4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
27 /* All Rights Reserved */
30 * University Copyright- Copyright (c) 1982, 1986, 1988
31 * The Regents of the University of California
34 * University Acknowledgment- Portions of this document are derived from
35 * software developed by the University of California, Berkeley, and its
39 #include <sys/types.h>
40 #include <sys/t_lock.h>
41 #include <sys/param.h>
42 #include <sys/errno.h>
44 #include <sys/fstyp.h>
46 #include <sys/systm.h>
48 #include <sys/mount.h>
50 #include <sys/vfs_opreg.h>
52 #include <sys/mntent.h>
54 #include <sys/statvfs.h>
55 #include <sys/statfs.h>
57 #include <sys/vnode.h>
58 #include <sys/rwstlock.h>
62 #include <sys/atomic.h>
63 #include <sys/cmn_err.h>
66 #include <sys/debug.h>
67 #include <sys/vnode.h>
68 #include <sys/modctl.h>
70 #include <sys/pathname.h>
71 #include <sys/bootconf.h>
72 #include <sys/dumphdr.h>
73 #include <sys/dc_ki.h>
75 #include <sys/sunddi.h>
76 #include <sys/sysmacros.h>
78 #include <sys/policy.h>
80 #include <sys/objfs.h>
81 #include <sys/console.h>
82 #include <sys/reboot.h>
87 #include <sys/bootprops.h>
91 #include <fs/fs_subr.h>
92 /* Private interfaces to create vopstats-related data structures */
93 extern void initialize_vopstats(vopstats_t
*);
94 extern vopstats_t
*get_fstype_vopstats(struct vfs
*, struct vfssw
*);
95 extern vsk_anchor_t
*get_vskstat_anchor(struct vfs
*);
97 static void vfs_clearmntopt_nolock(mntopts_t
*, const char *, int);
98 static void vfs_setmntopt_nolock(mntopts_t
*, const char *,
99 const char *, int, int);
100 static int vfs_optionisset_nolock(const mntopts_t
*, const char *, char **);
101 static void vfs_freemnttab(struct vfs
*);
102 static void vfs_freeopt(mntopt_t
*);
103 static void vfs_swapopttbl_nolock(mntopts_t
*, mntopts_t
*);
104 static void vfs_swapopttbl(mntopts_t
*, mntopts_t
*);
105 static void vfs_copyopttbl_extend(const mntopts_t
*, mntopts_t
*, int);
106 static void vfs_createopttbl_extend(mntopts_t
*, const char *,
108 static char **vfs_copycancelopt_extend(char **const, int);
109 static void vfs_freecancelopt(char **);
110 static void getrootfs(char **, char **);
111 static int getmacpath(dev_info_t
*, void *);
112 static void vfs_mnttabvp_setup(void);
115 struct ipmnt
*mip_next
;
117 struct vfs
*mip_vfsp
;
120 static kmutex_t vfs_miplist_mutex
;
121 static struct ipmnt
*vfs_miplist
= NULL
;
122 static struct ipmnt
*vfs_miplist_end
= NULL
;
124 static kmem_cache_t
*vfs_cache
; /* Pointer to VFS kmem cache */
129 vnode_t
*rootdir
; /* pointer to root inode vnode. */
130 vnode_t
*devicesdir
; /* pointer to inode of devices root */
131 vnode_t
*devdir
; /* pointer to inode of dev root */
133 char *server_rootpath
; /* root path for diskless clients */
134 char *server_hostname
; /* hostname of diskless server */
136 static struct vfs root
;
137 static struct vfs devices
;
138 static struct vfs dev
;
139 struct vfs
*rootvfs
= &root
; /* pointer to root vfs; head of VFS list. */
140 rvfs_t
*rvfs_list
; /* array of vfs ptrs for vfs hash list */
141 int vfshsz
= 512; /* # of heads/locks in vfs hash arrays */
142 /* must be power of 2! */
143 timespec_t vfs_mnttab_ctime
; /* mnttab created time */
144 timespec_t vfs_mnttab_mtime
; /* mnttab last modified time */
145 char *vfs_dummyfstype
= "\0";
146 struct pollhead vfs_pollhd
; /* for mnttab pollers */
147 struct vnode
*vfs_mntdummyvp
; /* to fake mnttab read/write for file events */
148 int mntfstype
; /* will be set once mnt fs is mounted */
151 * Table for generic options recognized in the VFS layer and acted
152 * on at this level before parsing file system specific options.
153 * The nosuid option is stronger than any of the devices and setuid
154 * options, so those are canceled when nosuid is seen.
156 * All options which are added here need to be added to the
157 * list of standard options in usr/src/cmd/fs.d/fslib.c as well.
160 * VFS Mount options table
162 static char *ro_cancel
[] = { MNTOPT_RW
, NULL
};
163 static char *rw_cancel
[] = { MNTOPT_RO
, NULL
};
164 static char *suid_cancel
[] = { MNTOPT_NOSUID
, NULL
};
165 static char *nosuid_cancel
[] = { MNTOPT_SUID
, MNTOPT_DEVICES
, MNTOPT_NODEVICES
,
166 MNTOPT_NOSETUID
, MNTOPT_SETUID
, NULL
};
167 static char *devices_cancel
[] = { MNTOPT_NODEVICES
, NULL
};
168 static char *nodevices_cancel
[] = { MNTOPT_DEVICES
, NULL
};
169 static char *setuid_cancel
[] = { MNTOPT_NOSETUID
, NULL
};
170 static char *nosetuid_cancel
[] = { MNTOPT_SETUID
, NULL
};
171 static char *nbmand_cancel
[] = { MNTOPT_NONBMAND
, NULL
};
172 static char *nonbmand_cancel
[] = { MNTOPT_NBMAND
, NULL
};
173 static char *exec_cancel
[] = { MNTOPT_NOEXEC
, NULL
};
174 static char *noexec_cancel
[] = { MNTOPT_EXEC
, NULL
};
176 static const mntopt_t mntopts
[] = {
178 * option name cancel options default arg flags
180 { MNTOPT_REMOUNT
, NULL
, NULL
,
181 MO_NODISPLAY
, (void *)0 },
182 { MNTOPT_RO
, ro_cancel
, NULL
, 0,
184 { MNTOPT_RW
, rw_cancel
, NULL
, 0,
186 { MNTOPT_SUID
, suid_cancel
, NULL
, 0,
188 { MNTOPT_NOSUID
, nosuid_cancel
, NULL
, 0,
190 { MNTOPT_DEVICES
, devices_cancel
, NULL
, 0,
192 { MNTOPT_NODEVICES
, nodevices_cancel
, NULL
, 0,
194 { MNTOPT_SETUID
, setuid_cancel
, NULL
, 0,
196 { MNTOPT_NOSETUID
, nosetuid_cancel
, NULL
, 0,
198 { MNTOPT_NBMAND
, nbmand_cancel
, NULL
, 0,
200 { MNTOPT_NONBMAND
, nonbmand_cancel
, NULL
, 0,
202 { MNTOPT_EXEC
, exec_cancel
, NULL
, 0,
204 { MNTOPT_NOEXEC
, noexec_cancel
, NULL
, 0,
208 const mntopts_t vfs_mntopts
= {
209 sizeof (mntopts
) / sizeof (mntopt_t
),
210 (mntopt_t
*)&mntopts
[0]
214 * File system operation dispatch functions.
218 fsop_mount(vfs_t
*vfsp
, vnode_t
*mvp
, struct mounta
*uap
, cred_t
*cr
)
220 return (*(vfsp
)->vfs_op
->vfs_mount
)(vfsp
, mvp
, uap
, cr
);
224 fsop_unmount(vfs_t
*vfsp
, int flag
, cred_t
*cr
)
226 return (*(vfsp
)->vfs_op
->vfs_unmount
)(vfsp
, flag
, cr
);
230 fsop_root(vfs_t
*vfsp
, vnode_t
**vpp
)
233 int ret
= (*(vfsp
)->vfs_op
->vfs_root
)(vfsp
, vpp
);
235 * Make sure this root has a path. With lofs, it is possible to have
238 if (ret
== 0 && vfsp
->vfs_mntpt
!= NULL
&& (*vpp
)->v_path
== NULL
) {
239 mntpt
= vfs_getmntpoint(vfsp
);
240 vn_setpath_str(*vpp
, refstr_value(mntpt
),
241 strlen(refstr_value(mntpt
)));
249 fsop_statfs(vfs_t
*vfsp
, statvfs64_t
*sp
)
251 return (*(vfsp
)->vfs_op
->vfs_statvfs
)(vfsp
, sp
);
255 fsop_sync(vfs_t
*vfsp
, short flag
, cred_t
*cr
)
257 return (*(vfsp
)->vfs_op
->vfs_sync
)(vfsp
, flag
, cr
);
261 fsop_vget(vfs_t
*vfsp
, vnode_t
**vpp
, fid_t
*fidp
)
264 * In order to handle system attribute fids in a manner
265 * transparent to the underlying fs, we embed the fid for
266 * the sysattr parent object in the sysattr fid and tack on
267 * some extra bytes that only the sysattr layer knows about.
269 * This guarantees that sysattr fids are larger than other fids
270 * for this vfs. If the vfs supports the sysattr view interface
271 * (as indicated by VFSFT_SYSATTR_VIEWS), we cannot have a size
272 * collision with XATTR_FIDSZ.
274 if (vfs_has_feature(vfsp
, VFSFT_SYSATTR_VIEWS
) &&
275 fidp
->fid_len
== XATTR_FIDSZ
)
276 return (xattr_dir_vget(vfsp
, vpp
, fidp
));
278 return (*(vfsp
)->vfs_op
->vfs_vget
)(vfsp
, vpp
, fidp
);
282 fsop_mountroot(vfs_t
*vfsp
, enum whymountroot reason
)
284 return (*(vfsp
)->vfs_op
->vfs_mountroot
)(vfsp
, reason
);
288 fsop_freefs(vfs_t
*vfsp
)
290 (*(vfsp
)->vfs_op
->vfs_freevfs
)(vfsp
);
294 fsop_vnstate(vfs_t
*vfsp
, vnode_t
*vp
, vntrans_t nstate
)
296 return ((*(vfsp
)->vfs_op
->vfs_vnstate
)(vfsp
, vp
, nstate
));
300 fsop_sync_by_kind(int fstype
, short flag
, cred_t
*cr
)
302 ASSERT((fstype
>= 0) && (fstype
< nfstype
));
304 if (ALLOCATED_VFSSW(&vfssw
[fstype
]) && VFS_INSTALLED(&vfssw
[fstype
]))
305 return (*vfssw
[fstype
].vsw_vfsops
.vfs_sync
) (NULL
, flag
, cr
);
311 * File system initialization. vfs_setfsops() must be called from a file
312 * system's init routine.
316 fs_copyfsops(const fs_operation_def_t
*template, vfsops_t
*actual
,
319 static const fs_operation_trans_def_t vfs_ops_table
[] = {
320 VFSNAME_MOUNT
, offsetof(vfsops_t
, vfs_mount
),
323 VFSNAME_UNMOUNT
, offsetof(vfsops_t
, vfs_unmount
),
326 VFSNAME_ROOT
, offsetof(vfsops_t
, vfs_root
),
329 VFSNAME_STATVFS
, offsetof(vfsops_t
, vfs_statvfs
),
332 VFSNAME_SYNC
, offsetof(vfsops_t
, vfs_sync
),
333 (fs_generic_func_p
) fs_sync
,
334 (fs_generic_func_p
) fs_sync
, /* No errors allowed */
336 VFSNAME_VGET
, offsetof(vfsops_t
, vfs_vget
),
339 VFSNAME_MOUNTROOT
, offsetof(vfsops_t
, vfs_mountroot
),
342 VFSNAME_FREEVFS
, offsetof(vfsops_t
, vfs_freevfs
),
343 (fs_generic_func_p
)fs_freevfs
,
344 (fs_generic_func_p
)fs_freevfs
, /* Shouldn't fail */
346 VFSNAME_VNSTATE
, offsetof(vfsops_t
, vfs_vnstate
),
347 (fs_generic_func_p
)fs_nosys
,
348 (fs_generic_func_p
)fs_nosys
,
353 return (fs_build_vector(actual
, unused_ops
, vfs_ops_table
, template));
359 if (strcmp(rootfs
.bo_fstype
, MNTTYPE_ZFS
) == 0)
364 vfs_setfsops(int fstype
, const fs_operation_def_t
*template, vfsops_t
**actual
)
370 * Verify that fstype refers to a valid fs. Note that
371 * 0 is valid since it's used to set "stray" ops.
373 if ((fstype
< 0) || (fstype
>= nfstype
))
376 if (!ALLOCATED_VFSSW(&vfssw
[fstype
]))
379 /* Set up the operations vector. */
381 error
= fs_copyfsops(template, &vfssw
[fstype
].vsw_vfsops
, &unused_ops
);
386 vfssw
[fstype
].vsw_flag
|= VSW_INSTALLED
;
389 *actual
= &vfssw
[fstype
].vsw_vfsops
;
393 cmn_err(CE_WARN
, "vfs_setfsops: %s: %d operations supplied "
394 "but not used", vfssw
[fstype
].vsw_name
, unused_ops
);
401 vfs_makefsops(const fs_operation_def_t
*template, vfsops_t
**actual
)
406 *actual
= (vfsops_t
*)kmem_alloc(sizeof (vfsops_t
), KM_SLEEP
);
408 error
= fs_copyfsops(template, *actual
, &unused_ops
);
410 kmem_free(*actual
, sizeof (vfsops_t
));
419 * Free a vfsops structure created as a result of vfs_makefsops().
420 * NOTE: For a vfsops structure initialized by vfs_setfsops(), use
421 * vfs_freevfsops_by_type().
424 vfs_freevfsops(vfsops_t
*vfsops
)
426 kmem_free(vfsops
, sizeof (vfsops_t
));
430 * Since the vfsops structure is part of the vfssw table and wasn't
431 * really allocated, we're not really freeing anything. We keep
432 * the name for consistency with vfs_freevfsops(). We do, however,
433 * need to take care of a little bookkeeping.
434 * NOTE: For a vfsops structure created by vfs_setfsops(), use
435 * vfs_freevfsops_by_type().
438 vfs_freevfsops_by_type(int fstype
)
441 /* Verify that fstype refers to a loaded fs (and not fsid 0). */
442 if ((fstype
<= 0) || (fstype
>= nfstype
))
446 if ((vfssw
[fstype
].vsw_flag
& VSW_INSTALLED
) == 0) {
451 vfssw
[fstype
].vsw_flag
&= ~VSW_INSTALLED
;
457 /* Support routines used to reference vfs_op */
459 /* Set the operations vector for a vfs */
461 vfs_setops(vfs_t
*vfsp
, vfsops_t
*vfsops
)
465 ASSERT(vfsp
!= NULL
);
466 ASSERT(vfsops
!= NULL
);
470 if (vfsp
->vfs_femhead
== NULL
&&
471 casptr(&vfsp
->vfs_op
, op
, vfsops
) == op
) {
474 fsem_setvfsops(vfsp
, vfsops
);
477 /* Retrieve the operations vector for a vfs */
479 vfs_getops(vfs_t
*vfsp
)
483 ASSERT(vfsp
!= NULL
);
487 if (vfsp
->vfs_femhead
== NULL
&& op
== vfsp
->vfs_op
) {
490 return (fsem_getvfsops(vfsp
));
495 * Returns non-zero (1) if the vfsops matches that of the vfs.
496 * Returns zero (0) if not.
499 vfs_matchops(vfs_t
*vfsp
, vfsops_t
*vfsops
)
501 return (vfs_getops(vfsp
) == vfsops
);
505 * Returns non-zero (1) if the file system has installed a non-default,
506 * non-error vfs_sync routine. Returns zero (0) otherwise.
509 vfs_can_sync(vfs_t
*vfsp
)
511 /* vfs_sync() routine is not the default/error function */
512 return (vfs_getops(vfsp
)->vfs_sync
!= fs_sync
);
516 * Initialize a vfs structure.
519 vfs_init(vfs_t
*vfsp
, vfsops_t
*op
, void *data
)
521 /* Other initialization has been moved to vfs_alloc() */
523 vfsp
->vfs_next
= vfsp
;
524 vfsp
->vfs_prev
= vfsp
;
525 vfsp
->vfs_zone_next
= vfsp
;
526 vfsp
->vfs_zone_prev
= vfsp
;
527 vfsp
->vfs_lofi_minor
= 0;
528 sema_init(&vfsp
->vfs_reflock
, 1, NULL
, SEMA_DEFAULT
, NULL
);
530 vfsp
->vfs_data
= (data
);
531 vfs_setops((vfsp
), (op
));
535 * Allocate and initialize the vfs implementation private data
536 * structure, vfs_impl_t.
