdir cookie didn't change 0x%llx, %d/%d\n", cookie, i, ndbhp->ndbh_count);
+ error = EIO;
+ }
+ nfs_buf_release(bp, 1);
}
- /*
- * Call nfs_bioread() to do the real work.
- */
- // LP64todo - fix this
- tresid = uio_uio_resid(uio);
- error = nfs_bioread(vp, uio, 0, cred, p);
+ if (!error)
+ nfs_dir_cookie_cache(dnp, nextcookie, lbn);
- if (!error && uio_uio_resid(uio) == tresid)
- OSAddAtomic(1, (SInt32*)&nfsstats.direofcache_misses);
+ if (ap->a_numdirent)
+ *ap->a_numdirent = numdirent;
+out:
return (error);
}
+
+/*
+ * Invalidate cached directory information, except for the actual directory
+ * blocks (which are invalidated separately).
+ */
+void
+nfs_invaldir(nfsnode_t dnp)
+{
+ if (vnode_vtype(NFSTOV(dnp)) != VDIR)
+ return;
+ dnp->n_eofcookie = 0;
+ dnp->n_cookieverf = 0;
+ if (!dnp->n_cookiecache)
+ return;
+ dnp->n_cookiecache->free = 0;
+ dnp->n_cookiecache->mru = -1;
+ memset(dnp->n_cookiecache->next, -1, NFSNUMCOOKIES);
+}
+
+/*
+ * calculate how much space is available for additional directory entries.
+ */
+uint32_t
+nfs_dir_buf_freespace(struct nfsbuf *bp, int rdirplus)
+{
+ struct nfs_dir_buf_header *ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
+ uint32_t space;
+
+ if (!ndbhp)
+ return (0);
+ space = bp->nb_bufsize - ndbhp->ndbh_entry_end;
+ if (rdirplus)
+ space -= ndbhp->ndbh_count * sizeof(struct nfs_vattr);
+ return (space);
+}
+
+/*
+ * add/update a cookie->lbn entry in the directory cookie cache
+ */
+void
+nfs_dir_cookie_cache(nfsnode_t dnp, uint64_t cookie, uint64_t lbn)
+{
+ struct nfsdmap *ndcc;
+ int8_t i, prev;
+
+ if (!cookie)
+ return;
+
+ if (nfs_node_lock(dnp))
+ return;
+
+ if (cookie == dnp->n_eofcookie) { /* EOF cookie */
+ nfs_node_unlock(dnp);
+ return;
+ }
+
+ ndcc = dnp->n_cookiecache;
+ if (!ndcc) {
+ /* allocate the cookie cache structure */
+ MALLOC_ZONE(dnp->n_cookiecache, struct nfsdmap *,
+ sizeof(struct nfsdmap), M_NFSDIROFF, M_WAITOK);
+ if (!dnp->n_cookiecache) {
+ nfs_node_unlock(dnp);
+ return;
+ }
+ ndcc = dnp->n_cookiecache;
+ ndcc->free = 0;
+ ndcc->mru = -1;
+ memset(ndcc->next, -1, NFSNUMCOOKIES);
+ }
+
+ /*
+ * Search the list for this cookie.
+ * Keep track of previous and last entries.
+ */
+ prev = -1;
+ i = ndcc->mru;
+ while ((i != -1) && (cookie != ndcc->cookies[i].key)) {
+ if (ndcc->next[i] == -1) /* stop on last entry so we can reuse */
+ break;
+ prev = i;
+ i = ndcc->next[i];
+ }
+ if ((i != -1) && (cookie == ndcc->cookies[i].key)) {
+ /* found it, remove from list */
+ if (prev != -1)
+ ndcc->next[prev] = ndcc->next[i];
+ else
+ ndcc->mru = ndcc->next[i];
+ } else {
+ /* not found, use next free entry or reuse last entry */
+ if (ndcc->free != NFSNUMCOOKIES)
+ i = ndcc->free++;
+ else
+ ndcc->next[prev] = -1;
+ ndcc->cookies[i].key = cookie;
+ ndcc->cookies[i].lbn = lbn;
+ }
+ /* insert cookie at head of MRU list */
+ ndcc->next[i] = ndcc->mru;
+ ndcc->mru = i;
+ nfs_node_unlock(dnp);
+}
+
/*
- * Readdir rpc call.
- * Called from below the buffer cache by nfs_doio().
+ * Try to map the given directory cookie to a directory buffer (return lbn).
+ * If we have a possibly truncated cookie (ptc), check for 32-bit matches too.
*/
int
-nfs_readdirrpc(
- vnode_t vp,
- struct uio *uiop,
- kauth_cred_t cred,
- proc_t p)
+nfs_dir_cookie_to_lbn(nfsnode_t dnp, uint64_t cookie, int *ptc, uint64_t *lbnp)
{
- register int len, skiplen, left;
- register struct dirent *dp;
- register u_long *tl;
- register caddr_t cp;
- register long t1, t2;
- register nfsuint64 *cookiep;
- caddr_t bpos, dpos, cp2;
- mbuf_t mreq, mrep, md, mb, mb2;
- nfsuint64 cookie;
+ struct nfsdmap *ndcc = dnp->n_cookiecache;
+ int8_t eofptc, found;
+ int i, iptc;
struct nfsmount *nmp;
- struct nfsnode *dnp = VTONFS(vp);
- u_quad_t fileno;
- int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
- int attrflag;
- int v3, nmreaddirsize;
- u_int64_t xid;
+ struct nfsbuf *bp, *lastbp;
+ struct nfsbuflists blist;
+ struct direntry *dp, *dpptc;
+ struct nfs_dir_buf_header *ndbhp;
-#ifndef nolint
- dp = (struct dirent *)0;
-#endif
-#if DIAGNOSTIC
- if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (NFS_DIRBLKSIZ - 1)) ||
- (uio_uio_resid(uiop) & (NFS_DIRBLKSIZ - 1)))
- panic("nfs_readdirrpc: bad uio");
-#endif
- nmp = VFSTONFS(vnode_mount(vp));
- if (!nmp)
- return (ENXIO);
- v3 = NFS_ISV3(vp);
- nmreaddirsize = nmp->nm_readdirsize;
+ if (!cookie) { /* initial cookie */
+ *lbnp = 0;
+ *ptc = 0;
+ return (0);
+ }
+
+ if (nfs_node_lock(dnp))
+ return (ENOENT);
+
+ if (cookie == dnp->n_eofcookie) { /* EOF cookie */
+ nfs_node_unlock(dnp);
+ OSAddAtomic64(1, &nfsstats.direofcache_hits);
+ *ptc = 0;
+ return (-1);
+ }
+ /* note if cookie is a 32-bit match with the EOF cookie */
+ eofptc = *ptc ? NFS_DIR_COOKIE_SAME32(cookie, dnp->n_eofcookie) : 0;
+ iptc = -1;
+
+ /* search the list for the cookie */
+ for (i = ndcc ? ndcc->mru : -1; i >= 0; i = ndcc->next[i]) {
+ if (ndcc->cookies[i].key == cookie) {
+ /* found a match for this cookie */
+ *lbnp = ndcc->cookies[i].lbn;
+ nfs_node_unlock(dnp);
+ OSAddAtomic64(1, &nfsstats.direofcache_hits);
+ *ptc = 0;
+ return (0);
+ }
+ /* check for 32-bit match */
+ if (*ptc && (iptc == -1) && NFS_DIR_COOKIE_SAME32(ndcc->cookies[i].key, cookie))
+ iptc = i;
+ }
+ /* exact match not found */
+ if (eofptc) {
+ /* but 32-bit match hit the EOF cookie */
+ nfs_node_unlock(dnp);
+ OSAddAtomic64(1, &nfsstats.direofcache_hits);
+ return (-1);
+ }
+ if (iptc >= 0) {
+ /* but 32-bit match got a hit */
+ *lbnp = ndcc->cookies[iptc].lbn;
+ nfs_node_unlock(dnp);
+ OSAddAtomic64(1, &nfsstats.direofcache_hits);
+ return (0);
+ }
+ nfs_node_unlock(dnp);
/*
- * If there is no cookie, assume directory was stale.
+ * No match found in the cookie cache... hmm...
+ * Let's search the directory's buffers for the cookie.
*/
- cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
- if (cookiep)
- cookie = *cookiep;
- else
- return (NFSERR_BAD_COOKIE);
+ nmp = NFSTONMP(dnp);
+ if (!nmp)
+ return (ENXIO);
+ dpptc = NULL;
+ found = 0;
+
+ lck_mtx_lock(nfs_buf_mutex);
/*
- * Loop around doing readdir rpc's of size nm_readdirsize
- * truncated to a multiple of DIRBLKSIZ.
- * The stopping criteria is EOF or buffer full.
+ * Scan the list of buffers, keeping them in order.
+ * Note that itercomplete inserts each of the remaining buffers
+ * into the head of list (thus reversing the elements). So, we
+ * make sure to iterate through all buffers, inserting them after
+ * each other, to keep them in order.
+ * Also note: the LIST_INSERT_AFTER(lastbp) is only safe because
+ * we don't drop nfs_buf_mutex.
*/
- while (more_dirs && bigenough) {
- nfsm_reqhead(NFSX_FH(v3) + NFSX_READDIR(v3));
- if (error)
- goto nfsmout;
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_READDIR]);
- nfsm_fhtom(vp, v3);
- if (v3) {
- nfsm_build(tl, u_long *, 5 * NFSX_UNSIGNED);
- *tl++ = cookie.nfsuquad[0];
- *tl++ = cookie.nfsuquad[1];
- *tl++ = dnp->n_cookieverf.nfsuquad[0];
- *tl++ = dnp->n_cookieverf.nfsuquad[1];
- } else {
- nfsm_build(tl, u_long *, 2 * NFSX_UNSIGNED);
- *tl++ = cookie.nfsuquad[0];
- }
- *tl = txdr_unsigned(nmreaddirsize);
- nfsm_request(vp, NFSPROC_READDIR, p, cred, &xid);
- if (v3) {
- if (mrep) {
- nfsm_postop_attr_update(vp, v3, attrflag, &xid);
- }
- if (!error) {
- nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED);
- dnp->n_cookieverf.nfsuquad[0] = *tl++;
- dnp->n_cookieverf.nfsuquad[1] = *tl;
- } else {
- mbuf_freem(mrep);
- goto nfsmout;
+ if (!nfs_buf_iterprepare(dnp, &blist, NBI_CLEAN)) {
+ lastbp = NULL;
+ while ((bp = LIST_FIRST(&blist))) {
+ LIST_REMOVE(bp, nb_vnbufs);
+ if (!lastbp)
+ LIST_INSERT_HEAD(&dnp->n_cleanblkhd, bp, nb_vnbufs);
+ else
+ LIST_INSERT_AFTER(lastbp, bp, nb_vnbufs);
+ lastbp = bp;
+ if (found)
+ continue;
+ nfs_buf_refget(bp);
+ if (nfs_buf_acquire(bp, NBAC_NOWAIT, 0, 0)) {
+ /* just skip this buffer */
+ nfs_buf_refrele(bp);
+ continue;
}
- } else if (!mrep) {
- // XXX assert error?
- goto nfsmout;
- }
- nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
- more_dirs = fxdr_unsigned(int, *tl);
-
- /* loop thru the dir entries, doctoring them to 4bsd form */
- while (more_dirs && bigenough) {
- if (v3) {
- nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED);
- fxdr_hyper(tl, &fileno);
- len = fxdr_unsigned(int, *(tl + 2));
- } else {
- nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED);
- fileno = fxdr_unsigned(u_quad_t, *tl++);
- len = fxdr_unsigned(int, *tl);
+ nfs_buf_refrele(bp);
+
+ /* scan the buffer for the cookie */
+ ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
+ dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
+ dpptc = NULL;
+ for (i=0; (i < ndbhp->ndbh_count) && (cookie != dp->d_seekoff); i++) {
+ if (*ptc && !dpptc && NFS_DIR_COOKIE_SAME32(cookie, dp->d_seekoff)) {
+ dpptc = dp;
+ iptc = i;
+ }
+ dp = NFS_DIRENTRY_NEXT(dp);
}
- /* Note: v3 supports longer names, but struct dirent doesn't */
- /* so we just truncate the names to fit */
- if (len <= 0) {
- error = EBADRPC;
- mbuf_freem(mrep);
- goto nfsmout;
+ if ((i == ndbhp->ndbh_count) && dpptc) {
+ /* found only a PTC match */
+ dp = dpptc;
+ i = iptc;
+ } else if (i < ndbhp->ndbh_count) {
+ *ptc = 0;
}
- if (len > MAXNAMLEN) {
- skiplen = len - MAXNAMLEN;
- len = MAXNAMLEN;
- } else {
- skiplen = 0;
+ if (i < (ndbhp->ndbh_count-1)) {
+ /* next entry is *in* this buffer: return this block */
+ *lbnp = bp->nb_lblkno;
+ found = 1;
+ } else if (i == (ndbhp->ndbh_count-1)) {
+ /* next entry refers to *next* buffer: return next block */
+ *lbnp = dp->d_seekoff;
+ found = 1;
}
- tlen = nfsm_rndup(len);
- if (tlen == len)
- tlen += 4; /* To ensure null termination */
- left = DIRBLKSIZ - blksiz;
- if ((tlen + (int)DIRHDSIZ) > left) {
- dp->d_reclen += left;
- uio_iov_base_add(uiop, left);
- uio_iov_len_add(uiop, -left);
- uiop->uio_offset += left;
- uio_uio_resid_add(uiop, -left);
- blksiz = 0;
+ nfs_buf_drop(bp);
+ }
+ nfs_buf_itercomplete(dnp, &blist, NBI_CLEAN);
+ }
+ lck_mtx_unlock(nfs_buf_mutex);
+ if (found) {
+ OSAddAtomic64(1, &nfsstats.direofcache_hits);
+ return (0);
+ }
+
+ /* still not found... oh well, just start a new block */
+ *lbnp = cookie;
+ OSAddAtomic64(1, &nfsstats.direofcache_misses);
+ return (0);
+}
+
+/*
+ * scan a directory buffer for the given name
+ * Returns: ESRCH if not found, ENOENT if found invalid, 0 if found
+ * Note: should only be called with RDIRPLUS directory buffers
+ */
+
+#define NDBS_PURGE 1
+#define NDBS_UPDATE 2
+
+int
+nfs_dir_buf_search(
+ struct nfsbuf *bp,
+ struct componentname *cnp,
+ fhandle_t *fhp,
+ struct nfs_vattr *nvap,
+ uint64_t *xidp,
+ time_t *attrstampp,
+ daddr64_t *nextlbnp,
+ int flags)
+{
+ struct direntry *dp;
+ struct nfs_dir_buf_header *ndbhp;
+ struct nfs_vattr *nvattrp;
+ daddr64_t nextlbn = 0;
+ int i, error = ESRCH, fhlen;
+
+ /* scan the buffer for the name */
+ ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
+ dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
+ for (i=0; i < ndbhp->ndbh_count; i++) {
+ nextlbn = dp->d_seekoff;
+ if ((cnp->cn_namelen == dp->d_namlen) && !strcmp(cnp->cn_nameptr, dp->d_name)) {
+ fhlen = dp->d_name[dp->d_namlen+1];
+ nvattrp = NFS_DIR_BUF_NVATTR(bp, i);
+ if ((ndbhp->ndbh_ncgen != bp->nb_np->n_ncgen) || (fhp->fh_len == 0) ||
+ (nvattrp->nva_type == VNON) || (nvattrp->nva_fileid == 0)) {
+ /* entry is not valid */
+ error = ENOENT;
+ break;
}
- if ((tlen + (int)DIRHDSIZ) > uio_uio_resid(uiop))
- bigenough = 0;
- if (bigenough) {
- // LP64todo - fix this!
- dp = (struct dirent *) CAST_DOWN(caddr_t, uio_iov_base(uiop));
- dp->d_fileno = (int)fileno;
- dp->d_namlen = len;
- dp->d_reclen = tlen + DIRHDSIZ;
- dp->d_type = DT_UNKNOWN;
- blksiz += dp->d_reclen;
- if (blksiz == DIRBLKSIZ)
- blksiz = 0;
- uiop->uio_offset += DIRHDSIZ;
-#if LP64KERN
- uio_uio_resid_add(uiop, -((int64_t)DIRHDSIZ));
- uio_iov_len_add(uiop, -((int64_t)DIRHDSIZ));
-#else
- uio_uio_resid_add(uiop, -((int)DIRHDSIZ));
- uio_iov_len_add(uiop, -((int)DIRHDSIZ));
-#endif
- uio_iov_base_add(uiop, DIRHDSIZ);
- nfsm_mtouio(uiop, len);
- // LP64todo - fix this!
- cp = CAST_DOWN(caddr_t, uio_iov_base(uiop));
- tlen -= len;
- *cp = '\0'; /* null terminate */
- uio_iov_base_add(uiop, tlen);
- uio_iov_len_add(uiop, -tlen);
- uiop->uio_offset += tlen;
- uio_uio_resid_add(uiop, -tlen);
- } else {
- nfsm_adv(nfsm_rndup(len));
+ if (flags == NDBS_PURGE) {
+ dp->d_fileno = 0;
+ bzero(nvattrp, sizeof(*nvattrp));
+ error = ENOENT;
+ break;
}
- if (skiplen)
- nfsm_adv(nfsm_rndup(skiplen));
- if (v3) {
- nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED);
- } else {
- nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED);
+ if (flags == NDBS_UPDATE) {
+ /* update direntry's attrs if fh matches */
+ if ((fhp->fh_len == fhlen) && !bcmp(&dp->d_name[dp->d_namlen+2], fhp->fh_data, fhlen)) {
+ bcopy(nvap, nvattrp, sizeof(*nvap));
+ dp->d_fileno = nvattrp->nva_fileid;
+ nvattrp->nva_fileid = *xidp;
+ *(time_t*)(&dp->d_name[dp->d_namlen+2+fhp->fh_len]) = *attrstampp;
+ }
+ error = 0;
+ break;
}
- if (bigenough) {
- cookie.nfsuquad[0] = *tl++;
- if (v3)
- cookie.nfsuquad[1] = *tl++;
- } else if (v3)
- tl += 2;
- else
- tl++;
- more_dirs = fxdr_unsigned(int, *tl);
+ /* copy out fh, attrs, attrstamp, and xid */
+ fhp->fh_len = fhlen;
+ bcopy(&dp->d_name[dp->d_namlen+2], fhp->fh_data, MAX(fhp->fh_len, (int)sizeof(fhp->fh_data)));
+ *attrstampp = *(time_t*)(&dp->d_name[dp->d_namlen+2+fhp->fh_len]);
+ bcopy(nvattrp, nvap, sizeof(*nvap));
+ *xidp = nvap->nva_fileid;
+ nvap->nva_fileid = dp->d_fileno;
+ error = 0;
+ break;
}
- /*
- * If at end of rpc data, get the eof boolean
- */
- if (!more_dirs) {
- nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
- more_dirs = (fxdr_unsigned(int, *tl) == 0);
+ dp = NFS_DIRENTRY_NEXT(dp);
+ }
+ if (nextlbnp)
+ *nextlbnp = nextlbn;
+ return (error);
+}
+
+/*
+ * Look up a name in a directory's buffers.
