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 (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
27 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
28 /* All Rights Reserved */
31 * University Copyright- Copyright (c) 1982, 1986, 1988
32 * The Regents of the University of California
35 * University Acknowledgment- Portions of this document are derived from
36 * software developed by the University of California, Berkeley, and its
40 #include <sys/types.h>
41 #include <sys/t_lock.h>
42 #include <sys/param.h>
43 #include <sys/errno.h>
45 #include <sys/fstyp.h>
47 #include <sys/systm.h>
49 #include <sys/mount.h>
51 #include <sys/vfs_opreg.h>
53 #include <sys/mntent.h>
55 #include <sys/statvfs.h>
56 #include <sys/statfs.h>
58 #include <sys/vnode.h>
59 #include <sys/rwstlock.h>
63 #include <sys/atomic.h>
64 #include <sys/cmn_err.h>
67 #include <sys/debug.h>
68 #include <sys/vnode.h>
69 #include <sys/modctl.h>
71 #include <sys/pathname.h>
72 #include <sys/bootconf.h>
73 #include <sys/dumphdr.h>
74 #include <sys/dc_ki.h>
76 #include <sys/sunddi.h>
77 #include <sys/sysmacros.h>
79 #include <sys/policy.h>
81 #include <sys/objfs.h>
82 #include <sys/console.h>
83 #include <sys/reboot.h>
88 #include <sys/bootprops.h>
92 #include <fs/fs_subr.h>
93 /* Private interfaces to create vopstats-related data structures */
94 extern void initialize_vopstats(vopstats_t
*);
95 extern vopstats_t
*get_fstype_vopstats(struct vfs
*, struct vfssw
*);
96 extern vsk_anchor_t
*get_vskstat_anchor(struct vfs
*);
98 static void vfs_clearmntopt_nolock(mntopts_t
*, const char *, int);
99 static void vfs_setmntopt_nolock(mntopts_t
*, const char *,
100 const char *, int, int);
101 static int vfs_optionisset_nolock(const mntopts_t
*, const char *, char **);
102 static void vfs_freemnttab(struct vfs
*);
103 static void vfs_freeopt(mntopt_t
*);
104 static void vfs_swapopttbl_nolock(mntopts_t
*, mntopts_t
*);
105 static void vfs_swapopttbl(mntopts_t
*, mntopts_t
*);
106 static void vfs_copyopttbl_extend(const mntopts_t
*, mntopts_t
*, int);
107 static void vfs_createopttbl_extend(mntopts_t
*, const char *,
109 static char **vfs_copycancelopt_extend(char **const, int);
110 static void vfs_freecancelopt(char **);
111 static void getrootfs(char **, char **);
112 static int getmacpath(dev_info_t
*, void *);
113 static void vfs_mnttabvp_setup(void);
116 struct ipmnt
*mip_next
;
118 struct vfs
*mip_vfsp
;
121 static kmutex_t vfs_miplist_mutex
;
122 static struct ipmnt
*vfs_miplist
= NULL
;
123 static struct ipmnt
*vfs_miplist_end
= NULL
;
125 static kmem_cache_t
*vfs_cache
; /* Pointer to VFS kmem cache */
130 vnode_t
*rootdir
; /* pointer to root inode vnode. */
131 vnode_t
*devicesdir
; /* pointer to inode of devices root */
132 vnode_t
*devdir
; /* pointer to inode of dev root */
134 char *server_rootpath
; /* root path for diskless clients */
135 char *server_hostname
; /* hostname of diskless server */
137 static struct vfs root
;
138 static struct vfs devices
;
139 static struct vfs dev
;
140 struct vfs
*rootvfs
= &root
; /* pointer to root vfs; head of VFS list. */
141 rvfs_t
*rvfs_list
; /* array of vfs ptrs for vfs hash list */
142 int vfshsz
= 512; /* # of heads/locks in vfs hash arrays */
143 /* must be power of 2! */
144 timespec_t vfs_mnttab_ctime
; /* mnttab created time */
145 timespec_t vfs_mnttab_mtime
; /* mnttab last modified time */
146 char *vfs_dummyfstype
= "\0";
147 struct pollhead vfs_pollhd
; /* for mnttab pollers */
148 struct vnode
*vfs_mntdummyvp
; /* to fake mnttab read/write for file events */
149 int mntfstype
; /* will be set once mnt fs is mounted */
152 * Table for generic options recognized in the VFS layer and acted
153 * on at this level before parsing file system specific options.
154 * The nosuid option is stronger than any of the devices and setuid
155 * options, so those are canceled when nosuid is seen.
157 * All options which are added here need to be added to the
158 * list of standard options in usr/src/cmd/fs.d/fslib.c as well.
161 * VFS Mount options table
163 static char *ro_cancel
[] = { MNTOPT_RW
, NULL
};
164 static char *rw_cancel
[] = { MNTOPT_RO
, NULL
};
165 static char *suid_cancel
[] = { MNTOPT_NOSUID
, NULL
};
166 static char *nosuid_cancel
[] = { MNTOPT_SUID
, MNTOPT_DEVICES
, MNTOPT_NODEVICES
,
167 MNTOPT_NOSETUID
, MNTOPT_SETUID
, NULL
};
168 static char *devices_cancel
[] = { MNTOPT_NODEVICES
, NULL
};
169 static char *nodevices_cancel
[] = { MNTOPT_DEVICES
, NULL
};
170 static char *setuid_cancel
[] = { MNTOPT_NOSETUID
, NULL
};
171 static char *nosetuid_cancel
[] = { MNTOPT_SETUID
, NULL
};
172 static char *nbmand_cancel
[] = { MNTOPT_NONBMAND
, NULL
};
173 static char *nonbmand_cancel
[] = { MNTOPT_NBMAND
, NULL
};
174 static char *exec_cancel
[] = { MNTOPT_NOEXEC
, NULL
};
175 static char *noexec_cancel
[] = { MNTOPT_EXEC
, NULL
};
177 static const mntopt_t mntopts
[] = {
179 * option name cancel options default arg flags
181 { MNTOPT_REMOUNT
, NULL
, NULL
,
182 MO_NODISPLAY
, (void *)0 },
183 { MNTOPT_RO
, ro_cancel
, NULL
, 0,
185 { MNTOPT_RW
, rw_cancel
, NULL
, 0,
187 { MNTOPT_SUID
, suid_cancel
, NULL
, 0,
189 { MNTOPT_NOSUID
, nosuid_cancel
, NULL
, 0,
191 { MNTOPT_DEVICES
, devices_cancel
, NULL
, 0,
193 { MNTOPT_NODEVICES
, nodevices_cancel
, NULL
, 0,
195 { MNTOPT_SETUID
, setuid_cancel
, NULL
, 0,
197 { MNTOPT_NOSETUID
, nosetuid_cancel
, NULL
, 0,
199 { MNTOPT_NBMAND
, nbmand_cancel
, NULL
, 0,
201 { MNTOPT_NONBMAND
, nonbmand_cancel
, NULL
, 0,
203 { MNTOPT_EXEC
, exec_cancel
, NULL
, 0,
205 { MNTOPT_NOEXEC
, noexec_cancel
, NULL
, 0,
209 const mntopts_t vfs_mntopts
= {
210 sizeof (mntopts
) / sizeof (mntopt_t
),
211 (mntopt_t
*)&mntopts
[0]
215 * File system operation dispatch functions.
219 fsop_mount(vfs_t
*vfsp
, vnode_t
*mvp
, struct mounta
*uap
, cred_t
*cr
)
221 return (*(vfsp
)->vfs_op
->vfs_mount
)(vfsp
, mvp
, uap
, cr
);
225 fsop_unmount(vfs_t
*vfsp
, int flag
, cred_t
*cr
)
227 return (*(vfsp
)->vfs_op
->vfs_unmount
)(vfsp
, flag
, cr
);
231 fsop_root(vfs_t
*vfsp
, vnode_t
**vpp
)
234 int ret
= (*(vfsp
)->vfs_op
->vfs_root
)(vfsp
, vpp
);
236 * Make sure this root has a path. With lofs, it is possible to have
239 if (ret
== 0 && vfsp
->vfs_mntpt
!= NULL
&& (*vpp
)->v_path
== NULL
) {
240 mntpt
= vfs_getmntpoint(vfsp
);
241 vn_setpath_str(*vpp
, refstr_value(mntpt
),
242 strlen(refstr_value(mntpt
)));
250 fsop_statfs(vfs_t
*vfsp
, statvfs64_t
*sp
)
252 return (*(vfsp
)->vfs_op
->vfs_statvfs
)(vfsp
, sp
);
256 fsop_sync(vfs_t
*vfsp
, short flag
, cred_t
*cr
)
258 return (*(vfsp
)->vfs_op
->vfs_sync
)(vfsp
, flag
, cr
);
262 fsop_vget(vfs_t
*vfsp
, vnode_t
**vpp
, fid_t
*fidp
)
265 * In order to handle system attribute fids in a manner
266 * transparent to the underlying fs, we embed the fid for
267 * the sysattr parent object in the sysattr fid and tack on
268 * some extra bytes that only the sysattr layer knows about.
270 * This guarantees that sysattr fids are larger than other fids
271 * for this vfs. If the vfs supports the sysattr view interface
272 * (as indicated by VFSFT_SYSATTR_VIEWS), we cannot have a size
273 * collision with XATTR_FIDSZ.
275 if (vfs_has_feature(vfsp
, VFSFT_SYSATTR_VIEWS
) &&
276 fidp
->fid_len
== XATTR_FIDSZ
)
277 return (xattr_dir_vget(vfsp
, vpp
, fidp
));
279 return (*(vfsp
)->vfs_op
->vfs_vget
)(vfsp
, vpp
, fidp
);
283 fsop_mountroot(vfs_t
*vfsp
, enum whymountroot reason
)
285 return (*(vfsp
)->vfs_op
->vfs_mountroot
)(vfsp
, reason
);
289 fsop_freefs(vfs_t
*vfsp
)
291 (*(vfsp
)->vfs_op
->vfs_freevfs
)(vfsp
);
295 fsop_vnstate(vfs_t
*vfsp
, vnode_t
*vp
, vntrans_t nstate
)
297 return ((*(vfsp
)->vfs_op
->vfs_vnstate
)(vfsp
, vp
, nstate
));
301 fsop_sync_by_kind(int fstype
, short flag
, cred_t
*cr
)
303 ASSERT((fstype
>= 0) && (fstype
< nfstype
));
305 if (ALLOCATED_VFSSW(&vfssw
[fstype
]) && VFS_INSTALLED(&vfssw
[fstype
]))
306 return (*vfssw
[fstype
].vsw_vfsops
.vfs_sync
) (NULL
, flag
, cr
);
312 * File system initialization. vfs_setfsops() must be called from a file
313 * system's init routine.
317 fs_copyfsops(const fs_operation_def_t
*template, vfsops_t
*actual
,
320 static const fs_operation_trans_def_t vfs_ops_table
[] = {
321 VFSNAME_MOUNT
, offsetof(vfsops_t
, vfs_mount
),
324 VFSNAME_UNMOUNT
, offsetof(vfsops_t
, vfs_unmount
),
327 VFSNAME_ROOT
, offsetof(vfsops_t
, vfs_root
),
330 VFSNAME_STATVFS
, offsetof(vfsops_t
, vfs_statvfs
),
333 VFSNAME_SYNC
, offsetof(vfsops_t
, vfs_sync
),
334 (fs_generic_func_p
) fs_sync
,
335 (fs_generic_func_p
) fs_sync
, /* No errors allowed */
337 VFSNAME_VGET
, offsetof(vfsops_t
, vfs_vget
),
340 VFSNAME_MOUNTROOT
, offsetof(vfsops_t
, vfs_mountroot
),
343 VFSNAME_FREEVFS
, offsetof(vfsops_t
, vfs_freevfs
),
344 (fs_generic_func_p
)fs_freevfs
,
345 (fs_generic_func_p
)fs_freevfs
, /* Shouldn't fail */
347 VFSNAME_VNSTATE
, offsetof(vfsops_t
, vfs_vnstate
),
348 (fs_generic_func_p
)fs_nosys
,
349 (fs_generic_func_p
)fs_nosys
,
354 return (fs_build_vector(actual
, unused_ops
, vfs_ops_table
, template));
360 if (strcmp(rootfs
.bo_fstype
, MNTTYPE_ZFS
) == 0)
365 vfs_setfsops(int fstype
, const fs_operation_def_t
*template, vfsops_t
**actual
)
371 * Verify that fstype refers to a valid fs. Note that
372 * 0 is valid since it's used to set "stray" ops.
374 if ((fstype
< 0) || (fstype
>= nfstype
))
377 if (!ALLOCATED_VFSSW(&vfssw
[fstype
]))
380 /* Set up the operations vector. */
382 error
= fs_copyfsops(template, &vfssw
[fstype
].vsw_vfsops
, &unused_ops
);
387 vfssw
[fstype
].vsw_flag
|= VSW_INSTALLED
;
390 *actual
= &vfssw
[fstype
].vsw_vfsops
;
394 cmn_err(CE_WARN
, "vfs_setfsops: %s: %d operations supplied "
395 "but not used", vfssw
[fstype
].vsw_name
, unused_ops
);
402 vfs_makefsops(const fs_operation_def_t
*template, vfsops_t
**actual
)
407 *actual
= (vfsops_t
*)kmem_alloc(sizeof (vfsops_t
), KM_SLEEP
);
409 error
= fs_copyfsops(template, *actual
, &unused_ops
);
411 kmem_free(*actual
, sizeof (vfsops_t
));
420 * Free a vfsops structure created as a result of vfs_makefsops().
421 * NOTE: For a vfsops structure initialized by vfs_setfsops(), use
422 * vfs_freevfsops_by_type().
425 vfs_freevfsops(vfsops_t
*vfsops
)
427 kmem_free(vfsops
, sizeof (vfsops_t
));
431 * Since the vfsops structure is part of the vfssw table and wasn't
432 * really allocated, we're not really freeing anything. We keep
433 * the name for consistency with vfs_freevfsops(). We do, however,
434 * need to take care of a little bookkeeping.
435 * NOTE: For a vfsops structure created by vfs_setfsops(), use
436 * vfs_freevfsops_by_type().
439 vfs_freevfsops_by_type(int fstype
)
442 /* Verify that fstype refers to a loaded fs (and not fsid 0). */
443 if ((fstype
<= 0) || (fstype
>= nfstype
))
447 if ((vfssw
[fstype
].vsw_flag
& VSW_INSTALLED
) == 0) {
452 vfssw
[fstype
].vsw_flag
&= ~VSW_INSTALLED
;
458 /* Support routines used to reference vfs_op */
460 /* Set the operations vector for a vfs */
462 vfs_setops(vfs_t
*vfsp
, vfsops_t
*vfsops
)
466 ASSERT(vfsp
!= NULL
);
467 ASSERT(vfsops
!= NULL
);
471 if (vfsp
->vfs_femhead
== NULL
&&
472 atomic_cas_ptr(&vfsp
->vfs_op
, op
, vfsops
) == op
) {
475 fsem_setvfsops(vfsp
, vfsops
);
478 /* Retrieve the operations vector for a vfs */
480 vfs_getops(vfs_t
*vfsp
)
484 ASSERT(vfsp
!= NULL
);
488 if (vfsp
->vfs_femhead
== NULL
&& op
== vfsp
->vfs_op
) {
491 return (fsem_getvfsops(vfsp
));
496 * Returns non-zero (1) if the vfsops matches that of the vfs.
497 * Returns zero (0) if not.
500 vfs_matchops(vfs_t
*vfsp
, vfsops_t
*vfsops
)
502 return (vfs_getops(vfsp
) == vfsops
);
506 * Returns non-zero (1) if the file system has installed a non-default,
507 * non-error vfs_sync routine. Returns zero (0) otherwise.
510 vfs_can_sync(vfs_t
*vfsp
)
512 /* vfs_sync() routine is not the default/error function */
513 return (vfs_getops(vfsp
)->vfs_sync
!= fs_sync
);
517 * Initialize a vfs structure.
520 vfs_init(vfs_t
*vfsp
, vfsops_t
*op
, void *data
)
522 /* Other initialization has been moved to vfs_alloc() */
524 vfsp
->vfs_next
= vfsp
;
525 vfsp
->vfs_prev
= vfsp
;
526 vfsp
->vfs_zone_next
= vfsp
;
527 vfsp
->vfs_zone_prev
= vfsp
;
528 vfsp
->vfs_lofi_minor
= 0;
529 sema_init(&vfsp
->vfs_reflock
, 1, NULL
, SEMA_DEFAULT
, NULL
);
531 vfsp
->vfs_data
= (data
);
532 vfs_setops((vfsp
), (op
));
536 * Allocate and initialize the vfs implementation private data
537 * structure, vfs_impl_t.