539 vfsimpl_setup(vfs_t
*vfsp
)
543 if (vfsp
->vfs_implp
!= NULL
) {
547 vfsp
->vfs_implp
= kmem_alloc(sizeof (vfs_impl_t
), KM_SLEEP
);
548 /* Note that these are #define'd in vfs.h */
549 vfsp
->vfs_vskap
= NULL
;
550 vfsp
->vfs_fstypevsp
= NULL
;
552 /* Set size of counted array, then zero the array */
553 vfsp
->vfs_featureset
[0] = VFS_FEATURE_MAXSZ
- 1;
554 for (i
= 1; i
< VFS_FEATURE_MAXSZ
; i
++) {
555 vfsp
->vfs_featureset
[i
] = 0;
560 * Release the vfs_impl_t structure, if it exists. Some unbundled
561 * filesystems may not use the newer version of vfs and thus
562 * would not contain this implementation private data structure.
565 vfsimpl_teardown(vfs_t
*vfsp
)
567 vfs_impl_t
*vip
= vfsp
->vfs_implp
;
572 kmem_free(vfsp
->vfs_implp
, sizeof (vfs_impl_t
));
573 vfsp
->vfs_implp
= NULL
;
577 * VFS system calls: mount, umount, syssync, statfs, fstatfs, statvfs,
578 * fstatvfs, and sysfs moved to common/syscall.
582 * Update every mounted file system. We call the vfs_sync operation of
583 * each file system type, passing it a NULL vfsp to indicate that all
584 * mounted file systems of that type should be updated.
591 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
592 if (ALLOCATED_VFSSW(vswp
) && VFS_INSTALLED(vswp
)) {
595 (void) (*vswp
->vsw_vfsops
.vfs_sync
)(NULL
, flag
,
597 vfs_unrefvfssw(vswp
);
614 krwlock_t vfssw_lock
; /* lock accesses to vfssw */
617 * Lock for accessing the vfs linked list. Initialized in vfs_mountroot(),
618 * but otherwise should be accessed only via vfs_list_lock() and
619 * vfs_list_unlock(). Also used to protect the timestamp for mods to the list.
621 static krwlock_t vfslist
;
624 * Mount devfs on /devices. This is done right after root is mounted
625 * to provide device access support for the system
628 vfs_mountdevices(void)
632 struct mounta mounta
= { /* fake mounta for devfs_mount() */
644 * _init devfs module to fill in the vfssw
646 if (modload("fs", "devfs") == -1)
647 panic("Cannot _init devfs module");
653 vsw
= vfs_getvfsswbyname("devfs");
654 VFS_INIT(&devices
, &vsw
->vsw_vfsops
, NULL
);
660 if (lookupname("/devices", UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
))
661 panic("Cannot find /devices");
664 * Perform the mount of /devices
666 if (VFS_MOUNT(&devices
, mvp
, &mounta
, CRED()))
667 panic("Cannot mount /devices");
672 * Set appropriate members and add to vfs list for mnttab display
674 vfs_setresource(&devices
, "/devices");
675 vfs_setmntpoint(&devices
, "/devices");
678 * Hold the root of /devices so it won't go away
680 if (VFS_ROOT(&devices
, &devicesdir
))
681 panic("vfs_mountdevices: not devices root");
683 if (vfs_lock(&devices
) != 0) {
685 cmn_err(CE_NOTE
, "Cannot acquire vfs_lock of /devices");
689 if (vn_vfswlock(mvp
) != 0) {
690 vfs_unlock(&devices
);
692 cmn_err(CE_NOTE
, "Cannot acquire vfswlock of /devices");
696 vfs_add(mvp
, &devices
, 0);
698 vfs_unlock(&devices
);
703 * mount the first instance of /dev to root and remain mounted
710 struct mounta mounta
= { /* fake mounta for sdev_mount() */
713 MS_SYSSPACE
| MS_OVERLAY
,
722 * _init dev module to fill in the vfssw
724 if (modload("fs", "dev") == -1)
725 cmn_err(CE_PANIC
, "Cannot _init dev module\n");
731 vsw
= vfs_getvfsswbyname("dev");
732 VFS_INIT(&dev
, &vsw
->vsw_vfsops
, NULL
);
738 if (lookupname("/dev", UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
))
739 cmn_err(CE_PANIC
, "Cannot find /dev\n");
742 * Perform the mount of /dev
744 if (VFS_MOUNT(&dev
, mvp
, &mounta
, CRED()))
745 cmn_err(CE_PANIC
, "Cannot mount /dev 1\n");
750 * Set appropriate members and add to vfs list for mnttab display
752 vfs_setresource(&dev
, "/dev");
753 vfs_setmntpoint(&dev
, "/dev");
756 * Hold the root of /dev so it won't go away
758 if (VFS_ROOT(&dev
, &devdir
))
759 cmn_err(CE_PANIC
, "vfs_mountdev1: not dev root");
761 if (vfs_lock(&dev
) != 0) {
763 cmn_err(CE_NOTE
, "Cannot acquire vfs_lock of /dev");
767 if (vn_vfswlock(mvp
) != 0) {
770 cmn_err(CE_NOTE
, "Cannot acquire vfswlock of /dev");
774 vfs_add(mvp
, &dev
, 0);
781 * Mount required filesystem. This is done right after root is mounted.
784 vfs_mountfs(char *module
, char *spec
, char *path
)
787 struct mounta mounta
;
790 mounta
.flags
= MS_SYSSPACE
| MS_DATA
;
791 mounta
.fstype
= module
;
794 if (lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
)) {
795 cmn_err(CE_WARN
, "Cannot find %s", path
);
798 if (domount(NULL
, &mounta
, mvp
, CRED(), &vfsp
))
799 cmn_err(CE_WARN
, "Cannot mount %s", path
);
806 * vfs_mountroot is called by main() to mount the root filesystem.
811 struct vnode
*rvp
= NULL
;
817 rw_init(&vfssw_lock
, NULL
, RW_DEFAULT
, NULL
);
818 rw_init(&vfslist
, NULL
, RW_DEFAULT
, NULL
);
821 * Alloc the vfs hash bucket array and locks
823 rvfs_list
= kmem_zalloc(vfshsz
* sizeof (rvfs_t
), KM_SLEEP
);
826 * Call machine-dependent routine "rootconf" to choose a root
830 panic("vfs_mountroot: cannot mount root");
832 * Get vnode for '/'. Set up rootdir, u.u_rdir and u.u_cdir
833 * to point to it. These are used by lookuppn() so that it
834 * knows where to start from ('/' or '.').
836 vfs_setmntpoint(rootvfs
, "/");
837 if (VFS_ROOT(rootvfs
, &rootdir
))
838 panic("vfs_mountroot: no root vnode");
841 * At this point, the process tree consists of p0 and possibly some
842 * direct children of p0. (i.e. there are no grandchildren)
844 * Walk through them all, setting their current directory.
846 mutex_enter(&pidlock
);
847 for (p
= practive
; p
!= NULL
; p
= p
->p_next
) {
848 ASSERT(p
== &p0
|| p
->p_parent
== &p0
);
850 PTOU(p
)->u_cdir
= rootdir
;
851 VN_HOLD(PTOU(p
)->u_cdir
);
852 PTOU(p
)->u_rdir
= NULL
;
854 mutex_exit(&pidlock
);
857 * Setup the global zone's rootvp, now that it exists.
859 global_zone
->zone_rootvp
= rootdir
;
860 VN_HOLD(global_zone
->zone_rootvp
);
863 * Notify the module code that it can begin using the
864 * root filesystem instead of the boot program's services.
869 * Special handling for a ZFS root file system.
874 * Set up mnttab information for root
876 vfs_setresource(rootvfs
, rootfs
.bo_name
);
879 * Notify cluster software that the root filesystem is available.
883 /* Now that we're all done with the root FS, set up its vopstats */
884 if ((vswp
= vfs_getvfsswbyvfsops(vfs_getops(rootvfs
))) != NULL
) {
885 /* Set flag for statistics collection */
886 if (vswp
->vsw_flag
& VSW_STATS
) {
887 initialize_vopstats(&rootvfs
->vfs_vopstats
);
888 rootvfs
->vfs_flag
|= VFS_STATS
;
889 rootvfs
->vfs_fstypevsp
=
890 get_fstype_vopstats(rootvfs
, vswp
);
891 rootvfs
->vfs_vskap
= get_vskstat_anchor(rootvfs
);
893 vfs_unrefvfssw(vswp
);
897 * Mount /devices, /dev instance 1, /system/contract, /etc/mnttab,
898 * /etc/svc/volatile, /etc/dfs/sharetab, /system/object, and /proc.
903 vfs_mountfs("ctfs", "ctfs", CTFS_ROOT
);
904 vfs_mountfs("proc", "/proc", "/proc");
905 vfs_mountfs("mntfs", "/etc/mnttab", "/etc/mnttab");
906 vfs_mountfs("tmpfs", "/etc/svc/volatile", "/etc/svc/volatile");
907 vfs_mountfs("objfs", "objfs", OBJFS_ROOT
);
909 if (getzoneid() == GLOBAL_ZONEID
) {
910 vfs_mountfs("sharefs", "sharefs", "/etc/dfs/sharetab");
915 * This bit of magic can go away when we convert sparc to
916 * the new boot architecture based on ramdisk.
918 * Booting off a mirrored root volume:
919 * At this point, we have booted and mounted root on a
920 * single component of the mirror. Complete the boot
921 * by configuring SVM and converting the root to the
922 * dev_t of the mirrored root device. This dev_t conversion
923 * only works because the underlying device doesn't change.
926 if (svm_rootconf()) {
927 panic("vfs_mountroot: cannot remount root");
931 * mnttab should reflect the new root device
933 vfs_lock_wait(rootvfs
);
934 vfs_setresource(rootvfs
, rootfs
.bo_name
);
940 * Look up the root device via devfs so that a dv_node is
941 * created for it. The vnode is never VN_RELE()ed.
942 * We allocate more than MAXPATHLEN so that the
943 * buffer passed to i_ddi_prompath_to_devfspath() is
944 * exactly MAXPATHLEN (the function expects a buffer
947 plen
= strlen("/devices");
948 path
= kmem_alloc(plen
+ MAXPATHLEN
, KM_SLEEP
);
949 (void) strcpy(path
, "/devices");
951 if (i_ddi_prompath_to_devfspath(rootfs
.bo_name
, path
+ plen
)
953 lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &rvp
)) {
955 /* NUL terminate in case "path" has garbage */
956 path
[plen
+ MAXPATHLEN
- 1] = '\0';
958 cmn_err(CE_WARN
, "!Cannot lookup root device: %s", path
);
961 kmem_free(path
, plen
+ MAXPATHLEN
);
962 vfs_mnttabvp_setup();
966 * If remount failed and we're in a zone we need to check for the zone
967 * root path and strip it before the call to vfs_setpath().
969 * If strpath doesn't begin with the zone_rootpath the original
970 * strpath is returned unchanged.
973 stripzonepath(const char *strpath
)
977 zone_t
*zonep
= curproc
->p_zone
;
979 if (zonep
->zone_rootpath
== NULL
|| strpath
== NULL
) {
984 * we check for the end of the string at one past the
985 * current position because the zone_rootpath always
986 * ends with "/" but we don't want to strip that off.
988 str1
= zonep
->zone_rootpath
;
989 str2
= (char *)strpath
;
990 ASSERT(str1
[0] != '\0');
991 for (i
= 0; str1
[i
+ 1] != '\0'; i
++) {
992 if (str1
[i
] != str2
[i
])
993 return ((char *)strpath
);
999 * Check to see if our "block device" is actually a file. If so,
1000 * automatically add a lofi device, and keep track of this fact.
1003 lofi_add(const char *fsname
, struct vfs
*vfsp
,
1004 mntopts_t
*mntopts
, struct mounta
*uap
)
1006 int fromspace
= (uap
->flags
& MS_SYSSPACE
) ?
1007 UIO_SYSSPACE
: UIO_USERSPACE
;
1008 struct lofi_ioctl
*li
= NULL
;
1009 struct vnode
*vp
= NULL
;
1010 struct pathname pn
= { NULL
};
1012 ldi_handle_t ldi_hdl
;
1017 if (fsname
== NULL
||
1018 (vfssw
= vfs_getvfssw(fsname
)) == NULL
)
1021 if (!(vfssw
->vsw_flag
& VSW_CANLOFI
)) {
1022 vfs_unrefvfssw(vfssw
);
1026 vfs_unrefvfssw(vfssw
);
1029 if (pn_get(uap
->spec
, fromspace
, &pn
) != 0)
1032 if (lookupname(uap
->spec
, fromspace
, FOLLOW
, NULL
, &vp
) != 0)
1035 if (vp
->v_type
!= VREG
)
1038 /* OK, this is a lofi mount. */
1040 if ((uap
->flags
& (MS_REMOUNT
|MS_GLOBAL
)) ||
1041 vfs_optionisset_nolock(mntopts
, MNTOPT_SUID
, NULL
) ||
1042 vfs_optionisset_nolock(mntopts
, MNTOPT_SETUID
, NULL
) ||
1043 vfs_optionisset_nolock(mntopts
, MNTOPT_DEVICES
, NULL
)) {
1048 ldi_id
= ldi_ident_from_anon();
1049 li
= kmem_zalloc(sizeof (*li
), KM_SLEEP
);
1050 (void) strlcpy(li
->li_filename
, pn
.pn_path
, MAXPATHLEN
);
1053 * The lofi control node is currently exclusive-open. We'd like
1054 * to improve this, but in the meantime, we'll loop waiting for
1058 err
= ldi_open_by_name("/dev/lofictl", FREAD
| FWRITE
| FEXCL
,
1059 kcred
, &ldi_hdl
, ldi_id
);
1064 if ((err
= delay_sig(hz
/ 8)) == EINTR
)
1071 err
= ldi_ioctl(ldi_hdl
, LOFI_MAP_FILE
, (intptr_t)li
,
1072 FREAD
| FWRITE
| FEXCL
| FKIOCTL
, kcred
, &minor
);
1074 (void) ldi_close(ldi_hdl
, FREAD
| FWRITE
| FEXCL
, kcred
);
1077 vfsp
->vfs_lofi_minor
= minor
;
1080 ldi_ident_release(ldi_id
);
1083 kmem_free(li
, sizeof (*li
));
1091 lofi_remove(struct vfs
*vfsp
)
1093 struct lofi_ioctl
*li
= NULL
;
1095 ldi_handle_t ldi_hdl
;
1098 if (vfsp
->vfs_lofi_minor
== 0)
1101 ldi_id
= ldi_ident_from_anon();
1103 li
= kmem_zalloc(sizeof (*li
), KM_SLEEP
);
1104 li
->li_minor
= vfsp
->vfs_lofi_minor
;
1105 li
->li_cleanup
= B_TRUE
;
1108 err
= ldi_open_by_name("/dev/lofictl", FREAD
| FWRITE
| FEXCL
,
1109 kcred
, &ldi_hdl
, ldi_id
);
1110 } while (err
== EBUSY
);
1115 err
= ldi_ioctl(ldi_hdl
, LOFI_UNMAP_FILE_MINOR
, (intptr_t)li
,
1116 FREAD
| FWRITE
| FEXCL
| FKIOCTL
, kcred
, NULL
);
1118 (void) ldi_close(ldi_hdl
, FREAD
| FWRITE
| FEXCL
, kcred
);
1121 vfsp
->vfs_lofi_minor
= 0;
1124 ldi_ident_release(ldi_id
);
1126 kmem_free(li
, sizeof (*li
));
1130 * Common mount code. Called from the system call entry point, from autofs,
1131 * nfsv4 trigger mounts, and from pxfs.
1133 * Takes the effective file system type, mount arguments, the mount point
1134 * vnode, flags specifying whether the mount is a remount and whether it
1135 * should be entered into the vfs list, and credentials. Fills in its vfspp
1136 * parameter with the mounted file system instance's vfs.
1138 * Note that the effective file system type is specified as a string. It may
1139 * be null, in which case it's determined from the mount arguments, and may
1140 * differ from the type specified in the mount arguments; this is a hook to
1141 * allow interposition when instantiating file system instances.
1143 * The caller is responsible for releasing its own hold on the mount point
1144 * vp (this routine does its own hold when necessary).
1145 * Also note that for remounts, the mount point vp should be the vnode for
1146 * the root of the file system rather than the vnode that the file system
1147 * is mounted on top of.
1150 domount(char *fsname
, struct mounta
*uap
, vnode_t
*vp
, struct cred
*credp
,
1158 mntopts_t mnt_mntopts
;
1160 int copyout_error
= 0;
1162 char *opts
= uap
->optptr
;
1163 char *inargs
= opts
;
1164 int optlen
= uap
->optlen
;
1170 int splice
= ((uap
->flags
& MS_NOSPLICE
) == 0);
1171 int fromspace
= (uap
->flags
& MS_SYSSPACE
) ?