+ * Note: should only be called with RDIRPLUS directory buffers
+ */
+int
+nfs_dir_buf_cache_lookup(nfsnode_t dnp, nfsnode_t *npp, struct componentname *cnp, vfs_context_t ctx, int purge)
+{
+ nfsnode_t newnp;
+ struct nfsmount *nmp;
+ int error = 0, i, found = 0, count = 0;
+ u_int64_t xid;
+ struct nfs_vattr nvattr;
+ fhandle_t fh;
+ time_t attrstamp = 0;
+ thread_t thd = vfs_context_thread(ctx);
+ struct nfsbuf *bp, *lastbp, *foundbp;
+ struct nfsbuflists blist;
+ daddr64_t lbn, nextlbn;
+ int dotunder = (cnp->cn_namelen > 2) && (cnp->cn_nameptr[0] == '.') && (cnp->cn_nameptr[1] == '_');
+
+ if (!(nmp = NFSTONMP(dnp)))
+ return (ENXIO);
+ if (!purge)
+ *npp = NULL;
+
+ /* first check most recent buffer (and next one too) */
+ lbn = dnp->n_lastdbl;
+ for (i=0; i < 2; i++) {
+ if ((error = nfs_buf_get(dnp, lbn, NFS_DIRBLKSIZ, thd, NBLK_READ|NBLK_ONLYVALID, &bp)))
+ return (error);
+ if (!bp)
+ break;
+ count++;
+ error = nfs_dir_buf_search(bp, cnp, &fh, &nvattr, &xid, &attrstamp, &nextlbn, purge ? NDBS_PURGE : 0);
+ nfs_buf_release(bp, 0);
+ if (error == ESRCH) {
+ error = 0;
+ } else {
+ found = 1;
+ break;
}
- mbuf_freem(mrep);
+ lbn = nextlbn;
}
- /*
- * Fill last record, iff any, out to a multiple of DIRBLKSIZ
- * by increasing d_reclen for the last record.
- */
- if (blksiz > 0) {
- left = DIRBLKSIZ - blksiz;
- dp->d_reclen += left;
- uio_iov_base_add(uiop, left);
- uio_iov_len_add(uiop, -left);
- uiop->uio_offset += left;
- uio_uio_resid_add(uiop, -left);
+
+ lck_mtx_lock(nfs_buf_mutex);
+ if (found) {
+ dnp->n_lastdbl = lbn;
+ goto done;
}
/*
- * We are now either at the end of the directory or have filled the
- * block.
+ * Scan the list of buffers, keeping them in order.
+ * Note that itercomplete inserts each of the remaining buffers
+ * into the head of list (thus reversing the elements). So, we
+ * make sure to iterate through all buffers, inserting them after
+ * each other, to keep them in order.
+ * Also note: the LIST_INSERT_AFTER(lastbp) is only safe because
+ * we don't drop nfs_buf_mutex.
*/
- if (bigenough)
- dnp->n_direofoffset = uiop->uio_offset;
- else {
- if (uio_uio_resid(uiop) > 0)
- printf("EEK! readdirrpc resid > 0\n");
- cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
- if (cookiep)
- *cookiep = cookie;
+ if (!nfs_buf_iterprepare(dnp, &blist, NBI_CLEAN)) {
+ lastbp = foundbp = NULL;
+ while ((bp = LIST_FIRST(&blist))) {
+ LIST_REMOVE(bp, nb_vnbufs);
+ if (!lastbp)
+ LIST_INSERT_HEAD(&dnp->n_cleanblkhd, bp, nb_vnbufs);
+ else
+ LIST_INSERT_AFTER(lastbp, bp, nb_vnbufs);
+ lastbp = bp;
+ if (error || found)
+ continue;
+ if (!purge && dotunder && (count > 100)) /* don't waste too much time looking for ._ files */
+ continue;
+ nfs_buf_refget(bp);
+ lbn = bp->nb_lblkno;
+ if (nfs_buf_acquire(bp, NBAC_NOWAIT, 0, 0)) {
+ /* just skip this buffer */
+ nfs_buf_refrele(bp);
+ continue;
+ }
+ nfs_buf_refrele(bp);
+ count++;
+ error = nfs_dir_buf_search(bp, cnp, &fh, &nvattr, &xid, &attrstamp, NULL, purge ? NDBS_PURGE : 0);
+ if (error == ESRCH) {
+ error = 0;
+ } else {
+ found = 1;
+ foundbp = bp;
+ }
+ nfs_buf_drop(bp);
+ }
+ if (found) {
+ LIST_REMOVE(foundbp, nb_vnbufs);
+ LIST_INSERT_HEAD(&dnp->n_cleanblkhd, foundbp, nb_vnbufs);
+ dnp->n_lastdbl = foundbp->nb_lblkno;
+ }
+ nfs_buf_itercomplete(dnp, &blist, NBI_CLEAN);
}
-nfsmout:
+done:
+ lck_mtx_unlock(nfs_buf_mutex);
+
+ if (!error && found && !purge) {
+ error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh.fh_data, fh.fh_len,
+ &nvattr, &xid, dnp->n_auth, NG_MAKEENTRY, &newnp);
+ if (error)
+ return (error);
+ newnp->n_attrstamp = attrstamp;
+ *npp = newnp;
+ nfs_node_unlock(newnp);
+ /* check if the dir buffer's attrs are out of date */
+ if (!nfs_getattr(newnp, &nvattr, ctx, NGA_CACHED) &&
+ (newnp->n_attrstamp != attrstamp)) {
+ /* they are, so update them */
+ error = nfs_buf_get(dnp, lbn, NFS_DIRBLKSIZ, thd, NBLK_READ|NBLK_ONLYVALID, &bp);
+ if (!error && bp) {
+ attrstamp = newnp->n_attrstamp;
+ xid = newnp->n_xid;
+ nfs_dir_buf_search(bp, cnp, &fh, &nvattr, &xid, &attrstamp, NULL, NDBS_UPDATE);
+ nfs_buf_release(bp, 0);
+ }
+ error = 0;
+ }
+ }
+
return (error);
}
/*
- * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
+ * Purge name cache entries for the given node.
+ * For RDIRPLUS, also invalidate the entry in the directory's buffers.
+ */
+void
+nfs_name_cache_purge(nfsnode_t dnp, nfsnode_t np, struct componentname *cnp, vfs_context_t ctx)
+{
+ struct nfsmount *nmp = NFSTONMP(dnp);
+
+ cache_purge(NFSTOV(np));
+ if (nmp && (nmp->nm_vers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS))
+ nfs_dir_buf_cache_lookup(dnp, NULL, cnp, ctx, 1);
+}
+
+/*
+ * NFS V3 readdir (plus) RPC.
*/
int
-nfs_readdirplusrpc(
- vnode_t vp,
- struct uio *uiop,
- kauth_cred_t cred,
- proc_t p)
+nfs3_readdir_rpc(nfsnode_t dnp, struct nfsbuf *bp, vfs_context_t ctx)
{
- int len, skiplen, left;
- struct dirent *dp;
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- vnode_t newvp;
- nfsuint64 *cookiep;
- caddr_t bpos, dpos, cp2;
- mbuf_t mreq, mrep, md, mb, mb2;
- struct componentname cn, *cnp = &cn;
- nfsuint64 cookie;
struct nfsmount *nmp;
- struct nfsnode *dnp = VTONFS(vp), *np;
- u_char *fhp;
- u_quad_t fileno;
- int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
- int attrflag, fhsize, nmreaddirsize, nmrsize;
- u_int64_t xid, savexid;
- struct nfs_vattr nvattr;
+ int error = 0, lockerror, nfsvers, rdirplus, bigcookies;
+ int i, status, attrflag, fhflag, more_entries = 1, eof, bp_dropped = 0;
+ uint32_t nmreaddirsize, nmrsize;
+ uint32_t namlen, skiplen, fhlen, xlen, attrlen, reclen, space_free, space_needed;
+ uint64_t cookie, lastcookie, xid, savedxid, fileno;
+ struct nfsm_chain nmreq, nmrep, nmrepsave;
+ fhandle_t fh;
+ struct nfs_vattr *nvattrp;
+ struct nfs_dir_buf_header *ndbhp;
+ struct direntry *dp;
+ char *padstart, padlen;
+ struct timeval now;
-#ifndef nolint
- dp = (struct dirent *)0;
-#endif
-#if DIAGNOSTIC
- if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
- (uio_uio_resid(uiop) & (DIRBLKSIZ - 1)))
- panic("nfs_readdirplusrpc: bad uio");
-#endif
- nmp = VFSTONFS(vnode_mount(vp));
+ nmp = NFSTONMP(dnp);
if (!nmp)
return (ENXIO);
+ nfsvers = nmp->nm_vers;
nmreaddirsize = nmp->nm_readdirsize;
nmrsize = nmp->nm_rsize;
-
- bzero(cnp, sizeof(*cnp));
- newvp = NULLVP;
+ bigcookies = nmp->nm_state & NFSSTA_BIGCOOKIES;
+noplus:
+ rdirplus = ((nfsvers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) ? 1 : 0;
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ return (lockerror);
+
+ /* determine cookie to use, and move dp to the right offset */
+ ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
+ dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
+ if (ndbhp->ndbh_count) {
+ for (i=0; i < ndbhp->ndbh_count-1; i++)
+ dp = NFS_DIRENTRY_NEXT(dp);
+ cookie = dp->d_seekoff;
+ dp = NFS_DIRENTRY_NEXT(dp);
+ } else {
+ cookie = bp->nb_lblkno;
+ /* increment with every buffer read */
+ OSAddAtomic64(1, &nfsstats.readdir_bios);
+ }
+ lastcookie = cookie;
/*
- * If there is no cookie, assume directory was stale.
+ * Loop around doing readdir(plus) RPCs of size nm_readdirsize until
+ * the buffer is full (or we hit EOF). Then put the remainder of the
+ * results in the next buffer(s).
*/
- cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
- if (cookiep)
- cookie = *cookiep;
- else
- return (NFSERR_BAD_COOKIE);
- /*
- * Loop around doing readdir rpc's of size nm_readdirsize
- * truncated to a multiple of DIRBLKSIZ.
- * The stopping criteria is EOF or buffer full.
- */
- while (more_dirs && bigenough) {
- nfsm_reqhead(NFSX_FH(1) + 6 * NFSX_UNSIGNED);
- if (error)
- goto nfsmout;
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_READDIRPLUS]);
- nfsm_fhtom(vp, 1);
- nfsm_build(tl, u_long *, 6 * NFSX_UNSIGNED);
- *tl++ = cookie.nfsuquad[0];
- *tl++ = cookie.nfsuquad[1];
- *tl++ = dnp->n_cookieverf.nfsuquad[0];
- *tl++ = dnp->n_cookieverf.nfsuquad[1];
- *tl++ = txdr_unsigned(nmreaddirsize);
- *tl = txdr_unsigned(nmrsize);
- nfsm_request(vp, NFSPROC_READDIRPLUS, p, cred, &xid);
- savexid = xid;
- if (mrep) {
- nfsm_postop_attr_update(vp, 1, attrflag, &xid);
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+ while (nfs_dir_buf_freespace(bp, rdirplus) && !(ndbhp->ndbh_flags & NDB_FULL)) {
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + NFSX_READDIR(nfsvers) + NFSX_UNSIGNED);
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ if (nfsvers == NFS_VER3) {
+ /* opaque values don't need swapping, but as long */
+ /* as we are consistent about it, it should be ok */
+ nfsm_chain_add_64(error, &nmreq, cookie);
+ nfsm_chain_add_64(error, &nmreq, dnp->n_cookieverf);
+ } else {
+ nfsm_chain_add_32(error, &nmreq, cookie);
}
- if (error) {
- mbuf_freem(mrep);
- goto nfsmout;
- }
- nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED);
- dnp->n_cookieverf.nfsuquad[0] = *tl++;
- dnp->n_cookieverf.nfsuquad[1] = *tl++;
- more_dirs = fxdr_unsigned(int, *tl);
-
- /* loop thru the dir entries, doctoring them to 4bsd form */
- while (more_dirs && bigenough) {
- nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED);
- fxdr_hyper(tl, &fileno);
- len = fxdr_unsigned(int, *(tl + 2));
- /* Note: v3 supports longer names, but struct dirent doesn't */
- /* so we just truncate the names to fit */
- if (len <= 0) {
+ nfsm_chain_add_32(error, &nmreq, nmreaddirsize);
+ if (rdirplus)
+ nfsm_chain_add_32(error, &nmreq, nmrsize);
+ nfsm_chain_build_done(error, &nmreq);
+ nfs_node_unlock(dnp);
+ lockerror = ENOENT;
+ nfsmout_if(error);
+
+ error = nfs_request(dnp, NULL, &nmreq,
+ rdirplus ? NFSPROC_READDIRPLUS : NFSPROC_READDIR,
+ ctx, NULL, &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+
+ savedxid = xid;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid);
+ if (!error)
+ error = status;
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_64(error, &nmrep, dnp->n_cookieverf);
+ nfsm_chain_get_32(error, &nmrep, more_entries);
+
+ if (!lockerror) {
+ nfs_node_unlock(dnp);
+ lockerror = ENOENT;
+ }
+ if (error == NFSERR_NOTSUPP) {
+ /* oops... it doesn't look like readdirplus is supported */
+ lck_mtx_lock(&nmp->nm_lock);
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_RDIRPLUS);
+ lck_mtx_unlock(&nmp->nm_lock);
+ goto noplus;
+ }
+ nfsmout_if(error);
+
+ if (rdirplus)
+ microuptime(&now);
+
+ /* loop through the entries packing them into the buffer */
+ while (more_entries) {
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_64(error, &nmrep, fileno);
+ else
+ nfsm_chain_get_32(error, &nmrep, fileno);
+ nfsm_chain_get_32(error, &nmrep, namlen);
+ nfsmout_if(error);
+ /* just truncate names that don't fit in direntry.d_name */
+ if (namlen <= 0) {
error = EBADRPC;
- mbuf_freem(mrep);
goto nfsmout;
}
- if (len > MAXNAMLEN) {
- skiplen = len - MAXNAMLEN;
- len = MAXNAMLEN;
+ if (namlen > (sizeof(dp->d_name)-1)) {
+ skiplen = namlen - sizeof(dp->d_name) + 1;
+ namlen = sizeof(dp->d_name) - 1;
} else {
skiplen = 0;
}
- tlen = nfsm_rndup(len);
- if (tlen == len)
- tlen += 4; /* To ensure null termination*/
- left = DIRBLKSIZ - blksiz;
- if ((tlen + (int)DIRHDSIZ) > left) {
- dp->d_reclen += left;
- uio_iov_base_add(uiop, left);
- uio_iov_len_add(uiop, -left);
- uiop->uio_offset += left;
- uio_uio_resid_add(uiop, -left);
- blksiz = 0;
- }
- if ((tlen + (int)DIRHDSIZ) > uio_uio_resid(uiop))
- bigenough = 0;
- if (bigenough) {
- // LP64todo - fix this!
- dp = (struct dirent *) CAST_DOWN(caddr_t, uio_iov_base(uiop));
- dp->d_fileno = (int)fileno;
- dp->d_namlen = len;
- dp->d_reclen = tlen + DIRHDSIZ;
- dp->d_type = DT_UNKNOWN;
- blksiz += dp->d_reclen;
- if (blksiz == DIRBLKSIZ)
- blksiz = 0;
- uiop->uio_offset += DIRHDSIZ;
-#if LP64KERN
- uio_uio_resid_add(uiop, -((int64_t)DIRHDSIZ));
- uio_iov_len_add(uiop, -((int64_t)DIRHDSIZ));
-#else
- uio_uio_resid_add(uiop, -((int)DIRHDSIZ));
- uio_iov_len_add(uiop, -((int)DIRHDSIZ));
-#endif
- uio_iov_base_add(uiop, DIRHDSIZ);
- // LP64todo - fix this!
- cnp->cn_nameptr = CAST_DOWN(caddr_t, uio_iov_base(uiop));
- cnp->cn_namelen = len;
- nfsm_mtouio(uiop, len);
- cp = CAST_DOWN(caddr_t, uio_iov_base(uiop));
- tlen -= len;
- *cp = '\0';
- uio_iov_base_add(uiop, tlen);
- uio_iov_len_add(uiop, -tlen);
- uiop->uio_offset += tlen;
- uio_uio_resid_add(uiop, -tlen);
- } else {
- nfsm_adv(nfsm_rndup(len));
+ /* guess that fh size will be same as parent */
+ fhlen = rdirplus ? (1 + dnp->n_fhsize) : 0;
+ xlen = rdirplus ? (fhlen + sizeof(time_t)) : 0;
+ attrlen = rdirplus ? sizeof(struct nfs_vattr) : 0;
+ reclen = NFS_DIRENTRY_LEN(namlen + xlen);
+ space_needed = reclen + attrlen;
+ space_free = nfs_dir_buf_freespace(bp, rdirplus);
+ if (space_needed > space_free) {
+ /*
+ * We still have entries to pack, but we've
+ * run out of room in the current buffer.
+ * So we need to move to the next buffer.
+ * The block# for the next buffer is the
+ * last cookie in the current buffer.
+ */
+nextbuffer:
+ ndbhp->ndbh_flags |= NDB_FULL;
+ nfs_buf_release(bp, 0);
+ bp_dropped = 1;
+ bp = NULL;
+ error = nfs_buf_get(dnp, lastcookie, NFS_DIRBLKSIZ, vfs_context_thread(ctx), NBLK_READ, &bp);
+ nfsmout_if(error);
+ /* initialize buffer */
+ ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
+ ndbhp->ndbh_flags = 0;
+ ndbhp->ndbh_count = 0;
+ ndbhp->ndbh_entry_end = sizeof(*ndbhp);
+ ndbhp->ndbh_ncgen = dnp->n_ncgen;
+ space_free = nfs_dir_buf_freespace(bp, rdirplus);
+ dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
+ /* increment with every buffer read */
+ OSAddAtomic64(1, &nfsstats.readdir_bios);
}
+ nmrepsave = nmrep;
+ dp->d_fileno = fileno;
+ dp->d_namlen = namlen;
+ dp->d_reclen = reclen;
+ dp->d_type = DT_UNKNOWN;
+ nfsm_chain_get_opaque(error, &nmrep, namlen, dp->d_name);
+ nfsmout_if(error);
+ dp->d_name[namlen] = '\0';
if (skiplen)
- nfsm_adv(nfsm_rndup(skiplen));
- nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED);
- if (bigenough) {
- cookie.nfsuquad[0] = *tl++;
- cookie.nfsuquad[1] = *tl++;
- } else
- tl += 2;
-
- /*
- * Since the attributes are before the file handle
- * (sigh), we must skip over the attributes and then
- * come back and get them.