540 vfsimpl_setup(vfs_t
*vfsp
)
544 if (vfsp
->vfs_implp
!= NULL
) {
548 vfsp
->vfs_implp
= kmem_alloc(sizeof (vfs_impl_t
), KM_SLEEP
);
549 /* Note that these are #define'd in vfs.h */
550 vfsp
->vfs_vskap
= NULL
;
551 vfsp
->vfs_fstypevsp
= NULL
;
553 /* Set size of counted array, then zero the array */
554 vfsp
->vfs_featureset
[0] = VFS_FEATURE_MAXSZ
- 1;
555 for (i
= 1; i
< VFS_FEATURE_MAXSZ
; i
++) {
556 vfsp
->vfs_featureset
[i
] = 0;
561 * Release the vfs_impl_t structure, if it exists. Some unbundled
562 * filesystems may not use the newer version of vfs and thus
563 * would not contain this implementation private data structure.
566 vfsimpl_teardown(vfs_t
*vfsp
)
568 vfs_impl_t
*vip
= vfsp
->vfs_implp
;
573 kmem_free(vfsp
->vfs_implp
, sizeof (vfs_impl_t
));
574 vfsp
->vfs_implp
= NULL
;
578 * VFS system calls: mount, umount, syssync, statfs, fstatfs, statvfs,
579 * fstatvfs, and sysfs moved to common/syscall.
583 * Update every mounted file system. We call the vfs_sync operation of
584 * each file system type, passing it a NULL vfsp to indicate that all
585 * mounted file systems of that type should be updated.
592 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
593 if (ALLOCATED_VFSSW(vswp
) && VFS_INSTALLED(vswp
)) {
596 (void) (*vswp
->vsw_vfsops
.vfs_sync
)(NULL
, flag
,
598 vfs_unrefvfssw(vswp
);
615 krwlock_t vfssw_lock
; /* lock accesses to vfssw */
618 * Lock for accessing the vfs linked list. Initialized in vfs_mountroot(),
619 * but otherwise should be accessed only via vfs_list_lock() and
620 * vfs_list_unlock(). Also used to protect the timestamp for mods to the list.
622 static krwlock_t vfslist
;
625 * Mount devfs on /devices. This is done right after root is mounted
626 * to provide device access support for the system
629 vfs_mountdevices(void)
633 struct mounta mounta
= { /* fake mounta for devfs_mount() */
645 * _init devfs module to fill in the vfssw
647 if (modload("fs", "devfs") == -1)
648 panic("Cannot _init devfs module");
654 vsw
= vfs_getvfsswbyname("devfs");
655 VFS_INIT(&devices
, &vsw
->vsw_vfsops
, NULL
);
661 if (lookupname("/devices", UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
))
662 panic("Cannot find /devices");
665 * Perform the mount of /devices
667 if (VFS_MOUNT(&devices
, mvp
, &mounta
, CRED()))
668 panic("Cannot mount /devices");
673 * Set appropriate members and add to vfs list for mnttab display
675 vfs_setresource(&devices
, "/devices", 0);
676 vfs_setmntpoint(&devices
, "/devices", 0);
679 * Hold the root of /devices so it won't go away
681 if (VFS_ROOT(&devices
, &devicesdir
))
682 panic("vfs_mountdevices: not devices root");
684 if (vfs_lock(&devices
) != 0) {
686 cmn_err(CE_NOTE
, "Cannot acquire vfs_lock of /devices");
690 if (vn_vfswlock(mvp
) != 0) {
691 vfs_unlock(&devices
);
693 cmn_err(CE_NOTE
, "Cannot acquire vfswlock of /devices");
697 vfs_add(mvp
, &devices
, 0);
699 vfs_unlock(&devices
);
704 * mount the first instance of /dev to root and remain mounted
711 struct mounta mounta
= { /* fake mounta for sdev_mount() */
714 MS_SYSSPACE
| MS_OVERLAY
,
723 * _init dev module to fill in the vfssw
725 if (modload("fs", "dev") == -1)
726 cmn_err(CE_PANIC
, "Cannot _init dev module\n");
732 vsw
= vfs_getvfsswbyname("dev");
733 VFS_INIT(&dev
, &vsw
->vsw_vfsops
, NULL
);
739 if (lookupname("/dev", UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
))
740 cmn_err(CE_PANIC
, "Cannot find /dev\n");
743 * Perform the mount of /dev
745 if (VFS_MOUNT(&dev
, mvp
, &mounta
, CRED()))
746 cmn_err(CE_PANIC
, "Cannot mount /dev 1\n");
751 * Set appropriate members and add to vfs list for mnttab display
753 vfs_setresource(&dev
, "/dev", 0);
754 vfs_setmntpoint(&dev
, "/dev", 0);
757 * Hold the root of /dev so it won't go away
759 if (VFS_ROOT(&dev
, &devdir
))
760 cmn_err(CE_PANIC
, "vfs_mountdev1: not dev root");
762 if (vfs_lock(&dev
) != 0) {
764 cmn_err(CE_NOTE
, "Cannot acquire vfs_lock of /dev");
768 if (vn_vfswlock(mvp
) != 0) {
771 cmn_err(CE_NOTE
, "Cannot acquire vfswlock of /dev");
775 vfs_add(mvp
, &dev
, 0);
782 * Mount required filesystem. This is done right after root is mounted.
785 vfs_mountfs(char *module
, char *spec
, char *path
)
788 struct mounta mounta
;
791 mounta
.flags
= MS_SYSSPACE
| MS_DATA
;
792 mounta
.fstype
= module
;
795 if (lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
)) {
796 cmn_err(CE_WARN
, "Cannot find %s", path
);
799 if (domount(NULL
, &mounta
, mvp
, CRED(), &vfsp
))
800 cmn_err(CE_WARN
, "Cannot mount %s", path
);
807 * vfs_mountroot is called by main() to mount the root filesystem.
812 struct vnode
*rvp
= NULL
;
818 rw_init(&vfssw_lock
, NULL
, RW_DEFAULT
, NULL
);
819 rw_init(&vfslist
, NULL
, RW_DEFAULT
, NULL
);
822 * Alloc the vfs hash bucket array and locks
824 rvfs_list
= kmem_zalloc(vfshsz
* sizeof (rvfs_t
), KM_SLEEP
);
827 * Call machine-dependent routine "rootconf" to choose a root
831 panic("vfs_mountroot: cannot mount root");
833 * Get vnode for '/'. Set up rootdir, u.u_rdir and u.u_cdir
834 * to point to it. These are used by lookuppn() so that it
835 * knows where to start from ('/' or '.').
837 vfs_setmntpoint(rootvfs
, "/", 0);
838 if (VFS_ROOT(rootvfs
, &rootdir
))
839 panic("vfs_mountroot: no root vnode");
842 * At this point, the process tree consists of p0 and possibly some
843 * direct children of p0. (i.e. there are no grandchildren)
845 * Walk through them all, setting their current directory.
847 mutex_enter(&pidlock
);
848 for (p
= practive
; p
!= NULL
; p
= p
->p_next
) {
849 ASSERT(p
== &p0
|| p
->p_parent
== &p0
);
851 PTOU(p
)->u_cdir
= rootdir
;
852 VN_HOLD(PTOU(p
)->u_cdir
);
853 PTOU(p
)->u_rdir
= NULL
;
855 mutex_exit(&pidlock
);
858 * Setup the global zone's rootvp, now that it exists.
860 global_zone
->zone_rootvp
= rootdir
;
861 VN_HOLD(global_zone
->zone_rootvp
);
864 * Notify the module code that it can begin using the
865 * root filesystem instead of the boot program's services.
870 * Special handling for a ZFS root file system.
875 * Set up mnttab information for root
877 vfs_setresource(rootvfs
, rootfs
.bo_name
, 0);
880 * Notify cluster software that the root filesystem is available.
884 /* Now that we're all done with the root FS, set up its vopstats */
885 if ((vswp
= vfs_getvfsswbyvfsops(vfs_getops(rootvfs
))) != NULL
) {
886 /* Set flag for statistics collection */
887 if (vswp
->vsw_flag
& VSW_STATS
) {
888 initialize_vopstats(&rootvfs
->vfs_vopstats
);
889 rootvfs
->vfs_flag
|= VFS_STATS
;
890 rootvfs
->vfs_fstypevsp
=
891 get_fstype_vopstats(rootvfs
, vswp
);
892 rootvfs
->vfs_vskap
= get_vskstat_anchor(rootvfs
);
894 vfs_unrefvfssw(vswp
);
898 * Mount /devices, /dev instance 1, /system/contract, /etc/mnttab,
899 * /etc/svc/volatile, /etc/dfs/sharetab, /system/object, and /proc.
904 vfs_mountfs("ctfs", "ctfs", CTFS_ROOT
);
905 vfs_mountfs("proc", "/proc", "/proc");
906 vfs_mountfs("mntfs", "/etc/mnttab", "/etc/mnttab");
907 vfs_mountfs("tmpfs", "/etc/svc/volatile", "/etc/svc/volatile");
908 vfs_mountfs("objfs", "objfs", OBJFS_ROOT
);
910 if (getzoneid() == GLOBAL_ZONEID
) {
911 vfs_mountfs("sharefs", "sharefs", "/etc/dfs/sharetab");
916 * This bit of magic can go away when we convert sparc to
917 * the new boot architecture based on ramdisk.
919 * Booting off a mirrored root volume:
920 * At this point, we have booted and mounted root on a
921 * single component of the mirror. Complete the boot
922 * by configuring SVM and converting the root to the
923 * dev_t of the mirrored root device. This dev_t conversion
924 * only works because the underlying device doesn't change.
927 if (svm_rootconf()) {
928 panic("vfs_mountroot: cannot remount root");
932 * mnttab should reflect the new root device
934 vfs_lock_wait(rootvfs
);
935 vfs_setresource(rootvfs
, rootfs
.bo_name
, 0);
940 if (strcmp(rootfs
.bo_fstype
, "zfs") != 0) {
942 * Look up the root device via devfs so that a dv_node is
943 * created for it. The vnode is never VN_RELE()ed.
944 * We allocate more than MAXPATHLEN so that the
945 * buffer passed to i_ddi_prompath_to_devfspath() is
946 * exactly MAXPATHLEN (the function expects a buffer
949 plen
= strlen("/devices");
950 path
= kmem_alloc(plen
+ MAXPATHLEN
, KM_SLEEP
);
951 (void) strcpy(path
, "/devices");
953 if (i_ddi_prompath_to_devfspath(rootfs
.bo_name
, path
+ plen
)
955 lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &rvp
)) {
957 /* NUL terminate in case "path" has garbage */
958 path
[plen
+ MAXPATHLEN
- 1] = '\0';
960 cmn_err(CE_WARN
, "!Cannot lookup root device: %s",
964 kmem_free(path
, plen
+ MAXPATHLEN
);
967 vfs_mnttabvp_setup();
971 * Check to see if our "block device" is actually a file. If so,
972 * automatically add a lofi device, and keep track of this fact.
975 lofi_add(const char *fsname
, struct vfs
*vfsp
,
976 mntopts_t
*mntopts
, struct mounta
*uap
)
978 int fromspace
= (uap
->flags
& MS_SYSSPACE
) ?
979 UIO_SYSSPACE
: UIO_USERSPACE
;
980 struct lofi_ioctl
*li
= NULL
;
981 struct vnode
*vp
= NULL
;
982 struct pathname pn
= { NULL
};
984 ldi_handle_t ldi_hdl
;
989 if ((vfssw
= vfs_getvfssw(fsname
)) == NULL
)
992 if (!(vfssw
->vsw_flag
& VSW_CANLOFI
)) {
993 vfs_unrefvfssw(vfssw
);
997 vfs_unrefvfssw(vfssw
);
1000 if (pn_get(uap
->spec
, fromspace
, &pn
) != 0)
1003 if (lookupname(uap
->spec
, fromspace
, FOLLOW
, NULL
, &vp
) != 0)
1006 if (vp
->v_type
!= VREG
)
1009 /* OK, this is a lofi mount. */
1011 if ((uap
->flags
& (MS_REMOUNT
|MS_GLOBAL
)) ||
1012 vfs_optionisset_nolock(mntopts
, MNTOPT_SUID
, NULL
) ||
1013 vfs_optionisset_nolock(mntopts
, MNTOPT_SETUID
, NULL
) ||
1014 vfs_optionisset_nolock(mntopts
, MNTOPT_DEVICES
, NULL
)) {
1019 ldi_id
= ldi_ident_from_anon();
1020 li
= kmem_zalloc(sizeof (*li
), KM_SLEEP
);
1021 (void) strlcpy(li
->li_filename
, pn
.pn_path
, MAXPATHLEN
);
1023 err
= ldi_open_by_name("/dev/lofictl", FREAD
| FWRITE
, kcred
,
1029 err
= ldi_ioctl(ldi_hdl
, LOFI_MAP_FILE
, (intptr_t)li
,
1030 FREAD
| FWRITE
| FKIOCTL
, kcred
, &minor
);
1032 (void) ldi_close(ldi_hdl
, FREAD
| FWRITE
, kcred
);
1035 vfsp
->vfs_lofi_minor
= minor
;
1038 ldi_ident_release(ldi_id
);
1041 kmem_free(li
, sizeof (*li
));
1049 lofi_remove(struct vfs
*vfsp
)
1051 struct lofi_ioctl
*li
= NULL
;
1053 ldi_handle_t ldi_hdl
;
1056 if (vfsp
->vfs_lofi_minor
== 0)
1059 ldi_id
= ldi_ident_from_anon();
1061 li
= kmem_zalloc(sizeof (*li
), KM_SLEEP
);
1062 li
->li_minor
= vfsp
->vfs_lofi_minor
;
1063 li
->li_cleanup
= B_TRUE
;
1065 err
= ldi_open_by_name("/dev/lofictl", FREAD
| FWRITE
, kcred
,
1071 err
= ldi_ioctl(ldi_hdl
, LOFI_UNMAP_FILE_MINOR
, (intptr_t)li
,
1072 FREAD
| FWRITE
| FKIOCTL
, kcred
, NULL
);
1074 (void) ldi_close(ldi_hdl
, FREAD
| FWRITE
, kcred
);
1077 vfsp
->vfs_lofi_minor
= 0;
1080 ldi_ident_release(ldi_id
);
1082 kmem_free(li
, sizeof (*li
));
1086 * Common mount code. Called from the system call entry point, from autofs,
1087 * nfsv4 trigger mounts, and from pxfs.
1089 * Takes the effective file system type, mount arguments, the mount point
1090 * vnode, flags specifying whether the mount is a remount and whether it
1091 * should be entered into the vfs list, and credentials. Fills in its vfspp
1092 * parameter with the mounted file system instance's vfs.
1094 * Note that the effective file system type is specified as a string. It may
1095 * be null, in which case it's determined from the mount arguments, and may
1096 * differ from the type specified in the mount arguments; this is a hook to
1097 * allow interposition when instantiating file system instances.
1099 * The caller is responsible for releasing its own hold on the mount point
1100 * vp (this routine does its own hold when necessary).
1101 * Also note that for remounts, the mount point vp should be the vnode for
1102 * the root of the file system rather than the vnode that the file system
1103 * is mounted on top of.
1106 domount(char *fsname
, struct mounta
*uap
, vnode_t
*vp
, struct cred
*credp
,
1114 mntopts_t mnt_mntopts
;
1116 int copyout_error
= 0;
1118 char *opts
= uap
->optptr
;
1119 char *inargs
= opts
;
1120 int optlen
= uap
->optlen
;
1126 int splice
= ((uap
->flags
& MS_NOSPLICE
) == 0);
1127 int fromspace
= (uap
->flags
& MS_SYSSPACE
) ?
1128 UIO_SYSSPACE
: UIO_USERSPACE
;
1129 char *resource
= NULL
, *mountpt
= NULL
;
1130 refstr_t
*oldresource
, *oldmntpt
;
1131 struct pathname pn
, rpn
;
1132 vsk_anchor_t
*vskap
;
1133 char fstname
[FSTYPSZ
];
1137 * The v_flag value for the mount point vp is permanently set
1138 * to VVFSLOCK so that no one bypasses the vn_vfs*locks routine
1139 * for mount point locking.
1141 mutex_enter(&vp
->v_lock
);
1142 vp
->v_flag
|= VVFSLOCK
;
1143 mutex_exit(&vp
->v_lock
);
1145 mnt_mntopts
.mo_count
= 0;
1147 * Find the ops vector to use to invoke the file system-specific mount
1148 * method. If the fsname argument is non-NULL, use it directly.
1149 * Otherwise, dig the file system type information out of the mount
1152 * A side effect is to hold the vfssw entry.
1154 * Mount arguments can be specified in several ways, which are
1155 * distinguished by flag bit settings. The preferred way is to set
1156 * MS_OPTIONSTR, indicating an 8 argument mount with the file system
1157 * type supplied as a character string and the last two arguments
1158 * being a pointer to a character buffer and the size of the buffer.