1172 UIO_SYSSPACE
: UIO_USERSPACE
;
1173 char *resource
= NULL
, *mountpt
= NULL
;
1174 refstr_t
*oldresource
, *oldmntpt
;
1175 struct pathname pn
, rpn
;
1176 vsk_anchor_t
*vskap
;
1177 char fstname
[FSTYPSZ
];
1180 * The v_flag value for the mount point vp is permanently set
1181 * to VVFSLOCK so that no one bypasses the vn_vfs*locks routine
1182 * for mount point locking.
1184 mutex_enter(&vp
->v_lock
);
1185 vp
->v_flag
|= VVFSLOCK
;
1186 mutex_exit(&vp
->v_lock
);
1188 mnt_mntopts
.mo_count
= 0;
1190 * Find the ops vector to use to invoke the file system-specific mount
1191 * method. If the fsname argument is non-NULL, use it directly.
1192 * Otherwise, dig the file system type information out of the mount
1195 * A side effect is to hold the vfssw entry.
1197 * Mount arguments can be specified in several ways, which are
1198 * distinguished by flag bit settings. The preferred way is to set
1199 * MS_OPTIONSTR, indicating an 8 argument mount with the file system
1200 * type supplied as a character string and the last two arguments
1201 * being a pointer to a character buffer and the size of the buffer.
1202 * On entry, the buffer holds a null terminated list of options; on
1203 * return, the string is the list of options the file system
1204 * recognized. If MS_DATA is set arguments five and six point to a
1205 * block of binary data which the file system interprets.
1206 * A further wrinkle is that some callers don't set MS_FSS and MS_DATA
1207 * consistently with these conventions. To handle them, we check to
1208 * see whether the pointer to the file system name has a numeric value
1209 * less than 256. If so, we treat it as an index.
1211 if (fsname
!= NULL
) {
1212 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
) {
1215 } else if (uap
->flags
& (MS_OPTIONSTR
| MS_DATA
| MS_FSS
)) {
1221 if ((fstype
= (uintptr_t)uap
->fstype
) < 256) {
1223 if (fstype
== 0 || fstype
>= nfstype
||
1224 !ALLOCATED_VFSSW(&vfssw
[fstype
])) {
1228 (void) strcpy(fsname
, vfssw
[fstype
].vsw_name
);
1230 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
)
1234 * Handle either kernel or user address space.
1236 if (uap
->flags
& MS_SYSSPACE
) {
1237 error
= copystr(uap
->fstype
, fsname
,
1240 error
= copyinstr(uap
->fstype
, fsname
,
1244 if (error
== ENAMETOOLONG
)
1248 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
)
1252 if ((vswp
= vfs_getvfsswbyvfsops(vfs_getops(rootvfs
))) == NULL
)
1255 if (!VFS_INSTALLED(vswp
))
1257 vfsops
= &vswp
->vsw_vfsops
;
1259 vfs_copyopttbl(&vswp
->vsw_optproto
, &mnt_mntopts
);
1261 * Fetch mount options and parse them for generic vfs options
1263 if (uap
->flags
& MS_OPTIONSTR
) {
1265 * Limit the buffer size
1267 if (optlen
< 0 || optlen
> MAX_MNTOPT_STR
) {
1271 if ((uap
->flags
& MS_SYSSPACE
) == 0) {
1272 inargs
= kmem_alloc(MAX_MNTOPT_STR
, KM_SLEEP
);
1275 error
= copyinstr(opts
, inargs
, (size_t)optlen
,
1282 vfs_parsemntopts(&mnt_mntopts
, inargs
, 0);
1285 * Flag bits override the options string.
1287 if (uap
->flags
& MS_REMOUNT
)
1288 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_REMOUNT
, NULL
, 0, 0);
1289 if (uap
->flags
& MS_RDONLY
)
1290 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_RO
, NULL
, 0, 0);
1291 if (uap
->flags
& MS_NOSUID
)
1292 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
, 0, 0);
1295 * Check if this is a remount; must be set in the option string and
1296 * the file system must support a remount option.
1298 if (remount
= vfs_optionisset_nolock(&mnt_mntopts
,
1299 MNTOPT_REMOUNT
, NULL
)) {
1300 if (!(vswp
->vsw_flag
& VSW_CANREMOUNT
)) {
1304 uap
->flags
|= MS_REMOUNT
;
1308 * uap->flags and vfs_optionisset() should agree.
1310 if (rdonly
= vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_RO
, NULL
)) {
1311 uap
->flags
|= MS_RDONLY
;
1313 if (vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
)) {
1314 uap
->flags
|= MS_NOSUID
;
1316 nbmand
= vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_NBMAND
, NULL
);
1317 ASSERT(splice
|| !remount
);
1319 * If we are splicing the fs into the namespace,
1320 * perform mount point checks.
1322 * We want to resolve the path for the mount point to eliminate
1323 * '.' and ".." and symlinks in mount points; we can't do the
1324 * same for the resource string, since it would turn
1325 * "/dev/dsk/c0t0d0s0" into "/devices/pci@...". We need to do
1326 * this before grabbing vn_vfswlock(), because otherwise we
1327 * would deadlock with lookuppn().
1330 ASSERT(vp
->v_count
> 0);
1333 * Pick up mount point and device from appropriate space.
1335 if (pn_get(uap
->spec
, fromspace
, &pn
) == 0) {
1336 resource
= kmem_alloc(pn
.pn_pathlen
+ 1,
1338 (void) strcpy(resource
, pn
.pn_path
);
1342 * Do a lookupname prior to taking the
1343 * writelock. Mark this as completed if
1344 * successful for later cleanup and addition to
1345 * the mount in progress table.
1347 if ((uap
->flags
& MS_GLOBAL
) == 0 &&
1348 lookupname(uap
->spec
, fromspace
,
1349 FOLLOW
, NULL
, &bvp
) == 0) {
1353 if ((error
= pn_get(uap
->dir
, fromspace
, &pn
)) == 0) {
1356 if (*pn
.pn_path
!= '/') {
1363 * Kludge to prevent autofs from deadlocking with
1364 * itself when it calls domount().
1366 * If autofs is calling, it is because it is doing
1367 * (autofs) mounts in the process of an NFS mount. A
1368 * lookuppn() here would cause us to block waiting for
1369 * said NFS mount to complete, which can't since this
1370 * is the thread that was supposed to doing it.
1372 if (fromspace
== UIO_USERSPACE
) {
1373 if ((error
= lookuppn(&pn
, &rpn
, FOLLOW
, NULL
,
1378 * The file disappeared or otherwise
1379 * became inaccessible since we opened
1380 * it; might as well fail the mount
1381 * since the mount point is no longer
1391 mountpt
= kmem_alloc(pnp
->pn_pathlen
+ 1, KM_SLEEP
);
1392 (void) strcpy(mountpt
, pnp
->pn_path
);
1395 * If the addition of the zone's rootpath
1396 * would push us over a total path length
1397 * of MAXPATHLEN, we fail the mount with
1398 * ENAMETOOLONG, which is what we would have
1399 * gotten if we were trying to perform the same
1400 * mount in the global zone.
1402 * strlen() doesn't count the trailing
1403 * '\0', but zone_rootpathlen counts both a
1404 * trailing '/' and the terminating '\0'.
1406 if ((curproc
->p_zone
->zone_rootpathlen
- 1 +
1407 strlen(mountpt
)) > MAXPATHLEN
||
1408 (resource
!= NULL
&&
1409 (curproc
->p_zone
->zone_rootpathlen
- 1 +
1410 strlen(resource
)) > MAXPATHLEN
)) {
1411 error
= ENAMETOOLONG
;
1422 * Prevent path name resolution from proceeding past
1425 if (vn_vfswlock(vp
) != 0) {
1431 * Verify that it's legitimate to establish a mount on
1432 * the prospective mount point.
1434 if (vn_mountedvfs(vp
) != NULL
) {
1436 * The mount point lock was obtained after some
1437 * other thread raced through and established a mount.
1443 if (vp
->v_flag
& VNOMOUNT
) {
1449 if ((uap
->flags
& (MS_DATA
| MS_OPTIONSTR
)) == 0) {
1450 uap
->dataptr
= NULL
;
1455 * If this is a remount, we don't want to create a new VFS.
1456 * Instead, we pass the existing one with a remount flag.
1460 * Confirm that the mount point is the root vnode of the
1461 * file system that is being remounted.
1462 * This can happen if the user specifies a different
1463 * mount point directory pathname in the (re)mount command.
1465 * Code below can only be reached if splice is true, so it's
1466 * safe to do vn_vfsunlock() here.
1468 if ((vp
->v_flag
& VROOT
) == 0) {
1474 * Disallow making file systems read-only unless file system
1475 * explicitly allows it in its vfssw. Ignore other flags.
1477 if (rdonly
&& vn_is_readonly(vp
) == 0 &&
1478 (vswp
->vsw_flag
& VSW_CANRWRO
) == 0) {
1484 * Disallow changing the NBMAND disposition of the file
1485 * system on remounts.
1487 if ((nbmand
&& ((vp
->v_vfsp
->vfs_flag
& VFS_NBMAND
) == 0)) ||
1488 (!nbmand
&& (vp
->v_vfsp
->vfs_flag
& VFS_NBMAND
))) {
1494 ovflags
= vfsp
->vfs_flag
;
1495 vfsp
->vfs_flag
|= VFS_REMOUNT
;
1496 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
1498 vfsp
= vfs_alloc(KM_SLEEP
);
1499 VFS_INIT(vfsp
, vfsops
, NULL
);
1504 if ((error
= lofi_add(fsname
, vfsp
, &mnt_mntopts
, uap
)) != 0) {
1517 * PRIV_SYS_MOUNT doesn't mean you can become root.
1519 if (vfsp
->vfs_lofi_minor
!= 0) {
1520 uap
->flags
|= MS_NOSUID
;
1521 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
, 0, 0);
1525 * The vfs_reflock is not used anymore the code below explicitly
1526 * holds it preventing others accesing it directly.
1528 if ((sema_tryp(&vfsp
->vfs_reflock
) == 0) &&
1529 !(vfsp
->vfs_flag
& VFS_REMOUNT
))
1531 "mount type %s couldn't get vfs_reflock", vswp
->vsw_name
);
1534 * Lock the vfs. If this is a remount we want to avoid spurious umount
1535 * failures that happen as a side-effect of fsflush() and other mount
1536 * and unmount operations that might be going on simultaneously and
1537 * may have locked the vfs currently. To not return EBUSY immediately
1538 * here we use vfs_lock_wait() instead vfs_lock() for the remount case.
1541 if (error
= vfs_lock(vfsp
)) {
1542 vfsp
->vfs_flag
= ovflags
;
1552 vfs_lock_wait(vfsp
);
1556 * Add device to mount in progress table, global mounts require special
1557 * handling. It is possible that we have already done the lookupname
1558 * on a spliced, non-global fs. If so, we don't want to do it again
1559 * since we cannot do a lookupname after taking the
1560 * wlock above. This case is for a non-spliced, non-global filesystem.
1563 if ((uap
->flags
& MS_GLOBAL
) == 0 &&
1564 lookupname(uap
->spec
, fromspace
, FOLLOW
, NULL
, &bvp
) == 0) {
1570 vnode_t
*lvp
= NULL
;
1572 error
= vfs_get_lofi(vfsp
, &lvp
);
1587 } else if (error
== -1) {
1596 vfs_addmip(bdev
, vfsp
);
1601 * Invalidate cached entry for the mount point.
1607 * If have an option string but the filesystem doesn't supply a
1608 * prototype options table, create a table with the global
1609 * options and sufficient room to accept all the options in the
1610 * string. Then parse the passed in option string
1611 * accepting all the options in the string. This gives us an
1612 * option table with all the proper cancel properties for the
1615 * Filesystems that supply a prototype options table are handled
1616 * earlier in this function.
1618 if (uap
->flags
& MS_OPTIONSTR
) {
1619 if (!(vswp
->vsw_flag
& VSW_HASPROTO
)) {
1620 mntopts_t tmp_mntopts
;
1622 tmp_mntopts
.mo_count
= 0;
1623 vfs_createopttbl_extend(&tmp_mntopts
, inargs
,
1625 vfs_parsemntopts(&tmp_mntopts
, inargs
, 1);
1626 vfs_swapopttbl_nolock(&mnt_mntopts
, &tmp_mntopts
);
1627 vfs_freeopttbl(&tmp_mntopts
);
1632 * Serialize with zone creations.
1634 mount_in_progress();
1636 * Instantiate (or reinstantiate) the file system. If appropriate,
1637 * splice it into the file system name space.
1639 * We want VFS_MOUNT() to be able to override the vfs_resource
1640 * string if necessary (ie, mntfs), and also for a remount to
1641 * change the same (necessary when remounting '/' during boot).
1642 * So we set up vfs_mntpt and vfs_resource to what we think they
1643 * should be, then hand off control to VFS_MOUNT() which can
1646 * For safety's sake, when changing vfs_resource or vfs_mntpt of
1647 * a vfs which is on the vfs list (i.e. during a remount), we must
1648 * never set those fields to NULL. Several bits of code make
1649 * assumptions that the fields are always valid.
1651 vfs_swapopttbl(&mnt_mntopts
, &vfsp
->vfs_mntopts
);
1653 if ((oldresource
= vfsp
->vfs_resource
) != NULL
)
1654 refstr_hold(oldresource
);
1655 if ((oldmntpt
= vfsp
->vfs_mntpt
) != NULL
)
1656 refstr_hold(oldmntpt
);
1658 vfs_setresource(vfsp
, resource
);
1659 vfs_setmntpoint(vfsp
, mountpt
);
1662 * going to mount on this vnode, so notify.
1664 vnevent_mountedover(vp
, NULL
);
1665 error
= VFS_MOUNT(vfsp
, vp
, uap
, credp
);
1667 if (uap
->flags
& MS_RDONLY
)
1668 vfs_setmntopt(vfsp
, MNTOPT_RO
, NULL
, 0);
1669 if (uap
->flags
& MS_NOSUID
)
1670 vfs_setmntopt(vfsp
, MNTOPT_NOSUID
, NULL
, 0);
1671 if (uap
->flags
& MS_GLOBAL
)
1672 vfs_setmntopt(vfsp
, MNTOPT_GLOBAL
, NULL
, 0);
1678 /* put back pre-remount options */
1679 vfs_swapopttbl(&mnt_mntopts
, &vfsp
->vfs_mntopts
);
1680 vfs_setmntpoint(vfsp
, (stripzonepath(
1681 refstr_value(oldmntpt
))));
1683 refstr_rele(oldmntpt
);
1684 vfs_setresource(vfsp
, (stripzonepath(
1685 refstr_value(oldresource
))));
1687 refstr_rele(oldresource
);
1688 vfsp
->vfs_flag
= ovflags
;
1693 vfs_freemnttab(vfsp
);
1698 * Set the mount time to now
1700 vfsp
->vfs_mtime
= ddi_get_time();
1702 vfsp
->vfs_flag
&= ~VFS_REMOUNT
;
1704 refstr_rele(oldresource
);
1706 refstr_rele(oldmntpt
);
1707 } else if (splice
) {
1709 * Link vfsp into the name space at the mount
1710 * point. Vfs_add() is responsible for
1711 * holding the mount point which will be
1712 * released when vfs_remove() is called.
1714 vfs_add(vp
, vfsp
, uap
->flags
);
1717 * Hold the reference to file system which is
1718 * not linked into the name space.
1720 vfsp
->vfs_zone
= NULL
;
1722 vfsp
->vfs_vnodecovered
= NULL
;
1725 * Set flags for global options encountered
1727 if (vfs_optionisset(vfsp
, MNTOPT_RO
, NULL
))
1728 vfsp
->vfs_flag
|= VFS_RDONLY
;
1730 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
1731 if (vfs_optionisset(vfsp
, MNTOPT_NOSUID
, NULL
)) {
1732 vfsp
->vfs_flag
|= (VFS_NOSETUID
|VFS_NODEVICES
);
1734 if (vfs_optionisset(vfsp
, MNTOPT_NODEVICES
, NULL
))
1735 vfsp
->vfs_flag
|= VFS_NODEVICES
;
1737 vfsp
->vfs_flag
&= ~VFS_NODEVICES
;
1738 if (vfs_optionisset(vfsp
, MNTOPT_NOSETUID
, NULL
))
1739 vfsp
->vfs_flag
|= VFS_NOSETUID
;
1741 vfsp
->vfs_flag
&= ~VFS_NOSETUID
;
1743 if (vfs_optionisset(vfsp
, MNTOPT_NBMAND
, NULL
))
1744 vfsp
->vfs_flag
|= VFS_NBMAND
;
1746 vfsp
->vfs_flag
&= ~VFS_NBMAND
;
1748 if (vfs_optionisset(vfsp
, MNTOPT_XATTR
, NULL
))
1749 vfsp
->vfs_flag
|= VFS_XATTR
;
1751 vfsp
->vfs_flag
&= ~VFS_XATTR
;
1753 if (vfs_optionisset(vfsp
, MNTOPT_NOEXEC
, NULL
))
1754 vfsp
->vfs_flag
|= VFS_NOEXEC
;
1756 vfsp
->vfs_flag
&= ~VFS_NOEXEC
;
1759 * Now construct the output option string of options
1762 if (uap
->flags
& MS_OPTIONSTR
) {
1763 vfs_list_read_lock();
1764 copyout_error
= vfs_buildoptionstr(
1765 &vfsp
->vfs_mntopts
, inargs
, optlen
);
1767 if (copyout_error
== 0 &&
1768 (uap
->flags
& MS_SYSSPACE
) == 0) {
1769 copyout_error
= copyoutstr(inargs
, opts
,
1775 * If this isn't a remount, set up the vopstats before
1776 * anyone can touch this. We only allow spliced file
1777 * systems (file systems which are in the namespace) to
1778 * have the VFS_STATS flag set.
