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]
23 * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2019, Joyent, Inc.
25 * Copyright 2016 Toomas Soome <tsoome@me.com>
26 * Copyright (c) 2016, 2017 by Delphix. All rights reserved.
27 * Copyright 2016 Nexenta Systems, Inc.
28 * Copyright 2017 RackTop Systems.
29 * Copyright 2024 Oxide Computer Company
32 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
33 /* All Rights Reserved */
36 * University Copyright- Copyright (c) 1982, 1986, 1988
37 * The Regents of the University of California
40 * University Acknowledgment- Portions of this document are derived from
41 * software developed by the University of California, Berkeley, and its
45 #include <sys/types.h>
46 #include <sys/t_lock.h>
47 #include <sys/param.h>
48 #include <sys/errno.h>
50 #include <sys/fstyp.h>
52 #include <sys/systm.h>
54 #include <sys/mount.h>
56 #include <sys/vfs_opreg.h>
58 #include <sys/mntent.h>
60 #include <sys/statvfs.h>
61 #include <sys/statfs.h>
63 #include <sys/vnode.h>
64 #include <sys/rwstlock.h>
68 #include <sys/atomic.h>
69 #include <sys/cmn_err.h>
72 #include <sys/debug.h>
73 #include <sys/vnode.h>
74 #include <sys/modctl.h>
76 #include <sys/pathname.h>
77 #include <sys/bootconf.h>
78 #include <sys/dumphdr.h>
79 #include <sys/dc_ki.h>
81 #include <sys/sunddi.h>
82 #include <sys/sysmacros.h>
84 #include <sys/policy.h>
86 #include <sys/objfs.h>
87 #include <sys/console.h>
88 #include <sys/reboot.h>
93 #include <sys/bootprops.h>
97 #include <fs/fs_subr.h>
98 /* Private interfaces to create vopstats-related data structures */
99 extern void initialize_vopstats(vopstats_t
*);
100 extern vopstats_t
*get_fstype_vopstats(struct vfs
*, struct vfssw
*);
101 extern vsk_anchor_t
*get_vskstat_anchor(struct vfs
*);
103 static void vfs_clearmntopt_nolock(mntopts_t
*, const char *, int);
104 static void vfs_setmntopt_nolock(mntopts_t
*, const char *,
105 const char *, int, int);
106 static int vfs_optionisset_nolock(const mntopts_t
*, const char *, char **);
107 static void vfs_freemnttab(struct vfs
*);
108 static void vfs_freeopt(mntopt_t
*);
109 static void vfs_swapopttbl_nolock(mntopts_t
*, mntopts_t
*);
110 static void vfs_swapopttbl(mntopts_t
*, mntopts_t
*);
111 static void vfs_copyopttbl_extend(const mntopts_t
*, mntopts_t
*, int);
112 static void vfs_createopttbl_extend(mntopts_t
*, const char *,
114 static char **vfs_copycancelopt_extend(char **const, int);
115 static void vfs_freecancelopt(char **);
116 static void getrootfs(char **, char **);
117 static int getmacpath(dev_info_t
*, void *);
118 static void vfs_mnttabvp_setup(void);
121 struct ipmnt
*mip_next
;
123 struct vfs
*mip_vfsp
;
126 static kmutex_t vfs_miplist_mutex
;
127 static struct ipmnt
*vfs_miplist
= NULL
;
128 static struct ipmnt
*vfs_miplist_end
= NULL
;
130 static kmem_cache_t
*vfs_cache
; /* Pointer to VFS kmem cache */
135 vnode_t
*rootdir
; /* pointer to root inode vnode. */
136 vnode_t
*devicesdir
; /* pointer to inode of devices root */
137 vnode_t
*devdir
; /* pointer to inode of dev root */
139 char *server_rootpath
; /* root path for diskless clients */
140 char *server_hostname
; /* hostname of diskless server */
142 static struct vfs root
;
143 static struct vfs devices
;
144 static struct vfs dev
;
145 struct vfs
*rootvfs
= &root
; /* pointer to root vfs; head of VFS list. */
146 rvfs_t
*rvfs_list
; /* array of vfs ptrs for vfs hash list */
147 int vfshsz
= 512; /* # of heads/locks in vfs hash arrays */
148 /* must be power of 2! */
149 timespec_t vfs_mnttab_ctime
; /* mnttab created time */
150 timespec_t vfs_mnttab_mtime
; /* mnttab last modified time */
151 char *vfs_dummyfstype
= "\0";
152 struct pollhead vfs_pollhd
; /* for mnttab pollers */
153 struct vnode
*vfs_mntdummyvp
; /* to fake mnttab read/write for file events */
154 int mntfstype
; /* will be set once mnt fs is mounted */
157 * Table for generic options recognized in the VFS layer and acted
158 * on at this level before parsing file system specific options.
159 * The nosuid option is stronger than any of the devices and setuid
160 * options, so those are canceled when nosuid is seen.
162 * All options which are added here need to be added to the
163 * list of standard options in usr/src/cmd/fs.d/fslib.c as well.
166 * VFS Mount options table
168 static char *ro_cancel
[] = { MNTOPT_RW
, NULL
};
169 static char *rw_cancel
[] = { MNTOPT_RO
, NULL
};
170 static char *suid_cancel
[] = { MNTOPT_NOSUID
, NULL
};
171 static char *nosuid_cancel
[] = { MNTOPT_SUID
, MNTOPT_DEVICES
, MNTOPT_NODEVICES
,
172 MNTOPT_NOSETUID
, MNTOPT_SETUID
, NULL
};
173 static char *devices_cancel
[] = { MNTOPT_NODEVICES
, NULL
};
174 static char *nodevices_cancel
[] = { MNTOPT_DEVICES
, NULL
};
175 static char *setuid_cancel
[] = { MNTOPT_NOSETUID
, NULL
};
176 static char *nosetuid_cancel
[] = { MNTOPT_SETUID
, NULL
};
177 static char *nbmand_cancel
[] = { MNTOPT_NONBMAND
, NULL
};
178 static char *nonbmand_cancel
[] = { MNTOPT_NBMAND
, NULL
};
179 static char *exec_cancel
[] = { MNTOPT_NOEXEC
, NULL
};
180 static char *noexec_cancel
[] = { MNTOPT_EXEC
, NULL
};
182 static const mntopt_t mntopts
[] = {
184 * option name cancel options default arg flags
186 { MNTOPT_REMOUNT
, NULL
, NULL
,
187 MO_NODISPLAY
, (void *)0 },
188 { MNTOPT_RO
, ro_cancel
, NULL
, 0,
190 { MNTOPT_RW
, rw_cancel
, NULL
, 0,
192 { MNTOPT_SUID
, suid_cancel
, NULL
, 0,
194 { MNTOPT_NOSUID
, nosuid_cancel
, NULL
, 0,
196 { MNTOPT_DEVICES
, devices_cancel
, NULL
, 0,
198 { MNTOPT_NODEVICES
, nodevices_cancel
, NULL
, 0,
200 { MNTOPT_SETUID
, setuid_cancel
, NULL
, 0,
202 { MNTOPT_NOSETUID
, nosetuid_cancel
, NULL
, 0,
204 { MNTOPT_NBMAND
, nbmand_cancel
, NULL
, 0,
206 { MNTOPT_NONBMAND
, nonbmand_cancel
, NULL
, 0,
208 { MNTOPT_EXEC
, exec_cancel
, NULL
, 0,
210 { MNTOPT_NOEXEC
, noexec_cancel
, NULL
, 0,
214 const mntopts_t vfs_mntopts
= {
215 sizeof (mntopts
) / sizeof (mntopt_t
),
216 (mntopt_t
*)&mntopts
[0]
220 * File system operation dispatch functions.