1159 * On entry, the buffer holds a null terminated list of options; on
1160 * return, the string is the list of options the file system
1161 * recognized. If MS_DATA is set arguments five and six point to a
1162 * block of binary data which the file system interprets.
1163 * A further wrinkle is that some callers don't set MS_FSS and MS_DATA
1164 * consistently with these conventions. To handle them, we check to
1165 * see whether the pointer to the file system name has a numeric value
1166 * less than 256. If so, we treat it as an index.
1168 if (fsname
!= NULL
) {
1169 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
) {
1172 } else if (uap
->flags
& (MS_OPTIONSTR
| MS_DATA
| MS_FSS
)) {
1178 if ((fstype
= (uintptr_t)uap
->fstype
) < 256) {
1180 if (fstype
== 0 || fstype
>= nfstype
||
1181 !ALLOCATED_VFSSW(&vfssw
[fstype
])) {
1185 (void) strcpy(fsname
, vfssw
[fstype
].vsw_name
);
1187 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
)
1191 * Handle either kernel or user address space.
1193 if (uap
->flags
& MS_SYSSPACE
) {
1194 error
= copystr(uap
->fstype
, fsname
,
1197 error
= copyinstr(uap
->fstype
, fsname
,
1201 if (error
== ENAMETOOLONG
)
1205 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
)
1209 if ((vswp
= vfs_getvfsswbyvfsops(vfs_getops(rootvfs
))) == NULL
)
1211 fsname
= vswp
->vsw_name
;
1213 if (!VFS_INSTALLED(vswp
))
1216 if ((error
= secpolicy_fs_allowed_mount(fsname
)) != 0) {
1217 vfs_unrefvfssw(vswp
);
1221 vfsops
= &vswp
->vsw_vfsops
;
1223 vfs_copyopttbl(&vswp
->vsw_optproto
, &mnt_mntopts
);
1225 * Fetch mount options and parse them for generic vfs options
1227 if (uap
->flags
& MS_OPTIONSTR
) {
1229 * Limit the buffer size
1231 if (optlen
< 0 || optlen
> MAX_MNTOPT_STR
) {
1235 if ((uap
->flags
& MS_SYSSPACE
) == 0) {
1236 inargs
= kmem_alloc(MAX_MNTOPT_STR
, KM_SLEEP
);
1239 error
= copyinstr(opts
, inargs
, (size_t)optlen
,
1246 vfs_parsemntopts(&mnt_mntopts
, inargs
, 0);
1249 * Flag bits override the options string.
1251 if (uap
->flags
& MS_REMOUNT
)
1252 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_REMOUNT
, NULL
, 0, 0);
1253 if (uap
->flags
& MS_RDONLY
)
1254 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_RO
, NULL
, 0, 0);
1255 if (uap
->flags
& MS_NOSUID
)
1256 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
, 0, 0);
1259 * Check if this is a remount; must be set in the option string and
1260 * the file system must support a remount option.
1262 if (remount
= vfs_optionisset_nolock(&mnt_mntopts
,
1263 MNTOPT_REMOUNT
, NULL
)) {
1264 if (!(vswp
->vsw_flag
& VSW_CANREMOUNT
)) {
1268 uap
->flags
|= MS_REMOUNT
;
1272 * uap->flags and vfs_optionisset() should agree.
1274 if (rdonly
= vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_RO
, NULL
)) {
1275 uap
->flags
|= MS_RDONLY
;
1277 if (vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
)) {
1278 uap
->flags
|= MS_NOSUID
;
1280 nbmand
= vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_NBMAND
, NULL
);
1281 ASSERT(splice
|| !remount
);
1283 * If we are splicing the fs into the namespace,
1284 * perform mount point checks.
1286 * We want to resolve the path for the mount point to eliminate
1287 * '.' and ".." and symlinks in mount points; we can't do the
1288 * same for the resource string, since it would turn
1289 * "/dev/dsk/c0t0d0s0" into "/devices/pci@...". We need to do
1290 * this before grabbing vn_vfswlock(), because otherwise we
1291 * would deadlock with lookuppn().
1294 ASSERT(vp
->v_count
> 0);
1297 * Pick up mount point and device from appropriate space.
1299 if (pn_get(uap
->spec
, fromspace
, &pn
) == 0) {
1300 resource
= kmem_alloc(pn
.pn_pathlen
+ 1,
1302 (void) strcpy(resource
, pn
.pn_path
);
1306 * Do a lookupname prior to taking the
1307 * writelock. Mark this as completed if
1308 * successful for later cleanup and addition to
1309 * the mount in progress table.
1311 if ((uap
->flags
& MS_GLOBAL
) == 0 &&
1312 lookupname(uap
->spec
, fromspace
,
1313 FOLLOW
, NULL
, &bvp
) == 0) {
1317 if ((error
= pn_get(uap
->dir
, fromspace
, &pn
)) == 0) {
1320 if (*pn
.pn_path
!= '/') {
1327 * Kludge to prevent autofs from deadlocking with
1328 * itself when it calls domount().
1330 * If autofs is calling, it is because it is doing
1331 * (autofs) mounts in the process of an NFS mount. A
1332 * lookuppn() here would cause us to block waiting for
1333 * said NFS mount to complete, which can't since this
1334 * is the thread that was supposed to doing it.
1336 if (fromspace
== UIO_USERSPACE
) {
1337 if ((error
= lookuppn(&pn
, &rpn
, FOLLOW
, NULL
,
1342 * The file disappeared or otherwise
1343 * became inaccessible since we opened
1344 * it; might as well fail the mount
1345 * since the mount point is no longer
1355 mountpt
= kmem_alloc(pnp
->pn_pathlen
+ 1, KM_SLEEP
);
1356 (void) strcpy(mountpt
, pnp
->pn_path
);
1359 * If the addition of the zone's rootpath
1360 * would push us over a total path length
1361 * of MAXPATHLEN, we fail the mount with
1362 * ENAMETOOLONG, which is what we would have
1363 * gotten if we were trying to perform the same
1364 * mount in the global zone.
1366 * strlen() doesn't count the trailing
1367 * '\0', but zone_rootpathlen counts both a
1368 * trailing '/' and the terminating '\0'.
1370 if ((curproc
->p_zone
->zone_rootpathlen
- 1 +
1371 strlen(mountpt
)) > MAXPATHLEN
||
1372 (resource
!= NULL
&&
1373 (curproc
->p_zone
->zone_rootpathlen
- 1 +
1374 strlen(resource
)) > MAXPATHLEN
)) {
1375 error
= ENAMETOOLONG
;
1386 * Prevent path name resolution from proceeding past
1389 if (vn_vfswlock(vp
) != 0) {
1395 * Verify that it's legitimate to establish a mount on
1396 * the prospective mount point.
1398 if (vn_mountedvfs(vp
) != NULL
) {
1400 * The mount point lock was obtained after some
1401 * other thread raced through and established a mount.
1407 if (vp
->v_flag
& VNOMOUNT
) {
1413 if ((uap
->flags
& (MS_DATA
| MS_OPTIONSTR
)) == 0) {
1414 uap
->dataptr
= NULL
;
1419 * If this is a remount, we don't want to create a new VFS.
1420 * Instead, we pass the existing one with a remount flag.
1424 * Confirm that the mount point is the root vnode of the
1425 * file system that is being remounted.
1426 * This can happen if the user specifies a different
1427 * mount point directory pathname in the (re)mount command.
1429 * Code below can only be reached if splice is true, so it's
1430 * safe to do vn_vfsunlock() here.
1432 if ((vp
->v_flag
& VROOT
) == 0) {
1438 * Disallow making file systems read-only unless file system
1439 * explicitly allows it in its vfssw. Ignore other flags.
1441 if (rdonly
&& vn_is_readonly(vp
) == 0 &&
1442 (vswp
->vsw_flag
& VSW_CANRWRO
) == 0) {
1448 * Disallow changing the NBMAND disposition of the file
1449 * system on remounts.
1451 if ((nbmand
&& ((vp
->v_vfsp
->vfs_flag
& VFS_NBMAND
) == 0)) ||
1452 (!nbmand
&& (vp
->v_vfsp
->vfs_flag
& VFS_NBMAND
))) {
1458 ovflags
= vfsp
->vfs_flag
;
1459 vfsp
->vfs_flag
|= VFS_REMOUNT
;
1460 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
1462 vfsp
= vfs_alloc(KM_SLEEP
);
1463 VFS_INIT(vfsp
, vfsops
, NULL
);
1468 if ((error
= lofi_add(fsname
, vfsp
, &mnt_mntopts
, uap
)) != 0) {
1481 * PRIV_SYS_MOUNT doesn't mean you can become root.
1483 if (vfsp
->vfs_lofi_minor
!= 0) {
1484 uap
->flags
|= MS_NOSUID
;
1485 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
, 0, 0);
1489 * The vfs_reflock is not used anymore the code below explicitly
1490 * holds it preventing others accesing it directly.
1492 if ((sema_tryp(&vfsp
->vfs_reflock
) == 0) &&
1493 !(vfsp
->vfs_flag
& VFS_REMOUNT
))
1495 "mount type %s couldn't get vfs_reflock", vswp
->vsw_name
);
1498 * Lock the vfs. If this is a remount we want to avoid spurious umount
1499 * failures that happen as a side-effect of fsflush() and other mount
1500 * and unmount operations that might be going on simultaneously and
1501 * may have locked the vfs currently. To not return EBUSY immediately
1502 * here we use vfs_lock_wait() instead vfs_lock() for the remount case.
1505 if (error
= vfs_lock(vfsp
)) {
1506 vfsp
->vfs_flag
= ovflags
;
1516 vfs_lock_wait(vfsp
);
1520 * Add device to mount in progress table, global mounts require special
1521 * handling. It is possible that we have already done the lookupname
1522 * on a spliced, non-global fs. If so, we don't want to do it again
1523 * since we cannot do a lookupname after taking the
1524 * wlock above. This case is for a non-spliced, non-global filesystem.
1527 if ((uap
->flags
& MS_GLOBAL
) == 0 &&
1528 lookupname(uap
->spec
, fromspace
, FOLLOW
, NULL
, &bvp
) == 0) {
1534 vnode_t
*lvp
= NULL
;
1536 error
= vfs_get_lofi(vfsp
, &lvp
);
1551 } else if (error
== -1) {
1560 vfs_addmip(bdev
, vfsp
);
1565 * Invalidate cached entry for the mount point.
1571 * If have an option string but the filesystem doesn't supply a
1572 * prototype options table, create a table with the global
1573 * options and sufficient room to accept all the options in the
1574 * string. Then parse the passed in option string
1575 * accepting all the options in the string. This gives us an
1576 * option table with all the proper cancel properties for the
1579 * Filesystems that supply a prototype options table are handled
1580 * earlier in this function.
1582 if (uap
->flags
& MS_OPTIONSTR
) {
1583 if (!(vswp
->vsw_flag
& VSW_HASPROTO
)) {
1584 mntopts_t tmp_mntopts
;
1586 tmp_mntopts
.mo_count
= 0;
1587 vfs_createopttbl_extend(&tmp_mntopts
, inargs
,
1589 vfs_parsemntopts(&tmp_mntopts
, inargs
, 1);
1590 vfs_swapopttbl_nolock(&mnt_mntopts
, &tmp_mntopts
);
1591 vfs_freeopttbl(&tmp_mntopts
);
1596 * Serialize with zone state transitions.
1597 * See vfs_list_add; zone mounted into is:
1598 * zone_find_by_path(refstr_value(vfsp->vfs_mntpt))
1599 * not the zone doing the mount (curproc->p_zone), but if we're already
1600 * inside a NGZ, then we know what zone we are.
1602 if (INGLOBALZONE(curproc
)) {
1603 zone
= zone_find_by_path(mountpt
);
1604 ASSERT(zone
!= NULL
);
1606 zone
= curproc
->p_zone
;
1608 * zone_find_by_path does a hold, so do one here too so that
1609 * we can do a zone_rele after mount_completed.
1613 mount_in_progress(zone
);
1615 * Instantiate (or reinstantiate) the file system. If appropriate,
1616 * splice it into the file system name space.
1618 * We want VFS_MOUNT() to be able to override the vfs_resource
1619 * string if necessary (ie, mntfs), and also for a remount to
1620 * change the same (necessary when remounting '/' during boot).
1621 * So we set up vfs_mntpt and vfs_resource to what we think they
1622 * should be, then hand off control to VFS_MOUNT() which can
1625 * For safety's sake, when changing vfs_resource or vfs_mntpt of
1626 * a vfs which is on the vfs list (i.e. during a remount), we must
1627 * never set those fields to NULL. Several bits of code make
1628 * assumptions that the fields are always valid.
1630 vfs_swapopttbl(&mnt_mntopts
, &vfsp
->vfs_mntopts
);
1632 if ((oldresource
= vfsp
->vfs_resource
) != NULL
)
1633 refstr_hold(oldresource
);
1634 if ((oldmntpt
= vfsp
->vfs_mntpt
) != NULL
)
1635 refstr_hold(oldmntpt
);
1637 vfs_setresource(vfsp
, resource
, 0);
1638 vfs_setmntpoint(vfsp
, mountpt
, 0);
1641 * going to mount on this vnode, so notify.
1643 vnevent_mountedover(vp
, NULL
);
1644 error
= VFS_MOUNT(vfsp
, vp
, uap
, credp
);
1646 if (uap
->flags
& MS_RDONLY
)
1647 vfs_setmntopt(vfsp
, MNTOPT_RO
, NULL
, 0);
1648 if (uap
->flags
& MS_NOSUID
)
1649 vfs_setmntopt(vfsp
, MNTOPT_NOSUID
, NULL
, 0);
1650 if (uap
->flags
& MS_GLOBAL
)
1651 vfs_setmntopt(vfsp
, MNTOPT_GLOBAL
, NULL
, 0);
1657 /* put back pre-remount options */
1658 vfs_swapopttbl(&mnt_mntopts
, &vfsp
->vfs_mntopts
);
1659 vfs_setmntpoint(vfsp
, refstr_value(oldmntpt
),
1662 refstr_rele(oldmntpt
);
1663 vfs_setresource(vfsp
, refstr_value(oldresource
),
1666 refstr_rele(oldresource
);
1667 vfsp
->vfs_flag
= ovflags
;
1672 vfs_freemnttab(vfsp
);
1677 * Set the mount time to now
1679 vfsp
->vfs_mtime
= ddi_get_time();
1681 vfsp
->vfs_flag
&= ~VFS_REMOUNT
;
1683 refstr_rele(oldresource
);
1685 refstr_rele(oldmntpt
);
1686 } else if (splice
) {
1688 * Link vfsp into the name space at the mount
1689 * point. Vfs_add() is responsible for
1690 * holding the mount point which will be
1691 * released when vfs_remove() is called.
1693 vfs_add(vp
, vfsp
, uap
->flags
);
1696 * Hold the reference to file system which is
1697 * not linked into the name space.
1699 vfsp
->vfs_zone
= NULL
;
1701 vfsp
->vfs_vnodecovered
= NULL
;
1704 * Set flags for global options encountered
1706 if (vfs_optionisset(vfsp
, MNTOPT_RO
, NULL
))
1707 vfsp
->vfs_flag
|= VFS_RDONLY
;
1709 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
1710 if (vfs_optionisset(vfsp
, MNTOPT_NOSUID
, NULL
)) {
1711 vfsp
->vfs_flag
|= (VFS_NOSETUID
|VFS_NODEVICES
);
1713 if (vfs_optionisset(vfsp
, MNTOPT_NODEVICES
, NULL
))
1714 vfsp
->vfs_flag
|= VFS_NODEVICES
;
1716 vfsp
->vfs_flag
&= ~VFS_NODEVICES
;
1717 if (vfs_optionisset(vfsp
, MNTOPT_NOSETUID
, NULL
))
1718 vfsp
->vfs_flag
|= VFS_NOSETUID
;
1720 vfsp
->vfs_flag
&= ~VFS_NOSETUID
;
1722 if (vfs_optionisset(vfsp
, MNTOPT_NBMAND
, NULL
))
1723 vfsp
->vfs_flag
|= VFS_NBMAND
;
1725 vfsp
->vfs_flag
&= ~VFS_NBMAND
;
1727 if (vfs_optionisset(vfsp
, MNTOPT_XATTR
, NULL
))
1728 vfsp
->vfs_flag
|= VFS_XATTR
;
1730 vfsp
->vfs_flag
&= ~VFS_XATTR
;
1732 if (vfs_optionisset(vfsp
, MNTOPT_NOEXEC
, NULL
))
1733 vfsp
->vfs_flag
|= VFS_NOEXEC
;
1735 vfsp
->vfs_flag
&= ~VFS_NOEXEC
;
1738 * Now construct the output option string of options
1741 if (uap
->flags
& MS_OPTIONSTR
) {
1742 vfs_list_read_lock();
1743 copyout_error
= vfs_buildoptionstr(
1744 &vfsp
->vfs_mntopts
, inargs
, optlen
);
1746 if (copyout_error
== 0 &&
1747 (uap
->flags
& MS_SYSSPACE
) == 0) {
1748 copyout_error
= copyoutstr(inargs
, opts
,
1754 * If this isn't a remount, set up the vopstats before
1755 * anyone can touch this. We only allow spliced file
1756 * systems (file systems which are in the namespace) to
1757 * have the VFS_STATS flag set.