- */
- attrflag = fxdr_unsigned(int, *tl);
- if (attrflag) {
- /* grab attributes */
- nfsm_attr_get(1, &nvattr);
- dp->d_type = IFTODT(VTTOIF(nvattr.nva_type));
- /* check for file handle */
- nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
- doit = fxdr_unsigned(int, *tl);
- if (doit) {
- nfsm_getfh(fhp, fhsize, 1);
- if (NFS_CMPFH(dnp, fhp, fhsize)) {
- error = vnode_ref(vp);
- if (error) {
- doit = 0;
- } else {
- newvp = vp;
- np = dnp;
- }
- } else if (!bigenough ||
- (cnp->cn_namelen == 2 &&
- cnp->cn_nameptr[1] == '.' &&
- cnp->cn_nameptr[0] == '.')) {
- /*
- * XXXmacko I don't think this ".." thing is a problem anymore.
- * don't doit if we can't guarantee
- * that this entry is NOT ".." because
- * we would have to drop the lock on
- * the directory before getting the
- * lock on the ".." vnode... and we
- * don't want to drop the dvp lock in
- * the middle of a readdirplus.
- */
- doit = 0;
+ nfsm_chain_adv(error, &nmrep,
+ nfsm_rndup(namlen + skiplen) - nfsm_rndup(namlen));
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_get_64(error, &nmrep, cookie);
+ else
+ nfsm_chain_get_32(error, &nmrep, cookie);
+ nfsmout_if(error);
+ dp->d_seekoff = cookie;
+ if (!bigcookies && (cookie >> 32) && (nmp == NFSTONMP(dnp))) {
+ /* we've got a big cookie, make sure flag is set */
+ lck_mtx_lock(&nmp->nm_lock);
+ nmp->nm_state |= NFSSTA_BIGCOOKIES;
+ lck_mtx_unlock(&nmp->nm_lock);
+ bigcookies = 1;
+ }
+ if (rdirplus) {
+ nvattrp = NFS_DIR_BUF_NVATTR(bp, ndbhp->ndbh_count);
+ /* check for attributes */
+ nfsm_chain_get_32(error, &nmrep, attrflag);
+ nfsmout_if(error);
+ if (attrflag) {
+ /* grab attributes */
+ error = nfs_parsefattr(&nmrep, NFS_VER3, nvattrp);
+ nfsmout_if(error);
+ dp->d_type = IFTODT(VTTOIF(nvattrp->nva_type));
+ /* fileid is already in d_fileno, so stash xid in attrs */
+ nvattrp->nva_fileid = savedxid;
} else {
- cnp->cn_hash = 0;
-
- error = nfs_nget(vnode_mount(vp), vp, cnp,
- fhp, fhsize, &nvattr, &xid,
- NG_MAKEENTRY, &np);
- if (error)
- doit = 0;
- else
- newvp = NFSTOV(np);
+ /* mark the attributes invalid */
+ bzero(nvattrp, sizeof(struct nfs_vattr));
}
- }
- /* update attributes if not already updated */
- if (doit && bigenough && (np->n_xid <= savexid)) {
- xid = savexid;
- nfs_loadattrcache(np, &nvattr, &xid, 0);
- /* any error can be ignored */
- }
- } else {
- /* Just skip over the file handle */
- nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
- i = fxdr_unsigned(int, *tl);
- nfsm_adv(nfsm_rndup(i));
- }
- if (newvp != NULLVP) {
- if (newvp == vp)
- vnode_rele(newvp);
- else
- vnode_put(newvp);
- newvp = NULLVP;
+ /* check for file handle */
+ nfsm_chain_get_32(error, &nmrep, fhflag);
+ nfsmout_if(error);
+ if (fhflag) {
+ nfsm_chain_get_fh(error, &nmrep, NFS_VER3, &fh);
+ nfsmout_if(error);
+ fhlen = fh.fh_len + 1;
+ xlen = fhlen + sizeof(time_t);
+ reclen = NFS_DIRENTRY_LEN(namlen + xlen);
+ space_needed = reclen + attrlen;
+ if (space_needed > space_free) {
+ /* didn't actually have the room... move on to next buffer */
+ nmrep = nmrepsave;
+ goto nextbuffer;
+ }
+ /* pack the file handle into the record */
+ dp->d_name[dp->d_namlen+1] = fh.fh_len;
+ bcopy(fh.fh_data, &dp->d_name[dp->d_namlen+2], fh.fh_len);
+ } else {
+ /* mark the file handle invalid */
+ fh.fh_len = 0;
+ fhlen = fh.fh_len + 1;
+ xlen = fhlen + sizeof(time_t);
+ reclen = NFS_DIRENTRY_LEN(namlen + xlen);
+ bzero(&dp->d_name[dp->d_namlen+1], fhlen);
+ }
+ *(time_t*)(&dp->d_name[dp->d_namlen+1+fhlen]) = now.tv_sec;
+ dp->d_reclen = reclen;
}
- nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
- more_dirs = fxdr_unsigned(int, *tl);
+ padstart = dp->d_name + dp->d_namlen + 1 + xlen;
+ ndbhp->ndbh_count++;
+ lastcookie = cookie;
+ /* advance to next direntry in buffer */
+ dp = NFS_DIRENTRY_NEXT(dp);
+ ndbhp->ndbh_entry_end = (char*)dp - bp->nb_data;
+ /* zero out the pad bytes */
+ padlen = (char*)dp - padstart;
+ if (padlen > 0)
+ bzero(padstart, padlen);
+ /* check for more entries */
+ nfsm_chain_get_32(error, &nmrep, more_entries);
+ nfsmout_if(error);
}
- /*
- * If at end of rpc data, get the eof boolean
- */
- if (!more_dirs) {
- nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
- more_dirs = (fxdr_unsigned(int, *tl) == 0);
+ /* Finally, get the eof boolean */
+ nfsm_chain_get_32(error, &nmrep, eof);
+ nfsmout_if(error);
+ if (eof) {
+ ndbhp->ndbh_flags |= (NDB_FULL|NDB_EOF);
+ nfs_node_lock_force(dnp);
+ dnp->n_eofcookie = lastcookie;
+ nfs_node_unlock(dnp);
+ } else {
+ more_entries = 1;
}
- mbuf_freem(mrep);
- }
- /*
- * Fill last record, iff any, out to a multiple of NFS_DIRBLKSIZ
- * by increasing d_reclen for the last record.
- */
- if (blksiz > 0) {
- left = DIRBLKSIZ - blksiz;
- dp->d_reclen += left;
- uio_iov_base_add(uiop, left);
- uio_iov_len_add(uiop, -left);
- uiop->uio_offset += left;
- uio_uio_resid_add(uiop, -left);
- }
-
- /*
- * We are now either at the end of the directory or have filled the
- * block.
- */
- if (bigenough)
- dnp->n_direofoffset = uiop->uio_offset;
- else {
- if (uio_uio_resid(uiop) > 0)
- printf("EEK! readdirplusrpc resid > 0\n");
- cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
- if (cookiep)
- *cookiep = cookie;
+ if (bp_dropped) {
+ nfs_buf_release(bp, 0);
+ bp = NULL;
+ break;
+ }
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+ nfsmout_if(error);
+ nfsm_chain_cleanup(&nmrep);
+ nfsm_chain_null(&nmreq);
}
nfsmout:
- return (error);
+ if (bp_dropped && bp)
+ nfs_buf_release(bp, 0);
+ if (!lockerror)
+ nfs_node_unlock(dnp);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (bp_dropped ? NFSERR_DIRBUFDROPPED : error);
}
/*
* Silly rename. To make the NFS filesystem that is stateless look a little
* more like the "ufs" a remove of an active vnode is translated to a rename
- * to a funny looking filename that is removed by nfs_inactive on the
+ * to a funny looking filename that is removed by nfs_vnop_inactive on the
* nfsnode. There is the potential for another process on a different client
- * to create the same funny name between the nfs_lookitup() fails and the
- * nfs_rename() completes, but...
+ * to create the same funny name between when the lookitup() fails and the
+ * rename() completes, but...
*/
-/* format of "random" names and next name to try */
-/* (note: shouldn't exceed size of sillyrename.s_name) */
-static char sillyrename_name[] = ".nfsAAA%04x4.4";
+/* format of "random" silly names - includes a number and pid */
+/* (note: shouldn't exceed size of nfs_sillyrename.nsr_name) */
+#define NFS_SILLYNAME_FORMAT ".nfs.%08x.%04x"
+/* starting from zero isn't silly enough */
+static uint32_t nfs_sillyrename_number = 0x20051025;
-static int
+int
nfs_sillyrename(
- vnode_t dvp,
- vnode_t vp,
+ nfsnode_t dnp,
+ nfsnode_t np,
struct componentname *cnp,
- kauth_cred_t cred,
- proc_t p)
+ vfs_context_t ctx)
{
- register struct sillyrename *sp;
- struct nfsnode *np;
+ struct nfs_sillyrename *nsp;
int error;
short pid;
- kauth_cred_t tmpcred;
- int i, j, k;
+ kauth_cred_t cred;
+ uint32_t num;
+ struct nfsmount *nmp;
- cache_purge(vp);
- np = VTONFS(vp);
-#if DIAGNOSTIC
- if (vnode_vtype(vp) == VDIR)
- panic("nfs_sillyrename: dir");
-#endif
- MALLOC_ZONE(sp, struct sillyrename *,
- sizeof (struct sillyrename), M_NFSREQ, M_WAITOK);
- if (!sp)
+ nmp = NFSTONMP(dnp);
+ if (!nmp)
+ return (ENXIO);
+
+ nfs_name_cache_purge(dnp, np, cnp, ctx);
+
+ MALLOC_ZONE(nsp, struct nfs_sillyrename *,
+ sizeof (struct nfs_sillyrename), M_NFSREQ, M_WAITOK);
+ if (!nsp)
return (ENOMEM);
+ cred = vfs_context_ucred(ctx);
kauth_cred_ref(cred);
- sp->s_cred = cred;
- sp->s_dvp = dvp;
- error = vnode_ref(dvp);
+ nsp->nsr_cred = cred;
+ nsp->nsr_dnp = dnp;
+ error = vnode_ref(NFSTOV(dnp));
if (error)
goto bad_norele;
/* Fudge together a funny name */
- pid = proc_pid(p);
- sp->s_namlen = sprintf(sp->s_name, sillyrename_name, pid);
+ pid = vfs_context_pid(ctx);
+ num = OSAddAtomic(1, &nfs_sillyrename_number);
+ nsp->nsr_namlen = snprintf(nsp->nsr_name, sizeof(nsp->nsr_name),
+ NFS_SILLYNAME_FORMAT, num, (pid & 0xffff));
+ if (nsp->nsr_namlen >= (int)sizeof(nsp->nsr_name))
+ nsp->nsr_namlen = sizeof(nsp->nsr_name) - 1;
/* Try lookitups until we get one that isn't there */
- i = j = k = 0;
- while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, p, NULL) == 0) {
- if (sp->s_name[4]++ >= 'z')
- sp->s_name[4] = 'A';
- if (++i > ('z' - 'A' + 1)) {
- i = 0;
- if (sp->s_name[5]++ >= 'z')
- sp->s_name[5] = 'A';
- if (++j > ('z' - 'A' + 1)) {
- j = 0;
- if (sp->s_name[6]++ >= 'z')
- sp->s_name[6] = 'A';
- if (++k > ('z' - 'A' + 1)) {
- error = EINVAL;
- goto bad;
- }
- }
- }
+ while (nfs_lookitup(dnp, nsp->nsr_name, nsp->nsr_namlen, ctx, NULL) == 0) {
+ num = OSAddAtomic(1, &nfs_sillyrename_number);
+ nsp->nsr_namlen = snprintf(nsp->nsr_name, sizeof(nsp->nsr_name),
+ NFS_SILLYNAME_FORMAT, num, (pid & 0xffff));
+ if (nsp->nsr_namlen >= (int)sizeof(nsp->nsr_name))
+ nsp->nsr_namlen = sizeof(nsp->nsr_name) - 1;
}
- /* make note of next "random" name to try */
- if ((sillyrename_name[4] = (sp->s_name[4] + 1)) > 'z') {
- sillyrename_name[4] = 'A';
- if ((sillyrename_name[5] = (sp->s_name[5] + 1)) > 'z') {
- sillyrename_name[5] = 'A';
- if ((sillyrename_name[6] = (sp->s_name[6] + 1)) > 'z')
- sillyrename_name[6] = 'A';
+
+ /* now, do the rename */
+ error = nmp->nm_funcs->nf_rename_rpc(dnp, cnp->cn_nameptr, cnp->cn_namelen,
+ dnp, nsp->nsr_name, nsp->nsr_namlen, ctx);
+
+ /* Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. */
+ if (error == ENOENT)
+ error = 0;
+ if (!error) {
+ nfs_node_lock_force(dnp);
+ if (dnp->n_flag & NNEGNCENTRIES) {
+ dnp->n_flag &= ~NNEGNCENTRIES;
+ cache_purge_negatives(NFSTOV(dnp));
}
+ nfs_node_unlock(dnp);
}
- /* now, do the rename */
- error = nfs_renamerpc(dvp, cnp->cn_nameptr, cnp->cn_namelen,
- dvp, sp->s_name, sp->s_namlen, sp->s_cred, p);
+ FSDBG(267, dnp, np, num, error);
if (error)
goto bad;
- error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, p, &np);
-#if DIAGNOSTIC
- kprintf("sillyrename: %s, vp=%x, np=%x, dvp=%x\n",
- &sp->s_name[0], (unsigned)vp, (unsigned)np, (unsigned)dvp);
-#endif
- np->n_sillyrename = sp;
+ error = nfs_lookitup(dnp, nsp->nsr_name, nsp->nsr_namlen, ctx, &np);
+ nfs_node_lock_force(np);
+ np->n_sillyrename = nsp;
+ nfs_node_unlock(np);
return (0);
bad:
- vnode_rele(sp->s_dvp);
+ vnode_rele(NFSTOV(dnp));
bad_norele:
- tmpcred = sp->s_cred;
- sp->s_cred = NOCRED;
- kauth_cred_rele(tmpcred);
- FREE_ZONE((caddr_t)sp, sizeof (struct sillyrename), M_NFSREQ);
- return (error);
-}
-
-/*
- * Look up a file name and optionally either update the file handle or
- * allocate an nfsnode, depending on the value of npp.
- * npp == NULL --> just do the lookup
- * *npp == NULL --> allocate a new nfsnode and make sure attributes are
- * handled too
- * *npp != NULL --> update the file handle in the vnode
- */
-static int
-nfs_lookitup(dvp, name, len, cred, procp, npp)
- vnode_t dvp;
- char *name;
- int len;
- kauth_cred_t cred;
- proc_t procp;
- struct nfsnode **npp;
-{
- u_long *tl;
- caddr_t cp;
- long t1, t2;
- vnode_t newvp = (vnode_t)0;
- struct nfsnode *np, *dnp = VTONFS(dvp);
- caddr_t bpos, dpos, cp2;
- int error = 0, fhlen, attrflag;
- mbuf_t mreq, mrep, md, mb, mb2;
- u_char *nfhp;
- int v3;
- u_int64_t xid, dxid, savedxid;
- struct nfs_vattr nvattr;
-
- if (!VFSTONFS(vnode_mount(dvp)))
- return (ENXIO);
- v3 = NFS_ISV3(dvp);
-
- nfsm_reqhead(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_LOOKUP]);
- nfsm_fhtom(dvp, v3);
- nfsm_strtom(name, len, NFS_MAXNAMLEN, v3);
- nfsm_request(dvp, NFSPROC_LOOKUP, procp, cred, &xid);
- if (npp && !error) {
- savedxid = xid;
- nfsm_getfh(nfhp, fhlen, v3);
- /* get attributes */
- if (v3) {
- nfsm_postop_attr_get(v3, attrflag, &nvattr);
- if (!attrflag) {
- /* We need valid attributes in order */
- /* to call nfs_nget/vnode_create(). */
- error = nfs_getattr_no_vnode(vnode_mount(dvp),
- nfhp, fhlen, cred, procp, &nvattr, &xid);
- if (error) {
- mbuf_freem(mrep);
- goto nfsmout;
- }
- }
- dxid = savedxid;
- nfsm_postop_attr_update(dvp, v3, attrflag, &dxid);
- } else {
- nfsm_attr_get(v3, &nvattr);
- }
- if (*npp) {
- np = *npp;
- if (fhlen != np->n_fhsize) {
- u_char *oldbuf = (np->n_fhsize > NFS_SMALLFH) ? np->n_fhp : NULL;
- if (fhlen > NFS_SMALLFH) {
- MALLOC_ZONE(np->n_fhp, u_char *, fhlen, M_NFSBIGFH, M_WAITOK);
- if (!np->n_fhp) {
- np->n_fhp = oldbuf;
- error = ENOMEM;
- mbuf_freem(mrep);
- goto nfsmout;
- }
- } else {
- np->n_fhp = &np->n_fh[0];
- }
- if (oldbuf) {
- FREE_ZONE(oldbuf, np->n_fhsize, M_NFSBIGFH);
- }
- }
- bcopy(nfhp, np->n_fhp, fhlen);
- np->n_fhsize = fhlen;
- newvp = NFSTOV(np);
- error = nfs_loadattrcache(np, &nvattr, &xid, 0);
- if (error) {
- mbuf_freem(mrep);
- goto nfsmout;
- }
- } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
- newvp = dvp;
- if (dnp->n_xid <= savedxid) {
- dxid = savedxid;
- error = nfs_loadattrcache(dnp, &nvattr, &dxid, 0);
- if (error) {
- mbuf_freem(mrep);
- goto nfsmout;
- }
- }
- } else {
- struct componentname cn, *cnp = &cn;
- bzero(cnp, sizeof(*cnp));
- cnp->cn_nameptr = name;
- cnp->cn_namelen = len;
-
- error = nfs_nget(vnode_mount(dvp), dvp, cnp, nfhp, fhlen,
- &nvattr, &xid, NG_MAKEENTRY, &np);
- if (error) {
- mbuf_freem(mrep);
- return (error);
- }
- newvp = NFSTOV(np);
- }
- }
- nfsm_reqdone;
- if (npp && *npp == NULL) {
- if (error) {
- if (newvp) {
- if (newvp == dvp)
- vnode_rele(newvp);
- else
- vnode_put(newvp);
- }
- } else
- *npp = np;
- }
+ nsp->nsr_cred = NOCRED;
+ kauth_cred_unref(&cred);
+ FREE_ZONE(nsp, sizeof(*nsp), M_NFSREQ);
return (error);
}
-/*
- * Nfs Version 3 commit rpc
- */
int
-nfs_commit(vp, offset, count, cred, procp)
- vnode_t vp;
- u_quad_t offset;
- u_int32_t count;
- kauth_cred_t cred;
- proc_t procp;
+nfs3_lookup_rpc_async(
+ nfsnode_t dnp,
+ char *name,
+ int namelen,
+ vfs_context_t ctx,
+ struct nfsreq **reqp)
{
- caddr_t cp;
- u_long *tl;
- int t1, t2;
- struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
- caddr_t bpos, dpos, cp2;
- int error = 0, wccpostattr = 0;
- struct timespec premtime = { 0, 0 };
- mbuf_t mreq, mrep, md, mb, mb2;
- u_int64_t xid, wverf;
+ struct nfsmount *nmp;
+ struct nfsm_chain nmreq;
+ int error = 0, nfsvers;
- FSDBG(521, vp, offset, count, nmp->nm_state);
+ nmp = NFSTONMP(dnp);
if (!nmp)
return (ENXIO);
- if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
- return (0);
- nfsm_reqhead(NFSX_FH(1));
- if (error)
- return (error);
- OSAddAtomic(1, (SInt32*)&nfsstats.rpccnt[NFSPROC_COMMIT]);
- nfsm_fhtom(vp, 1);
- nfsm_build(tl, u_long *, 3 * NFSX_UNSIGNED);
- txdr_hyper(&offset, tl);
- tl += 2;
- *tl = txdr_unsigned(count);
- nfsm_request(vp, NFSPROC_COMMIT, procp, cred, &xid);
- if (mrep) {
- nfsm_wcc_data(vp, &premtime, wccpostattr, &xid);
- /* XXX can we do anything useful with the wcc info? */
- }
- if (!error) {
- nfsm_dissect(tl, u_long *, NFSX_V3WRITEVERF);
- fxdr_hyper(tl, &wverf);
- if (wverf != nmp->nm_verf) {
- nmp->nm_verf = wverf;
- error = NFSERR_STALEWRITEVERF;
- }
- }
- nfsm_reqdone;
+ nfsvers = nmp->nm_vers;
+
+ nfsm_chain_null(&nmreq);
+
+ nfsm_chain_build_alloc_init(error, &nmreq,
+ NFSX_FH(nfsvers) + NFSX_UNSIGNED + nfsm_rndup(namelen));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ nfsm_chain_add_name(error, &nmreq, name, namelen, nmp);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_LOOKUP,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, reqp);
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
return (error);
}
-static int
-nfs_blockmap(
- __unused struct vnop_blockmap_args /* {
- struct vnodeop_desc *a_desc;
- vnode_t a_vp;
- off_t a_foffset;
- size_t a_size;
- daddr64_t *a_bpn;
- size_t *a_run;
- void *a_poff;
- int a_flags;
- } */ *ap)
-{
- return (ENOTSUP);
-}
-
-/*
- * Mmap a file
- *
- * NB Currently unsupported.