224 fsop_mount(vfs_t
*vfsp
, vnode_t
*mvp
, struct mounta
*uap
, cred_t
*cr
)
226 return (*(vfsp
)->vfs_op
->vfs_mount
)(vfsp
, mvp
, uap
, cr
);
230 fsop_unmount(vfs_t
*vfsp
, int flag
, cred_t
*cr
)
232 return (*(vfsp
)->vfs_op
->vfs_unmount
)(vfsp
, flag
, cr
);
236 fsop_root(vfs_t
*vfsp
, vnode_t
**vpp
)
239 int ret
= (*(vfsp
)->vfs_op
->vfs_root
)(vfsp
, vpp
);
241 * Make sure this root has a path. With lofs, it is possible to have
244 if (ret
== 0 && vfsp
->vfs_mntpt
!= NULL
&&
245 (*vpp
)->v_path
== vn_vpath_empty
) {
248 mntpt
= vfs_getmntpoint(vfsp
);
249 path
= refstr_value(mntpt
);
250 vn_setpath_str(*vpp
, path
, strlen(path
));
258 fsop_statfs(vfs_t
*vfsp
, statvfs64_t
*sp
)
260 return (*(vfsp
)->vfs_op
->vfs_statvfs
)(vfsp
, sp
);
264 fsop_sync(vfs_t
*vfsp
, short flag
, cred_t
*cr
)
266 return (*(vfsp
)->vfs_op
->vfs_sync
)(vfsp
, flag
, cr
);
270 fsop_vget(vfs_t
*vfsp
, vnode_t
**vpp
, fid_t
*fidp
)
273 * In order to handle system attribute fids in a manner
274 * transparent to the underlying fs, we embed the fid for
275 * the sysattr parent object in the sysattr fid and tack on
276 * some extra bytes that only the sysattr layer knows about.
278 * This guarantees that sysattr fids are larger than other fids
279 * for this vfs. If the vfs supports the sysattr view interface
280 * (as indicated by VFSFT_SYSATTR_VIEWS), we cannot have a size
281 * collision with XATTR_FIDSZ.
283 if (vfs_has_feature(vfsp
, VFSFT_SYSATTR_VIEWS
) &&
284 fidp
->fid_len
== XATTR_FIDSZ
)
285 return (xattr_dir_vget(vfsp
, vpp
, fidp
));
287 return (*(vfsp
)->vfs_op
->vfs_vget
)(vfsp
, vpp
, fidp
);
291 fsop_mountroot(vfs_t
*vfsp
, enum whymountroot reason
)
293 return (*(vfsp
)->vfs_op
->vfs_mountroot
)(vfsp
, reason
);
297 fsop_freefs(vfs_t
*vfsp
)
299 (*(vfsp
)->vfs_op
->vfs_freevfs
)(vfsp
);
303 fsop_vnstate(vfs_t
*vfsp
, vnode_t
*vp
, vntrans_t nstate
)
305 return ((*(vfsp
)->vfs_op
->vfs_vnstate
)(vfsp
, vp
, nstate
));
309 fsop_sync_by_kind(int fstype
, short flag
, cred_t
*cr
)
311 ASSERT((fstype
>= 0) && (fstype
< nfstype
));
313 if (ALLOCATED_VFSSW(&vfssw
[fstype
]) && VFS_INSTALLED(&vfssw
[fstype
]))
314 return (*vfssw
[fstype
].vsw_vfsops
.vfs_sync
) (NULL
, flag
, cr
);
320 fsop_syncfs(vfs_t
*vfsp
, uint64_t flags
, cred_t
*cr
)
322 return (*(vfsp
)->vfs_op
->vfs_syncfs
)(vfsp
, flags
, cr
);
326 * File system initialization. vfs_setfsops() must be called from a file
327 * system's init routine.
331 fs_copyfsops(const fs_operation_def_t
*template, vfsops_t
*actual
,
334 static const fs_operation_trans_def_t vfs_ops_table
[] = {
335 { VFSNAME_MOUNT
, offsetof(vfsops_t
, vfs_mount
),
336 fs_nosys
, fs_nosys
},
338 { VFSNAME_UNMOUNT
, offsetof(vfsops_t
, vfs_unmount
),
339 fs_nosys
, fs_nosys
},
341 { VFSNAME_ROOT
, offsetof(vfsops_t
, vfs_root
),
342 fs_nosys
, fs_nosys
},
344 { VFSNAME_STATVFS
, offsetof(vfsops_t
, vfs_statvfs
),
345 fs_nosys
, fs_nosys
},
347 { VFSNAME_SYNC
, offsetof(vfsops_t
, vfs_sync
),
348 (fs_generic_func_p
) fs_sync
,
349 (fs_generic_func_p
) fs_sync
}, /* No errors allowed */
351 { VFSNAME_VGET
, offsetof(vfsops_t
, vfs_vget
),
352 fs_nosys
, fs_nosys
},
354 { VFSNAME_MOUNTROOT
, offsetof(vfsops_t
, vfs_mountroot
),
355 fs_nosys
, fs_nosys
},
357 { VFSNAME_FREEVFS
, offsetof(vfsops_t
, vfs_freevfs
),
358 (fs_generic_func_p
)(uintptr_t)fs_freevfs
,
360 (fs_generic_func_p
)(uintptr_t)fs_freevfs
},
362 { VFSNAME_VNSTATE
, offsetof(vfsops_t
, vfs_vnstate
),
363 (fs_generic_func_p
)fs_nosys
, (fs_generic_func_p
)fs_nosys
},
366 * While it is tempting to say that a file system which does not
367 * implement a VFSNAME_SYNC likely doesn't need a VFSNAME_SYNCFS
368 * by default, implementing that policy is challenging with the
369 * way the fs_build_vector logic works and we'd rather a file
370 * system say that it doesn't support this by default rather
371 * than incorrectly claim to sync something that either doesn't
372 * make sense to sync (ala sockfs) or mislead when it didn't
375 { VFSNAME_SYNCFS
, offsetof(vfsops_t
, vfs_syncfs
),
376 (fs_generic_func_p
)fs_nosys_syncfs
,
377 (fs_generic_func_p
)fs_nosys_syncfs
},
379 { NULL
, 0, NULL
, NULL
}
382 return (fs_build_vector(actual
, unused_ops
, vfs_ops_table
, template));
388 if (strcmp(rootfs
.bo_fstype
, MNTTYPE_ZFS
) == 0)
393 vfs_setfsops(int fstype
, const fs_operation_def_t
*template, vfsops_t
**actual
)
399 * Verify that fstype refers to a valid fs. Note that
400 * 0 is valid since it's used to set "stray" ops.