1779 * NOTE: PxFS mounts the underlying file system with
1780 * MS_NOSPLICE set and copies those vfs_flags to its private
1781 * vfs structure. As a result, PxFS should never have
1782 * the VFS_STATS flag or else we might access the vfs
1783 * statistics-related fields prior to them being
1784 * properly initialized.
1786 if (!remount
&& (vswp
->vsw_flag
& VSW_STATS
) && splice
) {
1787 initialize_vopstats(&vfsp
->vfs_vopstats
);
1789 * We need to set vfs_vskap to NULL because there's
1790 * a chance it won't be set below. This is checked
1791 * in teardown_vopstats() so we can't have garbage.
1793 vfsp
->vfs_vskap
= NULL
;
1794 vfsp
->vfs_flag
|= VFS_STATS
;
1795 vfsp
->vfs_fstypevsp
= get_fstype_vopstats(vfsp
, vswp
);
1798 if (vswp
->vsw_flag
& VSW_XID
)
1799 vfsp
->vfs_flag
|= VFS_XID
;
1807 if ((error
== 0) && (copyout_error
== 0)) {
1810 * Don't call get_vskstat_anchor() while holding
1811 * locks since it allocates memory and calls
1812 * VFS_STATVFS(). For NFS, the latter can generate
1813 * an over-the-wire call.
1815 vskap
= get_vskstat_anchor(vfsp
);
1816 /* Only take the lock if we have something to do */
1817 if (vskap
!= NULL
) {
1818 vfs_lock_wait(vfsp
);
1819 if (vfsp
->vfs_flag
& VFS_STATS
) {
1820 vfsp
->vfs_vskap
= vskap
;
1825 /* Return vfsp to caller. */
1829 vfs_freeopttbl(&mnt_mntopts
);
1830 if (resource
!= NULL
)
1831 kmem_free(resource
, strlen(resource
) + 1);
1832 if (mountpt
!= NULL
)
1833 kmem_free(mountpt
, strlen(mountpt
) + 1);
1835 * It is possible we errored prior to adding to mount in progress
1836 * table. Must free vnode we acquired with successful lookupname.
1842 ASSERT(vswp
!= NULL
);
1843 vfs_unrefvfssw(vswp
);
1845 kmem_free(inargs
, MAX_MNTOPT_STR
);
1846 if (copyout_error
) {
1849 error
= copyout_error
;
1855 vfs_setpath(struct vfs
*vfsp
, refstr_t
**refp
, const char *newpath
)
1859 zone_t
*zone
= curproc
->p_zone
;
1861 int have_list_lock
= 0;
1863 ASSERT(!VFS_ON_LIST(vfsp
) || vfs_lock_held(vfsp
));
1866 * New path must be less than MAXPATHLEN because mntfs
1867 * will only display up to MAXPATHLEN bytes. This is currently
1868 * safe, because domount() uses pn_get(), and other callers
1869 * similarly cap the size to fewer than MAXPATHLEN bytes.
1872 ASSERT(strlen(newpath
) < MAXPATHLEN
);
1874 /* mntfs requires consistency while vfs list lock is held */
1876 if (VFS_ON_LIST(vfsp
)) {
1884 /* Do we need to modify the path? */
1886 if (zone
== global_zone
|| *newpath
!= '/') {
1887 ref
= refstr_alloc(newpath
);
1892 * Truncate the trailing '/' in the zoneroot, and merge
1893 * in the zone's rootpath with the "newpath" (resource
1894 * or mountpoint) passed in.
1896 * The size of the required buffer is thus the size of
1897 * the buffer required for the passed-in newpath
1898 * (strlen(newpath) + 1), plus the size of the buffer
1899 * required to hold zone_rootpath (zone_rootpathlen)
1900 * minus one for one of the now-superfluous NUL
1901 * terminations, minus one for the trailing '/'.
1905 * (strlen(newpath) + 1) + zone_rootpathlen - 1 - 1
1907 * Which is what we have below.
1910 len
= strlen(newpath
) + zone
->zone_rootpathlen
- 1;
1911 sp
= kmem_alloc(len
, KM_SLEEP
);
1914 * Copy everything including the trailing slash, which
1915 * we then overwrite with the NUL character.
1918 (void) strcpy(sp
, zone
->zone_rootpath
);
1919 sp
[zone
->zone_rootpathlen
- 2] = '\0';
1920 (void) strcat(sp
, newpath
);
1922 ref
= refstr_alloc(sp
);
1927 if (have_list_lock
) {
1928 vfs_mnttab_modtimeupd();
1934 * Record a mounted resource name in a vfs structure.
1935 * If vfsp is already mounted, caller must hold the vfs lock.
1938 vfs_setresource(struct vfs
*vfsp
, const char *resource
)
1940 if (resource
== NULL
|| resource
[0] == '\0')
1941 resource
= VFS_NORESOURCE
;
1942 vfs_setpath(vfsp
, &vfsp
->vfs_resource
, resource
);
1946 * Record a mount point name in a vfs structure.
1947 * If vfsp is already mounted, caller must hold the vfs lock.
1950 vfs_setmntpoint(struct vfs
*vfsp
, const char *mntpt
)
1952 if (mntpt
== NULL
|| mntpt
[0] == '\0')
1953 mntpt
= VFS_NOMNTPT
;
1954 vfs_setpath(vfsp
, &vfsp
->vfs_mntpt
, mntpt
);
1957 /* Returns the vfs_resource. Caller must call refstr_rele() when finished. */
1960 vfs_getresource(const struct vfs
*vfsp
)
1964 vfs_list_read_lock();
1965 resource
= vfsp
->vfs_resource
;
1966 refstr_hold(resource
);
1972 /* Returns the vfs_mntpt. Caller must call refstr_rele() when finished. */
1975 vfs_getmntpoint(const struct vfs
*vfsp
)
1979 vfs_list_read_lock();
1980 mntpt
= vfsp
->vfs_mntpt
;
1988 * Create an empty options table with enough empty slots to hold all
1989 * The options in the options string passed as an argument.
1990 * Potentially prepend another options table.
1992 * Note: caller is responsible for locking the vfs list, if needed,
1996 vfs_createopttbl_extend(mntopts_t
*mops
, const char *opts
,
1997 const mntopts_t
*mtmpl
)
1999 const char *s
= opts
;
2002 if (opts
== NULL
|| *opts
== '\0') {
2008 * Count number of options in the string
2010 for (s
= strchr(s
, ','); s
!= NULL
; s
= strchr(s
, ',')) {
2015 vfs_copyopttbl_extend(mtmpl
, mops
, count
);
2019 * Create an empty options table with enough empty slots to hold all
2020 * The options in the options string passed as an argument.
2022 * This function is *not* for general use by filesystems.
2024 * Note: caller is responsible for locking the vfs list, if needed,
2028 vfs_createopttbl(mntopts_t
*mops
, const char *opts
)
2030 vfs_createopttbl_extend(mops
, opts
, NULL
);
2035 * Swap two mount options tables
2038 vfs_swapopttbl_nolock(mntopts_t
*optbl1
, mntopts_t
*optbl2
)
2043 tmpcnt
= optbl2
->mo_count
;
2044 tmplist
= optbl2
->mo_list
;
2045 optbl2
->mo_count
= optbl1
->mo_count
;
2046 optbl2
->mo_list
= optbl1
->mo_list
;
2047 optbl1
->mo_count
= tmpcnt
;
2048 optbl1
->mo_list
= tmplist
;
2052 vfs_swapopttbl(mntopts_t
*optbl1
, mntopts_t
*optbl2
)
2055 vfs_swapopttbl_nolock(optbl1
, optbl2
);
2056 vfs_mnttab_modtimeupd();
2061 vfs_copycancelopt_extend(char **const moc
, int extend
)
2068 for (; moc
[i
] != NULL
; i
++)
2069 /* count number of options to cancel */;
2072 if (i
+ extend
== 0)
2075 result
= kmem_alloc((i
+ extend
+ 1) * sizeof (char *), KM_SLEEP
);
2077 for (j
= 0; j
< i
; j
++) {
2078 result
[j
] = kmem_alloc(strlen(moc
[j
]) + 1, KM_SLEEP
);
2079 (void) strcpy(result
[j
], moc
[j
]);
2081 for (; j
<= i
+ extend
; j
++)
2088 vfs_copyopt(const mntopt_t
*s
, mntopt_t
*d
)
2092 d
->mo_flags
= s
->mo_flags
;
2093 d
->mo_data
= s
->mo_data
;
2096 dp
= kmem_alloc(strlen(sp
) + 1, KM_SLEEP
);
2097 (void) strcpy(dp
, sp
);
2100 d
->mo_name
= NULL
; /* should never happen */
2103 d
->mo_cancel
= vfs_copycancelopt_extend(s
->mo_cancel
, 0);
2107 dp
= kmem_alloc(strlen(sp
) + 1, KM_SLEEP
);
2108 (void) strcpy(dp
, sp
);
2116 * Copy a mount options table, possibly allocating some spare
2117 * slots at the end. It is permissible to copy_extend the NULL table.
2120 vfs_copyopttbl_extend(const mntopts_t
*smo
, mntopts_t
*dmo
, int extra
)
2126 * Clear out any existing stuff in the options table being initialized
2128 vfs_freeopttbl(dmo
);
2129 count
= (smo
== NULL
) ? 0 : smo
->mo_count
;
2130 if ((count
+ extra
) == 0) /* nothing to do */
2132 dmo
->mo_count
= count
+ extra
;
2133 motbl
= kmem_zalloc((count
+ extra
) * sizeof (mntopt_t
), KM_SLEEP
);
2134 dmo
->mo_list
= motbl
;
2135 for (i
= 0; i
< count
; i
++) {
2136 vfs_copyopt(&smo
->mo_list
[i
], &motbl
[i
]);
2138 for (i
= count
; i
< count
+ extra
; i
++) {
2139 motbl
[i
].mo_flags
= MO_EMPTY
;
2144 * Copy a mount options table.
2146 * This function is *not* for general use by filesystems.
2148 * Note: caller is responsible for locking the vfs list, if needed,
2149 * to protect smo and dmo.
2152 vfs_copyopttbl(const mntopts_t
*smo
, mntopts_t
*dmo
)
2154 vfs_copyopttbl_extend(smo
, dmo
, 0);
2158 vfs_mergecancelopts(const mntopt_t
*mop1
, const mntopt_t
*mop2
)
2163 char **sp1
, **sp2
, **dp
;
2166 * First we count both lists of cancel options.
2167 * If either is NULL or has no elements, we return a copy of
2170 if (mop1
->mo_cancel
!= NULL
) {
2171 for (; mop1
->mo_cancel
[c1
] != NULL
; c1
++)
2172 /* count cancel options in mop1 */;
2176 return (vfs_copycancelopt_extend(mop2
->mo_cancel
, 0));
2178 if (mop2
->mo_cancel
!= NULL
) {
2179 for (; mop2
->mo_cancel
[c2
] != NULL
; c2
++)
2180 /* count cancel options in mop2 */;
2183 result
= vfs_copycancelopt_extend(mop1
->mo_cancel
, c2
);
2189 * When we get here, we've got two sets of cancel options;
2190 * we need to merge the two sets. We know that the result
2191 * array has "c1+c2+1" entries and in the end we might shrink
2193 * Result now has a copy of the c1 entries from mop1; we'll
2194 * now lookup all the entries of mop2 in mop1 and copy it if
2196 * This operation is O(n^2) but it's only called once per
2197 * filesystem per duplicate option. This is a situation
2198 * which doesn't arise with the filesystems in ON and
2203 for (sp2
= mop2
->mo_cancel
; *sp2
!= NULL
; sp2
++) {
2204 for (sp1
= mop1
->mo_cancel
; *sp1
!= NULL
; sp1
++) {
2205 if (strcmp(*sp1
, *sp2
) == 0)
2210 * Option *sp2 not found in mop1, so copy it.
2211 * The calls to vfs_copycancelopt_extend()
2212 * guarantee that there's enough room.
2214 *dp
= kmem_alloc(strlen(*sp2
) + 1, KM_SLEEP
);
2215 (void) strcpy(*dp
++, *sp2
);
2218 if (dp
!= &result
[c1
+c2
]) {
2219 size_t bytes
= (dp
- result
+ 1) * sizeof (char *);
2220 char **nres
= kmem_alloc(bytes
, KM_SLEEP
);
2222 bcopy(result
, nres
, bytes
);
2223 kmem_free(result
, (c1
+ c2
+ 1) * sizeof (char *));
2230 * Merge two mount option tables (outer and inner) into one. This is very
2231 * similar to "merging" global variables and automatic variables in C.
2233 * This isn't (and doesn't have to be) fast.
2235 * This function is *not* for general use by filesystems.
2237 * Note: caller is responsible for locking the vfs list, if needed,
2238 * to protect omo, imo & dmo.
2241 vfs_mergeopttbl(const mntopts_t
*omo
, const mntopts_t
*imo
, mntopts_t
*dmo
)
2244 mntopt_t
*mop
, *motbl
;
2248 * First determine how much space we need to allocate.
2250 count
= omo
->mo_count
;
2251 for (i
= 0; i
< imo
->mo_count
; i
++) {
2252 if (imo
->mo_list
[i
].mo_flags
& MO_EMPTY
)
2254 if (vfs_hasopt(omo
, imo
->mo_list
[i
].mo_name
) == NULL
)
2257 ASSERT(count
>= omo
->mo_count
&&
2258 count
<= omo
->mo_count
+ imo
->mo_count
);
2259 motbl
= kmem_alloc(count
* sizeof (mntopt_t
), KM_SLEEP
);
2260 for (i
= 0; i
< omo
->mo_count
; i
++)
2261 vfs_copyopt(&omo
->mo_list
[i
], &motbl
[i
]);
2262 freeidx
= omo
->mo_count
;
2263 for (i
= 0; i
< imo
->mo_count
; i
++) {
2264 if (imo
->mo_list
[i
].mo_flags
& MO_EMPTY
)
2266 if ((mop
= vfs_hasopt(omo
, imo
->mo_list
[i
].mo_name
)) != NULL
) {
2268 uint_t index
= mop
- omo
->mo_list
;
2270 newcanp
= vfs_mergecancelopts(mop
, &motbl
[index
]);
2272 vfs_freeopt(&motbl
[index
]);
2273 vfs_copyopt(&imo
->mo_list
[i
], &motbl
[index
]);
2275 vfs_freecancelopt(motbl
[index
].mo_cancel
);
2276 motbl
[index
].mo_cancel
= newcanp
;
2279 * If it's a new option, just copy it over to the first
2282 vfs_copyopt(&imo
->mo_list
[i
], &motbl
[freeidx
++]);
2285 dmo
->mo_count
= count
;
2286 dmo
->mo_list
= motbl
;
2290 * Functions to set and clear mount options in a mount options table.
2294 * Clear a mount option, if it exists.
2296 * The update_mnttab arg indicates whether mops is part of a vfs that is on
2300 vfs_clearmntopt_nolock(mntopts_t
*mops
, const char *opt
, int update_mnttab
)
2305 ASSERT(!update_mnttab
|| RW_WRITE_HELD(&vfslist
));
2307 count
= mops
->mo_count
;
2308 for (i
= 0; i
< count
; i
++) {
2309 mop
= &mops
->mo_list
[i
];
2311 if (mop
->mo_flags
& MO_EMPTY
)
2313 if (strcmp(opt
, mop
->mo_name
))
2315 mop
->mo_flags
&= ~MO_SET
;
2316 if (mop
->mo_arg
!= NULL
) {
2317 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2321 vfs_mnttab_modtimeupd();
2327 vfs_clearmntopt(struct vfs
*vfsp
, const char *opt
)
2331 if (VFS_ON_LIST(vfsp
)) {
2335 vfs_clearmntopt_nolock(&vfsp
->vfs_mntopts
, opt
, gotlock
);
2342 * Set a mount option on. If it's not found in the table, it's silently
2343 * ignored. If the option has MO_IGNORE set, it is still set unless the
2344 * VFS_NOFORCEOPT bit is set in the flags. Also, VFS_DISPLAY/VFS_NODISPLAY flag
2345 * bits can be used to toggle the MO_NODISPLAY bit for the option.
