/*
- * Copyright (c) 1995-2008 Apple Inc. All rights reserved.
+ * Copyright (c) 1995-2015 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <sys/quota.h>
#include <sys/kdebug.h>
#include <sys/fsevents.h>
+#include <sys/imgsrc.h>
#include <sys/sysproto.h>
#include <sys/xattr.h>
#include <sys/fcntl.h>
#include <sys/fsctl.h>
#include <sys/ubc_internal.h>
#include <sys/disk.h>
+#include <sys/content_protection.h>
#include <machine/cons.h>
#include <machine/limits.h>
#include <miscfs/specfs/specdev.h>
-#include <miscfs/union/union.h>
#include <security/audit/audit.h>
#include <bsm/audit_kevents.h>
#include <mach/mach_types.h>
#include <kern/kern_types.h>
#include <kern/kalloc.h>
+#include <kern/task.h>
#include <vm/vm_pageout.h>
#include <libkern/OSAtomic.h>
#include <pexpert/pexpert.h>
+#include <IOKit/IOBSD.h>
#if CONFIG_MACF
#include <security/mac.h>
static int getutimes(user_addr_t usrtvp, struct timespec *tsp);
static int setutimes(vfs_context_t ctx, vnode_t vp, const struct timespec *ts, int nullflag);
static int sync_callback(mount_t, void *);
+static void sync_thread(void *, __unused wait_result_t);
+static int sync_async(int);
static int munge_statfs(struct mount *mp, struct vfsstatfs *sfsp,
user_addr_t bufp, int *sizep, boolean_t is_64_bit,
boolean_t partial_copy);
static int statfs64_common(struct mount *mp, struct vfsstatfs *sfsp,
user_addr_t bufp);
static int fsync_common(proc_t p, struct fsync_args *uap, int flags);
+static int mount_common(char *fstypename, vnode_t pvp, vnode_t vp,
+ struct componentname *cnp, user_addr_t fsmountargs,
+ int flags, uint32_t internal_flags, char *labelstr, boolean_t kernelmount,
+ vfs_context_t ctx);
+void vfs_notify_mount(vnode_t pdvp);
+
+int prepare_coveredvp(vnode_t vp, vfs_context_t ctx, struct componentname *cnp, const char *fsname, boolean_t skip_auth);
+
+struct fd_vn_data * fg_vn_data_alloc(void);
+
+/*
+ * Max retries for ENOENT returns from vn_authorize_{rmdir, unlink, rename}
+ * Concurrent lookups (or lookups by ids) on hard links can cause the
+ * vn_getpath (which does not re-enter the filesystem as vn_getpath_fsenter
+ * does) to return ENOENT as the path cannot be returned from the name cache
+ * alone. We have no option but to retry and hope to get one namei->reverse path
+ * generation done without an intervening lookup, lookup by id on the hard link
+ * item. This is only an issue for MAC hooks which cannot reenter the filesystem
+ * which currently are the MAC hooks for rename, unlink and rmdir.
+ */
+#define MAX_AUTHORIZE_ENOENT_RETRIES 1024
+
+static int rmdirat_internal(vfs_context_t, int, user_addr_t, enum uio_seg);
+
+static int fsgetpath_internal(vfs_context_t, int, uint64_t, vm_size_t, caddr_t, int *);
#ifdef CONFIG_IMGSRC_ACCESS
-static int prepare_coveredvp(vnode_t vp, vfs_context_t ctx, struct componentname *cnp, const char *fsname);
static int authorize_devpath_and_update_mntfromname(mount_t mp, user_addr_t devpath, vnode_t *devvpp, vfs_context_t ctx);
static int place_mount_and_checkdirs(mount_t mp, vnode_t vp, vfs_context_t ctx);
static void undo_place_on_covered_vp(mount_t mp, vnode_t vp);
static int mount_begin_update(mount_t mp, vfs_context_t ctx, int flags);
static void mount_end_update(mount_t mp);
-static int relocate_imageboot_source(vnode_t vp, struct componentname *cnp, const char *fsname, vfs_context_t ctx, boolean_t is64bit, user_addr_t fsmountargs);
+static int relocate_imageboot_source(vnode_t pvp, vnode_t vp, struct componentname *cnp, const char *fsname, vfs_context_t ctx, boolean_t is64bit, user_addr_t fsmountargs, boolean_t by_index);
#endif /* CONFIG_IMGSRC_ACCESS */
int (*union_dircheckp)(struct vnode **, struct fileproc *, vfs_context_t);
int sync_internal(void);
__private_extern__
-int open1(vfs_context_t, struct nameidata *, int, struct vnode_attr *, int32_t *);
-
-__private_extern__
-int unlink1(vfs_context_t, struct nameidata *, int);
-
-
-#ifdef __APPLE_API_OBSOLETE
-struct fstatv_args {
- int fd; /* file descriptor of the target file */
- struct vstat *vsb; /* vstat structure for returned info */
-};
-struct lstatv_args {
- const char *path; /* pathname of the target file */
- struct vstat *vsb; /* vstat structure for returned info */
-};
-struct mkcomplex_args {
- const char *path; /* pathname of the file to be created */
- mode_t mode; /* access mode for the newly created file */
- u_int32_t type; /* format of the complex file */
-};
-struct statv_args {
- const char *path; /* pathname of the target file */
- struct vstat *vsb; /* vstat structure for returned info */
-};
-
-int fstatv(proc_t p, struct fstatv_args *uap, int32_t *retval);
-int lstatv(proc_t p, struct lstatv_args *uap, int32_t *retval);
-int mkcomplex(proc_t p, struct mkcomplex_args *uap, int32_t *retval);
-int statv(proc_t p, struct statv_args *uap, int32_t *retval);
+int unlink1(vfs_context_t, vnode_t, user_addr_t, enum uio_seg, int);
-#endif /* __APPLE_API_OBSOLETE */
+extern lck_grp_t *fd_vn_lck_grp;
+extern lck_grp_attr_t *fd_vn_lck_grp_attr;
+extern lck_attr_t *fd_vn_lck_attr;
/*
* incremented each time a mount or unmount operation occurs
/* counts number of mount and unmount operations */
unsigned int vfs_nummntops=0;
-extern struct fileops vnops;
+extern const struct fileops vnops;
+#if CONFIG_APPLEDOUBLE
extern errno_t rmdir_remove_orphaned_appleDouble(vnode_t, vfs_context_t, int *);
+#endif /* CONFIG_APPLEDOUBLE */
+typedef uint32_t vfs_rename_flags_t;
+#if CONFIG_SECLUDED_RENAME
+enum {
+ VFS_SECLUDE_RENAME = 0x00000001
+};
+#endif
/*
* Virtual File System System Calls
*/
+#if NFSCLIENT || DEVFS
+/*
+ * Private in-kernel mounting spi (NFS only, not exported)
+ */
+ __private_extern__
+boolean_t
+vfs_iskernelmount(mount_t mp)
+{
+ return ((mp->mnt_kern_flag & MNTK_KERNEL_MOUNT) ? TRUE : FALSE);
+}
+
+ __private_extern__
+int
+kernel_mount(char *fstype, vnode_t pvp, vnode_t vp, const char *path,
+ void *data, __unused size_t datalen, int syscall_flags, __unused uint32_t kern_flags, vfs_context_t ctx)
+{
+ struct nameidata nd;
+ boolean_t did_namei;
+ int error;
+
+ NDINIT(&nd, LOOKUP, OP_MOUNT, FOLLOW | AUDITVNPATH1 | WANTPARENT,
+ UIO_SYSSPACE, CAST_USER_ADDR_T(path), ctx);
+
+ /*
+ * Get the vnode to be covered if it's not supplied
+ */
+ if (vp == NULLVP) {
+ error = namei(&nd);
+ if (error)
+ return (error);
+ vp = nd.ni_vp;
+ pvp = nd.ni_dvp;
+ did_namei = TRUE;
+ } else {
+ char *pnbuf = CAST_DOWN(char *, path);
+
+ nd.ni_cnd.cn_pnbuf = pnbuf;
+ nd.ni_cnd.cn_pnlen = strlen(pnbuf) + 1;
+ did_namei = FALSE;
+ }
+
+ error = mount_common(fstype, pvp, vp, &nd.ni_cnd, CAST_USER_ADDR_T(data),
+ syscall_flags, kern_flags, NULL, TRUE, ctx);
+
+ if (did_namei) {
+ vnode_put(vp);
+ vnode_put(pvp);
+ nameidone(&nd);
+ }
+
+ return (error);
+}
+#endif /* NFSCLIENT || DEVFS */
+
/*
* Mount a file system.
*/
return (__mac_mount(p, &muap, retval));
}
+void
+vfs_notify_mount(vnode_t pdvp)
+{
+ vfs_event_signal(NULL, VQ_MOUNT, (intptr_t)NULL);
+ lock_vnode_and_post(pdvp, NOTE_WRITE);
+}
+
/*
* __mac_mount:
* Mount a file system taking into account MAC label behavior.
* Returns: 0 Success
* !0 Not success
*/
+boolean_t root_fs_upgrade_try = FALSE;
+
int
__mac_mount(struct proc *p, register struct __mac_mount_args *uap, __unused int32_t *retval)
{
- struct vnode *vp, *pvp;
+ vnode_t pvp = NULL;
+ vnode_t vp = NULL;
+ int need_nameidone = 0;
+ vfs_context_t ctx = vfs_context_current();
+ char fstypename[MFSNAMELEN];
+ struct nameidata nd;
+ size_t dummy=0;
+ char *labelstr = NULL;
+ int flags = uap->flags;
+ int error;
+#if CONFIG_IMGSRC_ACCESS || CONFIG_MACF
+ boolean_t is_64bit = IS_64BIT_PROCESS(p);
+#else
+#pragma unused(p)
+#endif
+ /*
+ * Get the fs type name from user space
+ */
+ error = copyinstr(uap->type, fstypename, MFSNAMELEN, &dummy);
+ if (error)
+ return (error);
+
+ /*
+ * Get the vnode to be covered
+ */
+ NDINIT(&nd, LOOKUP, OP_MOUNT, FOLLOW | AUDITVNPATH1 | WANTPARENT,
+ UIO_USERSPACE, uap->path, ctx);
+ error = namei(&nd);
+ if (error) {
+ goto out;
+ }
+ need_nameidone = 1;
+ vp = nd.ni_vp;
+ pvp = nd.ni_dvp;
+
+#ifdef CONFIG_IMGSRC_ACCESS
+ /* Mounting image source cannot be batched with other operations */
+ if (flags == MNT_IMGSRC_BY_INDEX) {
+ error = relocate_imageboot_source(pvp, vp, &nd.ni_cnd, fstypename,
+ ctx, is_64bit, uap->data, (flags == MNT_IMGSRC_BY_INDEX));
+ goto out;
+ }
+#endif /* CONFIG_IMGSRC_ACCESS */
+
+#if CONFIG_MACF
+ /*
+ * Get the label string (if any) from user space
+ */
+ if (uap->mac_p != USER_ADDR_NULL) {
+ struct user_mac mac;
+ size_t ulen = 0;
+
+ if (is_64bit) {
+ struct user64_mac mac64;
+ error = copyin(uap->mac_p, &mac64, sizeof(mac64));
+ mac.m_buflen = mac64.m_buflen;
+ mac.m_string = mac64.m_string;
+ } else {
+ struct user32_mac mac32;
+ error = copyin(uap->mac_p, &mac32, sizeof(mac32));
+ mac.m_buflen = mac32.m_buflen;
+ mac.m_string = mac32.m_string;
+ }
+ if (error)
+ goto out;
+ if ((mac.m_buflen > MAC_MAX_LABEL_BUF_LEN) ||
+ (mac.m_buflen < 2)) {
+ error = EINVAL;
+ goto out;
+ }
+ MALLOC(labelstr, char *, mac.m_buflen, M_MACTEMP, M_WAITOK);
+ error = copyinstr(mac.m_string, labelstr, mac.m_buflen, &ulen);
+ if (error) {
+ goto out;
+ }
+ AUDIT_ARG(mac_string, labelstr);
+ }
+#endif /* CONFIG_MACF */
+
+ AUDIT_ARG(fflags, flags);
+
+#if SECURE_KERNEL
+ if (flags & MNT_UNION) {
+ /* No union mounts on release kernels */
+ error = EPERM;
+ goto out;
+ }
+#endif
+
+ if ((vp->v_flag & VROOT) &&
+ (vp->v_mount->mnt_flag & MNT_ROOTFS)) {
+ if (!(flags & MNT_UNION)) {
+ flags |= MNT_UPDATE;
+ }
+ else {
+ /*
+ * For a union mount on '/', treat it as fresh
+ * mount instead of update.
+ * Otherwise, union mouting on '/' used to panic the
+ * system before, since mnt_vnodecovered was found to
+ * be NULL for '/' which is required for unionlookup
+ * after it gets ENOENT on union mount.
+ */
+ flags = (flags & ~(MNT_UPDATE));
+ }
+
+#if SECURE_KERNEL
+ if ((flags & MNT_RDONLY) == 0) {
+ /* Release kernels are not allowed to mount "/" as rw */
+ error = EPERM;
+ goto out;
+ }
+#endif
+ /*
+ * See 7392553 for more details on why this check exists.
+ * Suffice to say: If this check is ON and something tries
+ * to mount the rootFS RW, we'll turn off the codesign
+ * bitmap optimization.
+ */
+#if CHECK_CS_VALIDATION_BITMAP
+ if ((flags & MNT_RDONLY) == 0 ) {
+ root_fs_upgrade_try = TRUE;
+ }
+#endif
+ }
+
+ error = mount_common(fstypename, pvp, vp, &nd.ni_cnd, uap->data, flags, 0,
+ labelstr, FALSE, ctx);
+
+out:
+
+#if CONFIG_MACF
+ if (labelstr)
+ FREE(labelstr, M_MACTEMP);
+#endif /* CONFIG_MACF */
+
+ if (vp) {
+ vnode_put(vp);
+ }
+ if (pvp) {
+ vnode_put(pvp);
+ }
+ if (need_nameidone) {
+ nameidone(&nd);
+ }
+
+ return (error);
+}
+
+/*
+ * common mount implementation (final stage of mounting)
+
+ * Arguments:
+ * fstypename file system type (ie it's vfs name)
+ * pvp parent of covered vnode
+ * vp covered vnode
+ * cnp component name (ie path) of covered vnode
+ * flags generic mount flags
+ * fsmountargs file system specific data
+ * labelstr optional MAC label
+ * kernelmount TRUE for mounts initiated from inside the kernel
+ * ctx caller's context
+ */
+static int
+mount_common(char *fstypename, vnode_t pvp, vnode_t vp,
+ struct componentname *cnp, user_addr_t fsmountargs, int flags, uint32_t internal_flags,
+ char *labelstr, boolean_t kernelmount, vfs_context_t ctx)
+{
+#if !CONFIG_MACF
+#pragma unused(labelstr)
+#endif
struct vnode *devvp = NULLVP;
struct vnode *device_vnode = NULLVP;
#if CONFIG_MACF
#endif
struct mount *mp;
struct vfstable *vfsp = (struct vfstable *)0;
+ struct proc *p = vfs_context_proc(ctx);
int error, flag = 0;
- struct vnode_attr va;
- vfs_context_t ctx = vfs_context_current();
- struct nameidata nd;
- struct nameidata nd1;
- char fstypename[MFSNAMELEN];
- size_t dummy=0;
user_addr_t devpath = USER_ADDR_NULL;
- user_addr_t fsmountargs = uap->data;
int ronly = 0;
int mntalloc = 0;
boolean_t vfsp_ref = FALSE;
- mode_t accessmode;
- boolean_t is_64bit;
boolean_t is_rwlock_locked = FALSE;
boolean_t did_rele = FALSE;
boolean_t have_usecount = FALSE;
- AUDIT_ARG(fflags, uap->flags);
-
- is_64bit = proc_is64bit(p);
-
/*
- * Get vnode to be covered
+ * Process an update for an existing mount
*/
- NDINIT(&nd, LOOKUP, NOTRIGGER | FOLLOW | AUDITVNPATH1 | WANTPARENT,
- UIO_USERSPACE, uap->path, ctx);
- error = namei(&nd);
- if (error)
- return (error);
- vp = nd.ni_vp;
- pvp = nd.ni_dvp;
-
- if ((vp->v_flag & VROOT) &&
- (vp->v_mount->mnt_flag & MNT_ROOTFS))
- uap->flags |= MNT_UPDATE;
-
- error = copyinstr(uap->type, fstypename, MFSNAMELEN, &dummy);
- if (error)
- goto out1;
-
-#ifdef CONFIG_IMGSRC_ACCESS
- if (uap->flags == MNT_IMGSRC) {
- error = relocate_imageboot_source(vp, &nd.ni_cnd, fstypename, ctx, is_64bit, fsmountargs);
- vnode_put(pvp);
- vnode_put(vp);
- return error;
- }
-#endif /* CONFIG_IMGSRC_ACCESS */
-
- if (uap->flags & MNT_UPDATE) {
+ if (flags & MNT_UPDATE) {
if ((vp->v_flag & VROOT) == 0) {
error = EINVAL;
goto out1;
* We only allow the filesystem to be reloaded if it
* is currently mounted read-only.
*/
- if ((uap->flags & MNT_RELOAD) &&
+ if ((flags & MNT_RELOAD) &&
((mp->mnt_flag & MNT_RDONLY) == 0)) {
error = ENOTSUP;
goto out1;
}
+ /*
+ * If content protection is enabled, update mounts are not
+ * allowed to turn it off.
+ */
+ if ((mp->mnt_flag & MNT_CPROTECT) &&
+ ((flags & MNT_CPROTECT) == 0)) {
+ error = EINVAL;
+ goto out1;
+ }
+
#ifdef CONFIG_IMGSRC_ACCESS
/* Can't downgrade the backer of the root FS */
if ((mp->mnt_kern_flag & MNTK_BACKS_ROOT) &&
- (!vfs_isrdonly(mp)) && (uap->flags & MNT_RDONLY))
- {
+ (!vfs_isrdonly(mp)) && (flags & MNT_RDONLY)) {
error = ENOTSUP;
goto out1;
}
#if CONFIG_MACF
error = mac_mount_check_remount(ctx, mp);
if (error != 0) {
- lck_rw_done(&mp->mnt_rwlock);
goto out1;
}
#endif
* For non-root users, silently enforce MNT_NOSUID and MNT_NODEV,
* and MNT_NOEXEC if mount point is already MNT_NOEXEC.
*/
- if (suser(vfs_context_ucred(ctx), NULL)) {
- uap->flags |= MNT_NOSUID | MNT_NODEV;
+ if ((!kernelmount) && suser(vfs_context_ucred(ctx), NULL)) {
+ flags |= MNT_NOSUID | MNT_NODEV;
if (mp->mnt_flag & MNT_NOEXEC)
- uap->flags |= MNT_NOEXEC;
+ flags |= MNT_NOEXEC;
}
flag = mp->mnt_flag;
- mp->mnt_flag |=
- uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
+
+
+ mp->mnt_flag |= flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
vfsp = mp->mnt_vtable;
goto update;
}
- /*
- * If the user is not root, ensure that they own the directory
- * onto which we are attempting to mount.
- */
- VATTR_INIT(&va);
- VATTR_WANTED(&va, va_uid);
- if ((error = vnode_getattr(vp, &va, ctx)) ||
- (va.va_uid != kauth_cred_getuid(vfs_context_ucred(ctx)) &&
- (error = suser(vfs_context_ucred(ctx), &p->p_acflag)))) {
- goto out1;
- }
/*
* For non-root users, silently enforce MNT_NOSUID and MNT_NODEV, and
* MNT_NOEXEC if mount point is already MNT_NOEXEC.
*/
- if (suser(vfs_context_ucred(ctx), NULL)) {
- uap->flags |= MNT_NOSUID | MNT_NODEV;
+ if ((!kernelmount) && suser(vfs_context_ucred(ctx), NULL)) {
+ flags |= MNT_NOSUID | MNT_NODEV;
if (vp->v_mount->mnt_flag & MNT_NOEXEC)
- uap->flags |= MNT_NOEXEC;
- }
- if ( (error = VNOP_FSYNC(vp, MNT_WAIT, ctx)) )
- goto out1;
-
- if ( (error = buf_invalidateblks(vp, BUF_WRITE_DATA, 0, 0)) )
- goto out1;
-
- if (vp->v_type != VDIR) {
- error = ENOTDIR;
- goto out1;
+ flags |= MNT_NOEXEC;
}
/* XXXAUDIT: Should we capture the type on the error path as well? */
error = ENODEV;
goto out1;
}
-#if CONFIG_MACF
- error = mac_mount_check_mount(ctx, vp,
- &nd.ni_cnd, vfsp->vfc_name);
- if (error != 0)
+
+ /*
+ * VFC_VFSLOCALARGS is not currently supported for kernel mounts
+ */
+ if (kernelmount && (vfsp->vfc_vfsflags & VFC_VFSLOCALARGS)) {
+ error = EINVAL; /* unsupported request */
goto out1;
-#endif
- if (ISSET(vp->v_flag, VMOUNT) && (vp->v_mountedhere != NULL)) {
- error = EBUSY;
+ }
+
+ error = prepare_coveredvp(vp, ctx, cnp, fstypename, ((internal_flags & KERNEL_MOUNT_NOAUTH) != 0));
+ if (error != 0) {
goto out1;
}
- vnode_lock_spin(vp);
- SET(vp->v_flag, VMOUNT);
- vnode_unlock(vp);
/*
- * Allocate and initialize the filesystem.
+ * Allocate and initialize the filesystem (mount_t)
*/
MALLOC_ZONE(mp, struct mount *, (u_int32_t)sizeof(struct mount),
M_MOUNT, M_WAITOK);
mp->mnt_vtable = vfsp;
//mp->mnt_stat.f_type = vfsp->vfc_typenum;
mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
- strncpy(mp->mnt_vfsstat.f_fstypename, vfsp->vfc_name, MFSTYPENAMELEN);
- strncpy(mp->mnt_vfsstat.f_mntonname, nd.ni_cnd.cn_pnbuf, MAXPATHLEN);
+ strlcpy(mp->mnt_vfsstat.f_fstypename, vfsp->vfc_name, MFSTYPENAMELEN);
+ strlcpy(mp->mnt_vfsstat.f_mntonname, cnp->cn_pnbuf, MAXPATHLEN);
mp->mnt_vnodecovered = vp;
mp->mnt_vfsstat.f_owner = kauth_cred_getuid(vfs_context_ucred(ctx));
- mp->mnt_devbsdunit = LOWPRI_MAX_NUM_DEV - 1;
+ mp->mnt_throttle_mask = LOWPRI_MAX_NUM_DEV - 1;
+ mp->mnt_devbsdunit = 0;
/* XXX 3762912 hack to support HFS filesystem 'owner' - filesystem may update later */
vfs_setowner(mp, KAUTH_UID_NONE, KAUTH_GID_NONE);
-
+
+#if NFSCLIENT || DEVFS
+ if (kernelmount)
+ mp->mnt_kern_flag |= MNTK_KERNEL_MOUNT;
+ if ((internal_flags & KERNEL_MOUNT_PERMIT_UNMOUNT) != 0)
+ mp->mnt_kern_flag |= MNTK_PERMIT_UNMOUNT;
+#endif /* NFSCLIENT || DEVFS */
+
update:
/*
* Set the mount level flags.
*/
- if (uap->flags & MNT_RDONLY)
+ if (flags & MNT_RDONLY)
mp->mnt_flag |= MNT_RDONLY;
- else if (mp->mnt_flag & MNT_RDONLY)
+ else if (mp->mnt_flag & MNT_RDONLY) {
+ // disallow read/write upgrades of file systems that
+ // had the TYPENAME_OVERRIDE feature set.
+ if (mp->mnt_kern_flag & MNTK_TYPENAME_OVERRIDE) {
+ error = EPERM;
+ goto out1;
+ }
mp->mnt_kern_flag |= MNTK_WANTRDWR;
-
+ }
mp->mnt_flag &= ~(MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC |
- MNT_UNKNOWNPERMISSIONS | MNT_DONTBROWSE | MNT_AUTOMOUNTED |
- MNT_DEFWRITE | MNT_NOATIME | MNT_QUARANTINE | MNT_CPROTECT );
-
- mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
- MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC |
- MNT_UNKNOWNPERMISSIONS | MNT_DONTBROWSE | MNT_AUTOMOUNTED |
- MNT_DEFWRITE | MNT_NOATIME | MNT_QUARANTINE | MNT_CPROTECT );
+ MNT_UNKNOWNPERMISSIONS | MNT_DONTBROWSE |
+ MNT_AUTOMOUNTED | MNT_DEFWRITE | MNT_NOATIME |
+ MNT_QUARANTINE | MNT_CPROTECT);
+ mp->mnt_flag |= flags & (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
+ MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC |
+ MNT_UNKNOWNPERMISSIONS | MNT_DONTBROWSE |
+ MNT_AUTOMOUNTED | MNT_DEFWRITE | MNT_NOATIME |
+ MNT_QUARANTINE | MNT_CPROTECT);
#if CONFIG_MACF
- if (uap->flags & MNT_MULTILABEL) {
+ if (flags & MNT_MULTILABEL) {
if (vfsp->vfc_vfsflags & VFC_VFSNOMACLABEL) {
error = EINVAL;
goto out1;
mp->mnt_flag |= MNT_MULTILABEL;
}
#endif
-
+ /*
+ * Process device path for local file systems if requested
+ */
if (vfsp->vfc_vfsflags & VFC_VFSLOCALARGS) {
- if (is_64bit) {
+ if (vfs_context_is64bit(ctx)) {
if ( (error = copyin(fsmountargs, (caddr_t)&devpath, sizeof(devpath))) )
goto out1;
fsmountargs += sizeof(devpath);
fsmountargs += sizeof(tmp);
}
- /* if it is not update and device name needs to be parsed */
+ /* Lookup device and authorize access to it */
if ((devpath)) {
- NDINIT(&nd1, LOOKUP, FOLLOW, UIO_USERSPACE, devpath, ctx);
- if ( (error = namei(&nd1)) )
+ struct nameidata nd;
+
+ NDINIT(&nd, LOOKUP, OP_MOUNT, FOLLOW, UIO_USERSPACE, devpath, ctx);
+ if ( (error = namei(&nd)) )
goto out1;
- strncpy(mp->mnt_vfsstat.f_mntfromname, nd1.ni_cnd.cn_pnbuf, MAXPATHLEN);
- devvp = nd1.ni_vp;
+ strlcpy(mp->mnt_vfsstat.f_mntfromname, nd.ni_cnd.cn_pnbuf, MAXPATHLEN);
+ devvp = nd.ni_vp;
- nameidone(&nd1);
+ nameidone(&nd);
if (devvp->v_type != VBLK) {
error = ENOTBLK;
* permissions on the device.
*/
if (suser(vfs_context_ucred(ctx), NULL) != 0) {
- accessmode = KAUTH_VNODE_READ_DATA;
+ mode_t accessmode = KAUTH_VNODE_READ_DATA;
+
if ((mp->mnt_flag & MNT_RDONLY) == 0)
accessmode |= KAUTH_VNODE_WRITE_DATA;
if ((error = vnode_authorize(devvp, NULL, accessmode, ctx)) != 0)
goto out2;
}
}
- if (devpath && ((uap->flags & MNT_UPDATE) == 0)) {
+ /* On first mount, preflight and open device */
+ if (devpath && ((flags & MNT_UPDATE) == 0)) {
if ( (error = vnode_ref(devvp)) )
goto out2;
/*
mp->mnt_devvp = devvp;
device_vnode = devvp;
- } else {
- if ((mp->mnt_flag & MNT_RDONLY) && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
- dev_t dev;
- int maj;
- /*
- * If upgrade to read-write by non-root, then verify
- * that user has necessary permissions on the device.
- */
- device_vnode = mp->mnt_devvp;
-
- if (device_vnode) {
- vnode_getalways(device_vnode);
- if (suser(vfs_context_ucred(ctx), NULL)) {
- if ((error = vnode_authorize(device_vnode, NULL,
- KAUTH_VNODE_READ_DATA | KAUTH_VNODE_WRITE_DATA, ctx)) != 0) {
- vnode_put(device_vnode);
- goto out2;
- }
- }
+ } else if ((mp->mnt_flag & MNT_RDONLY) &&
+ (mp->mnt_kern_flag & MNTK_WANTRDWR) &&
+ (device_vnode = mp->mnt_devvp)) {
+ dev_t dev;
+ int maj;
+ /*
+ * If upgrade to read-write by non-root, then verify
+ * that user has necessary permissions on the device.
+ */
+ vnode_getalways(device_vnode);
- /* Tell the device that we're upgrading */
- dev = (dev_t)device_vnode->v_rdev;
- maj = major(dev);
+ if (suser(vfs_context_ucred(ctx), NULL) &&
+ (error = vnode_authorize(device_vnode, NULL,
+ KAUTH_VNODE_READ_DATA | KAUTH_VNODE_WRITE_DATA,
+ ctx)) != 0) {
+ vnode_put(device_vnode);
+ goto out2;
+ }
- if ((u_int)maj >= (u_int)nblkdev)
- panic("Volume mounted on a device with invalid major number.\n");
+ /* Tell the device that we're upgrading */
+ dev = (dev_t)device_vnode->v_rdev;
+ maj = major(dev);
- error = bdevsw[maj].d_open(dev, FREAD | FWRITE, S_IFBLK, p);
+ if ((u_int)maj >= (u_int)nblkdev)
+ panic("Volume mounted on a device with invalid major number.");
- vnode_put(device_vnode);
- if (error != 0) {
- goto out2;
- }
- }
- }
+ error = bdevsw[maj].d_open(dev, FREAD | FWRITE, S_IFBLK, p);
+ vnode_put(device_vnode);
device_vnode = NULLVP;
+ if (error != 0) {
+ goto out2;
+ }
}
}
#if CONFIG_MACF
- if ((uap->flags & MNT_UPDATE) == 0) {
+ if ((flags & MNT_UPDATE) == 0) {
mac_mount_label_init(mp);
mac_mount_label_associate(ctx, mp);
}
- if (uap->mac_p != USER_ADDR_NULL) {
- struct user_mac mac;
- char *labelstr = NULL;
- size_t ulen = 0;
-
- if ((uap->flags & MNT_UPDATE) != 0) {
- error = mac_mount_check_label_update(
- ctx, mp);
+ if (labelstr) {
+ if ((flags & MNT_UPDATE) != 0) {
+ error = mac_mount_check_label_update(ctx, mp);
if (error != 0)
goto out3;
}
- if (is_64bit) {
- error = copyin(uap->mac_p, &mac, sizeof(mac));
- } else {
- struct mac mac32;
- error = copyin(uap->mac_p, &mac32, sizeof(mac32));
- mac.m_buflen = mac32.m_buflen;
- mac.m_string = CAST_USER_ADDR_T(mac32.m_string);
- }
- if (error != 0)
- goto out3;
- if ((mac.m_buflen > MAC_MAX_LABEL_BUF_LEN) ||
- (mac.m_buflen < 2)) {
- error = EINVAL;
- goto out3;
- }
- MALLOC(labelstr, char *, mac.m_buflen, M_MACTEMP, M_WAITOK);
- error = copyinstr(mac.m_string, labelstr, mac.m_buflen, &ulen);
- if (error != 0) {
- FREE(labelstr, M_MACTEMP);
- goto out3;
- }
- AUDIT_ARG(mac_string, labelstr);
- error = mac_mount_label_internalize(mp->mnt_mntlabel, labelstr);
- FREE(labelstr, M_MACTEMP);
- if (error != 0)
- goto out3;
}
#endif
- if (device_vnode != NULL) {
- VNOP_IOCTL(device_vnode, DKIOCGETBSDUNIT, (caddr_t)&mp->mnt_devbsdunit, 0, NULL);
- mp->mnt_devbsdunit %= LOWPRI_MAX_NUM_DEV;
- }
-
/*
* Mount the filesystem.