- */
-/*ARGSUSED*/
-static int
-nfs_mmap(
- __unused struct vnop_mmap_args /* {
- struct vnodeop_desc *a_desc;
- vnode_t a_vp;
- int a_fflags;
- kauth_cred_t a_cred;
- proc_t a_p;
- } */ *ap)
-{
-
- return (EINVAL);
-}
-
-/*
- * fsync vnode op. Just call nfs_flush() with commit == 1.
- */
-/* ARGSUSED */
-static int
-nfs_fsync(ap)
- struct vnop_fsync_args /* {
- struct vnodeop_desc *a_desc;
- vnode_t a_vp;
- int a_waitfor;
- vfs_context_t a_context;
- } */ *ap;
-{
- kauth_cred_t cred = vfs_context_ucred(ap->a_context);
- proc_t p = vfs_context_proc(ap->a_context);
- struct nfsnode *np = VTONFS(ap->a_vp);
- int error;
-
- np->n_flag |= NWRBUSY;
- error = nfs_flush(ap->a_vp, ap->a_waitfor, cred, p, 0);
- np->n_flag &= ~NWRBUSY;
- return (error);
-}
-
int
-nfs_flushcommits(vnode_t vp, proc_t p, int nowait)
+nfs3_lookup_rpc_async_finish(
+ nfsnode_t dnp,
+ __unused char *name,
+ __unused int namelen,
+ vfs_context_t ctx,
+ struct nfsreq *req,
+ u_int64_t *xidp,
+ fhandle_t *fhp,
+ struct nfs_vattr *nvap)
{
- struct nfsnode *np = VTONFS(vp);
- struct nfsbuf *bp;
- struct nfsbuflists blist, commitlist;
- int error = 0, retv, wcred_set, flags;
- u_quad_t off, endoff, toff;
- u_int32_t count;
- kauth_cred_t wcred = NULL;
+ int error = 0, lockerror = ENOENT, status, nfsvers, attrflag;
+ u_int64_t xid;
+ struct nfsmount *nmp;
+ struct nfsm_chain nmrep;
- FSDBG_TOP(557, vp, np, 0, 0);
+ nmp = NFSTONMP(dnp);
+ nfsvers = nmp->nm_vers;
- /*
- * A nb_flags == (NB_DELWRI | NB_NEEDCOMMIT) block has been written to the
- * server, but nas not been committed to stable storage on the server
- * yet. The byte range is worked out for as many nfsbufs as we can handle
- * and the commit rpc is done.
- */
- if (!LIST_EMPTY(&np->n_dirtyblkhd))
- np->n_flag |= NMODIFIED;
+ nfsm_chain_null(&nmrep);
- off = (u_quad_t)-1;
- endoff = 0;
- wcred_set = 0;
- LIST_INIT(&commitlist);
+ error = nfs_request_async_finish(req, &nmrep, xidp, &status);
- if (!VFSTONFS(vnode_mount(vp))) {
- error = ENXIO;
- goto done;
- }
- if (!NFS_ISV3(vp)) {
- error = EINVAL;
- goto done;
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+ xid = *xidp;
+ if (error || status) {
+ if (nfsvers == NFS_VER3)
+ nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid);
+ if (!error)
+ error = status;
+ goto nfsmout;
}
- flags = NBI_DIRTY;
- if (nowait)
- flags |= NBI_NOWAIT;
- lck_mtx_lock(nfs_buf_mutex);
- if (!nfs_buf_iterprepare(np, &blist, flags)) {
- while ((bp = LIST_FIRST(&blist))) {
- LIST_REMOVE(bp, nb_vnbufs);
- LIST_INSERT_HEAD(&np->n_dirtyblkhd, bp, nb_vnbufs);
- error = nfs_buf_acquire(bp, NBAC_NOWAIT, 0, 0);
- if (error)
- continue;
- if (ISSET(bp->nb_flags, NB_NEEDCOMMIT))
- nfs_buf_check_write_verifier(np, bp);
- if (((bp->nb_flags & (NB_DELWRI | NB_NEEDCOMMIT))
- != (NB_DELWRI | NB_NEEDCOMMIT))) {
- nfs_buf_drop(bp);
- continue;
- }
- nfs_buf_remfree(bp);
- lck_mtx_unlock(nfs_buf_mutex);
- /*
- * we need a upl to see if the page has been
- * dirtied (think mmap) since the unstable write, and
- * also to prevent vm from paging it during our commit rpc
- */
- if (!ISSET(bp->nb_flags, NB_PAGELIST)) {
- retv = nfs_buf_upl_setup(bp);
- if (retv) {
- /* unable to create upl */
- /* vm object must no longer exist */
- /* this could be fatal if we need */
- /* to write the data again, we'll see... */
- printf("nfs_flushcommits: upl create failed %d\n", retv);
- bp->nb_valid = bp->nb_dirty = 0;
- }
- }
- nfs_buf_upl_check(bp);
- lck_mtx_lock(nfs_buf_mutex);
-
- FSDBG(557, bp, bp->nb_flags, bp->nb_valid, bp->nb_dirty);
- FSDBG(557, bp->nb_validoff, bp->nb_validend,
- bp->nb_dirtyoff, bp->nb_dirtyend);
+ nfsmout_if(error || !fhp || !nvap);
- /*
- * We used to check for dirty pages here; if there were any
- * we'd abort the commit and force the entire buffer to be
- * written again.
- *
- * Instead of doing that, we now go ahead and commit the dirty
- * range, and then leave the buffer around with dirty pages
- * that will be written out later.
- */
-
- /*
- * Work out if all buffers are using the same cred
- * so we can deal with them all with one commit.
- *
- * XXX creds in bp's must be obtained by kauth_cred_ref on
- * the same original cred in order for them to be equal.
- */
- if (wcred_set == 0) {
- wcred = bp->nb_wcred;
- if (wcred == NOCRED)
- panic("nfs: needcommit w/out wcred");
- wcred_set = 1;
- } else if ((wcred_set == 1) && wcred != bp->nb_wcred) {
- wcred_set = -1;
- }
- SET(bp->nb_flags, NB_WRITEINPROG);
+ /* get the file handle */
+ nfsm_chain_get_fh(error, &nmrep, nfsvers, fhp);
- /*
- * A list of these buffers is kept so that the
- * second loop knows which buffers have actually
- * been committed. This is necessary, since there
- * may be a race between the commit rpc and new
- * uncommitted writes on the file.
- */
- LIST_REMOVE(bp, nb_vnbufs);
- LIST_INSERT_HEAD(&commitlist, bp, nb_vnbufs);
- toff = NBOFF(bp) + bp->nb_dirtyoff;
- if (toff < off)
- off = toff;
- toff += (u_quad_t)(bp->nb_dirtyend - bp->nb_dirtyoff);
- if (toff > endoff)
- endoff = toff;
- }
- nfs_buf_itercomplete(np, &blist, NBI_DIRTY);
+ /* get the attributes */
+ if (nfsvers == NFS_VER3) {
+ nfsm_chain_postop_attr_get(error, &nmrep, attrflag, nvap);
+ nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid);
+ if (!error && !attrflag)
+ error = nfs3_getattr_rpc(NULL, NFSTOMP(dnp), fhp->fh_data, fhp->fh_len, 0, ctx, nvap, xidp);
+ } else {
+ error = nfs_parsefattr(&nmrep, nfsvers, nvap);
}
- lck_mtx_unlock(nfs_buf_mutex);
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(dnp);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
- if (LIST_EMPTY(&commitlist)) {
- error = ENOBUFS;
- goto done;
- }
+/*
+ * Look up a file name and optionally either update the file handle or
+ * allocate an nfsnode, depending on the value of npp.
+ * npp == NULL --> just do the lookup
+ * *npp == NULL --> allocate a new nfsnode and make sure attributes are
+ * handled too
+ * *npp != NULL --> update the file handle in the vnode
+ */
+int
+nfs_lookitup(
+ nfsnode_t dnp,
+ char *name,
+ int namelen,
+ vfs_context_t ctx,
+ nfsnode_t *npp)
+{
+ int error = 0;
+ nfsnode_t np, newnp = NULL;
+ u_int64_t xid;
+ fhandle_t fh;
+ struct nfsmount *nmp;
+ struct nfs_vattr nvattr;
+ struct nfsreq rq, *req = &rq;
- /*
- * Commit data on the server, as required.
- * If all bufs are using the same wcred, then use that with
- * one call for all of them, otherwise commit each one
- * separately.
- */
- if (wcred_set == 1) {
- /*
- * Note, it's possible the commit range could be >2^32-1.
- * If it is, we'll send one commit that covers the whole file.
- */
- if ((endoff - off) > 0xffffffff)
- count = 0;
- else
- count = (endoff - off);
- retv = nfs_commit(vp, off, count, wcred, p);
- } else {
- retv = 0;
- LIST_FOREACH(bp, &commitlist, nb_vnbufs) {
- toff = NBOFF(bp) + bp->nb_dirtyoff;
- count = bp->nb_dirtyend - bp->nb_dirtyoff;
- retv = nfs_commit(vp, toff, count, bp->nb_wcred, p);
- if (retv)
- break;
- }
- }
+ nmp = NFSTONMP(dnp);
+ if (!nmp)
+ return (ENXIO);
- /*
- * Now, either mark the blocks I/O done or mark the
- * blocks dirty, depending on whether the commit
- * succeeded.
- */
- while ((bp = LIST_FIRST(&commitlist))) {
- LIST_REMOVE(bp, nb_vnbufs);
- FSDBG(557, bp, retv, bp->nb_flags, bp->nb_dirty);
- CLR(bp->nb_flags, (NB_NEEDCOMMIT | NB_WRITEINPROG));
- np->n_needcommitcnt--;
- CHECK_NEEDCOMMITCNT(np);
-
- if (retv) {
- /* move back to dirty list */
- lck_mtx_lock(nfs_buf_mutex);
- LIST_INSERT_HEAD(&VTONFS(vp)->n_dirtyblkhd, bp, nb_vnbufs);
- lck_mtx_unlock(nfs_buf_mutex);
- nfs_buf_release(bp, 1);
- continue;
- }
+ if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME) &&
+ (namelen > (int)nmp->nm_fsattr.nfsa_maxname))
+ return (ENAMETOOLONG);
- vnode_startwrite(vp);
- if (ISSET(bp->nb_flags, NB_DELWRI)) {
- OSAddAtomic(-1, (SInt32*)&nfs_nbdwrite);
- NFSBUFCNTCHK(0);
- wakeup(&nfs_nbdwrite);
- }
- CLR(bp->nb_flags, (NB_READ|NB_DONE|NB_ERROR|NB_DELWRI));
- /* if block still has dirty pages, we don't want it to */
- /* be released in nfs_buf_iodone(). So, don't set NB_ASYNC. */
- if (!bp->nb_dirty)
- SET(bp->nb_flags, NB_ASYNC);
+ NVATTR_INIT(&nvattr);
- /* move to clean list */
- lck_mtx_lock(nfs_buf_mutex);
- LIST_INSERT_HEAD(&VTONFS(vp)->n_cleanblkhd, bp, nb_vnbufs);
- lck_mtx_unlock(nfs_buf_mutex);
+ /* check for lookup of "." */
+ if ((name[0] == '.') && (namelen == 1)) {
+ /* skip lookup, we know who we are */
+ fh.fh_len = 0;
+ newnp = dnp;
+ goto nfsmout;
+ }
- bp->nb_dirtyoff = bp->nb_dirtyend = 0;
+ error = nmp->nm_funcs->nf_lookup_rpc_async(dnp, name, namelen, ctx, &req);
+ nfsmout_if(error);
+ error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, name, namelen, ctx, req, &xid, &fh, &nvattr);
+ nfsmout_if(!npp || error);
- nfs_buf_iodone(bp);
- if (bp->nb_dirty) {
- /* throw it back in as a delayed write buffer */
- CLR(bp->nb_flags, NB_DONE);
- nfs_buf_write_delayed(bp, p);
+ if (*npp) {
+ np = *npp;
+ if (fh.fh_len != np->n_fhsize) {
+ u_char *oldbuf = (np->n_fhsize > NFS_SMALLFH) ? np->n_fhp : NULL;
+ if (fh.fh_len > NFS_SMALLFH) {
+ MALLOC_ZONE(np->n_fhp, u_char *, fh.fh_len, M_NFSBIGFH, M_WAITOK);
+ if (!np->n_fhp) {
+ np->n_fhp = oldbuf;
+ error = ENOMEM;
+ goto nfsmout;
+ }
+ } else {
+ np->n_fhp = &np->n_fh[0];
+ }
+ if (oldbuf)
+ FREE_ZONE(oldbuf, np->n_fhsize, M_NFSBIGFH);
}
+ bcopy(fh.fh_data, np->n_fhp, fh.fh_len);
+ np->n_fhsize = fh.fh_len;
+ nfs_node_lock_force(np);
+ error = nfs_loadattrcache(np, &nvattr, &xid, 0);
+ nfs_node_unlock(np);
+ nfsmout_if(error);
+ newnp = np;
+ } else if (NFS_CMPFH(dnp, fh.fh_data, fh.fh_len)) {
+ nfs_node_lock_force(dnp);
+ if (dnp->n_xid <= xid)
+ error = nfs_loadattrcache(dnp, &nvattr, &xid, 0);
+ nfs_node_unlock(dnp);
+ nfsmout_if(error);
+ newnp = dnp;
+ } else {
+ struct componentname cn, *cnp = &cn;
+ bzero(cnp, sizeof(*cnp));
+ cnp->cn_nameptr = name;
+ cnp->cn_namelen = namelen;
+ error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh.fh_data, fh.fh_len,
+ &nvattr, &xid, rq.r_auth, NG_MAKEENTRY, &np);
+ nfsmout_if(error);
+ newnp = np;
}
-done:
- FSDBG_BOT(557, vp, np, 0, error);
+nfsmout:
+ if (npp && !*npp && !error)
+ *npp = newnp;
+ NVATTR_CLEANUP(&nvattr);
return (error);
}
/*
- * Flush all the blocks associated with a vnode.
- * Walk through the buffer pool and push any dirty pages
- * associated with the vnode.
+ * set up and initialize a "._" file lookup structure used for
+ * performing async lookups.
*/
-int
-nfs_flush(
- vnode_t vp,
- int waitfor,
- __unused kauth_cred_t cred,
- proc_t p,
- int ignore_writeerr)
+void
+nfs_dulookup_init(struct nfs_dulookup *dulp, nfsnode_t dnp, const char *name, int namelen, vfs_context_t ctx)
{
- struct nfsnode *np = VTONFS(vp);
- struct nfsbuf *bp;
- struct nfsbuflists blist;
- struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
- int error = 0, error2, slptimeo = 0, slpflag = 0;
- int flags, passone = 1;
-
- FSDBG_TOP(517, vp, np, waitfor, 0);
-
- if (!nmp) {
- error = ENXIO;
- goto done;
- }
- if (nmp->nm_flag & NFSMNT_INT)
- slpflag = PCATCH;
-
- /*
- * On the first pass, start async/unstable writes on all
- * delayed write buffers. Then wait for all writes to complete
- * and call nfs_flushcommits() to commit any uncommitted buffers.
- * On all subsequent passes, start STABLE writes on any remaining
- * dirty buffers. Then wait for all writes to complete.