1758 * NOTE: PxFS mounts the underlying file system with
1759 * MS_NOSPLICE set and copies those vfs_flags to its private
1760 * vfs structure. As a result, PxFS should never have
1761 * the VFS_STATS flag or else we might access the vfs
1762 * statistics-related fields prior to them being
1763 * properly initialized.
1765 if (!remount
&& (vswp
->vsw_flag
& VSW_STATS
) && splice
) {
1766 initialize_vopstats(&vfsp
->vfs_vopstats
);
1768 * We need to set vfs_vskap to NULL because there's
1769 * a chance it won't be set below. This is checked
1770 * in teardown_vopstats() so we can't have garbage.
1772 vfsp
->vfs_vskap
= NULL
;
1773 vfsp
->vfs_flag
|= VFS_STATS
;
1774 vfsp
->vfs_fstypevsp
= get_fstype_vopstats(vfsp
, vswp
);
1777 if (vswp
->vsw_flag
& VSW_XID
)
1778 vfsp
->vfs_flag
|= VFS_XID
;
1782 mount_completed(zone
);
1787 if ((error
== 0) && (copyout_error
== 0)) {
1790 * Don't call get_vskstat_anchor() while holding
1791 * locks since it allocates memory and calls
1792 * VFS_STATVFS(). For NFS, the latter can generate
1793 * an over-the-wire call.
1795 vskap
= get_vskstat_anchor(vfsp
);
1796 /* Only take the lock if we have something to do */
1797 if (vskap
!= NULL
) {
1798 vfs_lock_wait(vfsp
);
1799 if (vfsp
->vfs_flag
& VFS_STATS
) {
1800 vfsp
->vfs_vskap
= vskap
;
1805 /* Return vfsp to caller. */
1809 vfs_freeopttbl(&mnt_mntopts
);
1810 if (resource
!= NULL
)
1811 kmem_free(resource
, strlen(resource
) + 1);
1812 if (mountpt
!= NULL
)
1813 kmem_free(mountpt
, strlen(mountpt
) + 1);
1815 * It is possible we errored prior to adding to mount in progress
1816 * table. Must free vnode we acquired with successful lookupname.
1822 ASSERT(vswp
!= NULL
);
1823 vfs_unrefvfssw(vswp
);
1825 kmem_free(inargs
, MAX_MNTOPT_STR
);
1826 if (copyout_error
) {
1829 error
= copyout_error
;
1836 struct vfs
*vfsp
, /* vfs being updated */
1837 refstr_t
**refp
, /* Ref-count string to contain the new path */
1838 const char *newpath
, /* Path to add to refp (above) */
1839 uint32_t flag
) /* flag */
1843 zone_t
*zone
= curproc
->p_zone
;
1845 int have_list_lock
= 0;
1847 ASSERT(!VFS_ON_LIST(vfsp
) || vfs_lock_held(vfsp
));
1850 * New path must be less than MAXPATHLEN because mntfs
1851 * will only display up to MAXPATHLEN bytes. This is currently
1852 * safe, because domount() uses pn_get(), and other callers
1853 * similarly cap the size to fewer than MAXPATHLEN bytes.
1856 ASSERT(strlen(newpath
) < MAXPATHLEN
);
1858 /* mntfs requires consistency while vfs list lock is held */
1860 if (VFS_ON_LIST(vfsp
)) {
1869 * If we are in a non-global zone then we prefix the supplied path,
1870 * newpath, with the zone's root path, with two exceptions. The first
1871 * is where we have been explicitly directed to avoid doing so; this
1872 * will be the case following a failed remount, where the path supplied
1873 * will be a saved version which must now be restored. The second
1874 * exception is where newpath is not a pathname but a descriptive name,
1877 if (zone
== global_zone
|| (flag
& VFSSP_VERBATIM
) || *newpath
!= '/') {
1878 ref
= refstr_alloc(newpath
);
1883 * Truncate the trailing '/' in the zoneroot, and merge
1884 * in the zone's rootpath with the "newpath" (resource
1885 * or mountpoint) passed in.
1887 * The size of the required buffer is thus the size of
1888 * the buffer required for the passed-in newpath
1889 * (strlen(newpath) + 1), plus the size of the buffer
1890 * required to hold zone_rootpath (zone_rootpathlen)
1891 * minus one for one of the now-superfluous NUL
1892 * terminations, minus one for the trailing '/'.
1896 * (strlen(newpath) + 1) + zone_rootpathlen - 1 - 1
1898 * Which is what we have below.
1901 len
= strlen(newpath
) + zone
->zone_rootpathlen
- 1;
1902 sp
= kmem_alloc(len
, KM_SLEEP
);
1905 * Copy everything including the trailing slash, which
1906 * we then overwrite with the NUL character.
1909 (void) strcpy(sp
, zone
->zone_rootpath
);
1910 sp
[zone
->zone_rootpathlen
- 2] = '\0';
1911 (void) strcat(sp
, newpath
);
1913 ref
= refstr_alloc(sp
);
1918 if (have_list_lock
) {
1919 vfs_mnttab_modtimeupd();
1925 * Record a mounted resource name in a vfs structure.
1926 * If vfsp is already mounted, caller must hold the vfs lock.
1929 vfs_setresource(struct vfs
*vfsp
, const char *resource
, uint32_t flag
)
1931 if (resource
== NULL
|| resource
[0] == '\0')
1932 resource
= VFS_NORESOURCE
;
1933 vfs_setpath(vfsp
, &vfsp
->vfs_resource
, resource
, flag
);
1937 * Record a mount point name in a vfs structure.
1938 * If vfsp is already mounted, caller must hold the vfs lock.
1941 vfs_setmntpoint(struct vfs
*vfsp
, const char *mntpt
, uint32_t flag
)
1943 if (mntpt
== NULL
|| mntpt
[0] == '\0')
1944 mntpt
= VFS_NOMNTPT
;
1945 vfs_setpath(vfsp
, &vfsp
->vfs_mntpt
, mntpt
, flag
);
1948 /* Returns the vfs_resource. Caller must call refstr_rele() when finished. */
1951 vfs_getresource(const struct vfs
*vfsp
)
1955 vfs_list_read_lock();
1956 resource
= vfsp
->vfs_resource
;
1957 refstr_hold(resource
);
1963 /* Returns the vfs_mntpt. Caller must call refstr_rele() when finished. */
1966 vfs_getmntpoint(const struct vfs
*vfsp
)
1970 vfs_list_read_lock();
1971 mntpt
= vfsp
->vfs_mntpt
;
1979 * Create an empty options table with enough empty slots to hold all
1980 * The options in the options string passed as an argument.
1981 * Potentially prepend another options table.
1983 * Note: caller is responsible for locking the vfs list, if needed,
1987 vfs_createopttbl_extend(mntopts_t
*mops
, const char *opts
,
1988 const mntopts_t
*mtmpl
)
1990 const char *s
= opts
;
1993 if (opts
== NULL
|| *opts
== '\0') {
1999 * Count number of options in the string
2001 for (s
= strchr(s
, ','); s
!= NULL
; s
= strchr(s
, ',')) {
2006 vfs_copyopttbl_extend(mtmpl
, mops
, count
);
2010 * Create an empty options table with enough empty slots to hold all
2011 * The options in the options string passed as an argument.
2013 * This function is *not* for general use by filesystems.
2015 * Note: caller is responsible for locking the vfs list, if needed,
2019 vfs_createopttbl(mntopts_t
*mops
, const char *opts
)
2021 vfs_createopttbl_extend(mops
, opts
, NULL
);
2026 * Swap two mount options tables
2029 vfs_swapopttbl_nolock(mntopts_t
*optbl1
, mntopts_t
*optbl2
)
2034 tmpcnt
= optbl2
->mo_count
;
2035 tmplist
= optbl2
->mo_list
;
2036 optbl2
->mo_count
= optbl1
->mo_count
;
2037 optbl2
->mo_list
= optbl1
->mo_list
;
2038 optbl1
->mo_count
= tmpcnt
;
2039 optbl1
->mo_list
= tmplist
;
2043 vfs_swapopttbl(mntopts_t
*optbl1
, mntopts_t
*optbl2
)
2046 vfs_swapopttbl_nolock(optbl1
, optbl2
);
2047 vfs_mnttab_modtimeupd();
2052 vfs_copycancelopt_extend(char **const moc
, int extend
)
2059 for (; moc
[i
] != NULL
; i
++)
2060 /* count number of options to cancel */;
2063 if (i
+ extend
== 0)
2066 result
= kmem_alloc((i
+ extend
+ 1) * sizeof (char *), KM_SLEEP
);
2068 for (j
= 0; j
< i
; j
++) {
2069 result
[j
] = kmem_alloc(strlen(moc
[j
]) + 1, KM_SLEEP
);
2070 (void) strcpy(result
[j
], moc
[j
]);
2072 for (; j
<= i
+ extend
; j
++)
2079 vfs_copyopt(const mntopt_t
*s
, mntopt_t
*d
)
2083 d
->mo_flags
= s
->mo_flags
;
2084 d
->mo_data
= s
->mo_data
;
2087 dp
= kmem_alloc(strlen(sp
) + 1, KM_SLEEP
);
2088 (void) strcpy(dp
, sp
);
2091 d
->mo_name
= NULL
; /* should never happen */
2094 d
->mo_cancel
= vfs_copycancelopt_extend(s
->mo_cancel
, 0);
2098 dp
= kmem_alloc(strlen(sp
) + 1, KM_SLEEP
);
2099 (void) strcpy(dp
, sp
);
2107 * Copy a mount options table, possibly allocating some spare
2108 * slots at the end. It is permissible to copy_extend the NULL table.
2111 vfs_copyopttbl_extend(const mntopts_t
*smo
, mntopts_t
*dmo
, int extra
)
2117 * Clear out any existing stuff in the options table being initialized
2119 vfs_freeopttbl(dmo
);
2120 count
= (smo
== NULL
) ? 0 : smo
->mo_count
;
2121 if ((count
+ extra
) == 0) /* nothing to do */
2123 dmo
->mo_count
= count
+ extra
;
2124 motbl
= kmem_zalloc((count
+ extra
) * sizeof (mntopt_t
), KM_SLEEP
);
2125 dmo
->mo_list
= motbl
;
2126 for (i
= 0; i
< count
; i
++) {
2127 vfs_copyopt(&smo
->mo_list
[i
], &motbl
[i
]);
2129 for (i
= count
; i
< count
+ extra
; i
++) {
2130 motbl
[i
].mo_flags
= MO_EMPTY
;
2135 * Copy a mount options table.
2137 * This function is *not* for general use by filesystems.
2139 * Note: caller is responsible for locking the vfs list, if needed,
2140 * to protect smo and dmo.
2143 vfs_copyopttbl(const mntopts_t
*smo
, mntopts_t
*dmo
)
2145 vfs_copyopttbl_extend(smo
, dmo
, 0);
2149 vfs_mergecancelopts(const mntopt_t
*mop1
, const mntopt_t
*mop2
)
2154 char **sp1
, **sp2
, **dp
;
2157 * First we count both lists of cancel options.
2158 * If either is NULL or has no elements, we return a copy of
2161 if (mop1
->mo_cancel
!= NULL
) {
2162 for (; mop1
->mo_cancel
[c1
] != NULL
; c1
++)
2163 /* count cancel options in mop1 */;
2167 return (vfs_copycancelopt_extend(mop2
->mo_cancel
, 0));
2169 if (mop2
->mo_cancel
!= NULL
) {
2170 for (; mop2
->mo_cancel
[c2
] != NULL
; c2
++)
2171 /* count cancel options in mop2 */;
2174 result
= vfs_copycancelopt_extend(mop1
->mo_cancel
, c2
);
2180 * When we get here, we've got two sets of cancel options;
2181 * we need to merge the two sets. We know that the result
2182 * array has "c1+c2+1" entries and in the end we might shrink
2184 * Result now has a copy of the c1 entries from mop1; we'll
2185 * now lookup all the entries of mop2 in mop1 and copy it if
2187 * This operation is O(n^2) but it's only called once per
2188 * filesystem per duplicate option. This is a situation
2189 * which doesn't arise with the filesystems in ON and
2194 for (sp2
= mop2
->mo_cancel
; *sp2
!= NULL
; sp2
++) {
2195 for (sp1
= mop1
->mo_cancel
; *sp1
!= NULL
; sp1
++) {
2196 if (strcmp(*sp1
, *sp2
) == 0)
2201 * Option *sp2 not found in mop1, so copy it.
2202 * The calls to vfs_copycancelopt_extend()
2203 * guarantee that there's enough room.
2205 *dp
= kmem_alloc(strlen(*sp2
) + 1, KM_SLEEP
);
2206 (void) strcpy(*dp
++, *sp2
);
2209 if (dp
!= &result
[c1
+c2
]) {
2210 size_t bytes
= (dp
- result
+ 1) * sizeof (char *);
2211 char **nres
= kmem_alloc(bytes
, KM_SLEEP
);
2213 bcopy(result
, nres
, bytes
);
2214 kmem_free(result
, (c1
+ c2
+ 1) * sizeof (char *));
2221 * Merge two mount option tables (outer and inner) into one. This is very
2222 * similar to "merging" global variables and automatic variables in C.
2224 * This isn't (and doesn't have to be) fast.
2226 * This function is *not* for general use by filesystems.
2228 * Note: caller is responsible for locking the vfs list, if needed,
2229 * to protect omo, imo & dmo.
2232 vfs_mergeopttbl(const mntopts_t
*omo
, const mntopts_t
*imo
, mntopts_t
*dmo
)
2235 mntopt_t
*mop
, *motbl
;
2239 * First determine how much space we need to allocate.
2241 count
= omo
->mo_count
;
2242 for (i
= 0; i
< imo
->mo_count
; i
++) {
2243 if (imo
->mo_list
[i
].mo_flags
& MO_EMPTY
)
2245 if (vfs_hasopt(omo
, imo
->mo_list
[i
].mo_name
) == NULL
)
2248 ASSERT(count
>= omo
->mo_count
&&
2249 count
<= omo
->mo_count
+ imo
->mo_count
);
2250 motbl
= kmem_alloc(count
* sizeof (mntopt_t
), KM_SLEEP
);
2251 for (i
= 0; i
< omo
->mo_count
; i
++)
2252 vfs_copyopt(&omo
->mo_list
[i
], &motbl
[i
]);
2253 freeidx
= omo
->mo_count
;
2254 for (i
= 0; i
< imo
->mo_count
; i
++) {
2255 if (imo
->mo_list
[i
].mo_flags
& MO_EMPTY
)
2257 if ((mop
= vfs_hasopt(omo
, imo
->mo_list
[i
].mo_name
)) != NULL
) {
2259 uint_t index
= mop
- omo
->mo_list
;
2261 newcanp
= vfs_mergecancelopts(mop
, &motbl
[index
]);
2263 vfs_freeopt(&motbl
[index
]);
2264 vfs_copyopt(&imo
->mo_list
[i
], &motbl
[index
]);
2266 vfs_freecancelopt(motbl
[index
].mo_cancel
);
2267 motbl
[index
].mo_cancel
= newcanp
;
2270 * If it's a new option, just copy it over to the first
2273 vfs_copyopt(&imo
->mo_list
[i
], &motbl
[freeidx
++]);
2276 dmo
->mo_count
= count
;
2277 dmo
->mo_list
= motbl
;
2281 * Functions to set and clear mount options in a mount options table.
2285 * Clear a mount option, if it exists.
2287 * The update_mnttab arg indicates whether mops is part of a vfs that is on
2291 vfs_clearmntopt_nolock(mntopts_t
*mops
, const char *opt
, int update_mnttab
)
2296 ASSERT(!update_mnttab
|| RW_WRITE_HELD(&vfslist
));
2298 count
= mops
->mo_count
;
2299 for (i
= 0; i
< count
; i
++) {
2300 mop
= &mops
->mo_list
[i
];
2302 if (mop
->mo_flags
& MO_EMPTY
)
2304 if (strcmp(opt
, mop
->mo_name
))
2306 mop
->mo_flags
&= ~MO_SET
;
2307 if (mop
->mo_arg
!= NULL
) {
2308 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2312 vfs_mnttab_modtimeupd();
2318 vfs_clearmntopt(struct vfs
*vfsp
, const char *opt
)
2322 if (VFS_ON_LIST(vfsp
)) {
2326 vfs_clearmntopt_nolock(&vfsp
->vfs_mntopts
, opt
, gotlock
);
2333 * Set a mount option on. If it's not found in the table, it's silently
2334 * ignored. If the option has MO_IGNORE set, it is still set unless the
2335 * VFS_NOFORCEOPT bit is set in the flags. Also, VFS_DISPLAY/VFS_NODISPLAY flag
2336 * bits can be used to toggle the MO_NODISPLAY bit for the option.
