X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..ecc0ceb4089d506a0b8d16686a95817b331af9cb:/bsd/nfs/nfs_vnops.c diff --git a/bsd/nfs/nfs_vnops.c b/bsd/nfs/nfs_vnops.c index 47c461076..ae1906aed 100644 --- a/bsd/nfs/nfs_vnops.c +++ b/bsd/nfs/nfs_vnops.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2015 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -112,48 +112,50 @@ #include #include #include + #include +#include #include #include -#include + +#define NFS_VNOP_DBG(...) NFS_DBG(NFS_FAC_VNOP, 7, ## __VA_ARGS__) +#define DEFAULT_READLINK_NOCACHE 0 /* * NFS vnode ops */ -static int nfs_vnop_lookup(struct vnop_lookup_args *); -static int nfsspec_vnop_read(struct vnop_read_args *); -static int nfsspec_vnop_write(struct vnop_write_args *); -static int nfsspec_vnop_close(struct vnop_close_args *); +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 -static int nfsfifo_vnop_read(struct vnop_read_args *); -static int nfsfifo_vnop_write(struct vnop_write_args *); -static int nfsfifo_vnop_close(struct vnop_close_args *); +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 -static int nfs_vnop_ioctl(struct vnop_ioctl_args *); -static int nfs_vnop_select(struct vnop_select_args *); -static int nfs_vnop_setattr(struct vnop_setattr_args *); -static int nfs_vnop_read(struct vnop_read_args *); -static int nfs_vnop_mmap(struct vnop_mmap_args *); -static int nfs_vnop_fsync(struct vnop_fsync_args *); -static int nfs_vnop_remove(struct vnop_remove_args *); -static int nfs_vnop_rename(struct vnop_rename_args *); -static int nfs_vnop_readdir(struct vnop_readdir_args *); -static int nfs_vnop_readlink(struct vnop_readlink_args *); -static int nfs_vnop_pathconf(struct vnop_pathconf_args *); -static int nfs_vnop_pagein(struct vnop_pagein_args *); -static int nfs_vnop_pageout(struct vnop_pageout_args *); -static int nfs_vnop_blktooff(struct vnop_blktooff_args *); -static int nfs_vnop_offtoblk(struct vnop_offtoblk_args *); -static int nfs_vnop_blockmap(struct vnop_blockmap_args *); - -static int nfs3_vnop_create(struct vnop_create_args *); -static int nfs3_vnop_mknod(struct vnop_mknod_args *); -static int nfs3_vnop_getattr(struct vnop_getattr_args *); -static int nfs3_vnop_link(struct vnop_link_args *); -static int nfs3_vnop_mkdir(struct vnop_mkdir_args *); -static int nfs3_vnop_rmdir(struct vnop_rmdir_args *); -static int nfs3_vnop_symlink(struct vnop_symlink_args *); +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[] = { @@ -161,8 +163,8 @@ static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { { &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 *)nfs3_vnop_open }, /* open */ - { &vnop_close_desc, (vnop_t *)nfs3_vnop_close }, /* close */ + { &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 */ @@ -172,6 +174,7 @@ static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { { &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 */ @@ -185,7 +188,7 @@ static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { { &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 *)nfs3_vnop_advlock }, /* advlock */ + { &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 */ @@ -193,6 +196,7 @@ static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = { { &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 = @@ -204,8 +208,8 @@ static struct vnodeopv_entry_desc nfsv4_vnodeop_entries[] = { { &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 *)nfs4_vnop_open }, /* open */ - { &vnop_close_desc, (vnop_t *)nfs4_vnop_close }, /* close */ + { &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 */ @@ -215,6 +219,7 @@ static struct vnodeopv_entry_desc nfsv4_vnodeop_entries[] = { { &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 */ @@ -228,7 +233,7 @@ static struct vnodeopv_entry_desc nfsv4_vnodeop_entries[] = { { &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 *)nfs4_vnop_advlock }, /* advlock */ + { &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 */ @@ -236,6 +241,16 @@ static struct vnodeopv_entry_desc nfsv4_vnodeop_entries[] = { { &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 = @@ -280,6 +295,7 @@ static struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = { { &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 = @@ -320,6 +336,16 @@ static struct vnodeopv_entry_desc spec_nfsv4nodeop_entries[] = { { &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 = @@ -362,6 +388,7 @@ static struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = { { &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 = @@ -403,42 +430,53 @@ static struct vnodeopv_entry_desc fifo_nfsv4nodeop_entries[] = { { &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_sillyrename(nfsnode_t,nfsnode_t,struct componentname *,vfs_context_t); +int nfs_sillyrename(nfsnode_t,nfsnode_t,struct componentname *,vfs_context_t); +int nfs_getattr_internal(nfsnode_t, struct nfs_vattr *, vfs_context_t, int); +int nfs_refresh_fh(nfsnode_t, vfs_context_t); /* * 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_mode array. + * The index of the next slot to use is kept in the last entry of the n_access array. */ int -nfs_node_mode_slot(nfsnode_t np, uid_t uid, int add) +nfs_node_access_slot(nfsnode_t np, uid_t uid, int add) { int slot; for (slot=0; slot < NFS_ACCESS_CACHE_SIZE; slot++) - if (np->n_modeuid[slot] == uid) + if (np->n_accessuid[slot] == uid) break; if (slot == NFS_ACCESS_CACHE_SIZE) { if (!add) return (-1); - slot = np->n_mode[NFS_ACCESS_CACHE_SIZE]; - np->n_mode[NFS_ACCESS_CACHE_SIZE] = (slot + 1) % NFS_ACCESS_CACHE_SIZE; + slot = np->n_access[NFS_ACCESS_CACHE_SIZE]; + np->n_access[NFS_ACCESS_CACHE_SIZE] = (slot + 1) % NFS_ACCESS_CACHE_SIZE; } return (slot); } int -nfs3_access_rpc(nfsnode_t np, u_long *mode, vfs_context_t ctx) +nfs3_access_rpc(nfsnode_t np, u_int32_t *access, int rpcflags, vfs_context_t ctx) { - int error = 0, status, slot; - uint32_t access = 0; + int error = 0, lockerror = ENOENT, status, slot; + uint32_t access_result = 0; u_int64_t xid; struct nfsm_chain nmreq, nmrep; struct timeval now; @@ -449,23 +487,26 @@ nfs3_access_rpc(nfsnode_t np, u_long *mode, vfs_context_t ctx) 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, *mode); + 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, - &nmrep, &xid, &status); + error = nfs_request2(np, NULL, &nmreq, NFSPROC_ACCESS, + vfs_context_thread(ctx), vfs_context_ucred(ctx), + NULL, rpcflags, &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); + nfsm_chain_get_32(error, &nmrep, access_result); nfsmout_if(error); uid = kauth_cred_getuid(vfs_context_ucred(ctx)); - slot = nfs_node_mode_slot(np, uid, 1); - np->n_modeuid[slot] = uid; + slot = nfs_node_access_slot(np, uid, 1); + np->n_accessuid[slot] = uid; microuptime(&now); - np->n_modestamp[slot] = now.tv_sec; - np->n_mode[slot] = access; + 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 @@ -474,12 +515,17 @@ nfs3_access_rpc(nfsnode_t np, u_long *mode, vfs_context_t ctx) * and just let any subsequent delete action fail if it * really isn't deletable. */ - if ((*mode & NFS_ACCESS_DELETE) && - !(np->n_mode[slot] & NFS_ACCESS_DELETE)) - np->n_mode[slot] |= NFS_ACCESS_DELETE; - /* pass back the mode returned with this request */ - *mode = np->n_mode[slot]; + 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); @@ -488,8 +534,8 @@ nfsmout: /* * 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. + * For NFS version 3+, use the access RPC to check accessibility. If file + * permissions are changed on the server, accesses might still fail later. */ int nfs_vnop_access( @@ -502,8 +548,8 @@ nfs_vnop_access( { vfs_context_t ctx = ap->a_context; vnode_t vp = ap->a_vp; - int error = 0, slot, dorpc; - u_long mode, wmode; + int error = 0, slot, dorpc, rpcflags = 0; + u_int32_t access, waccess; nfsnode_t np = VTONFS(vp); struct nfsmount *nmp; int nfsvers; @@ -511,7 +557,7 @@ nfs_vnop_access( uid_t uid; nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; @@ -534,91 +580,120 @@ nfs_vnop_access( /* * Convert KAUTH primitives to NFS access rights. */ - mode = 0; + access = 0; if (vnode_isdir(vp)) { /* directory */ if (ap->a_action & (KAUTH_VNODE_LIST_DIRECTORY | KAUTH_VNODE_READ_EXTATTRIBUTES)) - mode |= NFS_ACCESS_READ; + access |= NFS_ACCESS_READ; if (ap->a_action & KAUTH_VNODE_SEARCH) - mode |= NFS_ACCESS_LOOKUP; + access |= NFS_ACCESS_LOOKUP; if (ap->a_action & (KAUTH_VNODE_ADD_FILE | KAUTH_VNODE_ADD_SUBDIRECTORY)) - mode |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; + access |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; if (ap->a_action & KAUTH_VNODE_DELETE_CHILD) - mode |= NFS_ACCESS_MODIFY; + access |= NFS_ACCESS_MODIFY; } else { /* file */ if (ap->a_action & (KAUTH_VNODE_READ_DATA | KAUTH_VNODE_READ_EXTATTRIBUTES)) - mode |= NFS_ACCESS_READ; + access |= NFS_ACCESS_READ; if (ap->a_action & KAUTH_VNODE_WRITE_DATA) - mode |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; + access |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; if (ap->a_action & KAUTH_VNODE_APPEND_DATA) - mode |= NFS_ACCESS_EXTEND; + access |= NFS_ACCESS_EXTEND; if (ap->a_action & KAUTH_VNODE_EXECUTE) - mode |= NFS_ACCESS_EXECUTE; + access |= NFS_ACCESS_EXECUTE; } /* common */ if (ap->a_action & KAUTH_VNODE_DELETE) - mode |= NFS_ACCESS_DELETE; + access |= NFS_ACCESS_DELETE; if (ap->a_action & (KAUTH_VNODE_WRITE_ATTRIBUTES | KAUTH_VNODE_WRITE_EXTATTRIBUTES | KAUTH_VNODE_WRITE_SECURITY)) - mode |= NFS_ACCESS_MODIFY; + access |= NFS_ACCESS_MODIFY; /* XXX this is pretty dubious */ if (ap->a_action & KAUTH_VNODE_CHANGE_OWNER) - mode |= NFS_ACCESS_MODIFY; + access |= NFS_ACCESS_MODIFY; /* if caching, always ask for every right */ if (nfs_access_cache_timeout > 0) { - wmode = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | + waccess = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND | NFS_ACCESS_EXECUTE | NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; } else { - wmode = mode; + waccess = access; } - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_mode_slot(np, uid, 0); + if (auth_is_kerberized(np->n_auth) || auth_is_kerberized(nmp->nm_auth)) + uid = nfs_cred_getasid2uid(vfs_context_ucred(ctx)); + else + uid = kauth_cred_getuid(vfs_context_ucred(ctx)); + slot = nfs_node_access_slot(np, uid, 0); dorpc = 1; - if (NMODEVALID(np, slot)) { + 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_modestamp[slot] + nfs_access_cache_timeout)) && - ((np->n_mode[slot] & mode) == mode)) { - /* OSAddAtomic(1, (SInt32*)&nfsstats.accesscache_hits); */ + if (((now.tv_sec < (np->n_accessstamp[slot] + nfs_access_cache_timeout)) && + ((np->n_access[slot] & access) == access)) || nfs_use_cache(nmp)) { + /* OSAddAtomic(1, &nfsstats.accesscache_hits); */ dorpc = 0; - wmode = np->n_mode[slot]; + 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, (SInt32*)&nfsstats.accesscache_misses); */ - error = nmp->nm_funcs->nf_access_rpc(np, &wmode, ctx); + /* OSAddAtomic(1, &nfsstats.accesscache_misses); */ + + /* + * Allow an access call to timeout if we have it cached + * so we won't hang if the server isn't responding. + */ + if (NACCESSVALID(np, slot)) + rpcflags |= R_SOFT; + + error = nmp->nm_funcs->nf_access_rpc(np, &waccess, rpcflags, ctx); + + /* + * If the server didn't respond return the cached access. + */ + if ((error == ETIMEDOUT) && (rpcflags & R_SOFT)) { + error = 0; + waccess = np->n_access[slot]; + } } - if (!error && ((wmode & mode) != mode)) + if (!error && ((waccess & access) != access)) error = EACCES; - nfs_unlock(np); return (error); } + /* * 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. */ int -nfs3_vnop_open( +nfs_vnop_open( struct vnop_open_args /* { struct vnodeop_desc *a_desc; vnode_t a_vp; @@ -630,85 +705,213 @@ nfs3_vnop_open( vnode_t vp = ap->a_vp; nfsnode_t np = VTONFS(vp); struct nfsmount *nmp; - struct nfs_vattr nvattr; + int error, accessMode, denyMode, opened = 0; + struct nfs_open_owner *noop = NULL; + struct nfs_open_file *nofp = NULL; enum vtype vtype; - int error, nfsvers; + + if (!(ap->a_mode & (FREAD|FWRITE))) + return (EINVAL); nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); - nfsvers = nmp->nm_vers; + if (np->n_flag & NREVOKE) + return (EIO); vtype = vnode_vtype(vp); if ((vtype != VREG) && (vtype != VDIR) && (vtype != VLNK)) return (EACCES); + + /* First, check if we need to update/invalidate */ if (ISSET(np->n_flag, NUPDATESIZE)) nfs_data_update_size(np, 0); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + if ((error = nfs_node_lock(np))) return (error); if (np->n_flag & NNEEDINVALIDATE) { np->n_flag &= ~NNEEDINVALIDATE; - nfs_unlock(np); + if (vtype == VDIR) + nfs_invaldir(np); + nfs_node_unlock(np); nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + if ((error = nfs_node_lock(np))) return (error); } + if (vtype == VREG) + np->n_lastrahead = -1; if (np->n_flag & NMODIFIED) { - nfs_unlock(np); - if ((error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1)) == EINTR) - return (error); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) - return (error); if (vtype == VDIR) - np->n_direofoffset = 0; - NATTRINVALIDATE(np); /* For Open/Close consistency */ - error = nfs_getattr(np, &nvattr, ctx, 1); - if (error) { - nfs_unlock(np); + 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 (NFS_CHANGED_NC(nfsvers, np, &nvattr)) { - np->n_flag &= ~NNEGNCENTRIES; - cache_purge(vp); - } - NFS_CHANGED_UPDATE_NC(nfsvers, np, &nvattr); - } - NFS_CHANGED_UPDATE(nfsvers, np, &nvattr); } else { - NATTRINVALIDATE(np); /* For Open/Close consistency */ - error = nfs_getattr(np, &nvattr, ctx, 1); - if (error) { - nfs_unlock(np); - return (error); + 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 (NFS_CHANGED(nfsvers, np, &nvattr)) { - if (vtype == VDIR) { - np->n_direofoffset = 0; - nfs_invaldir(np); - /* purge name cache entries */ - if (NFS_CHANGED_NC(nfsvers, np, &nvattr)) { - np->n_flag &= ~NNEGNCENTRIES; - cache_purge(vp); - } + 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; } - nfs_unlock(np); - if ((error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1)) == EINTR) - return (error); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) - return (error); - if (vtype == VDIR) - NFS_CHANGED_UPDATE_NC(nfsvers, np, &nvattr); - NFS_CHANGED_UPDATE(nfsvers, np, &nvattr); } } - nfs_unlock(np); - 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 + * * 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: @@ -726,11 +929,11 @@ nfs3_vnop_open( * * 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 */ int -nfs3_vnop_close( +nfs_vnop_close( struct vnop_close_args /* { struct vnodeop_desc *a_desc; vnode_t a_vp; @@ -742,43 +945,328 @@ nfs3_vnop_close( vnode_t vp = ap->a_vp; nfsnode_t np = VTONFS(vp); struct nfsmount *nmp; - int nfsvers; - 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; - if (vnode_vtype(vp) != VREG) - return (0); nmp = VTONMP(vp); if (!nmp) return (ENXIO); nfsvers = nmp->nm_vers; + vtype = vnode_vtype(vp); + /* First, check if we need to update/flush/invalidate */ if (ISSET(np->n_flag, NUPDATESIZE)) nfs_data_update_size(np, 0); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) - return (error); + nfs_node_lock_force(np); if (np->n_flag & NNEEDINVALIDATE) { np->n_flag &= ~NNEEDINVALIDATE; - nfs_unlock(np); + nfs_node_unlock(np); nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) - return (error); + nfs_node_lock_force(np); } - if (np->n_flag & NMODIFIED) { - nfs_unlock(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); - if (error) - return (error); - nfs_lock(np, NFS_NODE_LOCK_FORCE); + nfs_node_lock_force(np); NATTRINVALIDATE(np); } if (np->n_flag & NWRITEERR) { np->n_flag &= ~NWRITEERR; error = np->n_error; } - nfs_unlock(np); + 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 || vfs_isforce(mp)); + + 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_close( + nfsnode_t np, + struct nfs_open_file *nofp, + uint32_t accessMode, + uint32_t denyMode, + vfs_context_t ctx) +{ + 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)); + + /* + * 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); + } + error = nfs4_close_rpc(np, nofp, vfs_context_thread(ctx), vfs_context_ucred(ctx), 0); + } + } 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); + } + } + + 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); } @@ -789,18 +1277,25 @@ nfs3_getattr_rpc( mount_t mp, u_char *fhp, size_t fhsize, + int flags, vfs_context_t ctx, struct nfs_vattr *nvap, u_int64_t *xidp) { struct nfsmount *nmp = mp ? VFSTONFS(mp) : NFSTONMP(np); - int error = 0, status, nfsvers; + int error = 0, status, nfsvers, rpcflags = 0; struct nfsm_chain nmreq, nmrep; - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; + if (flags & NGA_MONITOR) /* vnode monitor requests should be soft */ + rpcflags = R_RECOVER; + + if (flags & NGA_SOFT) /* Return ETIMEDOUT if server not responding */ + rpcflags |= R_SOFT; + nfsm_chain_null(&nmreq); nfsm_chain_null(&nmrep); @@ -810,8 +1305,9 @@ nfs3_getattr_rpc( nfsm_chain_add_opaque(error, &nmreq, fhp, fhsize); nfsm_chain_build_done(error, &nmreq); nfsmout_if(error); - error = nfs_request(np, mp, &nmreq, NFSPROC_GETATTR, ctx, - &nmrep, xidp, &status); + 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); @@ -822,113 +1318,366 @@ nfsmout: return (error); } +/* + * nfs_refresh_fh will attempt to update the file handle for the node. + * + * It only does this for symbolic links and regular files that are not currently opened. + * + * On Success returns 0 and the nodes file handle is updated, or ESTALE on failure. + */ +int +nfs_refresh_fh(nfsnode_t np, vfs_context_t ctx) +{ + vnode_t dvp, vp = NFSTOV(np); + nfsnode_t dnp; + const char *v_name = vnode_getname(vp); + char *name; + int namelen, fhsize, refreshed; + int error, wanted = 0; + uint8_t *fhp; + struct timespec ts = {2, 0}; + + NFS_VNOP_DBG("vnode is %d\n", vnode_vtype(vp)); + + dvp = vnode_parent(vp); + if ((vnode_vtype(vp) != VREG && vnode_vtype(vp) != VLNK) || + v_name == NULL || *v_name == '\0' || dvp == NULL) { + if (v_name != NULL) + vnode_putname(v_name); + return (ESTALE); + } + dnp = VTONFS(dvp); + + namelen = strlen(v_name); + MALLOC(name, char *, namelen + 1, M_TEMP, M_WAITOK); + if (name == NULL) { + vnode_putname(v_name); + return (ESTALE); + } + bcopy(v_name, name, namelen+1); + NFS_VNOP_DBG("Trying to refresh %s : %s\n", v_name, name); + vnode_putname(v_name); + + /* Allocate the maximum size file handle */ + MALLOC(fhp, uint8_t *, NFS4_FHSIZE, M_TEMP, M_WAITOK); + if (fhp == NULL) { + FREE(name, M_TEMP); + return (ESTALE); + } + + if ((error = nfs_node_lock(np))) { + FREE(name, M_TEMP); + FREE(fhp, M_TEMP); + return (ESTALE); + } + + fhsize = np->n_fhsize; + bcopy(np->n_fhp, fhp, fhsize); + while (ISSET(np->n_flag, NREFRESH)) { + SET(np->n_flag, NREFRESHWANT); + NFS_VNOP_DBG("Waiting for refresh of %s\n", name); + msleep(np, &np->n_lock, PZERO-1, "nfsrefreshwant", &ts); + if ((error = nfs_sigintr(NFSTONMP(np), NULL, vfs_context_thread(ctx), 0))) + break; + } + refreshed = error ? 0 : !NFS_CMPFH(np, fhp, fhsize); + SET(np->n_flag, NREFRESH); + nfs_node_unlock(np); + + NFS_VNOP_DBG("error = %d, refreshed = %d\n", error, refreshed); + if (error || refreshed) + goto nfsmout; + + /* Check that there are no open references for this file */ + lck_mtx_lock(&np->n_openlock); + if (np->n_openrefcnt || !TAILQ_EMPTY(&np->n_opens) || !TAILQ_EMPTY(&np->n_lock_owners)) { + int cnt = 0; + struct nfs_open_file *ofp; + + TAILQ_FOREACH(ofp, &np->n_opens, nof_link) { + cnt += ofp->nof_opencnt; + } + if (cnt) { + lck_mtx_unlock(&np->n_openlock); + NFS_VNOP_DBG("Can not refresh file handle for %s with open state\n", name); + NFS_VNOP_DBG("\topenrefcnt = %d, opens = %d lock_owners = %d\n", + np->n_openrefcnt, cnt, !TAILQ_EMPTY(&np->n_lock_owners)); + error = ESTALE; + goto nfsmout; + } + } + lck_mtx_unlock(&np->n_openlock); + /* + * Since the FH is currently stale we should not be able to + * establish any open state until the FH is refreshed. + */ + + error = nfs_node_lock(np); + nfsmout_if(error); + /* + * Symlinks should never need invalidations and are holding + * the one and only nfsbuf in an uncached acquired state + * trying to do a readlink. So we will hang if we invalidate + * in that case. Only in in the VREG case do we need to + * invalidate. + */ + if (vnode_vtype(vp) == VREG) { + np->n_flag &= ~NNEEDINVALIDATE; + nfs_node_unlock(np); + error = nfs_vinvalbuf(vp, V_IGNORE_WRITEERR, ctx, 1); + if (error) + NFS_VNOP_DBG("nfs_vinvalbuf returned %d\n", error); + nfsmout_if(error); + } else { + nfs_node_unlock(np); + } + + NFS_VNOP_DBG("Looking up %s\n", name); + error = nfs_lookitup(dnp, name, namelen, ctx, &np); + if (error) + NFS_VNOP_DBG("nfs_lookitup returned %d\n", error); + +nfsmout: + nfs_node_lock_force(np); + wanted = ISSET(np->n_flag, NREFRESHWANT); + CLR(np->n_flag, NREFRESH|NREFRESHWANT); + nfs_node_unlock(np); + if (wanted) + wakeup(np); + + if (error == 0) + NFS_VNOP_DBG("%s refreshed file handle\n", name); + + FREE(name, M_TEMP); + FREE(fhp, M_TEMP); + + return (error ? ESTALE : 0); +} + +int +nfs_getattr(nfsnode_t np, struct nfs_vattr *nvap, vfs_context_t ctx, int flags) +{ + int error; + +retry: + error = nfs_getattr_internal(np, nvap, ctx, flags); + if (error == ESTALE) { + error = nfs_refresh_fh(np, ctx); + if (!error) + goto retry; + } + return (error); +} int -nfs_getattr(nfsnode_t np, struct nfs_vattr *nvap, vfs_context_t ctx, int alreadylocked) +nfs_getattr_internal(nfsnode_t np, struct nfs_vattr *nvap, vfs_context_t ctx, int flags) { struct nfsmount *nmp; - int error = 0, lockerror = ENOENT, nfsvers, avoidfloods; + int error = 0, nfsvers, inprogset = 0, wanted = 0, avoidfloods; + struct nfs_vattr nvattr; + struct timespec ts = { 2, 0 }; u_int64_t xid; FSDBG_TOP(513, np->n_size, np, np->n_vattr.nva_size, np->n_flag); + nmp = NFSTONMP(np); + + if (nfs_mount_gone(nmp)) + 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)) { - if (!alreadylocked) - nfs_lock(np, NFS_NODE_LOCK_FORCE); + nfs_node_lock_force(np); np->n_flag |= NCHG; - if (!alreadylocked) - nfs_unlock(np); + nfs_node_unlock(np); } /* Update size, if necessary */ - if (!alreadylocked && ISSET(np->n_flag, NUPDATESIZE)) + if (ISSET(np->n_flag, NUPDATESIZE)) nfs_data_update_size(np, 0); - /* - * First look in the cache. - */ - if ((error = nfs_getattrcache(np, nvap, alreadylocked)) == 0) - goto nfsmout; - if (error != ENOENT) - goto nfsmout; - - nmp = NFSTONMP(np); - if (!nmp) { - error = ENXIO; - goto nfsmout; - } - nfsvers = nmp->nm_vers; + 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, or + * - the server isn't responding + */ + 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; + } + nfs_node_unlock(np); + + nmp = NFSTONMP(np); + if (nfs_mount_gone(nmp)) + error = ENXIO; + if (error) + goto nfsmout; + + /* + * Return cached attributes if they are valid, + * if the server doesn't respond, and this is + * some softened up style of mount. + */ + if (NATTRVALID(np) && nfs_use_cache(nmp)) + flags |= NGA_SOFT; /* - * 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 ((nfsvers != NFS_VER2) && (nfs_access_cache_timeout > 0)) { - if (!alreadylocked && ((error = lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE)))) - goto nfsmout; + if (!(flags & NGA_ACL) && (nfsvers != NFS_VER2) && nfs_access_for_getattr && (nfs_access_cache_timeout > 0)) { if (nfs_attrcachetimeout(np) > 0) { - /* OSAddAtomic(1, (SInt32*)&nfsstats.accesscache_misses); */ - u_long mode = NFS_ACCESS_ALL; - error = nmp->nm_funcs->nf_access_rpc(np, &mode, ctx); + /* OSAddAtomic(1, &nfsstats.accesscache_misses); */ + u_int32_t access = NFS_ACCESS_ALL; + int rpcflags = 0; + + /* Return cached attrs if server doesn't respond */ + if (flags & NGA_SOFT) + rpcflags |= R_SOFT; + + error = nmp->nm_funcs->nf_access_rpc(np, &access, rpcflags, ctx); + + if (error == ETIMEDOUT) + goto returncached; + if (error) goto nfsmout; - if ((error = nfs_getattrcache(np, nvap, 1)) == 0) - goto nfsmout; - if (error != ENOENT) + 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; } - } else if (!alreadylocked) { - error = lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE); - nfsmout_if(error); } + avoidfloods = 0; + tryagain: - error = nmp->nm_funcs->nf_getattr_rpc(np, NULL, np->n_fhp, np->n_fhsize, ctx, nvap, &xid); - nfsmout_if(error); - error = nfs_loadattrcache(np, nvap, &xid, 0); + 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); + } + + /* + * If the server didn't respond, return cached attributes. + */ +returncached: + if ((flags & NGA_SOFT) && (error == ETIMEDOUT)) { + nfs_node_lock_force(np); + error = nfs_getattrcache(np, nvap, flags); + if (!error || (error != ENOENT)) { + nfs_node_unlock(np); + goto nfsmout; + } + 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++ < 100) + if (avoidfloods++ < 20) goto tryagain; - /* avoidfloods>1 is bizarre. at 100 pull the plug */ - panic("nfs_getattr: getattr flood\n"); + /* avoidfloods>1 is bizarre. at 20 pull the plug */ + /* just return the last attributes we got */ } - if (NFS_CHANGED(nfsvers, np, nvap)) { +nfsmout: + nfs_node_lock_force(np); + if (inprogset) { + wanted = ISSET(np->n_flag, NGETATTRWANT); + CLR(np->n_flag, (NGETATTRINPROG | NGETATTRWANT)); + } + if (!error) { + /* check if the node changed on us */ vnode_t vp = NFSTOV(np); enum vtype vtype = vnode_vtype(vp); - FSDBG(513, -1, np, -1, np); - if (vtype == VDIR) { - nfs_invaldir(np); - /* purge name cache entries */ - if (NFS_CHANGED_NC(nfsvers, np, nvap)) { - np->n_flag &= ~NNEGNCENTRIES; - cache_purge(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 (!alreadylocked) { - nfs_unlock(np); - lockerror = ENOENT; + if (NFS_CHANGED(nfsvers, np, nvap)) { + FSDBG(513, -1, np, -1, np); + if (vtype == VDIR) { + NFS_VNOP_DBG("Invalidate directory 0x%llx\n", + (uint64_t)VM_KERNEL_ADDRPERM(vp)); + nfs_invaldir(np); + } + nfs_node_unlock(np); + if (wanted) + wakeup(np); error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1); FSDBG(513, -1, np, -2, error); - if (!error) - error = lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE); if (!error) { - if (vtype == VDIR) - NFS_CHANGED_UPDATE_NC(nfsvers, np, nvap); + nfs_node_lock_force(np); NFS_CHANGED_UPDATE(nfsvers, np, nvap); + nfs_node_unlock(np); } } else { - /* invalidate later */ - np->n_flag |= NNEEDINVALIDATE; + nfs_node_unlock(np); + if (wanted) + wakeup(np); + } + } else { + nfs_node_unlock(np); + if (wanted) + wakeup(np); + } + + 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; } } -nfsmout: - if (!lockerror) - nfs_unlock(np); FSDBG_BOT(513, np->n_size, error, np->n_vattr.nva_size, np->n_flag); return (error); } @@ -936,7 +1685,29 @@ nfsmout: /* * NFS getattr call from vfs. */ -static int + +/* + * 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) + +int nfs3_vnop_getattr( struct vnop_getattr_args /* { struct vnodeop_desc *a_desc; @@ -950,7 +1721,20 @@ nfs3_vnop_getattr( struct vnode_attr *vap = ap->a_vap; dev_t rdev; - error = nfs_getattr(VTONFS(ap->a_vp), &nva, ap->a_context, 0); + /* + * 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); + + 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); @@ -965,7 +1749,6 @@ nfs3_vnop_getattr( 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, nfs_iosize); 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); @@ -983,7 +1766,7 @@ nfs3_vnop_getattr( /* * NFS setattr call. */ -static int +int nfs_vnop_setattr( struct vnop_setattr_args /* { struct vnodeop_desc *a_desc; @@ -998,17 +1781,20 @@ nfs_vnop_setattr( struct nfsmount *nmp; struct vnode_attr *vap = ap->a_vap; int error = 0; - int biosize, nfsvers; - u_quad_t origsize; + 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) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; + namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR); biosize = nmp->nm_biosize; /* Disallow write attempts if the filesystem is mounted read-only. */ @@ -1043,19 +1829,63 @@ nfs_vnop_setattr( FSDBG_TOP(512, np->n_size, vap->va_data_size, np->n_vattr.nva_size, np->n_flag); /* clear NNEEDINVALIDATE, if set */ - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + if ((error = nfs_node_lock(np))) return (error); if (np->n_flag & NNEEDINVALIDATE) np->n_flag &= ~NNEEDINVALIDATE; - nfs_unlock(np); + nfs_node_unlock(np); /* flush everything */ error = nfs_vinvalbuf(vp, (vap->va_data_size ? V_SAVE : 0) , ctx, 1); if (error) { - printf("nfs_setattr: nfs_vinvalbuf %d\n", 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); } - nfs_data_lock(np, NFS_NODE_LOCK_EXCLUSIVE); + 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; + } + } + } + nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE); if (np->n_size > vap->va_data_size) { /* shrinking? */ daddr64_t obn, bn; int neweofoff, mustwrite; @@ -1111,7 +1941,7 @@ nfs_vnop_setattr( // Note: bp has been released if (error) { FSDBG(512, bp, 0xd00dee, 0xbad, error); - nfs_lock(np, NFS_NODE_LOCK_FORCE); + nfs_node_lock_force(np); np->n_error = error; np->n_flag |= NWRITEERR; /* @@ -1122,10 +1952,10 @@ nfs_vnop_setattr( * we may no longer know the correct size) */ NATTRINVALIDATE(np); - nfs_unlock(np); + nfs_node_unlock(np); nfs_data_unlock(np); nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1); - nfs_data_lock(np, NFS_NODE_LOCK_EXCLUSIVE); + nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE); error = 0; } } @@ -1134,79 +1964,118 @@ nfs_vnop_setattr( 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; + 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_lock(np, NFS_NODE_LOCK_SHARED))) + if ((error = nfs_node_lock(np))) return (error); if ((np->n_flag & NMODIFIED) && (vnode_vtype(vp) == VREG)) { - nfs_unlock(np); + nfs_node_unlock(np); error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1); if (error == EINTR) return (error); } else { - nfs_unlock(np); + nfs_node_unlock(np); } } - if (VATTR_IS_ACTIVE(vap, va_mode) || - VATTR_IS_ACTIVE(vap, va_uid) || - VATTR_IS_ACTIVE(vap, va_gid)) { - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) { - if (VATTR_IS_ACTIVE(vap, va_data_size)) - nfs_data_unlock(np); - return (error); - } - NMODEINVALIDATE(np); - nfs_unlock(np); - dvp = vnode_getparent(vp); - vname = vnode_getname(vp); - dnp = (dvp && vname) ? VTONFS(dvp) : NULL; - if (dnp) { - error = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE); - if (error) { - dnp = NULL; - error = 0; + 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 (dnp) { - nfs_dulookup_init(&dul, dnp, vname, strlen(vname)); - nfs_dulookup_start(&dul, dnp, ctx); } } - error = nmp->nm_funcs->nf_setattr_rpc(np, vap, ctx, 0); - - if (VATTR_IS_ACTIVE(vap, va_mode) || - VATTR_IS_ACTIVE(vap, va_uid) || - VATTR_IS_ACTIVE(vap, va_gid)) { - if (dnp) { - nfs_dulookup_finish(&dul, dnp, ctx); - nfs_unlock(dnp); + 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); } - if (dvp != NULLVP) - vnode_put(dvp); - if (vname != NULL) - vnode_putname(vname); } 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)) { + 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, 0); + err = nmp->nm_funcs->nf_setattr_rpc(np, vap, ctx); if (err) - printf("nfs_vnop_setattr: nfs%d_setattr_rpc %d %d\n", nfsvers, error, 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); } @@ -1218,15 +2087,14 @@ int nfs3_setattr_rpc( nfsnode_t np, struct vnode_attr *vap, - vfs_context_t ctx, - int alreadylocked) + vfs_context_t ctx) { struct nfsmount *nmp = NFSTONMP(np); int error = 0, lockerror = ENOENT, status, wccpostattr = 0, nfsvers; - u_int64_t xid; + u_int64_t xid, nextxid; struct nfsm_chain nmreq, nmrep; - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; @@ -1328,9 +2196,8 @@ nfs3_setattr_rpc( } nfsm_chain_build_done(error, &nmreq); nfsmout_if(error); - error = nfs_request(np, NULL, &nmreq, NFSPROC_SETATTR, ctx, - &nmrep, &xid, &status); - if (!alreadylocked && ((lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE)))) + 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 }; @@ -1349,11 +2216,26 @@ nfs3_setattr_rpc( } else { if (!error) error = status; - nfsm_chain_loadattr(error, &nmrep, np, nfsvers, NULL, &xid); + nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid); + } + /* + * 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 (!alreadylocked && !lockerror) - nfs_unlock(np); + if (!lockerror) + nfs_node_unlock(np); nfsm_chain_cleanup(&nmreq); nfsm_chain_cleanup(&nmrep); return (error); @@ -1364,7 +2246,7 @@ nfsmout: * First look in cache * If not found, unlock the directory nfsnode and do the RPC */ -static int +int nfs_vnop_lookup( struct vnop_lookup_args /* { struct vnodeop_desc *a_desc; @@ -1383,7 +2265,7 @@ nfs_vnop_lookup( nfsnode_t dnp, np; struct nfsmount *nmp; mount_t mp; - int nfsvers, error, lockerror = ENOENT, isdot, isdotdot, negnamecache; + int nfsvers, error, busyerror = ENOENT, isdot, isdotdot, negnamecache; u_int64_t xid; struct nfs_vattr nvattr; int ngflags; @@ -1394,30 +2276,22 @@ nfs_vnop_lookup( *vpp = NULLVP; dnp = VTONFS(dvp); + NVATTR_INIT(&nvattr); mp = vnode_mount(dvp); nmp = VFSTONFS(mp); - if (!nmp) { + if (nfs_mount_gone(nmp)) { error = ENXIO; goto error_return; } nfsvers = nmp->nm_vers; - negnamecache = !