*/
error = VFS_MOUNT(mp, device_vnode, fsmountargs, ctx);
- if (uap->flags & MNT_UPDATE) {
+ if (flags & MNT_UPDATE) {
if (mp->mnt_kern_flag & MNTK_WANTRDWR)
mp->mnt_flag &= ~MNT_RDONLY;
mp->mnt_flag &=~
(MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
if (error)
- mp->mnt_flag = flag;
+ mp->mnt_flag = flag; /* restore flag value */
vfs_event_signal(NULL, VQ_UPDATE, (intptr_t)NULL);
lck_rw_done(&mp->mnt_rwlock);
is_rwlock_locked = FALSE;
if (!error)
enablequotas(mp, ctx);
- goto out2;
+ goto exit;
}
+
/*
* Put the new filesystem on the mount list after root.
*/
*/
(void)VFS_START(mp, 0, ctx);
- error = mount_list_add(mp);
- if (error != 0) {
+ if (mount_list_add(mp) != 0) {
+ /*
+ * The system is shutting down trying to umount
+ * everything, so fail with a plausible errno.
+ */
+ error = EBUSY;
goto out4;
}
-
lck_rw_done(&mp->mnt_rwlock);
is_rwlock_locked = FALSE;
}
/* Now that mount is setup, notify the listeners */
- vfs_event_signal(NULL, VQ_MOUNT, (intptr_t)NULL);
+ vfs_notify_mount(pvp);
+ IOBSDMountChange(mp, kIOMountChangeMount);
+
} else {
+ /* If we fail a fresh mount, there should be no vnodes left hooked into the mountpoint. */
+ if (mp->mnt_vnodelist.tqh_first != NULL) {
+ panic("mount_common(): mount of %s filesystem failed with %d, but vnode list is not empty.",
+ mp->mnt_vtable->vfc_name, error);
+ }
+
vnode_lock_spin(vp);
CLR(vp->v_flag, VMOUNT);
vnode_unlock(vp);
}
lck_rw_done(&mp->mnt_rwlock);
is_rwlock_locked = FALSE;
+
+ /*
+ * if we get here, we have a mount structure that needs to be freed,
+ * but since the coveredvp hasn't yet been updated to point at it,
+ * no need to worry about other threads holding a crossref on this mp
+ * so it's ok to just free it
+ */
mount_lock_destroy(mp);
#if CONFIG_MACF
mac_mount_label_destroy(mp);
#endif
FREE_ZONE((caddr_t)mp, sizeof (struct mount), M_MOUNT);
}
- nameidone(&nd);
-
+exit:
/*
- * drop I/O count on covered 'vp' and
- * on the device vp if there was one
+ * drop I/O count on the device vp if there was one
*/
if (devpath && devvp)
vnode_put(devvp);
- vnode_put(vp);
-
- /* Note that we've changed something in the parent directory */
- post_event_if_success(pvp, error, NOTE_WRITE);
- vnode_put(pvp);
return(error);
+/* Error condition exits */
out4:
(void)VFS_UNMOUNT(mp, MNT_FORCE, ctx);
+
+ /*
+ * If the mount has been placed on the covered vp,
+ * it may have been discovered by now, so we have
+ * to treat this just like an unmount
+ */
+ mount_lock_spin(mp);
+ mp->mnt_lflag |= MNT_LDEAD;
+ mount_unlock(mp);
+
if (device_vnode != NULLVP) {
vnode_rele(device_vnode);
VNOP_CLOSE(device_vnode, mp->mnt_flag & MNT_RDONLY ? FREAD : FREAD|FWRITE,
ctx);
did_rele = TRUE;
}
+
vnode_lock_spin(vp);
+
+ mp->mnt_crossref++;
vp->v_mountedhere = (mount_t) 0;
+
vnode_unlock(vp);
-
+
if (have_usecount) {
vnode_rele(vp);
}
out3:
- if (devpath && ((uap->flags & MNT_UPDATE) == 0) && (!did_rele))
+ if (devpath && ((flags & MNT_UPDATE) == 0) && (!did_rele))
vnode_rele(devvp);
out2:
if (devpath && devvp)
if (is_rwlock_locked == TRUE) {
lck_rw_done(&mp->mnt_rwlock);
}
+
if (mntalloc) {
+ if (mp->mnt_crossref)
+ mount_dropcrossref(mp, vp, 0);
+ else {
+ mount_lock_destroy(mp);
#if CONFIG_MACF
- mac_mount_label_destroy(mp);
+ mac_mount_label_destroy(mp);
#endif
- FREE_ZONE((caddr_t)mp, sizeof (struct mount), M_MOUNT);
+ FREE_ZONE((caddr_t)mp, sizeof (struct mount), M_MOUNT);
+ }
}
-
if (vfsp_ref) {
mount_list_lock();
vfsp->vfc_refcount--;
mount_list_unlock();
}
- vnode_put(vp);
- vnode_put(pvp);
- nameidone(&nd);
return(error);
}
-#ifdef CONFIG_IMGSRC_ACCESS
/*
* Flush in-core data, check for competing mount attempts,
* and set VMOUNT
*/
-static int
-prepare_coveredvp(vnode_t vp, vfs_context_t ctx, struct componentname *cnp, const char *fsname)
+int
+prepare_coveredvp(vnode_t vp, vfs_context_t ctx, struct componentname *cnp, const char *fsname, boolean_t skip_auth)
{
+#if !CONFIG_MACF
+#pragma unused(cnp,fsname)
+#endif
struct vnode_attr va;
int error;
- /*
- * If the user is not root, ensure that they own the directory
- * onto which we are attempting to mount.
- */
- VATTR_INIT(&va);
- VATTR_WANTED(&va, va_uid);
- if ((error = vnode_getattr(vp, &va, ctx)) ||
- (va.va_uid != kauth_cred_getuid(vfs_context_ucred(ctx)) &&
- (!vfs_context_issuser(ctx)))) {
- error = EPERM;
- goto out;
+ if (!skip_auth) {
+ /*
+ * If the user is not root, ensure that they own the directory
+ * onto which we are attempting to mount.
+ */
+ VATTR_INIT(&va);
+ VATTR_WANTED(&va, va_uid);
+ if ((error = vnode_getattr(vp, &va, ctx)) ||
+ (va.va_uid != kauth_cred_getuid(vfs_context_ucred(ctx)) &&
+ (!vfs_context_issuser(ctx)))) {
+ error = EPERM;
+ goto out;
+ }
}
if ( (error = VNOP_FSYNC(vp, MNT_WAIT, ctx)) )
return error;
}
+#if CONFIG_IMGSRC_ACCESS
+
+#if DEBUG
+#define IMGSRC_DEBUG(args...) printf(args)
+#else
+#define IMGSRC_DEBUG(args...) do { } while(0)
+#endif
+
static int
authorize_devpath_and_update_mntfromname(mount_t mp, user_addr_t devpath, vnode_t *devvpp, vfs_context_t ctx)
{
struct nameidata nd;
- vnode_t vp;
+ vnode_t vp, realdevvp;
mode_t accessmode;
int error;
- NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, devpath, ctx);
- if ( (error = namei(&nd)) )
+ NDINIT(&nd, LOOKUP, OP_LOOKUP, FOLLOW, UIO_USERSPACE, devpath, ctx);
+ if ( (error = namei(&nd)) ) {
+ IMGSRC_DEBUG("namei() failed with %d\n", error);
return error;
+ }
- strncpy(mp->mnt_vfsstat.f_mntfromname, nd.ni_cnd.cn_pnbuf, MAXPATHLEN);
vp = nd.ni_vp;
- nameidone(&nd);
- if (vp->v_type != VBLK) {
+ if (!vnode_isblk(vp)) {
+ IMGSRC_DEBUG("Not block device.\n");
error = ENOTBLK;
goto out;
}
- if (major(vp->v_rdev) >= nblkdev) {
+
+ realdevvp = mp->mnt_devvp;
+ if (realdevvp == NULLVP) {
+ IMGSRC_DEBUG("No device backs the mount.\n");
error = ENXIO;
goto out;
}
+
+ error = vnode_getwithref(realdevvp);
+ if (error != 0) {
+ IMGSRC_DEBUG("Coudn't get iocount on device.\n");
+ goto out;
+ }
+
+ if (vnode_specrdev(vp) != vnode_specrdev(realdevvp)) {
+ IMGSRC_DEBUG("Wrong dev_t.\n");
+ error = ENXIO;
+ goto out1;
+ }
+
+ strlcpy(mp->mnt_vfsstat.f_mntfromname, nd.ni_cnd.cn_pnbuf, MAXPATHLEN);
+
/*
* If mount by non-root, then verify that user has necessary
* permissions on the device.
accessmode = KAUTH_VNODE_READ_DATA;
if ((mp->mnt_flag & MNT_RDONLY) == 0)
accessmode |= KAUTH_VNODE_WRITE_DATA;
- if ((error = vnode_authorize(vp, NULL, accessmode, ctx)) != 0)
- goto out;
+ if ((error = vnode_authorize(vp, NULL, accessmode, ctx)) != 0) {
+ IMGSRC_DEBUG("Access denied.\n");
+ goto out1;
+ }
}
*devvpp = vp;
+
+out1:
+ vnode_put(realdevvp);
out:
+ nameidone(&nd);
if (error) {
vnode_put(vp);
}
}
static int
-relocate_imageboot_source(vnode_t vp, struct componentname *cnp,
+get_imgsrc_rootvnode(uint32_t height, vnode_t *rvpp)
+{
+ vnode_t vp;
+
+ if (height >= MAX_IMAGEBOOT_NESTING) {
+ return EINVAL;
+ }
+
+ vp = imgsrc_rootvnodes[height];
+ if ((vp != NULLVP) && (vnode_get(vp) == 0)) {
+ *rvpp = vp;
+ return 0;
+ } else {
+ return ENOENT;
+ }
+}
+
+static int
+relocate_imageboot_source(vnode_t pvp, vnode_t vp, struct componentname *cnp,
const char *fsname, vfs_context_t ctx,
- boolean_t is64bit, user_addr_t fsmountargs)
+ boolean_t is64bit, user_addr_t fsmountargs, boolean_t by_index)
{
int error;
mount_t mp;
boolean_t placed = FALSE;
- vnode_t devvp;
+ vnode_t devvp = NULLVP;
struct vfstable *vfsp;
user_addr_t devpath;
char *old_mntonname;
+ vnode_t rvp;
+ uint32_t height;
+ uint32_t flags;
/* If we didn't imageboot, nothing to move */
- if (imgsrc_rootvnode == NULLVP) {
+ if (imgsrc_rootvnodes[0] == NULLVP) {
return EINVAL;
}
return EPERM;
}
- error = vnode_get(imgsrc_rootvnode);
+ IMGSRC_DEBUG("looking for root vnode.\n");
+
+ /*
+ * Get root vnode of filesystem we're moving.
+ */
+ if (by_index) {
+ if (is64bit) {
+ struct user64_mnt_imgsrc_args mia64;
+ error = copyin(fsmountargs, &mia64, sizeof(mia64));
+ if (error != 0) {
+ IMGSRC_DEBUG("Failed to copy in arguments.\n");
+ return error;
+ }
+
+ height = mia64.mi_height;
+ flags = mia64.mi_flags;
+ devpath = mia64.mi_devpath;
+ } else {
+ struct user32_mnt_imgsrc_args mia32;
+ error = copyin(fsmountargs, &mia32, sizeof(mia32));
+ if (error != 0) {
+ IMGSRC_DEBUG("Failed to copy in arguments.\n");
+ return error;
+ }
+
+ height = mia32.mi_height;
+ flags = mia32.mi_flags;
+ devpath = mia32.mi_devpath;
+ }
+ } else {
+ /*
+ * For binary compatibility--assumes one level of nesting.
+ */
+ if (is64bit) {
+ if ( (error = copyin(fsmountargs, (caddr_t)&devpath, sizeof(devpath))) )
+ return error;
+ } else {
+ user32_addr_t tmp;
+ if ( (error = copyin(fsmountargs, (caddr_t)&tmp, sizeof(tmp))) )
+ return error;
+
+ /* munge into LP64 addr */
+ devpath = CAST_USER_ADDR_T(tmp);
+ }
+
+ height = 0;
+ flags = 0;
+ }
+
+ if (flags != 0) {
+ IMGSRC_DEBUG("%s: Got nonzero flags.\n", __FUNCTION__);
+ return EINVAL;
+ }
+
+ error = get_imgsrc_rootvnode(height, &rvp);
if (error != 0) {
+ IMGSRC_DEBUG("getting root vnode failed with %d\n", error);
return error;
}
+ IMGSRC_DEBUG("got root vnode.\n");
+
MALLOC(old_mntonname, char*, MAXPATHLEN, M_TEMP, M_WAITOK);
/* Can only move once */
- mp = vnode_mount(imgsrc_rootvnode);
+ mp = vnode_mount(rvp);
if ((mp->mnt_kern_flag & MNTK_HAS_MOVED) == MNTK_HAS_MOVED) {
+ IMGSRC_DEBUG("Already moved.\n");
error = EBUSY;
goto out0;
}
+ IMGSRC_DEBUG("Starting updated.\n");
+
/* Get exclusive rwlock on mount, authorize update on mp */
error = mount_begin_update(mp , ctx, 0);
if (error != 0) {
+ IMGSRC_DEBUG("Starting updated failed with %d\n", error);
goto out0;
}
* so we're now safe to proceed.
*/
if ((mp->mnt_kern_flag & MNTK_HAS_MOVED) == MNTK_HAS_MOVED) {
+ IMGSRC_DEBUG("Already moved [2]\n");
goto out1;
}
+
+
+ IMGSRC_DEBUG("Preparing coveredvp.\n");
/* Mark covered vnode as mount in progress, authorize placing mount on top */
- error = prepare_coveredvp(vp, ctx, cnp, fsname);
+ error = prepare_coveredvp(vp, ctx, cnp, fsname, FALSE);
if (error != 0) {
+ IMGSRC_DEBUG("Preparing coveredvp failed with %d.\n", error);
goto out1;
}
+ IMGSRC_DEBUG("Covered vp OK.\n");
+
/* Sanity check the name caller has provided */
vfsp = mp->mnt_vtable;
if (strncmp(vfsp->vfc_name, fsname, MFSNAMELEN) != 0) {
+ IMGSRC_DEBUG("Wrong fs name.\n");
error = EINVAL;
goto out2;
}
/* Check the device vnode and update mount-from name, for local filesystems */
if (vfsp->vfc_vfsflags & VFC_VFSLOCALARGS) {
- if (is64bit) {
- if ( (error = copyin(fsmountargs, (caddr_t)&devpath, sizeof(devpath))) )
- goto out2;
- fsmountargs += sizeof(devpath);
- } else {
- user32_addr_t tmp;
- if ( (error = copyin(fsmountargs, (caddr_t)&tmp, sizeof(tmp))) )
- goto out2;
- /* munge into LP64 addr */
- devpath = CAST_USER_ADDR_T(tmp);
- fsmountargs += sizeof(tmp);
- }
+ IMGSRC_DEBUG("Local, doing device validation.\n");
if (devpath != USER_ADDR_NULL) {
error = authorize_devpath_and_update_mntfromname(mp, devpath, &devvp, ctx);
if (error) {
+ IMGSRC_DEBUG("authorize_devpath_and_update_mntfromname() failed.\n");
goto out2;
}
* Place mp on top of vnode, ref the vnode, call checkdirs(),
* and increment the name cache's mount generation
*/
+
+ IMGSRC_DEBUG("About to call place_mount_and_checkdirs().\n");
error = place_mount_and_checkdirs(mp, vp, ctx);
if (error != 0) {
goto out2;
placed = TRUE;
- strncpy(old_mntonname, mp->mnt_vfsstat.f_mntonname, MAXPATHLEN);
- strncpy(mp->mnt_vfsstat.f_mntonname, cnp->cn_pnbuf, MAXPATHLEN);
+ strlcpy(old_mntonname, mp->mnt_vfsstat.f_mntonname, MAXPATHLEN);
+ strlcpy(mp->mnt_vfsstat.f_mntonname, cnp->cn_pnbuf, MAXPATHLEN);
/* Forbid future moves */
mount_lock(mp);
mount_unlock(mp);
/* Finally, add to mount list, completely ready to go */
- error = mount_list_add(mp);
- if (error != 0) {
+ if (mount_list_add(mp) != 0) {
+ /*
+ * The system is shutting down trying to umount
+ * everything, so fail with a plausible errno.
+ */
+ error = EBUSY;
goto out3;
}
mount_end_update(mp);
- vnode_put(imgsrc_rootvnode);
+ vnode_put(rvp);
FREE(old_mntonname, M_TEMP);
+ vfs_notify_mount(pvp);
+
return 0;
out3:
- strncpy(mp->mnt_vfsstat.f_mntonname, old_mntonname, MAXPATHLEN);
+ strlcpy(mp->mnt_vfsstat.f_mntonname, old_mntonname, MAXPATHLEN);
mount_lock(mp);
mp->mnt_kern_flag &= ~(MNTK_HAS_MOVED);
mount_end_update(mp);
out0:
- vnode_put(imgsrc_rootvnode);
+ vnode_put(rvp);
FREE(old_mntonname, M_TEMP);
return error;
}
*/
for (type=0; type < MAXQUOTAS; type++) {
snprintf(qfpath, sizeof(qfpath), "%s/%s.%s", mp->mnt_vfsstat.f_mntonname, qfopsname, qfextension[type]);
- NDINIT(&qnd, LOOKUP, FOLLOW, UIO_SYSSPACE, CAST_USER_ADDR_T(qfpath), ctx);
+ NDINIT(&qnd, LOOKUP, OP_MOUNT, FOLLOW, UIO_SYSSPACE,
+ CAST_USER_ADDR_T(qfpath), ctx);
if (namei(&qnd) != 0)
continue; /* option file to trigger quotas is not present */
vnode_put(qnd.ni_vp);
vnode_t tvp;
int err;
struct cdirargs cdr;
- struct uthread * uth = get_bsdthread_info(current_thread());
if (olddp->v_usecount == 1)
return(0);
- if (uth != (struct uthread *)0)
- uth->uu_notrigger = 1;
err = VFS_ROOT(olddp->v_mountedhere, &newdp, ctx);
- if (uth != (struct uthread *)0)
- uth->uu_notrigger = 0;
if (err != 0) {
#if DIAGNOSTIC
struct nameidata nd;
vfs_context_t ctx = vfs_context_current();
- NDINIT(&nd, LOOKUP, NOTRIGGER | FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_UNMOUNT, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = namei(&nd);
if (error)
proc_t p = vfs_context_proc(ctx);
/*
- * Only root, or the user that did the original mount is
- * permitted to unmount this filesystem.
+ * If the file system is not responding and MNT_NOBLOCK
+ * is set and not a forced unmount then return EBUSY.
*/
- if ((mp->mnt_vfsstat.f_owner != kauth_cred_getuid(kauth_cred_get())) &&
- (error = suser(kauth_cred_get(), &p->p_acflag)))
+ if ((mp->mnt_kern_flag & MNT_LNOTRESP) &&
+ (flags & MNT_NOBLOCK) && ((flags & MNT_FORCE) == 0)) {
+ error = EBUSY;
goto out;
+ }
+ /*
+ * Skip authorization if the mount is tagged as permissive and
+ * this is not a forced-unmount attempt.
+ */
+ if (!(((mp->mnt_kern_flag & MNTK_PERMIT_UNMOUNT) != 0) && ((flags & MNT_FORCE) == 0))) {
+ /*
+ * Only root, or the user that did the original mount is
+ * permitted to unmount this filesystem.
+ */
+ if ((mp->mnt_vfsstat.f_owner != kauth_cred_getuid(kauth_cred_get())) &&
+ (error = suser(kauth_cred_get(), &p->p_acflag)))
+ goto out;
+ }
/*
* Don't allow unmounting the root file system.
*/
int forcedunmount = 0;
int lflags = 0;
struct vnode *devvp = NULLVP;
+#if CONFIG_TRIGGERS
+ proc_t p = vfs_context_proc(ctx);
+ int did_vflush = 0;
+ int pflags_save = 0;
+#endif /* CONFIG_TRIGGERS */
- if (flags & MNT_FORCE)
- forcedunmount = 1;
mount_lock(mp);
- /* XXX post jaguar fix LK_DRAIN - then clean this up */
- if ((flags & MNT_FORCE)) {
- mp->mnt_kern_flag |= MNTK_FRCUNMOUNT;
- mp->mnt_lflag |= MNT_LFORCE;
- }
+
+ /*
+ * If already an unmount in progress just return EBUSY.
+ * Even a forced unmount cannot override.
+ */
if (mp->mnt_lflag & MNT_LUNMOUNT) {
- mp->mnt_lflag |= MNT_LWAIT;
- if(withref != 0)
+ if (withref != 0)
mount_drop(mp, 1);
- msleep((caddr_t)mp, &mp->mnt_mlock, (PVFS | PDROP), "dounmount", NULL);
- /*
- * The prior unmount attempt has probably succeeded.
- * Do not dereference mp here - returning EBUSY is safest.
- */
+ mount_unlock(mp);
return (EBUSY);
}
+
+ if (flags & MNT_FORCE) {
+ forcedunmount = 1;
+ mp->mnt_lflag |= MNT_LFORCE;
+ }
+
+#if CONFIG_TRIGGERS
+ if (flags & MNT_NOBLOCK && p != kernproc)
+ pflags_save = OSBitOrAtomic(P_NOREMOTEHANG, &p->p_flag);
+#endif
+
mp->mnt_kern_flag |= MNTK_UNMOUNT;
mp->mnt_lflag |= MNT_LUNMOUNT;
mp->mnt_flag &=~ MNT_ASYNC;
mp->mnt_realrootvp = NULLVP;
mount_unlock(mp);
+ if (forcedunmount && (flags & MNT_LNOSUB) == 0) {
+ /*
+ * Force unmount any mounts in this filesystem.
+ * If any unmounts fail - just leave them dangling.
+ * Avoids recursion.
+ */
+ (void) dounmount_submounts(mp, flags | MNT_LNOSUB, ctx);
+ }
+
/*
* taking the name_cache_lock exclusively will
* insure that everyone is out of the fast path who
}
}
}
-
+
+ IOBSDMountChange(mp, kIOMountChangeUnmount);
+
+#if CONFIG_TRIGGERS
+ vfs_nested_trigger_unmounts(mp, flags, ctx);
+ did_vflush = 1;
+#endif
if (forcedunmount)
lflags |= FORCECLOSE;
error = vflush(mp, NULLVP, SKIPSWAP | SKIPSYSTEM | SKIPROOT | lflags);
lck_rw_done(&mp->mnt_rwlock);
mount_list_remove(mp);
lck_rw_lock_exclusive(&mp->mnt_rwlock);
-
+
/* mark the mount point hook in the vp but not drop the ref yet */
if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
- vnode_getwithref(coveredvp);
- vnode_lock_spin(coveredvp);
- coveredvp->v_mountedhere = (struct mount *)0;
- vnode_unlock(coveredvp);
- vnode_put(coveredvp);
+ /*
+ * The covered vnode needs special handling. Trying to get an
+ * iocount must not block here as this may lead to deadlocks
+ * if the Filesystem to which the covered vnode belongs is
+ * undergoing forced unmounts. Since we hold a usecount, the
+ * vnode cannot be reused (it can, however, still be terminated)
+ */
+ vnode_getalways(coveredvp);
+ vnode_lock_spin(coveredvp);
+
+ mp->mnt_crossref++;
+ coveredvp->v_mountedhere = (struct mount *)0;
+ CLR(coveredvp->v_flag, VMOUNT);
+
+ vnode_unlock(coveredvp);
+ vnode_put(coveredvp);
}
mount_list_lock();
mp->mnt_lflag &= ~MNT_LWAIT;
needwakeup = 1;
}
+
+#if CONFIG_TRIGGERS
+ if (flags & MNT_NOBLOCK && p != kernproc) {
+ // Restore P_NOREMOTEHANG bit to its previous value
+ if ((pflags_save & P_NOREMOTEHANG) == 0)
+ OSBitAndAtomic(~((uint32_t) P_NOREMOTEHANG), &p->p_flag);
+ }
+
+ /*
+ * Callback and context are set together under the mount lock, and
+ * never cleared, so we're safe to examine them here, drop the lock,
+ * and call out.
+ */
+ if (mp->mnt_triggercallback != NULL) {
+ mount_unlock(mp);
+ if (error == 0) {
+ mp->mnt_triggercallback(mp, VTC_RELEASE, mp->mnt_triggerdata, ctx);
+ } else if (did_vflush) {
+ mp->mnt_triggercallback(mp, VTC_REPLACE, mp->mnt_triggerdata, ctx);
+ }
+ } else {
+ mount_unlock(mp);
+ }
+#else
mount_unlock(mp);
+#endif /* CONFIG_TRIGGERS */
+
lck_rw_done(&mp->mnt_rwlock);
if (needwakeup)
wakeup((caddr_t)mp);
+
if (!error) {
if ((coveredvp != NULLVP)) {
- vnode_t pvp;
+ vnode_t pvp = NULLVP;
- vnode_getwithref(coveredvp);
- pvp = vnode_getparent(coveredvp);
- vnode_rele(coveredvp);
- vnode_lock_spin(coveredvp);
- if(mp->mnt_crossref == 0) {
- vnode_unlock(coveredvp);
- mount_lock_destroy(mp);
-#if CONFIG_MACF
- mac_mount_label_destroy(mp);
+ /*
+ * The covered vnode needs special handling. Trying to
+ * get an iocount must not block here as this may lead
+ * to deadlocks if the Filesystem to which the covered
+ * vnode belongs is undergoing forced unmounts. Since we
+ * hold a usecount, the vnode cannot be reused
+ * (it can, however, still be terminated).
+ */
+ vnode_getalways(coveredvp);
+
+ mount_dropcrossref(mp, coveredvp, 0);
+ /*
+ * We'll _try_ to detect if this really needs to be
+ * done. The coveredvp can only be in termination (or
+ * terminated) if the coveredvp's mount point is in a
+ * forced unmount (or has been) since we still hold the
+ * ref.
+ */
+ if (!vnode_isrecycled(coveredvp)) {
+ pvp = vnode_getparent(coveredvp);
+#if CONFIG_TRIGGERS
+ if (coveredvp->v_resolve) {
+ vnode_trigger_rearm(coveredvp, ctx);
+ }
#endif
- FREE_ZONE((caddr_t)mp, sizeof (struct mount), M_MOUNT);
- } else {
- coveredvp->v_lflag |= VL_MOUNTDEAD;
- vnode_unlock(coveredvp);
}
+
+ vnode_rele(coveredvp);
vnode_put(coveredvp);
+ coveredvp = NULLVP;
if (pvp) {
lock_vnode_and_post(pvp, NOTE_WRITE);
return (error);
}
-void
-mount_dropcrossref(mount_t mp, vnode_t dp, int need_put)
-{
- vnode_lock(dp);
- mp->mnt_crossref--;
- if (mp->mnt_crossref < 0)
- panic("mount cross refs -ve");
- if (((dp->v_lflag & VL_MOUNTDEAD) == VL_MOUNTDEAD) && (mp->mnt_crossref == 0)) {
- dp->v_lflag &= ~VL_MOUNTDEAD;
- if (need_put)
- vnode_put_locked(dp);
- vnode_unlock(dp);
- mount_lock_destroy(mp);
-#if CONFIG_MACF
- mac_mount_label_destroy(mp);
-#endif
- FREE_ZONE((caddr_t)mp, sizeof (struct mount), M_MOUNT);
- return;
- }
- if (need_put)
- vnode_put_locked(dp);
- vnode_unlock(dp);
-}
-
-
/*
- * Sync each mounted filesystem.
+ * Unmount any mounts in this filesystem.
*/
-#if DIAGNOSTIC
-int syncprt = 0;
-struct ctldebug debug0 = { "syncprt", &syncprt };
-#endif
-
-int print_vmpage_stat=0;
-
-static int
-sync_callback(mount_t mp, void * arg)
+void
+dounmount_submounts(struct mount *mp, int flags, vfs_context_t ctx)
{
- int asyncflag;
-
- if ((mp->mnt_flag & MNT_RDONLY) == 0) {
- asyncflag = mp->mnt_flag & MNT_ASYNC;
- mp->mnt_flag &= ~MNT_ASYNC;
- VFS_SYNC(mp, arg ? MNT_WAIT : MNT_NOWAIT, vfs_context_current());
- if (asyncflag)
- mp->mnt_flag |= MNT_ASYNC;
- }
- return(VFS_RETURNED);
-}
-
+ mount_t smp;
+ fsid_t *fsids, fsid;
+ int fsids_sz;
+ int count = 0, i, m = 0;
+ vnode_t vp;
-#include <kern/clock.h>
+ mount_list_lock();
-clock_sec_t sync_wait_time = 0;
+ // Get an array to hold the submounts fsids.
+ TAILQ_FOREACH(smp, &mountlist, mnt_list)
+ count++;
+ fsids_sz = count * sizeof(fsid_t);
+ MALLOC(fsids, fsid_t *, fsids_sz, M_TEMP, M_NOWAIT);
+ if (fsids == NULL) {
+ mount_list_unlock();
+ goto out;
+ }
+ fsids[0] = mp->mnt_vfsstat.f_fsid; // Prime the pump
-/* ARGSUSED */
-int
-sync(__unused proc_t p, __unused struct sync_args *uap, __unused int32_t *retval)
+ /*
+ * Fill the array with submount fsids.
+ * Since mounts are always added to the tail of the mount list, the
+ * list is always in mount order.
+ * For each mount check if the mounted-on vnode belongs to a
+ * mount that's already added to our array of mounts to be unmounted.
+ */
+ for (smp = TAILQ_NEXT(mp, mnt_list); smp; smp = TAILQ_NEXT(smp, mnt_list)) {
+ vp = smp->mnt_vnodecovered;
+ if (vp == NULL)
+ continue;
+ fsid = vnode_mount(vp)->mnt_vfsstat.f_fsid; // Underlying fsid
+ for (i = 0; i <= m; i++) {
+ if (fsids[i].val[0] == fsid.val[0] &&
+ fsids[i].val[1] == fsid.val[1]) {
+ fsids[++m] = smp->mnt_vfsstat.f_fsid;
+ break;
+ }
+ }
+ }
+ mount_list_unlock();
+
+ // Unmount the submounts in reverse order. Ignore errors.