- */
-again:
- lck_mtx_lock(nfs_buf_mutex);
- FSDBG(518, LIST_FIRST(&np->n_dirtyblkhd), np->n_flag, 0, 0);
- if (!LIST_EMPTY(&np->n_dirtyblkhd))
- np->n_flag |= NMODIFIED;
- if (!VFSTONFS(vnode_mount(vp))) {
- lck_mtx_unlock(nfs_buf_mutex);
- error = ENXIO;
- goto done;
- }
-
- /* Start/do any write(s) that are required. */
- if (!nfs_buf_iterprepare(np, &blist, NBI_DIRTY)) {
- while ((bp = LIST_FIRST(&blist))) {
- LIST_REMOVE(bp, nb_vnbufs);
- LIST_INSERT_HEAD(&np->n_dirtyblkhd, bp, nb_vnbufs);
- flags = (passone || (waitfor != MNT_WAIT)) ? NBAC_NOWAIT : 0;
- if (flags != NBAC_NOWAIT)
- nfs_buf_refget(bp);
- while ((error = nfs_buf_acquire(bp, flags, slpflag, slptimeo))) {
- FSDBG(524, bp, flags, bp->nb_lflags, bp->nb_flags);
- if (error == EBUSY)
- break;
- if (error) {
- error2 = nfs_sigintr(VFSTONFS(vnode_mount(vp)), NULL, p);
- if (error2) {
- if (flags != NBAC_NOWAIT)
- nfs_buf_refrele(bp);
- nfs_buf_itercomplete(np, &blist, NBI_DIRTY);
- lck_mtx_unlock(nfs_buf_mutex);
- error = error2;
- goto done;
- }
- if (slpflag == PCATCH) {
- slpflag = 0;
- slptimeo = 2 * hz;
- }
- }
- }
- if (flags != NBAC_NOWAIT)
- nfs_buf_refrele(bp);
- if (error == EBUSY)
- continue;
- if (!bp->nb_vp) {
- /* buffer is no longer valid */
- nfs_buf_drop(bp);
- continue;
- }
- if (ISSET(bp->nb_flags, NB_NEEDCOMMIT))
- nfs_buf_check_write_verifier(np, bp);
- if (!ISSET(bp->nb_flags, NB_DELWRI))
- panic("nfs_flush: not dirty");
- FSDBG(525, bp, passone, bp->nb_lflags, bp->nb_flags);
- if ((passone || (waitfor != MNT_WAIT)) &&
- ISSET(bp->nb_flags, NB_NEEDCOMMIT)) {
- nfs_buf_drop(bp);
- continue;
- }
- nfs_buf_remfree(bp);
- lck_mtx_unlock(nfs_buf_mutex);
- if (ISSET(bp->nb_flags, NB_ERROR)) {
- np->n_error = bp->nb_error ? bp->nb_error : EIO;
- np->n_flag |= NWRITEERR;
- nfs_buf_release(bp, 1);
- lck_mtx_lock(nfs_buf_mutex);
- continue;
- }
- SET(bp->nb_flags, NB_ASYNC);
- if (!passone) {
- /* NB_STABLE forces this to be written FILESYNC */
- SET(bp->nb_flags, NB_STABLE);
+ int error, du_namelen;
+ vnode_t du_vp;
+ struct nfsmount *nmp = NFSTONMP(dnp);
+
+ /* check for ._ file in name cache */
+ dulp->du_flags = 0;
+ bzero(&dulp->du_cn, sizeof(dulp->du_cn));
+ du_namelen = namelen + 2;
+ if (!nmp || NMFLAG(nmp, NONEGNAMECACHE))
+ return;
+ if ((namelen >= 2) && (name[0] == '.') && (name[1] == '_'))
+ return;
+ if (du_namelen >= (int)sizeof(dulp->du_smallname))
+ MALLOC(dulp->du_cn.cn_nameptr, char *, du_namelen + 1, M_TEMP, M_WAITOK);
+ else
+ dulp->du_cn.cn_nameptr = dulp->du_smallname;
+ if (!dulp->du_cn.cn_nameptr)
+ return;
+ dulp->du_cn.cn_namelen = du_namelen;
+ snprintf(dulp->du_cn.cn_nameptr, du_namelen + 1, "._%s", name);
+ dulp->du_cn.cn_nameptr[du_namelen] = '\0';
+ dulp->du_cn.cn_nameiop = LOOKUP;
+ dulp->du_cn.cn_flags = MAKEENTRY;
+
+ error = cache_lookup(NFSTOV(dnp), &du_vp, &dulp->du_cn);
+ if (error == -1) {
+ vnode_put(du_vp);
+ } else if (!error) {
+ nmp = NFSTONMP(dnp);
+ if (nmp && (nmp->nm_vers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) {
+ /* if rdirplus, try dir buf cache lookup */
+ nfsnode_t du_np = NULL;
+ if (!nfs_dir_buf_cache_lookup(dnp, &du_np, &dulp->du_cn, ctx, 0) && du_np) {
+ /* dir buf cache hit */
+ du_vp = NFSTOV(du_np);
+ vnode_put(du_vp);
+ error = -1;
}
- nfs_buf_write(bp);
- lck_mtx_lock(nfs_buf_mutex);
}
- nfs_buf_itercomplete(np, &blist, NBI_DIRTY);
+ if (!error)
+ dulp->du_flags |= NFS_DULOOKUP_DOIT;
}
- lck_mtx_unlock(nfs_buf_mutex);
+}
- if (waitfor == MNT_WAIT) {
- while ((error = vnode_waitforwrites(vp, 0, slpflag, slptimeo, "nfsflush"))) {
- error2 = nfs_sigintr(VFSTONFS(vnode_mount(vp)), NULL, p);
- if (error2) {
- error = error2;
- goto done;
- }
- if (slpflag == PCATCH) {
- slpflag = 0;
- slptimeo = 2 * hz;
- }
+/*
+ * start an async "._" file lookup request
+ */
+void
+nfs_dulookup_start(struct nfs_dulookup *dulp, nfsnode_t dnp, vfs_context_t ctx)
+{
+ struct nfsmount *nmp = NFSTONMP(dnp);
+ struct nfsreq *req = &dulp->du_req;
+
+ if (!nmp || !(dulp->du_flags & NFS_DULOOKUP_DOIT) || (dulp->du_flags & NFS_DULOOKUP_INPROG))
+ return;
+ if (!nmp->nm_funcs->nf_lookup_rpc_async(dnp, dulp->du_cn.cn_nameptr,
+ dulp->du_cn.cn_namelen, ctx, &req))
+ dulp->du_flags |= NFS_DULOOKUP_INPROG;
+}
+
+/*
+ * finish an async "._" file lookup request and clean up the structure
+ */
+void
+nfs_dulookup_finish(struct nfs_dulookup *dulp, nfsnode_t dnp, vfs_context_t ctx)
+{
+ struct nfsmount *nmp = NFSTONMP(dnp);
+ int error;
+ nfsnode_t du_np;
+ u_int64_t xid;
+ fhandle_t fh;
+ struct nfs_vattr nvattr;
+
+ if (!nmp || !(dulp->du_flags & NFS_DULOOKUP_INPROG))
+ goto out;
+
+ NVATTR_INIT(&nvattr);
+ error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, dulp->du_cn.cn_nameptr,
+ dulp->du_cn.cn_namelen, ctx, &dulp->du_req, &xid, &fh, &nvattr);
+ dulp->du_flags &= ~NFS_DULOOKUP_INPROG;
+ if (error == ENOENT) {
+ /* add a negative entry in the name cache */
+ nfs_node_lock_force(dnp);
+ cache_enter(NFSTOV(dnp), NULL, &dulp->du_cn);
+ dnp->n_flag |= NNEGNCENTRIES;
+ nfs_node_unlock(dnp);
+ } else if (!error) {
+ error = nfs_nget(NFSTOMP(dnp), dnp, &dulp->du_cn, fh.fh_data, fh.fh_len,
+ &nvattr, &xid, dulp->du_req.r_auth, NG_MAKEENTRY, &du_np);
+ if (!error) {
+ nfs_node_unlock(du_np);
+ vnode_put(NFSTOV(du_np));
}
}
+ NVATTR_CLEANUP(&nvattr);
+out:
+ if (dulp->du_flags & NFS_DULOOKUP_INPROG)
+ nfs_request_async_cancel(&dulp->du_req);
+ if (dulp->du_cn.cn_nameptr && (dulp->du_cn.cn_nameptr != dulp->du_smallname))
+ FREE(dulp->du_cn.cn_nameptr, M_TEMP);
+}
- if (NFS_ISV3(vp)) {
- /* loop while it looks like there are still buffers to be */
- /* commited and nfs_flushcommits() seems to be handling them. */
- while (np->n_needcommitcnt)
- if (nfs_flushcommits(vp, p, 0))
- break;
- }
- if (passone) {
- passone = 0;
- goto again;
- }
+/*
+ * NFS Version 3 commit RPC
+ */
+int
+nfs3_commit_rpc(
+ nfsnode_t np,
+ uint64_t offset,
+ uint64_t count,
+ kauth_cred_t cred,
+ uint64_t wverf)
+{
+ struct nfsmount *nmp;
+ int error = 0, lockerror, status, wccpostattr = 0, nfsvers;
+ struct timespec premtime = { 0, 0 };
+ u_int64_t xid, newwverf;
+ uint32_t count32;
+ struct nfsm_chain nmreq, nmrep;
- if (waitfor == MNT_WAIT) {
- if (!LIST_EMPTY(&np->n_dirtyblkhd))
- goto again;
- /* if we have no dirty blocks, we can clear the modified flag */
- np->n_flag &= ~NMODIFIED;
- }
+ nmp = NFSTONMP(np);
+ FSDBG(521, np, offset, count, nmp ? nmp->nm_state : 0);
+ if (!nmp)
+ return (ENXIO);
+ if (!(nmp->nm_state & NFSSTA_HASWRITEVERF))
+ return (0);
+ nfsvers = nmp->nm_vers;
- FSDBG(526, np->n_flag, np->n_error, 0, 0);
- if (!ignore_writeerr && (np->n_flag & NWRITEERR)) {
- error = np->n_error;
- np->n_flag &= ~NWRITEERR;
- }
-done:
- FSDBG_BOT(517, vp, np, error, 0);
+ if (count > UINT32_MAX)
+ count32 = 0;
+ else
+ count32 = count;
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(NFS_VER3));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
+ nfsm_chain_add_64(error, &nmreq, offset);
+ nfsm_chain_add_32(error, &nmreq, count32);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request2(np, NULL, &nmreq, NFSPROC_COMMIT,
+ current_thread(), cred, NULL, 0, &nmrep, &xid, &status);
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ /* can we do anything useful with the wcc info? */
+ nfsm_chain_get_wcc_data(error, &nmrep, np, &premtime, &wccpostattr, &xid);
+ if (!lockerror)
+ nfs_node_unlock(np);
+ if (!error)
+ error = status;
+ nfsm_chain_get_64(error, &nmrep, newwverf);
+ nfsmout_if(error);
+ lck_mtx_lock(&nmp->nm_lock);
+ if (nmp->nm_verf != newwverf)
+ nmp->nm_verf = newwverf;
+ if (wverf != newwverf)
+ error = NFSERR_STALEWRITEVERF;
+ lck_mtx_unlock(&nmp->nm_lock);
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
return (error);
}
+
+int
+nfs_vnop_blockmap(
+ __unused struct vnop_blockmap_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ off_t a_foffset;
+ size_t a_size;
+ daddr64_t *a_bpn;
+ size_t *a_run;
+ void *a_poff;
+ int a_flags;
+ } */ *ap)
+{
+ return (ENOTSUP);
+}
+
+
+/*
+ * fsync vnode op. Just call nfs_flush().
+ */
+/* ARGSUSED */
+int
+nfs_vnop_fsync(
+ struct vnop_fsync_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ int a_waitfor;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ return (nfs_flush(VTONFS(ap->a_vp), ap->a_waitfor, vfs_context_thread(ap->a_context), 0));
+}
+
+
/*
- * Do an nfs pathconf rpc.
+ * Do an NFS pathconf RPC.
*/
int
-nfs_pathconfrpc(
- vnode_t vp,
- struct nfsv3_pathconf *pc,
- kauth_cred_t cred,
- proc_t procp)
+nfs3_pathconf_rpc(
+ nfsnode_t np,
+ struct nfs_fsattr *nfsap,
+ vfs_context_t ctx)
{
- mbuf_t mreq, mrep, md, mb, mb2;
- caddr_t bpos, dpos, cp, cp2;
- int32_t t1, t2;
- u_long *tl;
u_int64_t xid;
- int attrflag, error = 0;
- struct nfsv3_pathconf *mpc;
+ int error = 0, lockerror, status, nfsvers;
+ struct nfsm_chain nmreq, nmrep;
+ struct nfsmount *nmp = NFSTONMP(np);
+ uint32_t val = 0;
- /* fetch pathconf info from server */
- nfsm_reqhead(NFSX_FH(1));
- if (error)
- return (error);
- nfsm_fhtom(vp, 1);
- nfsm_request(vp, NFSPROC_PATHCONF, procp, cred, &xid);
- nfsm_postop_attr_update(vp, 1, attrflag, &xid);
- if (!error) {
- nfsm_dissect(mpc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
- pc->pc_linkmax = fxdr_unsigned(long, mpc->pc_linkmax);
- pc->pc_namemax = fxdr_unsigned(long, mpc->pc_namemax);
- pc->pc_chownrestricted = fxdr_unsigned(long, mpc->pc_chownrestricted);
- pc->pc_notrunc = fxdr_unsigned(long, mpc->pc_notrunc);
- pc->pc_caseinsensitive = fxdr_unsigned(long, mpc->pc_caseinsensitive);
- pc->pc_casepreserving = fxdr_unsigned(long, mpc->pc_casepreserving);
- }
- nfsm_reqdone;
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ /* fetch pathconf info from server */
+ nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(NFS_VER3));
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsmout_if(error);
+ error = nfs_request(np, NULL, &nmreq, NFSPROC_PATHCONF, ctx, NULL, &nmrep, &xid, &status);
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
+ if (!lockerror)
+ nfs_node_unlock(np);
+ if (!error)
+ error = status;
+ nfsm_chain_get_32(error, &nmrep, nfsap->nfsa_maxlink);
+ nfsm_chain_get_32(error, &nmrep, nfsap->nfsa_maxname);
+ nfsm_chain_get_32(error, &nmrep, val);
+ if (val)
+ nfsap->nfsa_flags |= NFS_FSFLAG_NO_TRUNC;
+ nfsm_chain_get_32(error, &nmrep, val);
+ if (val)
+ nfsap->nfsa_flags |= NFS_FSFLAG_CHOWN_RESTRICTED;
+ nfsm_chain_get_32(error, &nmrep, val);
+ if (val)
+ nfsap->nfsa_flags |= NFS_FSFLAG_CASE_INSENSITIVE;
+ nfsm_chain_get_32(error, &nmrep, val);
+ if (val)
+ nfsap->nfsa_flags |= NFS_FSFLAG_CASE_PRESERVING;
+ NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_MAXLINK);
+ NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_MAXNAME);
+ NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_NO_TRUNC);
+ NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_CHOWN_RESTRICTED);
+ NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_INSENSITIVE);
+ NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_PRESERVING);
+nfsmout:
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
return (error);
}
+/* save pathconf info for NFSv3 mount */
void
-nfs_pathconf_cache(struct nfsmount *nmp, struct nfsv3_pathconf *pc)
+nfs3_pathconf_cache(struct nfsmount *nmp, struct nfs_fsattr *nfsap)
{
+ nmp->nm_fsattr.nfsa_maxlink = nfsap->nfsa_maxlink;
+ nmp->nm_fsattr.nfsa_maxname = nfsap->nfsa_maxname;
+ nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_NO_TRUNC;
+ nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_CHOWN_RESTRICTED;
+ nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_CASE_INSENSITIVE;
+ nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_CASE_PRESERVING;
+ NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXLINK);
+ NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME);
+ NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_NO_TRUNC);
+ NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_CHOWN_RESTRICTED);
+ NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_CASE_INSENSITIVE);
+ NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_CASE_PRESERVING);
nmp->nm_state |= NFSSTA_GOTPATHCONF;
- nmp->nm_fsinfo.linkmax = pc->pc_linkmax;
- nmp->nm_fsinfo.namemax = pc->pc_namemax;
- nmp->nm_fsinfo.pcflags = 0;
- if (pc->pc_notrunc)
- nmp->nm_fsinfo.pcflags |= NFSPCINFO_NOTRUNC;
- if (pc->pc_chownrestricted)
- nmp->nm_fsinfo.pcflags |= NFSPCINFO_CHOWN_RESTRICTED;
- if (pc->pc_caseinsensitive)
- nmp->nm_fsinfo.pcflags |= NFSPCINFO_CASE_INSENSITIVE;
- if (pc->pc_casepreserving)
- nmp->nm_fsinfo.pcflags |= NFSPCINFO_CASE_PRESERVING;
}
/*
* for V2.