402 if ((fstype
< 0) || (fstype
>= nfstype
))
405 if (!ALLOCATED_VFSSW(&vfssw
[fstype
]))
408 /* Set up the operations vector. */
410 error
= fs_copyfsops(template, &vfssw
[fstype
].vsw_vfsops
, &unused_ops
);
415 vfssw
[fstype
].vsw_flag
|= VSW_INSTALLED
;
418 *actual
= &vfssw
[fstype
].vsw_vfsops
;
422 cmn_err(CE_WARN
, "vfs_setfsops: %s: %d operations supplied "
423 "but not used", vfssw
[fstype
].vsw_name
, unused_ops
);
430 vfs_makefsops(const fs_operation_def_t
*template, vfsops_t
**actual
)
435 *actual
= (vfsops_t
*)kmem_alloc(sizeof (vfsops_t
), KM_SLEEP
);
437 error
= fs_copyfsops(template, *actual
, &unused_ops
);
439 kmem_free(*actual
, sizeof (vfsops_t
));
448 * Free a vfsops structure created as a result of vfs_makefsops().
449 * NOTE: For a vfsops structure initialized by vfs_setfsops(), use
450 * vfs_freevfsops_by_type().
453 vfs_freevfsops(vfsops_t
*vfsops
)
455 kmem_free(vfsops
, sizeof (vfsops_t
));
459 * Since the vfsops structure is part of the vfssw table and wasn't
460 * really allocated, we're not really freeing anything. We keep
461 * the name for consistency with vfs_freevfsops(). We do, however,
462 * need to take care of a little bookkeeping.
463 * NOTE: For a vfsops structure created by vfs_setfsops(), use
464 * vfs_freevfsops_by_type().
467 vfs_freevfsops_by_type(int fstype
)
470 /* Verify that fstype refers to a loaded fs (and not fsid 0). */
471 if ((fstype
<= 0) || (fstype
>= nfstype
))
475 if ((vfssw
[fstype
].vsw_flag
& VSW_INSTALLED
) == 0) {
480 vfssw
[fstype
].vsw_flag
&= ~VSW_INSTALLED
;
486 /* Support routines used to reference vfs_op */
488 /* Set the operations vector for a vfs */
490 vfs_setops(vfs_t
*vfsp
, vfsops_t
*vfsops
)
494 ASSERT(vfsp
!= NULL
);
495 ASSERT(vfsops
!= NULL
);
499 if (vfsp
->vfs_femhead
== NULL
&&
500 atomic_cas_ptr(&vfsp
->vfs_op
, op
, vfsops
) == op
) {
503 fsem_setvfsops(vfsp
, vfsops
);
506 /* Retrieve the operations vector for a vfs */
508 vfs_getops(vfs_t
*vfsp
)
512 ASSERT(vfsp
!= NULL
);
516 if (vfsp
->vfs_femhead
== NULL
&& op
== vfsp
->vfs_op
) {
519 return (fsem_getvfsops(vfsp
));
524 * Returns non-zero (1) if the vfsops matches that of the vfs.
525 * Returns zero (0) if not.
528 vfs_matchops(vfs_t
*vfsp
, vfsops_t
*vfsops
)
530 return (vfs_getops(vfsp
) == vfsops
);
534 * Returns non-zero (1) if the file system has installed a non-default,
535 * non-error vfs_sync routine. Returns zero (0) otherwise.
538 vfs_can_sync(vfs_t
*vfsp
)
540 /* vfs_sync() routine is not the default/error function */
541 return (vfs_getops(vfsp
)->vfs_sync
!= fs_sync
);
545 * Initialize a vfs structure.
548 vfs_init(vfs_t
*vfsp
, vfsops_t
*op
, void *data
)
550 /* Other initialization has been moved to vfs_alloc() */
552 vfsp
->vfs_next
= vfsp
;
553 vfsp
->vfs_prev
= vfsp
;
554 vfsp
->vfs_zone_next
= vfsp
;
555 vfsp
->vfs_zone_prev
= vfsp
;
556 vfsp
->vfs_lofi_id
= 0;
557 sema_init(&vfsp
->vfs_reflock
, 1, NULL
, SEMA_DEFAULT
, NULL
);
559 vfsp
->vfs_data
= (data
);
560 vfs_setops((vfsp
), (op
));
564 * Allocate and initialize the vfs implementation private data
565 * structure, vfs_impl_t.
568 vfsimpl_setup(vfs_t
*vfsp
)
572 if (vfsp
->vfs_implp
!= NULL
) {
576 vfsp
->vfs_implp
= kmem_alloc(sizeof (vfs_impl_t
), KM_SLEEP
);
577 /* Note that these are #define'd in vfs.h */
578 vfsp
->vfs_vskap
= NULL
;
579 vfsp
->vfs_fstypevsp
= NULL
;
581 /* Set size of counted array, then zero the array */
582 vfsp
->vfs_featureset
[0] = VFS_FEATURE_MAXSZ
- 1;
583 for (i
= 1; i
< VFS_FEATURE_MAXSZ
; i
++) {
584 vfsp
->vfs_featureset
[i
] = 0;
589 * Release the vfs_impl_t structure, if it exists. Some unbundled
590 * filesystems may not use the newer version of vfs and thus
591 * would not contain this implementation private data structure.
594 vfsimpl_teardown(vfs_t
*vfsp
)
596 vfs_impl_t
*vip
= vfsp
->vfs_implp
;
601 kmem_free(vfsp
->vfs_implp
, sizeof (vfs_impl_t
));
602 vfsp
->vfs_implp
= NULL
;
606 * VFS system calls: mount, umount, syssync, statfs, fstatfs, statvfs,
607 * fstatvfs, and sysfs moved to common/syscall.
611 * Update every mounted file system. We call the vfs_sync operation of
612 * each file system type, passing it a NULL vfsp to indicate that all
613 * mounted file systems of that type should be updated.
620 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
621 if (ALLOCATED_VFSSW(vswp
) && VFS_INSTALLED(vswp
)) {
624 (void) (*vswp
->vsw_vfsops
.vfs_sync
)(NULL
, flag
,
626 vfs_unrefvfssw(vswp
);
643 krwlock_t vfssw_lock
; /* lock accesses to vfssw */
646 * Lock for accessing the vfs linked list. Initialized in vfs_mountroot(),
647 * but otherwise should be accessed only via vfs_list_lock() and
648 * vfs_list_unlock(). Also used to protect the timestamp for mods to the list.