(nmp->nm_flag & NFSMNT_NONEGNAMECACHE); + negnamecache = !NMFLAG(nmp, NONEGNAMECACHE); - error = lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE); - if (!error) - error = nfs_getattr(dnp, &nvattr, ctx, 1); - if (error) + 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; - if (NFS_CHANGED_NC(nfsvers, dnp, &nvattr)) { - /* - * This directory has changed on us. - * Purge any name cache entries. - */ - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(dvp); - NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &nvattr); - } error = cache_lookup(dvp, vpp, cnp); switch (error) { @@ -1426,15 +2300,27 @@ nfs_vnop_lookup( 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 */ - OSAddAtomic(1, (SInt32*)&nfsstats.lookupcache_hits); + OSAddAtomic64(1, &nfsstats.lookupcache_hits); - nfs_unlock(dnp); - lockerror = ENOENT; + 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 = ctx; @@ -1459,15 +2345,20 @@ nfs_vnop_lookup( 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? */ nmp = VTONMP(dvp); - if (!nmp) { + if (nfs_mount_gone(nmp)) { error = ENXIO; goto error_return; } if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME) && - (cnp->cn_namelen > (long)nmp->nm_fsattr.nfsa_maxname)) { + (cnp->cn_namelen > (int)nmp->nm_fsattr.nfsa_maxname)) { error = ENAMETOOLONG; goto error_return; } @@ -1475,18 +2366,17 @@ nfs_vnop_lookup( error = 0; newvp = NULLVP; - OSAddAtomic(1, (SInt32*)&nfsstats.lookupcache_misses); + OSAddAtomic64(1, &nfsstats.lookupcache_misses); 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, ctx, req, &xid, &fh, &nvattr); + error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, req, &xid, &fh, &nvattr); nfsmout_if(error); /* is the file handle the same as this directory's file handle? */ isdot = NFS_CMPFH(dnp, fh.fh_data, fh.fh_len); found: - if (flags & ISLASTCN) { switch (cnp->cn_nameiop) { case DELETE: @@ -1503,8 +2393,6 @@ found: } if (isdotdot) { - nfs_unlock(dnp); - lockerror = ENOENT; newvp = vnode_getparent(dvp); if (!newvp) { error = ENOENT; @@ -1515,15 +2403,17 @@ found: if (error) goto error_return; newvp = dvp; + nfs_node_lock_force(dnp); if (fh.fh_len && (dnp->n_xid <= xid)) nfs_loadattrcache(dnp, &nvattr, &xid, 0); + nfs_node_unlock(dnp); } else { ngflags = (cnp->cn_flags & MAKEENTRY) ? NG_MAKEENTRY : 0; - error = nfs_nget(mp, dnp, cnp, fh.fh_data, fh.fh_len, &nvattr, &xid, ngflags, &np); + 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_unlock(np); + nfs_node_unlock(np); } *vpp = newvp; @@ -1540,12 +2430,15 @@ nfsmout: 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: - if (!lockerror) - nfs_unlock(dnp); + NVATTR_CLEANUP(&nvattr); + if (!busyerror) + nfs_node_clear_busy(dnp); if (error && *vpp) { vnode_put(*vpp); *vpp = NULLVP; @@ -1553,30 +2446,12 @@ error_return: return (error); } -/* - * NFS read call. - * Just call nfs_bioread() to do the work. - */ -static int -nfs_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) -{ - if (vnode_vtype(ap->a_vp) != VREG) - return (EPERM); - return (nfs_bioread(VTONFS(ap->a_vp), ap->a_uio, ap->a_ioflag, NULL, ap->a_context)); -} - +int nfs_readlink_nocache = DEFAULT_READLINK_NOCACHE; /* * NFS readlink call */ -static int +int nfs_vnop_readlink( struct vnop_readlink_args /* { struct vnodeop_desc *a_desc; @@ -1588,70 +2463,84 @@ nfs_vnop_readlink( vfs_context_t ctx = ap->a_context; nfsnode_t np = VTONFS(ap->a_vp); struct nfsmount *nmp; - int error = 0, lockerror, nfsvers, changed = 0, n; + int error = 0, nfsvers; uint32_t buflen; - struct uio *uio = ap->a_uio; - struct nfs_vattr nvattr; + uio_t uio = ap->a_uio; struct nfsbuf *bp = NULL; + struct timespec ts; + int timeo; if (vnode_vtype(ap->a_vp) != VLNK) return (EPERM); - if (uio_uio_resid(uio) == 0) + if (uio_resid(uio) == 0) return (0); - if (uio->uio_offset < 0) + if (uio_offset(uio) < 0) return (EINVAL); nmp = VTONMP(ap->a_vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; - error = lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE); - if (!error) - error = nfs_getattr(np, &nvattr, ctx, 1); - if (error) { - if (!lockerror) - nfs_unlock(np); + + /* nfs_getattr() will check changed and purge caches */ + if ((error = nfs_getattr(np, NULL, ctx, nfs_readlink_nocache ? NGA_UNCACHED : NGA_CACHED))) { FSDBG(531, np, 0xd1e0001, 0, error); return (error); } - if (NFS_CHANGED(nfsvers, np, &nvattr)) { - /* link changed, so just ignore NB_CACHE */ - changed = 1; - NFS_CHANGED_UPDATE(nfsvers, np, &nvattr); - } - nfs_unlock(np); - OSAddAtomic(1, (SInt32*)&nfsstats.biocache_readlinks); - error = nfs_buf_get(np, 0, NFS_MAXPATHLEN, vfs_context_thread(ctx), NBLK_READ, &bp); + if (nfs_readlink_nocache) { + timeo = nfs_attrcachetimeout(np); + nanouptime(&ts); + } + +retry: + OSAddAtomic64(1, &nfsstats.biocache_readlinks); + error = nfs_buf_get(np, 0, NFS_MAXPATHLEN, vfs_context_thread(ctx), NBLK_META, &bp); if (error) { FSDBG(531, np, 0xd1e0002, 0, error); return (error); } - if (changed) - CLR(bp->nb_flags, NB_CACHE); + + if (nfs_readlink_nocache) { + NFS_VNOP_DBG("timeo = %d ts.tv_sec = %ld need refresh = %d cached = %d\n", timeo, ts.tv_sec, + (np->n_rltim.tv_sec + timeo) < ts.tv_sec || nfs_readlink_nocache > 1, + ISSET(bp->nb_flags, NB_CACHE) == NB_CACHE); + /* n_rltim is synchronized by the associated nfs buf */ + if (ISSET(bp->nb_flags, NB_CACHE) && ((nfs_readlink_nocache > 1) || ((np->n_rltim.tv_sec + timeo) < ts.tv_sec))) { + SET(bp->nb_flags, NB_INVAL); + nfs_buf_release(bp, 0); + goto retry; + } + } if (!ISSET(bp->nb_flags, NB_CACHE)) { - SET(bp->nb_flags, NB_READ); - CLR(bp->nb_flags, NB_DONE); - OSAddAtomic(1, (SInt32*)&nfsstats.readlink_bios); +readagain: + OSAddAtomic64(1, &nfsstats.readlink_bios); buflen = bp->nb_bufsize; error = nmp->nm_funcs->nf_readlink_rpc(np, bp->nb_data, &buflen, ctx); if (error) { + if (error == ESTALE) { + NFS_VNOP_DBG("Stale FH from readlink rpc\n"); + error = nfs_refresh_fh(np, ctx); + if (error == 0) + goto readagain; + } SET(bp->nb_flags, NB_ERROR); bp->nb_error = error; + NFS_VNOP_DBG("readlink failed %d\n", error); } else { bp->nb_validoff = 0; bp->nb_validend = buflen; + np->n_rltim = ts; + NFS_VNOP_DBG("readlink of %.*s\n", bp->nb_validend, (char *)bp->nb_data); } - nfs_buf_iodone(bp); - } - if (!error) { - // LP64todo - fix this! - n = min(uio_uio_resid(uio), bp->nb_validend); - if (n > 0) - error = uiomove(bp->nb_data, n, uio); + } else { + NFS_VNOP_DBG("got cached link of %.*s\n", bp->nb_validend, (char *)bp->nb_data); } + + 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); @@ -1670,7 +2559,7 @@ nfs3_readlink_rpc(nfsnode_t np, char *buf, uint32_t *buflenp, vfs_context_t ctx) struct nfsm_chain nmreq, nmrep; nmp = NFSTONMP(np); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; nfsm_chain_null(&nmreq); @@ -1680,9 +2569,8 @@ nfs3_readlink_rpc(nfsnode_t np, char *buf, uint32_t *buflenp, vfs_context_t ctx) 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, - &nmrep, &xid, &status); - if ((lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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); @@ -1705,7 +2593,7 @@ nfs3_readlink_rpc(nfsnode_t np, char *buf, uint32_t *buflenp, vfs_context_t ctx) *buflenp = len; nfsmout: if (!lockerror) - nfs_unlock(np); + nfs_node_unlock(np); nfsm_chain_cleanup(&nmreq); nfsm_chain_cleanup(&nmrep); return (error); @@ -1716,37 +2604,60 @@ nfsmout: * Ditto above */ int -nfs_read_rpc(nfsnode_t np, struct uio *uiop, vfs_context_t ctx) +nfs_read_rpc(nfsnode_t np, uio_t uio, vfs_context_t ctx) { struct nfsmount *nmp; int error = 0, nfsvers, eof = 0; - size_t nmrsize, len, retlen, tsiz; + 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, uiop->uio_offset, uio_uio_resid(uiop), 0); + FSDBG_TOP(536, np, uio_offset(uio), uio_resid(uio), 0); nmp = NFSTONMP(np); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); 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) && (nfsvers == NFS_VER2)) { - FSDBG_BOT(536, np, 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); } - txoffset = uiop->uio_offset; - while (tsiz > 0) { - len = retlen = (tsiz > nmrsize) ? nmrsize : tsiz; + 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; + } + 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, uiop, &retlen, &eof); + 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); + } + 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; txoffset += retlen; @@ -1758,7 +2669,7 @@ nfs_read_rpc(nfsnode_t np, struct uio *uiop, vfs_context_t ctx) tsiz = 0; } - FSDBG_BOT(536, np, eof, uio_uio_resid(uiop), error); + FSDBG_BOT(536, np, eof, uio_resid(uio), error); return (error); } @@ -1777,7 +2688,7 @@ nfs3_read_rpc_async( struct nfsm_chain nmreq; nmp = NFSTONMP(np); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; @@ -1794,7 +2705,7 @@ nfs3_read_rpc_async( } nfsm_chain_build_done(error, &nmreq); nfsmout_if(error); - error = nfs_request_async(np, NULL, &nmreq, NFSPROC_READ, thd, cred, cb, reqp); + error = nfs_request_async(np, NULL, &nmreq, NFSPROC_READ, thd, cred, NULL, 0, cb, reqp); nfsmout: nfsm_chain_cleanup(&nmreq); return (error); @@ -1804,7 +2715,7 @@ int nfs3_read_rpc_async_finish( nfsnode_t np, struct nfsreq *req, - struct uio *uiop, + uio_t uio, size_t *lenp, int *eofp) { @@ -1815,7 +2726,7 @@ nfs3_read_rpc_async_finish( struct nfsm_chain nmrep; nmp = NFSTONMP(np); - if (!nmp) { + if (nfs_mount_gone(nmp)) { nfs_request_async_cancel(req); return (ENXIO); } @@ -1827,7 +2738,7 @@ nfs3_read_rpc_async_finish( if (error == EINPROGRESS) /* async request restarted */ return (error); - if ((lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + if ((lockerror = nfs_node_lock(np))) error = lockerror; if (nfsvers == NFS_VER3) nfsm_chain_postop_attr_update(error, &nmrep, np, &xid); @@ -1837,15 +2748,15 @@ nfs3_read_rpc_async_finish( nfsm_chain_adv(error, &nmrep, NFSX_UNSIGNED); nfsm_chain_get_32(error, &nmrep, eof); } else { - nfsm_chain_loadattr(error, &nmrep, np, nfsvers, NULL, &xid); + nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid); } if (!lockerror) - nfs_unlock(np); + 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), uiop); + error = nfsm_chain_get_uio(&nmrep, MIN(retlen, *lenp), uio); if (eofp) { if (nfsvers == NFS_VER3) { if (!eof && !retlen) @@ -1875,37 +2786,36 @@ nfs_vnop_write( } */ *ap) { vfs_context_t ctx = ap->a_context; - struct uio *uio = ap->a_uio; + 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 nfs_vattr nvattr; struct nfsmount *nmp = VTONMP(vp); daddr64_t lbn; int biosize; int n, on, error = 0; off_t boff, start, end; - struct iovec_32 iov; - struct uio auio; + uio_t auio; + char auio_buf [ UIO_SIZEOF(1) ]; thread_t thd; kauth_cred_t cred; - FSDBG_TOP(515, np, uio->uio_offset, uio_uio_resid(uio), ioflag); + FSDBG_TOP(515, np, uio_offset(uio), uio_resid(uio), ioflag); if (vnode_vtype(vp) != VREG) { - FSDBG_BOT(515, np, uio->uio_offset, uio_uio_resid(uio), EIO); + 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_NODE_LOCK_SHARED); + nfs_data_lock(np, NFS_DATA_LOCK_SHARED); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) { + if ((error = nfs_node_lock(np))) { nfs_data_unlock(np); - FSDBG_BOT(515, np, uio->uio_offset, uio_uio_resid(uio), error); + FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), error); return (error); } np->n_wrbusy++; @@ -1916,71 +2826,117 @@ nfs_vnop_write( } if (np->n_flag & NNEEDINVALIDATE) { np->n_flag &= ~NNEEDINVALIDATE; - nfs_unlock(np); + nfs_node_unlock(np); nfs_data_unlock(np); nfs_vinvalbuf(vp, V_SAVE|V_IGNORE_WRITEERR, ctx, 1); - nfs_data_lock(np, NFS_NODE_LOCK_SHARED); - if (error || ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE)))) - goto out; + nfs_data_lock(np, NFS_DATA_LOCK_SHARED); + } else { + nfs_node_unlock(np); } - if (error) { - nfs_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_unlock(np); + nfs_node_unlock(np); nfs_data_unlock(np); error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1); - nfs_data_lock(np, NFS_NODE_LOCK_SHARED); - if (error || ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE)))) { - FSDBG(515, np, uio->uio_offset, 0x10bad01, error); + nfs_data_lock(np, NFS_DATA_LOCK_SHARED); + if (error) { + FSDBG(515, np, uio_offset(uio), 0x10bad01, error); goto out; } + } else { + nfs_node_unlock(np); } if (ioflag & IO_APPEND) { - NATTRINVALIDATE(np); - nfs_unlock(np); nfs_data_unlock(np); - error = nfs_getattr(np, &nvattr, ctx, 0); + /* 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_NODE_LOCK_EXCLUSIVE); - if (error || ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE)))) { - FSDBG(515, np, uio->uio_offset, 0x10bad02, error); + nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE); + if (error) { + FSDBG(515, np, uio_offset(uio), 0x10bad02, error); goto out; } - uio->uio_offset = np->n_size; + uio_setoffset(uio, np->n_size); } } - if (uio->uio_offset < 0) { - nfs_unlock(np); + if (uio_offset(uio) < 0) { error = EINVAL; - FSDBG_BOT(515, np, uio->uio_offset, 0xbad0ff, error); + FSDBG_BOT(515, np, uio_offset(uio), 0xbad0ff, error); goto out; } - if (uio_uio_resid(uio) == 0) { - nfs_unlock(np); + if (uio_resid(uio) == 0) goto out; - } - nfs_unlock(np); - - if (((uio->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 */ + 