+ for (i = m; i > 0; i--) {
+ smp = mount_list_lookupby_fsid(&fsids[i], 0, 1);
+ if (smp) {
+ mount_ref(smp, 0);
+ mount_iterdrop(smp);
+ (void) dounmount(smp, flags, 1, ctx);
+ }
+ }
+out:
+ if (fsids)
+ FREE(fsids, M_TEMP);
+}
+
+void
+mount_dropcrossref(mount_t mp, vnode_t dp, int need_put)
{
- clock_nsec_t nsecs;
+ vnode_lock(dp);
+ mp->mnt_crossref--;
- vfs_iterate(LK_NOWAIT, sync_callback, (void *)0);
+ if (mp->mnt_crossref < 0)
+ panic("mount cross refs -ve");
- {
- static fsid_t fsid = { { 0, 0 } };
-
- clock_get_calendar_microtime(&sync_wait_time, &nsecs);
- vfs_event_signal(&fsid, VQ_SYNCEVENT, (intptr_t)NULL);
- wakeup((caddr_t)&sync_wait_time);
+ if ((mp != dp->v_mountedhere) && (mp->mnt_crossref == 0)) {
+
+ if (need_put)
+ vnode_put_locked(dp);
+ vnode_unlock(dp);
+
+ mount_lock_destroy(mp);
+#if CONFIG_MACF
+ mac_mount_label_destroy(mp);
+#endif
+ FREE_ZONE((caddr_t)mp, sizeof (struct mount), M_MOUNT);
+ return;
}
+ if (need_put)
+ vnode_put_locked(dp);
+ vnode_unlock(dp);
+}
- {
- if(print_vmpage_stat) {
+
+/*
+ * Sync each mounted filesystem.
+ */
+#if DIAGNOSTIC
+int syncprt = 0;
+#endif
+
+int print_vmpage_stat=0;
+int sync_timeout = 60; // Sync time limit (sec)
+
+static int
+sync_callback(mount_t mp, __unused void *arg)
+{
+ if ((mp->mnt_flag & MNT_RDONLY) == 0) {
+ int asyncflag = mp->mnt_flag & MNT_ASYNC;
+
+ mp->mnt_flag &= ~MNT_ASYNC;
+ VFS_SYNC(mp, arg ? MNT_WAIT : MNT_NOWAIT, vfs_context_kernel());
+ if (asyncflag)
+ mp->mnt_flag |= MNT_ASYNC;
+ }
+
+ return (VFS_RETURNED);
+}
+
+/* ARGSUSED */
+int
+sync(__unused proc_t p, __unused struct sync_args *uap, __unused int32_t *retval)
+{
+ vfs_iterate(LK_NOWAIT, sync_callback, NULL);
+
+ if (print_vmpage_stat) {
vm_countdirtypages();
}
+
+#if DIAGNOSTIC
+ if (syncprt)
+ vfs_bufstats();
+#endif /* DIAGNOSTIC */
+ return 0;
+}
+
+static void
+sync_thread(void *arg, __unused wait_result_t wr)
+{
+ int *timeout = (int *) arg;
+
+ vfs_iterate(LK_NOWAIT, sync_callback, NULL);
+
+ if (timeout)
+ wakeup((caddr_t) timeout);
+ if (print_vmpage_stat) {
+ vm_countdirtypages();
}
+
#if DIAGNOSTIC
if (syncprt)
vfs_bufstats();
#endif /* DIAGNOSTIC */
- return (0);
}
/*
- * Change filesystem quotas.
+ * Sync in a separate thread so we can time out if it blocks.
*/
-#if QUOTA
-static int quotactl_funneled(proc_t p, struct quotactl_args *uap, int32_t *retval);
-
-int
-quotactl(proc_t p, struct quotactl_args *uap, int32_t *retval)
+static int
+sync_async(int timeout)
{
- boolean_t funnel_state;
+ thread_t thd;
int error;
-
- funnel_state = thread_funnel_set(kernel_flock, TRUE);
- error = quotactl_funneled(p, uap, retval);
- thread_funnel_set(kernel_flock, funnel_state);
- return(error);
+ struct timespec ts = {timeout, 0};
+
+ lck_mtx_lock(sync_mtx_lck);
+ if (kernel_thread_start(sync_thread, &timeout, &thd) != KERN_SUCCESS) {
+ printf("sync_thread failed\n");
+ lck_mtx_unlock(sync_mtx_lck);
+ return (0);
+ }
+
+ error = msleep((caddr_t) &timeout, sync_mtx_lck, (PVFS | PDROP | PCATCH), "sync_thread", &ts);
+ if (error) {
+ printf("sync timed out: %d sec\n", timeout);
+ }
+ thread_deallocate(thd);
+
+ return (0);
}
-static int
-quotactl_funneled(proc_t p, struct quotactl_args *uap, __unused int32_t *retval)
+/*
+ * An in-kernel sync for power management to call.
+ */
+__private_extern__ int
+sync_internal(void)
+{
+ (void) sync_async(sync_timeout);
+
+ return 0;
+} /* end of sync_internal call */
+
+/*
+ * Change filesystem quotas.
+ */
+#if QUOTA
+int
+quotactl(proc_t p, struct quotactl_args *uap, __unused int32_t *retval)
{
struct mount *mp;
int error, quota_cmd, quota_status;
AUDIT_ARG(uid, uap->uid);
AUDIT_ARG(cmd, uap->cmd);
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
+ NDINIT(&nd, LOOKUP, OP_LOOKUP, FOLLOW | AUDITVNPATH1, UIO_USERSPACE,
+ uap->path, ctx);
error = namei(&nd);
if (error)
return (error);
/* uap->arg is a pointer to a dqblk structure we need to copy out to */
if (error == 0) {
if (proc_is64bit(p)) {
- struct user_dqblk my_dqblk64;
+ struct user_dqblk my_dqblk64 = {.dqb_bhardlimit = 0};
munge_dqblk(&my_dqblk, &my_dqblk64, TRUE);
error = copyout((caddr_t)&my_dqblk64, uap->arg, sizeof (my_dqblk64));
}
vfs_context_t ctx = vfs_context_current();
vnode_t vp;
- NDINIT(&nd, LOOKUP, NOTRIGGER | FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_STATFS, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = namei(&nd);
if (error)
nameidone(&nd);
error = vfs_update_vfsstat(mp, ctx, VFS_USER_EVENT);
- vnode_put(vp);
- if (error != 0)
+ if (error != 0) {
+ vnode_put(vp);
return (error);
+ }
error = munge_statfs(mp, sp, uap->buf, NULL, IS_64BIT_PROCESS(p), TRUE);
+ vnode_put(vp);
return (error);
}
sfs.f_type = mp->mnt_vtable->vfc_typenum;
sfs.f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
sfs.f_fssubtype = sfsp->f_fssubtype;
- strlcpy(&sfs.f_fstypename[0], &sfsp->f_fstypename[0], MFSTYPENAMELEN);
+ if (mp->mnt_kern_flag & MNTK_TYPENAME_OVERRIDE) {
+ strlcpy(&sfs.f_fstypename[0], &mp->fstypename_override[0], MFSTYPENAMELEN);
+ } else {
+ strlcpy(&sfs.f_fstypename[0], &sfsp->f_fstypename[0], MFSTYPENAMELEN);
+ }
strlcpy(&sfs.f_mntonname[0], &sfsp->f_mntonname[0], MAXPATHLEN);
strlcpy(&sfs.f_mntfromname[0], &sfsp->f_mntfromname[0], MAXPATHLEN);
vfs_context_t ctxp = vfs_context_current();
vnode_t vp;
- NDINIT(&nd, LOOKUP, NOTRIGGER | FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_STATFS, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctxp);
error = namei(&nd);
if (error)
nameidone(&nd);
error = vfs_update_vfsstat(mp, ctxp, VFS_USER_EVENT);
- vnode_put(vp);
- if (error != 0)
+ if (error != 0) {
+ vnode_put(vp);
return (error);
+ }
error = statfs64_common(mp, sp, uap->buf);
+ vnode_put(vp);
return (error);
}
fstp->sfsp += my_size;
if (fstp->mp) {
+#if CONFIG_MACF
error = mac_mount_label_get(mp, *fstp->mp);
if (error) {
fstp->error = error;
return(VFS_RETURNED_DONE);
}
+#endif
fstp->mp++;
}
}
return (0);
}
+/*
+ * gets the associated vnode with the file descriptor passed.
+ * as input
+ *
+ * INPUT
+ * ctx - vfs context of caller
+ * fd - file descriptor for which vnode is required.
+ * vpp - Pointer to pointer to vnode to be returned.
+ *
+ * The vnode is returned with an iocount so any vnode obtained
+ * by this call needs a vnode_put
+ *
+ */
+static int
+vnode_getfromfd(vfs_context_t ctx, int fd, vnode_t *vpp)
+{
+ int error;
+ vnode_t vp;
+ struct fileproc *fp;
+ proc_t p = vfs_context_proc(ctx);
+
+ *vpp = NULLVP;
+
+ error = fp_getfvp(p, fd, &fp, &vp);
+ if (error)
+ return (error);
+
+ error = vnode_getwithref(vp);
+ if (error) {
+ (void)fp_drop(p, fd, fp, 0);
+ return (error);
+ }
+
+ (void)fp_drop(p, fd, fp, 0);
+ *vpp = vp;
+ return (error);
+}
+
+/*
+ * Wrapper function around namei to start lookup from a directory
+ * specified by a file descriptor ni_dirfd.
+ *
+ * In addition to all the errors returned by namei, this call can
+ * return ENOTDIR if the file descriptor does not refer to a directory.
+ * and EBADF if the file descriptor is not valid.
+ */
+int
+nameiat(struct nameidata *ndp, int dirfd)
+{
+ if ((dirfd != AT_FDCWD) &&
+ !(ndp->ni_flag & NAMEI_CONTLOOKUP) &&
+ !(ndp->ni_cnd.cn_flags & USEDVP)) {
+ int error = 0;
+ char c;
+
+ if (UIO_SEG_IS_USER_SPACE(ndp->ni_segflg)) {
+ error = copyin(ndp->ni_dirp, &c, sizeof(char));
+ if (error)
+ return (error);
+ } else {
+ c = *((char *)(ndp->ni_dirp));
+ }
+
+ if (c != '/') {
+ vnode_t dvp_at;
+
+ error = vnode_getfromfd(ndp->ni_cnd.cn_context, dirfd,
+ &dvp_at);
+ if (error)
+ return (error);
+
+ if (vnode_vtype(dvp_at) != VDIR) {
+ vnode_put(dvp_at);
+ return (ENOTDIR);
+ }
+
+ ndp->ni_dvp = dvp_at;
+ ndp->ni_cnd.cn_flags |= USEDVP;
+ error = namei(ndp);
+ ndp->ni_cnd.cn_flags &= ~USEDVP;
+ vnode_put(dvp_at);
+ return (error);
+ }
+ }
+
+ return (namei(ndp));
+}
+
/*
* Change current working directory to a given file descriptor.
*/
vnode_t tvp;
vfs_context_t ctx = vfs_context_current();
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_CHDIR, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = change_dir(&nd, ctx);
if (error)
if ((error = suser(kauth_cred_get(), &p->p_acflag)))
return (error);
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_CHROOT, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = change_dir(&nd, ctx);
if (error)
return (error);
}
+/*
+ * Free the vnode data (for directories) associated with the file glob.
+ */
+struct fd_vn_data *
+fg_vn_data_alloc(void)
+{
+ struct fd_vn_data *fvdata;
+
+ /* Allocate per fd vnode data */
+ MALLOC(fvdata, struct fd_vn_data *, (sizeof(struct fd_vn_data)),
+ M_FD_VN_DATA, M_WAITOK | M_ZERO);
+ lck_mtx_init(&fvdata->fv_lock, fd_vn_lck_grp, fd_vn_lck_attr);
+ return fvdata;
+}
+
+/*
+ * Free the vnode data (for directories) associated with the file glob.
+ */
+void
+fg_vn_data_free(void *fgvndata)
+{
+ struct fd_vn_data *fvdata = (struct fd_vn_data *)fgvndata;
+
+ if (fvdata->fv_buf)
+ FREE(fvdata->fv_buf, M_FD_DIRBUF);
+ lck_mtx_destroy(&fvdata->fv_lock, fd_vn_lck_grp);
+ FREE(fvdata, M_FD_VN_DATA);
+}
+
/*
* Check permissions, allocate an open file structure,
* and call the device open routine if any.
* XXX Need to implement uid, gid
*/
int
-open1(vfs_context_t ctx, struct nameidata *ndp, int uflags, struct vnode_attr *vap, int32_t *retval)
+open1(vfs_context_t ctx, struct nameidata *ndp, int uflags,
+ struct vnode_attr *vap, fp_allocfn_t fp_zalloc, void *cra,
+ int32_t *retval)
{
proc_t p = vfs_context_proc(ctx);
uthread_t uu = get_bsdthread_info(vfs_context_thread(ctx));
- struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
vnode_t vp;
int flags, oflags;
- struct fileproc *nfp;
int type, indx, error;
struct flock lf;
- int no_controlling_tty = 0;
- int deny_controlling_tty = 0;
- struct session *sessp = SESSION_NULL;
- struct vfs_context context = *vfs_context_current(); /* local copy */
+ struct vfs_context context;
oflags = uflags;
if ((oflags & O_ACCMODE) == O_ACCMODE)
return(EINVAL);
+
flags = FFLAGS(uflags);
+ CLR(flags, FENCRYPTED);
+ CLR(flags, FUNENCRYPTED);
AUDIT_ARG(fflags, oflags);
AUDIT_ARG(mode, vap->va_mode);
- if ( (error = falloc(p, &nfp, &indx, ctx)) ) {
+ if ((error = falloc_withalloc(p,
+ &fp, &indx, ctx, fp_zalloc, cra)) != 0) {
return (error);
}
- fp = nfp;
uu->uu_dupfd = -indx - 1;
- if (!(p->p_flag & P_CONTROLT)) {
- sessp = proc_session(p);
- no_controlling_tty = 1;
- /*
- * If conditions would warrant getting a controlling tty if
- * the device being opened is a tty (see ttyopen in tty.c),
- * but the open flags deny it, set a flag in the session to
- * prevent it.
- */
- if (SESS_LEADER(p, sessp) &&
- sessp->s_ttyvp == NULL &&
- (flags & O_NOCTTY)) {
- session_lock(sessp);
- sessp->s_flags |= S_NOCTTY;
- session_unlock(sessp);
- deny_controlling_tty = 1;
- }
- }
-
if ((error = vn_open_auth(ndp, &flags, vap))) {
if ((error == ENODEV || error == ENXIO) && (uu->uu_dupfd >= 0)){ /* XXX from fdopen */
- if ((error = dupfdopen(fdp, indx, uu->uu_dupfd, flags, error)) == 0) {
+ if ((error = dupfdopen(p->p_fd, indx, uu->uu_dupfd, flags, error)) == 0) {
fp_drop(p, indx, NULL, 0);
*retval = indx;
- if (deny_controlling_tty) {
- session_lock(sessp);
- sessp->s_flags &= ~S_NOCTTY;
- session_unlock(sessp);
- }
- if (sessp != SESSION_NULL)
- session_rele(sessp);
return (0);
}
}
if (error == ERESTART)
error = EINTR;
fp_free(p, indx, fp);
-
- if (deny_controlling_tty) {
- session_lock(sessp);
- sessp->s_flags &= ~S_NOCTTY;
- session_unlock(sessp);
- }
- if (sessp != SESSION_NULL)
- session_rele(sessp);
return (error);
}
uu->uu_dupfd = 0;
vp = ndp->ni_vp;
- fp->f_fglob->fg_flag = flags & (FMASK | O_EVTONLY);
- fp->f_fglob->fg_type = DTYPE_VNODE;
+ fp->f_fglob->fg_flag = flags & (FMASK | O_EVTONLY | FENCRYPTED | FUNENCRYPTED);
fp->f_fglob->fg_ops = &vnops;
fp->f_fglob->fg_data = (caddr_t)vp;
if (error)
goto bad;
#endif
- if ((error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, type, ctx)))
+ if ((error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, type, ctx, NULL)))
goto bad;
fp->f_fglob->fg_flag |= FHASLOCK;
}
goto bad;
/*
- * If the open flags denied the acquisition of a controlling tty,
- * clear the flag in the session structure that prevented the lower
- * level code from assigning one.
+ * For directories we hold some additional information in the fd.
*/
- if (deny_controlling_tty) {
- session_lock(sessp);
- sessp->s_flags &= ~S_NOCTTY;
- session_unlock(sessp);
+ if (vnode_vtype(vp) == VDIR) {
+ fp->f_fglob->fg_vn_data = fg_vn_data_alloc();
+ } else {
+ fp->f_fglob->fg_vn_data = NULL;
}
+ vnode_put(vp);
+
/*
- * If a controlling tty was set by the tty line discipline, then we
- * want to set the vp of the tty into the session structure. We have
- * a race here because we can't get to the vp for the tp in ttyopen,
- * because it's not passed as a parameter in the open path.
+ * The first terminal open (without a O_NOCTTY) by a session leader
+ * results in it being set as the controlling terminal.
*/
- if (no_controlling_tty && (p->p_flag & P_CONTROLT)) {
- vnode_t ttyvp;
- vnode_ref(vp);
- session_lock(sessp);
- ttyvp = sessp->s_ttyvp;
- sessp->s_ttyvp = vp;
- sessp->s_ttyvid = vnode_vid(vp);
- session_unlock(sessp);
- if (ttyvp != NULLVP)
- vnode_rele(ttyvp);
- }
+ if (vnode_istty(vp) && !(p->p_flag & P_CONTROLT) &&
+ !(flags & O_NOCTTY)) {
+ int tmp = 0;
- vnode_put(vp);
+ (void)(*fp->f_fglob->fg_ops->fo_ioctl)(fp, (int)TIOCSCTTY,
+ (caddr_t)&tmp, ctx);
+ }
proc_fdlock(p);
+ if (flags & O_CLOEXEC)
+ *fdflags(p, indx) |= UF_EXCLOSE;
+ if (flags & O_CLOFORK)
+ *fdflags(p, indx) |= UF_FORKCLOSE;
procfdtbl_releasefd(p, indx, NULL);
fp_drop(p, indx, fp, 1);
proc_fdunlock(p);
*retval = indx;
- if (sessp != SESSION_NULL)
- session_rele(sessp);
return (0);
bad:
- if (deny_controlling_tty) {
- session_lock(sessp);
- sessp->s_flags &= ~S_NOCTTY;
- session_unlock(sessp);
- }
- if (sessp != SESSION_NULL)
- session_rele(sessp);
-
- /* Modify local copy (to not damage thread copy) */
+ context = *vfs_context_current();
context.vc_ucred = fp->f_fglob->fg_cred;
+
+ if ((fp->f_fglob->fg_flag & FHASLOCK) &&
+ (FILEGLOB_DTYPE(fp->f_fglob) == DTYPE_VNODE)) {
+ lf.l_whence = SEEK_SET;
+ lf.l_start = 0;
+ lf.l_len = 0;
+ lf.l_type = F_UNLCK;
+
+ (void)VNOP_ADVLOCK(
+ vp, (caddr_t)fp->f_fglob, F_UNLCK, &lf, F_FLOCK, ctx, NULL);
+ }
vn_close(vp, fp->f_fglob->fg_flag, &context);
vnode_put(vp);
fp_free(p, indx, fp);
return (error);
+}
+
+/*
+ * While most of the *at syscall handlers can call nameiat() which
+ * is a wrapper around namei, the use of namei and initialisation
+ * of nameidata are far removed and in different functions - namei
+ * gets called in vn_open_auth for open1. So we'll just do here what
+ * nameiat() does.
+ */
+static int
+open1at(vfs_context_t ctx, struct nameidata *ndp, int uflags,
+ struct vnode_attr *vap, fp_allocfn_t fp_zalloc, void *cra, int32_t *retval,
+ int dirfd)
+{
+ if ((dirfd != AT_FDCWD) && !(ndp->ni_cnd.cn_flags & USEDVP)) {
+ int error;
+ char c;
+
+ if (UIO_SEG_IS_USER_SPACE(ndp->ni_segflg)) {
+ error = copyin(ndp->ni_dirp, &c, sizeof(char));
+ if (error)
+ return (error);
+ } else {
+ c = *((char *)(ndp->ni_dirp));
+ }
+
+ if (c != '/') {
+ vnode_t dvp_at;
+
+ error = vnode_getfromfd(ndp->ni_cnd.cn_context, dirfd,
+ &dvp_at);
+ if (error)
+ return (error);
+
+ if (vnode_vtype(dvp_at) != VDIR) {
+ vnode_put(dvp_at);
+ return (ENOTDIR);
+ }
+ ndp->ni_dvp = dvp_at;
+ ndp->ni_cnd.cn_flags |= USEDVP;
+ error = open1(ctx, ndp, uflags, vap, fp_zalloc, cra,
+ retval);
+ vnode_put(dvp_at);
+ return (error);
+ }
+ }
+
+ return (open1(ctx, ndp, uflags, vap, fp_zalloc, cra, retval));
}
/*
if (xsecdst != NULL)
VATTR_SET(&va, va_acl, &xsecdst->fsec_acl);
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1, UIO_USERSPACE, uap->path, vfs_context_current());
+ NDINIT(&nd, LOOKUP, OP_OPEN, FOLLOW | AUDITVNPATH1, UIO_USERSPACE,
+ uap->path, vfs_context_current());
- ciferror = open1(vfs_context_current(), &nd, uap->flags, &va, retval);
+ ciferror = open1(vfs_context_current(), &nd, uap->flags, &va,
+ fileproc_alloc_init, NULL, retval);
if (xsecdst != NULL)
kauth_filesec_free(xsecdst);
return ciferror;
}
-int
-open(proc_t p, struct open_args *uap, int32_t *retval)
-{
- __pthread_testcancel(1);
- return(open_nocancel(p, (struct open_nocancel_args *)uap, retval));
-}
+/*
+ * Go through the data-protected atomically controlled open (2)
+ *
+ * int open_dprotected_np(user_addr_t path, int flags, int class, int dpflags, int mode)
+ */
+int open_dprotected_np (__unused proc_t p, struct open_dprotected_np_args *uap, int32_t *retval) {
+ int flags = uap->flags;
+ int class = uap->class;
+ int dpflags = uap->dpflags;
-int
-open_nocancel(proc_t p, struct open_nocancel_args *uap, int32_t *retval)
-{
+ /*
+ * Follow the same path as normal open(2)
+ * Look up the item if it exists, and acquire the vnode.
+ */
struct filedesc *fdp = p->p_fd;
struct vnode_attr va;
struct nameidata nd;
int cmode;
-
+ int error;
+
VATTR_INIT(&va);
/* Mask off all but regular access permissions */
cmode = ((uap->mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
VATTR_SET(&va, va_mode, cmode & ACCESSPERMS);
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1, UIO_USERSPACE, uap->path, vfs_context_current());
+ NDINIT(&nd, LOOKUP, OP_OPEN, FOLLOW | AUDITVNPATH1, UIO_USERSPACE,
+ uap->path, vfs_context_current());
- return(open1(vfs_context_current(), &nd, uap->flags, &va, retval));
-}
+ /*
+ * Initialize the extra fields in vnode_attr to pass down our
+ * extra fields.
+ * 1. target cprotect class.
+ * 2. set a flag to mark it as requiring open-raw-encrypted semantics.
+ */
+ if (flags & O_CREAT) {
+ /* lower level kernel code validates that the class is valid before applying it. */
+ if (class != PROTECTION_CLASS_DEFAULT) {
+ /*
+ * PROTECTION_CLASS_DEFAULT implies that we make the class for this
+ * file behave the same as open (2)
+ */
+ VATTR_SET(&va, va_dataprotect_class, class);
+ }
+ }
+
+ if (dpflags & (O_DP_GETRAWENCRYPTED|O_DP_GETRAWUNENCRYPTED)) {
+ if ( flags & (O_RDWR | O_WRONLY)) {
+ /* Not allowed to write raw encrypted bytes */
+ return EINVAL;
+ }
+ if (uap->dpflags & O_DP_GETRAWENCRYPTED) {
+ VATTR_SET(&va, va_dataprotect_flags, VA_DP_RAWENCRYPTED);
+ }
+ if (uap->dpflags & O_DP_GETRAWUNENCRYPTED) {
+ VATTR_SET(&va, va_dataprotect_flags, VA_DP_RAWUNENCRYPTED);
+ }
+ }
+ error = open1(vfs_context_current(), &nd, uap->flags, &va,
+ fileproc_alloc_init, NULL, retval);
-/*
- * Create a special file.
- */
-static int mkfifo1(vfs_context_t ctx, user_addr_t upath, struct vnode_attr *vap);
+ return error;
+}
-int
-mknod(proc_t p, struct mknod_args *uap, __unused int32_t *retval)
+static int
+openat_internal(vfs_context_t ctx, user_addr_t path, int flags, int mode,
+ int fd, enum uio_seg segflg, int *retval)
{
+ struct filedesc *fdp = (vfs_context_proc(ctx))->p_fd;
struct vnode_attr va;
- vfs_context_t ctx = vfs_context_current();
- int error;
- int whiteout = 0;
struct nameidata nd;
- vnode_t vp, dvp;
+ int cmode;
- VATTR_INIT(&va);
+ VATTR_INIT(&va);
+ /* Mask off all but regular access permissions */
+ cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
+ VATTR_SET(&va, va_mode, cmode & ACCESSPERMS);
+
+ NDINIT(&nd, LOOKUP, OP_OPEN, FOLLOW | AUDITVNPATH1,
+ segflg, path, ctx);
+
+ return (open1at(ctx, &nd, flags, &va, fileproc_alloc_init, NULL,
+ retval, fd));
+}
+
+int
+open(proc_t p, struct open_args *uap, int32_t *retval)
+{
+ __pthread_testcancel(1);
+ return(open_nocancel(p, (struct open_nocancel_args *)uap, retval));
+}
+
+int
+open_nocancel(__unused proc_t p, struct open_nocancel_args *uap,
+ int32_t *retval)
+{
+ return (openat_internal(vfs_context_current(), uap->path, uap->flags,
+ uap->mode, AT_FDCWD, UIO_USERSPACE, retval));
+}
+
+int
+openat_nocancel(__unused proc_t p, struct openat_nocancel_args *uap,
+ int32_t *retval)
+{
+ return (openat_internal(vfs_context_current(), uap->path, uap->flags,
+ uap->mode, uap->fd, UIO_USERSPACE, retval));
+}
+
+int
+openat(proc_t p, struct openat_args *uap, int32_t *retval)
+{
+ __pthread_testcancel(1);
+ return(openat_nocancel(p, (struct openat_nocancel_args *)uap, retval));
+}
+
+/*
+ * openbyid_np: open a file given a file system id and a file system object id
+ * the hfs file system object id is an fsobj_id_t {uint32, uint32}
+ * file systems that don't support object ids it is a node id (uint64_t).
+ *
+ * Parameters: p Process requesting the open
+ * uap User argument descriptor (see below)
+ * retval Pointer to an area to receive the
+ * return calue from the system call
+ *
+ * Indirect: uap->path Path to open (same as 'open')
+ *
+ * uap->fsid id of target file system
+ * uap->objid id of target file system object
+ * uap->flags Flags to open (same as 'open')
+ *
+ * Returns: 0 Success
+ * !0 errno value
+ *
+ *
+ * XXX: We should enummerate the possible errno values here, and where
+ * in the code they originated.
+ */
+int
+openbyid_np(__unused proc_t p, struct openbyid_np_args *uap, int *retval)
+{
+ fsid_t fsid;
+ uint64_t objid;
+ int error;
+ char *buf = NULL;
+ int buflen = MAXPATHLEN;
+ int pathlen = 0;
+ vfs_context_t ctx = vfs_context_current();
+
+ if ((error = copyin(uap->fsid, (caddr_t)&fsid, sizeof(fsid)))) {
+ return (error);
+ }
+
+ /*uap->obj is an fsobj_id_t defined as struct {uint32_t, uint32_t} */
+ if ((error = copyin(uap->objid, (caddr_t)&objid, sizeof(uint64_t)))) {
+ return (error);
+ }
+
+ AUDIT_ARG(value32, fsid.val[0]);
+ AUDIT_ARG(value64, objid);
+
+ /*resolve path from fsis, objid*/
+ do {
+ MALLOC(buf, char *, buflen + 1, M_TEMP, M_WAITOK);
+ if (buf == NULL) {
+ return (ENOMEM);
+ }
+
+ error = fsgetpath_internal(
+ ctx, fsid.val[0], objid,
+ buflen, buf, &pathlen);
+
+ if (error) {
+ FREE(buf, M_TEMP);
+ buf = NULL;
+ }
+ } while (error == ENOSPC && (buflen += MAXPATHLEN));
+
+ if (error) {
+ return error;
+ }
+
+ buf[pathlen] = 0;
+
+ error = openat_internal(
+ ctx, (user_addr_t)buf, uap->oflags, 0, AT_FDCWD, UIO_SYSSPACE, retval);
+
+ FREE(buf, M_TEMP);
+
+ return error;
+}
+
+
+/*
+ * Create a special file.
+ */
+static int mkfifo1(vfs_context_t ctx, user_addr_t upath, struct vnode_attr *vap);
+
+int
+mknod(proc_t p, struct mknod_args *uap, __unused int32_t *retval)
+{
+ struct vnode_attr va;
+ vfs_context_t ctx = vfs_context_current();
+ int error;
+ struct nameidata nd;
+ vnode_t vp, dvp;
+
+ VATTR_INIT(&va);
VATTR_SET(&va, va_mode, (uap->mode & ALLPERMS) & ~p->p_fd->fd_cmask);
VATTR_SET(&va, va_rdev, uap->dev);
if ((error = suser(vfs_context_ucred(ctx), &p->p_acflag)))
return (error);
- NDINIT(&nd, CREATE, LOCKPARENT | AUDITVNPATH1,
+ NDINIT(&nd, CREATE, OP_MKNOD, LOCKPARENT | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = namei(&nd);
if (error)
}
switch (uap->mode & S_IFMT) {
- case S_IFMT: /* used by badsect to flag bad sectors */
- VATTR_SET(&va, va_type, VBAD);
- break;
case S_IFCHR:
VATTR_SET(&va, va_type, VCHR);
break;
case S_IFBLK:
VATTR_SET(&va, va_type, VBLK);
break;
- case S_IFWHT:
- whiteout = 1;
- break;
default:
error = EINVAL;
goto out;
}
#if CONFIG_MACF
- if (!whiteout) {
- error = mac_vnode_check_create(ctx,
- nd.ni_dvp, &nd.ni_cnd, &va);
- if (error)
- goto out;
- }
+ error = mac_vnode_check_create(ctx,
+ nd.ni_dvp, &nd.ni_cnd, &va);
+ if (error)
+ goto out;
#endif
if ((error = vnode_authorize(dvp, NULL, KAUTH_VNODE_ADD_FILE, ctx)) != 0)
goto out;
- if (whiteout) {
- error = VNOP_WHITEOUT(dvp, &nd.ni_cnd, CREATE, ctx);
- } else {
- error = vn_create(dvp, &vp, &nd.ni_cnd, &va, 0, ctx);
- }
- if (error)
+ if ((error = vn_create(dvp, &vp, &nd, &va, 0, 0, NULL, ctx)) != 0)
goto out;
if (vp) {
int error;
struct nameidata nd;
- NDINIT(&nd, CREATE, LOCKPARENT | AUDITVNPATH1,
+ NDINIT(&nd, CREATE, OP_MKFIFO, LOCKPARENT | AUDITVNPATH1,
UIO_USERSPACE, upath, ctx);
error = namei(&nd);
if (error)
}
VATTR_SET(vap, va_type, VFIFO);
-#if CONFIG_MACF
- error = mac_vnode_check_create(ctx, nd.ni_dvp,
- &nd.ni_cnd, vap);
- if (error)
+ if ((error = vn_authorize_create(dvp, &nd.ni_cnd, vap, ctx, NULL)) != 0)
goto out;
-#endif
-
- if ((error = vnode_authorize(dvp, NULL, KAUTH_VNODE_ADD_FILE, ctx)) != 0)
- goto out;
-
-
- error = vn_create(dvp, &vp, &nd.ni_cnd, vap, 0, ctx);
+ error = vn_create(dvp, &vp, &nd, vap, 0, 0, NULL, ctx);
out:
/*
* nameidone has to happen before we vnode_put(dvp)
* VNOP_LINK:???