2346 * If the VFS_CREATEOPT flag bit is set then the first option slot with
2347 * MO_EMPTY set is created as the option passed in.
2349 * The update_mnttab arg indicates whether mops is part of a vfs that is on
2353 vfs_setmntopt_nolock(mntopts_t
*mops
, const char *opt
,
2354 const char *arg
, int flags
, int update_mnttab
)
2360 ASSERT(!update_mnttab
|| RW_WRITE_HELD(&vfslist
));
2362 if (flags
& VFS_CREATEOPT
) {
2363 if (vfs_hasopt(mops
, opt
) != NULL
) {
2364 flags
&= ~VFS_CREATEOPT
;
2367 count
= mops
->mo_count
;
2368 for (i
= 0; i
< count
; i
++) {
2369 mop
= &mops
->mo_list
[i
];
2371 if (mop
->mo_flags
& MO_EMPTY
) {
2372 if ((flags
& VFS_CREATEOPT
) == 0)
2374 sp
= kmem_alloc(strlen(opt
) + 1, KM_SLEEP
);
2375 (void) strcpy(sp
, opt
);
2378 mop
->mo_flags
= MO_HASVALUE
;
2381 } else if (strcmp(opt
, mop
->mo_name
)) {
2384 if ((mop
->mo_flags
& MO_IGNORE
) && (flags
& VFS_NOFORCEOPT
))
2386 if (arg
!= NULL
&& (mop
->mo_flags
& MO_HASVALUE
) != 0) {
2387 sp
= kmem_alloc(strlen(arg
) + 1, KM_SLEEP
);
2388 (void) strcpy(sp
, arg
);
2392 if (mop
->mo_arg
!= NULL
)
2393 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2395 if (flags
& VFS_DISPLAY
)
2396 mop
->mo_flags
&= ~MO_NODISPLAY
;
2397 if (flags
& VFS_NODISPLAY
)
2398 mop
->mo_flags
|= MO_NODISPLAY
;
2399 mop
->mo_flags
|= MO_SET
;
2400 if (mop
->mo_cancel
!= NULL
) {
2403 for (cp
= mop
->mo_cancel
; *cp
!= NULL
; cp
++)
2404 vfs_clearmntopt_nolock(mops
, *cp
, 0);
2407 vfs_mnttab_modtimeupd();
2413 vfs_setmntopt(struct vfs
*vfsp
, const char *opt
, const char *arg
, int flags
)
2417 if (VFS_ON_LIST(vfsp
)) {
2421 vfs_setmntopt_nolock(&vfsp
->vfs_mntopts
, opt
, arg
, flags
, gotlock
);
2428 * Add a "tag" option to a mounted file system's options list.
2430 * Note: caller is responsible for locking the vfs list, if needed,
2434 vfs_addtag(mntopts_t
*mops
, const char *tag
)
2437 mntopt_t
*mop
, *motbl
;
2439 count
= mops
->mo_count
+ 1;
2440 motbl
= kmem_zalloc(count
* sizeof (mntopt_t
), KM_SLEEP
);
2441 if (mops
->mo_count
) {
2442 size_t len
= (count
- 1) * sizeof (mntopt_t
);
2444 bcopy(mops
->mo_list
, motbl
, len
);
2445 kmem_free(mops
->mo_list
, len
);
2447 mops
->mo_count
= count
;
2448 mops
->mo_list
= motbl
;
2449 mop
= &motbl
[count
- 1];
2450 mop
->mo_flags
= MO_TAG
;
2451 mop
->mo_name
= kmem_alloc(strlen(tag
) + 1, KM_SLEEP
);
2452 (void) strcpy(mop
->mo_name
, tag
);
2457 * Allow users to set arbitrary "tags" in a vfs's mount options.
2458 * Broader use within the kernel is discouraged.
2461 vfs_settag(uint_t major
, uint_t minor
, const char *mntpt
, const char *tag
,
2468 dev_t dev
= makedevice(major
, minor
);
2470 char *buf
= kmem_alloc(MAX_MNTOPT_STR
, KM_SLEEP
);
2473 * Find the desired mounted file system
2478 if (vfsp
->vfs_dev
== dev
&&
2479 strcmp(mntpt
, refstr_value(vfsp
->vfs_mntpt
)) == 0) {
2483 vfsp
= vfsp
->vfs_next
;
2484 } while (vfsp
!= rootvfs
);
2490 err
= secpolicy_fs_config(cr
, vfsp
);
2494 mops
= &vfsp
->vfs_mntopts
;
2496 * Add tag if it doesn't already exist
2498 if ((mop
= vfs_hasopt(mops
, tag
)) == NULL
) {
2501 (void) vfs_buildoptionstr(mops
, buf
, MAX_MNTOPT_STR
);
2503 if (len
+ strlen(tag
) + 2 > MAX_MNTOPT_STR
) {
2507 mop
= vfs_addtag(mops
, tag
);
2509 if ((mop
->mo_flags
& MO_TAG
) == 0) {
2513 vfs_setmntopt_nolock(mops
, tag
, NULL
, 0, 1);
2516 kmem_free(buf
, MAX_MNTOPT_STR
);
2521 * Allow users to remove arbitrary "tags" in a vfs's mount options.
2522 * Broader use within the kernel is discouraged.
2525 vfs_clrtag(uint_t major
, uint_t minor
, const char *mntpt
, const char *tag
,
2531 dev_t dev
= makedevice(major
, minor
);
2535 * Find the desired mounted file system
2540 if (vfsp
->vfs_dev
== dev
&&
2541 strcmp(mntpt
, refstr_value(vfsp
->vfs_mntpt
)) == 0) {
2545 vfsp
= vfsp
->vfs_next
;
2546 } while (vfsp
!= rootvfs
);
2552 err
= secpolicy_fs_config(cr
, vfsp
);
2556 if ((mop
= vfs_hasopt(&vfsp
->vfs_mntopts
, tag
)) == NULL
) {
2560 if ((mop
->mo_flags
& MO_TAG
) == 0) {
2564 vfs_clearmntopt_nolock(&vfsp
->vfs_mntopts
, tag
, 1);
2571 * Function to parse an option string and fill in a mount options table.
2572 * Unknown options are silently ignored. The input option string is modified
2573 * by replacing separators with nulls. If the create flag is set, options
2574 * not found in the table are just added on the fly. The table must have
2575 * an option slot marked MO_EMPTY to add an option on the fly.
2577 * This function is *not* for general use by filesystems.
2579 * Note: caller is responsible for locking the vfs list, if needed,
2583 vfs_parsemntopts(mntopts_t
*mops
, char *osp
, int create
)
2585 char *s
= osp
, *p
, *nextop
, *valp
, *cp
, *ep
;
2586 int setflg
= VFS_NOFORCEOPT
;
2590 while (*s
!= '\0') {
2591 p
= strchr(s
, ','); /* find next option */
2596 cp
= p
; /* save location of comma */
2597 *p
++ = '\0'; /* mark end and point to next option */
2600 p
= strchr(s
, '='); /* look for value */
2602 valp
= NULL
; /* no value supplied */
2604 ep
= p
; /* save location of equals */
2605 *p
++ = '\0'; /* end option and point to value */
2609 * set option into options table
2612 setflg
|= VFS_CREATEOPT
;
2613 vfs_setmntopt_nolock(mops
, s
, valp
, setflg
, 0);
2615 *cp
= ','; /* restore the comma */
2617 *ep
= '='; /* restore the equals */
2623 * Function to inquire if an option exists in a mount options table.
2624 * Returns a pointer to the option if it exists, else NULL.
2626 * This function is *not* for general use by filesystems.
2628 * Note: caller is responsible for locking the vfs list, if needed,
2632 vfs_hasopt(const mntopts_t
*mops
, const char *opt
)
2637 count
= mops
->mo_count
;
2638 for (i
= 0; i
< count
; i
++) {
2639 mop
= &mops
->mo_list
[i
];
2641 if (mop
->mo_flags
& MO_EMPTY
)
2643 if (strcmp(opt
, mop
->mo_name
) == 0)
2650 * Function to inquire if an option is set in a mount options table.
2651 * Returns non-zero if set and fills in the arg pointer with a pointer to
2652 * the argument string or NULL if there is no argument string.
2655 vfs_optionisset_nolock(const mntopts_t
*mops
, const char *opt
, char **argp
)
2660 count
= mops
->mo_count
;
2661 for (i
= 0; i
< count
; i
++) {
2662 mop
= &mops
->mo_list
[i
];
2664 if (mop
->mo_flags
& MO_EMPTY
)
2666 if (strcmp(opt
, mop
->mo_name
))
2668 if ((mop
->mo_flags
& MO_SET
) == 0)
2670 if (argp
!= NULL
&& (mop
->mo_flags
& MO_HASVALUE
) != 0)
2671 *argp
= mop
->mo_arg
;
2679 vfs_optionisset(const struct vfs
*vfsp
, const char *opt
, char **argp
)
2683 vfs_list_read_lock();
2684 ret
= vfs_optionisset_nolock(&vfsp
->vfs_mntopts
, opt
, argp
);
2691 * Construct a comma separated string of the options set in the given
2692 * mount table, return the string in the given buffer. Return non-zero if
2693 * the buffer would overflow.
2695 * This function is *not* for general use by filesystems.
2697 * Note: caller is responsible for locking the vfs list, if needed,
2701 vfs_buildoptionstr(const mntopts_t
*mp
, char *buf
, int len
)
2708 for (i
= 0; i
< mp
->mo_count
; i
++) {
2711 mop
= &mp
->mo_list
[i
];
2712 if (mop
->mo_flags
& MO_SET
) {
2713 int optlen
, comma
= 0;
2717 optlen
= strlen(mop
->mo_name
);
2718 if (strlen(buf
) + comma
+ optlen
+ 1 > len
)
2722 (void) strcpy(cp
, mop
->mo_name
);
2725 * Append option value if there is one
2727 if (mop
->mo_arg
!= NULL
) {
2730 arglen
= strlen(mop
->mo_arg
);
2731 if (strlen(buf
) + arglen
+ 2 > len
)
2734 (void) strcpy(cp
, mop
->mo_arg
);
2745 vfs_freecancelopt(char **moc
)
2751 for (cp
= moc
; *cp
!= NULL
; cp
++) {
2752 kmem_free(*cp
, strlen(*cp
) + 1);
2755 kmem_free(moc
, (ccnt
+ 1) * sizeof (char *));
2760 vfs_freeopt(mntopt_t
*mop
)
2762 if (mop
->mo_name
!= NULL
)
2763 kmem_free(mop
->mo_name
, strlen(mop
->mo_name
) + 1);
2765 vfs_freecancelopt(mop
->mo_cancel
);
2767 if (mop
->mo_arg
!= NULL
)
2768 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2772 * Free a mount options table
2774 * This function is *not* for general use by filesystems.
2776 * Note: caller is responsible for locking the vfs list, if needed,
2780 vfs_freeopttbl(mntopts_t
*mp
)
2784 count
= mp
->mo_count
;
2785 for (i
= 0; i
< count
; i
++) {
2786 vfs_freeopt(&mp
->mo_list
[i
]);
2789 kmem_free(mp
->mo_list
, sizeof (mntopt_t
) * count
);
2798 vfs_mntdummyread(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cred
,
2799 caller_context_t
*ct
)
2806 vfs_mntdummywrite(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cred
,
2807 caller_context_t
*ct
)
2813 * The dummy vnode is currently used only by file events notification
2814 * module which is just interested in the timestamps.
2818 vfs_mntdummygetattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2819 caller_context_t
*ct
)
2821 bzero(vap
, sizeof (vattr_t
));
2822 vap
->va_type
= VREG
;
2824 vap
->va_ctime
= vfs_mnttab_ctime
;
2826 * it is ok to just copy mtime as the time will be monotonically
2829 vap
->va_mtime
= vfs_mnttab_mtime
;
2830 vap
->va_atime
= vap
->va_mtime
;
2835 vfs_mnttabvp_setup(void)
2838 vnodeops_t
*vfs_mntdummyvnops
;
2839 const fs_operation_def_t mnt_dummyvnodeops_template
[] = {
2840 VOPNAME_READ
, { .vop_read
= vfs_mntdummyread
},
2841 VOPNAME_WRITE
, { .vop_write
= vfs_mntdummywrite
},
2842 VOPNAME_GETATTR
, { .vop_getattr
= vfs_mntdummygetattr
},
2843 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
2847 if (vn_make_ops("mnttab", mnt_dummyvnodeops_template
,
2848 &vfs_mntdummyvnops
) != 0) {
2849 cmn_err(CE_WARN
, "vfs_mnttabvp_setup: vn_make_ops failed");
2850 /* Shouldn't happen, but not bad enough to panic */
2855 * A global dummy vnode is allocated to represent mntfs files.
2856 * The mntfs file (/etc/mnttab) can be monitored for file events
2857 * and receive an event when mnttab changes. Dummy VOP calls
2858 * will be made on this vnode. The file events notification module
2859 * intercepts this vnode and delivers relevant events.
2861 tvp
= vn_alloc(KM_SLEEP
);
2862 tvp
->v_flag
= VNOMOUNT
|VNOMAP
|VNOSWAP
|VNOCACHE
;
2863 vn_setops(tvp
, vfs_mntdummyvnops
);
2866 * The mnt dummy ops do not reference v_data.
2867 * No other module intercepting this vnode should either.
2868 * Just set it to point to itself.
2870 tvp
->v_data
= (caddr_t
)tvp
;
2871 tvp
->v_vfsp
= rootvfs
;
2872 vfs_mntdummyvp
= tvp
;
2876 * performs fake read/write ops
2879 vfs_mnttab_rwop(int rw
)
2885 if (vfs_mntdummyvp
== NULL
)
2888 bzero(&uio
, sizeof (uio
));
2889 bzero(&iov
, sizeof (iov
));
2894 uio
.uio_loffset
= 0;
2895 uio
.uio_segflg
= UIO_SYSSPACE
;
2898 (void) VOP_WRITE(vfs_mntdummyvp
, &uio
, 0, kcred
, NULL
);
2900 (void) VOP_READ(vfs_mntdummyvp
, &uio
, 0, kcred
, NULL
);
2905 * Generate a write operation.
2908 vfs_mnttab_writeop(void)
2914 * Generate a read operation.
2917 vfs_mnttab_readop(void)
2923 * Free any mnttab information recorded in the vfs struct.
2924 * The vfs must not be on the vfs list.
2927 vfs_freemnttab(struct vfs
*vfsp
)
2929 ASSERT(!VFS_ON_LIST(vfsp
));
2932 * Free device and mount point information
2934 if (vfsp
->vfs_mntpt
!= NULL
) {
2935 refstr_rele(vfsp
->vfs_mntpt
);
2936 vfsp
->vfs_mntpt
= NULL
;
2938 if (vfsp
->vfs_resource
!= NULL
) {
2939 refstr_rele(vfsp
->vfs_resource
);
2940 vfsp
->vfs_resource
= NULL
;
2943 * Now free mount options information
2945 vfs_freeopttbl(&vfsp
->vfs_mntopts
);
2949 * Return the last mnttab modification time
2952 vfs_mnttab_modtime(timespec_t
*ts
)
2954 ASSERT(RW_LOCK_HELD(&vfslist
));
2955 *ts
= vfs_mnttab_mtime
;
2959 * See if mnttab is changed
2962 vfs_mnttab_poll(timespec_t
*old
, struct pollhead
**phpp
)
2966 *phpp
= (struct pollhead
*)NULL
;
2969 * Note: don't grab vfs list lock before accessing vfs_mnttab_mtime.
2970 * Can lead to deadlock against vfs_mnttab_modtimeupd(). It is safe
2971 * to not grab the vfs list lock because tv_sec is monotonically
2975 changed
= (old
->tv_nsec
!= vfs_mnttab_mtime
.tv_nsec
) ||
2976 (old
->tv_sec
!= vfs_mnttab_mtime
.tv_sec
);
2978 *phpp
= &vfs_pollhd
;
2982 /* Provide a unique and monotonically-increasing timestamp. */
2984 vfs_mono_time(timespec_t
*ts
)
2986 static volatile hrtime_t hrt
; /* The saved time. */
2987 hrtime_t newhrt
, oldhrt
; /* For effecting the CAS. */
2991 * Try gethrestime() first, but be prepared to fabricate a sensible
2992 * answer at the first sign of any trouble.