650 static krwlock_t vfslist
;
653 * Mount devfs on /devices. This is done right after root is mounted
654 * to provide device access support for the system
657 vfs_mountdevices(void)
661 struct mounta mounta
= { /* fake mounta for devfs_mount() */
673 * _init devfs module to fill in the vfssw
675 if (modload("fs", "devfs") == -1)
676 panic("Cannot _init devfs module");
682 vsw
= vfs_getvfsswbyname("devfs");
683 VFS_INIT(&devices
, &vsw
->vsw_vfsops
, NULL
);
689 if (lookupname("/devices", UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
))
690 panic("Cannot find /devices");
693 * Perform the mount of /devices
695 if (VFS_MOUNT(&devices
, mvp
, &mounta
, CRED()))
696 panic("Cannot mount /devices");
701 * Set appropriate members and add to vfs list for mnttab display
703 vfs_setresource(&devices
, "/devices", 0);
704 vfs_setmntpoint(&devices
, "/devices", 0);
707 * Hold the root of /devices so it won't go away
709 if (VFS_ROOT(&devices
, &devicesdir
))
710 panic("vfs_mountdevices: not devices root");
712 if (vfs_lock(&devices
) != 0) {
714 cmn_err(CE_NOTE
, "Cannot acquire vfs_lock of /devices");
718 if (vn_vfswlock(mvp
) != 0) {
719 vfs_unlock(&devices
);
721 cmn_err(CE_NOTE
, "Cannot acquire vfswlock of /devices");
725 vfs_add(mvp
, &devices
, 0);
727 vfs_unlock(&devices
);
732 * mount the first instance of /dev to root and remain mounted
739 struct mounta mounta
= { /* fake mounta for sdev_mount() */
742 MS_SYSSPACE
| MS_OVERLAY
,
751 * _init dev module to fill in the vfssw
753 if (modload("fs", "dev") == -1)
754 cmn_err(CE_PANIC
, "Cannot _init dev module\n");
760 vsw
= vfs_getvfsswbyname("dev");
761 VFS_INIT(&dev
, &vsw
->vsw_vfsops
, NULL
);
767 if (lookupname("/dev", UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
))
768 cmn_err(CE_PANIC
, "Cannot find /dev\n");
771 * Perform the mount of /dev
773 if (VFS_MOUNT(&dev
, mvp
, &mounta
, CRED()))
774 cmn_err(CE_PANIC
, "Cannot mount /dev 1\n");
779 * Set appropriate members and add to vfs list for mnttab display
781 vfs_setresource(&dev
, "/dev", 0);
782 vfs_setmntpoint(&dev
, "/dev", 0);
785 * Hold the root of /dev so it won't go away
787 if (VFS_ROOT(&dev
, &devdir
))
788 cmn_err(CE_PANIC
, "vfs_mountdev1: not dev root");
790 if (vfs_lock(&dev
) != 0) {
792 cmn_err(CE_NOTE
, "Cannot acquire vfs_lock of /dev");
796 if (vn_vfswlock(mvp
) != 0) {
799 cmn_err(CE_NOTE
, "Cannot acquire vfswlock of /dev");
803 vfs_add(mvp
, &dev
, 0);
810 * Mount required filesystem. This is done right after root is mounted.
813 vfs_mountfs(char *module
, char *spec
, char *path
)
816 struct mounta mounta
;
819 bzero(&mounta
, sizeof (mounta
));
820 mounta
.flags
= MS_SYSSPACE
| MS_DATA
;
821 mounta
.fstype
= module
;
824 if (lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &mvp
)) {
825 cmn_err(CE_WARN
, "Cannot find %s", path
);
828 if (domount(NULL
, &mounta
, mvp
, CRED(), &vfsp
))
829 cmn_err(CE_WARN
, "Cannot mount %s", path
);
836 * vfs_mountroot is called by main() to mount the root filesystem.
841 struct vnode
*rvp
= NULL
;
847 rw_init(&vfssw_lock
, NULL
, RW_DEFAULT
, NULL
);
848 rw_init(&vfslist
, NULL
, RW_DEFAULT
, NULL
);
851 * Alloc the vfs hash bucket array and locks
853 rvfs_list
= kmem_zalloc(vfshsz
* sizeof (rvfs_t
), KM_SLEEP
);
856 * Call machine-dependent routine "rootconf" to choose a root
860 panic("vfs_mountroot: cannot mount root");
862 * Get vnode for '/'. Set up rootdir, u.u_rdir and u.u_cdir
863 * to point to it. These are used by lookuppn() so that it
864 * knows where to start from ('/' or '.').
866 vfs_setmntpoint(rootvfs
, "/", 0);
867 if (VFS_ROOT(rootvfs
, &rootdir
))
868 panic("vfs_mountroot: no root vnode");
871 * At this point, the process tree consists of p0 and possibly some
872 * direct children of p0. (i.e. there are no grandchildren)
874 * Walk through them all, setting their current directory.
876 mutex_enter(&pidlock
);
877 for (p
= practive
; p
!= NULL
; p
= p
->p_next
) {
878 ASSERT(p
== &p0
|| p
->p_parent
== &p0
);
880 PTOU(p
)->u_cdir
= rootdir
;
881 VN_HOLD(PTOU(p
)->u_cdir
);
882 PTOU(p
)->u_rdir
= NULL
;
884 mutex_exit(&pidlock
);
887 * Setup the global zone's rootvp, now that it exists.
889 global_zone
->zone_rootvp
= rootdir
;
890 VN_HOLD(global_zone
->zone_rootvp
);
893 * Notify the module code that it can begin using the
894 * root filesystem instead of the boot program's services.
899 * Special handling for a ZFS root file system.
904 * Set up mnttab information for root
906 vfs_setresource(rootvfs
, rootfs
.bo_name
, 0);
909 * Notify cluster software that the root filesystem is available.
913 /* Now that we're all done with the root FS, set up its vopstats */
914 if ((vswp
= vfs_getvfsswbyvfsops(vfs_getops(rootvfs
))) != NULL
) {
915 /* Set flag for statistics collection */
916 if (vswp
->vsw_flag
& VSW_STATS
) {
917 initialize_vopstats(&rootvfs
->vfs_vopstats
);
918 rootvfs
->vfs_flag
|= VFS_STATS
;
919 rootvfs
->vfs_fstypevsp
=
920 get_fstype_vopstats(rootvfs
, vswp
);
921 rootvfs
->vfs_vskap
= get_vskstat_anchor(rootvfs
);
923 vfs_unrefvfssw(vswp
);
927 * Mount /devices, /dev instance 1, /system/contract, /etc/mnttab,
928 * /etc/svc/volatile, /etc/dfs/sharetab, /system/object, and /proc.
933 vfs_mountfs("ctfs", "ctfs", CTFS_ROOT
);
934 vfs_mountfs("proc", "/proc", "/proc");
935 vfs_mountfs("mntfs", "/etc/mnttab", "/etc/mnttab");
936 vfs_mountfs("tmpfs", "/etc/svc/volatile", "/etc/svc/volatile");
937 vfs_mountfs("objfs", "objfs", OBJFS_ROOT
);
938 vfs_mountfs("bootfs", "bootfs", "/system/boot");
940 if (getzoneid() == GLOBAL_ZONEID
) {
941 vfs_mountfs("sharefs", "sharefs", "/etc/dfs/sharetab");
944 if (strcmp(rootfs
.bo_fstype
, "zfs") != 0) {
946 * Look up the root device via devfs so that a dv_node is
947 * created for it. The vnode is never VN_RELE()ed.