2337 * If the VFS_CREATEOPT flag bit is set then the first option slot with
2338 * MO_EMPTY set is created as the option passed in.
2340 * The update_mnttab arg indicates whether mops is part of a vfs that is on
2344 vfs_setmntopt_nolock(mntopts_t
*mops
, const char *opt
,
2345 const char *arg
, int flags
, int update_mnttab
)
2351 ASSERT(!update_mnttab
|| RW_WRITE_HELD(&vfslist
));
2353 if (flags
& VFS_CREATEOPT
) {
2354 if (vfs_hasopt(mops
, opt
) != NULL
) {
2355 flags
&= ~VFS_CREATEOPT
;
2358 count
= mops
->mo_count
;
2359 for (i
= 0; i
< count
; i
++) {
2360 mop
= &mops
->mo_list
[i
];
2362 if (mop
->mo_flags
& MO_EMPTY
) {
2363 if ((flags
& VFS_CREATEOPT
) == 0)
2365 sp
= kmem_alloc(strlen(opt
) + 1, KM_SLEEP
);
2366 (void) strcpy(sp
, opt
);
2369 mop
->mo_flags
= MO_HASVALUE
;
2372 } else if (strcmp(opt
, mop
->mo_name
)) {
2375 if ((mop
->mo_flags
& MO_IGNORE
) && (flags
& VFS_NOFORCEOPT
))
2377 if (arg
!= NULL
&& (mop
->mo_flags
& MO_HASVALUE
) != 0) {
2378 sp
= kmem_alloc(strlen(arg
) + 1, KM_SLEEP
);
2379 (void) strcpy(sp
, arg
);
2383 if (mop
->mo_arg
!= NULL
)
2384 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2386 if (flags
& VFS_DISPLAY
)
2387 mop
->mo_flags
&= ~MO_NODISPLAY
;
2388 if (flags
& VFS_NODISPLAY
)
2389 mop
->mo_flags
|= MO_NODISPLAY
;
2390 mop
->mo_flags
|= MO_SET
;
2391 if (mop
->mo_cancel
!= NULL
) {
2394 for (cp
= mop
->mo_cancel
; *cp
!= NULL
; cp
++)
2395 vfs_clearmntopt_nolock(mops
, *cp
, 0);
2398 vfs_mnttab_modtimeupd();
2404 vfs_setmntopt(struct vfs
*vfsp
, const char *opt
, const char *arg
, int flags
)
2408 if (VFS_ON_LIST(vfsp
)) {
2412 vfs_setmntopt_nolock(&vfsp
->vfs_mntopts
, opt
, arg
, flags
, gotlock
);
2419 * Add a "tag" option to a mounted file system's options list.
2421 * Note: caller is responsible for locking the vfs list, if needed,
2425 vfs_addtag(mntopts_t
*mops
, const char *tag
)
2428 mntopt_t
*mop
, *motbl
;
2430 count
= mops
->mo_count
+ 1;
2431 motbl
= kmem_zalloc(count
* sizeof (mntopt_t
), KM_SLEEP
);
2432 if (mops
->mo_count
) {
2433 size_t len
= (count
- 1) * sizeof (mntopt_t
);
2435 bcopy(mops
->mo_list
, motbl
, len
);
2436 kmem_free(mops
->mo_list
, len
);
2438 mops
->mo_count
= count
;
2439 mops
->mo_list
= motbl
;
2440 mop
= &motbl
[count
- 1];
2441 mop
->mo_flags
= MO_TAG
;
2442 mop
->mo_name
= kmem_alloc(strlen(tag
) + 1, KM_SLEEP
);
2443 (void) strcpy(mop
->mo_name
, tag
);
2448 * Allow users to set arbitrary "tags" in a vfs's mount options.
2449 * Broader use within the kernel is discouraged.
2452 vfs_settag(uint_t major
, uint_t minor
, const char *mntpt
, const char *tag
,
2459 dev_t dev
= makedevice(major
, minor
);
2461 char *buf
= kmem_alloc(MAX_MNTOPT_STR
, KM_SLEEP
);
2464 * Find the desired mounted file system
2469 if (vfsp
->vfs_dev
== dev
&&
2470 strcmp(mntpt
, refstr_value(vfsp
->vfs_mntpt
)) == 0) {
2474 vfsp
= vfsp
->vfs_next
;
2475 } while (vfsp
!= rootvfs
);
2481 err
= secpolicy_fs_config(cr
, vfsp
);
2485 mops
= &vfsp
->vfs_mntopts
;
2487 * Add tag if it doesn't already exist
2489 if ((mop
= vfs_hasopt(mops
, tag
)) == NULL
) {
2492 (void) vfs_buildoptionstr(mops
, buf
, MAX_MNTOPT_STR
);
2494 if (len
+ strlen(tag
) + 2 > MAX_MNTOPT_STR
) {
2498 mop
= vfs_addtag(mops
, tag
);
2500 if ((mop
->mo_flags
& MO_TAG
) == 0) {
2504 vfs_setmntopt_nolock(mops
, tag
, NULL
, 0, 1);
2507 kmem_free(buf
, MAX_MNTOPT_STR
);
2512 * Allow users to remove arbitrary "tags" in a vfs's mount options.
2513 * Broader use within the kernel is discouraged.
2516 vfs_clrtag(uint_t major
, uint_t minor
, const char *mntpt
, const char *tag
,
2522 dev_t dev
= makedevice(major
, minor
);
2526 * Find the desired mounted file system
2531 if (vfsp
->vfs_dev
== dev
&&
2532 strcmp(mntpt
, refstr_value(vfsp
->vfs_mntpt
)) == 0) {
2536 vfsp
= vfsp
->vfs_next
;
2537 } while (vfsp
!= rootvfs
);
2543 err
= secpolicy_fs_config(cr
, vfsp
);
2547 if ((mop
= vfs_hasopt(&vfsp
->vfs_mntopts
, tag
)) == NULL
) {
2551 if ((mop
->mo_flags
& MO_TAG
) == 0) {
2555 vfs_clearmntopt_nolock(&vfsp
->vfs_mntopts
, tag
, 1);
2562 * Function to parse an option string and fill in a mount options table.
2563 * Unknown options are silently ignored. The input option string is modified
2564 * by replacing separators with nulls. If the create flag is set, options
2565 * not found in the table are just added on the fly. The table must have
2566 * an option slot marked MO_EMPTY to add an option on the fly.
2568 * This function is *not* for general use by filesystems.
2570 * Note: caller is responsible for locking the vfs list, if needed,
2574 vfs_parsemntopts(mntopts_t
*mops
, char *osp
, int create
)
2576 char *s
= osp
, *p
, *nextop
, *valp
, *cp
, *ep
;
2577 int setflg
= VFS_NOFORCEOPT
;
2581 while (*s
!= '\0') {
2582 p
= strchr(s
, ','); /* find next option */
2587 cp
= p
; /* save location of comma */
2588 *p
++ = '\0'; /* mark end and point to next option */
2591 p
= strchr(s
, '='); /* look for value */
2593 valp
= NULL
; /* no value supplied */
2595 ep
= p
; /* save location of equals */
2596 *p
++ = '\0'; /* end option and point to value */
2600 * set option into options table
2603 setflg
|= VFS_CREATEOPT
;
2604 vfs_setmntopt_nolock(mops
, s
, valp
, setflg
, 0);
2606 *cp
= ','; /* restore the comma */
2608 *ep
= '='; /* restore the equals */
2614 * Function to inquire if an option exists in a mount options table.
2615 * Returns a pointer to the option if it exists, else NULL.
2617 * This function is *not* for general use by filesystems.
2619 * Note: caller is responsible for locking the vfs list, if needed,
2623 vfs_hasopt(const mntopts_t
*mops
, const char *opt
)
2628 count
= mops
->mo_count
;
2629 for (i
= 0; i
< count
; i
++) {
2630 mop
= &mops
->mo_list
[i
];
2632 if (mop
->mo_flags
& MO_EMPTY
)
2634 if (strcmp(opt
, mop
->mo_name
) == 0)
2641 * Function to inquire if an option is set in a mount options table.
2642 * Returns non-zero if set and fills in the arg pointer with a pointer to
2643 * the argument string or NULL if there is no argument string.
2646 vfs_optionisset_nolock(const mntopts_t
*mops
, const char *opt
, char **argp
)
2651 count
= mops
->mo_count
;
2652 for (i
= 0; i
< count
; i
++) {
2653 mop
= &mops
->mo_list
[i
];
2655 if (mop
->mo_flags
& MO_EMPTY
)
2657 if (strcmp(opt
, mop
->mo_name
))
2659 if ((mop
->mo_flags
& MO_SET
) == 0)
2661 if (argp
!= NULL
&& (mop
->mo_flags
& MO_HASVALUE
) != 0)
2662 *argp
= mop
->mo_arg
;
2670 vfs_optionisset(const struct vfs
*vfsp
, const char *opt
, char **argp
)
2674 vfs_list_read_lock();
2675 ret
= vfs_optionisset_nolock(&vfsp
->vfs_mntopts
, opt
, argp
);
2682 * Construct a comma separated string of the options set in the given
2683 * mount table, return the string in the given buffer. Return non-zero if
2684 * the buffer would overflow.
2686 * This function is *not* for general use by filesystems.
2688 * Note: caller is responsible for locking the vfs list, if needed,
2692 vfs_buildoptionstr(const mntopts_t
*mp
, char *buf
, int len
)
2699 for (i
= 0; i
< mp
->mo_count
; i
++) {
2702 mop
= &mp
->mo_list
[i
];
2703 if (mop
->mo_flags
& MO_SET
) {
2704 int optlen
, comma
= 0;
2708 optlen
= strlen(mop
->mo_name
);
2709 if (strlen(buf
) + comma
+ optlen
+ 1 > len
)
2713 (void) strcpy(cp
, mop
->mo_name
);
2716 * Append option value if there is one
2718 if (mop
->mo_arg
!= NULL
) {
2721 arglen
= strlen(mop
->mo_arg
);
2722 if (strlen(buf
) + arglen
+ 2 > len
)
2725 (void) strcpy(cp
, mop
->mo_arg
);
2736 vfs_freecancelopt(char **moc
)
2742 for (cp
= moc
; *cp
!= NULL
; cp
++) {
2743 kmem_free(*cp
, strlen(*cp
) + 1);
2746 kmem_free(moc
, (ccnt
+ 1) * sizeof (char *));
2751 vfs_freeopt(mntopt_t
*mop
)
2753 if (mop
->mo_name
!= NULL
)
2754 kmem_free(mop
->mo_name
, strlen(mop
->mo_name
) + 1);
2756 vfs_freecancelopt(mop
->mo_cancel
);
2758 if (mop
->mo_arg
!= NULL
)
2759 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2763 * Free a mount options table
2765 * This function is *not* for general use by filesystems.
2767 * Note: caller is responsible for locking the vfs list, if needed,
2771 vfs_freeopttbl(mntopts_t
*mp
)
2775 count
= mp
->mo_count
;
2776 for (i
= 0; i
< count
; i
++) {
2777 vfs_freeopt(&mp
->mo_list
[i
]);
2780 kmem_free(mp
->mo_list
, sizeof (mntopt_t
) * count
);
2789 vfs_mntdummyread(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cred
,
2790 caller_context_t
*ct
)
2797 vfs_mntdummywrite(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cred
,
2798 caller_context_t
*ct
)
2804 * The dummy vnode is currently used only by file events notification
2805 * module which is just interested in the timestamps.
2809 vfs_mntdummygetattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2810 caller_context_t
*ct
)
2812 bzero(vap
, sizeof (vattr_t
));
2813 vap
->va_type
= VREG
;
2815 vap
->va_ctime
= vfs_mnttab_ctime
;
2817 * it is ok to just copy mtime as the time will be monotonically
2820 vap
->va_mtime
= vfs_mnttab_mtime
;
2821 vap
->va_atime
= vap
->va_mtime
;
2826 vfs_mnttabvp_setup(void)
2829 vnodeops_t
*vfs_mntdummyvnops
;
2830 const fs_operation_def_t mnt_dummyvnodeops_template
[] = {
2831 VOPNAME_READ
, { .vop_read
= vfs_mntdummyread
},
2832 VOPNAME_WRITE
, { .vop_write
= vfs_mntdummywrite
},
2833 VOPNAME_GETATTR
, { .vop_getattr
= vfs_mntdummygetattr
},
2834 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
2838 if (vn_make_ops("mnttab", mnt_dummyvnodeops_template
,
2839 &vfs_mntdummyvnops
) != 0) {
2840 cmn_err(CE_WARN
, "vfs_mnttabvp_setup: vn_make_ops failed");
2841 /* Shouldn't happen, but not bad enough to panic */
2846 * A global dummy vnode is allocated to represent mntfs files.
2847 * The mntfs file (/etc/mnttab) can be monitored for file events
2848 * and receive an event when mnttab changes. Dummy VOP calls
2849 * will be made on this vnode. The file events notification module
2850 * intercepts this vnode and delivers relevant events.
2852 tvp
= vn_alloc(KM_SLEEP
);
2853 tvp
->v_flag
= VNOMOUNT
|VNOMAP
|VNOSWAP
|VNOCACHE
;
2854 vn_setops(tvp
, vfs_mntdummyvnops
);
2857 * The mnt dummy ops do not reference v_data.
2858 * No other module intercepting this vnode should either.
2859 * Just set it to point to itself.
2861 tvp
->v_data
= (caddr_t
)tvp
;
2862 tvp
->v_vfsp
= rootvfs
;
2863 vfs_mntdummyvp
= tvp
;
2867 * performs fake read/write ops
2870 vfs_mnttab_rwop(int rw
)
2876 if (vfs_mntdummyvp
== NULL
)
2879 bzero(&uio
, sizeof (uio
));
2880 bzero(&iov
, sizeof (iov
));
2885 uio
.uio_loffset
= 0;
2886 uio
.uio_segflg
= UIO_SYSSPACE
;
2889 (void) VOP_WRITE(vfs_mntdummyvp
, &uio
, 0, kcred
, NULL
);
2891 (void) VOP_READ(vfs_mntdummyvp
, &uio
, 0, kcred
, NULL
);
2896 * Generate a write operation.
2899 vfs_mnttab_writeop(void)
2905 * Generate a read operation.
2908 vfs_mnttab_readop(void)
2914 * Free any mnttab information recorded in the vfs struct.
2915 * The vfs must not be on the vfs list.
2918 vfs_freemnttab(struct vfs
*vfsp
)
2920 ASSERT(!VFS_ON_LIST(vfsp
));
2923 * Free device and mount point information
2925 if (vfsp
->vfs_mntpt
!= NULL
) {
2926 refstr_rele(vfsp
->vfs_mntpt
);
2927 vfsp
->vfs_mntpt
= NULL
;
2929 if (vfsp
->vfs_resource
!= NULL
) {
2930 refstr_rele(vfsp
->vfs_resource
);
2931 vfsp
->vfs_resource
= NULL
;
2934 * Now free mount options information
2936 vfs_freeopttbl(&vfsp
->vfs_mntopts
);
2940 * Return the last mnttab modification time
2943 vfs_mnttab_modtime(timespec_t
*ts
)
2945 ASSERT(RW_LOCK_HELD(&vfslist
));
2946 *ts
= vfs_mnttab_mtime
;
2950 * See if mnttab is changed
2953 vfs_mnttab_poll(timespec_t
*old
, struct pollhead
**phpp
)
2957 *phpp
= (struct pollhead
*)NULL
;
2960 * Note: don't grab vfs list lock before accessing vfs_mnttab_mtime.
2961 * Can lead to deadlock against vfs_mnttab_modtimeupd(). It is safe
2962 * to not grab the vfs list lock because tv_sec is monotonically
2966 changed
= (old
->tv_nsec
!= vfs_mnttab_mtime
.tv_nsec
) ||
2967 (old
->tv_sec
!= vfs_mnttab_mtime
.tv_sec
);
2969 *phpp
= &vfs_pollhd
;
2973 /* Provide a unique and monotonically-increasing timestamp. */
2975 vfs_mono_time(timespec_t
*ts
)
2977 static volatile hrtime_t hrt
; /* The saved time. */
2978 hrtime_t newhrt
, oldhrt
; /* For effecting the CAS. */
2982 * Try gethrestime() first, but be prepared to fabricate a sensible
2983 * answer at the first sign of any trouble.