*/
/* ARGSUSED */
-int
-link(__unused proc_t p, struct link_args *uap, __unused int32_t *retval)
+static int
+linkat_internal(vfs_context_t ctx, int fd1, user_addr_t path, int fd2,
+ user_addr_t link, int flag, enum uio_seg segflg)
{
vnode_t vp, dvp, lvp;
struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
+ int follow;
int error;
#if CONFIG_FSE
fse_info finfo;
vp = dvp = lvp = NULLVP;
/* look up the object we are linking to */
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
- error = namei(&nd);
+ follow = (flag & AT_SYMLINK_FOLLOW) ? FOLLOW : NOFOLLOW;
+ NDINIT(&nd, LOOKUP, OP_LOOKUP, AUDITVNPATH1 | follow,
+ segflg, path, ctx);
+
+ error = nameiat(&nd, fd1);
if (error)
return (error);
vp = nd.ni_vp;
}
/* lookup the target node */
+#if CONFIG_TRIGGERS
+ nd.ni_op = OP_LINK;
+#endif
nd.ni_cnd.cn_nameiop = CREATE;
nd.ni_cnd.cn_flags = LOCKPARENT | AUDITVNPATH2 | CN_NBMOUNTLOOK;
- nd.ni_dirp = uap->link;
- error = namei(&nd);
+ nd.ni_dirp = link;
+ error = nameiat(&nd, fd2);
if (error != 0)
goto out;
dvp = nd.ni_dvp;
if (error)
goto out2;
+#if CONFIG_MACF
+ (void)mac_vnode_notify_link(ctx, vp, dvp, &nd.ni_cnd);
+#endif
+
#if CONFIG_FSE
need_event = need_fsevent(FSE_CREATE_FILE, dvp);
#else
}
link_name_len = MAXPATHLEN;
- vn_getpath(vp, link_to_path, &link_name_len);
-
- /*
- * Call out to allow 3rd party notification of rename.
- * Ignore result of kauth_authorize_fileop call.
- */
- kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_LINK,
- (uintptr_t)link_to_path, (uintptr_t)target_path);
+ if (vn_getpath(vp, link_to_path, &link_name_len) == 0) {
+ /*
+ * Call out to allow 3rd party notification of rename.
+ * Ignore result of kauth_authorize_fileop call.
+ */
+ kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_LINK,
+ (uintptr_t)link_to_path,
+ (uintptr_t)target_path);
+ }
if (link_to_path != NULL) {
RELEASE_PATH(link_to_path);
}
return (error);
}
+int
+link(__unused proc_t p, struct link_args *uap, __unused int32_t *retval)
+{
+ return (linkat_internal(vfs_context_current(), AT_FDCWD, uap->path,
+ AT_FDCWD, uap->link, AT_SYMLINK_FOLLOW, UIO_USERSPACE));
+}
+
+int
+linkat(__unused proc_t p, struct linkat_args *uap, __unused int32_t *retval)
+{
+ if (uap->flag & ~AT_SYMLINK_FOLLOW)
+ return (EINVAL);
+
+ return (linkat_internal(vfs_context_current(), uap->fd1, uap->path,
+ uap->fd2, uap->link, uap->flag, UIO_USERSPACE));
+}
+
/*
* Make a symbolic link.
*
* We could add support for ACLs here too...
*/
/* ARGSUSED */
-int
-symlink(proc_t p, struct symlink_args *uap, __unused int32_t *retval)
+static int
+symlinkat_internal(vfs_context_t ctx, user_addr_t path_data, int fd,
+ user_addr_t link, enum uio_seg segflg)
{
struct vnode_attr va;
char *path;
int error;
struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
vnode_t vp, dvp;
+ uint32_t dfflags; // Directory file flags
size_t dummy=0;
-
- MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
- error = copyinstr(uap->path, path, MAXPATHLEN, &dummy);
+ proc_t p;
+
+ error = 0;
+ if (UIO_SEG_IS_USER_SPACE(segflg)) {
+ MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
+ error = copyinstr(path_data, path, MAXPATHLEN, &dummy);
+ } else {
+ path = (char *)path_data;
+ }
if (error)
goto out;
AUDIT_ARG(text, path); /* This is the link string */
- NDINIT(&nd, CREATE, LOCKPARENT | AUDITVNPATH1,
- UIO_USERSPACE, uap->link, ctx);
- error = namei(&nd);
+ NDINIT(&nd, CREATE, OP_SYMLINK, LOCKPARENT | AUDITVNPATH1,
+ segflg, link, ctx);
+
+ error = nameiat(&nd, fd);
if (error)
goto out;
dvp = nd.ni_dvp;
vp = nd.ni_vp;
+ p = vfs_context_proc(ctx);
VATTR_INIT(&va);
VATTR_SET(&va, va_type, VLNK);
VATTR_SET(&va, va_mode, ACCESSPERMS & ~p->p_fd->fd_cmask);
+
+ /*
+ * Handle inheritance of restricted flag
+ */
+ error = vnode_flags(dvp, &dfflags, ctx);
+ if (error)
+ goto skipit;
+ if (dfflags & SF_RESTRICTED)
+ VATTR_SET(&va, va_flags, SF_RESTRICTED);
+
#if CONFIG_MACF
error = mac_vnode_check_create(ctx,
dvp, &nd.ni_cnd, &va);
if (error == 0)
error = VNOP_SYMLINK(dvp, &vp, &nd.ni_cnd, &va, path, ctx);
+#if CONFIG_MACF
+ if (error == 0 && vp)
+ error = vnode_label(vnode_mount(vp), dvp, vp, &nd.ni_cnd, VNODE_LABEL_CREATE, ctx);
+#endif
+
/* do fallback attribute handling */
- if (error == 0)
+ if (error == 0 && vp)
error = vnode_setattr_fallback(vp, &va, ctx);
-
+
if (error == 0) {
int update_flags = 0;
+ /*check if a new vnode was created, else try to get one*/
if (vp == NULL) {
nd.ni_cnd.cn_nameiop = LOOKUP;
+#if CONFIG_TRIGGERS
+ nd.ni_op = OP_LOOKUP;
+#endif
nd.ni_cnd.cn_flags = 0;
- error = namei(&nd);
+ error = nameiat(&nd, fd);
vp = nd.ni_vp;
if (vp == NULL)
goto skipit;
}
-
+
#if 0 /* XXX - kauth_todo - is KAUTH_FILEOP_SYMLINK needed? */
- /* call out to allow 3rd party notification of rename.
+ /* call out to allow 3rd party notification of rename.
* Ignore result of kauth_authorize_fileop call.
*/
if (kauth_authorize_fileop_has_listeners() &&
namei(&nd) == 0) {
char *new_link_path = NULL;
int len;
-
+
/* build the path to the new link file */
new_link_path = get_pathbuff();
len = MAXPATHLEN;
new_link_path[len - 1] = '/';
strlcpy(&new_link_path[len], nd.ni_cnd.cn_nameptr, MAXPATHLEN-len);
}
-
- kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_SYMLINK,
+
+ kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_SYMLINK,
(uintptr_t)path, (uintptr_t)new_link_path);
if (new_link_path != NULL)
release_pathbuff(new_link_path);
}
-#endif
+#endif
// Make sure the name & parent pointers are hooked up
if (vp->v_name == NULL)
update_flags |= VNODE_UPDATE_NAME;
if (vp->v_parent == NULLVP)
update_flags |= VNODE_UPDATE_PARENT;
-
+
if (update_flags)
vnode_update_identity(vp, dvp, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen, nd.ni_cnd.cn_hash, update_flags);
vnode_put(vp);
vnode_put(dvp);
out:
- FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
+ if (path && (path != (char *)path_data))
+ FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
return (error);
}
+int
+symlink(__unused proc_t p, struct symlink_args *uap, __unused int32_t *retval)
+{
+ return (symlinkat_internal(vfs_context_current(), uap->path, AT_FDCWD,
+ uap->link, UIO_USERSPACE));
+}
+
+int
+symlinkat(__unused proc_t p, struct symlinkat_args *uap,
+ __unused int32_t *retval)
+{
+ return (symlinkat_internal(vfs_context_current(), uap->path1, uap->fd,
+ uap->path2, UIO_USERSPACE));
+}
+
/*
* Delete a whiteout from the filesystem.
- * XXX authorization not implmented for whiteouts
+ * No longer supported.
*/
int
-undelete(__unused proc_t p, struct undelete_args *uap, __unused int32_t *retval)
+undelete(__unused proc_t p, __unused struct undelete_args *uap, __unused int32_t *retval)
{
- int error;
- struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
- vnode_t vp, dvp;
-
- NDINIT(&nd, DELETE, LOCKPARENT|DOWHITEOUT|AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
- error = namei(&nd);
- if (error)
- return (error);
- dvp = nd.ni_dvp;
- vp = nd.ni_vp;
-
- if (vp == NULLVP && (nd.ni_cnd.cn_flags & ISWHITEOUT)) {
- error = VNOP_WHITEOUT(dvp, &nd.ni_cnd, DELETE, ctx);
- } else
- error = EEXIST;
-
- /*
- * nameidone has to happen before we vnode_put(dvp)
- * since it may need to release the fs_nodelock on the dvp
- */
- nameidone(&nd);
-
- if (vp)
- vnode_put(vp);
- vnode_put(dvp);
-
- return (error);
+ return (ENOTSUP);
}
-
/*
* Delete a name from the filesystem.
*/
/* ARGSUSED */
-int
-unlink1(vfs_context_t ctx, struct nameidata *ndp, int nodelbusy)
+static int
+unlinkat_internal(vfs_context_t ctx, int fd, vnode_t start_dvp,
+ user_addr_t path_arg, enum uio_seg segflg, int unlink_flags)
{
+ struct nameidata nd;
vnode_t vp, dvp;
int error;
struct componentname *cnp;
int len=0;
#if CONFIG_FSE
fse_info finfo;
+ struct vnode_attr va;
#endif
- int flags = 0;
- int need_event = 0;
- int has_listeners = 0;
- int truncated_path=0;
+ int flags;
+ int need_event;
+ int has_listeners;
+ int truncated_path;
+ int batched;
+ struct vnode_attr *vap;
+ int do_retry;
+ int retry_count = 0;
+ int cn_flags;
+
+ cn_flags = LOCKPARENT;
+ if (!(unlink_flags & VNODE_REMOVE_NO_AUDIT_PATH))
+ cn_flags |= AUDITVNPATH1;
+ /* If a starting dvp is passed, it trumps any fd passed. */
+ if (start_dvp)
+ cn_flags |= USEDVP;
+
#if NAMEDRSRCFORK
/* unlink or delete is allowed on rsrc forks and named streams */
- ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
+ cn_flags |= CN_ALLOWRSRCFORK;
#endif
- ndp->ni_cnd.cn_flags |= LOCKPARENT;
- cnp = &ndp->ni_cnd;
+retry:
+ do_retry = 0;
+ flags = 0;
+ need_event = 0;
+ has_listeners = 0;
+ truncated_path = 0;
+ vap = NULL;
+
+ NDINIT(&nd, DELETE, OP_UNLINK, cn_flags, segflg, path_arg, ctx);
+
+ nd.ni_dvp = start_dvp;
+ nd.ni_flag |= NAMEI_COMPOUNDREMOVE;
+ cnp = &nd.ni_cnd;
- error = namei(ndp);
+lookup_continue:
+ error = nameiat(&nd, fd);
if (error)
return (error);
- dvp = ndp->ni_dvp;
- vp = ndp->ni_vp;
+ dvp = nd.ni_dvp;
+ vp = nd.ni_vp;
+
/* With Carbon delete semantics, busy files cannot be deleted */
- if (nodelbusy) {
+ if (unlink_flags & VNODE_REMOVE_NODELETEBUSY) {
flags |= VNODE_REMOVE_NODELETEBUSY;
}
-
- /*
- * Normally, unlinking of directories is not supported.
- * However, some file systems may have limited support.
- */
- if ((vp->v_type == VDIR) &&
- !(vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSDIRLINKS)) {
- error = EPERM; /* POSIX */
- }
-
- /*
- * The root of a mounted filesystem cannot be deleted.
- */
- if (vp->v_flag & VROOT) {
- error = EBUSY;
+
+ /* Skip any potential upcalls if told to. */
+ if (unlink_flags & VNODE_REMOVE_SKIP_NAMESPACE_EVENT) {
+ flags |= VNODE_REMOVE_SKIP_NAMESPACE_EVENT;
}
- if (error)
- goto out;
+ if (vp) {
+ batched = vnode_compound_remove_available(vp);
+ /*
+ * The root of a mounted filesystem cannot be deleted.
+ */
+ if (vp->v_flag & VROOT) {
+ error = EBUSY;
+ }
- /* authorize the delete operation */
-#if CONFIG_MACF
- if (!error)
- error = mac_vnode_check_unlink(ctx,
- dvp, vp, cnp);
-#endif /* MAC */
- if (!error)
- error = vnode_authorize(vp, ndp->ni_dvp, KAUTH_VNODE_DELETE, ctx);
- if (error)
- goto out;
-
-#if CONFIG_FSE
- need_event = need_fsevent(FSE_DELETE, dvp);
- if (need_event) {
- if ((vp->v_flag & VISHARDLINK) == 0) {
- get_fse_info(vp, &finfo, ctx);
+ if (!batched) {
+ error = vn_authorize_unlink(dvp, vp, cnp, ctx, NULL);
+ if (error) {
+ if (error == ENOENT) {
+ assert(retry_count < MAX_AUTHORIZE_ENOENT_RETRIES);
+ if (retry_count < MAX_AUTHORIZE_ENOENT_RETRIES) {
+ do_retry = 1;
+ retry_count++;
+ }
+ }
+ goto out;
+ }
+ }
+ } else {
+ batched = 1;
+
+ if (!vnode_compound_remove_available(dvp)) {
+ panic("No vp, but no compound remove?");
+ }
+ }
+
+#if CONFIG_FSE
+ need_event = need_fsevent(FSE_DELETE, dvp);
+ if (need_event) {
+ if (!batched) {
+ if ((vp->v_flag & VISHARDLINK) == 0) {
+ /* XXX need to get these data in batched VNOP */
+ get_fse_info(vp, &finfo, ctx);
+ }
+ } else {
+ error = vfs_get_notify_attributes(&va);
+ if (error) {
+ goto out;
+ }
+
+ vap = &va;
}
}
#endif
has_listeners = kauth_authorize_fileop_has_listeners();
if (need_event || has_listeners) {
- GET_PATH(path);
if (path == NULL) {
- error = ENOMEM;
- goto out;
+ GET_PATH(path);
+ if (path == NULL) {
+ error = ENOMEM;
+ goto out;
+ }
}
-
- len = safe_getpath(dvp, ndp->ni_cnd.cn_nameptr, path, MAXPATHLEN, &truncated_path);
+ len = safe_getpath(dvp, nd.ni_cnd.cn_nameptr, path, MAXPATHLEN, &truncated_path);
}
#if NAMEDRSRCFORK
- if (ndp->ni_cnd.cn_flags & CN_WANTSRSRCFORK)
+ if (nd.ni_cnd.cn_flags & CN_WANTSRSRCFORK)
error = vnode_removenamedstream(dvp, vp, XATTR_RESOURCEFORK_NAME, 0, ctx);
else
#endif
- error = VNOP_REMOVE(dvp, vp, &ndp->ni_cnd, flags, ctx);
+ {
+ error = vn_remove(dvp, &nd.ni_vp, &nd, flags, vap, ctx);
+ vp = nd.ni_vp;
+ if (error == EKEEPLOOKING) {
+ if (!batched) {
+ panic("EKEEPLOOKING, but not a filesystem that supports compound VNOPs?");
+ }
+
+ if ((nd.ni_flag & NAMEI_CONTLOOKUP) == 0) {
+ panic("EKEEPLOOKING, but continue flag not set?");
+ }
+
+ if (vnode_isdir(vp)) {
+ error = EISDIR;
+ goto out;
+ }
+ goto lookup_continue;
+ } else if (error == ENOENT && batched) {
+ assert(retry_count < MAX_AUTHORIZE_ENOENT_RETRIES);
+ if (retry_count < MAX_AUTHORIZE_ENOENT_RETRIES) {
+ /*
+ * For compound VNOPs, the authorization callback may
+ * return ENOENT in case of racing hardlink lookups
+ * hitting the name cache, redrive the lookup.
+ */
+ do_retry = 1;
+ retry_count += 1;
+ goto out;
+ }
+ }
+ }
/*
* Call out to allow 3rd party notification of delete.
if (need_event) {
if (vp->v_flag & VISHARDLINK) {
get_fse_info(vp, &finfo, ctx);
+ } else if (vap) {
+ vnode_get_fse_info_from_vap(vp, &finfo, vap);
}
if (truncated_path) {
finfo.mode |= FSE_TRUNCATED_PATH;
}
#endif
}
+
+out:
if (path != NULL)
RELEASE_PATH(path);
- /*
- * nameidone has to happen before we vnode_put(dvp)
- * since it may need to release the fs_nodelock on the dvp
- */
-out:
#if NAMEDRSRCFORK
/* recycle the deleted rsrc fork vnode to force a reclaim, which
* will cause its shadow file to go away if necessary.
*/
- if ((vnode_isnamedstream(ndp->ni_vp)) &&
- (ndp->ni_vp->v_parent != NULLVP) &&
- vnode_isshadow(ndp->ni_vp)) {
- vnode_recycle(ndp->ni_vp);
+ if (vp && (vnode_isnamedstream(vp)) &&
+ (vp->v_parent != NULLVP) &&
+ vnode_isshadow(vp)) {
+ vnode_recycle(vp);
}
#endif
- nameidone(ndp);
+ /*
+ * nameidone has to happen before we vnode_put(dvp)
+ * since it may need to release the fs_nodelock on the dvp
+ */
+ nameidone(&nd);
vnode_put(dvp);
- vnode_put(vp);
+ if (vp) {
+ vnode_put(vp);
+ }
+
+ if (do_retry) {
+ goto retry;
+ }
+
return (error);
}
-/*
- * Delete a name from the filesystem using POSIX semantics.
- */
int
-unlink(__unused proc_t p, struct unlink_args *uap, __unused int32_t *retval)
+unlink1(vfs_context_t ctx, vnode_t start_dvp, user_addr_t path_arg,
+ enum uio_seg segflg, int unlink_flags)
{
- struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
-
- NDINIT(&nd, DELETE, AUDITVNPATH1, UIO_USERSPACE, uap->path, ctx);
- return unlink1(ctx, &nd, 0);
+ return (unlinkat_internal(ctx, AT_FDCWD, start_dvp, path_arg, segflg,
+ unlink_flags));
}
/*
int
delete(__unused proc_t p, struct delete_args *uap, __unused int32_t *retval)
{
- struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
+ return (unlinkat_internal(vfs_context_current(), AT_FDCWD, NULLVP,
+ uap->path, UIO_USERSPACE, VNODE_REMOVE_NODELETEBUSY));
+}
+
+/*
+ * Delete a name from the filesystem using POSIX semantics.
+ */
+int
+unlink(__unused proc_t p, struct unlink_args *uap, __unused int32_t *retval)
+{
+ return (unlinkat_internal(vfs_context_current(), AT_FDCWD, NULLVP,
+ uap->path, UIO_USERSPACE, 0));
+}
+
+int
+unlinkat(__unused proc_t p, struct unlinkat_args *uap, __unused int32_t *retval)
+{
+ if (uap->flag & ~AT_REMOVEDIR)
+ return (EINVAL);
- NDINIT(&nd, DELETE, AUDITVNPATH1, UIO_USERSPACE, uap->path, ctx);
- return unlink1(ctx, &nd, 1);
+ if (uap->flag & AT_REMOVEDIR)
+ return (rmdirat_internal(vfs_context_current(), uap->fd,
+ uap->path, UIO_USERSPACE));
+ else
+ return (unlinkat_internal(vfs_context_current(), uap->fd,
+ NULLVP, uap->path, UIO_USERSPACE, 0));
}
/*
niopts |= WANTPARENT;
/* do the lookup */
- NDINIT(&nd, LOOKUP, niopts, UIO_SYSSPACE, CAST_USER_ADDR_T(((const char *)input) + input[i].ad_name_offset), &context);
+ NDINIT(&nd, LOOKUP, OP_ACCESS, niopts, UIO_SYSSPACE,
+ CAST_USER_ADDR_T(((const char *)input) + input[i].ad_name_offset),
+ &context);
error = namei(&nd);
if (!error) {
vp = nd.ni_vp;
* namei:???
* access1:
*/
-int
-access(__unused proc_t p, struct access_args *uap, __unused int32_t *retval)
+static int
+faccessat_internal(vfs_context_t ctx, int fd, user_addr_t path, int amode,
+ int flag, enum uio_seg segflg)
{
int error;
struct nameidata nd;
#endif
/*
- * Access is defined as checking against the process'
- * real identity, even if operations are checking the
- * effective identity. So we need to tweak the credential
- * in the context.
+ * Unless the AT_EACCESS option is used, Access is defined as checking
+ * against the process' real identity, even if operations are checking
+ * the effective identity. So we need to tweak the credential
+ * in the context for that case.
*/
- context.vc_ucred = kauth_cred_copy_real(kauth_cred_get());
- context.vc_thread = current_thread();
+ if (!(flag & AT_EACCESS))
+ context.vc_ucred = kauth_cred_copy_real(kauth_cred_get());
+ else
+ context.vc_ucred = ctx->vc_ucred;
+ context.vc_thread = ctx->vc_thread;
+
niopts = FOLLOW | AUDITVNPATH1;
/* need parent for vnode_authorize for deletion test */
- if (uap->flags & _DELETE_OK)
+ if (amode & _DELETE_OK)
niopts |= WANTPARENT;
- NDINIT(&nd, LOOKUP, niopts, UIO_USERSPACE, uap->path, &context);
+ NDINIT(&nd, LOOKUP, OP_ACCESS, niopts, segflg,
+ path, &context);
#if NAMEDRSRCFORK
/* access(F_OK) calls are allowed for resource forks. */
- if (uap->flags == F_OK)
+ if (amode == F_OK)
nd.ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
#endif
- error = namei(&nd);
+ error = nameiat(&nd, fd);
if (error)
goto out;
}
#endif
- error = access1(nd.ni_vp, nd.ni_dvp, uap->flags, &context);
+ error = access1(nd.ni_vp, nd.ni_dvp, amode, &context);
#if NAMEDRSRCFORK
if (is_namedstream) {
#endif
vnode_put(nd.ni_vp);
- if (uap->flags & _DELETE_OK)
+ if (amode & _DELETE_OK)
vnode_put(nd.ni_dvp);
nameidone(&nd);
out:
- kauth_cred_unref(&context.vc_ucred);
- return(error);
+ if (!(flag & AT_EACCESS))
+ kauth_cred_unref(&context.vc_ucred);
+ return (error);
+}
+
+int
+access(__unused proc_t p, struct access_args *uap, __unused int32_t *retval)
+{
+ return (faccessat_internal(vfs_context_current(), AT_FDCWD,
+ uap->path, uap->flags, 0, UIO_USERSPACE));
}
+int
+faccessat(__unused proc_t p, struct faccessat_args *uap,
+ __unused int32_t *retval)
+{
+ if (uap->flag & ~AT_EACCESS)
+ return (EINVAL);
+
+ return (faccessat_internal(vfs_context_current(), uap->fd,
+ uap->path, uap->amode, uap->flag, UIO_USERSPACE));
+}
/*
* Returns: 0 Success
* vn_stat:???
*/
static int
-stat2(vfs_context_t ctx, struct nameidata *ndp, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64)
+fstatat_internal(vfs_context_t ctx, user_addr_t path, user_addr_t ub,
+ user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64,
+ enum uio_seg segflg, int fd, int flag)
{
+ struct nameidata nd;
+ int follow;
union {
struct stat sb;
struct stat64 sb64;
size_t xsecurity_bufsize;
void * statptr;
+ follow = (flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW;
+ NDINIT(&nd, LOOKUP, OP_GETATTR, follow | AUDITVNPATH1,
+ segflg, path, ctx);
+
#if NAMEDRSRCFORK
int is_namedstream = 0;
/* stat calls are allowed for resource forks. */
- ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
+ nd.ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
#endif
- error = namei(ndp);
+ error = nameiat(&nd, fd);
if (error)
return (error);
fsec = KAUTH_FILESEC_NONE;
* force an inactive on release which will mark it
* for recycle.
*/
- if (vnode_isnamedstream(ndp->ni_vp) &&
- (ndp->ni_vp->v_parent != NULLVP) &&
- vnode_isshadow(ndp->ni_vp)) {
+ if (vnode_isnamedstream(nd.ni_vp) &&
+ (nd.ni_vp->v_parent != NULLVP) &&
+ vnode_isshadow(nd.ni_vp)) {
is_namedstream = 1;
- vnode_ref(ndp->ni_vp);
+ vnode_ref(nd.ni_vp);
}
#endif
- error = vn_stat(ndp->ni_vp, statptr, (xsecurity != USER_ADDR_NULL ? &fsec : NULL), isstat64, ctx);
+ error = vn_stat(nd.ni_vp, statptr, (xsecurity != USER_ADDR_NULL ? &fsec : NULL), isstat64, ctx);
#if NAMEDRSRCFORK
if (is_namedstream) {
- vnode_rele(ndp->ni_vp);
+ vnode_rele(nd.ni_vp);
}
#endif
- vnode_put(ndp->ni_vp);
- nameidone(ndp);
+ vnode_put(nd.ni_vp);
+ nameidone(&nd);
if (error)
return (error);
return (error);
}
-/*
- * Get file status; this version follows links.
- *
- * Returns: 0 Success
- * stat2:??? [see stat2() in this file]
- */
-static int
-stat1(user_addr_t path, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64)
-{
- struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
-
- NDINIT(&nd, LOOKUP, NOTRIGGER | FOLLOW | AUDITVNPATH1,
- UIO_USERSPACE, path, ctx);
- return(stat2(ctx, &nd, ub, xsecurity, xsecurity_size, isstat64));
-}
-
/*
* stat_extended: Get file status; with extended security (ACL).
*
*
*/
int
-stat_extended(__unused proc_t p, struct stat_extended_args *uap, __unused int32_t *retval)
+stat_extended(__unused proc_t p, struct stat_extended_args *uap,
+ __unused int32_t *retval)
{
- return (stat1(uap->path, uap->ub, uap->xsecurity, uap->xsecurity_size, 0));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ uap->xsecurity, uap->xsecurity_size, 0, UIO_USERSPACE, AT_FDCWD,
+ 0));
}
/*
* Returns: 0 Success
- * stat1:??? [see stat1() in this file]
+ * fstatat_internal:??? [see fstatat_internal() in this file]
*/
int
stat(__unused proc_t p, struct stat_args *uap, __unused int32_t *retval)
{
- return(stat1(uap->path, uap->ub, 0, 0, 0));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ 0, 0, 0, UIO_USERSPACE, AT_FDCWD, 0));
}
int
stat64(__unused proc_t p, struct stat64_args *uap, __unused int32_t *retval)
{
- return(stat1(uap->path, uap->ub, 0, 0, 1));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ 0, 0, 1, UIO_USERSPACE, AT_FDCWD, 0));
}
/*
int
stat64_extended(__unused proc_t p, struct stat64_extended_args *uap, __unused int32_t *retval)
{
- return (stat1(uap->path, uap->ub, uap->xsecurity, uap->xsecurity_size, 1));
-}
-/*
- * Get file status; this version does not follow links.
- */
-static int
-lstat1(user_addr_t path, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64)
-{
- struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
-
- NDINIT(&nd, LOOKUP, NOTRIGGER | NOFOLLOW | AUDITVNPATH1,
- UIO_USERSPACE, path, ctx);
-
- return(stat2(ctx, &nd, ub, xsecurity, xsecurity_size, isstat64));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ uap->xsecurity, uap->xsecurity_size, 1, UIO_USERSPACE, AT_FDCWD,
+ 0));
}
/*
int
lstat_extended(__unused proc_t p, struct lstat_extended_args *uap, __unused int32_t *retval)
{
- return (lstat1(uap->path, uap->ub, uap->xsecurity, uap->xsecurity_size, 0));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ uap->xsecurity, uap->xsecurity_size, 0, UIO_USERSPACE, AT_FDCWD,
+ AT_SYMLINK_NOFOLLOW));
}
+/*
+ * Get file status; this version does not follow links.
+ */
int
lstat(__unused proc_t p, struct lstat_args *uap, __unused int32_t *retval)
{
- return(lstat1(uap->path, uap->ub, 0, 0, 0));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ 0, 0, 0, UIO_USERSPACE, AT_FDCWD, AT_SYMLINK_NOFOLLOW));
}
int
lstat64(__unused proc_t p, struct lstat64_args *uap, __unused int32_t *retval)
{
- return(lstat1(uap->path, uap->ub, 0, 0, 1));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ 0, 0, 1, UIO_USERSPACE, AT_FDCWD, AT_SYMLINK_NOFOLLOW));
}
/*
int
lstat64_extended(__unused proc_t p, struct lstat64_extended_args *uap, __unused int32_t *retval)
{
- return (lstat1(uap->path, uap->ub, uap->xsecurity, uap->xsecurity_size, 1));
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ uap->xsecurity, uap->xsecurity_size, 1, UIO_USERSPACE, AT_FDCWD,
+ AT_SYMLINK_NOFOLLOW));
+}
+
+int
+fstatat(__unused proc_t p, struct fstatat_args *uap, __unused int32_t *retval)
+{
+ if (uap->flag & ~AT_SYMLINK_NOFOLLOW)
+ return (EINVAL);
+
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ 0, 0, 0, UIO_USERSPACE, uap->fd, uap->flag));
+}
+
+int
+fstatat64(__unused proc_t p, struct fstatat64_args *uap,
+ __unused int32_t *retval)
+{
+ if (uap->flag & ~AT_SYMLINK_NOFOLLOW)
+ return (EINVAL);
+
+ return (fstatat_internal(vfs_context_current(), uap->path, uap->ub,
+ 0, 0, 1, UIO_USERSPACE, uap->fd, uap->flag));
}
/*
struct nameidata nd;
vfs_context_t ctx = vfs_context_current();
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_PATHCONF, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = namei(&nd);
if (error)
* Return target name of a symbolic link.