948 * We allocate more than MAXPATHLEN so that the
949 * buffer passed to i_ddi_prompath_to_devfspath() is
950 * exactly MAXPATHLEN (the function expects a buffer
953 plen
= strlen("/devices");
954 path
= kmem_alloc(plen
+ MAXPATHLEN
, KM_SLEEP
);
955 (void) strcpy(path
, "/devices");
957 if (i_ddi_prompath_to_devfspath(rootfs
.bo_name
, path
+ plen
)
959 lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, &rvp
)) {
961 /* NUL terminate in case "path" has garbage */
962 path
[plen
+ MAXPATHLEN
- 1] = '\0';
964 cmn_err(CE_WARN
, "!Cannot lookup root device: %s",
968 kmem_free(path
, plen
+ MAXPATHLEN
);
971 vfs_mnttabvp_setup();
975 * Check to see if our "block device" is actually a file. If so,
976 * automatically add a lofi device, and keep track of this fact.
979 lofi_add(const char *fsname
, struct vfs
*vfsp
,
980 mntopts_t
*mntopts
, struct mounta
*uap
)
982 int fromspace
= (uap
->flags
& MS_SYSSPACE
) ?
983 UIO_SYSSPACE
: UIO_USERSPACE
;
984 struct lofi_ioctl
*li
= NULL
;
985 struct vnode
*vp
= NULL
;
986 struct pathname pn
= { NULL
};
988 ldi_handle_t ldi_hdl
;
993 if ((vfssw
= vfs_getvfssw(fsname
)) == NULL
)
996 if (!(vfssw
->vsw_flag
& VSW_CANLOFI
)) {
997 vfs_unrefvfssw(vfssw
);
1001 vfs_unrefvfssw(vfssw
);
1004 if (pn_get(uap
->spec
, fromspace
, &pn
) != 0)
1007 if (lookupname(uap
->spec
, fromspace
, FOLLOW
, NULL
, &vp
) != 0)
1010 if (vp
->v_type
!= VREG
)
1013 /* OK, this is a lofi mount. */
1015 if ((uap
->flags
& (MS_REMOUNT
|MS_GLOBAL
)) ||
1016 vfs_optionisset_nolock(mntopts
, MNTOPT_SUID
, NULL
) ||
1017 vfs_optionisset_nolock(mntopts
, MNTOPT_SETUID
, NULL
) ||
1018 vfs_optionisset_nolock(mntopts
, MNTOPT_DEVICES
, NULL
)) {
1023 ldi_id
= ldi_ident_from_anon();
1024 li
= kmem_zalloc(sizeof (*li
), KM_SLEEP
);
1025 (void) strlcpy(li
->li_filename
, pn
.pn_path
, MAXPATHLEN
);
1027 err
= ldi_open_by_name("/dev/lofictl", FREAD
| FWRITE
, kcred
,
1033 err
= ldi_ioctl(ldi_hdl
, LOFI_MAP_FILE
, (intptr_t)li
,
1034 FREAD
| FWRITE
| FKIOCTL
, kcred
, &id
);
1036 (void) ldi_close(ldi_hdl
, FREAD
| FWRITE
, kcred
);
1039 vfsp
->vfs_lofi_id
= id
;
1042 ldi_ident_release(ldi_id
);
1045 kmem_free(li
, sizeof (*li
));
1053 lofi_remove(struct vfs
*vfsp
)
1055 struct lofi_ioctl
*li
;
1057 ldi_handle_t ldi_hdl
;
1060 if (vfsp
->vfs_lofi_id
== 0)
1063 ldi_id
= ldi_ident_from_anon();
1065 li
= kmem_zalloc(sizeof (*li
), KM_SLEEP
);
1066 li
->li_id
= vfsp
->vfs_lofi_id
;
1067 li
->li_cleanup
= B_TRUE
;
1069 err
= ldi_open_by_name("/dev/lofictl", FREAD
| FWRITE
, kcred
,
1075 err
= ldi_ioctl(ldi_hdl
, LOFI_UNMAP_FILE_MINOR
, (intptr_t)li
,
1076 FREAD
| FWRITE
| FKIOCTL
, kcred
, NULL
);
1078 (void) ldi_close(ldi_hdl
, FREAD
| FWRITE
, kcred
);
1081 vfsp
->vfs_lofi_id
= 0;
1084 ldi_ident_release(ldi_id
);
1085 kmem_free(li
, sizeof (*li
));
1089 * Common mount code. Called from the system call entry point, from autofs,
1090 * nfsv4 trigger mounts, and from pxfs.
1092 * Takes the effective file system type, mount arguments, the mount point
1093 * vnode, flags specifying whether the mount is a remount and whether it
1094 * should be entered into the vfs list, and credentials. Fills in its vfspp
1095 * parameter with the mounted file system instance's vfs.
1097 * Note that the effective file system type is specified as a string. It may
1098 * be null, in which case it's determined from the mount arguments, and may
1099 * differ from the type specified in the mount arguments; this is a hook to
1100 * allow interposition when instantiating file system instances.
1102 * The caller is responsible for releasing its own hold on the mount point
1103 * vp (this routine does its own hold when necessary).
1104 * Also note that for remounts, the mount point vp should be the vnode for
1105 * the root of the file system rather than the vnode that the file system
1106 * is mounted on top of.
1109 domount(char *fsname
, struct mounta
*uap
, vnode_t
*vp
, struct cred
*credp
,
1117 mntopts_t mnt_mntopts
;
1119 int copyout_error
= 0;
1121 char *opts
= uap
->optptr
;
1122 char *inargs
= opts
;
1123 int optlen
= uap
->optlen
;
1129 int splice
= ((uap
->flags
& MS_NOSPLICE
) == 0);
1130 int fromspace
= (uap
->flags
& MS_SYSSPACE
) ?
1131 UIO_SYSSPACE
: UIO_USERSPACE
;
1132 char *resource
= NULL
, *mountpt
= NULL
;
1133 refstr_t
*oldresource
, *oldmntpt
;
1134 struct pathname pn
, rpn
;
1135 vsk_anchor_t
*vskap
;
1136 char fstname
[FSTYPSZ
];
1140 * The v_flag value for the mount point vp is permanently set
1141 * to VVFSLOCK so that no one bypasses the vn_vfs*locks routine
1142 * for mount point locking.
1144 mutex_enter(&vp
->v_lock
);
1145 vp
->v_flag
|= VVFSLOCK
;
1146 mutex_exit(&vp
->v_lock
);
1148 mnt_mntopts
.mo_count
= 0;
1150 * Find the ops vector to use to invoke the file system-specific mount
1151 * method. If the fsname argument is non-NULL, use it directly.
1152 * Otherwise, dig the file system type information out of the mount
1155 * A side effect is to hold the vfssw entry.
1157 * Mount arguments can be specified in several ways, which are
1158 * distinguished by flag bit settings. The preferred way is to set
1159 * MS_OPTIONSTR, indicating an 8 argument mount with the file system
1160 * type supplied as a character string and the last two arguments
1161 * being a pointer to a character buffer and the size of the buffer.