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_NODE_LOCK_EXCLUSIVE); + nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE); + + /* + * Also, if the write begins after the previous EOF buffer, make sure to zero + * and validate the new bytes in that buffer. + */ + struct nfsbuf *eofbp = NULL; + daddr64_t eofbn = np->n_size / biosize; + int eofoff = np->n_size % biosize; + lbn = uio_offset(uio) / biosize; + + if (eofoff && (eofbn < lbn)) { + if ((error = nfs_buf_get(np, eofbn, biosize, thd, NBLK_WRITE|NBLK_ONLYVALID, &eofbp))) + goto out; + np->n_size += (biosize - eofoff); + 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) { + /* + * 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. + */ + 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); + } + eofoff += PAGE_SIZE - poff; + i++; + } + nfs_buf_release(eofbp, 1); + } + } } do { - OSAddAtomic(1, (SInt32*)&nfsstats.biocache_writes); - lbn = uio->uio_offset / biosize; - on = uio->uio_offset % biosize; - // LP64todo - fix this - n = min((unsigned)(biosize - on), uio_uio_resid(uio)); + 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 @@ -2017,7 +2973,7 @@ again: */ if (bp->nb_dirtyend > 0) { if (on > bp->nb_dirtyend || (on + n) < bp->nb_dirtyoff || bp->nb_dirty) { - FSDBG(515, np, uio->uio_offset, bp, 0xd15c001); + 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)); @@ -2036,7 +2992,7 @@ again: 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->uio_offset, bp, 0xd15c002); + 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)); @@ -2049,7 +3005,7 @@ 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->uio_offset, bp, 0xd15c003); + 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; @@ -2070,17 +3026,11 @@ again: * If there was a partial buf at the old eof, validate * and zero the new bytes. */ - if ((uio->uio_offset + n) > (off_t)np->n_size) { - struct nfsbuf *eofbp = NULL; + if ((uio_offset(uio) + n) > (off_t)np->n_size) { daddr64_t eofbn = np->n_size / biosize; - int eofoff = np->n_size % biosize; - int neweofoff = (uio->uio_offset + n) % biosize; + int neweofoff = (uio_offset(uio) + n) % biosize; - FSDBG(515, 0xb1ffa000, uio->uio_offset + n, eofoff, neweofoff); - - if (eofoff && (eofbn < lbn) && - ((error = nfs_buf_get(np, eofbn, biosize, thd, NBLK_WRITE|NBLK_ONLYVALID, &eofbp)))) - goto out; + FSDBG(515, 0xb1ffa000, uio_offset(uio) + n, eofoff, neweofoff); /* if we're extending within the same last block */ /* and the block is flagged as being cached... */ @@ -2111,45 +3061,13 @@ again: } } } - np->n_size = uio->uio_offset + n; - nfs_lock(np, NFS_NODE_LOCK_FORCE); + np->n_size = uio_offset(uio) + n; + nfs_node_lock_force(np); CLR(np->n_flag, NUPDATESIZE); np->n_flag |= NMODIFIED; - nfs_unlock(np); + 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) { - /* - * 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. - */ - 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); - } } /* * If dirtyend exceeds file size, chop it down. This should @@ -2176,12 +3094,14 @@ again: * * 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_NOCACHE) && !ISSET(bp->nb_flags, NB_CACHE) && (n < biosize)) { + if (!ISSET(bp->nb_flags, NB_CACHE) && (n < biosize)) { int firstpg, lastpg, dirtypg; int firstpgoff, lastpgoff; start = end = -1; @@ -2200,6 +3120,22 @@ again: 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 */ @@ -2222,29 +3158,23 @@ again: NFS_BUF_MAP(bp); /* setup uio for read(s) */ boff = NBOFF(bp); - auio.uio_iovs.iov32p = &iov; - auio.uio_iovcnt = 1; -#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_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 */ - auio.uio_offset = boff + start; - iov.iov_len = on - start; - uio_uio_resid_set(&auio, iov.iov_len); - iov.iov_base = (uintptr_t) bp->nb_data + start; - error = nfs_read_rpc(np, &auio, ctx); - if (error) /* couldn't read the data, so treat buffer as NOCACHE */ + 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)); - if (uio_uio_resid(&auio) > 0) { - FSDBG(516, bp, (caddr_t)iov.iov_base - bp->nb_data, uio_uio_resid(&auio), 0xd00dee01); - // LP64todo - fix this - bzero((caddr_t)iov.iov_base, uio_uio_resid(&auio)); + 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 */ @@ -2283,19 +3213,19 @@ again: FSDBG(516, bp, start, end - start, 0xd00dee00); bzero(bp->nb_data + start, end - start); error = 0; - } else if (!ISSET(bp->nb_flags, NB_NOCACHE)) { + } else { /* now we'll read the (rest of the) data */ - auio.uio_offset = boff + start; - iov.iov_len = end - start; - uio_uio_resid_set(&auio, iov.iov_len); - iov.iov_base = (uintptr_t) (bp->nb_data + start); - error = nfs_read_rpc(np, &auio, ctx); - if (error) /* couldn't read the data, so treat buffer as NOCACHE */ + 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)); - if (uio_uio_resid(&auio) > 0) { - FSDBG(516, bp, (caddr_t)iov.iov_base - bp->nb_data, uio_uio_resid(&auio), 0xd00dee02); - // LP64todo - fix this - bzero((caddr_t)iov.iov_base, uio_uio_resid(&auio)); + 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) { @@ -2307,15 +3237,16 @@ again: 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 < trunc_page_32(on); start+=PAGE_SIZE) + 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) > round_page_32(on+n-1); end-=PAGE_SIZE) + 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; @@ -2323,9 +3254,9 @@ again: goto out; } - nfs_lock(np, NFS_NODE_LOCK_FORCE); + nfs_node_lock_force(np); np->n_flag |= NMODIFIED; - nfs_unlock(np); + nfs_node_unlock(np); NFS_BUF_MAP(bp); error = uiomove((char *)bp->nb_data + on, n, uio); @@ -2372,20 +3303,21 @@ again: * again and not just committed. */ if (ISSET(bp->nb_flags, NB_NEEDCOMMIT)) { - nfs_lock(np, NFS_NODE_LOCK_FORCE); + 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_unlock(np); + nfs_node_unlock(np); } if (ioflag & IO_SYNC) { error = nfs_buf_write(bp); if (error) goto out; - } else if (((n + on) == biosize) || (ioflag & IO_NOCACHE) || ISSET(bp->nb_flags, NB_NOCACHE)) { + } 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) @@ -2395,21 +3327,21 @@ again: if (!ISSET(bp->nb_flags, NB_DELWRI)) { proc_t p = vfs_context_proc(ctx); if (p && p->p_stats) - OSIncrementAtomic(&p->p_stats->p_ru.ru_oublock); + 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_uio_resid(uio) > 0 && n > 0); + } while (uio_resid(uio) > 0 && n > 0); out: - nfs_lock(np, NFS_NODE_LOCK_FORCE); + nfs_node_lock_force(np); np->n_wrbusy--; - nfs_unlock(np); + nfs_node_unlock(np); nfs_data_unlock(np); - FSDBG_BOT(515, np, uio->uio_offset, uio_uio_resid(uio), error); + FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), error); return (error); } @@ -2420,61 +3352,90 @@ out: int nfs_write_rpc( nfsnode_t np, - struct uio *uiop, + uio_t uio, vfs_context_t ctx, int *iomodep, uint64_t *wverfp) { - return nfs_write_rpc2(np, uiop, vfs_context_thread(ctx), vfs_context_ucred(ctx), iomodep, wverfp); + return nfs_write_rpc2(np, uio, vfs_context_thread(ctx), vfs_context_ucred(ctx), iomodep, wverfp); } int nfs_write_rpc2( nfsnode_t np, - struct uio *uiop, + uio_t uio, thread_t thd, kauth_cred_t cred, int *iomodep, uint64_t *wverfp) { struct nfsmount *nmp; - int error = 0, nfsvers, restart; - int backup, wverfset, commit, committed; + 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 (uiop->uio_iovcnt != 1) + if (uio_iovcnt(uio) != 1) panic("nfs3_write_rpc: iovcnt > 1"); #endif - FSDBG_TOP(537, np, uiop->uio_offset, uio_uio_resid(uiop), *iomodep); + FSDBG_TOP(537, np, uio_offset(uio), uio_resid(uio), *iomodep); nmp = NFSTONMP(np); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; nmwsize = nmp->nm_wsize; - restart = wverfset = 0; + wverfset = 0; committed = NFS_WRITE_FILESYNC; - // LP64todo - fix this - totalsize = tsiz = uio_uio_resid(uiop); - if (((u_int64_t)uiop->uio_offset + (unsigned int)tsiz > 0xffffffff) && (nfsvers == NFS_VER2)) { - FSDBG_BOT(537, np, uiop->uio_offset, uio_uio_resid(uiop), EFBIG); + 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, uiop->uio_offset, len, 0); - error = nmp->nm_funcs->nf_write_rpc_async(np, uiop, len, thd, cred, *iomodep, NULL, &req); + 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) + if (nfs_mount_gone(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) { @@ -2484,11 +3445,9 @@ nfs_write_rpc2( /* check for a short write */ 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); + /* Reset the uio to reflect the actual transfer */ + *uio = *uio_save; + uio_update(uio, totalsize - (tsiz - rlen)); len = rlen; } @@ -2504,34 +3463,32 @@ nfs_write_rpc2( wverfset = 1; } else if (wverf != wverf2) { /* verifier changed, so we need to restart all the writes */ - if (++restart > 10) { + if (++vrestart > 100) { /* give up after too many restarts */ error = EIO; break; } - backup = totalsize - tsiz; - uio_iov_base_add(uiop, -backup); - uio_iov_len_add(uiop, backup); - uiop->uio_offset -= backup; - uio_uio_resid_add(uiop, backup); + *uio = *uio_save; // Reset the uio back to the start committed = NFS_WRITE_FILESYNC; wverfset = 0; tsiz = totalsize; } } + 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, np, 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, - struct uio *uiop, + uio_t uio, size_t len, thread_t thd, kauth_cred_t cred, @@ -2540,36 +3497,39 @@ nfs3_write_rpc_async( struct nfsreq **reqp) { struct nfsmount *nmp; + mount_t mp; int error = 0, nfsvers; - off_t offset; struct nfsm_chain nmreq; nmp = NFSTONMP(np); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; - offset = uiop->uio_offset; + /* 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, offset); + 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, offset); + 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, uiop, len); + 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, cb, reqp); + error = nfs_request_async(np, NULL, &nmreq, NFSPROC_WRITE, thd, cred, NULL, 0, cb, reqp); nfsmout: nfsm_chain_cleanup(&nmreq); return (error); @@ -2591,7 +3551,7 @@ nfs3_write_rpc_async_finish( struct nfsm_chain nmrep; nmp = NFSTONMP(np); - if (!nmp) { + if (nfs_mount_gone(nmp)) { nfs_request_async_cancel(req); return (ENXIO); } @@ -2603,9 +3563,9 @@ nfs3_write_rpc_async_finish( if (error == EINPROGRESS) /* async request restarted */ return (error); nmp = NFSTONMP(np); - if (!nmp) + if (nfs_mount_gone(nmp)) error = ENXIO; - if (!error && (lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + if (!error && (lockerror = nfs_node_lock(np))) error = lockerror; if (nfsvers == NFS_VER3) { struct timespec premtime = { 0, 0 }; @@ -2635,14 +3595,14 @@ nfs3_write_rpc_async_finish( } else { if (!error) error = status; - nfsm_chain_loadattr(error, &nmrep, np, nfsvers, NULL, &xid); + 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_unlock(np); + nfs_node_unlock(np); nfsm_chain_cleanup(&nmrep); if ((committed != NFS_WRITE_FILESYNC) && nfs_allow_async && ((mp = NFSTOMP(np))) && (vfs_flags(mp) & MNT_ASYNC)) @@ -2657,7 +3617,7 @@ nfsmout: * 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 +int nfs3_vnop_mknod( struct vnop_mknod_args /* { struct vnodeop_desc *a_desc; @@ -2677,17 +3637,18 @@ nfs3_vnop_mknod( nfsnode_t np = NULL; struct nfsmount *nmp; nfsnode_t dnp = VTONFS(dvp); - struct nfs_vattr nvattr, dnvattr; + struct nfs_vattr nvattr; fhandle_t fh; - int error = 0, lockerror = ENOENT, status, wccpostattr = 0; + int error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0; struct timespec premtime = { 0, 0 }; - u_long rdev; - u_int64_t xid, dxid; + 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) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; @@ -2705,6 +3666,8 @@ nfs3_vnop_mknod( 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); VATTR_SET_SUPPORTED(vap, va_gid); @@ -2721,7 +3684,7 @@ nfs3_vnop_mknod( 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_string(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen); + 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); @@ -2733,13 +3696,17 @@ nfs3_vnop_mknod( nfsm_chain_add_v2sattr(error, &nmreq, vap, rdev); } nfsm_chain_build_done(error, &nmreq); - nfsmout_if(error); - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; + if (!error) + error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx)); nfsmout_if(error); - error = nfs_request(dnp, NULL, &nmreq, NFSPROC_MKNOD, ctx, &nmrep, &xid, &status); + 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? */ dxid = xid; if (!error && !status) { @@ -2762,28 +3729,22 @@ nfsmout: /* if directory hadn't changed, update namecache mtime */ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr); - if (!wccpostattr) - NATTRINVALIDATE(dnp); - if (!nfs_getattr(dnp, &dnvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, dnp, &dnvattr)) { - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(dvp); - NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnvattr); - } - } + 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, NG_MAKEENTRY, &np); + 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 (!lockerror) - nfs_unlock(dnp); + if (!busyerror) + nfs_node_clear_busy(dnp); if (!error && (gotuid || gotgid) && - (!newvp || nfs_getattrcache(np, &nvattr, 1) || + (!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) */ @@ -2792,21 +3753,21 @@ nfsmout: } if (error) { if (newvp) { - nfs_unlock(np); + nfs_node_unlock(np); vnode_put(newvp); } } else { *vpp = newvp; - nfs_unlock(np); + nfs_node_unlock(np); } return (error); } -static u_long create_verf; +static uint32_t create_verf; /* * NFS file create call */ -static int +int nfs3_vnop_create( struct vnop_create_args /* { struct vnodeop_desc *a_desc; @@ -2821,29 +3782,31 @@ nfs3_vnop_create( vnode_t dvp = ap->a_dvp; struct vnode_attr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; - struct nfs_vattr nvattr, dnvattr; + struct nfs_vattr nvattr; fhandle_t fh; nfsnode_t np = NULL; struct nfsmount *nmp; nfsnode_t dnp = VTONFS(dvp); vnode_t newvp = NULL; - int error = 0, lockerror = ENOENT, status, wccpostattr = 0, fmode = 0; + int error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0, fmode = 0; struct timespec premtime = { 0, 0 }; int nfsvers, gotuid, gotgid; u_int64_t xid, dxid; uint32_t val; struct nfsm_chain nmreq, nmrep; - struct nfsreq *req; + struct nfsreq rq, *req = &rq; struct nfs_dulookup dul; nmp = VTONMP(dvp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; 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); VATTR_SET_SUPPORTED(vap, va_gid); @@ -2853,12 +3816,15 @@ nfs3_vnop_create( 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; + if (!VATTR_IS_ACTIVE(vap, va_access_time) || !VATTR_IS_ACTIVE(vap, va_modify_time)) + vap->va_vaflags |= VA_UTIMES_NULL; + } again: - req = NULL; - nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen); + 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); @@ -2867,14 +3833,16 @@ again: 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_string(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen); + 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); @@ -2887,17 +3855,16 @@ again: } nfsm_chain_build_done(error, &nmreq); nfsmout_if(error); - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; - nfsmout_if(error); error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_CREATE, - vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, &req); + 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) { @@ -2919,27 +3886,21 @@ nfsmout: /* if directory hadn't changed, update namecache mtime */ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr); - if (!wccpostattr) - NATTRINVALIDATE(dnp); - if (!nfs_getattr(dnp, &dnvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, dnp, &dnvattr)) { - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(dvp); - NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnvattr); - } - } + 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, NG_MAKEENTRY, &np); + 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 (!lockerror) - nfs_unlock(dnp); + if (!busyerror) + nfs_node_clear_busy(dnp); if (error) { if ((nfsvers == NFS_VER3) && (fmode & O_EXCL) && (error == NFSERR_NOTSUPP)) { @@ -2947,27 +3908,28 @@ nfsmout: goto again; } if (newvp) { - nfs_unlock(np); + nfs_node_unlock(np); vnode_put(newvp); } } else if ((nfsvers == NFS_VER3) && (fmode & O_EXCL)) { - error = nfs3_setattr_rpc(np, vap, ctx, 1); + 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 = nfs3_setattr_rpc(np, vap, ctx, 1); + error = nfs3_setattr_rpc(np, vap, ctx); } - if (error) { - nfs_unlock(np); + if (error) vnode_put(newvp); - } + else + nfs_node_lock_force(np); } if (!error) *ap->a_vpp = newvp; if (!error && (gotuid || gotgid) && - (!newvp || nfs_getattrcache(np, &nvattr, 1) || + (!