*/
/* ARGSUSED */
-int
-readlink(proc_t p, struct readlink_args *uap, int32_t *retval)
+static int
+readlinkat_internal(vfs_context_t ctx, int fd, user_addr_t path,
+ enum uio_seg seg, user_addr_t buf, size_t bufsize, enum uio_seg bufseg,
+ int *retval)
{
vnode_t vp;
uio_t auio;
- int spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
int error;
struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
char uio_buf[ UIO_SIZEOF(1) ];
- NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
- error = namei(&nd);
+ NDINIT(&nd, LOOKUP, OP_READLINK, NOFOLLOW | AUDITVNPATH1,
+ seg, path, ctx);
+
+ error = nameiat(&nd, fd);
if (error)
return (error);
vp = nd.ni_vp;
nameidone(&nd);
- auio = uio_createwithbuffer(1, 0, spacetype, UIO_READ,
- &uio_buf[0], sizeof(uio_buf));
- uio_addiov(auio, uap->buf, uap->count);
- if (vp->v_type != VLNK)
+ auio = uio_createwithbuffer(1, 0, bufseg, UIO_READ,
+ &uio_buf[0], sizeof(uio_buf));
+ uio_addiov(auio, buf, bufsize);
+ if (vp->v_type != VLNK) {
error = EINVAL;
- else {
+ } else {
#if CONFIG_MACF
- error = mac_vnode_check_readlink(ctx,
- vp);
+ error = mac_vnode_check_readlink(ctx, vp);
#endif
if (error == 0)
- error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_DATA, ctx);
+ error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_DATA,
+ ctx);
if (error == 0)
error = VNOP_READLINK(vp, auio, ctx);
}
vnode_put(vp);
- /* Safe: uio_resid() is bounded above by "count", and "count" is an int */
- *retval = uap->count - (int)uio_resid(auio);
+ *retval = bufsize - (int)uio_resid(auio);
return (error);
}
+int
+readlink(proc_t p, struct readlink_args *uap, int32_t *retval)
+{
+ enum uio_seg procseg;
+
+ procseg = IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
+ return (readlinkat_internal(vfs_context_current(), AT_FDCWD,
+ CAST_USER_ADDR_T(uap->path), procseg, CAST_USER_ADDR_T(uap->buf),
+ uap->count, procseg, retval));
+}
+
+int
+readlinkat(proc_t p, struct readlinkat_args *uap, int32_t *retval)
+{
+ enum uio_seg procseg;
+
+ procseg = IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
+ return (readlinkat_internal(vfs_context_current(), uap->fd, uap->path,
+ procseg, uap->buf, uap->bufsize, procseg, retval));
+}
+
/*
* Change file flags.
*/
struct nameidata nd;
AUDIT_ARG(fflags, uap->flags);
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_SETATTR, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = namei(&nd);
if (error)
* translated to EPERM before being returned.
*/
static int
-chmod2(vfs_context_t ctx, vnode_t vp, struct vnode_attr *vap)
+chmod_vnode(vfs_context_t ctx, vnode_t vp, struct vnode_attr *vap)
{
kauth_action_t action;
int error;
#endif
#if CONFIG_MACF
- error = mac_vnode_check_setmode(ctx, vp, (mode_t)vap->va_mode);
- if (error)
+ if (VATTR_IS_ACTIVE(vap, va_mode) &&
+ (error = mac_vnode_check_setmode(ctx, vp, (mode_t)vap->va_mode)) != 0)
return (error);
#endif
*
* Returns: 0 Success
* namei:??? [anything namei can return]
- * chmod2:??? [anything chmod2 can return]
+ * chmod_vnode:??? [anything chmod_vnode can return]
*/
static int
-chmod1(vfs_context_t ctx, user_addr_t path, struct vnode_attr *vap)
+chmodat(vfs_context_t ctx, user_addr_t path, struct vnode_attr *vap,
+ int fd, int flag, enum uio_seg segflg)
{
struct nameidata nd;
- int error;
+ int follow, error;
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
- UIO_USERSPACE, path, ctx);
- if ((error = namei(&nd)))
+ follow = (flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW;
+ NDINIT(&nd, LOOKUP, OP_SETATTR, follow | AUDITVNPATH1,
+ segflg, path, ctx);
+ if ((error = nameiat(&nd, fd)))
return (error);
- error = chmod2(ctx, nd.ni_vp, vap);
+ error = chmod_vnode(ctx, nd.ni_vp, vap);
vnode_put(nd.ni_vp);
nameidone(&nd);
return(error);
KAUTH_DEBUG("CHMOD - setting ACL with %d entries", va.va_acl->acl_entrycount);
}
- error = chmod1(vfs_context_current(), uap->path, &va);
+ error = chmodat(vfs_context_current(), uap->path, &va, AT_FDCWD, 0,
+ UIO_USERSPACE);
if (xsecdst != NULL)
kauth_filesec_free(xsecdst);
/*
* Returns: 0 Success
- * chmod1:??? [anything chmod1 can return]
+ * chmodat:??? [anything chmodat can return]
*/
-int
-chmod(__unused proc_t p, struct chmod_args *uap, __unused int32_t *retval)
+static int
+fchmodat_internal(vfs_context_t ctx, user_addr_t path, int mode, int fd,
+ int flag, enum uio_seg segflg)
{
struct vnode_attr va;
VATTR_INIT(&va);
- VATTR_SET(&va, va_mode, uap->mode & ALLPERMS);
+ VATTR_SET(&va, va_mode, mode & ALLPERMS);
+
+ return (chmodat(ctx, path, &va, fd, flag, segflg));
+}
+
+int
+chmod(__unused proc_t p, struct chmod_args *uap, __unused int32_t *retval)
+{
+ return (fchmodat_internal(vfs_context_current(), uap->path, uap->mode,
+ AT_FDCWD, 0, UIO_USERSPACE));
+}
+
+int
+fchmodat(__unused proc_t p, struct fchmodat_args *uap, __unused int32_t *retval)
+{
+ if (uap->flag & ~AT_SYMLINK_NOFOLLOW)
+ return (EINVAL);
- return(chmod1(vfs_context_current(), uap->path, &va));
+ return (fchmodat_internal(vfs_context_current(), uap->path, uap->mode,
+ uap->fd, uap->flag, UIO_USERSPACE));
}
/*
}
AUDIT_ARG(vnpath, vp, ARG_VNODE1);
- error = chmod2(vfs_context_current(), vp, vap);
+ error = chmod_vnode(vfs_context_current(), vp, vap);
(void)vnode_put(vp);
file_drop(fd);
case USER_ADDR_NULL:
VATTR_SET(&va, va_acl, NULL);
break;
+ case CAST_USER_ADDR_T((void *)1): /* _FILESEC_REMOVE_ACL */
+ VATTR_SET(&va, va_acl, NULL);
+ break;
+ /* not being set */
case CAST_USER_ADDR_T(-1):
break;
default:
*/
/* ARGSUSED */
static int
-chown1(vfs_context_t ctx, struct chown_args *uap, __unused int32_t *retval, int follow)
+fchownat_internal(vfs_context_t ctx, int fd, user_addr_t path, uid_t uid,
+ gid_t gid, int flag, enum uio_seg segflg)
{
vnode_t vp;
struct vnode_attr va;
int error;
struct nameidata nd;
+ int follow;
kauth_action_t action;
- AUDIT_ARG(owner, uap->uid, uap->gid);
+ AUDIT_ARG(owner, uid, gid);
- NDINIT(&nd, LOOKUP, (follow ? FOLLOW : 0) | NOTRIGGER | AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
- error = namei(&nd);
+ follow = (flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW;
+ NDINIT(&nd, LOOKUP, OP_SETATTR, follow | AUDITVNPATH1, segflg,
+ path, ctx);
+ error = nameiat(&nd, fd);
if (error)
return (error);
vp = nd.ni_vp;
nameidone(&nd);
VATTR_INIT(&va);
- if (uap->uid != VNOVAL)
- VATTR_SET(&va, va_uid, uap->uid);
- if (uap->gid != VNOVAL)
- VATTR_SET(&va, va_gid, uap->gid);
+ if (uid != (uid_t)VNOVAL)
+ VATTR_SET(&va, va_uid, uid);
+ if (gid != (gid_t)VNOVAL)
+ VATTR_SET(&va, va_gid, gid);
#if CONFIG_MACF
- error = mac_vnode_check_setowner(ctx, vp, uap->uid, uap->gid);
+ error = mac_vnode_check_setowner(ctx, vp, uid, gid);
if (error)
goto out;
#endif
*/
if (error == EACCES)
error = EPERM;
-
+
vnode_put(vp);
return (error);
}
int
-chown(__unused proc_t p, struct chown_args *uap, int32_t *retval)
+chown(__unused proc_t p, struct chown_args *uap, __unused int32_t *retval)
+{
+ return (fchownat_internal(vfs_context_current(), AT_FDCWD, uap->path,
+ uap->uid, uap->gid, 0, UIO_USERSPACE));
+}
+
+int
+lchown(__unused proc_t p, struct lchown_args *uap, __unused int32_t *retval)
{
- return chown1(vfs_context_current(), uap, retval, 1);
+ return (fchownat_internal(vfs_context_current(), AT_FDCWD, uap->path,
+ uap->owner, uap->group, AT_SYMLINK_NOFOLLOW, UIO_USERSPACE));
}
int
-lchown(__unused proc_t p, struct lchown_args *uap, int32_t *retval)
+fchownat(__unused proc_t p, struct fchownat_args *uap, __unused int32_t *retval)
{
- /* Argument list identical, but machine generated; cast for chown1() */
- return chown1(vfs_context_current(), (struct chown_args *)uap, retval, 0);
+ if (uap->flag & ~AT_SYMLINK_NOFOLLOW)
+ return (EINVAL);
+
+ return (fchownat_internal(vfs_context_current(), uap->fd, uap->path,
+ uap->uid, uap->gid, uap->flag, UIO_USERSPACE));
}
/*
* AUDIT: Needed to change the order of operations to do the
* name lookup first because auditing wants the path.
*/
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_SETATTR, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
error = namei(&nd);
if (error)
if (uap->length < 0)
return(EINVAL);
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
+ NDINIT(&nd, LOOKUP, OP_TRUNCATE, FOLLOW | AUDITVNPATH1,
UIO_USERSPACE, uap->path, ctx);
if ((error = namei(&nd)))
return (error);
return(error);
}
- if (fp->f_fglob->fg_type == DTYPE_PSXSHM) {
+ switch (FILEGLOB_DTYPE(fp->f_fglob)) {
+ case DTYPE_PSXSHM:
error = pshm_truncate(p, fp, uap->fd, uap->length, retval);
goto out;
- }
- if (fp->f_fglob->fg_type != DTYPE_VNODE) {
+ case DTYPE_VNODE:
+ break;
+ default:
error = EINVAL;
goto out;
}
return(EINVAL);
}
- NDINIT(&fromnd, LOOKUP, SAVESTART | AUDITVNPATH1,
+ NDINIT(&fromnd, LOOKUP, OP_COPYFILE, AUDITVNPATH1,
UIO_USERSPACE, uap->from, ctx);
if ((error = namei(&fromnd)))
return (error);
fvp = fromnd.ni_vp;
- NDINIT(&tond, CREATE, LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNPATH2 | CN_NBMOUNTLOOK,
- UIO_USERSPACE, uap->to, ctx);
+ NDINIT(&tond, CREATE, OP_LINK,
+ LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNPATH2 | CN_NBMOUNTLOOK,
+ UIO_USERSPACE, uap->to, ctx);
if ((error = namei(&tond))) {
goto out1;
}
out1:
vnode_put(fvp);
- if (fromnd.ni_startdir)
- vnode_put(fromnd.ni_startdir);
nameidone(&fromnd);
if (error == -1)
* or both not be directories. If target is a directory, it must be empty.
*/
/* ARGSUSED */
-int
-rename(__unused proc_t p, struct rename_args *uap, __unused int32_t *retval)
+static int
+renameat_internal(vfs_context_t ctx, int fromfd, user_addr_t from,
+ int tofd, user_addr_t to, int segflg, vfs_rename_flags_t flags)
{
vnode_t tvp, tdvp;
vnode_t fvp, fdvp;
- struct nameidata fromnd, tond;
- vfs_context_t ctx = vfs_context_current();
+ struct nameidata *fromnd, *tond;
int error;
int do_retry;
+ int retry_count;
int mntrename;
int need_event;
- const char *oname;
+ const char *oname = NULL;
char *from_name = NULL, *to_name = NULL;
int from_len=0, to_len=0;
int holding_mntlock;
mount_t locked_mp = NULL;
- vnode_t oparent;
+ vnode_t oparent = NULLVP;
#if CONFIG_FSE
fse_info from_finfo, to_finfo;
#endif
int from_truncated=0, to_truncated;
-
+ int batched = 0;
+ struct vnode_attr *fvap, *tvap;
+ int continuing = 0;
+ /* carving out a chunk for structs that are too big to be on stack. */
+ struct {
+ struct nameidata from_node, to_node;
+ struct vnode_attr fv_attr, tv_attr;
+ } * __rename_data;
+ MALLOC(__rename_data, void *, sizeof(*__rename_data), M_TEMP, M_WAITOK);
+ fromnd = &__rename_data->from_node;
+ tond = &__rename_data->to_node;
+
holding_mntlock = 0;
- do_retry = 0;
+ do_retry = 0;
+ retry_count = 0;
retry:
fvp = tvp = NULL;
fdvp = tdvp = NULL;
+ fvap = tvap = NULL;
mntrename = FALSE;
- NDINIT(&fromnd, DELETE, WANTPARENT | AUDITVNPATH1, UIO_USERSPACE, uap->from, ctx);
-
- if ( (error = namei(&fromnd)) )
- goto out1;
- fdvp = fromnd.ni_dvp;
- fvp = fromnd.ni_vp;
+ NDINIT(fromnd, DELETE, OP_UNLINK, WANTPARENT | AUDITVNPATH1,
+ segflg, from, ctx);
+ fromnd->ni_flag = NAMEI_COMPOUNDRENAME;
-#if CONFIG_MACF
- error = mac_vnode_check_rename_from(ctx, fdvp, fvp, &fromnd.ni_cnd);
- if (error)
- goto out1;
-#endif
+ NDINIT(tond, RENAME, OP_RENAME, WANTPARENT | AUDITVNPATH2 | CN_NBMOUNTLOOK,
+ segflg, to, ctx);
+ tond->ni_flag = NAMEI_COMPOUNDRENAME;
- NDINIT(&tond, RENAME, WANTPARENT | AUDITVNPATH2 | CN_NBMOUNTLOOK , UIO_USERSPACE, uap->to, ctx);
- if (fvp->v_type == VDIR)
- tond.ni_cnd.cn_flags |= WILLBEDIR;
+continue_lookup:
+ if ((fromnd->ni_flag & NAMEI_CONTLOOKUP) != 0 || !continuing) {
+ if ( (error = nameiat(fromnd, fromfd)) )
+ goto out1;
+ fdvp = fromnd->ni_dvp;
+ fvp = fromnd->ni_vp;
- if ( (error = namei(&tond)) ) {
- /*
- * Translate error code for rename("dir1", "dir2/.").
- */
- if (error == EISDIR && fvp->v_type == VDIR)
- error = EINVAL;
- goto out1;
+ if (fvp && fvp->v_type == VDIR)
+ tond->ni_cnd.cn_flags |= WILLBEDIR;
}
- tdvp = tond.ni_dvp;
- tvp = tond.ni_vp;
-#if CONFIG_MACF
- error = mac_vnode_check_rename_to(ctx,
- tdvp, tvp, fdvp == tdvp, &tond.ni_cnd);
- if (error)
- goto out1;
-#endif
-
- if (tvp != NULL) {
- if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
- error = ENOTDIR;
+ if ((tond->ni_flag & NAMEI_CONTLOOKUP) != 0 || !continuing) {
+ if ( (error = nameiat(tond, tofd)) ) {
+ /*
+ * Translate error code for rename("dir1", "dir2/.").
+ */
+ if (error == EISDIR && fvp->v_type == VDIR)
+ error = EINVAL;
goto out1;
- } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
- error = EISDIR;
+ }
+ tdvp = tond->ni_dvp;
+ tvp = tond->ni_vp;
+ }
+
+ batched = vnode_compound_rename_available(fdvp);
+ if (!fvp) {
+ /*
+ * Claim: this check will never reject a valid rename.
+ * For success, either fvp must be on the same mount as tdvp, or fvp must sit atop a vnode on the same mount as tdvp.
+ * Suppose fdvp and tdvp are not on the same mount.
+ * If fvp is on the same mount as tdvp, then fvp is not on the same mount as fdvp, so fvp is the root of its filesystem. If fvp is the root,
+ * then you can't move it to within another dir on the same mountpoint.
+ * If fvp sits atop a vnode on the same mount as fdvp, then that vnode must be part of the same mount as fdvp, which is a contradiction.
+ *
+ * If this check passes, then we are safe to pass these vnodes to the same FS.
+ */
+ if (fdvp->v_mount != tdvp->v_mount) {
+ error = EXDEV;
goto out1;
}
+ goto skipped_lookup;
}
- if (fvp == tdvp) {
- error = EINVAL;
- goto out1;
+
+ if (!batched) {
+ error = vn_authorize_rename(fdvp, fvp, &fromnd->ni_cnd, tdvp, tvp, &tond->ni_cnd, ctx, NULL);
+ if (error) {
+ if (error == ENOENT) {
+ assert(retry_count < MAX_AUTHORIZE_ENOENT_RETRIES);
+ if (retry_count < MAX_AUTHORIZE_ENOENT_RETRIES) {
+ /*
+ * We encountered a race where after doing the namei, tvp stops
+ * being valid. If so, simply re-drive the rename call from the
+ * top.
+ */
+ do_retry = 1;
+ retry_count += 1;
+ }
+ }
+ goto out1;
+ }
}
+
/*
* If the source and destination are the same (i.e. they're
* links to the same vnode) and the target file system is
* case sensitive, then there is nothing to do.
+ *
+ * XXX Come back to this.
*/
if (fvp == tvp) {
int pathconf_val;
-
+
/*
* Note: if _PC_CASE_SENSITIVE selector isn't supported,
* then assume that this file system is case sensitive.
if (VNOP_PATHCONF(fvp, _PC_CASE_SENSITIVE, &pathconf_val, ctx) != 0 ||
pathconf_val != 0) {
goto out1;
- }
+ }
}
- /*
- * Authorization.
- *
- * If tvp is a directory and not the same as fdvp, or tdvp is not
- * the same as fdvp, the node is moving between directories and we
- * need rights to remove from the old and add to the new.
- *
- * If tvp already exists and is not a directory, we need to be
- * allowed to delete it.
- *
- * Note that we do not inherit when renaming.
- *
- * XXX This needs to be revisited to implement the deferred-inherit bit
- */
- {
- int moving = 0;
-
- error = 0;
- if ((tvp != NULL) && vnode_isdir(tvp)) {
- if (tvp != fdvp)
- moving = 1;
- } else if (tdvp != fdvp) {
- moving = 1;
- }
- /*
- * must have delete rights to remove the old name even in
- * the simple case of fdvp == tdvp.
- *
- * If fvp is a directory, and we are changing it's parent,
- * then we also need rights to rewrite its ".." entry as well.
- */
- if (vnode_isdir(fvp)) {
- if ((error = vnode_authorize(fvp, fdvp, KAUTH_VNODE_DELETE | KAUTH_VNODE_ADD_SUBDIRECTORY, ctx)) != 0)
- goto auth_exit;
- } else {
- if ((error = vnode_authorize(fvp, fdvp, KAUTH_VNODE_DELETE, ctx)) != 0)
- goto auth_exit;
- }
- if (moving) {
- /* moving into tdvp or tvp, must have rights to add */
- if ((error = vnode_authorize(((tvp != NULL) && vnode_isdir(tvp)) ? tvp : tdvp,
- NULL,
- vnode_isdir(fvp) ? KAUTH_VNODE_ADD_SUBDIRECTORY : KAUTH_VNODE_ADD_FILE,
- ctx)) != 0) {
- /*
- * We could encounter a race where after doing the namei, tvp stops
- * being valid. If so, simply re-drive the rename call from the
- * top.
- */
- if (error == ENOENT) {
- do_retry = 1;
- }
- goto auth_exit;
- }
- } else {
- /* node staying in same directory, must be allowed to add new name */
- if ((error = vnode_authorize(fdvp, NULL,
- vnode_isdir(fvp) ? KAUTH_VNODE_ADD_SUBDIRECTORY : KAUTH_VNODE_ADD_FILE, ctx)) != 0)
- goto auth_exit;
- }
- /* overwriting tvp */
- if ((tvp != NULL) && !vnode_isdir(tvp) &&
- ((error = vnode_authorize(tvp, tdvp, KAUTH_VNODE_DELETE, ctx)) != 0)) {
- /*
- * We could encounter a race where after doing the namei, tvp stops
- * being valid. If so, simply re-drive the rename call from the
- * top.
- */
- if (error == ENOENT) {
- do_retry = 1;
- }
- goto auth_exit;
- }
-
- /* XXX more checks? */
-
-auth_exit:
- /* authorization denied */
- if (error != 0)
- goto out1;
- }
/*
* Allow the renaming of mount points.
* - target must not exist
* - target must reside in the same directory as source
* - union mounts cannot be renamed
* - "/" cannot be renamed
+ *
+ * XXX Handle this in VFS after a continued lookup (if we missed
+ * in the cache to start off)
*/
if ((fvp->v_flag & VROOT) &&
(fvp->v_type == VDIR) &&
((fvp->v_mount->mnt_flag & (MNT_UNION | MNT_ROOTFS)) == 0) &&
(fvp->v_mount->mnt_vnodecovered != NULLVP)) {
vnode_t coveredvp;
-
+
/* switch fvp to the covered vnode */
coveredvp = fvp->v_mount->mnt_vnodecovered;
if ( (vnode_getwithref(coveredvp)) ) {
error = EXDEV;
goto out1;
}
- /*
- * Avoid renaming "." and "..".
- */
- if (fvp->v_type == VDIR &&
- ((fdvp == fvp) ||
- (fromnd.ni_cnd.cn_namelen == 1 && fromnd.ni_cnd.cn_nameptr[0] == '.') ||
- ((fromnd.ni_cnd.cn_flags | tond.ni_cnd.cn_flags) & ISDOTDOT)) ) {
- error = EINVAL;
- goto out1;
- }
- /*
- * The following edge case is caught here:
- * (to cannot be a descendent of from)
- *
- * o fdvp
- * /
- * /
- * o fvp
- * \
- * \
- * o tdvp
- * /
- * /
- * o tvp
- */
- if (tdvp->v_parent == fvp) {
- error = EINVAL;
- goto out1;
- }
/*
* If source is the same as the destination (that is the
* source. NOTE: Then the target is unlocked going into vnop_rename,
* so not to cause locking problems. There is a single reference on tvp.
*
- * NOTE - that fvp == tvp also occurs if they are hard linked and
+ * NOTE - that fvp == tvp also occurs if they are hard linked and
* that correct behaviour then is just to return success without doing
* anything.
+ *
+ * XXX filesystem should take care of this itself, perhaps...
*/
if (fvp == tvp && fdvp == tdvp) {
- if (fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
- !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
- fromnd.ni_cnd.cn_namelen)) {
+ if (fromnd->ni_cnd.cn_namelen == tond->ni_cnd.cn_namelen &&
+ !bcmp(fromnd->ni_cnd.cn_nameptr, tond->ni_cnd.cn_nameptr,
+ fromnd->ni_cnd.cn_namelen)) {
goto out1;
}
}
* nameidone has to happen before we vnode_put(tvp)
* since it may need to release the fs_nodelock on the tvp
*/
- nameidone(&tond);
+ nameidone(tond);
if (tvp)
vnode_put(tvp);
* nameidone has to happen before we vnode_put(fdvp)
* since it may need to release the fs_nodelock on the fvp
*/
- nameidone(&fromnd);
+ nameidone(fromnd);
vnode_put(fvp);
vnode_put(fdvp);
} else {
/*
* when we dropped the iocounts to take
- * the lock, we allowed the identity of
+ * the lock, we allowed the identity of
* the various vnodes to change... if they did,
* we may no longer be dealing with a rename
* that reshapes the tree... once we're holding
holding_mntlock = 0;
}
}
+
// save these off so we can later verify that fvp is the same
oname = fvp->v_name;
oparent = fvp->v_parent;
+skipped_lookup:
#if CONFIG_FSE
- need_event = need_fsevent(FSE_RENAME, fvp);
- if (need_event) {
- get_fse_info(fvp, &from_finfo, ctx);
+ need_event = need_fsevent(FSE_RENAME, fdvp);
+ if (need_event) {
+ if (fvp) {
+ get_fse_info(fvp, &from_finfo, ctx);
+ } else {
+ error = vfs_get_notify_attributes(&__rename_data->fv_attr);
+ if (error) {
+ goto out1;
+ }
+
+ fvap = &__rename_data->fv_attr;
+ }
if (tvp) {
get_fse_info(tvp, &to_finfo, ctx);
+ } else if (batched) {
+ error = vfs_get_notify_attributes(&__rename_data->tv_attr);
+ if (error) {
+ goto out1;
+ }
+
+ tvap = &__rename_data->tv_attr;
}
}
#else
#endif /* CONFIG_FSE */
if (need_event || kauth_authorize_fileop_has_listeners()) {
- GET_PATH(from_name);
if (from_name == NULL) {
- error = ENOMEM;
- goto out1;
+ GET_PATH(from_name);
+ if (from_name == NULL) {
+ error = ENOMEM;
+ goto out1;
+ }
}
- from_len = safe_getpath(fdvp, fromnd.ni_cnd.cn_nameptr, from_name, MAXPATHLEN, &from_truncated);
+ from_len = safe_getpath(fdvp, fromnd->ni_cnd.cn_nameptr, from_name, MAXPATHLEN, &from_truncated);
- GET_PATH(to_name);
if (to_name == NULL) {
- error = ENOMEM;
- goto out1;
+ GET_PATH(to_name);
+ if (to_name == NULL) {
+ error = ENOMEM;
+ goto out1;
+ }
}
- to_len = safe_getpath(tdvp, tond.ni_cnd.cn_nameptr, to_name, MAXPATHLEN, &to_truncated);
- }
-
- error = VNOP_RENAME(fdvp, fvp, &fromnd.ni_cnd,
- tdvp, tvp, &tond.ni_cnd,
- ctx);
+ to_len = safe_getpath(tdvp, tond->ni_cnd.cn_nameptr, to_name, MAXPATHLEN, &to_truncated);
+ }
+#if CONFIG_SECLUDED_RENAME
+ if (flags & VFS_SECLUDE_RENAME) {
+ fromnd->ni_cnd.cn_flags |= CN_SECLUDE_RENAME;
+ }
+#else
+ #pragma unused(flags)
+#endif
+ error = vn_rename(fdvp, &fvp, &fromnd->ni_cnd, fvap,
+ tdvp, &tvp, &tond->ni_cnd, tvap,
+ 0, ctx);
if (holding_mntlock) {
/*
holding_mntlock = 0;
}
if (error) {
- /*
- * We may encounter a race in the VNOP where the destination didn't
- * exist when we did the namei, but it does by the time we go and
- * try to create the entry. In this case, we should re-drive this rename
- * call from the top again. Currently, only HFS bubbles out ERECYCLE,
- * but other filesystems susceptible to this race could return it, too.
- */
- if (error == ERECYCLE) {
- do_retry = 1;
- }
+ if (error == EKEEPLOOKING) {
+ if ((fromnd->ni_flag & NAMEI_CONTLOOKUP) == 0) {
+ if ((tond->ni_flag & NAMEI_CONTLOOKUP) == 0) {
+ panic("EKEEPLOOKING without NAMEI_CONTLOOKUP on either ndp?");
+ }
+ }
+
+ fromnd->ni_vp = fvp;
+ tond->ni_vp = tvp;
+
+ goto continue_lookup;
+ }
+
+ /*
+ * We may encounter a race in the VNOP where the destination didn't
+ * exist when we did the namei, but it does by the time we go and
+ * try to create the entry. In this case, we should re-drive this rename
+ * call from the top again. Currently, only HFS bubbles out ERECYCLE,
+ * but other filesystems susceptible to this race could return it, too.
+ */
+ if (error == ERECYCLE) {
+ do_retry = 1;
+ }
+
+ /*
+ * For compound VNOPs, the authorization callback may return
+ * ENOENT in case of racing hardlink lookups hitting the name
+ * cache, redrive the lookup.
+ */
+ if (batched && error == ENOENT) {
+ assert(retry_count < MAX_AUTHORIZE_ENOENT_RETRIES);
+ if (retry_count < MAX_AUTHORIZE_ENOENT_RETRIES) {
+ do_retry = 1;
+ retry_count += 1;
+ }
+ }
goto out1;
- }
-
- /* call out to allow 3rd party notification of rename.
+ }
+
+ /* call out to allow 3rd party notification of rename.
* Ignore result of kauth_authorize_fileop call.
*/
- kauth_authorize_fileop(vfs_context_ucred(ctx),
- KAUTH_FILEOP_RENAME,
+ kauth_authorize_fileop(vfs_context_ucred(ctx),
+ KAUTH_FILEOP_RENAME,
(uintptr_t)from_name, (uintptr_t)to_name);
#if CONFIG_FSE
// set it here since only the from_finfo gets reported up to user space
from_finfo.mode |= FSE_TRUNCATED_PATH;
}
+
+ if (tvap && tvp) {
+ vnode_get_fse_info_from_vap(tvp, &to_finfo, tvap);
+ }
+ if (fvap) {
+ vnode_get_fse_info_from_vap(fvp, &from_finfo, fvap);
+ }
+
if (tvp) {
add_fsevent(FSE_RENAME, ctx,
FSE_ARG_STRING, from_len, from_name,
}
}
#endif /* CONFIG_FSE */
-
+
/*
* update filesystem's mount point data
*/
}
MALLOC_ZONE(tobuf, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
- error = copyinstr(uap->to, tobuf, MAXPATHLEN, &len);
+ if (UIO_SEG_IS_USER_SPACE(segflg))
+ error = copyinstr(to, tobuf, MAXPATHLEN, &len);
+ else
+ error = copystr((void *)to, tobuf, MAXPATHLEN, &len);
if (!error) {
/* find current mount point prefix */
pathend = &mp->mnt_vfsstat.f_mntonname[0];
vfs_unbusy(mp);
}
/*
- * fix up name & parent pointers. note that we first
+ * fix up name & parent pointers. note that we first
* check that fvp has the same name/parent pointers it
* had before the rename call... this is a 'weak' check
* at best...
+ *
+ * XXX oparent and oname may not be set in the compound vnop case
*/
- if (oname == fvp->v_name && oparent == fvp->v_parent) {
+ if (batched || (oname == fvp->v_name && oparent == fvp->v_parent)) {
int update_flags;
update_flags = VNODE_UPDATE_NAME;
if (fdvp != tdvp)
update_flags |= VNODE_UPDATE_PARENT;
- vnode_update_identity(fvp, tdvp, tond.ni_cnd.cn_nameptr, tond.ni_cnd.cn_namelen, tond.ni_cnd.cn_hash, update_flags);
+ vnode_update_identity(fvp, tdvp, tond->ni_cnd.cn_nameptr, tond->ni_cnd.cn_namelen, tond->ni_cnd.cn_hash, update_flags);
}
out1:
if (to_name != NULL) {
* nameidone has to happen before we vnode_put(tdvp)
* since it may need to release the fs_nodelock on the tdvp
*/
- nameidone(&tond);
+ nameidone(tond);
if (tvp)
vnode_put(tvp);
* nameidone has to happen before we vnode_put(fdvp)
* since it may need to release the fs_nodelock on the fdvp
*/
- nameidone(&fromnd);
+ nameidone(fromnd);
if (fvp)
vnode_put(fvp);
vnode_put(fdvp);
}
-
- /*
- * If things changed after we did the namei, then we will re-drive
- * this rename call from the top.
- */
- if(do_retry) {
- do_retry = 0;
+
+ /*
+ * If things changed after we did the namei, then we will re-drive
+ * this rename call from the top.