2985 gethrestime(&newts
);
2986 newhrt
= ts2hrt(&newts
);
2991 if (atomic_cas_64((uint64_t *)&hrt
, oldhrt
, newhrt
) == oldhrt
)
2998 * Update the mnttab modification time and wake up any waiters for
3002 vfs_mnttab_modtimeupd()
3004 hrtime_t oldhrt
, newhrt
;
3006 ASSERT(RW_WRITE_HELD(&vfslist
));
3007 oldhrt
= ts2hrt(&vfs_mnttab_mtime
);
3008 gethrestime(&vfs_mnttab_mtime
);
3009 newhrt
= ts2hrt(&vfs_mnttab_mtime
);
3010 if (oldhrt
== (hrtime_t
)0)
3011 vfs_mnttab_ctime
= vfs_mnttab_mtime
;
3013 * Attempt to provide unique mtime (like uniqtime but not).
3015 if (newhrt
== oldhrt
) {
3017 hrt2ts(newhrt
, &vfs_mnttab_mtime
);
3019 pollwakeup(&vfs_pollhd
, (short)POLLRDBAND
);
3020 vfs_mnttab_writeop();
3024 dounmount(struct vfs
*vfsp
, int flag
, cred_t
*cr
)
3028 extern void teardown_vopstats(vfs_t
*);
3031 * Get covered vnode. This will be NULL if the vfs is not linked
3032 * into the file system name space (i.e., domount() with MNT_NOSPICE).
3034 coveredvp
= vfsp
->vfs_vnodecovered
;
3035 ASSERT(coveredvp
== NULL
|| vn_vfswlock_held(coveredvp
));
3038 * Purge all dnlc entries for this vfs.
3040 (void) dnlc_purge_vfsp(vfsp
, 0);
3042 /* For forcible umount, skip VFS_SYNC() since it may hang */
3043 if ((flag
& MS_FORCE
) == 0)
3044 (void) VFS_SYNC(vfsp
, 0, cr
);
3047 * Lock the vfs to maintain fs status quo during unmount. This
3048 * has to be done after the sync because ufs_update tries to acquire
3051 vfs_lock_wait(vfsp
);
3053 if (error
= VFS_UNMOUNT(vfsp
, flag
, cr
)) {
3055 if (coveredvp
!= NULL
)
3056 vn_vfsunlock(coveredvp
);
3057 } else if (coveredvp
!= NULL
) {
3058 teardown_vopstats(vfsp
);
3060 * vfs_remove() will do a VN_RELE(vfsp->vfs_vnodecovered)
3061 * when it frees vfsp so we do a VN_HOLD() so we can
3062 * continue to use coveredvp afterwards.
3066 vn_vfsunlock(coveredvp
);
3069 teardown_vopstats(vfsp
);
3071 * Release the reference to vfs that is not linked
3072 * into the name space.
3082 * Vfs_unmountall() is called by uadmin() to unmount all
3083 * mounted file systems (except the root file system) during shutdown.
3084 * It follows the existing locking protocol when traversing the vfs list
3085 * to sync and unmount vfses. Even though there should be no
3086 * other thread running while the system is shutting down, it is prudent
3087 * to still follow the locking protocol.
3090 vfs_unmountall(void)
3093 struct vfs
*prev_vfsp
= NULL
;
3097 * Toss all dnlc entries now so that the per-vfs sync
3098 * and unmount operations don't have to slog through
3099 * a bunch of uninteresting vnodes over and over again.
3104 for (vfsp
= rootvfs
->vfs_prev
; vfsp
!= rootvfs
; vfsp
= prev_vfsp
) {
3105 prev_vfsp
= vfsp
->vfs_prev
;
3107 if (vfs_lock(vfsp
) != 0)
3109 error
= vn_vfswlock(vfsp
->vfs_vnodecovered
);
3116 (void) VFS_SYNC(vfsp
, SYNC_CLOSE
, CRED());
3117 (void) dounmount(vfsp
, 0, CRED());
3120 * Since we dropped the vfslist lock above we must
3121 * verify that next_vfsp still exists, else start over.
3124 for (vfsp
= rootvfs
->vfs_prev
;
3125 vfsp
!= rootvfs
; vfsp
= vfsp
->vfs_prev
)
3126 if (vfsp
== prev_vfsp
)
3128 if (vfsp
== rootvfs
&& prev_vfsp
!= rootvfs
)
3129 prev_vfsp
= rootvfs
->vfs_prev
;
3135 * Called to add an entry to the end of the vfs mount in progress list
3138 vfs_addmip(dev_t dev
, struct vfs
*vfsp
)
3142 mipp
= (struct ipmnt
*)kmem_alloc(sizeof (struct ipmnt
), KM_SLEEP
);
3143 mipp
->mip_next
= NULL
;
3144 mipp
->mip_dev
= dev
;
3145 mipp
->mip_vfsp
= vfsp
;
3146 mutex_enter(&vfs_miplist_mutex
);
3147 if (vfs_miplist_end
!= NULL
)
3148 vfs_miplist_end
->mip_next
= mipp
;
3151 vfs_miplist_end
= mipp
;
3152 mutex_exit(&vfs_miplist_mutex
);
3156 * Called to remove an entry from the mount in progress list
3157 * Either because the mount completed or it failed.
3160 vfs_delmip(struct vfs
*vfsp
)
3162 struct ipmnt
*mipp
, *mipprev
;
3164 mutex_enter(&vfs_miplist_mutex
);
3166 for (mipp
= vfs_miplist
;
3167 mipp
&& mipp
->mip_vfsp
!= vfsp
; mipp
= mipp
->mip_next
) {
3171 return; /* shouldn't happen */
3172 if (mipp
== vfs_miplist_end
)
3173 vfs_miplist_end
= mipprev
;
3174 if (mipprev
== NULL
)
3175 vfs_miplist
= mipp
->mip_next
;
3177 mipprev
->mip_next
= mipp
->mip_next
;
3178 mutex_exit(&vfs_miplist_mutex
);
3179 kmem_free(mipp
, sizeof (struct ipmnt
));
3183 * vfs_add is called by a specific filesystem's mount routine to add
3184 * the new vfs into the vfs list/hash and to cover the mounted-on vnode.
3185 * The vfs should already have been locked by the caller.
3187 * coveredvp is NULL if this is the root.
3190 vfs_add(vnode_t
*coveredvp
, struct vfs
*vfsp
, int mflag
)
3194 ASSERT(vfs_lock_held(vfsp
));
3196 newflag
= vfsp
->vfs_flag
;
3197 if (mflag
& MS_RDONLY
)
3198 newflag
|= VFS_RDONLY
;
3200 newflag
&= ~VFS_RDONLY
;
3201 if (mflag
& MS_NOSUID
)
3202 newflag
|= (VFS_NOSETUID
|VFS_NODEVICES
);
3204 newflag
&= ~(VFS_NOSETUID
|VFS_NODEVICES
);
3205 if (mflag
& MS_NOMNTTAB
)
3206 newflag
|= VFS_NOMNTTAB
;
3208 newflag
&= ~VFS_NOMNTTAB
;
3210 if (coveredvp
!= NULL
) {
3211 ASSERT(vn_vfswlock_held(coveredvp
));
3212 coveredvp
->v_vfsmountedhere
= vfsp
;
3215 vfsp
->vfs_vnodecovered
= coveredvp
;
3216 vfsp
->vfs_flag
= newflag
;
3222 * Remove a vfs from the vfs list, null out the pointer from the
3223 * covered vnode to the vfs (v_vfsmountedhere), and null out the pointer
3224 * from the vfs to the covered vnode (vfs_vnodecovered). Release the
3225 * reference to the vfs and to the covered vnode.
3227 * Called from dounmount after it's confirmed with the file system
3228 * that the unmount is legal.
3231 vfs_remove(struct vfs
*vfsp
)
3235 ASSERT(vfs_lock_held(vfsp
));
3238 * Can't unmount root. Should never happen because fs will
3241 if (vfsp
== rootvfs
)
3242 panic("vfs_remove: unmounting root");
3244 vfs_list_remove(vfsp
);
3247 * Unhook from the file system name space.
3249 vp
= vfsp
->vfs_vnodecovered
;
3250 ASSERT(vn_vfswlock_held(vp
));
3251 vp
->v_vfsmountedhere
= NULL
;
3252 vfsp
->vfs_vnodecovered
= NULL
;
3256 * Release lock and wakeup anybody waiting.
3263 * Lock a filesystem to prevent access to it while mounting,
3264 * unmounting and syncing. Return EBUSY immediately if lock
3265 * can't be acquired.
3268 vfs_lock(vfs_t
*vfsp
)
3270 vn_vfslocks_entry_t
*vpvfsentry
;
3272 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3273 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_WRITER
))
3276 vn_vfslocks_rele(vpvfsentry
);
3281 vfs_rlock(vfs_t
*vfsp
)
3283 vn_vfslocks_entry_t
*vpvfsentry
;
3285 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3287 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_READER
))
3290 vn_vfslocks_rele(vpvfsentry
);
3295 vfs_lock_wait(vfs_t
*vfsp
)
3297 vn_vfslocks_entry_t
*vpvfsentry
;
3299 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3300 rwst_enter(&vpvfsentry
->ve_lock
, RW_WRITER
);
3304 vfs_rlock_wait(vfs_t
*vfsp
)
3306 vn_vfslocks_entry_t
*vpvfsentry
;
3308 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3309 rwst_enter(&vpvfsentry
->ve_lock
, RW_READER
);
3313 * Unlock a locked filesystem.
3316 vfs_unlock(vfs_t
*vfsp
)
3318 vn_vfslocks_entry_t
*vpvfsentry
;
3321 * vfs_unlock will mimic sema_v behaviour to fix 4748018.
3322 * And these changes should remain for the patch changes as it is.
3328 * ve_refcount needs to be dropped twice here.
3329 * 1. To release refernce after a call to vfs_locks_getlock()
3330 * 2. To release the reference from the locking routines like
3331 * vfs_rlock_wait/vfs_wlock_wait/vfs_wlock etc,.
3334 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3335 vn_vfslocks_rele(vpvfsentry
);
3337 rwst_exit(&vpvfsentry
->ve_lock
);
3338 vn_vfslocks_rele(vpvfsentry
);
3342 * Utility routine that allows a filesystem to construct its
3343 * fsid in "the usual way" - by munging some underlying dev_t and
3344 * the filesystem type number into the 64-bit fsid. Note that
3345 * this implicitly relies on dev_t persistence to make filesystem
3348 * There's nothing to prevent an individual fs from constructing its
3349 * fsid in a different way, and indeed they should.
3351 * Since we want fsids to be 32-bit quantities (so that they can be
3352 * exported identically by either 32-bit or 64-bit APIs, as well as
3353 * the fact that fsid's are "known" to NFS), we compress the device
3354 * number given down to 32-bits, and panic if that isn't possible.
3357 vfs_make_fsid(fsid_t
*fsi
, dev_t dev
, int val
)
3359 if (!cmpldev((dev32_t
*)&fsi
->val
[0], dev
))
3360 panic("device number too big for fsid!");
3365 vfs_lock_held(vfs_t
*vfsp
)
3368 vn_vfslocks_entry_t
*vpvfsentry
;
3371 * vfs_lock_held will mimic sema_held behaviour
3372 * if panicstr is set. And these changes should remain
3373 * for the patch changes as it is.
3378 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3379 held
= rwst_lock_held(&vpvfsentry
->ve_lock
, RW_WRITER
);
3381 vn_vfslocks_rele(vpvfsentry
);
3386 vfs_lock_owner(vfs_t
*vfsp
)
3388 struct _kthread
*owner
;
3389 vn_vfslocks_entry_t
*vpvfsentry
;
3392 * vfs_wlock_held will mimic sema_held behaviour
3393 * if panicstr is set. And these changes should remain
3394 * for the patch changes as it is.
3399 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3400 owner
= rwst_owner(&vpvfsentry
->ve_lock
);
3402 vn_vfslocks_rele(vpvfsentry
);
3409 * Rather than manipulate the vfslist lock directly, we abstract into lock
3410 * and unlock routines to allow the locking implementation to be changed for
3413 * Whenever the vfs list is modified through its hash links, the overall list
3414 * lock must be obtained before locking the relevant hash bucket. But to see
3415 * whether a given vfs is on the list, it suffices to obtain the lock for the
3416 * hash bucket without getting the overall list lock. (See getvfs() below.)
3422 rw_enter(&vfslist
, RW_WRITER
);
3426 vfs_list_read_lock()
3428 rw_enter(&vfslist
, RW_READER
);
3438 * Low level worker routines for adding entries to and removing entries from
3443 vfs_hash_add(struct vfs
*vfsp
, int insert_at_head
)
3449 ASSERT(RW_WRITE_HELD(&vfslist
));
3451 dev
= expldev(vfsp
->vfs_fsid
.val
[0]);
3452 vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3454 mutex_enter(&rvfs_list
[vhno
].rvfs_lock
);
3457 * Link into the hash table, inserting it at the end, so that LOFS
3458 * with the same fsid as UFS (or other) file systems will not hide the
3461 if (insert_at_head
) {
3462 vfsp
->vfs_hash
= rvfs_list
[vhno
].rvfs_head
;
3463 rvfs_list
[vhno
].rvfs_head
= vfsp
;
3465 for (hp
= &rvfs_list
[vhno
].rvfs_head
; *hp
!= NULL
;
3466 hp
= &(*hp
)->vfs_hash
)
3469 * hp now contains the address of the pointer to update
3470 * to effect the insertion.
3472 vfsp
->vfs_hash
= NULL
;
3476 rvfs_list
[vhno
].rvfs_len
++;
3477 mutex_exit(&rvfs_list
[vhno
].rvfs_lock
);
3482 vfs_hash_remove(struct vfs
*vfsp
)
3488 ASSERT(RW_WRITE_HELD(&vfslist
));
3490 dev
= expldev(vfsp
->vfs_fsid
.val
[0]);
3491 vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3493 mutex_enter(&rvfs_list
[vhno
].rvfs_lock
);
3498 if (rvfs_list
[vhno
].rvfs_head
== vfsp
) {
3499 rvfs_list
[vhno
].rvfs_head
= vfsp
->vfs_hash
;
3500 rvfs_list
[vhno
].rvfs_len
--;
3503 for (tvfsp
= rvfs_list
[vhno
].rvfs_head
; tvfsp
!= NULL
;
3504 tvfsp
= tvfsp
->vfs_hash
) {
3505 if (tvfsp
->vfs_hash
== vfsp
) {
3506 tvfsp
->vfs_hash
= vfsp
->vfs_hash
;
3507 rvfs_list
[vhno
].rvfs_len
--;
3511 cmn_err(CE_WARN
, "vfs_list_remove: vfs not found in hash");
3515 mutex_exit(&rvfs_list
[vhno
].rvfs_lock
);
3520 vfs_list_add(struct vfs
*vfsp
)
3525 * Typically, the vfs_t will have been created on behalf of the file
3526 * system in vfs_init, where it will have been provided with a
3527 * vfs_impl_t. This, however, might be lacking if the vfs_t was created
3528 * by an unbundled file system. We therefore check for such an example
3529 * before stamping the vfs_t with its creation time for the benefit of
3532 if (vfsp
->vfs_implp
== NULL
)
3533 vfsimpl_setup(vfsp
);
3534 vfs_mono_time(&vfsp
->vfs_hrctime
);
3537 * The zone that owns the mount is the one that performed the mount.
3538 * Note that this isn't necessarily the same as the zone mounted into.
3539 * The corresponding zone_rele_ref() will be done when the vfs_t
3542 vfsp
->vfs_zone
= curproc
->p_zone
;
3543 zone_init_ref(&vfsp
->vfs_implp
->vi_zone_ref
);
3544 zone_hold_ref(vfsp
->vfs_zone
, &vfsp
->vfs_implp
->vi_zone_ref
,
3548 * Find the zone mounted into, and put this mount on its vfs list.
3550 zone
= zone_find_by_path(refstr_value(vfsp
->vfs_mntpt
));
3551 ASSERT(zone
!= NULL
);
3553 * Special casing for the root vfs. This structure is allocated
3554 * statically and hooked onto rootvfs at link time. During the
3555 * vfs_mountroot call at system startup time, the root file system's
3556 * VFS_MOUNTROOT routine will call vfs_add with this root vfs struct
3557 * as argument. The code below must detect and handle this special
3558 * case. The only apparent justification for this special casing is
3559 * to ensure that the root file system appears at the head of the
3562 * XXX: I'm assuming that it's ok to do normal list locking when
3563 * adding the entry for the root file system (this used to be
3564 * done with no locks held).
3568 * Link into the vfs list proper.
3570 if (vfsp
== &root
) {
3572 * Assert: This vfs is already on the list as its first entry.
3573 * Thus, there's nothing to do.
3575 ASSERT(rootvfs
== vfsp
);
3577 * Add it to the head of the global zone's vfslist.
3579 ASSERT(zone
== global_zone
);
3580 ASSERT(zone
->zone_vfslist
== NULL
);
3581 zone
->zone_vfslist
= vfsp
;
3584 * Link to end of list using vfs_prev (as rootvfs is now a
3585 * doubly linked circular list) so list is in mount order for
3588 rootvfs
->vfs_prev
->vfs_next
= vfsp
;
3589 vfsp
->vfs_prev
= rootvfs
->vfs_prev
;
3590 rootvfs
->vfs_prev
= vfsp
;
3591 vfsp
->vfs_next
= rootvfs
;
3594 * Do it again for the zone-private list (which may be NULL).