2994 gethrestime(&newts
);
2995 newhrt
= ts2hrt(&newts
);
3000 if (cas64((uint64_t *)&hrt
, oldhrt
, newhrt
) == oldhrt
)
3007 * Update the mnttab modification time and wake up any waiters for
3011 vfs_mnttab_modtimeupd()
3013 hrtime_t oldhrt
, newhrt
;
3015 ASSERT(RW_WRITE_HELD(&vfslist
));
3016 oldhrt
= ts2hrt(&vfs_mnttab_mtime
);
3017 gethrestime(&vfs_mnttab_mtime
);
3018 newhrt
= ts2hrt(&vfs_mnttab_mtime
);
3019 if (oldhrt
== (hrtime_t
)0)
3020 vfs_mnttab_ctime
= vfs_mnttab_mtime
;
3022 * Attempt to provide unique mtime (like uniqtime but not).
3024 if (newhrt
== oldhrt
) {
3026 hrt2ts(newhrt
, &vfs_mnttab_mtime
);
3028 pollwakeup(&vfs_pollhd
, (short)POLLRDBAND
);
3029 vfs_mnttab_writeop();
3033 dounmount(struct vfs
*vfsp
, int flag
, cred_t
*cr
)
3037 extern void teardown_vopstats(vfs_t
*);
3040 * Get covered vnode. This will be NULL if the vfs is not linked
3041 * into the file system name space (i.e., domount() with MNT_NOSPICE).
3043 coveredvp
= vfsp
->vfs_vnodecovered
;
3044 ASSERT(coveredvp
== NULL
|| vn_vfswlock_held(coveredvp
));
3047 * Purge all dnlc entries for this vfs.
3049 (void) dnlc_purge_vfsp(vfsp
, 0);
3051 /* For forcible umount, skip VFS_SYNC() since it may hang */
3052 if ((flag
& MS_FORCE
) == 0)
3053 (void) VFS_SYNC(vfsp
, 0, cr
);
3056 * Lock the vfs to maintain fs status quo during unmount. This
3057 * has to be done after the sync because ufs_update tries to acquire
3060 vfs_lock_wait(vfsp
);
3062 if (error
= VFS_UNMOUNT(vfsp
, flag
, cr
)) {
3064 if (coveredvp
!= NULL
)
3065 vn_vfsunlock(coveredvp
);
3066 } else if (coveredvp
!= NULL
) {
3067 teardown_vopstats(vfsp
);
3069 * vfs_remove() will do a VN_RELE(vfsp->vfs_vnodecovered)
3070 * when it frees vfsp so we do a VN_HOLD() so we can
3071 * continue to use coveredvp afterwards.
3075 vn_vfsunlock(coveredvp
);
3078 teardown_vopstats(vfsp
);
3080 * Release the reference to vfs that is not linked
3081 * into the name space.
3091 * Vfs_unmountall() is called by uadmin() to unmount all
3092 * mounted file systems (except the root file system) during shutdown.
3093 * It follows the existing locking protocol when traversing the vfs list
3094 * to sync and unmount vfses. Even though there should be no
3095 * other thread running while the system is shutting down, it is prudent
3096 * to still follow the locking protocol.
3099 vfs_unmountall(void)
3102 struct vfs
*prev_vfsp
= NULL
;
3106 * Toss all dnlc entries now so that the per-vfs sync
3107 * and unmount operations don't have to slog through
3108 * a bunch of uninteresting vnodes over and over again.
3113 for (vfsp
= rootvfs
->vfs_prev
; vfsp
!= rootvfs
; vfsp
= prev_vfsp
) {
3114 prev_vfsp
= vfsp
->vfs_prev
;
3116 if (vfs_lock(vfsp
) != 0)
3118 error
= vn_vfswlock(vfsp
->vfs_vnodecovered
);
3125 (void) VFS_SYNC(vfsp
, SYNC_CLOSE
, CRED());
3126 (void) dounmount(vfsp
, 0, CRED());
3129 * Since we dropped the vfslist lock above we must
3130 * verify that next_vfsp still exists, else start over.
3133 for (vfsp
= rootvfs
->vfs_prev
;
3134 vfsp
!= rootvfs
; vfsp
= vfsp
->vfs_prev
)
3135 if (vfsp
== prev_vfsp
)
3137 if (vfsp
== rootvfs
&& prev_vfsp
!= rootvfs
)
3138 prev_vfsp
= rootvfs
->vfs_prev
;
3144 * Called to add an entry to the end of the vfs mount in progress list
3147 vfs_addmip(dev_t dev
, struct vfs
*vfsp
)
3151 mipp
= (struct ipmnt
*)kmem_alloc(sizeof (struct ipmnt
), KM_SLEEP
);
3152 mipp
->mip_next
= NULL
;
3153 mipp
->mip_dev
= dev
;
3154 mipp
->mip_vfsp
= vfsp
;
3155 mutex_enter(&vfs_miplist_mutex
);
3156 if (vfs_miplist_end
!= NULL
)
3157 vfs_miplist_end
->mip_next
= mipp
;
3160 vfs_miplist_end
= mipp
;
3161 mutex_exit(&vfs_miplist_mutex
);
3165 * Called to remove an entry from the mount in progress list
3166 * Either because the mount completed or it failed.
3169 vfs_delmip(struct vfs
*vfsp
)
3171 struct ipmnt
*mipp
, *mipprev
;
3173 mutex_enter(&vfs_miplist_mutex
);
3175 for (mipp
= vfs_miplist
;
3176 mipp
&& mipp
->mip_vfsp
!= vfsp
; mipp
= mipp
->mip_next
) {
3180 return; /* shouldn't happen */
3181 if (mipp
== vfs_miplist_end
)
3182 vfs_miplist_end
= mipprev
;
3183 if (mipprev
== NULL
)
3184 vfs_miplist
= mipp
->mip_next
;
3186 mipprev
->mip_next
= mipp
->mip_next
;
3187 mutex_exit(&vfs_miplist_mutex
);
3188 kmem_free(mipp
, sizeof (struct ipmnt
));
3192 * vfs_add is called by a specific filesystem's mount routine to add
3193 * the new vfs into the vfs list/hash and to cover the mounted-on vnode.
3194 * The vfs should already have been locked by the caller.
3196 * coveredvp is NULL if this is the root.
3199 vfs_add(vnode_t
*coveredvp
, struct vfs
*vfsp
, int mflag
)
3203 ASSERT(vfs_lock_held(vfsp
));
3205 newflag
= vfsp
->vfs_flag
;
3206 if (mflag
& MS_RDONLY
)
3207 newflag
|= VFS_RDONLY
;
3209 newflag
&= ~VFS_RDONLY
;
3210 if (mflag
& MS_NOSUID
)
3211 newflag
|= (VFS_NOSETUID
|VFS_NODEVICES
);
3213 newflag
&= ~(VFS_NOSETUID
|VFS_NODEVICES
);
3214 if (mflag
& MS_NOMNTTAB
)
3215 newflag
|= VFS_NOMNTTAB
;
3217 newflag
&= ~VFS_NOMNTTAB
;
3219 if (coveredvp
!= NULL
) {
3220 ASSERT(vn_vfswlock_held(coveredvp
));
3221 coveredvp
->v_vfsmountedhere
= vfsp
;
3224 vfsp
->vfs_vnodecovered
= coveredvp
;
3225 vfsp
->vfs_flag
= newflag
;
3231 * Remove a vfs from the vfs list, null out the pointer from the
3232 * covered vnode to the vfs (v_vfsmountedhere), and null out the pointer
3233 * from the vfs to the covered vnode (vfs_vnodecovered). Release the
3234 * reference to the vfs and to the covered vnode.
3236 * Called from dounmount after it's confirmed with the file system
3237 * that the unmount is legal.
3240 vfs_remove(struct vfs
*vfsp
)
3244 ASSERT(vfs_lock_held(vfsp
));
3247 * Can't unmount root. Should never happen because fs will
3250 if (vfsp
== rootvfs
)
3251 panic("vfs_remove: unmounting root");
3253 vfs_list_remove(vfsp
);
3256 * Unhook from the file system name space.
3258 vp
= vfsp
->vfs_vnodecovered
;
3259 ASSERT(vn_vfswlock_held(vp
));
3260 vp
->v_vfsmountedhere
= NULL
;
3261 vfsp
->vfs_vnodecovered
= NULL
;
3265 * Release lock and wakeup anybody waiting.
3272 * Lock a filesystem to prevent access to it while mounting,
3273 * unmounting and syncing. Return EBUSY immediately if lock
3274 * can't be acquired.
3277 vfs_lock(vfs_t
*vfsp
)
3279 vn_vfslocks_entry_t
*vpvfsentry
;
3281 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3282 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_WRITER
))
3285 vn_vfslocks_rele(vpvfsentry
);
3290 vfs_rlock(vfs_t
*vfsp
)
3292 vn_vfslocks_entry_t
*vpvfsentry
;
3294 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3296 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_READER
))
3299 vn_vfslocks_rele(vpvfsentry
);
3304 vfs_lock_wait(vfs_t
*vfsp
)
3306 vn_vfslocks_entry_t
*vpvfsentry
;
3308 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3309 rwst_enter(&vpvfsentry
->ve_lock
, RW_WRITER
);
3313 vfs_rlock_wait(vfs_t
*vfsp
)
3315 vn_vfslocks_entry_t
*vpvfsentry
;
3317 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3318 rwst_enter(&vpvfsentry
->ve_lock
, RW_READER
);
3322 * Unlock a locked filesystem.
3325 vfs_unlock(vfs_t
*vfsp
)
3327 vn_vfslocks_entry_t
*vpvfsentry
;
3330 * vfs_unlock will mimic sema_v behaviour to fix 4748018.
3331 * And these changes should remain for the patch changes as it is.
3337 * ve_refcount needs to be dropped twice here.
3338 * 1. To release refernce after a call to vfs_locks_getlock()
3339 * 2. To release the reference from the locking routines like
3340 * vfs_rlock_wait/vfs_wlock_wait/vfs_wlock etc,.
3343 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3344 vn_vfslocks_rele(vpvfsentry
);
3346 rwst_exit(&vpvfsentry
->ve_lock
);
3347 vn_vfslocks_rele(vpvfsentry
);
3351 * Utility routine that allows a filesystem to construct its
3352 * fsid in "the usual way" - by munging some underlying dev_t and
3353 * the filesystem type number into the 64-bit fsid. Note that
3354 * this implicitly relies on dev_t persistence to make filesystem
3357 * There's nothing to prevent an individual fs from constructing its
3358 * fsid in a different way, and indeed they should.
3360 * Since we want fsids to be 32-bit quantities (so that they can be
3361 * exported identically by either 32-bit or 64-bit APIs, as well as
3362 * the fact that fsid's are "known" to NFS), we compress the device
3363 * number given down to 32-bits, and panic if that isn't possible.
3366 vfs_make_fsid(fsid_t
*fsi
, dev_t dev
, int val
)
3368 if (!cmpldev((dev32_t
*)&fsi
->val
[0], dev
))
3369 panic("device number too big for fsid!");
3374 vfs_lock_held(vfs_t
*vfsp
)
3377 vn_vfslocks_entry_t
*vpvfsentry
;
3380 * vfs_lock_held will mimic sema_held behaviour
3381 * if panicstr is set. And these changes should remain
3382 * for the patch changes as it is.
3387 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3388 held
= rwst_lock_held(&vpvfsentry
->ve_lock
, RW_WRITER
);
3390 vn_vfslocks_rele(vpvfsentry
);
3395 vfs_lock_owner(vfs_t
*vfsp
)
3397 struct _kthread
*owner
;
3398 vn_vfslocks_entry_t
*vpvfsentry
;
3401 * vfs_wlock_held will mimic sema_held behaviour
3402 * if panicstr is set. And these changes should remain
3403 * for the patch changes as it is.
3408 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3409 owner
= rwst_owner(&vpvfsentry
->ve_lock
);
3411 vn_vfslocks_rele(vpvfsentry
);
3418 * Rather than manipulate the vfslist lock directly, we abstract into lock
3419 * and unlock routines to allow the locking implementation to be changed for
3422 * Whenever the vfs list is modified through its hash links, the overall list
3423 * lock must be obtained before locking the relevant hash bucket. But to see
3424 * whether a given vfs is on the list, it suffices to obtain the lock for the
3425 * hash bucket without getting the overall list lock. (See getvfs() below.)
3431 rw_enter(&vfslist
, RW_WRITER
);
3435 vfs_list_read_lock()
3437 rw_enter(&vfslist
, RW_READER
);
3447 * Low level worker routines for adding entries to and removing entries from
3452 vfs_hash_add(struct vfs
*vfsp
, int insert_at_head
)
3458 ASSERT(RW_WRITE_HELD(&vfslist
));
3460 dev
= expldev(vfsp
->vfs_fsid
.val
[0]);
3461 vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3463 mutex_enter(&rvfs_list
[vhno
].rvfs_lock
);
3466 * Link into the hash table, inserting it at the end, so that LOFS
3467 * with the same fsid as UFS (or other) file systems will not hide the
3470 if (insert_at_head
) {
3471 vfsp
->vfs_hash
= rvfs_list
[vhno
].rvfs_head
;
3472 rvfs_list
[vhno
].rvfs_head
= vfsp
;
3474 for (hp
= &rvfs_list
[vhno
].rvfs_head
; *hp
!= NULL
;
3475 hp
= &(*hp
)->vfs_hash
)
3478 * hp now contains the address of the pointer to update
3479 * to effect the insertion.
3481 vfsp
->vfs_hash
= NULL
;
3485 rvfs_list
[vhno
].rvfs_len
++;
3486 mutex_exit(&rvfs_list
[vhno
].rvfs_lock
);
3491 vfs_hash_remove(struct vfs
*vfsp
)
3497 ASSERT(RW_WRITE_HELD(&vfslist
));
3499 dev
= expldev(vfsp
->vfs_fsid
.val
[0]);
3500 vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3502 mutex_enter(&rvfs_list
[vhno
].rvfs_lock
);
3507 if (rvfs_list
[vhno
].rvfs_head
== vfsp
) {
3508 rvfs_list
[vhno
].rvfs_head
= vfsp
->vfs_hash
;
3509 rvfs_list
[vhno
].rvfs_len
--;
3512 for (tvfsp
= rvfs_list
[vhno
].rvfs_head
; tvfsp
!= NULL
;
3513 tvfsp
= tvfsp
->vfs_hash
) {
3514 if (tvfsp
->vfs_hash
== vfsp
) {
3515 tvfsp
->vfs_hash
= vfsp
->vfs_hash
;
3516 rvfs_list
[vhno
].rvfs_len
--;
3520 cmn_err(CE_WARN
, "vfs_list_remove: vfs not found in hash");
3524 mutex_exit(&rvfs_list
[vhno
].rvfs_lock
);
3529 vfs_list_add(struct vfs
*vfsp
)
3534 * Typically, the vfs_t will have been created on behalf of the file
3535 * system in vfs_init, where it will have been provided with a
3536 * vfs_impl_t. This, however, might be lacking if the vfs_t was created
3537 * by an unbundled file system. We therefore check for such an example
3538 * before stamping the vfs_t with its creation time for the benefit of
3541 if (vfsp
->vfs_implp
== NULL
)
3542 vfsimpl_setup(vfsp
);
3543 vfs_mono_time(&vfsp
->vfs_hrctime
);
3546 * The zone that owns the mount is the one that performed the mount.
3547 * Note that this isn't necessarily the same as the zone mounted into.
3548 * The corresponding zone_rele() will be done when the vfs_t is
3551 vfsp
->vfs_zone
= curproc
->p_zone
;
3552 zone_hold(vfsp
->vfs_zone
);
3555 * Find the zone mounted into, and put this mount on its vfs list.
3557 zone
= zone_find_by_path(refstr_value(vfsp
->vfs_mntpt
));
3558 ASSERT(zone
!= NULL
);
3560 * Special casing for the root vfs. This structure is allocated
3561 * statically and hooked onto rootvfs at link time. During the
3562 * vfs_mountroot call at system startup time, the root file system's
3563 * VFS_MOUNTROOT routine will call vfs_add with this root vfs struct
3564 * as argument. The code below must detect and handle this special
3565 * case. The only apparent justification for this special casing is
3566 * to ensure that the root file system appears at the head of the
3569 * XXX: I'm assuming that it's ok to do normal list locking when
3570 * adding the entry for the root file system (this used to be
3571 * done with no locks held).
3575 * Link into the vfs list proper.
3577 if (vfsp
== &root
) {
3579 * Assert: This vfs is already on the list as its first entry.
3580 * Thus, there's nothing to do.
3582 ASSERT(rootvfs
== vfsp
);
3584 * Add it to the head of the global zone's vfslist.