+ */
+ if (do_retry) {
+ do_retry = 0;
goto retry;
}
-
+
+ FREE(__rename_data, M_TEMP);
return (error);
}
+int
+rename(__unused proc_t p, struct rename_args *uap, __unused int32_t *retval)
+{
+ return (renameat_internal(vfs_context_current(), AT_FDCWD, uap->from,
+ AT_FDCWD, uap->to, UIO_USERSPACE, 0));
+}
+
+#if CONFIG_SECLUDED_RENAME
+int rename_ext(__unused proc_t p, struct rename_ext_args *uap, __unused int32_t *retval)
+{
+ return renameat_internal(
+ vfs_context_current(),
+ AT_FDCWD, uap->from,
+ AT_FDCWD, uap->to,
+ UIO_USERSPACE, uap->flags);
+}
+#endif
+
+int
+renameat(__unused proc_t p, struct renameat_args *uap, __unused int32_t *retval)
+{
+ return (renameat_internal(vfs_context_current(), uap->fromfd, uap->from,
+ uap->tofd, uap->to, UIO_USERSPACE, 0));
+}
+
/*
* Make a directory file.
*
*/
/* ARGSUSED */
static int
-mkdir1(vfs_context_t ctx, user_addr_t path, struct vnode_attr *vap)
+mkdir1at(vfs_context_t ctx, user_addr_t path, struct vnode_attr *vap, int fd,
+ enum uio_seg segflg)
{
vnode_t vp, dvp;
int error;
int update_flags = 0;
+ int batched;
struct nameidata nd;
AUDIT_ARG(mode, vap->va_mode);
- NDINIT(&nd, CREATE, LOCKPARENT | AUDITVNPATH1,
- UIO_USERSPACE, path, ctx);
+ NDINIT(&nd, CREATE, OP_MKDIR, LOCKPARENT | AUDITVNPATH1, segflg,
+ path, ctx);
nd.ni_cnd.cn_flags |= WILLBEDIR;
- error = namei(&nd);
+ nd.ni_flag = NAMEI_COMPOUNDMKDIR;
+
+continue_lookup:
+ error = nameiat(&nd, fd);
if (error)
return (error);
dvp = nd.ni_dvp;
vp = nd.ni_vp;
- if (vp != NULL) {
- error = EEXIST;
- goto out;
- }
+ if (vp != NULL) {
+ error = EEXIST;
+ goto out;
+ }
+
+ batched = vnode_compound_mkdir_available(dvp);
VATTR_SET(vap, va_type, VDIR);
-
-#if CONFIG_MACF
- error = mac_vnode_check_create(ctx,
- nd.ni_dvp, &nd.ni_cnd, vap);
- if (error)
+
+ /*
+ * XXX
+ * Don't authorize in VFS for compound VNOP.... mkdir -p today assumes that it will
+ * only get EXISTS or EISDIR for existing path components, and not that it could see
+ * EACCESS/EPERM--so if we authorize for mkdir on "/" for "mkdir -p /tmp/foo/bar/baz"
+ * it will fail in a spurious manner. Need to figure out if this is valid behavior.
+ */
+ if ((error = vn_authorize_mkdir(dvp, &nd.ni_cnd, vap, ctx, NULL)) != 0) {
+ if (error == EACCES || error == EPERM) {
+ int error2;
+
+ nameidone(&nd);
+ vnode_put(dvp);
+ dvp = NULLVP;
+
+ /*
+ * Try a lookup without "NAMEI_COMPOUNDVNOP" to make sure we return EEXIST
+ * rather than EACCESS if the target exists.
+ */
+ NDINIT(&nd, LOOKUP, OP_MKDIR, AUDITVNPATH1, segflg,
+ path, ctx);
+ error2 = nameiat(&nd, fd);
+ if (error2) {
+ goto out;
+ } else {
+ vp = nd.ni_vp;
+ error = EEXIST;
+ goto out;
+ }
+ }
+
goto out;
-#endif
+ }
+
+ /*
+ * make the directory
+ */
+ if ((error = vn_create(dvp, &vp, &nd, vap, 0, 0, NULL, ctx)) != 0) {
+ if (error == EKEEPLOOKING) {
+ nd.ni_vp = vp;
+ goto continue_lookup;
+ }
+
+ goto out;
+ }
- /* authorize addition of a directory to the parent */
- if ((error = vnode_authorize(dvp, NULL, KAUTH_VNODE_ADD_SUBDIRECTORY, ctx)) != 0)
- goto out;
-
-
- /* make the directory */
- if ((error = vn_create(dvp, &vp, &nd.ni_cnd, vap, 0, ctx)) != 0)
- goto out;
-
// Make sure the name & parent pointers are hooked up
if (vp->v_name == NULL)
update_flags |= VNODE_UPDATE_NAME;
nameidone(&nd);
if (vp)
- vnode_put(vp);
- vnode_put(dvp);
+ vnode_put(vp);
+ if (dvp)
+ vnode_put(dvp);
return (error);
}
*
* Parameters: p Process requesting to create the directory
* uap User argument descriptor (see below)
- * retval (ignored)
+ * retval (ignored)
*
* Indirect: uap->path Path of directory to create
* uap->mode Access permissions to set
* uap->xsecurity ACL to set
- *
+ *
* Returns: 0 Success
* !0 Not success
*
return ciferror;
VATTR_INIT(&va);
- VATTR_SET(&va, va_mode, (uap->mode & ACCESSPERMS) & ~p->p_fd->fd_cmask);
+ VATTR_SET(&va, va_mode, (uap->mode & ACCESSPERMS) & ~p->p_fd->fd_cmask);
if (xsecdst != NULL)
VATTR_SET(&va, va_acl, &xsecdst->fsec_acl);
- ciferror = mkdir1(vfs_context_current(), uap->path, &va);
+ ciferror = mkdir1at(vfs_context_current(), uap->path, &va, AT_FDCWD,
+ UIO_USERSPACE);
if (xsecdst != NULL)
kauth_filesec_free(xsecdst);
return ciferror;
struct vnode_attr va;
VATTR_INIT(&va);
- VATTR_SET(&va, va_mode, (uap->mode & ACCESSPERMS) & ~p->p_fd->fd_cmask);
+ VATTR_SET(&va, va_mode, (uap->mode & ACCESSPERMS) & ~p->p_fd->fd_cmask);
- return(mkdir1(vfs_context_current(), uap->path, &va));
+ return (mkdir1at(vfs_context_current(), uap->path, &va, AT_FDCWD,
+ UIO_USERSPACE));
}
-/*
- * Remove a directory file.
- */
-/* ARGSUSED */
int
-rmdir(__unused proc_t p, struct rmdir_args *uap, __unused int32_t *retval)
+mkdirat(proc_t p, struct mkdirat_args *uap, __unused int32_t *retval)
+{
+ struct vnode_attr va;
+
+ VATTR_INIT(&va);
+ VATTR_SET(&va, va_mode, (uap->mode & ACCESSPERMS) & ~p->p_fd->fd_cmask);
+
+ return(mkdir1at(vfs_context_current(), uap->path, &va, uap->fd,
+ UIO_USERSPACE));
+}
+
+static int
+rmdirat_internal(vfs_context_t ctx, int fd, user_addr_t dirpath,
+ enum uio_seg segflg)
{
vnode_t vp, dvp;
int error;
struct nameidata nd;
- vfs_context_t ctx = vfs_context_current();
+ char *path = NULL;
+ int len=0;
+ int has_listeners = 0;
+ int need_event = 0;
+ int truncated = 0;
+#if CONFIG_FSE
+ struct vnode_attr va;
+#endif /* CONFIG_FSE */
+ struct vnode_attr *vap = NULL;
+ int restart_count = 0;
+ int batched;
int restart_flag;
- uint32_t oldvp_id = UINT32_MAX;
- /*
+ /*
* This loop exists to restart rmdir in the unlikely case that two
* processes are simultaneously trying to remove the same directory
* containing orphaned appleDouble files.
*/
do {
+ NDINIT(&nd, DELETE, OP_RMDIR, LOCKPARENT | AUDITVNPATH1,
+ segflg, dirpath, ctx);
+ nd.ni_flag = NAMEI_COMPOUNDRMDIR;
+continue_lookup:
restart_flag = 0;
+ vap = NULL;
- NDINIT(&nd, DELETE, LOCKPARENT | AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
- error = namei(&nd);
+ error = nameiat(&nd, fd);
if (error)
return (error);
dvp = nd.ni_dvp;
vp = nd.ni_vp;
+ if (vp) {
+ batched = vnode_compound_rmdir_available(vp);
- /*
- * If being restarted check if the new vp
- * still has the same v_id.
- */
- if (oldvp_id != UINT32_MAX && oldvp_id != vp->v_id) {
- error = ENOENT;
- goto out;
- }
+ if (vp->v_flag & VROOT) {
+ /*
+ * The root of a mounted filesystem cannot be deleted.
+ */
+ error = EBUSY;
+ goto out;
+ }
- if (vp->v_type != VDIR) {
- /*
- * rmdir only deals with directories
- */
- error = ENOTDIR;
- } else if (dvp == vp) {
- /*
- * No rmdir "." please.
- */
- error = EINVAL;
- } else if (vp->v_flag & VROOT) {
/*
- * The root of a mounted filesystem cannot be deleted.
+ * Removed a check here; we used to abort if vp's vid
+ * was not the same as what we'd seen the last time around.
+ * I do not think that check was valid, because if we retry
+ * and all dirents are gone, the directory could legitimately
+ * be recycled but still be present in a situation where we would
+ * have had permission to delete. Therefore, we won't make
+ * an effort to preserve that check now that we may not have a
+ * vp here.
*/
- error = EBUSY;
+
+ if (!batched) {
+ error = vn_authorize_rmdir(dvp, vp, &nd.ni_cnd, ctx, NULL);
+ if (error) {
+ if (error == ENOENT) {
+ assert(restart_count < MAX_AUTHORIZE_ENOENT_RETRIES);
+ if (restart_count < MAX_AUTHORIZE_ENOENT_RETRIES) {
+ restart_flag = 1;
+ restart_count += 1;
+ }
+ }
+ goto out;
+ }
+ }
} else {
-#if CONFIG_MACF
- error = mac_vnode_check_unlink(ctx, dvp,
- vp, &nd.ni_cnd);
- if (!error)
-#endif
- error = vnode_authorize(vp, nd.ni_dvp, KAUTH_VNODE_DELETE, ctx);
+ batched = 1;
+
+ if (!vnode_compound_rmdir_available(dvp)) {
+ panic("No error, but no compound rmdir?");
+ }
}
- if (!error) {
- char *path = NULL;
- int len=0;
- int has_listeners = 0;
- int need_event = 0;
- int truncated = 0;
+
#if CONFIG_FSE
- fse_info finfo;
+ fse_info finfo;
- need_event = need_fsevent(FSE_DELETE, dvp);
- if (need_event) {
+ need_event = need_fsevent(FSE_DELETE, dvp);
+ if (need_event) {
+ if (!batched) {
get_fse_info(vp, &finfo, ctx);
+ } else {
+ error = vfs_get_notify_attributes(&va);
+ if (error) {
+ goto out;
+ }
+
+ vap = &va;
}
+ }
#endif
- has_listeners = kauth_authorize_fileop_has_listeners();
- if (need_event || has_listeners) {
+ has_listeners = kauth_authorize_fileop_has_listeners();
+ if (need_event || has_listeners) {
+ if (path == NULL) {
GET_PATH(path);
if (path == NULL) {
error = ENOMEM;
goto out;
}
+ }
- len = safe_getpath(vp, NULL, path, MAXPATHLEN, &truncated);
+ len = safe_getpath(dvp, nd.ni_cnd.cn_nameptr, path, MAXPATHLEN, &truncated);
#if CONFIG_FSE
- if (truncated) {
- finfo.mode |= FSE_TRUNCATED_PATH;
- }
-#endif
+ if (truncated) {
+ finfo.mode |= FSE_TRUNCATED_PATH;
}
+#endif
+ }
- error = VNOP_RMDIR(dvp, vp, &nd.ni_cnd, ctx);
-
- /*
- * Special case to remove orphaned AppleDouble
- * files. I don't like putting this in the kernel,
- * but carbon does not like putting this in carbon either,
- * so here we are.
- */
- if (error == ENOTEMPTY) {
- error = rmdir_remove_orphaned_appleDouble(vp, ctx, &restart_flag);
- if (error == EBUSY) {
- oldvp_id = vp->v_id;
- goto out;
- }
-
+ error = vn_rmdir(dvp, &vp, &nd, vap, ctx);
+ nd.ni_vp = vp;
+ if (vp == NULLVP) {
+ /* Couldn't find a vnode */
+ goto out;
+ }
+ if (error == EKEEPLOOKING) {
+ goto continue_lookup;
+ } else if (batched && error == ENOENT) {
+ assert(restart_count < MAX_AUTHORIZE_ENOENT_RETRIES);
+ if (restart_count < MAX_AUTHORIZE_ENOENT_RETRIES) {
/*
- * Assuming everything went well, we will try the RMDIR again
+ * For compound VNOPs, the authorization callback
+ * may return ENOENT in case of racing hard link lookups
+ * redrive the lookup.
*/
- if (!error)
- error = VNOP_RMDIR(dvp, vp, &nd.ni_cnd, ctx);
+ restart_flag = 1;
+ restart_count += 1;
+ goto out;
+ }
+ }
+#if CONFIG_APPLEDOUBLE
+ /*
+ * Special case to remove orphaned AppleDouble
+ * files. I don't like putting this in the kernel,
+ * but carbon does not like putting this in carbon either,
+ * so here we are.
+ */
+ if (error == ENOTEMPTY) {
+ error = rmdir_remove_orphaned_appleDouble(vp, ctx, &restart_flag);
+ if (error == EBUSY) {
+ goto out;
}
+
/*
- * Call out to allow 3rd party notification of delete.
- * Ignore result of kauth_authorize_fileop call.
+ * Assuming everything went well, we will try the RMDIR again
*/
- if (!error) {
- if (has_listeners) {
- kauth_authorize_fileop(vfs_context_ucred(ctx),
- KAUTH_FILEOP_DELETE,
- (uintptr_t)vp,
- (uintptr_t)path);
- }
+ if (!error)
+ error = vn_rmdir(dvp, &vp, &nd, vap, ctx);
+ }
+#endif /* CONFIG_APPLEDOUBLE */
+ /*
+ * Call out to allow 3rd party notification of delete.
+ * Ignore result of kauth_authorize_fileop call.
+ */
+ if (!error) {
+ if (has_listeners) {
+ kauth_authorize_fileop(vfs_context_ucred(ctx),
+ KAUTH_FILEOP_DELETE,
+ (uintptr_t)vp,
+ (uintptr_t)path);
+ }
- if (vp->v_flag & VISHARDLINK) {
- // see the comment in unlink1() about why we update
- // the parent of a hard link when it is removed
- vnode_update_identity(vp, NULL, NULL, 0, 0, VNODE_UPDATE_PARENT);
- }
+ if (vp->v_flag & VISHARDLINK) {
+ // see the comment in unlink1() about why we update
+ // the parent of a hard link when it is removed
+ vnode_update_identity(vp, NULL, NULL, 0, 0, VNODE_UPDATE_PARENT);
+ }
#if CONFIG_FSE
- if (need_event) {
- add_fsevent(FSE_DELETE, ctx,
- FSE_ARG_STRING, len, path,
- FSE_ARG_FINFO, &finfo,
- FSE_ARG_DONE);
+ if (need_event) {
+ if (vap) {
+ vnode_get_fse_info_from_vap(vp, &finfo, vap);
}
-#endif
+ add_fsevent(FSE_DELETE, ctx,
+ FSE_ARG_STRING, len, path,
+ FSE_ARG_FINFO, &finfo,
+ FSE_ARG_DONE);
}
- if (path != NULL)
- RELEASE_PATH(path);
+#endif
}
out:
+ if (path != NULL) {
+ RELEASE_PATH(path);
+ path = NULL;
+ }
/*
* nameidone has to happen before we vnode_put(dvp)
* since it may need to release the fs_nodelock on the dvp
*/
nameidone(&nd);
-
vnode_put(dvp);
- vnode_put(vp);
+
+ if (vp)
+ vnode_put(vp);
if (restart_flag == 0) {
wakeup_one((caddr_t)vp);
}
-/* Get direntry length padded to 8 byte alignment */
-#define DIRENT64_LEN(namlen) \
- ((sizeof(struct direntry) + (namlen) - (MAXPATHLEN-1) + 7) & ~7)
-
-static errno_t
-vnode_readdir64(struct vnode *vp, struct uio *uio, int flags, int *eofflag,
+/*
+ * Remove a directory file.
+ */
+/* ARGSUSED */
+int
+rmdir(__unused proc_t p, struct rmdir_args *uap, __unused int32_t *retval)
+{
+ return (rmdirat_internal(vfs_context_current(), AT_FDCWD,
+ CAST_USER_ADDR_T(uap->path), UIO_USERSPACE));
+}
+
+/* Get direntry length padded to 8 byte alignment */
+#define DIRENT64_LEN(namlen) \
+ ((sizeof(struct direntry) + (namlen) - (MAXPATHLEN-1) + 7) & ~7)
+
+errno_t
+vnode_readdir64(struct vnode *vp, struct uio *uio, int flags, int *eofflag,
int *numdirent, vfs_context_t ctxp)
{
/* Check if fs natively supports VNODE_READDIR_EXTENDED */
- if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSREADDIR_EXTENDED) {
+ if ((vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSREADDIR_EXTENDED) &&
+ ((vp->v_mount->mnt_kern_flag & MNTK_DENY_READDIREXT) == 0)) {
return VNOP_READDIR(vp, uio, flags, eofflag, numdirent, ctxp);
} else {
size_t bufsize;
void * bufptr;
uio_t auio;
- struct direntry entry64;
+ struct direntry *entry64;
struct dirent *dep;
int bytesread;
int error;
* use 32K in the MIN(), but we use magic number 87371 to
* prevent uio_resid() * 3 / 8 from overflowing.
*/
- bufsize = 3 * MIN(uio_resid(uio), 87371) / 8;
+ bufsize = 3 * MIN((user_size_t)uio_resid(uio), 87371u) / 8;
MALLOC(bufptr, void *, bufsize, M_TEMP, M_WAITOK);
if (bufptr == NULL) {
return ENOMEM;
dep = (struct dirent *)bufptr;
bytesread = bufsize - uio_resid(auio);
+ MALLOC(entry64, struct direntry *, sizeof(struct direntry),
+ M_TEMP, M_WAITOK);
/*
* Convert all the entries and copy them out to user's buffer.
*/
while (error == 0 && (char *)dep < ((char *)bufptr + bytesread)) {
+ size_t enbufsize = DIRENT64_LEN(dep->d_namlen);
+
+ bzero(entry64, enbufsize);
/* Convert a dirent to a dirent64. */
- entry64.d_ino = dep->d_ino;
- entry64.d_seekoff = 0;
- entry64.d_reclen = DIRENT64_LEN(dep->d_namlen);
- entry64.d_namlen = dep->d_namlen;
- entry64.d_type = dep->d_type;
- bcopy(dep->d_name, entry64.d_name, dep->d_namlen + 1);
+ entry64->d_ino = dep->d_ino;
+ entry64->d_seekoff = 0;
+ entry64->d_reclen = enbufsize;
+ entry64->d_namlen = dep->d_namlen;
+ entry64->d_type = dep->d_type;
+ bcopy(dep->d_name, entry64->d_name, dep->d_namlen + 1);
/* Move to next entry. */
dep = (struct dirent *)((char *)dep + dep->d_reclen);
/* Copy entry64 to user's buffer. */
- error = uiomove((caddr_t)&entry64, entry64.d_reclen, uio);
+ error = uiomove((caddr_t)entry64, entry64->d_reclen, uio);
}
/* Update the real offset using the offset we got from VNOP_READDIR. */
}
uio_free(auio);
FREE(bufptr, M_TEMP);
+ FREE(entry64, M_TEMP);
return (error);
}
}
+#define GETDIRENTRIES_MAXBUFSIZE (128 * 1024 * 1024U)
+
/*
* Read a block of directory entries in a file system independent format.
*/
goto out;
}
+ if (bufsize > GETDIRENTRIES_MAXBUFSIZE)
+ bufsize = GETDIRENTRIES_MAXBUFSIZE;
+
#if CONFIG_MACF
error = mac_file_check_change_offset(vfs_context_ucred(&context), fp->f_fglob);
if (error)
goto out;
}
- if ((vp->v_flag & VROOT) && (vp->v_mount->mnt_flag & MNT_UNION)) {
+ if ((vp->v_mount->mnt_flag & MNT_UNION)) {
struct vnode *tvp = vp;
- vp = vp->v_mount->mnt_vnodecovered;
- vnode_getwithref(vp);
- vnode_ref(vp);
- fp->f_fglob->fg_data = (caddr_t) vp;
- fp->f_fglob->fg_offset = 0;
- vnode_rele(tvp);
- vnode_put(tvp);
- goto unionread;
+ if (lookup_traverse_union(tvp, &vp, &context) == 0) {
+ vnode_ref(vp);
+ fp->f_fglob->fg_data = (caddr_t) vp;
+ fp->f_fglob->fg_offset = 0;
+ vnode_rele(tvp);
+ vnode_put(tvp);
+ goto unionread;
+ }
+ vp = tvp;
}
}
int error;
struct nameidata nd;
- NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
+ NDINIT(&nd, LOOKUP, OP_REVOKE, FOLLOW | AUDITVNPATH1, UIO_USERSPACE,
+ uap->path, ctx);
error = namei(&nd);
if (error)
return (error);
* which are specific to the HFS & HFS Plus volume formats
*/
-#ifdef __APPLE_API_OBSOLETE
-
-/************************************************/
-/* *** Following calls will be deleted soon *** */
-/************************************************/
-
-/*
- * Make a complex file. A complex file is one with multiple forks (data streams)
- */
-/* ARGSUSED */
-int
-mkcomplex(__unused proc_t p, __unused struct mkcomplex_args *uap, __unused int32_t *retval)
-{
- return (ENOTSUP);
-}
/*
- * Extended stat call which returns volumeid and vnodeid as well as other info
+ * Obtain attribute information on objects in a directory while enumerating
+ * the directory.
*/
-/* ARGSUSED */
-int
-statv(__unused proc_t p,
- __unused struct statv_args *uap,
- __unused int32_t *retval)
-{
- return (ENOTSUP); /* We'll just return an error for now */
-
-} /* end of statv system call */
-
-/*
-* Extended lstat call which returns volumeid and vnodeid as well as other info
-*/
-/* ARGSUSED */
-int
-lstatv(__unused proc_t p,
- __unused struct lstatv_args *uap,
- __unused int32_t *retval)
-{
- return (ENOTSUP); /* We'll just return an error for now */
-} /* end of lstatv system call */
-
-/*
-* Extended fstat call which returns volumeid and vnodeid as well as other info
-*/
-/* ARGSUSED */
-int
-fstatv(__unused proc_t p,
- __unused struct fstatv_args *uap,
- __unused int32_t *retval)
-{
- return (ENOTSUP); /* We'll just return an error for now */
-} /* end of fstatv system call */
-
-
-/************************************************/
-/* *** Preceding calls will be deleted soon *** */
-/************************************************/
-
-#endif /* __APPLE_API_OBSOLETE */
-
-/*
-* Obtain attribute information on objects in a directory while enumerating
-* the directory. This call does not yet support union mounted directories.
-* TO DO
-* 1.union mounted directories.
-*/
-
/* ARGSUSED */
int
getdirentriesattr (proc_t p, struct getdirentriesattr_args *uap, int32_t *retval)
struct fileproc *fp;
uio_t auio = NULL;
int spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
- uint32_t count;
+ uint32_t count, savecount;
uint32_t newstate;
int error, eofflag;
uint32_t loff;
if ((error = copyin(uap->count, (caddr_t)&count, sizeof(count)))) {
return(error);
}
+ savecount = count;
if ( (error = fp_getfvp(p, fd, &fp, &vp)) ) {
return (error);
}
AUDIT_ARG(vnpath, vp, ARG_VNODE1);
+unionread:
if (vp->v_type != VDIR) {
(void)vnode_put(vp);
error = EINVAL;
/* set up the uio structure which will contain the users return buffer */
loff = fp->f_fglob->fg_offset;
- auio = uio_createwithbuffer(1, loff, spacetype, UIO_READ,
- &uio_buf[0], sizeof(uio_buf));
+ auio = uio_createwithbuffer(1, loff, spacetype, UIO_READ, &uio_buf[0], sizeof(uio_buf));
uio_addiov(auio, uap->buffer, uap->buffersize);
/*
/* Believe it or not, uap->options only has 32-bits of valid
* info, so truncate before extending again */
- error = VNOP_READDIRATTR(vp, &attributelist, auio,
- count,
- (u_long)(uint32_t)uap->options, &newstate, &eofflag,
- &count, ctx);
+
+ error = VNOP_READDIRATTR(vp, &attributelist, auio, count,
+ (u_long)(uint32_t)uap->options, &newstate, &eofflag, &count, ctx);
+ }
+
+ if (error) {
+ (void) vnode_put(vp);
+ goto out;
+ }
+
+ /*
+ * If we've got the last entry of a directory in a union mount
+ * then reset the eofflag and pretend there's still more to come.
+ * The next call will again set eofflag and the buffer will be empty,
+ * so traverse to the underlying directory and do the directory
+ * read there.
+ */
+ if (eofflag && vp->v_mount->mnt_flag & MNT_UNION) {
+ if (uio_resid(auio) < (user_ssize_t) uap->buffersize) { // Got some entries
+ eofflag = 0;
+ } else { // Empty buffer
+ struct vnode *tvp = vp;
+ if (lookup_traverse_union(tvp, &vp, ctx) == 0) {
+ vnode_ref_ext(vp, fp->f_fglob->fg_flag & O_EVTONLY, 0);
+ fp->f_fglob->fg_data = (caddr_t) vp;
+ fp->f_fglob->fg_offset = 0; // reset index for new dir
+ count = savecount;
+ vnode_rele_internal(tvp, fp->f_fglob->fg_flag & O_EVTONLY, 0, 0);
+ vnode_put(tvp);
+ goto unionread;
+ }
+ vp = tvp;
+ }
}
+
(void)vnode_put(vp);
if (error)
file_drop(fd);
return (error); /* return error earlier, an retval of 0 or 1 now */
-} /* end of getdirentryattr system call */
+} /* end of getdirentriesattr system call */
/*
* Exchange data between two files
nameiflags = 0;
if ((uap->options & FSOPT_NOFOLLOW) == 0) nameiflags |= FOLLOW;
- NDINIT(&fnd, LOOKUP, nameiflags | AUDITVNPATH1,
- UIO_USERSPACE, uap->path1, ctx);
+ NDINIT(&fnd, LOOKUP, OP_EXCHANGEDATA, nameiflags | AUDITVNPATH1,
+ UIO_USERSPACE, uap->path1, ctx);
- error = namei(&fnd);
- if (error)
- goto out2;
+ error = namei(&fnd);
+ if (error)
+ goto out2;
nameidone(&fnd);
fvp = fnd.ni_vp;
- NDINIT(&snd, LOOKUP | CN_NBMOUNTLOOK, nameiflags | AUDITVNPATH2,
- UIO_USERSPACE, uap->path2, ctx);
+ NDINIT(&snd, LOOKUP, OP_EXCHANGEDATA, CN_NBMOUNTLOOK | nameiflags | AUDITVNPATH2,
+ UIO_USERSPACE, uap->path2, ctx);
- error = namei(&snd);
- if (error) {
+ error = namei(&snd);
+ if (error) {
vnode_put(fvp);
goto out2;
- }
+ }
nameidone(&snd);
svp = snd.ni_vp;
goto out;
}
+ /* If they're not files, return an error */
+ if ( (vnode_isreg(fvp) == 0) || (vnode_isreg(svp) == 0)) {
+ error = EINVAL;
+ goto out;
+ }
+
#if CONFIG_MACF
error = mac_vnode_check_exchangedata(ctx,
fvp, svp);
return (error);
}
+/*
+ * Return (in MB) the amount of freespace on the given vnode's volume.
+ */
+uint32_t freespace_mb(vnode_t vp);
+
+uint32_t
+freespace_mb(vnode_t vp)
+{
+ vfs_update_vfsstat(vp->v_mount, vfs_context_current(), VFS_USER_EVENT);
+ return (((uint64_t)vp->v_mount->mnt_vfsstat.f_bavail *
+ vp->v_mount->mnt_vfsstat.f_bsize) >> 20);
+}
+
+#if CONFIG_SEARCHFS
/* ARGSUSED */
int
searchfs(proc_t p, struct searchfs_args *uap, __unused int32_t *retval)
{
- vnode_t vp;
- int error=0;
+ vnode_t vp, tvp;
+ int i, error=0;
int fserror = 0;
struct nameidata nd;
struct user64_fssearchblock searchblock;
vfs_context_t ctx = vfs_context_current();
char uio_buf[ UIO_SIZEOF(1) ];
- /* Start by copying in fsearchblock paramater list */
+ /* Start by copying in fsearchblock parameter list */
if (IS_64BIT_PROCESS(p)) {
error = copyin(uap->searchblock, (caddr_t) &searchblock, sizeof(searchblock));
timelimit.tv_sec = searchblock.timelimit.tv_sec;
/* It all has to do into local memory and it's not that big so we might as well put it all together. */
/* Searchparams1 shall be first so we might as well use that to hold the base address of the allocated*/
/* block. */
-
+ /* */
+ /* NOTE: we allocate an extra 8 bytes to account for the difference in size of the searchstate */
+ /* due to the changes in rdar://problem/12438273. That way if a 3rd party file system */
+ /* assumes the size is still 556 bytes it will continue to work */
+
mallocsize = searchblock.sizeofsearchparams1 + searchblock.sizeofsearchparams2 +
- sizeof(struct attrlist) + sizeof(struct searchstate);
+ sizeof(struct attrlist) + sizeof(struct searchstate) + (2*sizeof(uint32_t));
MALLOC(searchparams1, void *, mallocsize, M_TEMP, M_WAITOK);
if ((error = copyin(uap->state, (caddr_t) state, sizeof(struct searchstate))))
goto freeandexit;
+ /*
+ * When searching a union mount, need to set the
+ * start flag at the first call on each layer to
+ * reset state for the new volume.
+ */
+ if (uap->options & SRCHFS_START)
+ state->ss_union_layer = 0;
+ else
+ uap->options |= state->ss_union_flags;
+ state->ss_union_flags = 0;
/*
* Because searchparams1 and searchparams2 may contain an ATTR_CMN_NAME search parameter,
}
/* set up the uio structure which will contain the users return buffer */
- auio = uio_createwithbuffer(1, 0, spacetype, UIO_READ,
- &uio_buf[0], sizeof(uio_buf));
- uio_addiov(auio, searchblock.returnbuffer, searchblock.returnbuffersize);
+ auio = uio_createwithbuffer(1, 0, spacetype, UIO_READ, &uio_buf[0], sizeof(uio_buf));
+ uio_addiov(auio, searchblock.returnbuffer, searchblock.returnbuffersize);
nameiflags = 0;
if ((uap->options & FSOPT_NOFOLLOW) == 0) nameiflags |= FOLLOW;
- NDINIT(&nd, LOOKUP, nameiflags | AUDITVNPATH1,
- UIO_USERSPACE, uap->path, ctx);
+ NDINIT(&nd, LOOKUP, OP_SEARCHFS, nameiflags | AUDITVNPATH1,
+ UIO_USERSPACE, uap->path, ctx);
error = namei(&nd);
if (error)
goto freeandexit;
-
+ vp = nd.ni_vp;
nameidone(&nd);
- vp = nd.ni_vp;
+
+ /*
+ * Switch to the root vnode for the volume
+ */
+ error = VFS_ROOT(vnode_mount(vp), &tvp, ctx);
+ vnode_put(vp);
+ if (error)
+ goto freeandexit;
+ vp = tvp;
+
+ /*
+ * If it's a union mount, the path lookup takes
+ * us to the top layer. But we may need to descend
+ * to a lower layer. For non-union mounts the layer
+ * is always zero.