*/
/* ARGSUSED */
-static int
-nfs_pathconf(ap)
+int
+nfs_vnop_pathconf(
struct vnop_pathconf_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
int a_name;
- register_t *a_retval;
+ int32_t *a_retval;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
vnode_t vp = ap->a_vp;
+ nfsnode_t np = VTONFS(vp);
struct nfsmount *nmp;
- struct nfsv3_pathconf pc;
- int error = 0, cached;
+ struct nfs_fsattr nfsa, *nfsap;
+ int error = 0;
+ uint64_t maxFileSize;
+ uint nbits;
- nmp = VFSTONFS(vnode_mount(vp));
+ nmp = VTONMP(vp);
if (!nmp)
return (ENXIO);
- if (!NFS_ISV3(vp))
- return (EINVAL);
switch (ap->a_name) {
case _PC_LINK_MAX:
case _PC_CASE_SENSITIVE:
case _PC_CASE_PRESERVING:
break;
+ case _PC_FILESIZEBITS:
+ if (nmp->nm_vers == NFS_VER2) {
+ *ap->a_retval = 32;
+ return (0);
+ }
+ break;
+ case _PC_XATTR_SIZE_BITS:
+ /* Do we support xattrs natively? */
+ if (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR)
+ break; /* Yes */
+ /* No... so just return an error */
+ /* FALLTHROUGH */
default:
/* don't bother contacting the server if we know the answer */
return (EINVAL);
}
- if (!(nmp->nm_state & NFSSTA_GOTPATHCONF)) {
+ if (nmp->nm_vers == NFS_VER2)
+ return (EINVAL);
+
+ lck_mtx_lock(&nmp->nm_lock);
+ if (nmp->nm_vers == NFS_VER3) {
+ if (!(nmp->nm_state & NFSSTA_GOTPATHCONF)) {
+ /* no pathconf info cached */
+ lck_mtx_unlock(&nmp->nm_lock);
+ NFS_CLEAR_ATTRIBUTES(nfsa.nfsa_bitmap);
+ error = nfs3_pathconf_rpc(np, &nfsa, ap->a_context);
+ if (error)
+ return (error);
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+ lck_mtx_lock(&nmp->nm_lock);
+ if (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS) {
+ /* all files have the same pathconf info, */
+ /* so cache a copy of the results */
+ nfs3_pathconf_cache(nmp, &nfsa);
+ }
+ nfsap = &nfsa;
+ } else {
+ nfsap = &nmp->nm_fsattr;
+ }
+ } else if (!(nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS)) {
/* no pathconf info cached */
- kauth_cred_t cred = vfs_context_ucred(ap->a_context);
- proc_t p = vfs_context_proc(ap->a_context);
- error = nfs_pathconfrpc(vp, &pc, cred, p);
+ lck_mtx_unlock(&nmp->nm_lock);
+ NFS_CLEAR_ATTRIBUTES(nfsa.nfsa_bitmap);
+ error = nfs4_pathconf_rpc(np, &nfsa, ap->a_context);
if (error)
return (error);
- nmp = VFSTONFS(vnode_mount(vp));
+ nmp = VTONMP(vp);
if (!nmp)
return (ENXIO);
- if (!(nmp->nm_state & NFSSTA_GOTFSINFO)) {
- nfs_fsinfo(nmp, vp, cred, p);
- nmp = VFSTONFS(vnode_mount(vp));
- if (!nmp)
- return (ENXIO);
- }
- if ((nmp->nm_state & NFSSTA_GOTFSINFO) &&
- (nmp->nm_fsinfo.fsproperties & NFSV3FSINFO_HOMOGENEOUS)) {
- /* all files have the same pathconf info, */
- /* so cache a copy of the results */
- nfs_pathconf_cache(nmp, &pc);
- }
+ lck_mtx_lock(&nmp->nm_lock);
+ nfsap = &nfsa;
+ } else {
+ nfsap = &nmp->nm_fsattr;
}
- cached = (nmp->nm_state & NFSSTA_GOTPATHCONF);
-
switch (ap->a_name) {
case _PC_LINK_MAX:
- *ap->a_retval = cached ? nmp->nm_fsinfo.linkmax : pc.pc_linkmax;
+ if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_MAXLINK))
+ *ap->a_retval = nfsap->nfsa_maxlink;
+ else if ((nmp->nm_vers == NFS_VER4) && NFS_BITMAP_ISSET(np->n_vattr.nva_bitmap, NFS_FATTR_MAXLINK))
+ *ap->a_retval = np->n_vattr.nva_maxlink;
+ else
+ error = EINVAL;
break;
case _PC_NAME_MAX:
- *ap->a_retval = cached ? nmp->nm_fsinfo.namemax : pc.pc_namemax;
+ if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_MAXNAME))
+ *ap->a_retval = nfsap->nfsa_maxname;
+ else
+ error = EINVAL;
break;
case _PC_CHOWN_RESTRICTED:
- if (cached)
- *ap->a_retval = (nmp->nm_fsinfo.pcflags & NFSPCINFO_CHOWN_RESTRICTED) ? 1 : 0;
+ if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CHOWN_RESTRICTED))
+ *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CHOWN_RESTRICTED) ? 200112 /* _POSIX_CHOWN_RESTRICTED */ : 0;
else
- *ap->a_retval = pc.pc_chownrestricted;
+ error = EINVAL;
break;
case _PC_NO_TRUNC:
- if (cached)
- *ap->a_retval = (nmp->nm_fsinfo.pcflags & NFSPCINFO_NOTRUNC) ? 1 : 0;
+ if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_NO_TRUNC))
+ *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_NO_TRUNC) ? 200112 /* _POSIX_NO_TRUNC */ : 0;
else
- *ap->a_retval = pc.pc_notrunc;
+ error = EINVAL;
break;
case _PC_CASE_SENSITIVE:
- if (cached)
- *ap->a_retval = (nmp->nm_fsinfo.pcflags & NFSPCINFO_CASE_INSENSITIVE) ? 0 : 1;
+ if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_INSENSITIVE))
+ *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CASE_INSENSITIVE) ? 0 : 1;
else
- *ap->a_retval = !pc.pc_caseinsensitive;
+ error = EINVAL;
break;
case _PC_CASE_PRESERVING:
- if (cached)
- *ap->a_retval = (nmp->nm_fsinfo.pcflags & NFSPCINFO_CASE_PRESERVING) ? 1 : 0;
+ if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_PRESERVING))
+ *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CASE_PRESERVING) ? 1 : 0;
else
- *ap->a_retval = pc.pc_casepreserving;
+ error = EINVAL;
+ break;
+ case _PC_XATTR_SIZE_BITS: /* same as file size bits if named attrs supported */
+ case _PC_FILESIZEBITS:
+ if (!NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_MAXFILESIZE)) {
+ *ap->a_retval = 64;
+ error = 0;
+ break;
+ }
+ maxFileSize = nfsap->nfsa_maxfilesize;
+ nbits = 1;
+ if (maxFileSize & 0xffffffff00000000ULL) {
+ nbits += 32;
+ maxFileSize >>= 32;
+ }
+ if (maxFileSize & 0xffff0000) {
+ nbits += 16;
+ maxFileSize >>= 16;
+ }
+ if (maxFileSize & 0xff00) {
+ nbits += 8;
+ maxFileSize >>= 8;
+ }
+ if (maxFileSize & 0xf0) {
+ nbits += 4;
+ maxFileSize >>= 4;
+ }
+ if (maxFileSize & 0xc) {
+ nbits += 2;
+ maxFileSize >>= 2;
+ }
+ if (maxFileSize & 0x2) {
+ nbits += 1;
+ }
+ *ap->a_retval = nbits;
break;
default:
error = EINVAL;
}
- return (error);
-}
-
-/*
- * NFS advisory byte-level locks (client)
- */
-static int
-nfs_advlock(ap)
- struct vnop_advlock_args /* {
- struct vnodeop_desc *a_desc;
- vnode_t a_vp;
- caddr_t a_id;
- int a_op;
- struct flock *a_fl;
- int a_flags;
- vfs_context_t a_context;
- } */ *ap;
-{
- return (nfs_dolock(ap));
-}
-
-/*
- * write (or commit) the given NFS buffer
- */
-int
-nfs_buf_write(struct nfsbuf *bp)
-{
- int oldflags = bp->nb_flags, rv = 0;
- vnode_t vp = bp->nb_vp;
- struct nfsnode *np = VTONFS(vp);
- kauth_cred_t cr;
- proc_t p = current_proc(); // XXX
-
- FSDBG_TOP(553, bp, NBOFF(bp), bp->nb_flags, 0);
-
- if (!ISSET(bp->nb_lflags, NBL_BUSY))
- panic("nfs_buf_write: buffer is not busy???");
-
- CLR(bp->nb_flags, (NB_READ|NB_DONE|NB_ERROR|NB_DELWRI));
- if (ISSET(oldflags, NB_DELWRI)) {
- OSAddAtomic(-1, (SInt32*)&nfs_nbdwrite);
- NFSBUFCNTCHK(0);
- wakeup(&nfs_nbdwrite);
- }
-
- /* move to clean list */
- if (ISSET(oldflags, (NB_ASYNC|NB_DELWRI))) {
- lck_mtx_lock(nfs_buf_mutex);
- if (bp->nb_vnbufs.le_next != NFSNOLIST)
- LIST_REMOVE(bp, nb_vnbufs);
- LIST_INSERT_HEAD(&VTONFS(vp)->n_cleanblkhd, bp, nb_vnbufs);
- lck_mtx_unlock(nfs_buf_mutex);
- }
- vnode_startwrite(vp);
-
- if (p && p->p_stats)
- p->p_stats->p_ru.ru_oublock++;
-
- /*
- * For async requests when nfsiod(s) are running, queue the request by
- * calling nfs_asyncio(), otherwise just all nfs_doio() to do the request.
- */
- if (ISSET(bp->nb_flags, NB_ASYNC))
- p = NULL;
- if (ISSET(bp->nb_flags, NB_READ))
- cr = bp->nb_rcred;
- else
- cr = bp->nb_wcred;
- if (!ISSET(bp->nb_flags, NB_ASYNC) || nfs_asyncio(bp, NOCRED))
- rv = nfs_doio(bp, cr, p);
-
- if ((oldflags & NB_ASYNC) == 0) {
- rv = nfs_buf_iowait(bp);
- /* move to clean list */
- if (oldflags & NB_DELWRI) {
- lck_mtx_lock(nfs_buf_mutex);
- if (bp->nb_vnbufs.le_next != NFSNOLIST)
- LIST_REMOVE(bp, nb_vnbufs);
- LIST_INSERT_HEAD(&VTONFS(vp)->n_cleanblkhd, bp, nb_vnbufs);
- lck_mtx_unlock(nfs_buf_mutex);
- }
- oldflags = bp->nb_flags;
- FSDBG_BOT(553, bp, NBOFF(bp), bp->nb_flags, rv);
- if (cr) {
- kauth_cred_ref(cr);
- }
- nfs_buf_release(bp, 1);
- if (ISSET(oldflags, NB_ERROR) && !(np->n_flag & NFLUSHINPROG)) {
- /*
- * There was a write error and we need to
- * invalidate attrs and flush buffers in
- * order to sync up with the server.
- * (if this write was extending the file,
- * we may no longer know the correct size)
- *
- * But we couldn't call vinvalbuf while holding
- * the buffer busy. So we call vinvalbuf() after
- * releasing the buffer.
- */
- nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, cr, p, 1);
- }
- if (cr)
- kauth_cred_rele(cr);
- return (rv);
- }
+ lck_mtx_unlock(&nmp->nm_lock);
- FSDBG_BOT(553, bp, NBOFF(bp), bp->nb_flags, rv);
- return (rv);
+ return (error);
}
/*
* Read wrapper for special devices.
*/
-static int
-nfsspec_read(ap)
+int
+nfsspec_vnop_read(
struct vnop_read_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
struct uio *a_uio;
int a_ioflag;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
- register struct nfsnode *np = VTONFS(ap->a_vp);
+ nfsnode_t np = VTONFS(ap->a_vp);
struct timeval now;
+ int error;
/*
* Set access flag.
*/
+ if ((error = nfs_node_lock(np)))
+ return (error);
np->n_flag |= NACC;
microtime(&now);
np->n_atim.tv_sec = now.tv_sec;
np->n_atim.tv_nsec = now.tv_usec * 1000;
+ nfs_node_unlock(np);
return (VOCALL(spec_vnodeop_p, VOFFSET(vnop_read), ap));
}
/*
* Write wrapper for special devices.
*/
-static int
-nfsspec_write(ap)
+int
+nfsspec_vnop_write(
struct vnop_write_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
struct uio *a_uio;
int a_ioflag;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
- register struct nfsnode *np = VTONFS(ap->a_vp);
+ nfsnode_t np = VTONFS(ap->a_vp);
struct timeval now;
+ int error;
/*
* Set update flag.
*/
+ if ((error = nfs_node_lock(np)))
+ return (error);
np->n_flag |= NUPD;
microtime(&now);
np->n_mtim.tv_sec = now.tv_sec;
np->n_mtim.tv_nsec = now.tv_usec * 1000;
+ nfs_node_unlock(np);
return (VOCALL(spec_vnodeop_p, VOFFSET(vnop_write), ap));
}
*
* Update the times on the nfsnode then do device close.
*/
-static int
-nfsspec_close(ap)
+int
+nfsspec_vnop_close(
struct vnop_close_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
int a_fflag;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
vnode_t vp = ap->a_vp;
- struct nfsnode *np = VTONFS(vp);
+ nfsnode_t np = VTONFS(vp);
struct vnode_attr vattr;
mount_t mp;
+ int error;
+ if ((error = nfs_node_lock(np)))
+ return (error);
if (np->n_flag & (NACC | NUPD)) {
np->n_flag |= NCHG;
- if (!vnode_isinuse(vp, 1) && (mp = vnode_mount(vp)) && !vfs_isrdonly(mp)) {
+ if (!vnode_isinuse(vp, 0) && (mp = vnode_mount(vp)) && !vfs_isrdonly(mp)) {
VATTR_INIT(&vattr);
if (np->n_flag & NACC) {
vattr.va_access_time = np->n_atim;
vattr.va_modify_time = np->n_mtim;
VATTR_SET_ACTIVE(&vattr, va_modify_time);
}
+ nfs_node_unlock(np);
vnode_setattr(vp, &vattr, ap->a_context);
+ } else {
+ nfs_node_unlock(np);
}
+ } else {
+ nfs_node_unlock(np);
}
return (VOCALL(spec_vnodeop_p, VOFFSET(vnop_close), ap));
}
+#if FIFO
extern vnop_t **fifo_vnodeop_p;
/*
* Read wrapper for fifos.
*/
-static int
-nfsfifo_read(ap)
+int
+nfsfifo_vnop_read(
struct vnop_read_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
struct uio *a_uio;
int a_ioflag;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
- register struct nfsnode *np = VTONFS(ap->a_vp);
+ nfsnode_t np = VTONFS(ap->a_vp);
struct timeval now;
+ int error;
/*
* Set access flag.
*/
+ if ((error = nfs_node_lock(np)))
+ return (error);
np->n_flag |= NACC;
microtime(&now);
np->n_atim.tv_sec = now.tv_sec;
np->n_atim.tv_nsec = now.tv_usec * 1000;
+ nfs_node_unlock(np);
return (VOCALL(fifo_vnodeop_p, VOFFSET(vnop_read), ap));
}
/*
* Write wrapper for fifos.
*/
-static int
-nfsfifo_write(ap)
+int
+nfsfifo_vnop_write(
struct vnop_write_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
struct uio *a_uio;
int a_ioflag;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
- register struct nfsnode *np = VTONFS(ap->a_vp);
+ nfsnode_t np = VTONFS(ap->a_vp);
struct timeval now;
+ int error;
/*
* Set update flag.
*/
+ if ((error = nfs_node_lock(np)))
+ return (error);
np->n_flag |= NUPD;
microtime(&now);
np->n_mtim.tv_sec = now.tv_sec;
np->n_mtim.tv_nsec = now.tv_usec * 1000;
+ nfs_node_unlock(np);
return (VOCALL(fifo_vnodeop_p, VOFFSET(vnop_write), ap));
}
*
* Update the times on the nfsnode then do fifo close.
*/
-static int
-nfsfifo_close(ap)
+int
+nfsfifo_vnop_close(
struct vnop_close_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
int a_fflag;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
vnode_t vp = ap->a_vp;
- struct nfsnode *np = VTONFS(vp);
+ nfsnode_t np = VTONFS(vp);
struct vnode_attr vattr;
struct timeval now;
mount_t mp;
+ int error;
+ if ((error = nfs_node_lock(np)))
+ return (error);
if (np->n_flag & (NACC | NUPD)) {
microtime(&now);
if (np->n_flag & NACC) {
vattr.va_modify_time = np->n_mtim;
VATTR_SET_ACTIVE(&vattr, va_modify_time);
}
+ nfs_node_unlock(np);
vnode_setattr(vp, &vattr, ap->a_context);
+ } else {
+ nfs_node_unlock(np);
}
+ } else {
+ nfs_node_unlock(np);
}
return (VOCALL(fifo_vnodeop_p, VOFFSET(vnop_close), ap));
}
+#endif /* FIFO */
/*ARGSUSED*/
-static int
-nfs_ioctl(
- __unused struct vnop_ioctl_args /* {
+int
+nfs_vnop_ioctl(
+ struct vnop_ioctl_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
- u_long a_command;
+ u_int32_t a_command;
caddr_t a_data;
int a_fflag;
- kauth_cred_t a_cred;
- proc_t a_p;
+ vfs_context_t a_context;
} */ *ap)
{
+ vfs_context_t ctx = ap->a_context;
+ vnode_t vp = ap->a_vp;
+ struct nfsmount *mp = VTONMP(vp);
+ int error = ENOTTY;
- /*
- * XXX we were once bogusly enoictl() which returned this (ENOTTY).
- * Probably we should return ENODEV.
- */
- return (ENOTTY);
+ if (mp == NULL)
+ return (ENXIO);
+
+ switch (ap->a_command) {
+
+ case F_FULLFSYNC:
+ if (vnode_vfsisrdonly(vp))
+ return (EROFS);
+ error = nfs_flush(VTONFS(vp), MNT_WAIT, vfs_context_thread(ctx), 0);
+ break;
+ case NFS_FSCTL_DESTROY_CRED:
+ error = nfs_gss_clnt_ctx_destroy(mp, vfs_context_ucred(ctx));
+ break;
+ }
+
+ return (error);
}
/*ARGSUSED*/
-static int
-nfs_select(
+int
+nfs_vnop_select(
__unused struct vnop_select_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
int a_which;
int a_fflags;
- kauth_cred_t a_cred;
void *a_wql;
- proc_t a_p;
+ vfs_context_t a_context;
} */ *ap)
{
}
/*
- * Vnode op for pagein using getblk_pages
- * derived from nfs_bioread()
- * No read aheads are started from pagein operation
+ * vnode OP for pagein using UPL
+ *
+ * No buffer I/O, just RPCs straight into the mapped pages.
*/
-static int
-nfs_pagein(ap)
+int
+nfs_vnop_pagein(
struct vnop_pagein_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
size_t a_size;
int a_flags;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
vnode_t vp = ap->a_vp;
upl_t pl = ap->a_pl;
- size_t size= ap->a_size;
+ size_t size = ap->a_size;
off_t f_offset = ap->a_f_offset;
vm_offset_t pl_offset = ap->a_pl_offset;
- int flags = ap->a_flags;
+ int flags = ap->a_flags;
+ thread_t thd;
kauth_cred_t cred;
- proc_t p;
- struct nfsnode *np = VTONFS(vp);
- int biosize, xsize, iosize;
+ nfsnode_t np = VTONFS(vp);
+ size_t nmrsize, iosize, txsize, rxsize, retsize;
+ off_t txoffset;
struct nfsmount *nmp;
int error = 0;
- vm_offset_t ioaddr;
- struct uio auio;
- struct iovec_32 aiov;
- struct uio * uio = &auio;
+ vm_offset_t ioaddr, rxaddr;
+ uio_t uio;
+ char uio_buf [ UIO_SIZEOF(1) ];
int nofreeupl = flags & UPL_NOCOMMIT;
upl_page_info_t *plinfo;
+#define MAXPAGINGREQS 16 /* max outstanding RPCs for pagein/pageout */
+ struct nfsreq *req[MAXPAGINGREQS];
+ int nextsend, nextwait;
+ uint32_t stategenid = 0, restart = 0;
+ kern_return_t kret;
- FSDBG(322, vp, f_offset, size, flags);
+ FSDBG(322, np, f_offset, size, flags);
if (pl == (upl_t)NULL)
panic("nfs_pagein: no upl");
- if (UBCINVALID(vp)) {
- printf("nfs_pagein: invalid vnode 0x%x", (int)vp);
- if (!nofreeupl)
- (void) ubc_upl_abort(pl, 0);
- return (EPERM);
- }
- UBCINFOCHECK("nfs_pagein", vp);
-
if (size <= 0) {
- printf("nfs_pagein: invalid size %d", size);
+ printf("nfs_pagein: invalid size %ld", size);
if (!nofreeupl)
- (void) ubc_upl_abort(pl, 0);
+ (void) ubc_upl_abort_range(pl, pl_offset, size, 0);
return (EINVAL);
}
if (f_offset < 0 || f_offset >= (off_t)np->n_size || (f_offset & PAGE_MASK_64)) {
if (!nofreeupl)
- ubc_upl_abort_range(pl, pl_offset, size,
+ ubc_upl_abort_range(pl, pl_offset, size,
UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
return (EINVAL);
}
+ thd = vfs_context_thread(ap->a_context);
cred = ubc_getcred(vp);
- if (cred == NOCRED)
+ if (!IS_VALID_CRED(cred))
cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
- auio.uio_offset = f_offset;
-#if 1 /* LP64todo - can't use new segment flags until the drivers are ready */
- auio.uio_segflg = UIO_SYSSPACE;
-#else
- auio.uio_segflg = UIO_SYSSPACE32;
-#endif
- auio.uio_rw = UIO_READ;
- auio.uio_procp = p;
+ uio = uio_createwithbuffer(1, f_offset, UIO_SYSSPACE, UIO_READ,
+ &uio_buf, sizeof(uio_buf));
- nmp = VFSTONFS(vnode_mount(vp));
+ nmp = VTONMP(vp);
if (!nmp) {
if (!nofreeupl)
- ubc_upl_abort_range(pl, pl_offset, size,
+ ubc_upl_abort_range(pl, pl_offset, size,
UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
return (ENXIO);
}
- if ((nmp->nm_flag & NFSMNT_NFSV3) && !(nmp->nm_state & NFSSTA_GOTFSINFO))
- (void)nfs_fsinfo(nmp, vp, cred, p);
- biosize = vfs_statfs(vnode_mount(vp))->f_iosize;
+ nmrsize = nmp->nm_rsize;
plinfo = ubc_upl_pageinfo(pl);
- ubc_upl_map(pl, &ioaddr);
+ kret = ubc_upl_map(pl, &ioaddr);
+ if (kret != KERN_SUCCESS)
+ panic("nfs_vnop_pagein: ubc_upl_map() failed with (%d)", kret);
ioaddr += pl_offset;
- xsize = size;
+tryagain:
+ if (nmp->nm_vers >= NFS_VER4)
+ stategenid = nmp->nm_stategenid;
+ txsize = rxsize = size;
+ txoffset = f_offset;
+ rxaddr = ioaddr;
+
+ bzero(req, sizeof(req));
+ nextsend = nextwait = 0;
do {
- /*
- * It would be nice to be able to issue all these requests
- * in parallel instead of waiting for each one to complete
- * before sending the next one.