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) */ @@ -2975,7 +3937,7 @@ nfsmout: VATTR_CLEAR_SUPPORTED(vap, va_gid); } if (!error) - nfs_unlock(np); + nfs_node_unlock(np); return (error); } @@ -2990,7 +3952,7 @@ nfsmout: * else * do the remove RPC */ -static int +int nfs_vnop_remove( struct vnop_remove_args /* { struct vnodeop_desc *a_desc; @@ -3007,7 +3969,7 @@ nfs_vnop_remove( struct componentname *cnp = ap->a_cnp; nfsnode_t dnp = VTONFS(dvp); nfsnode_t np = VTONFS(vp); - int error = 0, nfsvers, inuse, gotattr = 0, flushed = 0, setsize = 0; + int error = 0, nfsvers, namedattrs, inuse, gotattr = 0, flushed = 0, setsize = 0; struct nfs_vattr nvattr; struct nfsmount *nmp; struct nfs_dulookup dul; @@ -3015,12 +3977,13 @@ nfs_vnop_remove( /* XXX prevent removing a sillyrenamed file? */ nmp = NFSTONMP(dnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; + namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR); again_relock: - error = nfs_lock2(dnp, np, NFS_NODE_LOCK_EXCLUSIVE); + error = nfs_node_set_busy2(dnp, np, vfs_context_thread(ctx)); if (error) return (error); @@ -3033,7 +3996,8 @@ again_relock: np->n_hflag |= NHLOCKED; lck_mtx_unlock(nfs_node_hash_mutex); - nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen); + 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) { @@ -3042,7 +4006,7 @@ again: goto out; } if (inuse && !gotattr) { - if (nfs_getattr(np, &nvattr, ctx, 1)) + if (nfs_getattr(np, &nvattr, ctx, NGA_CACHED)) nvattr.nva_nlink = 1; gotattr = 1; goto again; @@ -3058,27 +4022,33 @@ again: wakeup(np); } lck_mtx_unlock(nfs_node_hash_mutex); - nfs_unlock2(dnp, np); + 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_lock(np, NFS_NODE_LOCK_FORCE); + nfs_node_lock_force(np); NATTRINVALIDATE(np); - nfs_unlock(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); + nfs_name_cache_purge(dnp, np, cnp, ctx); - nfs_dulookup_start(&dul, dnp, ctx); + if (!namedattrs) + nfs_dulookup_start(&dul, dnp, ctx); /* Do the rpc */ error = nmp->nm_funcs->nf_remove_rpc(dnp, cnp->cn_nameptr, cnp->cn_namelen, @@ -3109,30 +4079,36 @@ again: lck_mtx_unlock(nfs_node_hash_mutex); /* clear flags now: won't get nfs_vnop_inactive for recycled vnode */ /* clear all flags other than these */ + nfs_node_lock_force(np); np->n_flag &= (NMODIFIED); - vnode_recycle(vp); 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) { - nfs_dulookup_start(&dul, dnp, ctx); + 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_dulookup_start(&dul, dnp, ctx); + nfs_node_unlock(np); + if (!namedattrs) + nfs_dulookup_start(&dul, dnp, ctx); } - if (!nfs_getattr(dnp, &nvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, dnp, &nvattr)) { - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(dvp); - NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &nvattr); - } - } - nfs_dulookup_finish(&dul, dnp, ctx); + /* 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); @@ -3142,7 +4118,7 @@ out: wakeup(np); } lck_mtx_unlock(nfs_node_hash_mutex); - nfs_unlock2(dnp, np); + nfs_node_clear_busy2(dnp, np); if (setsize) ubc_setsize(vp, 0); return (error); @@ -3155,7 +4131,7 @@ int nfs_removeit(struct nfs_sillyrename *nsp) { struct nfsmount *nmp = NFSTONMP(nsp->nsr_dnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); return nmp->nm_funcs->nf_remove_rpc(nsp->nsr_dnp, nsp->nsr_name, nsp->nsr_namlen, NULL, nsp->nsr_cred); } @@ -3171,7 +4147,7 @@ nfs3_remove_rpc( thread_t thd, kauth_cred_t cred) { - int error = 0, status, wccpostattr = 0; + int error = 0, lockerror = ENOENT, status, wccpostattr = 0; struct timespec premtime = { 0, 0 }; struct nfsmount *nmp; int nfsvers; @@ -3179,7 +4155,7 @@ nfs3_remove_rpc( struct nfsm_chain nmreq, nmrep; nmp = NFSTONMP(dnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; if ((nfsvers == NFS_VER2) && (namelen > NFS_MAXNAMLEN)) @@ -3191,14 +4167,17 @@ nfs3_remove_rpc( 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_string(error, &nmreq, name, namelen); + 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, 0, &nmrep, &xid, &status); + 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(&dnp->n_ncmtime, &premtime, ==)) @@ -3208,6 +4187,8 @@ nfs3_remove_rpc( if (!error) error = status; nfsmout: + if (!lockerror) + nfs_node_unlock(dnp); nfsm_chain_cleanup(&nmreq); nfsm_chain_cleanup(&nmrep); return (error); @@ -3216,7 +4197,7 @@ nfsmout: /* * NFS file rename call */ -static int +int nfs_vnop_rename( struct vnop_rename_args /* { struct vnodeop_desc *a_desc; @@ -3241,7 +4222,6 @@ nfs_vnop_rename( mount_t fmp, tdmp, tmp; struct nfs_vattr nvattr; struct nfsmount *nmp; - struct nfs_dulookup fdul, tdul; fdnp = VTONFS(fdvp); fnp = VTONFS(fvp); @@ -3249,11 +4229,11 @@ nfs_vnop_rename( tnp = tvp ? VTONFS(tvp) : NULL; nmp = NFSTONMP(fdnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; - error = nfs_lock4(fdnp, fnp, tdnp, tnp, NFS_NODE_LOCK_EXCLUSIVE); + error = nfs_node_set_busy4(fdnp, fnp, tdnp, tnp, vfs_context_thread(ctx)); if (error) return (error); @@ -3269,9 +4249,6 @@ nfs_vnop_rename( locked = 1; } - nfs_dulookup_init(&fdul, fdnp, fcnp->cn_nameptr, fcnp->cn_namelen); - nfs_dulookup_init(&tdul, tdnp, tcnp->cn_nameptr, tcnp->cn_namelen); - /* Check for cross-device rename */ fmp = vnode_mount(fvp); tmp = tvp ? vnode_mount(tvp) : NULL; @@ -3301,11 +4278,10 @@ nfs_vnop_rename( /* 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)); } - nfs_dulookup_start(&fdul, fdnp, ctx); - nfs_dulookup_start(&tdul, tdnp, ctx); - error = nmp->nm_funcs->nf_rename_rpc(fdnp, fcnp->cn_nameptr, fcnp->cn_namelen, tdnp, tcnp->cn_nameptr, tcnp->cn_namelen, ctx); @@ -3316,8 +4292,10 @@ nfs_vnop_rename( error = 0; if (tvp && (tvp != fvp) && !tnp->n_sillyrename) { + nfs_node_lock_force(tnp); tvprecycle = (!error && !vnode_isinuse(tvp, 0) && - (nfs_getattrcache(tnp, &nvattr, 1) || (nvattr.nva_nlink == 1))); + (nfs_getattrcache(tnp, &nvattr, 0) || (nvattr.nva_nlink == 1))); + nfs_node_unlock(tnp); lck_mtx_lock(nfs_node_hash_mutex); if (tvprecycle && (tnp->n_hflag & NHHASHED)) { /* @@ -3333,21 +4311,26 @@ nfs_vnop_rename( } /* 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); + 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 */ + nfs_node_lock_force(tnp); tnp->n_flag &= (NMODIFIED); + nfs_node_unlock(tnp); vnode_recycle(tvp); } } 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)) { @@ -3364,24 +4347,12 @@ nfs_vnop_rename( fnp->n_parent = NULL; } } + nfs_node_unlock(fnp); } out: - if (!nfs_getattr(fdnp, &nvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, fdnp, &nvattr)) { - fdnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(fdvp); - NFS_CHANGED_UPDATE_NC(nfsvers, fdnp, &nvattr); - } - } - if (!nfs_getattr(tdnp, &nvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, tdnp, &nvattr)) { - tdnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(tdvp); - NFS_CHANGED_UPDATE_NC(nfsvers, tdnp, &nvattr); - } - } - nfs_dulookup_finish(&fdul, fdnp, ctx); - nfs_dulookup_finish(&tdul, tdnp, ctx); + /* 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); @@ -3392,7 +4363,7 @@ out: } lck_mtx_unlock(nfs_node_hash_mutex); } - nfs_unlock4(fdnp, fnp, tdnp, tnp); + nfs_node_clear_busy4(fdnp, fnp, tdnp, tnp); return (error); } @@ -3409,7 +4380,7 @@ nfs3_rename_rpc( int tnamelen, vfs_context_t ctx) { - int error = 0, status, fwccpostattr = 0, twccpostattr = 0; + int error = 0, lockerror = ENOENT, status, fwccpostattr = 0, twccpostattr = 0; struct timespec fpremtime = { 0, 0 }, tpremtime = { 0, 0 }; struct nfsmount *nmp; int nfsvers; @@ -3417,7 +4388,7 @@ nfs3_rename_rpc( struct nfsm_chain nmreq, nmrep; nmp = NFSTONMP(fdnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; if ((nfsvers == NFS_VER2) && @@ -3431,14 +4402,16 @@ nfs3_rename_rpc( (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_string(error, &nmreq, fnameptr, fnamelen); + 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_string(error, &nmreq, tnameptr, tnamelen); + 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, &nmrep, &xid, &status); + 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); @@ -3449,25 +4422,28 @@ nfs3_rename_rpc( nfsmout: nfsm_chain_cleanup(&nmreq); nfsm_chain_cleanup(&nmrep); - 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); + 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 */ -static int +int nfs3_vnop_link( struct vnop_link_args /* { struct vnodeop_desc *a_desc; @@ -3481,7 +4457,7 @@ nfs3_vnop_link( vnode_t vp = ap->a_vp; vnode_t tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; - int error = 0, status, wccpostattr = 0, attrflag = 0; + int error = 0, lockerror = ENOENT, status, wccpostattr = 0, attrflag = 0; struct timespec premtime = { 0, 0 }; struct nfsmount *nmp; nfsnode_t np = VTONFS(vp); @@ -3494,7 +4470,7 @@ nfs3_vnop_link( return (EXDEV); nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN)) @@ -3507,7 +4483,7 @@ nfs3_vnop_link( */ nfs_flush(np, MNT_WAIT, vfs_context_thread(ctx), V_IGNORE_WRITEERR); - error = nfs_lock2(tdnp, np, NFS_NODE_LOCK_EXCLUSIVE); + error = nfs_node_set_busy2(tdnp, np, vfs_context_thread(ctx)); if (error) return (error); @@ -3518,11 +4494,15 @@ nfs3_vnop_link( 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_string(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen); + 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, - &nmrep, &xid, &status); + 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); @@ -3533,19 +4513,22 @@ nfs3_vnop_link( nfsmout: nfsm_chain_cleanup(&nmreq); nfsm_chain_cleanup(&nmrep); - tdnp->n_flag |= NMODIFIED; - if (!attrflag) - NATTRINVALIDATE(np); - /* 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); + 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_unlock2(tdnp, np); + nfs_node_clear_busy2(tdnp, np); /* * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. */ @@ -3557,7 +4540,7 @@ nfsmout: /* * NFS symbolic link create call */ -static int +int nfs3_vnop_symlink( struct vnop_symlink_args /* { struct vnodeop_desc *a_desc; @@ -3573,22 +4556,22 @@ nfs3_vnop_symlink( vnode_t dvp = ap->a_dvp; struct vnode_attr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; - struct nfs_vattr nvattr, dnvattr; + struct nfs_vattr nvattr; fhandle_t fh; - int slen, error = 0, lockerror = ENOENT, status, wccpostattr = 0; + int slen, error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0; struct timespec premtime = { 0, 0 }; vnode_t newvp = NULL; int nfsvers, gotuid, gotgid; - u_int64_t xid, dxid; + 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 *req = NULL; + struct nfsreq rq, *req = &rq; struct nfs_dulookup dul; nmp = VTONMP(dvp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; @@ -3597,6 +4580,8 @@ nfs3_vnop_symlink( ((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); VATTR_SET_SUPPORTED(vap, va_gid); @@ -3606,7 +4591,8 @@ nfs3_vnop_symlink( gotuid = VATTR_IS_ACTIVE(vap, va_uid); gotgid = VATTR_IS_ACTIVE(vap, va_gid); - nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen); + 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); @@ -3615,25 +4601,24 @@ nfs3_vnop_symlink( 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_string(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen); + 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_string(error, &nmreq, ap->a_target, slen); + 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); - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; - nfsmout_if(error); error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_SYMLINK, - vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, &req); + 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) { @@ -3658,19 +4643,13 @@ nfsmout: /* if directory hadn't changed, update namecache mtime */ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr); - if (!wccpostattr) - NATTRINVALIDATE(dnp); - if (!nfs_getattr(dnp, &dnvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, dnp, &dnvattr)) { - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(dvp); - NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnvattr); - } - } + 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, NG_MAKEENTRY, &np); + 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); @@ -3682,7 +4661,7 @@ nfsmout: */ if ((error == EEXIST) || (!error && !newvp)) { if (newvp) { - nfs_unlock(np); + nfs_node_unlock(np); vnode_put(newvp); newvp = NULL; } @@ -3693,10 +4672,10 @@ nfsmout: error = EEXIST; } } - if (!lockerror) - nfs_unlock(dnp); + if (!busyerror) + nfs_node_clear_busy(dnp); if (!error && (gotuid || gotgid) && - (!newvp || nfs_getattrcache(np, &nvattr, 1) || + (!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) */ @@ -3705,11 +4684,11 @@ nfsmout: } if (error) { if (newvp) { - nfs_unlock(np); + nfs_node_unlock(np); vnode_put(newvp); } } else { - nfs_unlock(np); + nfs_node_unlock(np); *ap->a_vpp = newvp; } return (error); @@ -3718,7 +4697,7 @@ nfsmout: /* * NFS make dir call */ -static int +int nfs3_vnop_mkdir( struct vnop_mkdir_args /* { struct vnodeop_desc *a_desc; @@ -3733,27 +4712,29 @@ nfs3_vnop_mkdir( vnode_t dvp = ap->a_dvp; struct vnode_attr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; - struct nfs_vattr nvattr, dnvattr; + struct nfs_vattr nvattr; nfsnode_t np = NULL; struct nfsmount *nmp; nfsnode_t dnp = VTONFS(dvp); vnode_t newvp = NULL; - int error = 0, lockerror = ENOENT, status, wccpostattr = 0; + int error = 0, lockerror = ENOENT, busyerror = ENOENT, status, wccpostattr = 0; struct timespec premtime = { 0, 0 }; int nfsvers, gotuid, gotgid; - u_int64_t xid, dxid; + u_int64_t xid= 0, dxid; fhandle_t fh; struct nfsm_chain nmreq, nmrep; - struct nfsreq *req = NULL; + struct nfsreq rq, *req = &rq; struct nfs_dulookup dul; nmp = VTONMP(dvp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; 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); VATTR_SET_SUPPORTED(vap, va_gid); @@ -3763,7 +4744,8 @@ nfs3_vnop_mkdir( gotuid = VATTR_IS_ACTIVE(vap, va_uid); gotgid = VATTR_IS_ACTIVE(vap, va_gid); - nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen); + 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); @@ -3772,24 +4754,23 @@ nfs3_vnop_mkdir( 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_string(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen); + 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); - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; - nfsmout_if(error); error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_MKDIR, - vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, &req); + 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) { @@ -3811,19 +4792,13 @@ nfsmout: /* if directory hadn't changed, update namecache mtime */ if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr); - if (!wccpostattr) - NATTRINVALIDATE(dnp); - if (!nfs_getattr(dnp, &dnvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, dnp, &dnvattr)) { - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(dvp); - NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnvattr); - } - } + 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, NG_MAKEENTRY, &np); + 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); @@ -3833,9 +4808,9 @@ nfsmout: * 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 && !newvp)) { + if ((error == EEXIST) || (!error && !newvp)) { if (newvp) { - nfs_unlock(np); + nfs_node_unlock(np); vnode_put(newvp); newvp = NULL; } @@ -3846,10 +4821,10 @@ nfsmout: error = EEXIST; } } - if (!lockerror) - nfs_unlock(dnp); + if (!busyerror) + nfs_node_clear_busy(dnp); if (!error && (gotuid || gotgid) && - (!newvp || nfs_getattrcache(np, &nvattr, 1) || + (!