1162 * On entry, the buffer holds a null terminated list of options; on
1163 * return, the string is the list of options the file system
1164 * recognized. If MS_DATA is set arguments five and six point to a
1165 * block of binary data which the file system interprets.
1166 * A further wrinkle is that some callers don't set MS_FSS and MS_DATA
1167 * consistently with these conventions. To handle them, we check to
1168 * see whether the pointer to the file system name has a numeric value
1169 * less than 256. If so, we treat it as an index.
1171 if (fsname
!= NULL
) {
1172 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
) {
1175 } else if (uap
->flags
& (MS_OPTIONSTR
| MS_DATA
| MS_FSS
)) {
1181 if ((fstype
= (uintptr_t)uap
->fstype
) < 256) {
1183 if (fstype
== 0 || fstype
>= nfstype
||
1184 !ALLOCATED_VFSSW(&vfssw
[fstype
])) {
1188 (void) strcpy(fsname
, vfssw
[fstype
].vsw_name
);
1190 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
)
1194 * Handle either kernel or user address space.
1196 if (uap
->flags
& MS_SYSSPACE
) {
1197 error
= copystr(uap
->fstype
, fsname
,
1200 error
= copyinstr(uap
->fstype
, fsname
,
1204 if (error
== ENAMETOOLONG
)
1208 if ((vswp
= vfs_getvfssw(fsname
)) == NULL
)
1212 if ((vswp
= vfs_getvfsswbyvfsops(vfs_getops(rootvfs
))) == NULL
)
1214 fsname
= vswp
->vsw_name
;
1216 if (!VFS_INSTALLED(vswp
))
1219 if ((error
= secpolicy_fs_allowed_mount(fsname
)) != 0) {
1220 vfs_unrefvfssw(vswp
);
1224 vfsops
= &vswp
->vsw_vfsops
;
1226 vfs_copyopttbl(&vswp
->vsw_optproto
, &mnt_mntopts
);
1228 * Fetch mount options and parse them for generic vfs options
1230 if (uap
->flags
& MS_OPTIONSTR
) {
1232 * Limit the buffer size
1234 if (optlen
< 0 || optlen
> MAX_MNTOPT_STR
) {
1238 if ((uap
->flags
& MS_SYSSPACE
) == 0) {
1239 inargs
= kmem_alloc(MAX_MNTOPT_STR
, KM_SLEEP
);
1242 error
= copyinstr(opts
, inargs
, (size_t)optlen
,
1249 vfs_parsemntopts(&mnt_mntopts
, inargs
, 0);
1252 * Flag bits override the options string.
1254 if (uap
->flags
& MS_REMOUNT
)
1255 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_REMOUNT
, NULL
, 0, 0);
1256 if (uap
->flags
& MS_RDONLY
)
1257 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_RO
, NULL
, 0, 0);
1258 if (uap
->flags
& MS_NOSUID
)
1259 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
, 0, 0);
1262 * Check if this is a remount; must be set in the option string and
1263 * the file system must support a remount option.
1265 if (remount
= vfs_optionisset_nolock(&mnt_mntopts
,
1266 MNTOPT_REMOUNT
, NULL
)) {
1267 if (!(vswp
->vsw_flag
& VSW_CANREMOUNT
)) {
1271 uap
->flags
|= MS_REMOUNT
;
1275 * uap->flags and vfs_optionisset() should agree.
1277 if (rdonly
= vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_RO
, NULL
)) {
1278 uap
->flags
|= MS_RDONLY
;
1280 if (vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
)) {
1281 uap
->flags
|= MS_NOSUID
;
1283 nbmand
= vfs_optionisset_nolock(&mnt_mntopts
, MNTOPT_NBMAND
, NULL
);
1284 ASSERT(splice
|| !remount
);
1286 * If we are splicing the fs into the namespace,
1287 * perform mount point checks.
1289 * We want to resolve the path for the mount point to eliminate
1290 * '.' and ".." and symlinks in mount points; we can't do the
1291 * same for the resource string, since it would turn
1292 * "/dev/dsk/c0t0d0s0" into "/devices/pci@...". We need to do
1293 * this before grabbing vn_vfswlock(), because otherwise we
1294 * would deadlock with lookuppn().
1297 ASSERT(vp
->v_count
> 0);
1300 * Pick up mount point and device from appropriate space.
1302 if (pn_get(uap
->spec
, fromspace
, &pn
) == 0) {
1303 resource
= kmem_alloc(pn
.pn_pathlen
+ 1,
1305 (void) strcpy(resource
, pn
.pn_path
);
1309 * Do a lookupname prior to taking the
1310 * writelock. Mark this as completed if
1311 * successful for later cleanup and addition to
1312 * the mount in progress table.
1314 if ((vswp
->vsw_flag
& VSW_MOUNTDEV
) &&
1315 (uap
->flags
& MS_GLOBAL
) == 0 &&
1316 lookupname(uap
->spec
, fromspace
,
1317 FOLLOW
, NULL
, &bvp
) == 0) {
1321 if ((error
= pn_get(uap
->dir
, fromspace
, &pn
)) == 0) {
1324 if (*pn
.pn_path
!= '/') {
1331 * Kludge to prevent autofs from deadlocking with
1332 * itself when it calls domount().
1334 * If autofs is calling, it is because it is doing
1335 * (autofs) mounts in the process of an NFS mount. A
1336 * lookuppn() here would cause us to block waiting for
1337 * said NFS mount to complete, which can't since this
1338 * is the thread that was supposed to doing it.
1340 if (fromspace
== UIO_USERSPACE
) {
1341 if ((error
= lookuppn(&pn
, &rpn
, FOLLOW
, NULL
,
1346 * The file disappeared or otherwise
1347 * became inaccessible since we opened
1348 * it; might as well fail the mount
1349 * since the mount point is no longer
1359 mountpt
= kmem_alloc(pnp
->pn_pathlen
+ 1, KM_SLEEP
);
1360 (void) strcpy(mountpt
, pnp
->pn_path
);
1363 * If the addition of the zone's rootpath
1364 * would push us over a total path length
1365 * of MAXPATHLEN, we fail the mount with
1366 * ENAMETOOLONG, which is what we would have
1367 * gotten if we were trying to perform the same
1368 * mount in the global zone.
1370 * strlen() doesn't count the trailing
1371 * '\0', but zone_rootpathlen counts both a
1372 * trailing '/' and the terminating '\0'.
1374 if ((curproc
->p_zone
->zone_rootpathlen
- 1 +
1375 strlen(mountpt
)) > MAXPATHLEN
||
1376 (resource
!= NULL
&&
1377 (curproc
->p_zone
->zone_rootpathlen
- 1 +
1378 strlen(resource
)) > MAXPATHLEN
)) {
1379 error
= ENAMETOOLONG
;
1390 * Prevent path name resolution from proceeding past
1393 if (vn_vfswlock(vp
) != 0) {
1399 * Verify that it's legitimate to establish a mount on
1400 * the prospective mount point.
1402 if (vn_mountedvfs(vp
) != NULL
) {
1404 * The mount point lock was obtained after some
1405 * other thread raced through and established a mount.