3586 ASSERT(zone
== global_zone
);
3587 ASSERT(zone
->zone_vfslist
== NULL
);
3588 zone
->zone_vfslist
= vfsp
;
3591 * Link to end of list using vfs_prev (as rootvfs is now a
3592 * doubly linked circular list) so list is in mount order for
3595 rootvfs
->vfs_prev
->vfs_next
= vfsp
;
3596 vfsp
->vfs_prev
= rootvfs
->vfs_prev
;
3597 rootvfs
->vfs_prev
= vfsp
;
3598 vfsp
->vfs_next
= rootvfs
;
3601 * Do it again for the zone-private list (which may be NULL).
3603 if (zone
->zone_vfslist
== NULL
) {
3604 ASSERT(zone
!= global_zone
);
3605 zone
->zone_vfslist
= vfsp
;
3607 zone
->zone_vfslist
->vfs_zone_prev
->vfs_zone_next
= vfsp
;
3608 vfsp
->vfs_zone_prev
= zone
->zone_vfslist
->vfs_zone_prev
;
3609 zone
->zone_vfslist
->vfs_zone_prev
= vfsp
;
3610 vfsp
->vfs_zone_next
= zone
->zone_vfslist
;
3615 * Link into the hash table, inserting it at the end, so that LOFS
3616 * with the same fsid as UFS (or other) file systems will not hide
3619 vfs_hash_add(vfsp
, 0);
3622 * update the mnttab modification time
3624 vfs_mnttab_modtimeupd();
3630 vfs_list_remove(struct vfs
*vfsp
)
3634 zone
= zone_find_by_path(refstr_value(vfsp
->vfs_mntpt
));
3635 ASSERT(zone
!= NULL
);
3637 * Callers are responsible for preventing attempts to unmount the
3640 ASSERT(vfsp
!= rootvfs
);
3647 vfs_hash_remove(vfsp
);
3650 * Remove from vfs list.
3652 vfsp
->vfs_prev
->vfs_next
= vfsp
->vfs_next
;
3653 vfsp
->vfs_next
->vfs_prev
= vfsp
->vfs_prev
;
3654 vfsp
->vfs_next
= vfsp
->vfs_prev
= NULL
;
3657 * Remove from zone-specific vfs list.
3659 if (zone
->zone_vfslist
== vfsp
)
3660 zone
->zone_vfslist
= vfsp
->vfs_zone_next
;
3662 if (vfsp
->vfs_zone_next
== vfsp
) {
3663 ASSERT(vfsp
->vfs_zone_prev
== vfsp
);
3664 ASSERT(zone
->zone_vfslist
== vfsp
);
3665 zone
->zone_vfslist
= NULL
;
3668 vfsp
->vfs_zone_prev
->vfs_zone_next
= vfsp
->vfs_zone_next
;
3669 vfsp
->vfs_zone_next
->vfs_zone_prev
= vfsp
->vfs_zone_prev
;
3670 vfsp
->vfs_zone_next
= vfsp
->vfs_zone_prev
= NULL
;
3673 * update the mnttab modification time
3675 vfs_mnttab_modtimeupd();
3681 getvfs(fsid_t
*fsid
)
3684 int val0
= fsid
->val
[0];
3685 int val1
= fsid
->val
[1];
3686 dev_t dev
= expldev(val0
);
3687 int vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3688 kmutex_t
*hmp
= &rvfs_list
[vhno
].rvfs_lock
;
3691 for (vfsp
= rvfs_list
[vhno
].rvfs_head
; vfsp
; vfsp
= vfsp
->vfs_hash
) {
3692 if (vfsp
->vfs_fsid
.val
[0] == val0
&&
3693 vfsp
->vfs_fsid
.val
[1] == val1
) {
3704 * Search the vfs mount in progress list for a specified device/vfs entry.
3705 * Returns 0 if the first entry in the list that the device matches has the
3706 * given vfs pointer as well. If the device matches but a different vfs
3707 * pointer is encountered in the list before the given vfs pointer then
3712 vfs_devmounting(dev_t dev
, struct vfs
*vfsp
)
3717 mutex_enter(&vfs_miplist_mutex
);
3718 for (mipp
= vfs_miplist
; mipp
!= NULL
; mipp
= mipp
->mip_next
) {
3719 if (mipp
->mip_dev
== dev
) {
3720 if (mipp
->mip_vfsp
!= vfsp
)
3725 mutex_exit(&vfs_miplist_mutex
);
3730 * Search the vfs list for a specified device. Returns 1, if entry is found
3731 * or 0 if no suitable entry is found.
3735 vfs_devismounted(dev_t dev
)
3740 vfs_list_read_lock();
3744 if (vfsp
->vfs_dev
== dev
) {
3748 vfsp
= vfsp
->vfs_next
;
3749 } while (vfsp
!= rootvfs
);
3756 * Search the vfs list for a specified device. Returns a pointer to it
3757 * or NULL if no suitable entry is found. The caller of this routine
3758 * is responsible for releasing the returned vfs pointer.
3761 vfs_dev2vfsp(dev_t dev
)
3766 vfs_list_read_lock();
3771 * The following could be made more efficient by making
3772 * the entire loop use vfs_zone_next if the call is from
3773 * a zone. The only callers, however, ustat(2) and
3774 * umount2(2), don't seem to justify the added
3775 * complexity at present.
3777 if (vfsp
->vfs_dev
== dev
&&
3778 ZONE_PATH_VISIBLE(refstr_value(vfsp
->vfs_mntpt
),
3784 vfsp
= vfsp
->vfs_next
;
3785 } while (vfsp
!= rootvfs
);
3787 return (found
? vfsp
: NULL
);
3791 * Search the vfs list for a specified mntpoint. Returns a pointer to it
3792 * or NULL if no suitable entry is found. The caller of this routine
3793 * is responsible for releasing the returned vfs pointer.
3795 * Note that if multiple mntpoints match, the last one matching is
3796 * returned in an attempt to return the "top" mount when overlay
3797 * mounts are covering the same mount point. This is accomplished by starting
3798 * at the end of the list and working our way backwards, stopping at the first
3802 vfs_mntpoint2vfsp(const char *mp
)
3805 struct vfs
*retvfsp
= NULL
;
3806 zone_t
*zone
= curproc
->p_zone
;
3809 vfs_list_read_lock();
3810 if (getzoneid() == GLOBAL_ZONEID
) {
3812 * The global zone may see filesystems in any zone.
3814 vfsp
= rootvfs
->vfs_prev
;
3816 if (strcmp(refstr_value(vfsp
->vfs_mntpt
), mp
) == 0) {
3820 vfsp
= vfsp
->vfs_prev
;
3821 } while (vfsp
!= rootvfs
->vfs_prev
);
3822 } else if ((list
= zone
->zone_vfslist
) != NULL
) {
3825 vfsp
= list
->vfs_zone_prev
;
3827 mntpt
= refstr_value(vfsp
->vfs_mntpt
);
3828 mntpt
= ZONE_PATH_TRANSLATE(mntpt
, zone
);
3829 if (strcmp(mntpt
, mp
) == 0) {
3833 vfsp
= vfsp
->vfs_zone_prev
;
3834 } while (vfsp
!= list
->vfs_zone_prev
);
3843 * Search the vfs list for a specified vfsops.
3844 * if vfs entry is found then return 1, else 0.
3847 vfs_opsinuse(vfsops_t
*ops
)
3852 vfs_list_read_lock();
3856 if (vfs_getops(vfsp
) == ops
) {
3860 vfsp
= vfsp
->vfs_next
;
3861 } while (vfsp
!= rootvfs
);
3867 * Allocate an entry in vfssw for a file system type
3870 allocate_vfssw(const char *type
)
3874 if (type
[0] == '\0' || strlen(type
) + 1 > _ST_FSTYPSZ
) {
3876 * The vfssw table uses the empty string to identify an
3877 * available entry; we cannot add any type which has
3878 * a leading NUL. The string length is limited to
3879 * the size of the st_fstype array in struct stat.
3884 ASSERT(VFSSW_WRITE_LOCKED());
3885 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++)
3886 if (!ALLOCATED_VFSSW(vswp
)) {
3887 vswp
->vsw_name
= kmem_alloc(strlen(type
) + 1, KM_SLEEP
);
3888 (void) strcpy(vswp
->vsw_name
, type
);
3889 ASSERT(vswp
->vsw_count
== 0);
3890 vswp
->vsw_count
= 1;
3891 mutex_init(&vswp
->vsw_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
3898 * Impose additional layer of translation between vfstype names
3899 * and module names in the filesystem.
3902 vfs_to_modname(const char *vfstype
)
3904 if (strcmp(vfstype
, "proc") == 0) {
3906 } else if (strcmp(vfstype
, "fd") == 0) {
3908 } else if (strncmp(vfstype
, "nfs", 3) == 0) {
3916 * Find a vfssw entry given a file system type name.
3917 * Try to autoload the filesystem if it's not found.
3918 * If it's installed, return the vfssw locked to prevent unloading.
3921 vfs_getvfssw(const char *type
)
3924 const char *modname
;
3927 vswp
= vfs_getvfsswbyname(type
);
3928 modname
= vfs_to_modname(type
);
3930 if (rootdir
== NULL
) {
3932 * If we haven't yet loaded the root file system, then our
3933 * _init won't be called until later. Allocate vfssw entry,
3934 * because mod_installfs won't be called.
3939 if ((vswp
= vfs_getvfsswbyname(type
)) == NULL
) {
3940 if ((vswp
= allocate_vfssw(type
)) == NULL
) {
3948 if (!VFS_INSTALLED(vswp
)) {
3950 (void) modloadonly("fs", modname
);
3957 * Try to load the filesystem. Before calling modload(), we drop
3958 * our lock on the VFS switch table, and pick it up after the
3959 * module is loaded. However, there is a potential race: the
3960 * module could be unloaded after the call to modload() completes
3961 * but before we pick up the lock and drive on. Therefore,
3962 * we keep reloading the module until we've loaded the module
3963 * _and_ we have the lock on the VFS switch table.
3965 while (vswp
== NULL
|| !VFS_INSTALLED(vswp
)) {
3967 if (modload("fs", modname
) == -1)
3971 if ((vswp
= vfs_getvfsswbyname(type
)) == NULL
)
3980 * Find a vfssw entry given a file system type name.
3983 vfs_getvfsswbyname(const char *type
)
3987 ASSERT(VFSSW_LOCKED());
3988 if (type
== NULL
|| *type
== '\0')
3991 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
3992 if (strcmp(type
, vswp
->vsw_name
) == 0) {
4002 * Find a vfssw entry given a set of vfsops.
4005 vfs_getvfsswbyvfsops(vfsops_t
*vfsops
)
4010 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
4011 if (ALLOCATED_VFSSW(vswp
) && &vswp
->vsw_vfsops
== vfsops
) {
4023 * Reference a vfssw entry.
4026 vfs_refvfssw(struct vfssw
*vswp
)
4029 mutex_enter(&vswp
->vsw_lock
);
4031 mutex_exit(&vswp
->vsw_lock
);
4035 * Unreference a vfssw entry.
4038 vfs_unrefvfssw(struct vfssw
*vswp
)
4041 mutex_enter(&vswp
->vsw_lock
);
4043 mutex_exit(&vswp
->vsw_lock
);
4046 int sync_timeout
= 30; /* timeout for syncing a page during panic */
4047 int sync_timeleft
; /* portion of sync_timeout remaining */
4049 static int sync_retries
= 20; /* number of retries when not making progress */
4050 static int sync_triesleft
; /* portion of sync_retries remaining */
4052 static pgcnt_t old_pgcnt
, new_pgcnt
;
4053 static int new_bufcnt
, old_bufcnt
;
4056 * Sync all of the mounted filesystems, and then wait for the actual i/o to
4057 * complete. We wait by counting the number of dirty pages and buffers,
4058 * pushing them out using bio_busy() and page_busy(), and then counting again.
4059 * This routine is used during both the uadmin A_SHUTDOWN code as well as
4060 * the SYNC phase of the panic code (see comments in panic.c). It should only
4061 * be used after some higher-level mechanism has quiesced the system so that
4062 * new writes are not being initiated while we are waiting for completion.
4064 * To ensure finite running time, our algorithm uses two timeout mechanisms:
4065 * sync_timeleft (a timer implemented by the omnipresent deadman() cyclic), and
4066 * sync_triesleft (a progress counter used by the vfs_syncall() loop below).
4067 * Together these ensure that syncing completes if our i/o paths are stuck.
4068 * The counters are declared above so they can be found easily in the debugger.
4070 * The sync_timeleft counter is reset by bio_busy() and page_busy() using the
4071 * vfs_syncprogress() subroutine whenever we make progress through the lists of
4072 * pages and buffers. It is decremented and expired by the deadman() cyclic.
4073 * When vfs_syncall() decides it is done, we disable the deadman() counter by
4074 * setting sync_timeleft to zero. This timer guards against vfs_syncall()
4075 * deadlocking or hanging inside of a broken filesystem or driver routine.
4077 * The sync_triesleft counter is updated by vfs_syncall() itself. If we make
4078 * sync_retries consecutive calls to bio_busy() and page_busy() without
4079 * decreasing either the number of dirty buffers or dirty pages below the
4080 * lowest count we have seen so far, we give up and return from vfs_syncall().
4082 * Each loop iteration ends with a call to delay() one second to allow time for
4083 * i/o completion and to permit the user time to read our progress messages.
4088 if (rootdir
== NULL
&& !modrootloaded
)
4089 return; /* panic during boot - no filesystems yet */
4091 printf("syncing file systems...");
4096 sync_triesleft
= sync_retries
;
4098 old_bufcnt
= new_bufcnt
= INT_MAX
;
4099 old_pgcnt
= new_pgcnt
= ULONG_MAX
;
4101 while (sync_triesleft
> 0) {
4102 old_bufcnt
= MIN(old_bufcnt
, new_bufcnt
);
4103 old_pgcnt
= MIN(old_pgcnt
, new_pgcnt
);
4105 new_bufcnt
= bio_busy(B_TRUE
);
4106 new_pgcnt
= page_busy(B_TRUE
);
4109 if (new_bufcnt
== 0 && new_pgcnt
== 0)
4112 if (new_bufcnt
< old_bufcnt
|| new_pgcnt
< old_pgcnt
)
4113 sync_triesleft
= sync_retries
;
4118 printf(" [%d]", new_bufcnt
);
4120 printf(" %lu", new_pgcnt
);
4125 if (new_bufcnt
!= 0 || new_pgcnt
!= 0)
4126 printf(" done (not all i/o completed)\n");
4135 * If we are in the middle of the sync phase of panic, reset sync_timeleft to
4136 * sync_timeout to indicate that we are making progress and the deadman()
4137 * omnipresent cyclic should not yet time us out. Note that it is safe to
4138 * store to sync_timeleft here since the deadman() is firing at high-level
4139 * on top of us. If we are racing with the deadman(), either the deadman()
4140 * will decrement the old value and then we will reset it, or we will
4141 * reset it and then the deadman() will immediately decrement it. In either
4142 * case, correct behavior results.
4145 vfs_syncprogress(void)
4148 sync_timeleft
= sync_timeout
;
4152 * Map VFS flags to statvfs flags. These shouldn't really be separate
4156 vf_to_stf(uint_t vf
)
4160 if (vf
& VFS_RDONLY
)
4162 if (vf
& VFS_NOSETUID
)
4164 if (vf
& VFS_NOTRUNC
)
4171 * Entries for (illegal) fstype 0.
4175 vfsstray_sync(struct vfs
*vfsp
, short arg
, struct cred
*cr
)
4177 cmn_err(CE_PANIC
, "stray vfs operation");
4182 * Entries for (illegal) fstype 0.
4187 cmn_err(CE_PANIC
, "stray vfs operation");
4192 * Support for dealing with forced UFS unmount and its interaction with
4193 * LOFS. Could be used by any filesystem.
4203 * We've gotta define the op for sync separately, since the compiler gets
4204 * confused if we mix and match ANSI and normal style prototypes when
4205 * a "short" argument is present and spits out a warning.