- * XXX Should we align these requests to block boundaries?
- */
- iosize = min(biosize, xsize);
- aiov.iov_len = iosize;
- aiov.iov_base = (uintptr_t)ioaddr;
- auio.uio_iovs.iov32p = &aiov;
- auio.uio_iovcnt = 1;
- uio_uio_resid_set(&auio, iosize);
-
- FSDBG(322, uio->uio_offset, uio_uio_resid(uio), ioaddr, xsize);
- /*
- * With UBC we get here only when the file data is not in the VM
- * page cache, so go ahead and read in.
- */
-#ifdef UPL_DEBUG
- upl_ubc_alias_set(pl, current_thread(), 2);
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ break;
+ }
+ /* send requests while we need to and have available slots */
+ while ((txsize > 0) && (req[nextsend] == NULL)) {
+ iosize = MIN(nmrsize, txsize);
+ if ((error = nmp->nm_funcs->nf_read_rpc_async(np, txoffset, iosize, thd, cred, NULL, &req[nextsend]))) {
+ req[nextsend] = NULL;
+ break;
+ }
+ txoffset += iosize;
+ txsize -= iosize;
+ nextsend = (nextsend + 1) % MAXPAGINGREQS;
+ }
+ /* wait while we need to and break out if more requests to send */
+ while ((rxsize > 0) && req[nextwait]) {
+ iosize = retsize = MIN(nmrsize, rxsize);
+ uio_reset(uio, uio_offset(uio), UIO_SYSSPACE, UIO_READ);
+ uio_addiov(uio, CAST_USER_ADDR_T(rxaddr), iosize);
+ FSDBG(322, uio_offset(uio), uio_resid(uio), rxaddr, rxsize);
+#if UPL_DEBUG
+ upl_ubc_alias_set(pl, (uintptr_t) current_thread(), (uintptr_t) 2);
#endif /* UPL_DEBUG */
- OSAddAtomic(1, (SInt32*)&nfsstats.pageins);
-
- error = nfs_readrpc(vp, uio, cred, p);
+ OSAddAtomic64(1, &nfsstats.pageins);
+ error = nmp->nm_funcs->nf_read_rpc_async_finish(np, req[nextwait], uio, &retsize, NULL);
+ req[nextwait] = NULL;
+ nextwait = (nextwait + 1) % MAXPAGINGREQS;
+ if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error)) {
+ lck_mtx_lock(&nmp->nm_lock);
+ if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
+ NP(np, "nfs_vnop_pagein: error %d, initiating recovery", error);
+ nfs_need_recover(nmp, error);
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+ restart++;
+ goto cancel;
+ }
+ if (error) {
+ FSDBG(322, uio_offset(uio), uio_resid(uio), error, -1);
+ break;
+ }
+ if (retsize < iosize) {
+ /* Just zero fill the rest of the valid area. */
+ int zcnt = iosize - retsize;
+ bzero((char *)rxaddr + retsize, zcnt);
+ FSDBG(324, uio_offset(uio), retsize, zcnt, rxaddr);
+ uio_update(uio, zcnt);
+ }
+ rxaddr += iosize;
+ rxsize -= iosize;
+ if (txsize)
+ break;
+ }
+ } while (!error && (txsize || rxsize));
- if (!error) {
- if (uio_uio_resid(uio)) {
- /*
- * If uio_resid > 0, there is a hole in the file
- * and no writes after the hole have been pushed
- * to the server yet... or we're at the EOF
- * Just zero fill the rest of the valid area.
- */
- // LP64todo - fix this
- int zcnt = uio_uio_resid(uio);
- int zoff = iosize - zcnt;
- bzero((char *)ioaddr + zoff, zcnt);
+ restart = 0;
- FSDBG(324, uio->uio_offset, zoff, zcnt, ioaddr);
- uio->uio_offset += zcnt;
+ if (error) {
+cancel:
+ /* cancel any outstanding requests */
+ while (req[nextwait]) {
+ nfs_request_async_cancel(req[nextwait]);
+ req[nextwait] = NULL;
+ nextwait = (nextwait + 1) % MAXPAGINGREQS;
+ }
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ } else if (restart) {
+ if (restart <= nfs_mount_state_max_restarts(nmp)) { /* guard against no progress */
+ if (error == NFSERR_GRACE)
+ tsleep(&nmp->nm_state, (PZERO-1), "nfsgrace", 2*hz);
+ if (!(error = nfs_mount_state_wait_for_recovery(nmp)))
+ goto tryagain;
+ } else {
+ NP(np, "nfs_pagein: too many restarts, aborting");
}
- ioaddr += iosize;
- xsize -= iosize;
- } else {
- FSDBG(322, uio->uio_offset, uio_uio_resid(uio), error, -1);
}
-
- nmp = VFSTONFS(vnode_mount(vp));
- } while (error == 0 && xsize > 0);
+ }
ubc_upl_unmap(pl);
if (!nofreeupl) {
- if (error)
- ubc_upl_abort_range(pl, pl_offset, size,
+ if (error)
+ ubc_upl_abort_range(pl, pl_offset, size,
UPL_ABORT_ERROR |
UPL_ABORT_FREE_ON_EMPTY);
else
/*
- * Vnode op for pageout using UPL
- * Derived from nfs_write()
+ * the following are needed only by nfs_pageout to know how to handle errors
+ * see nfs_pageout comments on explanation of actions.
+ * the errors here are copied from errno.h and errors returned by servers
+ * are expected to match the same numbers here. If not, our actions maybe
+ * erroneous.
+ */
+char nfs_pageouterrorhandler(int);
+enum actiontype {NOACTION, DUMP, DUMPANDLOG, RETRY, SEVER};
+#define NFS_ELAST 88
+static u_char errorcount[NFS_ELAST+1]; /* better be zeros when initialized */
+static const char errortooutcome[NFS_ELAST+1] = {
+ NOACTION,
+ DUMP, /* EPERM 1 Operation not permitted */
+ DUMP, /* ENOENT 2 No such file or directory */
+ DUMPANDLOG, /* ESRCH 3 No such process */
+ RETRY, /* EINTR 4 Interrupted system call */
+ DUMP, /* EIO 5 Input/output error */
+ DUMP, /* ENXIO 6 Device not configured */
+ DUMPANDLOG, /* E2BIG 7 Argument list too long */
+ DUMPANDLOG, /* ENOEXEC 8 Exec format error */
+ DUMPANDLOG, /* EBADF 9 Bad file descriptor */
+ DUMPANDLOG, /* ECHILD 10 No child processes */
+ DUMPANDLOG, /* EDEADLK 11 Resource deadlock avoided - was EAGAIN */
+ RETRY, /* ENOMEM 12 Cannot allocate memory */
+ DUMP, /* EACCES 13 Permission denied */
+ DUMPANDLOG, /* EFAULT 14 Bad address */
+ DUMPANDLOG, /* ENOTBLK 15 POSIX - Block device required */
+ RETRY, /* EBUSY 16 Device busy */
+ DUMP, /* EEXIST 17 File exists */
+ DUMP, /* EXDEV 18 Cross-device link */
+ DUMP, /* ENODEV 19 Operation not supported by device */
+ DUMP, /* ENOTDIR 20 Not a directory */
+ DUMP, /* EISDIR 21 Is a directory */
+ DUMP, /* EINVAL 22 Invalid argument */
+ DUMPANDLOG, /* ENFILE 23 Too many open files in system */
+ DUMPANDLOG, /* EMFILE 24 Too many open files */
+ DUMPANDLOG, /* ENOTTY 25 Inappropriate ioctl for device */
+ DUMPANDLOG, /* ETXTBSY 26 Text file busy - POSIX */
+ DUMP, /* EFBIG 27 File too large */
+ DUMP, /* ENOSPC 28 No space left on device */
+ DUMPANDLOG, /* ESPIPE 29 Illegal seek */
+ DUMP, /* EROFS 30 Read-only file system */
+ DUMP, /* EMLINK 31 Too many links */
+ RETRY, /* EPIPE 32 Broken pipe */
+ /* math software */
+ DUMPANDLOG, /* EDOM 33 Numerical argument out of domain */
+ DUMPANDLOG, /* ERANGE 34 Result too large */
+ RETRY, /* EAGAIN/EWOULDBLOCK 35 Resource temporarily unavailable */
+ DUMPANDLOG, /* EINPROGRESS 36 Operation now in progress */
+ DUMPANDLOG, /* EALREADY 37 Operation already in progress */
+ /* ipc/network software -- argument errors */
+ DUMPANDLOG, /* ENOTSOC 38 Socket operation on non-socket */
+ DUMPANDLOG, /* EDESTADDRREQ 39 Destination address required */
+ DUMPANDLOG, /* EMSGSIZE 40 Message too long */
+ DUMPANDLOG, /* EPROTOTYPE 41 Protocol wrong type for socket */
+ DUMPANDLOG, /* ENOPROTOOPT 42 Protocol not available */
+ DUMPANDLOG, /* EPROTONOSUPPORT 43 Protocol not supported */
+ DUMPANDLOG, /* ESOCKTNOSUPPORT 44 Socket type not supported */
+ DUMPANDLOG, /* ENOTSUP 45 Operation not supported */
+ DUMPANDLOG, /* EPFNOSUPPORT 46 Protocol family not supported */
+ DUMPANDLOG, /* EAFNOSUPPORT 47 Address family not supported by protocol family */
+ DUMPANDLOG, /* EADDRINUSE 48 Address already in use */
+ DUMPANDLOG, /* EADDRNOTAVAIL 49 Can't assign requested address */
+ /* ipc/network software -- operational errors */
+ RETRY, /* ENETDOWN 50 Network is down */
+ RETRY, /* ENETUNREACH 51 Network is unreachable */
+ RETRY, /* ENETRESET 52 Network dropped connection on reset */
+ RETRY, /* ECONNABORTED 53 Software caused connection abort */
+ RETRY, /* ECONNRESET 54 Connection reset by peer */
+ RETRY, /* ENOBUFS 55 No buffer space available */
+ RETRY, /* EISCONN 56 Socket is already connected */
+ RETRY, /* ENOTCONN 57 Socket is not connected */
+ RETRY, /* ESHUTDOWN 58 Can't send after socket shutdown */
+ RETRY, /* ETOOMANYREFS 59 Too many references: can't splice */
+ RETRY, /* ETIMEDOUT 60 Operation timed out */
+ RETRY, /* ECONNREFUSED 61 Connection refused */
+
+ DUMPANDLOG, /* ELOOP 62 Too many levels of symbolic links */
+ DUMP, /* ENAMETOOLONG 63 File name too long */
+ RETRY, /* EHOSTDOWN 64 Host is down */
+ RETRY, /* EHOSTUNREACH 65 No route to host */
+ DUMP, /* ENOTEMPTY 66 Directory not empty */
+ /* quotas & mush */
+ DUMPANDLOG, /* PROCLIM 67 Too many processes */
+ DUMPANDLOG, /* EUSERS 68 Too many users */
+ DUMPANDLOG, /* EDQUOT 69 Disc quota exceeded */
+ /* Network File System */
+ DUMP, /* ESTALE 70 Stale NFS file handle */
+ DUMP, /* EREMOTE 71 Too many levels of remote in path */
+ DUMPANDLOG, /* EBADRPC 72 RPC struct is bad */
+ DUMPANDLOG, /* ERPCMISMATCH 73 RPC version wrong */
+ DUMPANDLOG, /* EPROGUNAVAIL 74 RPC prog. not avail */
+ DUMPANDLOG, /* EPROGMISMATCH 75 Program version wrong */
+ DUMPANDLOG, /* EPROCUNAVAIL 76 Bad procedure for program */
+
+ DUMPANDLOG, /* ENOLCK 77 No locks available */
+ DUMPANDLOG, /* ENOSYS 78 Function not implemented */
+ DUMPANDLOG, /* EFTYPE 79 Inappropriate file type or format */
+ DUMPANDLOG, /* EAUTH 80 Authentication error */
+ DUMPANDLOG, /* ENEEDAUTH 81 Need authenticator */
+ /* Intelligent device errors */
+ DUMPANDLOG, /* EPWROFF 82 Device power is off */
+ DUMPANDLOG, /* EDEVERR 83 Device error, e.g. paper out */
+ DUMPANDLOG, /* EOVERFLOW 84 Value too large to be stored in data type */
+ /* Program loading errors */
+ DUMPANDLOG, /* EBADEXEC 85 Bad executable */
+ DUMPANDLOG, /* EBADARCH 86 Bad CPU type in executable */
+ DUMPANDLOG, /* ESHLIBVERS 87 Shared library version mismatch */
+ DUMPANDLOG, /* EBADMACHO 88 Malformed Macho file */
+};
+
+char
+nfs_pageouterrorhandler(int error)
+{
+ if (error > NFS_ELAST)
+ return(DUMP);
+ else
+ return(errortooutcome[error]);
+}
+
+
+/*
+ * vnode OP for pageout using UPL
+ *
+ * No buffer I/O, just RPCs straight from the mapped pages.
* File size changes are not permitted in pageout.
*/
-static int
-nfs_pageout(ap)
+int
+nfs_vnop_pageout(
struct vnop_pageout_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
size_t a_size;
int a_flags;
vfs_context_t a_context;
- } */ *ap;
+ } */ *ap)
{
vnode_t vp = ap->a_vp;
upl_t pl = ap->a_pl;
- size_t size= ap->a_size;
+ size_t size = ap->a_size;
off_t f_offset = ap->a_f_offset;
vm_offset_t pl_offset = ap->a_pl_offset;
- int flags = ap->a_flags;
- struct nfsnode *np = VTONFS(vp);
+ int flags = ap->a_flags;
+ nfsnode_t np = VTONFS(vp);
+ thread_t thd;
kauth_cred_t cred;
- proc_t p;
struct nfsbuf *bp;
- struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
+ struct nfsmount *nmp = VTONMP(vp);
daddr64_t lbn;
int error = 0, iomode;
- off_t off;
- vm_offset_t ioaddr;
- struct uio auio;
- struct iovec_32 aiov;
+ off_t off, txoffset, rxoffset;
+ vm_offset_t ioaddr, txaddr, rxaddr;
+ uio_t auio;
+ char uio_buf [ UIO_SIZEOF(1) ];
int nofreeupl = flags & UPL_NOCOMMIT;
- size_t biosize, iosize, pgsize, xsize;
+ size_t nmwsize, biosize, iosize, pgsize, txsize, rxsize, xsize, remsize;
+ struct nfsreq *req[MAXPAGINGREQS];
+ int nextsend, nextwait, wverfset, commit;
+ uint64_t wverf, wverf2;
+ uint32_t stategenid = 0, vrestart = 0, restart = 0, vrestarts = 0, restarts = 0;
+ kern_return_t kret;
FSDBG(323, f_offset, size, pl, pl_offset);
if (pl == (upl_t)NULL)
panic("nfs_pageout: no upl");
- if (UBCINVALID(vp)) {
- printf("nfs_pageout: invalid vnode 0x%x", (int)vp);
- if (!nofreeupl)
- ubc_upl_abort(pl, 0);
- return (EIO);
- }
- UBCINFOCHECK("nfs_pageout", vp);
-
if (size <= 0) {
- printf("nfs_pageout: invalid size %d", size);
+ printf("nfs_pageout: invalid size %ld", size);
if (!nofreeupl)
- ubc_upl_abort(pl, 0);
+ ubc_upl_abort_range(pl, pl_offset, size, 0);
return (EINVAL);
}
ubc_upl_abort(pl, UPL_ABORT_DUMP_PAGES|UPL_ABORT_FREE_ON_EMPTY);
return (ENXIO);
}
- biosize = vfs_statfs(vnode_mount(vp))->f_iosize;
+ biosize = nmp->nm_biosize;
+ nmwsize = nmp->nm_wsize;
+
+ nfs_data_lock_noupdate(np, NFS_DATA_LOCK_SHARED);
/*
* Check to see whether the buffer is incore.
xsize = biosize - (off % biosize);
if (off + xsize > f_offset + size)
xsize = f_offset + size - off;
- lbn = ubc_offtoblk(vp, off);
+ lbn = (daddr64_t)(off / biosize);
lck_mtx_lock(nfs_buf_mutex);
- if ((bp = nfs_buf_incore(vp, lbn))) {
+ if ((bp = nfs_buf_incore(np, lbn))) {
FSDBG(323, off, bp, bp->nb_lflags, bp->nb_flags);
if (nfs_buf_acquire(bp, NBAC_NOWAIT, 0, 0)) {
lck_mtx_unlock(nfs_buf_mutex);
+ nfs_data_unlock_noupdate(np);
/* no panic. just tell vm we are busy */
if (!nofreeupl)
- ubc_upl_abort(pl, 0);
+ ubc_upl_abort_range(pl, pl_offset, size, 0);
return (EBUSY);
}
if (bp->nb_dirtyend > 0) {
end -= boff;
if ((bp->nb_dirtyoff < start) &&
(bp->nb_dirtyend > end)) {
- /* not gonna be able to clip the dirty region */
- FSDBG(323, vp, bp, 0xd00deebc, EBUSY);
+ /*
+ * not gonna be able to clip the dirty region
+ *
+ * But before returning the bad news, move the
+ * buffer to the start of the delwri list and
+ * give the list a push to try to flush the
+ * buffer out.