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) */ @@ -3858,11 +4833,11 @@ nfsmout: } if (error) { if (newvp) { - nfs_unlock(np); + nfs_node_unlock(np); vnode_put(newvp); } } else { - nfs_unlock(np); + nfs_node_unlock(np); *ap->a_vpp = newvp; } return (error); @@ -3871,7 +4846,7 @@ nfsmout: /* * NFS remove directory call */ -static int +int nfs3_vnop_rmdir( struct vnop_rmdir_args /* { struct vnodeop_desc *a_desc; @@ -3885,47 +4860,48 @@ nfs3_vnop_rmdir( vnode_t vp = ap->a_vp; vnode_t dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; - int error = 0, status, wccpostattr = 0; + int error = 0, lockerror = ENOENT, status, wccpostattr = 0; struct timespec premtime = { 0, 0 }; struct nfsmount *nmp; nfsnode_t np = VTONFS(vp); nfsnode_t dnp = VTONFS(dvp); - struct nfs_vattr dnvattr; int nfsvers; u_int64_t xid; struct nfsm_chain nmreq, nmrep; - struct nfsreq *req = NULL; + struct nfsreq rq, *req = &rq; struct nfs_dulookup dul; nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN)) return (ENAMETOOLONG); - nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen); - - if ((error = nfs_lock2(dnp, np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_string(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen); + 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, &req); + 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) @@ -3934,22 +4910,18 @@ nfsmout: nfsm_chain_cleanup(&nmreq); nfsm_chain_cleanup(&nmrep); - 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); - if (!wccpostattr) - NATTRINVALIDATE(dnp); - cache_purge(vp); - if (!nfs_getattr(dnp, &dnvattr, ctx, 1)) { - if (NFS_CHANGED_NC(nfsvers, dnp, &dnvattr)) { - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge(dvp); - NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnvattr); - } + 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_unlock2(dnp, np); + nfs_node_clear_busy2(dnp, np); /* * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. @@ -3975,609 +4947,980 @@ nfsmout: /* * 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 +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) { vfs_context_t ctx = ap->a_context; - vnode_t vp = ap->a_vp; - nfsnode_t np = VTONFS(vp); + vnode_t dvp = ap->a_vp; + nfsnode_t dnp = VTONFS(dvp); struct nfsmount *nmp; - struct uio *uio = ap->a_uio; - int tresid, error, nfsvers; - struct nfs_vattr nvattr; - - if (vnode_vtype(vp) != VDIR) - return (EPERM); + 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(vp); - if (!nmp) + nmp = VTONMP(dvp); + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; + bigcookies = (nmp->nm_state & NFSSTA_BIGCOOKIES); + extended = (ap->a_flags & VNODE_READDIR_EXTENDED); - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) - return (error); + if (vnode_vtype(dvp) != VDIR) + return (EPERM); + + 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(np, &nvattr, ctx, 1)) { - if (!NFS_CHANGED(nfsvers, np, &nvattr)) { - nfs_unlock(np); - OSAddAtomic(1, (SInt32*)&nfsstats.direofcache_hits); + 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 ((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; - return (0); - } - if (NFS_CHANGED_NC(nfsvers, np, &nvattr)) { - /* directory changed, purge any name cache entries */ - np->n_flag &= ~NNEGNCENTRIES; - cache_purge(vp); } } + 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); } - nfs_unlock(np); - if (ap->a_eofflag) - *ap->a_eofflag = 0; - /* - * Call nfs_bioread() to do the real work. - */ - // LP64todo - fix this - tresid = uio_uio_resid(uio); - error = nfs_bioread(np, uio, 0, ap->a_eofflag, ctx); + 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); } + /* - * Readdir RPC call. - * Called from below the buffer cache by nfs_buf_readdir(). + * Invalidate cached directory information, except for the actual directory + * blocks (which are invalidated separately). */ -#define DIRHDSIZ ((int)(sizeof(struct dirent) - (MAXNAMLEN + 1))) -int -nfs3_readdir_rpc(nfsnode_t dnp, struct uio *uiop, vfs_context_t ctx) +void +nfs_invaldir(nfsnode_t dnp) { - int len, skiplen, left; - struct dirent *dp = NULL; - nfsuint64 *cookiep; - nfsuint64 cookie; - struct nfsmount *nmp; - u_quad_t fileno; - int error = 0, lockerror, status, tlen, more_dirs = 1, blksiz = 0, bigenough = 1, eof; - int nfsvers, nmreaddirsize; - u_int64_t xid; - struct nfsm_chain nmreq, nmrep; - char *cp; + 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); +} -#if DIAGNOSTIC - /* XXX limitation based on need to adjust uio */ - if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || - (uio_uio_resid(uiop) & (DIRBLKSIZ - 1))) - panic("nfs_readdirrpc: bad uio"); -#endif - nmp = NFSTONMP(dnp); - if (!nmp) - return (ENXIO); - nfsvers = nmp->nm_vers; - nmreaddirsize = nmp->nm_readdirsize; +/* + * 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 ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_SHARED))) - return (lockerror); + if (!ndbhp) + return (0); + space = bp->nb_bufsize - ndbhp->ndbh_entry_end; + if (rdirplus) + space -= ndbhp->ndbh_count * sizeof(struct nfs_vattr); + return (space); +} - /* - * If there is no cookie, assume directory was stale. - */ - cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); - if (cookiep) - cookie = *cookiep; - else { - nfs_unlock(dnp); - return (NFSERR_BAD_COOKIE); +/* + * 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); } /* - * Loop around doing readdir rpc's of size nm_readdirsize - * truncated to a multiple of DIRBLKSIZ. - * The stopping criteria is EOF or buffer full. + * Search the list for this cookie. + * Keep track of previous and last entries. */ - nfsm_chain_null(&nmreq); - nfsm_chain_null(&nmrep); - while (more_dirs && bigenough) { - nfsm_chain_build_alloc_init(error, &nmreq, - NFSX_FH(nfsvers) + NFSX_READDIR(nfsvers)); - 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_32(error, &nmreq, cookie.nfsuquad[0]); - nfsm_chain_add_32(error, &nmreq, cookie.nfsuquad[1]); - nfsm_chain_add_32(error, &nmreq, dnp->n_cookieverf.nfsuquad[0]); - nfsm_chain_add_32(error, &nmreq, dnp->n_cookieverf.nfsuquad[1]); - } else { - nfsm_chain_add_32(error, &nmreq, cookie.nfsuquad[0]); - } - nfsm_chain_add_32(error, &nmreq, nmreaddirsize); - nfsm_chain_build_done(error, &nmreq); - nfs_unlock(dnp); - lockerror = ENOENT; - nfsmout_if(error); + 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); +} - error = nfs_request(dnp, NULL, &nmreq, NFSPROC_READDIR, ctx, - &nmrep, &xid, &status); +/* + * 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_dir_cookie_to_lbn(nfsnode_t dnp, uint64_t cookie, int *ptc, uint64_t *lbnp) +{ + struct nfsdmap *ndcc = dnp->n_cookiecache; + int8_t eofptc, found; + int i, iptc; + struct nfsmount *nmp; + struct nfsbuf *bp, *lastbp; + struct nfsbuflists blist; + struct direntry *dp, *dpptc; + struct nfs_dir_buf_header *ndbhp; + + if (!cookie) { /* initial cookie */ + *lbnp = 0; + *ptc = 0; + return (0); + } - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; + if (nfs_node_lock(dnp)) + return (ENOENT); - if (nfsvers == NFS_VER3) - nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid); - if (!error) - error = status; - if (nfsvers == NFS_VER3) { - nfsm_chain_get_32(error, &nmrep, dnp->n_cookieverf.nfsuquad[0]); - nfsm_chain_get_32(error, &nmrep, dnp->n_cookieverf.nfsuquad[1]); - } - nfsm_chain_get_32(error, &nmrep, more_dirs); + 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; - if (!lockerror) { - nfs_unlock(dnp); - lockerror = ENOENT; + /* 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); } - nfsmout_if(error); + /* 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); - /* loop thru the dir entries, doctoring them to 4bsd form */ - while (more_dirs && bigenough) { - if (nfsvers == NFS_VER3) - nfsm_chain_get_64(error, &nmrep, fileno); + /* + * No match found in the cookie cache... hmm... + * Let's search the directory's buffers for the cookie. + */ + nmp = NFSTONMP(dnp); + if (nfs_mount_gone(nmp)) + return (ENXIO); + dpptc = NULL; + found = 0; + + lck_mtx_lock(nfs_buf_mutex); + /* + * 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 (!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 - nfsm_chain_get_32(error, &nmrep, fileno); - nfsm_chain_get_32(error, &nmrep, len); - nfsmout_if(error); - /* Note: v3 supports longer names, but struct dirent doesn't */ - /* so we just truncate the names to fit */ - if (len <= 0) { - error = EBADRPC; - goto nfsmout; + 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; } - if (len > MAXNAMLEN) { - skiplen = len - MAXNAMLEN; - len = MAXNAMLEN; - } else { - skiplen = 0; + 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); } - tlen = nfsm_rndup(len); - if (tlen == len) - tlen += 4; /* To ensure null termination */ - left = DIRBLKSIZ - blksiz; - if ((tlen + 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 ((i == ndbhp->ndbh_count) && dpptc) { + /* found only a PTC match */ + dp = dpptc; + i = iptc; + } else if (i < ndbhp->ndbh_count) { + *ptc = 0; } - if ((tlen + 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); - error = nfsm_chain_get_uio(&nmrep, len, uiop); - nfsmout_if(error); - // 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); - if (skiplen) - nfsm_chain_adv(error, &nmrep, - nfsm_rndup(len + skiplen) - nfsm_rndup(len)); - } else { - nfsm_chain_adv(error, &nmrep, nfsm_rndup(len + skiplen)); + 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; } - if (bigenough) { - nfsm_chain_get_32(error, &nmrep, cookie.nfsuquad[0]); - if (nfsvers == NFS_VER3) - nfsm_chain_get_32(error, &nmrep, cookie.nfsuquad[1]); - } else if (nfsvers == NFS_VER3) - nfsm_chain_adv(error, &nmrep, 2 * NFSX_UNSIGNED); - else - nfsm_chain_adv(error, &nmrep, NFSX_UNSIGNED); - nfsm_chain_get_32(error, &nmrep, more_dirs); - nfsmout_if(error); - } - /* - * If at end of rpc data, get the eof boolean - */ - if (!more_dirs) { - nfsm_chain_get_32(error, &nmrep, eof); - if (!error) - more_dirs = (eof == 0); + nfs_buf_drop(bp); } - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_SHARED))) - error = lockerror; - nfsmout_if(error); - nfsm_chain_cleanup(&nmrep); - nfsm_chain_null(&nmreq); + nfs_buf_itercomplete(dnp, &blist, NBI_CLEAN); } - if (!lockerror) { - nfs_unlock(dnp); - lockerror = ENOENT; + lck_mtx_unlock(nfs_buf_mutex); + if (found) { + OSAddAtomic64(1, &nfsstats.direofcache_hits); + return (0); } - /* - * 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); + + /* 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 (flags == NDBS_PURGE) { + dp->d_fileno = 0; + bzero(nvattrp, sizeof(*nvattrp)); + error = ENOENT; + break; + } + 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; + } + /* 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; + } + dp = NFS_DIRENTRY_NEXT(dp); } + if (nextlbnp) + *nextlbnp = nextlbn; + return (error); +} - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; - nfsmout_if(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] == '_'); + + nmp = NFSTONMP(dnp); + if (nfs_mount_gone(nmp)) + 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; + } + lbn = nextlbn; + } + + 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); + } +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; + } } -nfsmout: - if (!lockerror) - nfs_unlock(dnp); - nfsm_chain_cleanup(&nmreq); - nfsm_chain_cleanup(&nmrep); 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 -nfs3_readdirplus_rpc(nfsnode_t dnp, struct uio *uiop, vfs_context_t ctx) +nfs3_readdir_rpc(nfsnode_t dnp, struct nfsbuf *bp, vfs_context_t ctx) { - size_t len, tlen, skiplen, left; - struct dirent *dp = NULL; - vnode_t newvp; - nfsuint64 *cookiep; - struct componentname cn, *cnp = &cn; - nfsuint64 cookie; struct nfsmount *nmp; - nfsnode_t np; - u_char *fhp; - u_quad_t fileno; - int error = 0, lockerror, status, more_dirs = 1, blksiz = 0, doit, bigenough = 1; - int nfsvers, nmreaddirsize, nmrsize, attrflag, eof; - size_t fhsize; - u_int64_t xid, savexid; - struct nfs_vattr nvattr; - struct nfsm_chain nmreq, nmrep; - char *cp; + 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; -#if DIAGNOSTIC - /* XXX limitation based on need to adjust uio */ - if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) || - (uio_uio_resid(uiop) & (DIRBLKSIZ - 1))) - panic("nfs3_readdirplus_rpc: bad uio"); -#endif nmp = NFSTONMP(dnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; nmreaddirsize = nmp->nm_readdirsize; nmrsize = nmp->nm_rsize; + bigcookies = nmp->nm_state & NFSSTA_BIGCOOKIES; +noplus: + rdirplus = ((nfsvers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) ? 1 : 0; - bzero(cnp, sizeof(*cnp)); - newvp = NULLVP; - - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_SHARED))) + if ((lockerror = nfs_node_lock(dnp))) return (lockerror); - /* - * If there is no cookie, assume directory was stale. - */ - cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); - if (cookiep) - cookie = *cookiep; - else { - nfs_unlock(dnp); - return (NFSERR_BAD_COOKIE); + /* 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; /* - * Loop around doing readdir rpc's of size nm_readdirsize - * truncated to a multiple of DIRBLKSIZ. - * The stopping criteria is EOF or buffer full. + * 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). */ nfsm_chain_null(&nmreq); nfsm_chain_null(&nmrep); - while (more_dirs && bigenough) { + while (nfs_dir_buf_freespace(bp, rdirplus) && !