4209 vfs_EIO_sync(struct vfs
*vfsp
, short arg
, struct cred
*cr
)
4215 vfsops_t
*EIO_vfsops
;
4218 * Called from startup() to initialize all loaded vfs's
4225 extern int vopstats_enabled
;
4226 extern void vopstats_startup();
4228 static const fs_operation_def_t EIO_vfsops_template
[] = {
4229 VFSNAME_MOUNT
, { .error
= vfs_EIO
},
4230 VFSNAME_UNMOUNT
, { .error
= vfs_EIO
},
4231 VFSNAME_ROOT
, { .error
= vfs_EIO
},
4232 VFSNAME_STATVFS
, { .error
= vfs_EIO
},
4233 VFSNAME_SYNC
, { .vfs_sync
= vfs_EIO_sync
},
4234 VFSNAME_VGET
, { .error
= vfs_EIO
},
4235 VFSNAME_MOUNTROOT
, { .error
= vfs_EIO
},
4236 VFSNAME_FREEVFS
, { .error
= vfs_EIO
},
4237 VFSNAME_VNSTATE
, { .error
= vfs_EIO
},
4241 static const fs_operation_def_t stray_vfsops_template
[] = {
4242 VFSNAME_MOUNT
, { .error
= vfsstray
},
4243 VFSNAME_UNMOUNT
, { .error
= vfsstray
},
4244 VFSNAME_ROOT
, { .error
= vfsstray
},
4245 VFSNAME_STATVFS
, { .error
= vfsstray
},
4246 VFSNAME_SYNC
, { .vfs_sync
= vfsstray_sync
},
4247 VFSNAME_VGET
, { .error
= vfsstray
},
4248 VFSNAME_MOUNTROOT
, { .error
= vfsstray
},
4249 VFSNAME_FREEVFS
, { .error
= vfsstray
},
4250 VFSNAME_VNSTATE
, { .error
= vfsstray
},
4254 /* Create vfs cache */
4255 vfs_cache
= kmem_cache_create("vfs_cache", sizeof (struct vfs
),
4256 sizeof (uintptr_t), NULL
, NULL
, NULL
, NULL
, NULL
, 0);
4258 /* Initialize the vnode cache (file systems may use it during init). */
4261 /* Setup event monitor framework */
4264 /* Initialize the dummy stray file system type. */
4265 error
= vfs_setfsops(0, stray_vfsops_template
, NULL
);
4267 /* Initialize the dummy EIO file system. */
4268 error
= vfs_makefsops(EIO_vfsops_template
, &EIO_vfsops
);
4270 cmn_err(CE_WARN
, "vfsinit: bad EIO vfs ops template");
4271 /* Shouldn't happen, but not bad enough to panic */
4274 VFS_INIT(&EIO_vfs
, EIO_vfsops
, (caddr_t
)NULL
);
4277 * Default EIO_vfs.vfs_flag to VFS_UNMOUNTED so a lookup
4278 * on this vfs can immediately notice it's invalid.
4280 EIO_vfs
.vfs_flag
|= VFS_UNMOUNTED
;
4283 * Call the init routines of non-loadable filesystems only.
4284 * Filesystems which are loaded as separate modules will be
4285 * initialized by the module loading code instead.
4288 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
4290 if (vswp
->vsw_init
!= NULL
)
4291 (*vswp
->vsw_init
)(vswp
- vfssw
, vswp
->vsw_name
);
4297 if (vopstats_enabled
) {
4298 /* EIO_vfs can collect stats, but we don't retrieve them */
4299 initialize_vopstats(&EIO_vfs
.vfs_vopstats
);
4300 EIO_vfs
.vfs_fstypevsp
= NULL
;
4301 EIO_vfs
.vfs_vskap
= NULL
;
4302 EIO_vfs
.vfs_flag
|= VFS_STATS
;
4307 reparse_point_init();
4311 vfs_alloc(int kmflag
)
4315 vfsp
= kmem_cache_alloc(vfs_cache
, kmflag
);
4318 * Do the simplest initialization here.
4319 * Everything else gets done in vfs_init()
4321 bzero(vfsp
, sizeof (vfs_t
));
4326 vfs_free(vfs_t
*vfsp
)
4329 * One would be tempted to assert that "vfsp->vfs_count == 0".
4330 * The problem is that this gets called out of domount() with
4331 * a partially initialized vfs and a vfs_count of 1. This is
4332 * also called from vfs_rele() with a vfs_count of 0. We can't
4333 * call VFS_RELE() from domount() if VFS_MOUNT() hasn't successfully
4334 * returned. This is because VFS_MOUNT() fully initializes the
4335 * vfs structure and its associated data. VFS_RELE() will call
4336 * VFS_FREEVFS() which may panic the system if the data structures
4337 * aren't fully initialized from a successful VFS_MOUNT()).
4340 /* If FEM was in use, make sure everything gets cleaned up */
4341 if (vfsp
->vfs_femhead
) {
4342 ASSERT(vfsp
->vfs_femhead
->femh_list
== NULL
);
4343 mutex_destroy(&vfsp
->vfs_femhead
->femh_lock
);
4344 kmem_free(vfsp
->vfs_femhead
, sizeof (*(vfsp
->vfs_femhead
)));
4345 vfsp
->vfs_femhead
= NULL
;
4348 if (vfsp
->vfs_implp
)
4349 vfsimpl_teardown(vfsp
);
4350 sema_destroy(&vfsp
->vfs_reflock
);
4351 kmem_cache_free(vfs_cache
, vfsp
);
4355 * Increments the vfs reference count by one atomically.
4358 vfs_hold(vfs_t
*vfsp
)
4360 atomic_add_32(&vfsp
->vfs_count
, 1);
4361 ASSERT(vfsp
->vfs_count
!= 0);
4365 * Decrements the vfs reference count by one atomically. When
4366 * vfs reference count becomes zero, it calls the file system
4367 * specific vfs_freevfs() to free up the resources.
4370 vfs_rele(vfs_t
*vfsp
)
4372 ASSERT(vfsp
->vfs_count
!= 0);
4373 if (atomic_add_32_nv(&vfsp
->vfs_count
, -1) == 0) {
4377 zone_rele(vfsp
->vfs_zone
);
4378 vfs_freemnttab(vfsp
);
4384 * Generic operations vector support.
4386 * This is used to build operations vectors for both the vfs and vnode.
4387 * It's normally called only when a file system is loaded.
4389 * There are many possible algorithms for this, including the following:
4391 * (1) scan the list of known operations; for each, see if the file system
4392 * includes an entry for it, and fill it in as appropriate.
4394 * (2) set up defaults for all known operations. scan the list of ops
4395 * supplied by the file system; for each which is both supplied and
4396 * known, fill it in.
4398 * (3) sort the lists of known ops & supplied ops; scan the list, filling
4399 * in entries as we go.
4401 * we choose (1) for simplicity, and because performance isn't critical here.
4402 * note that (2) could be sped up using a precomputed hash table on known ops.
4403 * (3) could be faster than either, but only if the lists were very large or
4404 * supplied in sorted order.
4409 fs_build_vector(void *vector
, int *unused_ops
,
4410 const fs_operation_trans_def_t
*translation
,
4411 const fs_operation_def_t
*operations
)
4413 int i
, num_trans
, num_ops
, used
;
4416 * Count the number of translations and the number of supplied
4421 const fs_operation_trans_def_t
*p
;
4423 for (num_trans
= 0, p
= translation
;
4430 const fs_operation_def_t
*p
;
4432 for (num_ops
= 0, p
= operations
;
4438 /* Walk through each operation known to our caller. There will be */
4439 /* one entry in the supplied "translation table" for each. */
4443 for (i
= 0; i
< num_trans
; i
++) {
4446 fs_generic_func_p result
;
4447 fs_generic_func_p
*location
;
4449 curname
= translation
[i
].name
;
4451 /* Look for a matching operation in the list supplied by the */
4456 for (j
= 0; j
< num_ops
; j
++) {
4457 if (strcmp(operations
[j
].name
, curname
) == 0) {
4465 * If the file system is using a "placeholder" for default
4466 * or error functions, grab the appropriate function out of
4467 * the translation table. If the file system didn't supply
4468 * this operation at all, use the default function.
4472 result
= operations
[j
].func
.fs_generic
;
4473 if (result
== fs_default
) {
4474 result
= translation
[i
].defaultFunc
;
4475 } else if (result
== fs_error
) {
4476 result
= translation
[i
].errorFunc
;
4477 } else if (result
== NULL
) {
4478 /* Null values are PROHIBITED */
4482 result
= translation
[i
].defaultFunc
;
4485 /* Now store the function into the operations vector. */
4487 location
= (fs_generic_func_p
*)
4488 (((char *)vector
) + translation
[i
].offset
);
4493 *unused_ops
= num_ops
- used
;
4498 /* Placeholder functions, should never be called. */
4503 cmn_err(CE_PANIC
, "fs_error called");
4510 cmn_err(CE_PANIC
, "fs_default called");
4517 * Part of the implementation of booting off a mirrored root
4518 * involves a change of dev_t for the root device. To
4519 * accomplish this, first remove the existing hash table
4520 * entry for the root device, convert to the new dev_t,
4521 * then re-insert in the hash table at the head of the list.
4524 vfs_root_redev(vfs_t
*vfsp
, dev_t ndev
, int fstype
)
4528 vfs_hash_remove(vfsp
);
4530 vfsp
->vfs_dev
= ndev
;
4531 vfs_make_fsid(&vfsp
->vfs_fsid
, ndev
, fstype
);
4533 vfs_hash_add(vfsp
, 1);
4538 #else /* x86 NEWBOOT */
4541 extern int hvmboot_rootconf();
4544 extern ib_boot_prop_t
*iscsiboot_prop
;
4551 extern void pm_init();
4552 char *fstyp
, *fsmod
;
4555 getrootfs(&fstyp
, &fsmod
);
4559 * hvmboot_rootconf() is defined in the hvm_bootstrap misc module,
4560 * which lives in /platform/i86hvm, and hence is only available when
4561 * booted in an x86 hvm environment. If the hvm_bootstrap misc module
4562 * is not available then the modstub for this function will return 0.
4563 * If the hvm_bootstrap misc module is available it will be loaded
4564 * and hvmboot_rootconf() will be invoked.
4566 if (error
= hvmboot_rootconf())
4570 if (error
= clboot_rootconf())
4573 if (modload("fs", fsmod
) == -1)
4574 panic("Cannot _init %s module", fsmod
);
4577 vsw
= vfs_getvfsswbyname(fstyp
);
4580 cmn_err(CE_CONT
, "Cannot find %s filesystem\n", fstyp
);
4583 VFS_INIT(rootvfs
, &vsw
->vsw_vfsops
, 0);
4586 /* always mount readonly first */
4587 rootvfs
->vfs_flag
|= VFS_RDONLY
;
4591 if (netboot
&& iscsiboot_prop
) {
4592 cmn_err(CE_WARN
, "NFS boot and iSCSI boot"
4593 " shouldn't happen in the same time");
4597 if (netboot
|| iscsiboot_prop
) {
4600 cmn_err(CE_WARN
, "Cannot plumb network device %d", ret
);
4605 if ((ret
== 0) && iscsiboot_prop
) {
4606 ret
= modload("drv", "iscsi");
4607 /* -1 indicates fail */
4609 cmn_err(CE_WARN
, "Failed to load iscsi module");
4610 iscsi_boot_prop_free();
4613 if (!i_ddi_attach_pseudo_node("iscsi")) {
4615 "Failed to attach iscsi driver");
4616 iscsi_boot_prop_free();
4622 error
= VFS_MOUNTROOT(rootvfs
, ROOT_INIT
);
4623 vfs_unrefvfssw(vsw
);
4624 rootdev
= rootvfs
->vfs_dev
;
4627 cmn_err(CE_CONT
, "Cannot mount root on %s fstype %s\n",
4628 rootfs
.bo_name
, fstyp
);
4630 cmn_err(CE_CONT
, "?root on %s fstype %s\n",
4631 rootfs
.bo_name
, fstyp
);
4636 * XXX this is called by nfs only and should probably be removed
4637 * If booted with ASKNAME, prompt on the console for a filesystem
4638 * name and return it.
4641 getfsname(char *askfor
, char *name
, size_t namelen
)
4643 if (boothowto
& RB_ASKNAME
) {
4644 printf("%s name: ", askfor
);
4645 console_gets(name
, namelen
);
4650 * Init the root filesystem type (rootfs.bo_fstype) from the "fstype"
4653 * Filesystem types starting with the prefix "nfs" are diskless clients;
4654 * init the root filename name (rootfs.bo_name), too.
4656 * If we are booting via NFS we currently have these options:
4657 * nfs - dynamically choose NFS V2, V3, or V4 (default)
4658 * nfs2 - force NFS V2
4659 * nfs3 - force NFS V3
4660 * nfs4 - force NFS V4
4661 * Because we need to maintain backward compatibility with the naming
4662 * convention that the NFS V2 filesystem name is "nfs" (see vfs_conf.c)
4663 * we need to map "nfs" => "nfsdyn" and "nfs2" => "nfs". The dynamic
4664 * nfs module will map the type back to either "nfs", "nfs3", or "nfs4".
4665 * This is only for root filesystems, all other uses such as cachefs
4666 * will expect that "nfs" == NFS V2.
4669 getrootfs(char **fstypp
, char **fsmodp
)
4671 extern char *strplumb_get_netdev_path(void);
4672 char *propstr
= NULL
;
4675 * Check fstype property; for diskless it should be one of "nfs",
4676 * "nfs2", "nfs3" or "nfs4".
4678 if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4679 DDI_PROP_DONTPASS
, "fstype", &propstr
)
4681 (void) strncpy(rootfs
.bo_fstype
, propstr
, BO_MAXFSNAME
);
4682 ddi_prop_free(propstr
);
4685 * if the boot property 'fstype' is not set, but 'zfs-bootfs' is set,
4686 * assume the type of this root filesystem is 'zfs'.
4688 } else if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4689 DDI_PROP_DONTPASS
, "zfs-bootfs", &propstr
)
4691 (void) strncpy(rootfs
.bo_fstype
, "zfs", BO_MAXFSNAME
);
4692 ddi_prop_free(propstr
);
4695 if (strncmp(rootfs
.bo_fstype
, "nfs", 3) != 0) {
4696 *fstypp
= *fsmodp
= rootfs
.bo_fstype
;
4702 if (strcmp(rootfs
.bo_fstype
, "nfs2") == 0)
4703 (void) strcpy(rootfs
.bo_fstype
, "nfs");
4704 else if (strcmp(rootfs
.bo_fstype
, "nfs") == 0)
4705 (void) strcpy(rootfs
.bo_fstype
, "nfsdyn");
4708 * check if path to network interface is specified in bootpath
4709 * or by a hypervisor domain configuration file.
4710 * XXPV - enable strlumb_get_netdev_path()
4712 if (ddi_prop_exists(DDI_DEV_T_ANY
, ddi_root_node(), DDI_PROP_DONTPASS
,
4714 (void) strcpy(rootfs
.bo_name
, "/xpvd/xnf@0");
4715 } else if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4716 DDI_PROP_DONTPASS
, "bootpath", &propstr
)
4718 (void) strncpy(rootfs
.bo_name
, propstr
, BO_MAXOBJNAME
);
4719 ddi_prop_free(propstr
);
4721 /* attempt to determine netdev_path via boot_mac address */
4722 netdev_path
= strplumb_get_netdev_path();
4723 if (netdev_path
== NULL
)
4724 panic("cannot find boot network interface");
4725 (void) strncpy(rootfs
.bo_name
, netdev_path
, BO_MAXOBJNAME
);
4727 *fstypp
= rootfs
.bo_fstype
;
4733 * VFS feature routines
4736 #define VFTINDEX(feature) (((feature) >> 32) & 0xFFFFFFFF)
4737 #define VFTBITS(feature) ((feature) & 0xFFFFFFFFLL)
4739 /* Register a feature in the vfs */
4741 vfs_set_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4743 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4744 if (vfsp
->vfs_implp
== NULL
)
4747 vfsp
->vfs_featureset
[VFTINDEX(feature
)] |= VFTBITS(feature
);
4751 * Query a vfs for a feature.
4752 * Returns 1 if feature is present, 0 if not
4755 vfs_has_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4759 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4760 if (vfsp
->vfs_implp
== NULL
)
4763 if (vfsp
->vfs_featureset
[VFTINDEX(feature
)] & VFTBITS(feature
))
4770 * Propagate feature set from one vfs to another
4773 vfs_propagate_features(vfs_t
*from
, vfs_t
*to
)
4777 if (to
->vfs_implp
== NULL
|| from
->vfs_implp
== NULL
)
4780 for (i
= 1; i
<= to
->vfs_featureset
[0]; i
++) {
4781 to
->vfs_featureset
[i
] = from
->vfs_featureset
[i
];
4785 #define LOFICTL_PATH "/devices/pseudo/lofi@0:%d"
4788 * Return the vnode for the lofi node if there's a lofi mount in place.
4789 * Returns -1 when there's no lofi node, 0 on success, and > 0 on
4793 vfs_get_lofi(vfs_t
*vfsp
, vnode_t
**vpp
)
4799 if (vfsp
->vfs_lofi_minor
== 0) {
4804 strsize
= snprintf(NULL
, 0, LOFICTL_PATH
, vfsp
->vfs_lofi_minor
);
4805 path
= kmem_alloc(strsize
+ 1, KM_SLEEP
);
4806 (void) snprintf(path
, strsize
+ 1, LOFICTL_PATH
, vfsp
->vfs_lofi_minor
);
4808 err
= lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, vpp
);
4813 kmem_free(path
, strsize
+ 1);