3596 if (zone
->zone_vfslist
== NULL
) {
3597 ASSERT(zone
!= global_zone
);
3598 zone
->zone_vfslist
= vfsp
;
3600 zone
->zone_vfslist
->vfs_zone_prev
->vfs_zone_next
= vfsp
;
3601 vfsp
->vfs_zone_prev
= zone
->zone_vfslist
->vfs_zone_prev
;
3602 zone
->zone_vfslist
->vfs_zone_prev
= vfsp
;
3603 vfsp
->vfs_zone_next
= zone
->zone_vfslist
;
3608 * Link into the hash table, inserting it at the end, so that LOFS
3609 * with the same fsid as UFS (or other) file systems will not hide
3612 vfs_hash_add(vfsp
, 0);
3615 * update the mnttab modification time
3617 vfs_mnttab_modtimeupd();
3623 vfs_list_remove(struct vfs
*vfsp
)
3627 zone
= zone_find_by_path(refstr_value(vfsp
->vfs_mntpt
));
3628 ASSERT(zone
!= NULL
);
3630 * Callers are responsible for preventing attempts to unmount the
3633 ASSERT(vfsp
!= rootvfs
);
3640 vfs_hash_remove(vfsp
);
3643 * Remove from vfs list.
3645 vfsp
->vfs_prev
->vfs_next
= vfsp
->vfs_next
;
3646 vfsp
->vfs_next
->vfs_prev
= vfsp
->vfs_prev
;
3647 vfsp
->vfs_next
= vfsp
->vfs_prev
= NULL
;
3650 * Remove from zone-specific vfs list.
3652 if (zone
->zone_vfslist
== vfsp
)
3653 zone
->zone_vfslist
= vfsp
->vfs_zone_next
;
3655 if (vfsp
->vfs_zone_next
== vfsp
) {
3656 ASSERT(vfsp
->vfs_zone_prev
== vfsp
);
3657 ASSERT(zone
->zone_vfslist
== vfsp
);
3658 zone
->zone_vfslist
= NULL
;
3661 vfsp
->vfs_zone_prev
->vfs_zone_next
= vfsp
->vfs_zone_next
;
3662 vfsp
->vfs_zone_next
->vfs_zone_prev
= vfsp
->vfs_zone_prev
;
3663 vfsp
->vfs_zone_next
= vfsp
->vfs_zone_prev
= NULL
;
3666 * update the mnttab modification time
3668 vfs_mnttab_modtimeupd();
3674 getvfs(fsid_t
*fsid
)
3677 int val0
= fsid
->val
[0];
3678 int val1
= fsid
->val
[1];
3679 dev_t dev
= expldev(val0
);
3680 int vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3681 kmutex_t
*hmp
= &rvfs_list
[vhno
].rvfs_lock
;
3684 for (vfsp
= rvfs_list
[vhno
].rvfs_head
; vfsp
; vfsp
= vfsp
->vfs_hash
) {
3685 if (vfsp
->vfs_fsid
.val
[0] == val0
&&
3686 vfsp
->vfs_fsid
.val
[1] == val1
) {
3697 * Search the vfs mount in progress list for a specified device/vfs entry.
3698 * Returns 0 if the first entry in the list that the device matches has the
3699 * given vfs pointer as well. If the device matches but a different vfs
3700 * pointer is encountered in the list before the given vfs pointer then
3705 vfs_devmounting(dev_t dev
, struct vfs
*vfsp
)
3710 mutex_enter(&vfs_miplist_mutex
);
3711 for (mipp
= vfs_miplist
; mipp
!= NULL
; mipp
= mipp
->mip_next
) {
3712 if (mipp
->mip_dev
== dev
) {
3713 if (mipp
->mip_vfsp
!= vfsp
)
3718 mutex_exit(&vfs_miplist_mutex
);
3723 * Search the vfs list for a specified device. Returns 1, if entry is found
3724 * or 0 if no suitable entry is found.
3728 vfs_devismounted(dev_t dev
)
3733 vfs_list_read_lock();
3737 if (vfsp
->vfs_dev
== dev
) {
3741 vfsp
= vfsp
->vfs_next
;
3742 } while (vfsp
!= rootvfs
);
3749 * Search the vfs list for a specified device. Returns a pointer to it
3750 * or NULL if no suitable entry is found. The caller of this routine
3751 * is responsible for releasing the returned vfs pointer.
3754 vfs_dev2vfsp(dev_t dev
)
3759 vfs_list_read_lock();
3764 * The following could be made more efficient by making
3765 * the entire loop use vfs_zone_next if the call is from
3766 * a zone. The only callers, however, ustat(2) and
3767 * umount2(2), don't seem to justify the added
3768 * complexity at present.
3770 if (vfsp
->vfs_dev
== dev
&&
3771 ZONE_PATH_VISIBLE(refstr_value(vfsp
->vfs_mntpt
),
3777 vfsp
= vfsp
->vfs_next
;
3778 } while (vfsp
!= rootvfs
);
3780 return (found
? vfsp
: NULL
);
3784 * Search the vfs list for a specified mntpoint. Returns a pointer to it
3785 * or NULL if no suitable entry is found. The caller of this routine
3786 * is responsible for releasing the returned vfs pointer.
3788 * Note that if multiple mntpoints match, the last one matching is
3789 * returned in an attempt to return the "top" mount when overlay
3790 * mounts are covering the same mount point. This is accomplished by starting
3791 * at the end of the list and working our way backwards, stopping at the first
3795 vfs_mntpoint2vfsp(const char *mp
)
3798 struct vfs
*retvfsp
= NULL
;
3799 zone_t
*zone
= curproc
->p_zone
;
3802 vfs_list_read_lock();
3803 if (getzoneid() == GLOBAL_ZONEID
) {
3805 * The global zone may see filesystems in any zone.
3807 vfsp
= rootvfs
->vfs_prev
;
3809 if (strcmp(refstr_value(vfsp
->vfs_mntpt
), mp
) == 0) {
3813 vfsp
= vfsp
->vfs_prev
;
3814 } while (vfsp
!= rootvfs
->vfs_prev
);
3815 } else if ((list
= zone
->zone_vfslist
) != NULL
) {
3818 vfsp
= list
->vfs_zone_prev
;
3820 mntpt
= refstr_value(vfsp
->vfs_mntpt
);
3821 mntpt
= ZONE_PATH_TRANSLATE(mntpt
, zone
);
3822 if (strcmp(mntpt
, mp
) == 0) {
3826 vfsp
= vfsp
->vfs_zone_prev
;
3827 } while (vfsp
!= list
->vfs_zone_prev
);
3836 * Search the vfs list for a specified vfsops.
3837 * if vfs entry is found then return 1, else 0.
3840 vfs_opsinuse(vfsops_t
*ops
)
3845 vfs_list_read_lock();
3849 if (vfs_getops(vfsp
) == ops
) {
3853 vfsp
= vfsp
->vfs_next
;
3854 } while (vfsp
!= rootvfs
);
3860 * Allocate an entry in vfssw for a file system type
3863 allocate_vfssw(const char *type
)
3867 if (type
[0] == '\0' || strlen(type
) + 1 > _ST_FSTYPSZ
) {
3869 * The vfssw table uses the empty string to identify an
3870 * available entry; we cannot add any type which has
3871 * a leading NUL. The string length is limited to
3872 * the size of the st_fstype array in struct stat.
3877 ASSERT(VFSSW_WRITE_LOCKED());
3878 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++)
3879 if (!ALLOCATED_VFSSW(vswp
)) {
3880 vswp
->vsw_name
= kmem_alloc(strlen(type
) + 1, KM_SLEEP
);
3881 (void) strcpy(vswp
->vsw_name
, type
);
3882 ASSERT(vswp
->vsw_count
== 0);
3883 vswp
->vsw_count
= 1;
3884 mutex_init(&vswp
->vsw_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
3891 * Impose additional layer of translation between vfstype names
3892 * and module names in the filesystem.
3895 vfs_to_modname(const char *vfstype
)
3897 if (strcmp(vfstype
, "proc") == 0) {
3899 } else if (strcmp(vfstype
, "fd") == 0) {
3901 } else if (strncmp(vfstype
, "nfs", 3) == 0) {
3909 * Find a vfssw entry given a file system type name.
3910 * Try to autoload the filesystem if it's not found.
3911 * If it's installed, return the vfssw locked to prevent unloading.
3914 vfs_getvfssw(const char *type
)
3917 const char *modname
;
3920 vswp
= vfs_getvfsswbyname(type
);
3921 modname
= vfs_to_modname(type
);
3923 if (rootdir
== NULL
) {
3925 * If we haven't yet loaded the root file system, then our
3926 * _init won't be called until later. Allocate vfssw entry,
3927 * because mod_installfs won't be called.
3932 if ((vswp
= vfs_getvfsswbyname(type
)) == NULL
) {
3933 if ((vswp
= allocate_vfssw(type
)) == NULL
) {
3941 if (!VFS_INSTALLED(vswp
)) {
3943 (void) modloadonly("fs", modname
);
3950 * Try to load the filesystem. Before calling modload(), we drop
3951 * our lock on the VFS switch table, and pick it up after the
3952 * module is loaded. However, there is a potential race: the
3953 * module could be unloaded after the call to modload() completes
3954 * but before we pick up the lock and drive on. Therefore,
3955 * we keep reloading the module until we've loaded the module
3956 * _and_ we have the lock on the VFS switch table.
3958 while (vswp
== NULL
|| !VFS_INSTALLED(vswp
)) {
3960 if (modload("fs", modname
) == -1)
3964 if ((vswp
= vfs_getvfsswbyname(type
)) == NULL
)
3973 * Find a vfssw entry given a file system type name.
3976 vfs_getvfsswbyname(const char *type
)
3980 ASSERT(VFSSW_LOCKED());
3981 if (type
== NULL
|| *type
== '\0')
3984 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
3985 if (strcmp(type
, vswp
->vsw_name
) == 0) {
3995 * Find a vfssw entry given a set of vfsops.
3998 vfs_getvfsswbyvfsops(vfsops_t
*vfsops
)
4003 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
4004 if (ALLOCATED_VFSSW(vswp
) && &vswp
->vsw_vfsops
== vfsops
) {
4016 * Reference a vfssw entry.
4019 vfs_refvfssw(struct vfssw
*vswp
)
4022 mutex_enter(&vswp
->vsw_lock
);
4024 mutex_exit(&vswp
->vsw_lock
);
4028 * Unreference a vfssw entry.
4031 vfs_unrefvfssw(struct vfssw
*vswp
)
4034 mutex_enter(&vswp
->vsw_lock
);
4036 mutex_exit(&vswp
->vsw_lock
);
4039 int sync_timeout
= 30; /* timeout for syncing a page during panic */
4040 int sync_timeleft
; /* portion of sync_timeout remaining */
4042 static int sync_retries
= 20; /* number of retries when not making progress */
4043 static int sync_triesleft
; /* portion of sync_retries remaining */
4045 static pgcnt_t old_pgcnt
, new_pgcnt
;
4046 static int new_bufcnt
, old_bufcnt
;
4049 * Sync all of the mounted filesystems, and then wait for the actual i/o to
4050 * complete. We wait by counting the number of dirty pages and buffers,
4051 * pushing them out using bio_busy() and page_busy(), and then counting again.
4052 * This routine is used during both the uadmin A_SHUTDOWN code as well as
4053 * the SYNC phase of the panic code (see comments in panic.c). It should only
4054 * be used after some higher-level mechanism has quiesced the system so that
4055 * new writes are not being initiated while we are waiting for completion.
4057 * To ensure finite running time, our algorithm uses two timeout mechanisms:
4058 * sync_timeleft (a timer implemented by the omnipresent deadman() cyclic), and
4059 * sync_triesleft (a progress counter used by the vfs_syncall() loop below).
4060 * Together these ensure that syncing completes if our i/o paths are stuck.
4061 * The counters are declared above so they can be found easily in the debugger.
4063 * The sync_timeleft counter is reset by bio_busy() and page_busy() using the
4064 * vfs_syncprogress() subroutine whenever we make progress through the lists of
4065 * pages and buffers. It is decremented and expired by the deadman() cyclic.
4066 * When vfs_syncall() decides it is done, we disable the deadman() counter by
4067 * setting sync_timeleft to zero. This timer guards against vfs_syncall()
4068 * deadlocking or hanging inside of a broken filesystem or driver routine.
4070 * The sync_triesleft counter is updated by vfs_syncall() itself. If we make
4071 * sync_retries consecutive calls to bio_busy() and page_busy() without
4072 * decreasing either the number of dirty buffers or dirty pages below the
4073 * lowest count we have seen so far, we give up and return from vfs_syncall().
4075 * Each loop iteration ends with a call to delay() one second to allow time for
4076 * i/o completion and to permit the user time to read our progress messages.
4081 if (rootdir
== NULL
&& !modrootloaded
)
4082 return; /* panic during boot - no filesystems yet */
4084 printf("syncing file systems...");
4089 sync_triesleft
= sync_retries
;
4091 old_bufcnt
= new_bufcnt
= INT_MAX
;
4092 old_pgcnt
= new_pgcnt
= ULONG_MAX
;
4094 while (sync_triesleft
> 0) {
4095 old_bufcnt
= MIN(old_bufcnt
, new_bufcnt
);
4096 old_pgcnt
= MIN(old_pgcnt
, new_pgcnt
);
4098 new_bufcnt
= bio_busy(B_TRUE
);
4099 new_pgcnt
= page_busy(B_TRUE
);
4102 if (new_bufcnt
== 0 && new_pgcnt
== 0)
4105 if (new_bufcnt
< old_bufcnt
|| new_pgcnt
< old_pgcnt
)
4106 sync_triesleft
= sync_retries
;
4111 printf(" [%d]", new_bufcnt
);
4113 printf(" %lu", new_pgcnt
);
4118 if (new_bufcnt
!= 0 || new_pgcnt
!= 0)
4119 printf(" done (not all i/o completed)\n");
4128 * If we are in the middle of the sync phase of panic, reset sync_timeleft to
4129 * sync_timeout to indicate that we are making progress and the deadman()
4130 * omnipresent cyclic should not yet time us out. Note that it is safe to
4131 * store to sync_timeleft here since the deadman() is firing at high-level
4132 * on top of us. If we are racing with the deadman(), either the deadman()
4133 * will decrement the old value and then we will reset it, or we will
4134 * reset it and then the deadman() will immediately decrement it. In either
4135 * case, correct behavior results.
4138 vfs_syncprogress(void)
4141 sync_timeleft
= sync_timeout
;
4145 * Map VFS flags to statvfs flags. These shouldn't really be separate
4149 vf_to_stf(uint_t vf
)
4153 if (vf
& VFS_RDONLY
)
4155 if (vf
& VFS_NOSETUID
)
4157 if (vf
& VFS_NOTRUNC
)
4164 * Entries for (illegal) fstype 0.
4168 vfsstray_sync(struct vfs
*vfsp
, short arg
, struct cred
*cr
)
4170 cmn_err(CE_PANIC
, "stray vfs operation");
4175 * Entries for (illegal) fstype 0.
4180 cmn_err(CE_PANIC
, "stray vfs operation");
4185 * Support for dealing with forced UFS unmount and its interaction with
4186 * LOFS. Could be used by any filesystem.
4196 * We've gotta define the op for sync separately, since the compiler gets
4197 * confused if we mix and match ANSI and normal style prototypes when
4198 * a "short" argument is present and spits out a warning.