(ndbhp->ndbh_flags & NDB_FULL)) { nfsm_chain_build_alloc_init(error, &nmreq, - NFSX_FH(NFS_VER3) + 6 * NFSX_UNSIGNED); + NFSX_FH(nfsvers) + NFSX_READDIR(nfsvers) + NFSX_UNSIGNED); nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize); - /* opaque values don't need swapping, but as long */ - /* as we are consistent about it, it should be ok */ - nfsm_chain_add_32(error, &nmreq, cookie.nfsuquad[0]); - nfsm_chain_add_32(error, &nmreq, cookie.nfsuquad[1]); - nfsm_chain_add_32(error, &nmreq, dnp->n_cookieverf.nfsuquad[0]); - nfsm_chain_add_32(error, &nmreq, dnp->n_cookieverf.nfsuquad[1]); + 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); + } nfsm_chain_add_32(error, &nmreq, nmreaddirsize); - nfsm_chain_add_32(error, &nmreq, nmrsize); + if (rdirplus) + nfsm_chain_add_32(error, &nmreq, nmrsize); nfsm_chain_build_done(error, &nmreq); - nfs_unlock(dnp); + nfs_node_unlock(dnp); lockerror = ENOENT; nfsmout_if(error); - error = nfs_request(dnp, NULL, &nmreq, NFSPROC_READDIRPLUS, ctx, - &nmrep, &xid, &status); + error = nfs_request(dnp, NULL, &nmreq, + rdirplus ? NFSPROC_READDIRPLUS : NFSPROC_READDIR, + ctx, NULL, &nmrep, &xid, &status); - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) + if ((lockerror = nfs_node_lock(dnp))) error = lockerror; - savexid = xid; - nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid); + savedxid = xid; + if (nfsvers == NFS_VER3) + nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid); if (!error) error = status; - nfsm_chain_get_32(error, &nmrep, dnp->n_cookieverf.nfsuquad[0]); - nfsm_chain_get_32(error, &nmrep, dnp->n_cookieverf.nfsuquad[1]); - nfsm_chain_get_32(error, &nmrep, more_dirs); + if (nfsvers == NFS_VER3) + nfsm_chain_get_64(error, &nmrep, dnp->n_cookieverf); + nfsm_chain_get_32(error, &nmrep, more_entries); if (!lockerror) { - nfs_unlock(dnp); + nfs_node_unlock(dnp); lockerror = ENOENT; } - nfsmout_if(error); + 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); - /* loop thru the dir entries, doctoring them to 4bsd form */ - while (more_dirs && bigenough) { - nfsm_chain_get_64(error, &nmrep, fileno); - nfsm_chain_get_32(error, &nmrep, len); + 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); - /* Note: v3 supports longer names, but struct dirent doesn't */ - /* so we just truncate the names to fit */ - if (len <= 0) { + /* just truncate names that don't fit in direntry.d_name */ + if (namlen <= 0) { error = EBADRPC; 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 + 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 + 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; - error = nfsm_chain_get_uio(&nmrep, len, uiop); + /* 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); - 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); - if (skiplen) - nfsm_chain_adv(error, &nmrep, - nfsm_rndup(len + skiplen) - nfsm_rndup(len)); - } else { - nfsm_chain_adv(error, &nmrep, nfsm_rndup(len + skiplen)); + /* 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); } - if (bigenough) { - nfsm_chain_get_32(error, &nmrep, cookie.nfsuquad[0]); - nfsm_chain_get_32(error, &nmrep, cookie.nfsuquad[1]); - } else - nfsm_chain_adv(error, &nmrep, 2 * NFSX_UNSIGNED); - - nfsm_chain_get_32(error, &nmrep, attrflag); + 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_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); - if (attrflag) { - /* grab attributes */ - error = nfs_parsefattr(&nmrep, NFS_VER3, &nvattr); - nfsmout_if(error); - dp->d_type = IFTODT(VTTOIF(nvattr.nva_type)); - /* check for file handle */ - nfsm_chain_get_32(error, &nmrep, doit); - nfsmout_if(error); - if (doit) { - nfsm_chain_get_fh_ptr(error, &nmrep, NFS_VER3, fhp, fhsize); + 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 (NFS_CMPFH(dnp, fhp, fhsize)) { - error = vnode_ref(NFSTOV(dnp)); - if (error) { - doit = 0; - } else { - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; - if (error) { - vnode_rele(NFSTOV(dnp)); - goto nfsmout; + 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 { + /* mark the attributes invalid */ + bzero(nvattrp, sizeof(struct nfs_vattr)); + } + /* 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; } - newvp = NFSTOV(dnp); - 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; + /* 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 { - cnp->cn_hash = 0; - - error = nfs_nget(NFSTOMP(dnp), dnp, cnp, - fhp, fhsize, &nvattr, &xid, NG_MAKEENTRY, &np); - if (error) - doit = 0; - else - newvp = NFSTOV(np); + /* 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); } - } - /* 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_chain_get_32(error, &nmrep, fhsize); - nfsm_chain_adv(error, &nmrep, nfsm_rndup(fhsize)); + *(time_t*)(&dp->d_name[dp->d_namlen+1+fhlen]) = now.tv_sec; + dp->d_reclen = reclen; } - if (newvp != NULLVP) { - nfs_unlock(np); - if (newvp == NFSTOV(dnp)) - vnode_rele(newvp); - else - vnode_put(newvp); - newvp = NULLVP; - } - nfsm_chain_get_32(error, &nmrep, more_dirs); + 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_chain_get_32(error, &nmrep, eof); - if (!error) - more_dirs = (eof == 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; } - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_SHARED))) + 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); } - if (!lockerror) { - nfs_unlock(dnp); - lockerror = ENOENT; - } - /* - * 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); - } - - if ((lockerror = nfs_lock(dnp, NFS_NODE_LOCK_EXCLUSIVE))) - error = lockerror; - nfsmout_if(error); - - /* - * 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! readdirplus_rpc resid > 0\n"); - cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); - if (cookiep) - *cookiep = cookie; - } - nfsmout: + if (bp_dropped && bp) + nfs_buf_release(bp, 0); if (!lockerror) - nfs_unlock(dnp); + nfs_node_unlock(dnp); nfsm_chain_cleanup(&nmreq); nfsm_chain_cleanup(&nmrep); - return (error); + return (bp_dropped ? NFSERR_DIRBUFDROPPED : error); } /* @@ -4595,7 +5938,7 @@ nfsmout: /* starting from zero isn't silly enough */ static uint32_t nfs_sillyrename_number = 0x20051025; -static int +int nfs_sillyrename( nfsnode_t dnp, nfsnode_t np, @@ -4610,10 +5953,10 @@ nfs_sillyrename( struct nfsmount *nmp; nmp = NFSTONMP(dnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); - cache_purge(NFSTOV(np)); + nfs_name_cache_purge(dnp, np, cnp, ctx); MALLOC_ZONE(nsp, struct nfs_sillyrename *, sizeof (struct nfs_sillyrename), M_NFSREQ, M_WAITOK); @@ -4629,7 +5972,7 @@ nfs_sillyrename( /* Fudge together a funny name */ pid = vfs_context_pid(ctx); - num = OSAddAtomic(1, (SInt32*)&nfs_sillyrename_number); + 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)) @@ -4637,7 +5980,7 @@ nfs_sillyrename( /* Try lookitups until we get one that isn't there */ while (nfs_lookitup(dnp, nsp->nsr_name, nsp->nsr_namlen, ctx, NULL) == 0) { - num = OSAddAtomic(1, (SInt32*)&nfs_sillyrename_number); + 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)) @@ -4647,15 +5990,25 @@ nfs_sillyrename( /* 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); - if (!error && (dnp->n_flag & NNEGNCENTRIES)) { - dnp->n_flag &= ~NNEGNCENTRIES; - cache_purge_negatives(NFSTOV(dnp)); + + /* 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); } FSDBG(267, dnp, np, num, error); if (error) goto bad; 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(NFSTOV(dnp)); @@ -4679,7 +6032,7 @@ nfs3_lookup_rpc_async( int error = 0, nfsvers; nmp = NFSTONMP(dnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; @@ -4688,11 +6041,11 @@ nfs3_lookup_rpc_async( 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_string(error, &nmreq, name, namelen); + 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, reqp); + vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, reqp); nfsmout: nfsm_chain_cleanup(&nmreq); return (error); @@ -4701,24 +6054,30 @@ nfsmout: int 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) { - int error = 0, status, nfsvers, attrflag; + int error = 0, lockerror = ENOENT, status, nfsvers, attrflag; u_int64_t xid; struct nfsmount *nmp; struct nfsm_chain nmrep; nmp = NFSTONMP(dnp); + if (nmp == NULL) + return (ENXIO); nfsvers = nmp->nm_vers; nfsm_chain_null(&nmrep); error = nfs_request_async_finish(req, &nmrep, xidp, &status); + if ((lockerror = nfs_node_lock(dnp))) + error = lockerror; xid = *xidp; if (error || status) { if (nfsvers == NFS_VER3) @@ -4738,11 +6097,13 @@ nfs3_lookup_rpc_async_finish( 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, ctx, nvap, xidp); + error = nfs3_getattr_rpc(NULL, NFSTOMP(dnp), fhp->fh_data, fhp->fh_len, 0, ctx, nvap, xidp); } else { error = nfs_parsefattr(&nmrep, nfsvers, nvap); } nfsmout: + if (!lockerror) + nfs_node_unlock(dnp); nfsm_chain_cleanup(&nmrep); return (error); } @@ -4772,13 +6133,15 @@ nfs_lookitup( struct nfsreq rq, *req = &rq; nmp = NFSTONMP(dnp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME) && - (namelen > (long)nmp->nm_fsattr.nfsa_maxname)) + (namelen > (int)nmp->nm_fsattr.nfsa_maxname)) return (ENAMETOOLONG); + NVATTR_INIT(&nvattr); + /* check for lookup of "." */ if ((name[0] == '.') && (namelen == 1)) { /* skip lookup, we know who we are */ @@ -4789,7 +6152,7 @@ nfs_lookitup( 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, ctx, req, &xid, &fh, &nvattr); + error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, name, namelen, ctx, req, &xid, &fh, &nvattr); nfsmout_if(!npp || error); if (*npp) { @@ -4811,12 +6174,16 @@ nfs_lookitup( } 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 { @@ -4825,7 +6192,7 @@ nfs_lookitup( cnp->cn_nameptr = name; cnp->cn_namelen = namelen; error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh.fh_data, fh.fh_len, - &nvattr, &xid, NG_MAKEENTRY, &np); + &nvattr, &xid, rq.r_auth, NG_MAKEENTRY, &np); nfsmout_if(error); newnp = np; } @@ -4833,6 +6200,7 @@ nfs_lookitup( nfsmout: if (npp && !*npp && !error) *npp = newnp; + NVATTR_CLEANUP(&nvattr); return (error); } @@ -4841,15 +6209,18 @@ nfsmout: * performing async lookups. */ void -nfs_dulookup_init(struct nfs_dulookup *dulp, nfsnode_t dnp, const char *name, int namelen) +nfs_dulookup_init(struct nfs_dulookup *dulp, nfsnode_t dnp, const char *name, int namelen, vfs_context_t ctx) { 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)) @@ -4861,14 +6232,27 @@ nfs_dulookup_init(struct nfs_dulookup *dulp, nfsnode_t dnp, const char *name, in 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) + if (error == -1) { vnode_put(du_vp); - else if (!error) - dulp->du_flags |= NFS_DULOOKUP_DOIT; - else if (dulp->du_cn.cn_nameptr != dulp->du_smallname) - FREE(dulp->du_cn.cn_nameptr, M_TEMP); + } 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; + } + } + if (!error) + dulp->du_flags |= NFS_DULOOKUP_DOIT; + } } /* @@ -4880,7 +6264,7 @@ 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)) + 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)) @@ -4903,20 +6287,25 @@ nfs_dulookup_finish(struct nfs_dulookup *dulp, nfsnode_t dnp, vfs_context_t ctx) if (!nmp || !(dulp->du_flags & NFS_DULOOKUP_INPROG)) goto out; - error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, ctx, &dulp->du_req, &xid, &fh, &nvattr); + 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, NG_MAKEENTRY, &du_np); + &nvattr, &xid, dulp->du_req.r_auth, NG_MAKEENTRY, &du_np); if (!error) { - nfs_unlock(du_np); + 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); @@ -4931,20 +6320,21 @@ out: int nfs3_commit_rpc( nfsnode_t np, - u_int64_t offset, - u_int64_t count, - kauth_cred_t cred) + 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, wverf; + u_int64_t xid, newwverf; uint32_t count32; struct nfsm_chain nmreq, nmrep; nmp = NFSTONMP(np); FSDBG(521, np, offset, count, nmp ? nmp->nm_state : 0); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); if (!(nmp->nm_state & NFSSTA_HASWRITEVERF)) return (0); @@ -4965,22 +6355,22 @@ nfs3_commit_rpc( nfsm_chain_build_done(error, &nmreq); nfsmout_if(error); error = nfs_request2(np, NULL, &nmreq, NFSPROC_COMMIT, - current_thread(), cred, 0, &nmrep, &xid, &status); - if ((lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_unlock(np); + nfs_node_unlock(np); if (!error) error = status; - nfsm_chain_get_64(error, &nmrep, wverf); + nfsm_chain_get_64(error, &nmrep, newwverf); nfsmout_if(error); lck_mtx_lock(&nmp->nm_lock); - if (nmp->nm_verf != wverf) { - nmp->nm_verf = wverf; + 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); @@ -4989,7 +6379,7 @@ nfsmout: } -static int +int nfs_vnop_blockmap( __unused struct vnop_blockmap_args /* { struct vnodeop_desc *a_desc; @@ -5005,29 +6395,12 @@ nfs_vnop_blockmap( return (ENOTSUP); } -/* - * Mmap a file - * - * NB Currently unsupported. - */ -/*ARGSUSED*/ -static int -nfs_vnop_mmap( - __unused struct vnop_mmap_args /* { - struct vnodeop_desc *a_desc; - vnode_t a_vp; - int a_fflags; - vfs_context_t a_context; - } */ *ap) -{ - return (EINVAL); -} /* * fsync vnode op. Just call nfs_flush(). */ /* ARGSUSED */ -static int +int nfs_vnop_fsync( struct vnop_fsync_args /* { struct vnodeop_desc *a_desc; @@ -5055,7 +6428,7 @@ nfs3_pathconf_rpc( struct nfsmount *nmp = NFSTONMP(np); uint32_t val = 0; - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); nfsvers = nmp->nm_vers; @@ -5067,17 +6440,17 @@ nfs3_pathconf_rpc( 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, - &nmrep, &xid, &status); - if ((lockerror = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_unlock(np); + 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); + nfsap->nfsa_flags &= ~(NFS_FSFLAG_NO_TRUNC|NFS_FSFLAG_CHOWN_RESTRICTED|NFS_FSFLAG_CASE_INSENSITIVE|NFS_FSFLAG_CASE_PRESERVING); nfsm_chain_get_32(error, &nmrep, val); if (val) nfsap->nfsa_flags |= NFS_FSFLAG_NO_TRUNC; @@ -5108,6 +6481,7 @@ 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 &= ~(NFS_FSFLAG_NO_TRUNC|NFS_FSFLAG_CHOWN_RESTRICTED|NFS_FSFLAG_CASE_INSENSITIVE|NFS_FSFLAG_CASE_PRESERVING); 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; @@ -5128,13 +6502,13 @@ nfs3_pathconf_cache(struct nfsmount *nmp, struct nfs_fsattr *nfsap) * for V2. */ /* ARGSUSED */ -static int +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) { @@ -5147,7 +6521,7 @@ nfs_vnop_pathconf( uint nbits; nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); switch (ap->a_name) { @@ -5164,6 +6538,12 @@ nfs_vnop_pathconf( 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); @@ -5182,7 +6562,7 @@ nfs_vnop_pathconf( if (error) return (error); nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); lck_mtx_lock(&nmp->nm_lock); if (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS) { @@ -5202,7 +6582,7 @@ nfs_vnop_pathconf( if (error) return (error); nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); lck_mtx_lock(&nmp->nm_lock); nfsap = &nfsa; @@ -5227,35 +6607,36 @@ nfs_vnop_pathconf( break; case _PC_CHOWN_RESTRICTED: if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CHOWN_RESTRICTED)) - *ap->a_retval = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_CHOWN_RESTRICTED) ? 200112 /* _POSIX_CHOWN_RESTRICTED */ : 0; + *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CHOWN_RESTRICTED) ? 200112 /* _POSIX_CHOWN_RESTRICTED */ : 0; else error = EINVAL; break; case _PC_NO_TRUNC: if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_NO_TRUNC)) - *ap->a_retval = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NO_TRUNC) ? 200112 /* _POSIX_NO_TRUNC */ : 0; + *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_NO_TRUNC) ? 200112 /* _POSIX_NO_TRUNC */ : 0; else error = EINVAL; break; case _PC_CASE_SENSITIVE: if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_INSENSITIVE)) - *ap->a_retval = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_CASE_INSENSITIVE) ? 0 : 1; + *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CASE_INSENSITIVE) ? 0 : 1; else error = EINVAL; break; case _PC_CASE_PRESERVING: if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_PRESERVING)) - *ap->a_retval = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_CASE_PRESERVING) ? 1 : 0; + *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CASE_PRESERVING) ? 