+ */
+ FSDBG(323, np, bp, 0xd00deebc, EBUSY);
+ nfs_buf_remfree(bp);
+ TAILQ_INSERT_HEAD(&nfsbufdelwri, bp, nb_free);
+ nfsbufdelwricnt++;
nfs_buf_drop(bp);
+ nfs_buf_delwri_push(1);
lck_mtx_unlock(nfs_buf_mutex);
+ nfs_data_unlock_noupdate(np);
if (!nofreeupl)
- ubc_upl_abort(pl, 0);
+ ubc_upl_abort_range(pl, pl_offset, size, 0);
return (EBUSY);
}
if ((bp->nb_dirtyoff < start) ||
nfs_buf_remfree(bp);
lck_mtx_unlock(nfs_buf_mutex);
SET(bp->nb_flags, NB_INVAL);
+ nfs_node_lock_force(np);
if (ISSET(bp->nb_flags, NB_NEEDCOMMIT)) {
CLR(bp->nb_flags, NB_NEEDCOMMIT);
np->n_needcommitcnt--;
CHECK_NEEDCOMMITCNT(np);
}
+ nfs_node_unlock(np);
nfs_buf_release(bp, 1);
} else {
lck_mtx_unlock(nfs_buf_mutex);
}
}
+ thd = vfs_context_thread(ap->a_context);
cred = ubc_getcred(vp);
- if (cred == NOCRED)
+ if (!IS_VALID_CRED(cred))
cred = vfs_context_ucred(ap->a_context);
- p = vfs_context_proc(ap->a_context);
+ nfs_node_lock_force(np);
if (np->n_flag & NWRITEERR) {
- np->n_flag &= ~NWRITEERR;
+ error = np->n_error;
+ nfs_node_unlock(np);
+ nfs_data_unlock_noupdate(np);
if (!nofreeupl)
ubc_upl_abort_range(pl, pl_offset, size,
UPL_ABORT_FREE_ON_EMPTY);
- return (np->n_error);
+ return (error);
}
- if ((nmp->nm_flag & NFSMNT_NFSV3) && !(nmp->nm_state & NFSSTA_GOTFSINFO))
- nfs_fsinfo(nmp, vp, cred, p);
+ nfs_node_unlock(np);
if (f_offset < 0 || f_offset >= (off_t)np->n_size ||
f_offset & PAGE_MASK_64 || size & PAGE_MASK_64) {
+ nfs_data_unlock_noupdate(np);
if (!nofreeupl)
ubc_upl_abort_range(pl, pl_offset, size,
UPL_ABORT_FREE_ON_EMPTY);
return (EINVAL);
}
- ubc_upl_map(pl, &ioaddr);
+ kret = ubc_upl_map(pl, &ioaddr);
+ if (kret != KERN_SUCCESS)
+ panic("nfs_vnop_pageout: ubc_upl_map() failed with (%d)", kret);
ioaddr += pl_offset;
if ((u_quad_t)f_offset + size > np->n_size)
xsize = size;
pgsize = round_page_64(xsize);
- if (size > pgsize) {
- if (!nofreeupl)
- ubc_upl_abort_range(pl, pl_offset + pgsize,
- size - pgsize,
- UPL_ABORT_FREE_ON_EMPTY);
- }
+ if ((size > pgsize) && !nofreeupl)
+ ubc_upl_abort_range(pl, pl_offset + pgsize, size - pgsize,
+ UPL_ABORT_FREE_ON_EMPTY);
- /*
+ /*
* check for partial page and clear the
* contents past end of the file before
* releasing it in the VM page cache
bzero((caddr_t)(ioaddr + io), size - io);
FSDBG(321, np->n_size, f_offset, f_offset + io, size - io);
}
+ nfs_data_unlock_noupdate(np);
- auio.uio_offset = f_offset;
-#if 1 /* LP64todo - can't use new segment flags until the drivers are ready */
- auio.uio_segflg = UIO_SYSSPACE;
-#else
- auio.uio_segflg = UIO_SYSSPACE32;
-#endif
- auio.uio_rw = UIO_READ;
- auio.uio_procp = p;
+ auio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_WRITE,
+ &uio_buf, sizeof(uio_buf));
+tryagain:
+ if (nmp->nm_vers >= NFS_VER4)
+ stategenid = nmp->nm_stategenid;
+ wverf = wverf2 = wverfset = 0;
+ txsize = rxsize = xsize;
+ txoffset = rxoffset = f_offset;
+ txaddr = rxaddr = ioaddr;
+ commit = NFS_WRITE_FILESYNC;
+
+ bzero(req, sizeof(req));
+ nextsend = nextwait = 0;
do {
- /*
- * It would be nice to be able to issue all these requests
- * in parallel instead of waiting for each one to complete
- * before sending the next one.
- * XXX Should we align these requests to block boundaries?
- */
- iosize = min(biosize, xsize);
- uio_uio_resid_set(&auio, iosize);
- aiov.iov_len = iosize;
- aiov.iov_base = (uintptr_t)ioaddr;
- auio.uio_iovs.iov32p = &aiov;
- auio.uio_iovcnt = 1;
-
- FSDBG(323, auio.uio_offset, uio_uio_resid(&auio), ioaddr, xsize);
- OSAddAtomic(1, (SInt32*)&nfsstats.pageouts);
-
- vnode_startwrite(vp);
-
- /* NMODIFIED would be set here if doing unstable writes */
- iomode = NFSV3WRITE_FILESYNC;
- error = nfs_writerpc(vp, &auio, cred, p, &iomode, NULL);
- vnode_writedone(vp);
- if (error)
- goto cleanup;
- /* Note: no need to check uio_resid, because */
- /* it'll only be set if there was an error. */
- ioaddr += iosize;
- xsize -= iosize;
- } while (xsize > 0);
-
-cleanup:
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ break;
+ }
+ /* send requests while we need to and have available slots */
+ while ((txsize > 0) && (req[nextsend] == NULL)) {
+ iosize = MIN(nmwsize, txsize);
+ uio_reset(auio, txoffset, UIO_SYSSPACE, UIO_WRITE);
+ uio_addiov(auio, CAST_USER_ADDR_T(txaddr), iosize);
+ FSDBG(323, uio_offset(auio), iosize, txaddr, txsize);
+ OSAddAtomic64(1, &nfsstats.pageouts);
+ nfs_node_lock_force(np);
+ np->n_numoutput++;
+ nfs_node_unlock(np);
+ vnode_startwrite(vp);
+ iomode = NFS_WRITE_UNSTABLE;
+ if ((error = nmp->nm_funcs->nf_write_rpc_async(np, auio, iosize, thd, cred, iomode, NULL, &req[nextsend]))) {
+ req[nextsend] = NULL;
+ vnode_writedone(vp);
+ nfs_node_lock_force(np);
+ np->n_numoutput--;
+ nfs_node_unlock(np);
+ break;
+ }
+ txaddr += iosize;
+ txoffset += iosize;
+ txsize -= iosize;
+ nextsend = (nextsend + 1) % MAXPAGINGREQS;
+ }
+ /* wait while we need to and break out if more requests to send */
+ while ((rxsize > 0) && req[nextwait]) {
+ iosize = remsize = MIN(nmwsize, rxsize);
+ error = nmp->nm_funcs->nf_write_rpc_async_finish(np, req[nextwait], &iomode, &iosize, &wverf2);
+ req[nextwait] = NULL;
+ nextwait = (nextwait + 1) % MAXPAGINGREQS;
+ vnode_writedone(vp);
+ nfs_node_lock_force(np);
+ np->n_numoutput--;
+ nfs_node_unlock(np);
+ if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error)) {
+ lck_mtx_lock(&nmp->nm_lock);
+ if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
+ NP(np, "nfs_vnop_pageout: error %d, initiating recovery", error);
+ nfs_need_recover(nmp, error);
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+ restart = 1;
+ goto cancel;
+ }
+ if (error) {
+ FSDBG(323, rxoffset, rxsize, error, -1);
+ break;
+ }
+ if (!wverfset) {
+ wverf = wverf2;
+ wverfset = 1;
+ } else if (wverf != wverf2) {
+ /* verifier changed, so we need to restart all the writes */
+ vrestart = 1;
+ goto cancel;
+ }
+ /* Retain the lowest commitment level returned. */
+ if (iomode < commit)
+ commit = iomode;
+ rxaddr += iosize;
+ rxoffset += iosize;
+ rxsize -= iosize;
+ remsize -= iosize;
+ if (remsize > 0) {
+ /* need to try sending the remainder */
+ iosize = remsize;
+ uio_reset(auio, rxoffset, UIO_SYSSPACE, UIO_WRITE);
+ uio_addiov(auio, CAST_USER_ADDR_T(rxaddr), remsize);
+ iomode = NFS_WRITE_UNSTABLE;
+ error = nfs_write_rpc2(np, auio, thd, cred, &iomode, &wverf2);
+ if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error)) {
+ NP(np, "nfs_vnop_pageout: restart: error %d", error);
+ lck_mtx_lock(&nmp->nm_lock);
+ if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
+ NP(np, "nfs_vnop_pageout: error %d, initiating recovery", error);
+ nfs_need_recover(nmp, error);
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+ restart = 1;
+ goto cancel;
+ }
+ if (error) {
+ FSDBG(323, rxoffset, rxsize, error, -1);
+ break;
+ }
+ if (wverf != wverf2) {
+ /* verifier changed, so we need to restart all the writes */
+ vrestart = 1;
+ goto cancel;
+ }
+ if (iomode < commit)
+ commit = iomode;
+ rxaddr += iosize;
+ rxoffset += iosize;
+ rxsize -= iosize;
+ }
+ if (txsize)
+ break;
+ }
+ } while (!error && (txsize || rxsize));
+
+ vrestart = 0;
+
+ if (!error && (commit != NFS_WRITE_FILESYNC)) {
+ error = nmp->nm_funcs->nf_commit_rpc(np, f_offset, xsize, cred, wverf);
+ if (error == NFSERR_STALEWRITEVERF) {
+ vrestart = 1;
+ error = EIO;
+ }
+ }
+
+ if (error) {
+cancel:
+ /* cancel any outstanding requests */
+ while (req[nextwait]) {
+ nfs_request_async_cancel(req[nextwait]);
+ req[nextwait] = NULL;
+ nextwait = (nextwait + 1) % MAXPAGINGREQS;
+ vnode_writedone(vp);
+ nfs_node_lock_force(np);
+ np->n_numoutput--;
+ nfs_node_unlock(np);
+ }
+ if (np->n_flag & NREVOKE) {
+ error = EIO;
+ } else {
+ if (vrestart) {
+ if (++vrestarts <= 100) /* guard against no progress */
+ goto tryagain;
+ NP(np, "nfs_pageout: too many restarts, aborting");
+ FSDBG(323, f_offset, xsize, ERESTART, -1);
+ }
+ if (restart) {
+ if (restarts <= nfs_mount_state_max_restarts(nmp)) { /* guard against no progress */
+ if (error == NFSERR_GRACE)
+ tsleep(&nmp->nm_state, (PZERO-1), "nfsgrace", 2*hz);
+ if (!(error = nfs_mount_state_wait_for_recovery(nmp)))
+ goto tryagain;
+ } else {
+ NP(np, "nfs_pageout: too many restarts, aborting");
+ FSDBG(323, f_offset, xsize, ERESTART, -1);
+ }
+ }
+ }
+ }
+
ubc_upl_unmap(pl);
+
/*
* We've had several different solutions on what to do when the pageout
- * gets an error. If we don't handle it, and return an error to the
- * caller, vm, it will retry . This can end in endless looping
+ * gets an error. If we don't handle it, and return an error to the
+ * caller, vm, it will retry . This can end in endless looping
* between vm and here doing retries of the same page. Doing a dump
* back to vm, will get it out of vm's knowledge and we lose whatever
* data existed. This is risky, but in some cases necessary. For
* example, the initial fix here was to do that for ESTALE. In that case
- * the server is telling us that the file is no longer the same. We
- * would not want to keep paging out to that. We also saw some 151
+ * the server is telling us that the file is no longer the same. We
+ * would not want to keep paging out to that. We also saw some 151
* errors from Auspex server and NFSv3 can return errors higher than
* ELAST. Those along with NFS known server errors we will "dump" from
* vm. Errors we don't expect to occur, we dump and log for further
* have some kind of persistant store when the vm cannot dump nor keep
* retrying as a solution, but this would be a file architectural change
*/
-
if (!nofreeupl) { /* otherwise stacked file system has to handle this */
if (error) {
- int abortflags = 0;
- short action = nfs_pageouterrorhandler(error);
+ int abortflags = 0;
+ char action = nfs_pageouterrorhandler(error);
switch (action) {
case DUMP:
break;
case DUMPANDLOG:
abortflags = UPL_ABORT_DUMP_PAGES|UPL_ABORT_FREE_ON_EMPTY;
- if (error <= ELAST &&
- (errorcount[error] % 100 == 0))
- printf("nfs_pageout: unexpected error %d. dumping vm page\n", error);
- errorcount[error]++;
+ if (error <= NFS_ELAST) {
+ if ((errorcount[error] % 100) == 0)
+ NP(np, "nfs_pageout: unexpected error %d. dumping vm page", error);
+ errorcount[error]++;
+ }
break;
case RETRY:
abortflags = UPL_ABORT_FREE_ON_EMPTY;
break;
- case RETRYWITHSLEEP:
- abortflags = UPL_ABORT_FREE_ON_EMPTY;
- /* pri unused. PSOCK for placeholder. */
- tsleep(&lbolt, PSOCK, "nfspageout", 0);
- break;
case SEVER: /* not implemented */
default:
- printf("nfs_pageout: action %d not expected\n", action);
+ NP(np, "nfs_pageout: action %d not expected", action);
break;
}
-
- ubc_upl_abort_range(pl, pl_offset, size, abortflags);
+
+ ubc_upl_abort_range(pl, pl_offset, pgsize, abortflags);
/* return error in all cases above */
-
- } else
+
+ } else {
ubc_upl_commit_range(pl, pl_offset, pgsize,
UPL_COMMIT_CLEAR_DIRTY |
UPL_COMMIT_FREE_ON_EMPTY);
+ }
}
return (error);
}
/* Blktooff derives file offset given a logical block number */
-static int
-nfs_blktooff(ap)
+int
+nfs_vnop_blktooff(
struct vnop_blktooff_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
daddr64_t a_lblkno;
off_t *a_offset;
- } */ *ap;
+ } */ *ap)
{
int biosize;
vnode_t vp = ap->a_vp;
- mount_t mp = vnode_mount(vp);
+ struct nfsmount *nmp = VTONMP(vp);
- if (!mp)
+ if (!nmp)
return (ENXIO);
-
- biosize = vfs_statfs(mp)->f_iosize;
+ biosize = nmp->nm_biosize;
*ap->a_offset = (off_t)(ap->a_lblkno * biosize);
return (0);
}
-static int
-nfs_offtoblk(ap)
+int
+nfs_vnop_offtoblk(
struct vnop_offtoblk_args /* {
struct vnodeop_desc *a_desc;
vnode_t a_vp;
off_t a_offset;
daddr64_t *a_lblkno;
- } */ *ap;
+ } */ *ap)
{
int biosize;
vnode_t vp = ap->a_vp;
- mount_t mp = vnode_mount(vp);
+ struct nfsmount *nmp = VTONMP(vp);
- if (!mp)
+ if (!nmp)
return (ENXIO);
-
- biosize = vfs_statfs(mp)->f_iosize;
+ biosize = nmp->nm_biosize;
*ap->a_lblkno = (daddr64_t)(ap->a_offset / biosize);
return (0);
}
+/*
+ * vnode change monitoring
+ */
+int
+nfs_vnop_monitor(
+ struct vnop_monitor_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ uint32_t a_events;
+ uint32_t a_flags;
+ void *a_handle;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ nfsnode_t np = VTONFS(ap->a_vp);
+ struct nfsmount *nmp = VTONMP(ap->a_vp);
+ int error = 0;
+
+ if (!nmp)
+ return (ENXIO);
+
+ /* make sure that the vnode's monitoring status is up to date */
+ lck_mtx_lock(&nmp->nm_lock);
+ if (vnode_ismonitored(ap->a_vp)) {
+ /* This vnode is currently being monitored, make sure we're tracking it. */
+ if (np->n_monlink.le_next == NFSNOLIST) {
+ LIST_INSERT_HEAD(&nmp->nm_monlist, np, n_monlink);
+ nfs_mount_sock_thread_wake(nmp);
+ }
+ } else {
+ /* This vnode is no longer being monitored, make sure we're not tracking it. */
+ /* Wait for any in-progress getattr to complete first. */
+ while (np->n_mflag & NMMONSCANINPROG) {
+ struct timespec ts = { 1, 0 };
+ np->n_mflag |= NMMONSCANWANT;
+ msleep(&np->n_mflag, &nmp->nm_lock, PZERO-1, "nfswaitmonscan", &ts);
+ }
+ if (np->n_monlink.le_next != NFSNOLIST) {
+ LIST_REMOVE(np, n_monlink);
+ np->n_monlink.le_next = NFSNOLIST;
+ }
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+
+ return (error);
+}
+
+/*
+ * Send a vnode notification for the given events.
+ */
+void
+nfs_vnode_notify(nfsnode_t np, uint32_t events)
+{
+ struct nfsmount *nmp = NFSTONMP(np);
+ struct nfs_vattr nvattr;
+ struct vnode_attr vattr, *vap = NULL;
+ struct timeval now;
+
+ microuptime(&now);
+ if ((np->n_evtstamp == now.tv_sec) || !nmp) {
+ /* delay sending this notify */
+ np->n_events |= events;
+ return;
+ }
+ events |= np->n_events;
+ np->n_events = 0;
+ np->n_evtstamp = now.tv_sec;
+
+ vfs_get_notify_attributes(&vattr);
+ if (!nfs_getattrcache(np, &nvattr, 0)) {
+ vap = &vattr;
+ VATTR_INIT(vap);
+ VATTR_RETURN(vap, va_fsid, vfs_statfs(nmp->nm_mountp)->f_fsid.val[0]);
+ VATTR_RETURN(vap, va_fileid, nvattr.nva_fileid);
+ VATTR_RETURN(vap, va_mode, nvattr.nva_mode);
+ VATTR_RETURN(vap, va_uid, nvattr.nva_uid);
+ VATTR_RETURN(vap, va_gid, nvattr.nva_gid);
+ VATTR_RETURN(vap, va_nlink, nvattr.nva_nlink);
+ }
+ vnode_notify(NFSTOV(np), events, vap);
+}
projects / apple / xnu.git / blobdiff