+ */
+ for (i = 0; i < (int) state->ss_union_layer; i++) {
+ if ((vp->v_mount->mnt_flag & MNT_UNION) == 0)
+ break;
+ tvp = vp;
+ vp = vp->v_mount->mnt_vnodecovered;
+ if (vp == NULL) {
+ vnode_put(tvp);
+ error = ENOENT;
+ goto freeandexit;
+ }
+ vnode_getwithref(vp);
+ vnode_put(tvp);
+ }
+
+#if CONFIG_MACF
+ error = mac_vnode_check_searchfs(ctx, vp, &searchblock.searchattrs);
+ if (error) {
+ vnode_put(vp);
+ goto freeandexit;
+ }
+#endif
/*
* If searchblock.maxmatches == 0, then skip the search. This has happened
- * before and sometimes the underlyning code doesnt deal with it well.
+ * before and sometimes the underlying code doesnt deal with it well.
*/
if (searchblock.maxmatches == 0) {
nummatches = 0;
}
/*
- Allright, we have everything we need, so lets make that call.
-
- We keep special track of the return value from the file system:
- EAGAIN is an acceptable error condition that shouldn't keep us
- from copying out any results...
+ * Allright, we have everything we need, so lets make that call.
+ *
+ * We keep special track of the return value from the file system:
+ * EAGAIN is an acceptable error condition that shouldn't keep us
+ * from copying out any results...
*/
fserror = VNOP_SEARCHFS(vp,
- searchparams1,
- searchparams2,
- &searchblock.searchattrs,
- (u_long)searchblock.maxmatches,
- &timelimit,
- returnattrs,
- &nummatches,
- (u_long)uap->scriptcode,
- (u_long)uap->options,
- auio,
- state,
- ctx);
+ searchparams1,
+ searchparams2,
+ &searchblock.searchattrs,
+ (u_long)searchblock.maxmatches,
+ &timelimit,
+ returnattrs,
+ &nummatches,
+ (u_long)uap->scriptcode,
+ (u_long)uap->options,
+ auio,
+ (struct searchstate *) &state->ss_fsstate,
+ ctx);
+ /*
+ * If it's a union mount we need to be called again
+ * to search the mounted-on filesystem.
+ */
+ if ((vp->v_mount->mnt_flag & MNT_UNION) && fserror == 0) {
+ state->ss_union_flags = SRCHFS_START;
+ state->ss_union_layer++; // search next layer down
+ fserror = EAGAIN;
+ }
+
saveandexit:
vnode_put(vp);
if ((error = copyout((caddr_t) state, uap->state, sizeof(struct searchstate))) != 0)
goto freeandexit;
- if ((error = suulong(uap->nummatches, (uint64_t)nummatches)) != 0)
- goto freeandexit;
-
- error = fserror;
+ if ((error = suulong(uap->nummatches, (uint64_t)nummatches)) != 0)
+ goto freeandexit;
+
+ error = fserror;
+
+freeandexit:
+
+ FREE(searchparams1,M_TEMP);
+
+ return(error);
+
+
+} /* end of searchfs system call */
+
+#else /* CONFIG_SEARCHFS */
+
+int
+searchfs(__unused proc_t p, __unused struct searchfs_args *uap, __unused int32_t *retval)
+{
+ return (ENOTSUP);
+}
+
+#endif /* CONFIG_SEARCHFS */
+
+
+lck_grp_attr_t * nspace_group_attr;
+lck_attr_t * nspace_lock_attr;
+lck_grp_t * nspace_mutex_group;
+
+lck_mtx_t nspace_handler_lock;
+lck_mtx_t nspace_handler_exclusion_lock;
+
+time_t snapshot_timestamp=0;
+int nspace_allow_virtual_devs=0;
+
+void nspace_handler_init(void);
+
+typedef struct nspace_item_info {
+ struct vnode *vp;
+ void *arg;
+ uint64_t op;
+ uint32_t vid;
+ uint32_t flags;
+ uint32_t token;
+ uint32_t refcount;
+} nspace_item_info;
+
+#define MAX_NSPACE_ITEMS 128
+nspace_item_info nspace_items[MAX_NSPACE_ITEMS];
+uint32_t nspace_item_idx=0; // also used as the sleep/wakeup rendezvous address
+uint32_t nspace_token_id=0;
+uint32_t nspace_handler_timeout = 15; // seconds
+
+#define NSPACE_ITEM_NEW 0x0001
+#define NSPACE_ITEM_PROCESSING 0x0002
+#define NSPACE_ITEM_DEAD 0x0004
+#define NSPACE_ITEM_CANCELLED 0x0008
+#define NSPACE_ITEM_DONE 0x0010
+#define NSPACE_ITEM_RESET_TIMER 0x0020
+
+#define NSPACE_ITEM_NSPACE_EVENT 0x0040
+#define NSPACE_ITEM_SNAPSHOT_EVENT 0x0080
+
+#define NSPACE_ITEM_ALL_EVENT_TYPES (NSPACE_ITEM_NSPACE_EVENT | NSPACE_ITEM_SNAPSHOT_EVENT)
+
+//#pragma optimization_level 0
+
+typedef enum {
+ NSPACE_HANDLER_NSPACE = 0,
+ NSPACE_HANDLER_SNAPSHOT = 1,
+
+ NSPACE_HANDLER_COUNT,
+} nspace_type_t;
+
+typedef struct {
+ uint64_t handler_tid;
+ struct proc *handler_proc;
+ int handler_busy;
+} nspace_handler_t;
+
+nspace_handler_t nspace_handlers[NSPACE_HANDLER_COUNT];
+
+/* namespace fsctl functions */
+static int nspace_flags_matches_handler(uint32_t event_flags, nspace_type_t nspace_type);
+static int nspace_item_flags_for_type(nspace_type_t nspace_type);
+static int nspace_open_flags_for_type(nspace_type_t nspace_type);
+static nspace_type_t nspace_type_for_op(uint64_t op);
+static int nspace_is_special_process(struct proc *proc);
+static int vn_open_with_vp(vnode_t vp, int fmode, vfs_context_t ctx);
+static int wait_for_namespace_event(namespace_handler_data *nhd, nspace_type_t nspace_type);
+static int validate_namespace_args (int is64bit, int size);
+static int process_namespace_fsctl(nspace_type_t nspace_type, int is64bit, u_int size, caddr_t data);
+
+
+static inline int nspace_flags_matches_handler(uint32_t event_flags, nspace_type_t nspace_type)
+{
+ switch(nspace_type) {
+ case NSPACE_HANDLER_NSPACE:
+ return (event_flags & NSPACE_ITEM_ALL_EVENT_TYPES) == NSPACE_ITEM_NSPACE_EVENT;
+ case NSPACE_HANDLER_SNAPSHOT:
+ return (event_flags & NSPACE_ITEM_ALL_EVENT_TYPES) == NSPACE_ITEM_SNAPSHOT_EVENT;
+ default:
+ printf("nspace_flags_matches_handler: invalid type %u\n", (int)nspace_type);
+ return 0;
+ }
+}
+
+static inline int nspace_item_flags_for_type(nspace_type_t nspace_type)
+{
+ switch(nspace_type) {
+ case NSPACE_HANDLER_NSPACE:
+ return NSPACE_ITEM_NSPACE_EVENT;
+ case NSPACE_HANDLER_SNAPSHOT:
+ return NSPACE_ITEM_SNAPSHOT_EVENT;
+ default:
+ printf("nspace_item_flags_for_type: invalid type %u\n", (int)nspace_type);
+ return 0;
+ }
+}
+
+static inline int nspace_open_flags_for_type(nspace_type_t nspace_type)
+{
+ switch(nspace_type) {
+ case NSPACE_HANDLER_NSPACE:
+ return FREAD | FWRITE | O_EVTONLY;
+ case NSPACE_HANDLER_SNAPSHOT:
+ return FREAD | O_EVTONLY;
+ default:
+ printf("nspace_open_flags_for_type: invalid type %u\n", (int)nspace_type);
+ return 0;
+ }
+}
+
+static inline nspace_type_t nspace_type_for_op(uint64_t op)
+{
+ switch(op & NAMESPACE_HANDLER_EVENT_TYPE_MASK) {
+ case NAMESPACE_HANDLER_NSPACE_EVENT:
+ return NSPACE_HANDLER_NSPACE;
+ case NAMESPACE_HANDLER_SNAPSHOT_EVENT:
+ return NSPACE_HANDLER_SNAPSHOT;
+ default:
+ printf("nspace_type_for_op: invalid op mask %llx\n", op & NAMESPACE_HANDLER_EVENT_TYPE_MASK);
+ return NSPACE_HANDLER_NSPACE;
+ }
+}
+
+static inline int nspace_is_special_process(struct proc *proc)
+{
+ int i;
+ for (i = 0; i < NSPACE_HANDLER_COUNT; i++) {
+ if (proc == nspace_handlers[i].handler_proc)
+ return 1;
+ }
+ return 0;
+}
+
+void
+nspace_handler_init(void)
+{
+ nspace_lock_attr = lck_attr_alloc_init();
+ nspace_group_attr = lck_grp_attr_alloc_init();
+ nspace_mutex_group = lck_grp_alloc_init("nspace-mutex", nspace_group_attr);
+ lck_mtx_init(&nspace_handler_lock, nspace_mutex_group, nspace_lock_attr);
+ lck_mtx_init(&nspace_handler_exclusion_lock, nspace_mutex_group, nspace_lock_attr);
+ memset(&nspace_items[0], 0, sizeof(nspace_items));
+}
+
+void
+nspace_proc_exit(struct proc *p)
+{
+ int i, event_mask = 0;
+
+ for (i = 0; i < NSPACE_HANDLER_COUNT; i++) {
+ if (p == nspace_handlers[i].handler_proc) {
+ event_mask |= nspace_item_flags_for_type(i);
+ nspace_handlers[i].handler_tid = 0;
+ nspace_handlers[i].handler_proc = NULL;
+ }
+ }
+
+ if (event_mask == 0) {
+ return;
+ }
+
+ if (event_mask & NSPACE_ITEM_SNAPSHOT_EVENT) {
+ // if this process was the snapshot handler, zero snapshot_timeout
+ snapshot_timestamp = 0;
+ }
+
+ //
+ // unblock anyone that's waiting for the handler that died
+ //
+ lck_mtx_lock(&nspace_handler_lock);
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].flags & (NSPACE_ITEM_NEW | NSPACE_ITEM_PROCESSING)) {
+
+ if ( nspace_items[i].flags & event_mask ) {
+
+ if (nspace_items[i].vp && (nspace_items[i].vp->v_flag & VNEEDSSNAPSHOT)) {
+ vnode_lock_spin(nspace_items[i].vp);
+ nspace_items[i].vp->v_flag &= ~VNEEDSSNAPSHOT;
+ vnode_unlock(nspace_items[i].vp);
+ }
+ nspace_items[i].vp = NULL;
+ nspace_items[i].vid = 0;
+ nspace_items[i].flags = NSPACE_ITEM_DONE;
+ nspace_items[i].token = 0;
+
+ wakeup((caddr_t)&(nspace_items[i].vp));
+ }
+ }
+ }
+
+ wakeup((caddr_t)&nspace_item_idx);
+ lck_mtx_unlock(&nspace_handler_lock);
+}
+
+
+int
+resolve_nspace_item(struct vnode *vp, uint64_t op)
+{
+ return resolve_nspace_item_ext(vp, op, NULL);
+}
+
+int
+resolve_nspace_item_ext(struct vnode *vp, uint64_t op, void *arg)
+{
+ int i, error, keep_waiting;
+ struct timespec ts;
+ nspace_type_t nspace_type = nspace_type_for_op(op);
+
+ // only allow namespace events on regular files, directories and symlinks.
+ if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
+ return 0;
+ }
+
+ //
+ // if this is a snapshot event and the vnode is on a
+ // disk image just pretend nothing happened since any
+ // change to the disk image will cause the disk image
+ // itself to get backed up and this avoids multi-way
+ // deadlocks between the snapshot handler and the ever
+ // popular diskimages-helper process. the variable
+ // nspace_allow_virtual_devs allows this behavior to
+ // be overridden (for use by the Mobile TimeMachine
+ // testing infrastructure which uses disk images)
+ //
+ if ( (op & NAMESPACE_HANDLER_SNAPSHOT_EVENT)
+ && (vp->v_mount != NULL)
+ && (vp->v_mount->mnt_kern_flag & MNTK_VIRTUALDEV)
+ && !nspace_allow_virtual_devs) {
+
+ return 0;
+ }
+
+ // if (thread_tid(current_thread()) == namespace_handler_tid) {
+ if (nspace_handlers[nspace_type].handler_proc == NULL) {
+ return 0;
+ }
+
+ if (nspace_is_special_process(current_proc())) {
+ return EDEADLK;
+ }
+
+ lck_mtx_lock(&nspace_handler_lock);
+
+retry:
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (vp == nspace_items[i].vp && op == nspace_items[i].op) {
+ break;
+ }
+ }
+
+ if (i >= MAX_NSPACE_ITEMS) {
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].flags == 0) {
+ break;
+ }
+ }
+ } else {
+ nspace_items[i].refcount++;
+ }
+
+ if (i >= MAX_NSPACE_ITEMS) {
+ ts.tv_sec = nspace_handler_timeout;
+ ts.tv_nsec = 0;
+
+ error = msleep((caddr_t)&nspace_token_id, &nspace_handler_lock, PVFS|PCATCH, "nspace-no-space", &ts);
+ if (error == 0) {
+ // an entry got free'd up, go see if we can get a slot
+ goto retry;
+ } else {
+ lck_mtx_unlock(&nspace_handler_lock);
+ return error;
+ }
+ }
+
+ //
+ // if it didn't already exist, add it. if it did exist
+ // we'll get woken up when someone does a wakeup() on
+ // the slot in the nspace_items table.
+ //
+ if (vp != nspace_items[i].vp) {
+ nspace_items[i].vp = vp;
+ nspace_items[i].arg = (arg == NSPACE_REARM_NO_ARG) ? NULL : arg; // arg is {NULL, true, uio *} - only pass uio thru to the user
+ nspace_items[i].op = op;
+ nspace_items[i].vid = vnode_vid(vp);
+ nspace_items[i].flags = NSPACE_ITEM_NEW;
+ nspace_items[i].flags |= nspace_item_flags_for_type(nspace_type);
+ if (nspace_items[i].flags & NSPACE_ITEM_SNAPSHOT_EVENT) {
+ if (arg) {
+ vnode_lock_spin(vp);
+ vp->v_flag |= VNEEDSSNAPSHOT;
+ vnode_unlock(vp);
+ }
+ }
+
+ nspace_items[i].token = 0;
+ nspace_items[i].refcount = 1;
+
+ wakeup((caddr_t)&nspace_item_idx);
+ }
+
+ //
+ // Now go to sleep until the handler does a wakeup on this
+ // slot in the nspace_items table (or we timeout).
+ //
+ keep_waiting = 1;
+ while(keep_waiting) {
+ ts.tv_sec = nspace_handler_timeout;
+ ts.tv_nsec = 0;
+ error = msleep((caddr_t)&(nspace_items[i].vp), &nspace_handler_lock, PVFS|PCATCH, "namespace-done", &ts);
+
+ if (nspace_items[i].flags & NSPACE_ITEM_DONE) {
+ error = 0;
+ } else if (nspace_items[i].flags & NSPACE_ITEM_CANCELLED) {
+ error = nspace_items[i].token;
+ } else if (error == EWOULDBLOCK || error == ETIMEDOUT) {
+ if (nspace_items[i].flags & NSPACE_ITEM_RESET_TIMER) {
+ nspace_items[i].flags &= ~NSPACE_ITEM_RESET_TIMER;
+ continue;
+ } else {
+ error = ETIMEDOUT;
+ }
+ } else if (error == 0) {
+ // hmmm, why did we get woken up?
+ printf("woken up for token %d but it's not done, cancelled or timedout and error == 0.\n",
+ nspace_items[i].token);
+ }
+
+ if (--nspace_items[i].refcount == 0) {
+ nspace_items[i].vp = NULL; // clear this so that no one will match on it again
+ nspace_items[i].arg = NULL;
+ nspace_items[i].token = 0; // clear this so that the handler will not find it anymore
+ nspace_items[i].flags = 0; // this clears it for re-use
+ }
+ wakeup(&nspace_token_id);
+ keep_waiting = 0;
+ }
+
+ lck_mtx_unlock(&nspace_handler_lock);
+
+ return error;
+}
+
+
+int
+get_nspace_item_status(struct vnode *vp, int32_t *status)
+{
+ int i;
+
+ lck_mtx_lock(&nspace_handler_lock);
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].vp == vp) {
+ break;
+ }
+ }
+
+ if (i >= MAX_NSPACE_ITEMS) {
+ lck_mtx_unlock(&nspace_handler_lock);
+ return ENOENT;
+ }
+
+ *status = nspace_items[i].flags;
+ lck_mtx_unlock(&nspace_handler_lock);
+ return 0;
+}
+
+
+#if 0
+static int
+build_volfs_path(struct vnode *vp, char *path, int *len)
+{
+ struct vnode_attr va;
+ int ret;
+
+ VATTR_INIT(&va);
+ VATTR_WANTED(&va, va_fsid);
+ VATTR_WANTED(&va, va_fileid);
+
+ if (vnode_getattr(vp, &va, vfs_context_kernel()) != 0) {
+ *len = snprintf(path, *len, "/non/existent/path/because/vnode_getattr/failed") + 1;
+ ret = -1;
+ } else {
+ *len = snprintf(path, *len, "/.vol/%d/%lld", (dev_t)va.va_fsid, va.va_fileid) + 1;
+ ret = 0;
+ }
+
+ return ret;
+}
+#endif
+
+//
+// Note: this function does NOT check permissions on all of the
+// parent directories leading to this vnode. It should only be
+// called on behalf of a root process. Otherwise a process may
+// get access to a file because the file itself is readable even
+// though its parent directories would prevent access.
+//
+static int
+vn_open_with_vp(vnode_t vp, int fmode, vfs_context_t ctx)
+{
+ int error, action;
+
+ if ((error = suser(kauth_cred_get(), &(current_proc()->p_acflag)))) {
+ return error;
+ }
+
+#if CONFIG_MACF
+ error = mac_vnode_check_open(ctx, vp, fmode);
+ if (error)
+ return error;
+#endif
+
+ /* compute action to be authorized */
+ action = 0;
+ if (fmode & FREAD) {
+ action |= KAUTH_VNODE_READ_DATA;
+ }
+ if (fmode & (FWRITE | O_TRUNC)) {
+ /*
+ * If we are writing, appending, and not truncating,
+ * indicate that we are appending so that if the
+ * UF_APPEND or SF_APPEND bits are set, we do not deny
+ * the open.
+ */
+ if ((fmode & O_APPEND) && !(fmode & O_TRUNC)) {
+ action |= KAUTH_VNODE_APPEND_DATA;
+ } else {
+ action |= KAUTH_VNODE_WRITE_DATA;
+ }
+ }
+
+ if ((error = vnode_authorize(vp, NULL, action, ctx)) != 0)
+ return error;
+
+
+ //
+ // if the vnode is tagged VOPENEVT and the current process
+ // has the P_CHECKOPENEVT flag set, then we or in the O_EVTONLY
+ // flag to the open mode so that this open won't count against
+ // the vnode when carbon delete() does a vnode_isinuse() to see
+ // if a file is currently in use. this allows spotlight
+ // importers to not interfere with carbon apps that depend on
+ // the no-delete-if-busy semantics of carbon delete().
+ //
+ if ((vp->v_flag & VOPENEVT) && (current_proc()->p_flag & P_CHECKOPENEVT)) {
+ fmode |= O_EVTONLY;
+ }
+
+ if ( (error = VNOP_OPEN(vp, fmode, ctx)) ) {
+ return error;
+ }
+ if ( (error = vnode_ref_ext(vp, fmode, 0)) ) {
+ VNOP_CLOSE(vp, fmode, ctx);
+ return error;
+ }
+
+ /* Call out to allow 3rd party notification of open.
+ * Ignore result of kauth_authorize_fileop call.
+ */
+#if CONFIG_MACF
+ mac_vnode_notify_open(ctx, vp, fmode);
+#endif
+ kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_OPEN,
+ (uintptr_t)vp, 0);
+
+
+ return 0;
+}
+
+static int
+wait_for_namespace_event(namespace_handler_data *nhd, nspace_type_t nspace_type)
+{
+ int i, error=0, unblock=0;
+ task_t curtask;
+
+ lck_mtx_lock(&nspace_handler_exclusion_lock);
+ if (nspace_handlers[nspace_type].handler_busy) {
+ lck_mtx_unlock(&nspace_handler_exclusion_lock);
+ return EBUSY;
+ }
+ nspace_handlers[nspace_type].handler_busy = 1;
+ lck_mtx_unlock(&nspace_handler_exclusion_lock);
+
+ /*
+ * Any process that gets here will be one of the namespace handlers.
+ * As such, they should be prevented from acquiring DMG vnodes during vnode reclamation
+ * as we can cause deadlocks to occur, because the namespace handler may prevent
+ * VNOP_INACTIVE from proceeding. Mark the current task as a P_DEPENDENCY_CAPABLE
+ * process.
+ */
+ curtask = current_task();
+ bsd_set_dependency_capable (curtask);
+
+ lck_mtx_lock(&nspace_handler_lock);
+ if (nspace_handlers[nspace_type].handler_proc == NULL) {
+ nspace_handlers[nspace_type].handler_tid = thread_tid(current_thread());
+ nspace_handlers[nspace_type].handler_proc = current_proc();
+ }
+
+ while (error == 0) {
+
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].flags & NSPACE_ITEM_NEW) {
+ if (!nspace_flags_matches_handler(nspace_items[i].flags, nspace_type)) {
+ continue;
+ }
+ break;
+ }
+ }
+
+ if (i < MAX_NSPACE_ITEMS) {
+ nspace_items[i].flags &= ~NSPACE_ITEM_NEW;
+ nspace_items[i].flags |= NSPACE_ITEM_PROCESSING;
+ nspace_items[i].token = ++nspace_token_id;
+
+ if (nspace_items[i].vp) {
+ struct fileproc *fp;
+ int32_t indx, fmode;
+ struct proc *p = current_proc();
+ vfs_context_t ctx = vfs_context_current();
+ struct vnode_attr va;
+
+
+ /*
+ * Use vnode pointer to acquire a file descriptor for
+ * hand-off to userland
+ */
+ fmode = nspace_open_flags_for_type(nspace_type);
+ error = vnode_getwithvid(nspace_items[i].vp, nspace_items[i].vid);
+ if (error) {
+ unblock = 1;
+ break;
+ }
+ error = vn_open_with_vp(nspace_items[i].vp, fmode, ctx);
+ if (error) {
+ unblock = 1;
+ vnode_put(nspace_items[i].vp);
+ break;
+ }
+
+ if ((error = falloc(p, &fp, &indx, ctx))) {
+ vn_close(nspace_items[i].vp, fmode, ctx);
+ vnode_put(nspace_items[i].vp);
+ unblock = 1;
+ break;
+ }
+
+ fp->f_fglob->fg_flag = fmode;
+ fp->f_fglob->fg_ops = &vnops;
+ fp->f_fglob->fg_data = (caddr_t)nspace_items[i].vp;
+
+ proc_fdlock(p);
+ procfdtbl_releasefd(p, indx, NULL);
+ fp_drop(p, indx, fp, 1);
+ proc_fdunlock(p);
+
+ /*
+ * All variants of the namespace handler struct support these three fields:
+ * token, flags, and the FD pointer
+ */
+ error = copyout(&nspace_items[i].token, nhd->token, sizeof(uint32_t));
+ error = copyout(&nspace_items[i].op, nhd->flags, sizeof(uint64_t));
+ error = copyout(&indx, nhd->fdptr, sizeof(uint32_t));
+
+ /*
+ * Handle optional fields:
+ * extended version support an info ptr (offset, length), and the
+ *
+ * namedata version supports a unique per-link object ID
+ *
+ */
+ if (nhd->infoptr) {
+ uio_t uio = (uio_t)nspace_items[i].arg;
+ uint64_t u_offset, u_length;
+
+ if (uio) {
+ u_offset = uio_offset(uio);
+ u_length = uio_resid(uio);
+ } else {
+ u_offset = 0;
+ u_length = 0;
+ }
+ error = copyout(&u_offset, nhd->infoptr, sizeof(uint64_t));
+ error = copyout(&u_length, nhd->infoptr+sizeof(uint64_t), sizeof(uint64_t));
+ }
+
+ if (nhd->objid) {
+ VATTR_INIT(&va);
+ VATTR_WANTED(&va, va_linkid);
+ error = vnode_getattr(nspace_items[i].vp, &va, ctx);
+ if (error == 0 ) {
+ uint64_t linkid = 0;
+ if (VATTR_IS_SUPPORTED (&va, va_linkid)) {
+ linkid = (uint64_t)va.va_linkid;
+ }
+ error = copyout (&linkid, nhd->objid, sizeof(uint64_t));
+ }
+ }
+
+ if (error) {
+ vn_close(nspace_items[i].vp, fmode, ctx);
+ fp_free(p, indx, fp);
+ unblock = 1;
+ }
+
+ vnode_put(nspace_items[i].vp);
+
+ break;
+ } else {
+ printf("wait_for_nspace_event: failed (nspace_items[%d] == %p error %d, name %s)\n",
+ i, nspace_items[i].vp, error, nspace_items[i].vp->v_name);
+ }
+
+ } else {
+ error = msleep((caddr_t)&nspace_item_idx, &nspace_handler_lock, PVFS|PCATCH, "namespace-items", 0);
+ if ((nspace_type == NSPACE_HANDLER_SNAPSHOT) && (snapshot_timestamp == 0 || snapshot_timestamp == ~0)) {
+ error = EINVAL;
+ break;
+ }
+
+ }
+ }
+
+ if (unblock) {
+ if (nspace_items[i].vp && (nspace_items[i].vp->v_flag & VNEEDSSNAPSHOT)) {
+ vnode_lock_spin(nspace_items[i].vp);
+ nspace_items[i].vp->v_flag &= ~VNEEDSSNAPSHOT;
+ vnode_unlock(nspace_items[i].vp);
+ }
+ nspace_items[i].vp = NULL;
+ nspace_items[i].vid = 0;
+ nspace_items[i].flags = NSPACE_ITEM_DONE;
+ nspace_items[i].token = 0;
+
+ wakeup((caddr_t)&(nspace_items[i].vp));
+ }
+
+ if (nspace_type == NSPACE_HANDLER_SNAPSHOT) {
+ // just go through every snapshot event and unblock it immediately.
+ if (error && (snapshot_timestamp == 0 || snapshot_timestamp == ~0)) {
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].flags & NSPACE_ITEM_NEW) {
+ if (nspace_flags_matches_handler(nspace_items[i].flags, nspace_type)) {
+ nspace_items[i].vp = NULL;
+ nspace_items[i].vid = 0;
+ nspace_items[i].flags = NSPACE_ITEM_DONE;
+ nspace_items[i].token = 0;
+
+ wakeup((caddr_t)&(nspace_items[i].vp));
+ }
+ }
+ }
+ }
+ }
+
+ lck_mtx_unlock(&nspace_handler_lock);
+
+ lck_mtx_lock(&nspace_handler_exclusion_lock);
+ nspace_handlers[nspace_type].handler_busy = 0;
+ lck_mtx_unlock(&nspace_handler_exclusion_lock);
+
+ return error;
+}
+
+static inline int validate_namespace_args (int is64bit, int size) {
+
+ if (is64bit) {
+ /* Must be one of these */
+ if (size == sizeof(user64_namespace_handler_info)) {
+ goto sizeok;
+ }
+ if (size == sizeof(user64_namespace_handler_info_ext)) {
+ goto sizeok;
+ }
+ if (size == sizeof(user64_namespace_handler_data)) {
+ goto sizeok;
+ }
+ return EINVAL;
+ }
+ else {
+ /* 32 bit -- must be one of these */
+ if (size == sizeof(user32_namespace_handler_info)) {
+ goto sizeok;
+ }
+ if (size == sizeof(user32_namespace_handler_info_ext)) {
+ goto sizeok;
+ }
+ if (size == sizeof(user32_namespace_handler_data)) {
+ goto sizeok;
+ }
+ return EINVAL;
+ }
+
+sizeok:
+
+ return 0;
+
+}
+
+static int process_namespace_fsctl(nspace_type_t nspace_type, int is64bit, u_int size, caddr_t data)
+{
+ int error = 0;
+ namespace_handler_data nhd;
+
+ bzero (&nhd, sizeof(namespace_handler_data));
+
+ if (nspace_type == NSPACE_HANDLER_SNAPSHOT &&
+ (snapshot_timestamp == 0 || snapshot_timestamp == ~0)) {
+ return EINVAL;
+ }
+
+ if ((error = suser(kauth_cred_get(), &(current_proc()->p_acflag)))) {
+ return error;
+ }
+
+ error = validate_namespace_args (is64bit, size);
+ if (error) {
+ return error;
+ }
+
+ /* Copy in the userland pointers into our kernel-only struct */
+
+ if (is64bit) {
+ /* 64 bit userland structures */
+ nhd.token = (user_addr_t)((user64_namespace_handler_info *)data)->token;
+ nhd.flags = (user_addr_t)((user64_namespace_handler_info *)data)->flags;
+ nhd.fdptr = (user_addr_t)((user64_namespace_handler_info *)data)->fdptr;
+
+ /* If the size is greater than the standard info struct, add in extra fields */
+ if (size > (sizeof(user64_namespace_handler_info))) {
+ if (size >= (sizeof(user64_namespace_handler_info_ext))) {
+ nhd.infoptr = (user_addr_t)((user64_namespace_handler_info_ext *)data)->infoptr;
+ }
+ if (size == (sizeof(user64_namespace_handler_data))) {
+ nhd.objid = (user_addr_t)((user64_namespace_handler_data*)data)->objid;
+ }
+ /* Otherwise the fields were pre-zeroed when we did the bzero above. */
+ }
+ }
+ else {
+ /* 32 bit userland structures */
+ nhd.token = CAST_USER_ADDR_T(((user32_namespace_handler_info *)data)->token);
+ nhd.flags = CAST_USER_ADDR_T(((user32_namespace_handler_info *)data)->flags);
+ nhd.fdptr = CAST_USER_ADDR_T(((user32_namespace_handler_info *)data)->fdptr);
+
+ if (size > (sizeof(user32_namespace_handler_info))) {
+ if (size >= (sizeof(user32_namespace_handler_info_ext))) {
+ nhd.infoptr = CAST_USER_ADDR_T(((user32_namespace_handler_info_ext *)data)->infoptr);
+ }
+ if (size == (sizeof(user32_namespace_handler_data))) {
+ nhd.objid = (user_addr_t)((user32_namespace_handler_data*)data)->objid;
+ }
+ /* Otherwise the fields were pre-zeroed when we did the bzero above. */
+ }
+ }
+
+ return wait_for_namespace_event(&nhd, nspace_type);
+}
+
+/*
+ * Make a filesystem-specific control call:
+ */
+/* ARGSUSED */
+static int
+fsctl_internal(proc_t p, vnode_t *arg_vp, u_long cmd, user_addr_t udata, u_long options, vfs_context_t ctx)
+{
+ int error=0;
+ boolean_t is64bit;
+ u_int size;
+#define STK_PARAMS 128
+ char stkbuf[STK_PARAMS];
+ caddr_t data, memp;
+ vnode_t vp = *arg_vp;
+
+ size = IOCPARM_LEN(cmd);
+ if (size > IOCPARM_MAX) return (EINVAL);
+
+ is64bit = proc_is64bit(p);
+
+ memp = NULL;
+
+
+ /*
+ * ensure the buffer is large enough for underlying calls
+ */
+#ifndef HFSIOC_GETPATH
+ typedef char pn_t[MAXPATHLEN];
+#define HFSIOC_GETPATH _IOWR('h', 13, pn_t)
+#endif
+
+#ifndef HFS_GETPATH
+#define HFS_GETPATH IOCBASECMD(HFSIOC_GETPATH)
+#endif
+ if (IOCBASECMD(cmd) == HFS_GETPATH) {
+ /* Round up to MAXPATHLEN regardless of user input */
+ size = MAXPATHLEN;
+ }
+
+ if (size > sizeof (stkbuf)) {
+ if ((memp = (caddr_t)kalloc(size)) == 0) return ENOMEM;
+ data = memp;
+ } else {
+ data = &stkbuf[0];
+ };
+
+ if (cmd & IOC_IN) {
+ if (size) {
+ error = copyin(udata, data, size);
+ if (error) {
+ if (memp) {
+ kfree (memp, size);
+ }
+ return error;
+ }
+ } else {
+ if (is64bit) {
+ *(user_addr_t *)data = udata;
+ }
+ else {
+ *(uint32_t *)data = (uint32_t)udata;
+ }
+ };
+ } else if ((cmd & IOC_OUT) && size) {
+ /*
+ * Zero the buffer so the user always
+ * gets back something deterministic.