1 : 0; else error = EINVAL; break; + case _PC_XATTR_SIZE_BITS: /* same as file size bits if named attrs supported */ case _PC_FILESIZEBITS: - if (!NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXFILESIZE)) { + if (!NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_MAXFILESIZE)) { *ap->a_retval = 64; error = 0; break; } - maxFileSize = nmp->nm_fsattr.nfsa_maxfilesize; + maxFileSize = nfsap->nfsa_maxfilesize; nbits = 1; if (maxFileSize & 0xffffffff00000000ULL) { nbits += 32; @@ -5294,7 +6675,7 @@ nfs_vnop_pathconf( /* * Read wrapper for special devices. */ -static int +int nfsspec_vnop_read( struct vnop_read_args /* { struct vnodeop_desc *a_desc; @@ -5311,20 +6692,20 @@ nfsspec_vnop_read( /* * Set access flag. */ - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_unlock(np); + nfs_node_unlock(np); return (VOCALL(spec_vnodeop_p, VOFFSET(vnop_read), ap)); } /* * Write wrapper for special devices. */ -static int +int nfsspec_vnop_write( struct vnop_write_args /* { struct vnodeop_desc *a_desc; @@ -5341,13 +6722,13 @@ nfsspec_vnop_write( /* * Set update flag. */ - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_unlock(np); + nfs_node_unlock(np); return (VOCALL(spec_vnodeop_p, VOFFSET(vnop_write), ap)); } @@ -5356,7 +6737,7 @@ nfsspec_vnop_write( * * Update the times on the nfsnode then do device close. */ -static int +int nfsspec_vnop_close( struct vnop_close_args /* { struct vnodeop_desc *a_desc; @@ -5371,11 +6752,11 @@ nfsspec_vnop_close( mount_t mp; int error; - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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; @@ -5385,13 +6766,13 @@ nfsspec_vnop_close( vattr.va_modify_time = np->n_mtim; VATTR_SET_ACTIVE(&vattr, va_modify_time); } - nfs_unlock(np); + nfs_node_unlock(np); vnode_setattr(vp, &vattr, ap->a_context); } else { - nfs_unlock(np); + nfs_node_unlock(np); } } else { - nfs_unlock(np); + nfs_node_unlock(np); } return (VOCALL(spec_vnodeop_p, VOFFSET(vnop_close), ap)); } @@ -5402,7 +6783,7 @@ extern vnop_t **fifo_vnodeop_p; /* * Read wrapper for fifos. */ -static int +int nfsfifo_vnop_read( struct vnop_read_args /* { struct vnodeop_desc *a_desc; @@ -5419,20 +6800,20 @@ nfsfifo_vnop_read( /* * Set access flag. */ - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_unlock(np); + nfs_node_unlock(np); return (VOCALL(fifo_vnodeop_p, VOFFSET(vnop_read), ap)); } /* * Write wrapper for fifos. */ -static int +int nfsfifo_vnop_write( struct vnop_write_args /* { struct vnodeop_desc *a_desc; @@ -5449,13 +6830,13 @@ nfsfifo_vnop_write( /* * Set update flag. */ - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + 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_unlock(np); + nfs_node_unlock(np); return (VOCALL(fifo_vnodeop_p, VOFFSET(vnop_write), ap)); } @@ -5464,7 +6845,7 @@ nfsfifo_vnop_write( * * Update the times on the nfsnode then do fifo close. */ -static int +int nfsfifo_vnop_close( struct vnop_close_args /* { struct vnodeop_desc *a_desc; @@ -5480,7 +6861,7 @@ nfsfifo_vnop_close( mount_t mp; int error; - if ((error = nfs_lock(np, NFS_NODE_LOCK_EXCLUSIVE))) + if ((error = nfs_node_lock(np))) return (error); if (np->n_flag & (NACC | NUPD)) { microtime(&now); @@ -5503,40 +6884,127 @@ nfsfifo_vnop_close( vattr.va_modify_time = np->n_mtim; VATTR_SET_ACTIVE(&vattr, va_modify_time); } - nfs_unlock(np); + nfs_node_unlock(np); vnode_setattr(vp, &vattr, ap->a_context); } else { - nfs_unlock(np); + nfs_node_unlock(np); } } else { - nfs_unlock(np); + nfs_node_unlock(np); } return (VOCALL(fifo_vnodeop_p, VOFFSET(vnop_close), ap)); } #endif /* FIFO */ /*ARGSUSED*/ -static int +int nfs_vnop_ioctl( - __unused struct vnop_ioctl_args /* { + 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; vfs_context_t a_context; } */ *ap) { + vfs_context_t ctx = ap->a_context; + vnode_t vp = ap->a_vp; + struct nfsmount *mp = VTONMP(vp); + struct user_nfs_gss_principal gprinc; + uint32_t len; + 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: + if (!auth_is_kerberized(mp->nm_auth)) + return (ENOTSUP); + error = nfs_gss_clnt_ctx_remove(mp, vfs_context_ucred(ctx)); + break; + case NFS_FSCTL_SET_CRED: + if (!auth_is_kerberized(mp->nm_auth)) + return (ENOTSUP); + NFS_DBG(NFS_FAC_GSS, 7, "Enter NFS_FSCTL_SET_CRED (proc %d) data = %p\n", vfs_context_is64bit(ctx), (void *)ap->a_data); + if (vfs_context_is64bit(ctx)) { + gprinc = *(struct user_nfs_gss_principal *)ap->a_data; + } else { + struct nfs_gss_principal *tp; + tp = (struct nfs_gss_principal *)ap->a_data; + gprinc.princlen = tp->princlen; + gprinc.nametype = tp->nametype; + gprinc.principal = CAST_USER_ADDR_T(tp->principal); + } + if (gprinc.princlen > MAXPATHLEN) + return (EINVAL); + NFS_DBG(NFS_FAC_GSS, 7, "Received principal length %d name type = %d\n", gprinc.princlen, gprinc.nametype); + uint8_t *p; + MALLOC(p, uint8_t *, gprinc.princlen+1, M_TEMP, M_WAITOK|M_ZERO); + if (p == NULL) + return (ENOMEM); + error = copyin(gprinc.principal, p, gprinc.princlen); + if (error) { + NFS_DBG(NFS_FAC_GSS, 7, "NFS_FSCTL_SET_CRED could not copy in princiapl data of len %d: %d\n", + gprinc.princlen, error); + FREE(p, M_TEMP); + return (error); + } + NFS_DBG(NFS_FAC_GSS, 7, "Seting credential to principal %s\n", p); + error = nfs_gss_clnt_ctx_set_principal(mp, ctx, p, gprinc.princlen, gprinc.nametype); + NFS_DBG(NFS_FAC_GSS, 7, "Seting credential to principal %s returned %d\n", p, error); + FREE(p, M_TEMP); + break; + case NFS_FSCTL_GET_CRED: + if (!auth_is_kerberized(mp->nm_auth)) + return (ENOTSUP); + error = nfs_gss_clnt_ctx_get_principal(mp, ctx, &gprinc); + if (error) + break; + if (vfs_context_is64bit(ctx)) { + struct user_nfs_gss_principal *upp = (struct user_nfs_gss_principal *)ap->a_data; + len = upp->princlen; + if (gprinc.princlen < len) + len = gprinc.princlen; + upp->princlen = gprinc.princlen; + upp->nametype = gprinc.nametype; + upp->flags = gprinc.flags; + if (gprinc.principal) + error = copyout((void *)gprinc.principal, upp->principal, len); + else + upp->principal = USER_ADDR_NULL; + } else { + struct nfs_gss_principal *u32pp = (struct nfs_gss_principal *)ap->a_data; + len = u32pp->princlen; + if (gprinc.princlen < len) + len = gprinc.princlen; + u32pp->princlen = gprinc.princlen; + u32pp->nametype = gprinc.nametype; + u32pp->flags = gprinc.flags; + if (gprinc.principal) + error = copyout((void *)gprinc.principal, u32pp->principal, len); + else + u32pp->principal = (user32_addr_t)0; + } + if (error) { + NFS_DBG(NFS_FAC_GSS, 7, "NFS_FSCTL_GET_CRED could not copy out princiapl data of len %d: %d\n", + gprinc.princlen, error); + } + FREE(gprinc.principal, M_TEMP); + } + + return (error); } /*ARGSUSED*/ -static int +int nfs_vnop_select( __unused struct vnop_select_args /* { struct vnodeop_desc *a_desc; @@ -5559,7 +7027,7 @@ nfs_vnop_select( * * No buffer I/O, just RPCs straight into the mapped pages. */ -static int +int nfs_vnop_pagein( struct vnop_pagein_args /* { struct vnodeop_desc *a_desc; @@ -5585,15 +7053,16 @@ nfs_vnop_pagein( 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, np, f_offset, size, flags); if (pl == (upl_t)NULL) @@ -5602,7 +7071,7 @@ nfs_vnop_pagein( if (size <= 0) { 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)) { @@ -5617,17 +7086,11 @@ nfs_vnop_pagein( if (!IS_VALID_CRED(cred)) cred = vfs_context_ucred(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 = vfs_context_proc(ap->a_context); + uio = uio_createwithbuffer(1, f_offset, UIO_SYSSPACE, UIO_READ, + &uio_buf, sizeof(uio_buf)); nmp = VTONMP(vp); - if (!nmp) { + if (nfs_mount_gone(nmp)) { if (!nofreeupl) ubc_upl_abort_range(pl, pl_offset, size, UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY); @@ -5636,14 +7099,25 @@ nfs_vnop_pagein( 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; + +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 { + 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); @@ -5658,49 +7132,70 @@ nfs_vnop_pagein( /* wait while we need to and break out if more requests to send */ while ((rxsize > 0) && req[nextwait]) { iosize = retsize = MIN(nmrsize, rxsize); - 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, rxsize); -#ifdef UPL_DEBUG - upl_ubc_alias_set(pl, current_thread(), 2); + 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); + 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->uio_offset, uio_uio_resid(uio), error, -1); + FSDBG(322, uio_offset(uio), uio_resid(uio), error, -1); break; } if (retsize < iosize) { /* Just zero fill the rest of the valid area. */ - // LP64todo - fix this int zcnt = iosize - retsize; - bzero((char *)ioaddr + retsize, zcnt); - FSDBG(324, uio->uio_offset, retsize, zcnt, ioaddr); - uio->uio_offset += zcnt; + bzero((char *)rxaddr + retsize, zcnt); + FSDBG(324, uio_offset(uio), retsize, zcnt, rxaddr); + uio_update(uio, zcnt); } - ioaddr += iosize; + rxaddr += iosize; rxsize -= iosize; if (txsize) break; } } while (!error && (txsize || rxsize)); - ubc_upl_unmap(pl); + restart = 0; 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"); + } + } } + ubc_upl_unmap(pl); + if (!nofreeupl) { if (error) ubc_upl_abort_range(pl, pl_offset, size, @@ -5722,7 +7217,8 @@ nfs_vnop_pagein( * are expected to match the same numbers here. If not, our actions maybe * erroneous. */ -enum actiontype {NOACTION, DUMP, DUMPANDLOG, RETRY, RETRYWITHSLEEP, SEVER}; +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] = { @@ -5826,7 +7322,7 @@ static const char errortooutcome[NFS_ELAST+1] = { DUMPANDLOG, /* EBADMACHO 88 Malformed Macho file */ }; -static char +char nfs_pageouterrorhandler(int error) { if (error > NFS_ELAST) @@ -5842,7 +7338,7 @@ nfs_pageouterrorhandler(int error) * No buffer I/O, just RPCs straight from the mapped pages. * File size changes are not permitted in pageout. */ -static int +int nfs_vnop_pageout( struct vnop_pageout_args /* { struct vnodeop_desc *a_desc; @@ -5870,13 +7366,15 @@ nfs_vnop_pageout( int error = 0, iomode; off_t off, txoffset, rxoffset; vm_offset_t ioaddr, txaddr, rxaddr; - struct uio auio; - struct iovec_32 aiov; + uio_t auio; + char uio_buf [ UIO_SIZEOF(1) ]; int nofreeupl = flags & UPL_NOCOMMIT; size_t nmwsize, biosize, iosize, pgsize, txsize, rxsize, xsize, remsize; struct nfsreq *req[MAXPAGINGREQS]; - int nextsend, nextwait, wverfset, commit, restart = 0; + 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); @@ -5886,7 +7384,7 @@ nfs_vnop_pageout( if (size <= 0) { 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); } @@ -5898,7 +7396,7 @@ nfs_vnop_pageout( biosize = nmp->nm_biosize; nmwsize = nmp->nm_wsize; - nfs_data_lock2(np, NFS_NODE_LOCK_SHARED, 0); + nfs_data_lock_noupdate(np, NFS_DATA_LOCK_SHARED); /* * Check to see whether the buffer is incore. @@ -5916,10 +7414,10 @@ nfs_vnop_pageout( 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_unlock2(np, 0); + 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) { @@ -5949,13 +7447,24 @@ nfs_vnop_pageout( end -= boff; if ((bp->nb_dirtyoff < start) && (bp->nb_dirtyend > end)) { - /* not gonna be able to clip the dirty region */ + /* + * 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_unlock2(np, 0); + 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) || @@ -5975,13 +7484,13 @@ nfs_vnop_pageout( nfs_buf_remfree(bp); lck_mtx_unlock(nfs_buf_mutex); SET(bp->nb_flags, NB_INVAL); - nfs_lock(np, NFS_NODE_LOCK_FORCE); + 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_unlock(np); + nfs_node_unlock(np); nfs_buf_release(bp, 1); } else { lck_mtx_unlock(nfs_buf_mutex); @@ -5993,28 +7502,30 @@ nfs_vnop_pageout( if (!IS_VALID_CRED(cred)) cred = vfs_context_ucred(ap->a_context); - nfs_lock(np, NFS_NODE_LOCK_FORCE); + nfs_node_lock_force(np); if (np->n_flag & NWRITEERR) { error = np->n_error; - nfs_unlock(np); - nfs_data_unlock2(np, 0); + 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 (error); } - nfs_unlock(np); + 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_unlock2(np, 0); + 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) @@ -6037,17 +7548,14 @@ nfs_vnop_pageout( bzero((caddr_t)(ioaddr + io), size - io); FSDBG(321, np->n_size, f_offset, f_offset + io, size - io); } - nfs_data_unlock2(np, 0); + nfs_data_unlock_noupdate(np); -#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_WRITE; - auio.uio_procp = vfs_context_proc(ap->a_context); + 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; @@ -6057,22 +7565,28 @@ tryagain: bzero(req, sizeof(req)); nextsend = nextwait = 0; do { + 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); - aiov.iov_len = iosize; - aiov.iov_base = (uintptr_t)txaddr; - auio.uio_iovs.iov32p = &aiov; - auio.uio_iovcnt = 1; - auio.uio_offset = txoffset; - uio_uio_resid_set(&auio, iosize); - FSDBG(323, auio.uio_offset, iosize, txaddr, txsize); - OSAddAtomic(1, (SInt32*)&nfsstats.pageouts); + 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]))) { + 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; @@ -6087,6 +7601,19 @@ tryagain: 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; @@ -6096,7 +7623,7 @@ tryagain: wverfset = 1; } else if (wverf != wverf2) { /* verifier changed, so we need to restart all the writes */ - restart++; + vrestart = 1; goto cancel; } /* Retain the lowest commitment level returned. */ @@ -6109,21 +7636,28 @@ tryagain: if (remsize > 0) { /* need to try sending the remainder */ iosize = remsize; - aiov.iov_len = remsize; - aiov.iov_base = (uintptr_t)rxaddr; - auio.uio_iovs.iov32p = &aiov; - auio.uio_iovcnt = 1; - auio.uio_offset = rxoffset; - uio_uio_resid_set(&auio, 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); + 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 */ - restart++; + vrestart = 1; goto cancel; } if (iomode < commit) @@ -6137,12 +7671,12 @@ tryagain: } } while (!error && (txsize || rxsize)); - restart = 0; + vrestart = 0; if (!error && (commit != NFS_WRITE_FILESYNC)) { - error = nmp->nm_funcs->nf_commit_rpc(np, f_offset, xsize, cred); + error = nmp->nm_funcs->nf_commit_rpc(np, f_offset, xsize, cred, wverf); if (error == NFSERR_STALEWRITEVERF) { - restart++; + vrestart = 1; error = EIO; } } @@ -6155,12 +7689,30 @@ cancel: req[nextwait] = NULL; nextwait = (nextwait + 1) % MAXPAGINGREQS; vnode_writedone(vp); + nfs_node_lock_force(np); + np->n_numoutput--; + nfs_node_unlock(np); } - if (restart) { - if (restart <= 10) - goto tryagain; - printf("nfs_pageout: too many restarts, aborting.\n"); - FSDBG(323, f_offset, xsize, ERESTART, -1); + 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); + } + } } } @@ -6202,21 +7754,16 @@ cancel: abortflags = UPL_ABORT_DUMP_PAGES|UPL_ABORT_FREE_ON_EMPTY; if (error <= NFS_ELAST) { if ((errorcount[error] % 100) == 0) - printf("nfs_pageout: unexpected error %d. dumping vm page\n", error); + 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; } @@ -6233,7 +7780,7 @@ cancel: } /* Blktooff derives file offset given a logical block number */ -static int +int nfs_vnop_blktooff( struct vnop_blktooff_args /* { struct vnodeop_desc *a_desc; @@ -6246,7 +7793,7 @@ nfs_vnop_blktooff( vnode_t vp = ap->a_vp; struct nfsmount *nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); biosize = nmp->nm_biosize; @@ -6255,7 +7802,7 @@ nfs_vnop_blktooff( return (0); } -static int +int nfs_vnop_offtoblk( struct vnop_offtoblk_args /* { struct vnodeop_desc *a_desc; @@ -6268,7 +7815,7 @@ nfs_vnop_offtoblk( vnode_t vp = ap->a_vp; struct nfsmount *nmp = VTONMP(vp); - if (!nmp) + if (nfs_mount_gone(nmp)) return (ENXIO); biosize = nmp->nm_biosize; @@ -6277,3 +7824,84 @@ nfs_vnop_offtoblk( 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 (nfs_mount_gone(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); +}