1411 if (vp
->v_flag
& VNOMOUNT
) {
1417 if ((uap
->flags
& (MS_DATA
| MS_OPTIONSTR
)) == 0) {
1418 uap
->dataptr
= NULL
;
1423 * If this is a remount, we don't want to create a new VFS.
1424 * Instead, we pass the existing one with a remount flag.
1428 * Confirm that the mount point is the root vnode of the
1429 * file system that is being remounted.
1430 * This can happen if the user specifies a different
1431 * mount point directory pathname in the (re)mount command.
1433 * Code below can only be reached if splice is true, so it's
1434 * safe to do vn_vfsunlock() here.
1436 if ((vp
->v_flag
& VROOT
) == 0) {
1442 * Disallow making file systems read-only unless file system
1443 * explicitly allows it in its vfssw. Ignore other flags.
1445 if (rdonly
&& vn_is_readonly(vp
) == 0 &&
1446 (vswp
->vsw_flag
& VSW_CANRWRO
) == 0) {
1452 * Disallow changing the NBMAND disposition of the file
1453 * system on remounts.
1455 if ((nbmand
&& ((vp
->v_vfsp
->vfs_flag
& VFS_NBMAND
) == 0)) ||
1456 (!nbmand
&& (vp
->v_vfsp
->vfs_flag
& VFS_NBMAND
))) {
1462 ovflags
= vfsp
->vfs_flag
;
1463 vfsp
->vfs_flag
|= VFS_REMOUNT
;
1464 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
1466 vfsp
= vfs_alloc(KM_SLEEP
);
1467 VFS_INIT(vfsp
, vfsops
, NULL
);
1472 if ((error
= lofi_add(fsname
, vfsp
, &mnt_mntopts
, uap
)) != 0) {
1485 * PRIV_SYS_MOUNT doesn't mean you can become root.
1487 if (vfsp
->vfs_lofi_id
!= 0) {
1488 uap
->flags
|= MS_NOSUID
;
1489 vfs_setmntopt_nolock(&mnt_mntopts
, MNTOPT_NOSUID
, NULL
, 0, 0);
1493 * The vfs_reflock is not used anymore the code below explicitly
1494 * holds it preventing others accesing it directly.
1496 if ((sema_tryp(&vfsp
->vfs_reflock
) == 0) &&
1497 !(vfsp
->vfs_flag
& VFS_REMOUNT
))
1499 "mount type %s couldn't get vfs_reflock", vswp
->vsw_name
);
1502 * Lock the vfs. If this is a remount we want to avoid spurious umount
1503 * failures that happen as a side-effect of fsflush() and other mount
1504 * and unmount operations that might be going on simultaneously and
1505 * may have locked the vfs currently. To not return EBUSY immediately
1506 * here we use vfs_lock_wait() instead vfs_lock() for the remount case.
1509 if (error
= vfs_lock(vfsp
)) {
1518 vfs_lock_wait(vfsp
);
1522 * Add device to mount in progress table, global mounts require special
1523 * handling. It is possible that we have already done the lookupname
1524 * on a spliced, non-global fs. If so, we don't want to do it again
1525 * since we cannot do a lookupname after taking the
1526 * wlock above. This case is for a non-spliced, non-global filesystem.
1529 if ((vswp
->vsw_flag
& VSW_MOUNTDEV
) &&
1530 (uap
->flags
& MS_GLOBAL
) == 0 &&
1531 lookupname(uap
->spec
, fromspace
, FOLLOW
, NULL
, &bvp
) == 0) {
1537 vnode_t
*lvp
= NULL
;
1539 error
= vfs_get_lofi(vfsp
, &lvp
);
1554 } else if (error
== -1) {
1563 vfs_addmip(bdev
, vfsp
);
1568 * Invalidate cached entry for the mount point.
1574 * If have an option string but the filesystem doesn't supply a
1575 * prototype options table, create a table with the global
1576 * options and sufficient room to accept all the options in the
1577 * string. Then parse the passed in option string
1578 * accepting all the options in the string. This gives us an
1579 * option table with all the proper cancel properties for the
1582 * Filesystems that supply a prototype options table are handled
1583 * earlier in this function.
1585 if (uap
->flags
& MS_OPTIONSTR
) {
1586 if (!(vswp
->vsw_flag
& VSW_HASPROTO
)) {
1587 mntopts_t tmp_mntopts
;
1589 tmp_mntopts
.mo_count
= 0;
1590 vfs_createopttbl_extend(&tmp_mntopts
, inargs
,
1592 vfs_parsemntopts(&tmp_mntopts
, inargs
, 1);
1593 vfs_swapopttbl_nolock(&mnt_mntopts
, &tmp_mntopts
);
1594 vfs_freeopttbl(&tmp_mntopts
);
1599 * Serialize with zone state transitions.
1600 * See vfs_list_add; zone mounted into is:
1601 * zone_find_by_path(refstr_value(vfsp->vfs_mntpt))
1602 * not the zone doing the mount (curproc->p_zone), but if we're already
1603 * inside a NGZ, then we know what zone we are.
1605 if (INGLOBALZONE(curproc
)) {
1606 zone
= zone_find_by_path(mountpt
);
1607 ASSERT(zone
!= NULL
);
1609 zone
= curproc
->p_zone
;
1611 * zone_find_by_path does a hold, so do one here too so that
1612 * we can do a zone_rele after mount_completed.
1616 mount_in_progress(zone
);
1618 * Instantiate (or reinstantiate) the file system. If appropriate,
1619 * splice it into the file system name space.
1621 * We want VFS_MOUNT() to be able to override the vfs_resource
1622 * string if necessary (ie, mntfs), and also for a remount to
1623 * change the same (necessary when remounting '/' during boot).
1624 * So we set up vfs_mntpt and vfs_resource to what we think they
1625 * should be, then hand off control to VFS_MOUNT() which can
1628 * For safety's sake, when changing vfs_resource or vfs_mntpt of
1629 * a vfs which is on the vfs list (i.e. during a remount), we must
1630 * never set those fields to NULL. Several bits of code make
1631 * assumptions that the fields are always valid.
1633 vfs_swapopttbl(&mnt_mntopts
, &vfsp
->vfs_mntopts
);
1635 if ((oldresource
= vfsp
->vfs_resource
) != NULL
)
1636 refstr_hold(oldresource
);
1637 if ((oldmntpt
= vfsp
->vfs_mntpt
) != NULL
)
1638 refstr_hold(oldmntpt
);
1640 vfs_setresource(vfsp
, resource
, 0);
1641 vfs_setmntpoint(vfsp
, mountpt
, 0);
1644 * going to mount on this vnode, so notify.