+ */
+ bzero(data, size);
+ } else if (cmd & IOC_VOID) {
+ if (is64bit) {
+ *(user_addr_t *)data = udata;
+ }
+ else {
+ *(uint32_t *)data = (uint32_t)udata;
+ }
+ }
+
+ /* Check to see if it's a generic command */
+ switch (IOCBASECMD(cmd)) {
+
+ case FSCTL_SYNC_VOLUME: {
+ mount_t mp = vp->v_mount;
+ int arg = *(uint32_t*)data;
+
+ /* record vid of vp so we can drop it below. */
+ uint32_t vvid = vp->v_id;
+
+ /*
+ * Then grab mount_iterref so that we can release the vnode.
+ * Without this, a thread may call vnode_iterate_prepare then
+ * get into a deadlock because we've never released the root vp
+ */
+ error = mount_iterref (mp, 0);
+ if (error) {
+ break;
+ }
+ vnode_put(vp);
+
+ /* issue the sync for this volume */
+ (void)sync_callback(mp, (arg & FSCTL_SYNC_WAIT) ? &arg : NULL);
+
+ /*
+ * Then release the mount_iterref once we're done syncing; it's not
+ * needed for the VNOP_IOCTL below
+ */
+ mount_iterdrop(mp);
+
+ if (arg & FSCTL_SYNC_FULLSYNC) {
+ /* re-obtain vnode iocount on the root vp, if possible */
+ error = vnode_getwithvid (vp, vvid);
+ if (error == 0) {
+ error = VNOP_IOCTL(vp, F_FULLFSYNC, (caddr_t)NULL, 0, ctx);
+ vnode_put (vp);
+ }
+ }
+ /* mark the argument VP as having been released */
+ *arg_vp = NULL;
+ }
+ break;
+
+ case FSCTL_SET_PACKAGE_EXTS: {
+ user_addr_t ext_strings;
+ uint32_t num_entries;
+ uint32_t max_width;
+
+ if ( (is64bit && size != sizeof(user64_package_ext_info))
+ || (is64bit == 0 && size != sizeof(user32_package_ext_info))) {
+
+ // either you're 64-bit and passed a 64-bit struct or
+ // you're 32-bit and passed a 32-bit struct. otherwise
+ // it's not ok.
+ error = EINVAL;
+ break;
+ }
+
+ if (is64bit) {
+ ext_strings = ((user64_package_ext_info *)data)->strings;
+ num_entries = ((user64_package_ext_info *)data)->num_entries;
+ max_width = ((user64_package_ext_info *)data)->max_width;
+ } else {
+ ext_strings = CAST_USER_ADDR_T(((user32_package_ext_info *)data)->strings);
+ num_entries = ((user32_package_ext_info *)data)->num_entries;
+ max_width = ((user32_package_ext_info *)data)->max_width;
+ }
+ error = set_package_extensions_table(ext_strings, num_entries, max_width);
+ }
+ break;
+
+ /* namespace handlers */
+ case FSCTL_NAMESPACE_HANDLER_GET: {
+ error = process_namespace_fsctl(NSPACE_HANDLER_NSPACE, is64bit, size, data);
+ }
+ break;
+
+ /* Snapshot handlers */
+ case FSCTL_OLD_SNAPSHOT_HANDLER_GET: {
+ error = process_namespace_fsctl(NSPACE_HANDLER_SNAPSHOT, is64bit, size, data);
+ }
+ break;
+
+ case FSCTL_SNAPSHOT_HANDLER_GET_EXT: {
+ error = process_namespace_fsctl(NSPACE_HANDLER_SNAPSHOT, is64bit, size, data);
+ }
+ break;
+
+ case FSCTL_NAMESPACE_HANDLER_UPDATE: {
+ uint32_t token, val;
+ int i;
+
+ if ((error = suser(kauth_cred_get(), &(p->p_acflag)))) {
+ break;
+ }
+
+ if (!nspace_is_special_process(p)) {
+ error = EINVAL;
+ break;
+ }
+
+ token = ((uint32_t *)data)[0];
+ val = ((uint32_t *)data)[1];
+
+ lck_mtx_lock(&nspace_handler_lock);
+
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].token == token) {
+ break; /* exit for loop, not case stmt */
+ }
+ }
+
+ if (i >= MAX_NSPACE_ITEMS) {
+ error = ENOENT;
+ } else {
+ //
+ // if this bit is set, when resolve_nspace_item() times out
+ // it will loop and go back to sleep.
+ //
+ nspace_items[i].flags |= NSPACE_ITEM_RESET_TIMER;
+ }
+
+ lck_mtx_unlock(&nspace_handler_lock);
+
+ if (error) {
+ printf("nspace-handler-update: did not find token %u\n", token);
+ }
+ }
+ break;
+
+ case FSCTL_NAMESPACE_HANDLER_UNBLOCK: {
+ uint32_t token, val;
+ int i;
+
+ if ((error = suser(kauth_cred_get(), &(p->p_acflag)))) {
+ break;
+ }
+
+ if (!nspace_is_special_process(p)) {
+ error = EINVAL;
+ break;
+ }
+
+ token = ((uint32_t *)data)[0];
+ val = ((uint32_t *)data)[1];
+
+ lck_mtx_lock(&nspace_handler_lock);
+
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].token == token) {
+ break; /* exit for loop, not case statement */
+ }
+ }
+
+ if (i >= MAX_NSPACE_ITEMS) {
+ printf("nspace-handler-unblock: did not find token %u\n", token);
+ error = ENOENT;
+ } else {
+ if (val == 0 && nspace_items[i].vp) {
+ vnode_lock_spin(nspace_items[i].vp);
+ nspace_items[i].vp->v_flag &= ~VNEEDSSNAPSHOT;
+ vnode_unlock(nspace_items[i].vp);
+ }
+
+ nspace_items[i].vp = NULL;
+ nspace_items[i].arg = NULL;
+ nspace_items[i].op = 0;
+ nspace_items[i].vid = 0;
+ nspace_items[i].flags = NSPACE_ITEM_DONE;
+ nspace_items[i].token = 0;
+
+ wakeup((caddr_t)&(nspace_items[i].vp));
+ }
+
+ lck_mtx_unlock(&nspace_handler_lock);
+ }
+ break;
+
+ case FSCTL_NAMESPACE_HANDLER_CANCEL: {
+ uint32_t token, val;
+ int i;
+
+ if ((error = suser(kauth_cred_get(), &(p->p_acflag)))) {
+ break;
+ }
+
+ if (!nspace_is_special_process(p)) {
+ error = EINVAL;
+ break;
+ }
+
+ token = ((uint32_t *)data)[0];
+ val = ((uint32_t *)data)[1];
-freeandexit:
+ lck_mtx_lock(&nspace_handler_lock);
- FREE(searchparams1,M_TEMP);
+ for(i=0; i < MAX_NSPACE_ITEMS; i++) {
+ if (nspace_items[i].token == token) {
+ break; /* exit for loop, not case stmt */
+ }
+ }
- return(error);
+ if (i >= MAX_NSPACE_ITEMS) {
+ printf("nspace-handler-cancel: did not find token %u\n", token);
+ error = ENOENT;
+ } else {
+ if (nspace_items[i].vp) {
+ vnode_lock_spin(nspace_items[i].vp);
+ nspace_items[i].vp->v_flag &= ~VNEEDSSNAPSHOT;
+ vnode_unlock(nspace_items[i].vp);
+ }
+ nspace_items[i].vp = NULL;
+ nspace_items[i].arg = NULL;
+ nspace_items[i].vid = 0;
+ nspace_items[i].token = val;
+ nspace_items[i].flags &= ~NSPACE_ITEM_PROCESSING;
+ nspace_items[i].flags |= NSPACE_ITEM_CANCELLED;
-} /* end of searchfs system call */
+ wakeup((caddr_t)&(nspace_items[i].vp));
+ }
+ lck_mtx_unlock(&nspace_handler_lock);
+ }
+ break;
-/*
- * Make a filesystem-specific control call:
- */
-/* ARGSUSED */
-static int
-fsctl_internal(proc_t p, vnode_t *arg_vp, u_long cmd, user_addr_t udata, u_long options, vfs_context_t ctx)
-{
- int error=0;
- boolean_t is64bit;
- u_int size;
-#define STK_PARAMS 128
- char stkbuf[STK_PARAMS];
- caddr_t data, memp;
- vnode_t vp = *arg_vp;
+ case FSCTL_NAMESPACE_HANDLER_SET_SNAPSHOT_TIME: {
+ if ((error = suser(kauth_cred_get(), &(current_proc()->p_acflag)))) {
+ break;
+ }
- size = IOCPARM_LEN(cmd);
- if (size > IOCPARM_MAX) return (EINVAL);
+ // we explicitly do not do the namespace_handler_proc check here
- is64bit = proc_is64bit(p);
+ lck_mtx_lock(&nspace_handler_lock);
+ snapshot_timestamp = ((uint32_t *)data)[0];
+ wakeup(&nspace_item_idx);
+ lck_mtx_unlock(&nspace_handler_lock);
+ printf("nspace-handler-set-snapshot-time: %d\n", (int)snapshot_timestamp);
- memp = NULL;
- if (size > sizeof (stkbuf)) {
- if ((memp = (caddr_t)kalloc(size)) == 0) return ENOMEM;
- data = memp;
- } else {
- data = &stkbuf[0];
- };
-
- if (cmd & IOC_IN) {
- if (size) {
- error = copyin(udata, data, size);
- if (error) goto FSCtl_Exit;
- } else {
- if (is64bit) {
- *(user_addr_t *)data = udata;
- }
- else {
- *(uint32_t *)data = (uint32_t)udata;
- }
- };
- } else if ((cmd & IOC_OUT) && size) {
- /*
- * Zero the buffer so the user always
- * gets back something deterministic.
- */
- bzero(data, size);
- } else if (cmd & IOC_VOID) {
- if (is64bit) {
- *(user_addr_t *)data = udata;
- }
- else {
- *(uint32_t *)data = (uint32_t)udata;
- }
- }
+ }
+ break;
- /* Check to see if it's a generic command */
- if (IOCBASECMD(cmd) == FSCTL_SYNC_VOLUME) {
- mount_t mp = vp->v_mount;
- int arg = *(uint32_t*)data;
-
- /* record vid of vp so we can drop it below. */
- uint32_t vvid = vp->v_id;
+ case FSCTL_NAMESPACE_ALLOW_DMG_SNAPSHOT_EVENTS:
+ {
+ if ((error = suser(kauth_cred_get(), &(current_proc()->p_acflag)))) {
+ break;
+ }
- /*
- * Then grab mount_iterref so that we can release the vnode.
- * Without this, a thread may call vnode_iterate_prepare then
- * get into a deadlock because we've never released the root vp
- */
- error = mount_iterref (mp, 0);
- if (error) {
- goto FSCtl_Exit;
- }
- vnode_put(vp);
+ lck_mtx_lock(&nspace_handler_lock);
+ nspace_allow_virtual_devs = ((uint32_t *)data)[0];
+ lck_mtx_unlock(&nspace_handler_lock);
+ printf("nspace-snapshot-handler will%s allow events on disk-images\n",
+ nspace_allow_virtual_devs ? "" : " NOT");
+ error = 0;
- /* issue the sync for this volume */
- (void)sync_callback(mp, (arg & FSCTL_SYNC_WAIT) ? &arg : NULL);
-
- /*
- * Then release the mount_iterref once we're done syncing; it's not
- * needed for the VNOP_IOCTL below
- */
- mount_iterdrop(mp);
+ }
+ break;
- if (arg & FSCTL_SYNC_FULLSYNC) {
- /* re-obtain vnode iocount on the root vp, if possible */
- error = vnode_getwithvid (vp, vvid);
- if (error == 0) {
- error = VNOP_IOCTL(vp, F_FULLFSYNC, (caddr_t)NULL, 0, ctx);
- vnode_put (vp);
+ case FSCTL_SET_FSTYPENAME_OVERRIDE:
+ {
+ if ((error = suser(kauth_cred_get(), &(current_proc()->p_acflag)))) {
+ break;
+ }
+ if (vp->v_mount) {
+ mount_lock(vp->v_mount);
+ if (data[0] != 0) {
+ strlcpy(&vp->v_mount->fstypename_override[0], data, MFSTYPENAMELEN);
+ vp->v_mount->mnt_kern_flag |= MNTK_TYPENAME_OVERRIDE;
+ if (vfs_isrdonly(vp->v_mount) && strcmp(vp->v_mount->fstypename_override, "mtmfs") == 0) {
+ vp->v_mount->mnt_kern_flag |= MNTK_EXTENDED_SECURITY;
+ vp->v_mount->mnt_kern_flag &= ~MNTK_AUTH_OPAQUE;
+ }
+ } else {
+ if (strcmp(vp->v_mount->fstypename_override, "mtmfs") == 0) {
+ vp->v_mount->mnt_kern_flag &= ~MNTK_EXTENDED_SECURITY;
+ }
+ vp->v_mount->mnt_kern_flag &= ~MNTK_TYPENAME_OVERRIDE;
+ vp->v_mount->fstypename_override[0] = '\0';
+ }
+ mount_unlock(vp->v_mount);
}
}
- /* mark the argument VP as having been released */
- *arg_vp = NULL;
-
- } else if (IOCBASECMD(cmd) == FSCTL_SET_PACKAGE_EXTS) {
- user_addr_t ext_strings;
- uint32_t num_entries;
- uint32_t max_width;
-
- if ( (is64bit && size != sizeof(user64_package_ext_info))
- || (is64bit == 0 && size != sizeof(user32_package_ext_info))) {
-
- // either you're 64-bit and passed a 64-bit struct or
- // you're 32-bit and passed a 32-bit struct. otherwise
- // it's not ok.
- error = EINVAL;
- goto FSCtl_Exit;
- }
+ break;
+
+ default: {
+ /* Invoke the filesystem-specific code */
+ error = VNOP_IOCTL(vp, IOCBASECMD(cmd), data, options, ctx);
+ }
- if (is64bit) {
- ext_strings = ((user64_package_ext_info *)data)->strings;
- num_entries = ((user64_package_ext_info *)data)->num_entries;
- max_width = ((user64_package_ext_info *)data)->max_width;
- } else {
- ext_strings = CAST_USER_ADDR_T(((user32_package_ext_info *)data)->strings);
- num_entries = ((user32_package_ext_info *)data)->num_entries;
- max_width = ((user32_package_ext_info *)data)->max_width;
- }
-
- error = set_package_extensions_table(ext_strings, num_entries, max_width);
+ } /* end switch stmt */
- } else if (IOCBASECMD(cmd) == FSCTL_WAIT_FOR_SYNC) {
- error = tsleep((caddr_t)&sync_wait_time, PVFS|PCATCH, "sync-wait", 0);
- if (error == 0) {
- *(uint32_t *)data = (uint32_t)sync_wait_time;
- error = 0;
- } else {
- error *= -1;
- }
-
- } else {
- /* Invoke the filesystem-specific code */
- error = VNOP_IOCTL(vp, IOCBASECMD(cmd), data, options, ctx);
- }
-
-
/*
- * Copy any data to user, size was
+ * if no errors, copy any data to user. Size was
* already set and checked above.
*/
if (error == 0 && (cmd & IOC_OUT) && size)
error = copyout(data, udata, size);
-FSCtl_Exit:
- if (memp) kfree(memp, size);
+ if (memp) {
+ kfree(memp, size);
+ }
return error;
}
/* Get the vnode for the file we are getting info on: */
nameiflags = 0;
if ((uap->options & FSOPT_NOFOLLOW) == 0) nameiflags |= FOLLOW;
- NDINIT(&nd, LOOKUP, nameiflags | AUDITVNPATH1, UIO_USERSPACE,
- uap->path, ctx);
+ NDINIT(&nd, LOOKUP, OP_FSCTL, nameiflags | AUDITVNPATH1,
+ UIO_USERSPACE, uap->path, ctx);
if ((error = namei(&nd))) goto done;
vp = nd.ni_vp;
nameidone(&nd);
/* Get the vnode for the file we are getting info on: */
if ((error = file_vnode(uap->fd, &vp)))
- goto done;
+ return error;
fd = uap->fd;
if ((error = vnode_getwithref(vp))) {
- goto done;
+ file_drop(fd);
+ return error;
}
#if CONFIG_MACF
- error = mac_mount_check_fsctl(ctx, vnode_mount(vp), uap->cmd);
- if (error) {
- goto done;
+ if ((error = mac_mount_check_fsctl(ctx, vnode_mount(vp), uap->cmd))) {
+ file_drop(fd);
+ vnode_put(vp);
+ return error;
}
#endif
error = fsctl_internal(p, &vp, uap->cmd, (user_addr_t)uap->data, uap->options, ctx);
-done:
- if (fd != -1)
- file_drop(fd);
+ file_drop(fd);
- if (vp)
+ /*validate vp; fsctl_internal() can drop iocount and reset vp to NULL*/
+ if (vp) {
vnode_put(vp);
+ }
+
return error;
}
/* end of fsctl system call */
-/*
- * An in-kernel sync for power management to call.
- */
-__private_extern__ int
-sync_internal(void)
-{
- int error;
-
- struct sync_args data;
-
- int retval[2];
-
-
- error = sync(current_proc(), &data, &retval[0]);
-
-
- return (error);
-} /* end of sync_internal call */
-
-
/*
* Retrieve the data of an extended attribute.
*/
return (EINVAL);
nameiflags = (uap->options & XATTR_NOFOLLOW) ? 0 : FOLLOW;
- NDINIT(&nd, LOOKUP, nameiflags, spacetype, uap->path, ctx);
+ NDINIT(&nd, LOOKUP, OP_GETXATTR, nameiflags, spacetype, uap->path, ctx);
if ((error = namei(&nd))) {
return (error);
}
goto out;
}
if (xattr_protected(attrname)) {
- error = EPERM;
- goto out;
+ if (!vfs_context_issuser(ctx) || strcmp(attrname, "com.apple.system.Security") != 0) {
+ error = EPERM;
+ goto out;
+ }
}
/*
* the specific check for 0xffffffff is a hack to preserve
if (uap->size == 0xffffffff || uap->size == (size_t)-1)
goto no_uio;
- if (uap->size > (size_t)XATTR_MAXSIZE)
- uap->size = XATTR_MAXSIZE;
-
if (uap->value) {
+ if (uap->size > (size_t)XATTR_MAXSIZE)
+ uap->size = XATTR_MAXSIZE;
+
auio = uio_createwithbuffer(1, uap->position, spacetype, UIO_READ,
&uio_buf[0], sizeof(uio_buf));
uio_addiov(auio, uap->value, uap->size);
return (EINVAL);
if ((error = copyinstr(uap->attrname, attrname, sizeof(attrname), &namelen) != 0)) {
- return (error);
+ if (error == EPERM) {
+ /* if the string won't fit in attrname, copyinstr emits EPERM */
+ return (ENAMETOOLONG);
+ }
+ /* Otherwise return the default error from copyinstr to detect ERANGE, etc */
+ return error;
}
if (xattr_protected(attrname))
return(EPERM);
}
nameiflags = (uap->options & XATTR_NOFOLLOW) ? 0 : FOLLOW;
- NDINIT(&nd, LOOKUP, nameiflags, spacetype, uap->path, ctx);
+ NDINIT(&nd, LOOKUP, OP_SETXATTR, nameiflags, spacetype, uap->path, ctx);
if ((error = namei(&nd))) {
return (error);
}
size_t namelen;
int error;
char uio_buf[ UIO_SIZEOF(1) ];
+#if CONFIG_FSE
vfs_context_t ctx = vfs_context_current();
+#endif
if (uap->options & (XATTR_NOFOLLOW | XATTR_NOSECURITY | XATTR_NODEFAULT))
return (EINVAL);
if ((error = copyinstr(uap->attrname, attrname, sizeof(attrname), &namelen) != 0)) {
- return (error);
+ if (error == EPERM) {
+ /* if the string won't fit in attrname, copyinstr emits EPERM */
+ return (ENAMETOOLONG);
+ }
+ /* Otherwise return the default error from copyinstr to detect ERANGE, etc */
+ return error;
}
if (xattr_protected(attrname))
return(EPERM);
if (xattr_protected(attrname))
return(EPERM);
nameiflags = (uap->options & XATTR_NOFOLLOW) ? 0 : FOLLOW;
- NDINIT(&nd, LOOKUP, nameiflags, spacetype, uap->path, ctx);
+ NDINIT(&nd, LOOKUP, OP_REMOVEXATTR, nameiflags, spacetype, uap->path, ctx);
if ((error = namei(&nd))) {
return (error);
}
char attrname[XATTR_MAXNAMELEN+1];
size_t namelen;
int error;
+#if CONFIG_FSE
vfs_context_t ctx = vfs_context_current();
+#endif
if (uap->options & (XATTR_NOFOLLOW | XATTR_NOSECURITY | XATTR_NODEFAULT))
return (EINVAL);
if (uap->options & (XATTR_NOSECURITY | XATTR_NODEFAULT))
return (EINVAL);
- nameiflags = ((uap->options & XATTR_NOFOLLOW) ? 0 : FOLLOW) | NOTRIGGER;
- NDINIT(&nd, LOOKUP, nameiflags, spacetype, uap->path, ctx);
+ nameiflags = (uap->options & XATTR_NOFOLLOW) ? 0 : FOLLOW;
+ NDINIT(&nd, LOOKUP, OP_LISTXATTR, nameiflags, spacetype, uap->path, ctx);
if ((error = namei(&nd))) {
return (error);
}
vp = nd.ni_vp;
nameidone(&nd);
if (uap->namebuf != 0 && uap->bufsize > 0) {
- auio = uio_createwithbuffer(1, 0, spacetype,
- UIO_READ, &uio_buf[0], sizeof(uio_buf));
+ auio = uio_createwithbuffer(1, 0, spacetype, UIO_READ,
+ &uio_buf[0], sizeof(uio_buf));
uio_addiov(auio, uap->namebuf, uap->bufsize);
}
return (error);
}
+static int fsgetpath_internal(
+ vfs_context_t ctx, int volfs_id, uint64_t objid,
+ vm_size_t bufsize, caddr_t buf, int *pathlen)
+{
+ int error;
+ struct mount *mp = NULL;
+ vnode_t vp;
+ int length;
+ int bpflags;
+
+ if (bufsize > PAGE_SIZE) {
+ return (EINVAL);
+ }
+
+ if (buf == NULL) {
+ return (ENOMEM);
+ }
+
+ if ((mp = mount_lookupby_volfsid(volfs_id, 1)) == NULL) {
+ error = ENOTSUP; /* unexpected failure */
+ return ENOTSUP;
+ }
+
+unionget:
+ if (objid == 2) {
+ error = VFS_ROOT(mp, &vp, ctx);
+ } else {
+ error = VFS_VGET(mp, (ino64_t)objid, &vp, ctx);
+ }
+
+ if (error == ENOENT && (mp->mnt_flag & MNT_UNION)) {
+ /*
+ * If the fileid isn't found and we're in a union
+ * mount volume, then see if the fileid is in the
+ * mounted-on volume.
+ */
+ struct mount *tmp = mp;
+ mp = vnode_mount(tmp->mnt_vnodecovered);
+ vfs_unbusy(tmp);
+ if (vfs_busy(mp, LK_NOWAIT) == 0)
+ goto unionget;
+ } else {
+ vfs_unbusy(mp);
+ }
+
+ if (error) {
+ return error;
+ }
+
+#if CONFIG_MACF
+ error = mac_vnode_check_fsgetpath(ctx, vp);
+ if (error) {
+ vnode_put(vp);
+ return error;
+ }
+#endif
+
+ /* Obtain the absolute path to this vnode. */
+ bpflags = vfs_context_suser(ctx) ? BUILDPATH_CHECKACCESS : 0;
+ bpflags |= BUILDPATH_CHECK_MOVED;
+ error = build_path(vp, buf, bufsize, &length, bpflags, ctx);
+ vnode_put(vp);
+
+ if (error) {
+ goto out;
+ }
+
+ AUDIT_ARG(text, buf);
+
+ if (kdebug_enable) {
+ long dbg_parms[NUMPARMS];
+ int dbg_namelen;
+
+ dbg_namelen = (int)sizeof(dbg_parms);
+
+ if (length < dbg_namelen) {
+ memcpy((char *)dbg_parms, buf, length);
+ memset((char *)dbg_parms + length, 0, dbg_namelen - length);
+
+ dbg_namelen = length;
+ } else {
+ memcpy((char *)dbg_parms, buf + (length - dbg_namelen), dbg_namelen);
+ }
+
+ kdebug_lookup_gen_events(dbg_parms, dbg_namelen, (void *)vp, TRUE);
+ }
+
+ *pathlen = (user_ssize_t)length; /* may be superseded by error */
+
+out:
+ return (error);
+}
+
/*
* Obtain the full pathname of a file system object by id.
*
int
fsgetpath(__unused proc_t p, struct fsgetpath_args *uap, user_ssize_t *retval)
{
- vnode_t vp;
- struct mount *mp = NULL;
vfs_context_t ctx = vfs_context_current();
fsid_t fsid;
char *realpath;
- int bpflags;
int length;
int error;
AUDIT_ARG(value32, fsid.val[0]);
AUDIT_ARG(value64, uap->objid);
/* Restrict output buffer size for now. */
+
if (uap->bufsize > PAGE_SIZE) {
return (EINVAL);
}
if (realpath == NULL) {
return (ENOMEM);
}
- /* Find the target mountpoint. */
- if ((mp = mount_lookupby_volfsid(fsid.val[0], 1)) == NULL) {
- error = ENOTSUP; /* unexpected failure */
- goto out;
- }
- /* Find the target vnode. */
- if (uap->objid == 2) {
- error = VFS_ROOT(mp, &vp, ctx);
- } else {
- error = VFS_VGET(mp, (ino64_t)uap->objid, &vp, ctx);
- }
- vfs_unbusy(mp);
- if (error) {
- goto out;
- }
- /* Obtain the absolute path to this vnode. */
- bpflags = vfs_context_suser(ctx) ? BUILDPATH_CHECKACCESS : 0;
- error = build_path(vp, realpath, uap->bufsize, &length, bpflags, ctx);
- vnode_put(vp);
+
+ error = fsgetpath_internal(
+ ctx, fsid.val[0], uap->objid,
+ uap->bufsize, realpath, &length);
+
if (error) {
goto out;
}
- AUDIT_ARG(text, realpath);
+
error = copyout((caddr_t)realpath, uap->buf, length);
*retval = (user_ssize_t)length; /* may be superseded by error */
sfs.f_ffree = (user64_long_t)sfsp->f_ffree;
sfs.f_fsid = sfsp->f_fsid;
sfs.f_owner = sfsp->f_owner;
- strlcpy(&sfs.f_fstypename[0], &sfsp->f_fstypename[0], MFSNAMELEN);
+ if (mp->mnt_kern_flag & MNTK_TYPENAME_OVERRIDE) {
+ strlcpy(&sfs.f_fstypename[0], &mp->fstypename_override[0], MFSNAMELEN);
+ } else {
+ strlcpy(&sfs.f_fstypename[0], &sfsp->f_fstypename[0], MFSNAMELEN);
+ }
strlcpy(&sfs.f_mntonname[0], &sfsp->f_mntonname[0], MNAMELEN);
strlcpy(&sfs.f_mntfromname[0], &sfsp->f_mntfromname[0], MNAMELEN);
sfs.f_ffree = (user32_long_t)sfsp->f_ffree;
sfs.f_fsid = sfsp->f_fsid;
sfs.f_owner = sfsp->f_owner;
- strlcpy(&sfs.f_fstypename[0], &sfsp->f_fstypename[0], MFSNAMELEN);
+ if (mp->mnt_kern_flag & MNTK_TYPENAME_OVERRIDE) {
+ strlcpy(&sfs.f_fstypename[0], &mp->fstypename_override[0], MFSNAMELEN);
+ } else {
+ strlcpy(&sfs.f_fstypename[0], &sfsp->f_fstypename[0], MFSNAMELEN);
+ }
strlcpy(&sfs.f_mntonname[0], &sfsp->f_mntonname[0], MNAMELEN);
strlcpy(&sfs.f_mntfromname[0], &sfsp->f_mntfromname[0], MNAMELEN);
usbp->st_qspare[0] = sbp->st_qspare[0];
usbp->st_qspare[1] = sbp->st_qspare[1];
}
+
+/*
+ * Purge buffer cache for simulating cold starts
+ */
+static int vnode_purge_callback(struct vnode *vp, __unused void *cargs)
+{
+ ubc_msync(vp, (off_t)0, ubc_getsize(vp), NULL /* off_t *resid_off */, UBC_PUSHALL | UBC_INVALIDATE);
+
+ return VNODE_RETURNED;
+}
+
+static int vfs_purge_callback(mount_t mp, __unused void * arg)
+{
+ vnode_iterate(mp, VNODE_WAIT | VNODE_ITERATE_ALL, vnode_purge_callback, NULL);
+
+ return VFS_RETURNED;
+}
+
+int
+vfs_purge(__unused struct proc *p, __unused struct vfs_purge_args *uap, __unused int32_t *retval)
+{
+ if (!kauth_cred_issuser(kauth_cred_get()))
+ return EPERM;
+
+ vfs_iterate(0/* flags */, vfs_purge_callback, NULL);
+
+ return 0;
+}
+
projects / apple / xnu.git / blobdiff