4202 vfs_EIO_sync(struct vfs
*vfsp
, short arg
, struct cred
*cr
)
4208 vfsops_t
*EIO_vfsops
;
4211 * Called from startup() to initialize all loaded vfs's
4218 extern int vopstats_enabled
;
4219 extern void vopstats_startup();
4221 static const fs_operation_def_t EIO_vfsops_template
[] = {
4222 VFSNAME_MOUNT
, { .error
= vfs_EIO
},
4223 VFSNAME_UNMOUNT
, { .error
= vfs_EIO
},
4224 VFSNAME_ROOT
, { .error
= vfs_EIO
},
4225 VFSNAME_STATVFS
, { .error
= vfs_EIO
},
4226 VFSNAME_SYNC
, { .vfs_sync
= vfs_EIO_sync
},
4227 VFSNAME_VGET
, { .error
= vfs_EIO
},
4228 VFSNAME_MOUNTROOT
, { .error
= vfs_EIO
},
4229 VFSNAME_FREEVFS
, { .error
= vfs_EIO
},
4230 VFSNAME_VNSTATE
, { .error
= vfs_EIO
},
4234 static const fs_operation_def_t stray_vfsops_template
[] = {
4235 VFSNAME_MOUNT
, { .error
= vfsstray
},
4236 VFSNAME_UNMOUNT
, { .error
= vfsstray
},
4237 VFSNAME_ROOT
, { .error
= vfsstray
},
4238 VFSNAME_STATVFS
, { .error
= vfsstray
},
4239 VFSNAME_SYNC
, { .vfs_sync
= vfsstray_sync
},
4240 VFSNAME_VGET
, { .error
= vfsstray
},
4241 VFSNAME_MOUNTROOT
, { .error
= vfsstray
},
4242 VFSNAME_FREEVFS
, { .error
= vfsstray
},
4243 VFSNAME_VNSTATE
, { .error
= vfsstray
},
4247 /* Create vfs cache */
4248 vfs_cache
= kmem_cache_create("vfs_cache", sizeof (struct vfs
),
4249 sizeof (uintptr_t), NULL
, NULL
, NULL
, NULL
, NULL
, 0);
4251 /* Initialize the vnode cache (file systems may use it during init). */
4254 /* Setup event monitor framework */
4257 /* Initialize the dummy stray file system type. */
4258 error
= vfs_setfsops(0, stray_vfsops_template
, NULL
);
4260 /* Initialize the dummy EIO file system. */
4261 error
= vfs_makefsops(EIO_vfsops_template
, &EIO_vfsops
);
4263 cmn_err(CE_WARN
, "vfsinit: bad EIO vfs ops template");
4264 /* Shouldn't happen, but not bad enough to panic */
4267 VFS_INIT(&EIO_vfs
, EIO_vfsops
, (caddr_t
)NULL
);
4270 * Default EIO_vfs.vfs_flag to VFS_UNMOUNTED so a lookup
4271 * on this vfs can immediately notice it's invalid.
4273 EIO_vfs
.vfs_flag
|= VFS_UNMOUNTED
;
4276 * Call the init routines of non-loadable filesystems only.
4277 * Filesystems which are loaded as separate modules will be
4278 * initialized by the module loading code instead.
4281 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
4283 if (vswp
->vsw_init
!= NULL
)
4284 (*vswp
->vsw_init
)(vswp
- vfssw
, vswp
->vsw_name
);
4290 if (vopstats_enabled
) {
4291 /* EIO_vfs can collect stats, but we don't retrieve them */
4292 initialize_vopstats(&EIO_vfs
.vfs_vopstats
);
4293 EIO_vfs
.vfs_fstypevsp
= NULL
;
4294 EIO_vfs
.vfs_vskap
= NULL
;
4295 EIO_vfs
.vfs_flag
|= VFS_STATS
;
4300 reparse_point_init();
4304 vfs_alloc(int kmflag
)
4308 vfsp
= kmem_cache_alloc(vfs_cache
, kmflag
);
4311 * Do the simplest initialization here.
4312 * Everything else gets done in vfs_init()
4314 bzero(vfsp
, sizeof (vfs_t
));
4319 vfs_free(vfs_t
*vfsp
)
4322 * One would be tempted to assert that "vfsp->vfs_count == 0".
4323 * The problem is that this gets called out of domount() with
4324 * a partially initialized vfs and a vfs_count of 1. This is
4325 * also called from vfs_rele() with a vfs_count of 0. We can't
4326 * call VFS_RELE() from domount() if VFS_MOUNT() hasn't successfully
4327 * returned. This is because VFS_MOUNT() fully initializes the
4328 * vfs structure and its associated data. VFS_RELE() will call
4329 * VFS_FREEVFS() which may panic the system if the data structures
4330 * aren't fully initialized from a successful VFS_MOUNT()).
4333 /* If FEM was in use, make sure everything gets cleaned up */
4334 if (vfsp
->vfs_femhead
) {
4335 ASSERT(vfsp
->vfs_femhead
->femh_list
== NULL
);
4336 mutex_destroy(&vfsp
->vfs_femhead
->femh_lock
);
4337 kmem_free(vfsp
->vfs_femhead
, sizeof (*(vfsp
->vfs_femhead
)));
4338 vfsp
->vfs_femhead
= NULL
;
4341 if (vfsp
->vfs_implp
)
4342 vfsimpl_teardown(vfsp
);
4343 sema_destroy(&vfsp
->vfs_reflock
);
4344 kmem_cache_free(vfs_cache
, vfsp
);
4348 * Increments the vfs reference count by one atomically.
4351 vfs_hold(vfs_t
*vfsp
)
4353 atomic_inc_32(&vfsp
->vfs_count
);
4354 ASSERT(vfsp
->vfs_count
!= 0);
4358 * Decrements the vfs reference count by one atomically. When
4359 * vfs reference count becomes zero, it calls the file system
4360 * specific vfs_freevfs() to free up the resources.
4363 vfs_rele(vfs_t
*vfsp
)
4365 ASSERT(vfsp
->vfs_count
!= 0);
4366 if (atomic_dec_32_nv(&vfsp
->vfs_count
) == 0) {
4370 zone_rele_ref(&vfsp
->vfs_implp
->vi_zone_ref
,
4372 vfs_freemnttab(vfsp
);
4378 * Generic operations vector support.
4380 * This is used to build operations vectors for both the vfs and vnode.
4381 * It's normally called only when a file system is loaded.
4383 * There are many possible algorithms for this, including the following:
4385 * (1) scan the list of known operations; for each, see if the file system
4386 * includes an entry for it, and fill it in as appropriate.
4388 * (2) set up defaults for all known operations. scan the list of ops
4389 * supplied by the file system; for each which is both supplied and
4390 * known, fill it in.
4392 * (3) sort the lists of known ops & supplied ops; scan the list, filling
4393 * in entries as we go.
4395 * we choose (1) for simplicity, and because performance isn't critical here.
4396 * note that (2) could be sped up using a precomputed hash table on known ops.
4397 * (3) could be faster than either, but only if the lists were very large or
4398 * supplied in sorted order.
4403 fs_build_vector(void *vector
, int *unused_ops
,
4404 const fs_operation_trans_def_t
*translation
,
4405 const fs_operation_def_t
*operations
)
4407 int i
, num_trans
, num_ops
, used
;
4410 * Count the number of translations and the number of supplied
4415 const fs_operation_trans_def_t
*p
;
4417 for (num_trans
= 0, p
= translation
;
4424 const fs_operation_def_t
*p
;
4426 for (num_ops
= 0, p
= operations
;
4432 /* Walk through each operation known to our caller. There will be */
4433 /* one entry in the supplied "translation table" for each. */
4437 for (i
= 0; i
< num_trans
; i
++) {
4440 fs_generic_func_p result
;
4441 fs_generic_func_p
*location
;
4443 curname
= translation
[i
].name
;
4445 /* Look for a matching operation in the list supplied by the */
4450 for (j
= 0; j
< num_ops
; j
++) {
4451 if (strcmp(operations
[j
].name
, curname
) == 0) {
4459 * If the file system is using a "placeholder" for default
4460 * or error functions, grab the appropriate function out of
4461 * the translation table. If the file system didn't supply
4462 * this operation at all, use the default function.
4466 result
= operations
[j
].func
.fs_generic
;
4467 if (result
== fs_default
) {
4468 result
= translation
[i
].defaultFunc
;
4469 } else if (result
== fs_error
) {
4470 result
= translation
[i
].errorFunc
;
4471 } else if (result
== NULL
) {
4472 /* Null values are PROHIBITED */
4476 result
= translation
[i
].defaultFunc
;
4479 /* Now store the function into the operations vector. */
4481 location
= (fs_generic_func_p
*)
4482 (((char *)vector
) + translation
[i
].offset
);
4487 *unused_ops
= num_ops
- used
;
4492 /* Placeholder functions, should never be called. */
4497 cmn_err(CE_PANIC
, "fs_error called");
4504 cmn_err(CE_PANIC
, "fs_default called");
4511 * Part of the implementation of booting off a mirrored root
4512 * involves a change of dev_t for the root device. To
4513 * accomplish this, first remove the existing hash table
4514 * entry for the root device, convert to the new dev_t,
4515 * then re-insert in the hash table at the head of the list.
4518 vfs_root_redev(vfs_t
*vfsp
, dev_t ndev
, int fstype
)
4522 vfs_hash_remove(vfsp
);
4524 vfsp
->vfs_dev
= ndev
;
4525 vfs_make_fsid(&vfsp
->vfs_fsid
, ndev
, fstype
);
4527 vfs_hash_add(vfsp
, 1);
4532 #else /* x86 NEWBOOT */
4535 extern int hvmboot_rootconf();
4538 extern ib_boot_prop_t
*iscsiboot_prop
;
4545 extern void pm_init();
4546 char *fstyp
, *fsmod
;
4549 getrootfs(&fstyp
, &fsmod
);
4553 * hvmboot_rootconf() is defined in the hvm_bootstrap misc module,
4554 * which lives in /platform/i86hvm, and hence is only available when
4555 * booted in an x86 hvm environment. If the hvm_bootstrap misc module
4556 * is not available then the modstub for this function will return 0.
4557 * If the hvm_bootstrap misc module is available it will be loaded
4558 * and hvmboot_rootconf() will be invoked.
4560 if (error
= hvmboot_rootconf())
4564 if (error
= clboot_rootconf())
4567 if (modload("fs", fsmod
) == -1)
4568 panic("Cannot _init %s module", fsmod
);
4571 vsw
= vfs_getvfsswbyname(fstyp
);
4574 cmn_err(CE_CONT
, "Cannot find %s filesystem\n", fstyp
);
4577 VFS_INIT(rootvfs
, &vsw
->vsw_vfsops
, 0);
4580 /* always mount readonly first */
4581 rootvfs
->vfs_flag
|= VFS_RDONLY
;
4585 if (netboot
&& iscsiboot_prop
) {
4586 cmn_err(CE_WARN
, "NFS boot and iSCSI boot"
4587 " shouldn't happen in the same time");
4591 if (netboot
|| iscsiboot_prop
) {
4594 cmn_err(CE_WARN
, "Cannot plumb network device %d", ret
);
4599 if ((ret
== 0) && iscsiboot_prop
) {
4600 ret
= modload("drv", "iscsi");
4601 /* -1 indicates fail */
4603 cmn_err(CE_WARN
, "Failed to load iscsi module");
4604 iscsi_boot_prop_free();
4607 if (!i_ddi_attach_pseudo_node("iscsi")) {
4609 "Failed to attach iscsi driver");
4610 iscsi_boot_prop_free();
4616 error
= VFS_MOUNTROOT(rootvfs
, ROOT_INIT
);
4617 vfs_unrefvfssw(vsw
);
4618 rootdev
= rootvfs
->vfs_dev
;
4621 cmn_err(CE_CONT
, "Cannot mount root on %s fstype %s\n",
4622 rootfs
.bo_name
, fstyp
);
4624 cmn_err(CE_CONT
, "?root on %s fstype %s\n",
4625 rootfs
.bo_name
, fstyp
);
4630 * XXX this is called by nfs only and should probably be removed
4631 * If booted with ASKNAME, prompt on the console for a filesystem
4632 * name and return it.
4635 getfsname(char *askfor
, char *name
, size_t namelen
)
4637 if (boothowto
& RB_ASKNAME
) {
4638 printf("%s name: ", askfor
);
4639 console_gets(name
, namelen
);
4644 * Init the root filesystem type (rootfs.bo_fstype) from the "fstype"
4647 * Filesystem types starting with the prefix "nfs" are diskless clients;
4648 * init the root filename name (rootfs.bo_name), too.
4650 * If we are booting via NFS we currently have these options:
4651 * nfs - dynamically choose NFS V2, V3, or V4 (default)
4652 * nfs2 - force NFS V2
4653 * nfs3 - force NFS V3
4654 * nfs4 - force NFS V4
4655 * Because we need to maintain backward compatibility with the naming
4656 * convention that the NFS V2 filesystem name is "nfs" (see vfs_conf.c)
4657 * we need to map "nfs" => "nfsdyn" and "nfs2" => "nfs". The dynamic
4658 * nfs module will map the type back to either "nfs", "nfs3", or "nfs4".
4659 * This is only for root filesystems, all other uses will expect
4660 * that "nfs" == NFS V2.
4663 getrootfs(char **fstypp
, char **fsmodp
)
4665 extern char *strplumb_get_netdev_path(void);
4666 char *propstr
= NULL
;
4669 * Check fstype property; for diskless it should be one of "nfs",
4670 * "nfs2", "nfs3" or "nfs4".
4672 if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4673 DDI_PROP_DONTPASS
, "fstype", &propstr
)
4675 (void) strncpy(rootfs
.bo_fstype
, propstr
, BO_MAXFSNAME
);
4676 ddi_prop_free(propstr
);
4679 * if the boot property 'fstype' is not set, but 'zfs-bootfs' is set,
4680 * assume the type of this root filesystem is 'zfs'.
4682 } else if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4683 DDI_PROP_DONTPASS
, "zfs-bootfs", &propstr
)
4685 (void) strncpy(rootfs
.bo_fstype
, "zfs", BO_MAXFSNAME
);
4686 ddi_prop_free(propstr
);
4689 if (strncmp(rootfs
.bo_fstype
, "nfs", 3) != 0) {
4690 *fstypp
= *fsmodp
= rootfs
.bo_fstype
;
4696 if (strcmp(rootfs
.bo_fstype
, "nfs2") == 0)
4697 (void) strcpy(rootfs
.bo_fstype
, "nfs");
4698 else if (strcmp(rootfs
.bo_fstype
, "nfs") == 0)
4699 (void) strcpy(rootfs
.bo_fstype
, "nfsdyn");
4702 * check if path to network interface is specified in bootpath
4703 * or by a hypervisor domain configuration file.
4704 * XXPV - enable strlumb_get_netdev_path()
4706 if (ddi_prop_exists(DDI_DEV_T_ANY
, ddi_root_node(), DDI_PROP_DONTPASS
,
4708 (void) strcpy(rootfs
.bo_name
, "/xpvd/xnf@0");
4709 } else if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4710 DDI_PROP_DONTPASS
, "bootpath", &propstr
)
4712 (void) strncpy(rootfs
.bo_name
, propstr
, BO_MAXOBJNAME
);
4713 ddi_prop_free(propstr
);
4715 /* attempt to determine netdev_path via boot_mac address */
4716 netdev_path
= strplumb_get_netdev_path();
4717 if (netdev_path
== NULL
)
4718 panic("cannot find boot network interface");
4719 (void) strncpy(rootfs
.bo_name
, netdev_path
, BO_MAXOBJNAME
);
4721 *fstypp
= rootfs
.bo_fstype
;
4727 * VFS feature routines
4730 #define VFTINDEX(feature) (((feature) >> 32) & 0xFFFFFFFF)
4731 #define VFTBITS(feature) ((feature) & 0xFFFFFFFFLL)
4733 /* Register a feature in the vfs */
4735 vfs_set_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4737 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4738 if (vfsp
->vfs_implp
== NULL
)
4741 vfsp
->vfs_featureset
[VFTINDEX(feature
)] |= VFTBITS(feature
);
4745 vfs_clear_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4747 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4748 if (vfsp
->vfs_implp
== NULL
)
4750 vfsp
->vfs_featureset
[VFTINDEX(feature
)] &= VFTBITS(~feature
);
4754 * Query a vfs for a feature.
4755 * Returns 1 if feature is present, 0 if not
4758 vfs_has_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4762 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4763 if (vfsp
->vfs_implp
== NULL
)
4766 if (vfsp
->vfs_featureset
[VFTINDEX(feature
)] & VFTBITS(feature
))
4773 * Propagate feature set from one vfs to another
4776 vfs_propagate_features(vfs_t
*from
, vfs_t
*to
)
4780 if (to
->vfs_implp
== NULL
|| from
->vfs_implp
== NULL
)
4783 for (i
= 1; i
<= to
->vfs_featureset
[0]; i
++) {
4784 to
->vfs_featureset
[i
] = from
->vfs_featureset
[i
];
4788 #define LOFINODE_PATH "/dev/lofi/%d"
4791 * Return the vnode for the lofi node if there's a lofi mount in place.
4792 * Returns -1 when there's no lofi node, 0 on success, and > 0 on
4796 vfs_get_lofi(vfs_t
*vfsp
, vnode_t
**vpp
)
4802 if (vfsp
->vfs_lofi_minor
== 0) {
4807 strsize
= snprintf(NULL
, 0, LOFINODE_PATH
, vfsp
->vfs_lofi_minor
);
4808 path
= kmem_alloc(strsize
+ 1, KM_SLEEP
);
4809 (void) snprintf(path
, strsize
+ 1, LOFINODE_PATH
, vfsp
->vfs_lofi_minor
);
4812 * We may be inside a zone, so we need to use the /dev path, but
4813 * it's created asynchronously, so we wait here.
4816 err
= lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, vpp
);
4821 if ((err
= delay_sig(hz
/ 8)) == EINTR
)
4828 kmem_free(path
, strsize
+ 1);