1646 vnevent_mountedover(vp
, NULL
);
1647 error
= VFS_MOUNT(vfsp
, vp
, uap
, credp
);
1649 if (uap
->flags
& MS_RDONLY
)
1650 vfs_setmntopt(vfsp
, MNTOPT_RO
, NULL
, 0);
1651 if (uap
->flags
& MS_NOSUID
)
1652 vfs_setmntopt(vfsp
, MNTOPT_NOSUID
, NULL
, 0);
1653 if (uap
->flags
& MS_GLOBAL
)
1654 vfs_setmntopt(vfsp
, MNTOPT_GLOBAL
, NULL
, 0);
1660 /* put back pre-remount options */
1661 vfs_swapopttbl(&mnt_mntopts
, &vfsp
->vfs_mntopts
);
1662 vfs_setmntpoint(vfsp
, refstr_value(oldmntpt
),
1665 refstr_rele(oldmntpt
);
1666 vfs_setresource(vfsp
, refstr_value(oldresource
),
1669 refstr_rele(oldresource
);
1670 vfsp
->vfs_flag
= ovflags
;
1675 vfs_freemnttab(vfsp
);
1680 * Set the mount time to now
1682 vfsp
->vfs_mtime
= ddi_get_time();
1684 vfsp
->vfs_flag
&= ~VFS_REMOUNT
;
1686 refstr_rele(oldresource
);
1688 refstr_rele(oldmntpt
);
1689 } else if (splice
) {
1691 * Link vfsp into the name space at the mount
1692 * point. Vfs_add() is responsible for
1693 * holding the mount point which will be
1694 * released when vfs_remove() is called.
1696 vfs_add(vp
, vfsp
, uap
->flags
);
1699 * Hold the reference to file system which is
1700 * not linked into the name space.
1702 vfsp
->vfs_zone
= NULL
;
1704 vfsp
->vfs_vnodecovered
= NULL
;
1707 * Set flags for global options encountered
1709 if (vfs_optionisset(vfsp
, MNTOPT_RO
, NULL
))
1710 vfsp
->vfs_flag
|= VFS_RDONLY
;
1712 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
1713 if (vfs_optionisset(vfsp
, MNTOPT_NOSUID
, NULL
)) {
1714 vfsp
->vfs_flag
|= (VFS_NOSETUID
|VFS_NODEVICES
);
1716 if (vfs_optionisset(vfsp
, MNTOPT_NODEVICES
, NULL
))
1717 vfsp
->vfs_flag
|= VFS_NODEVICES
;
1719 vfsp
->vfs_flag
&= ~VFS_NODEVICES
;
1720 if (vfs_optionisset(vfsp
, MNTOPT_NOSETUID
, NULL
))
1721 vfsp
->vfs_flag
|= VFS_NOSETUID
;
1723 vfsp
->vfs_flag
&= ~VFS_NOSETUID
;
1725 if (vfs_optionisset(vfsp
, MNTOPT_NBMAND
, NULL
))
1726 vfsp
->vfs_flag
|= VFS_NBMAND
;
1728 vfsp
->vfs_flag
&= ~VFS_NBMAND
;
1730 if (vfs_optionisset(vfsp
, MNTOPT_XATTR
, NULL
))
1731 vfsp
->vfs_flag
|= VFS_XATTR
;
1733 vfsp
->vfs_flag
&= ~VFS_XATTR
;
1735 if (vfs_optionisset(vfsp
, MNTOPT_NOEXEC
, NULL
))
1736 vfsp
->vfs_flag
|= VFS_NOEXEC
;
1738 vfsp
->vfs_flag
&= ~VFS_NOEXEC
;
1741 * Now construct the output option string of options
1744 if (uap
->flags
& MS_OPTIONSTR
) {
1745 vfs_list_read_lock();
1746 copyout_error
= vfs_buildoptionstr(
1747 &vfsp
->vfs_mntopts
, inargs
, optlen
);
1749 if (copyout_error
== 0 &&
1750 (uap
->flags
& MS_SYSSPACE
) == 0) {
1751 copyout_error
= copyoutstr(inargs
, opts
,
1757 * If this isn't a remount, set up the vopstats before
1758 * anyone can touch this. We only allow spliced file
1759 * systems (file systems which are in the namespace) to
1760 * have the VFS_STATS flag set.
1761 * NOTE: PxFS mounts the underlying file system with
1762 * MS_NOSPLICE set and copies those vfs_flags to its private
1763 * vfs structure. As a result, PxFS should never have
1764 * the VFS_STATS flag or else we might access the vfs
1765 * statistics-related fields prior to them being
1766 * properly initialized.
1768 if (!remount
&& (vswp
->vsw_flag
& VSW_STATS
) && splice
) {
1769 initialize_vopstats(&vfsp
->vfs_vopstats
);
1771 * We need to set vfs_vskap to NULL because there's
1772 * a chance it won't be set below. This is checked
1773 * in teardown_vopstats() so we can't have garbage.
1775 vfsp
->vfs_vskap
= NULL
;
1776 vfsp
->vfs_flag
|= VFS_STATS
;
1777 vfsp
->vfs_fstypevsp
= get_fstype_vopstats(vfsp
, vswp
);
1780 if (vswp
->vsw_flag
& VSW_XID
)
1781 vfsp
->vfs_flag
|= VFS_XID
;
1785 mount_completed(zone
);
1790 if ((error
== 0) && (copyout_error
== 0)) {
1793 * Don't call get_vskstat_anchor() while holding
1794 * locks since it allocates memory and calls
1795 * VFS_STATVFS(). For NFS, the latter can generate
1796 * an over-the-wire call.
1798 vskap
= get_vskstat_anchor(vfsp
);
1799 /* Only take the lock if we have something to do */
1800 if (vskap
!= NULL
) {
1801 vfs_lock_wait(vfsp
);
1802 if (vfsp
->vfs_flag
& VFS_STATS
) {
1803 vfsp
->vfs_vskap
= vskap
;
1808 /* Return vfsp to caller. */
1812 vfs_freeopttbl(&mnt_mntopts
);
1813 if (resource
!= NULL
)
1814 kmem_free(resource
, strlen(resource
) + 1);
1815 if (mountpt
!= NULL
)
1816 kmem_free(mountpt
, strlen(mountpt
) + 1);
1818 * It is possible we errored prior to adding to mount in progress
1819 * table. Must free vnode we acquired with successful lookupname.
1825 ASSERT(vswp
!= NULL
);
1826 vfs_unrefvfssw(vswp
);
1828 kmem_free(inargs
, MAX_MNTOPT_STR
);
1829 if (copyout_error
) {
1832 error
= copyout_error
;
1839 struct vfs
*vfsp
, /* vfs being updated */
1840 refstr_t
**refp
, /* Ref-count string to contain the new path */
1841 const char *newpath
, /* Path to add to refp (above) */
1842 uint32_t flag
) /* flag */
1846 zone_t
*zone
= curproc
->p_zone
;
1848 int have_list_lock
= 0;
1850 ASSERT(!VFS_ON_LIST(vfsp
) || vfs_lock_held(vfsp
));
1853 * New path must be less than MAXPATHLEN because mntfs
1854 * will only display up to MAXPATHLEN bytes. This is currently
1855 * safe, because domount() uses pn_get(), and other callers
1856 * similarly cap the size to fewer than MAXPATHLEN bytes.
1859 ASSERT(strlen(newpath
) < MAXPATHLEN
);
1861 /* mntfs requires consistency while vfs list lock is held */
1863 if (VFS_ON_LIST(vfsp
)) {
1872 * If we are in a non-global zone then we prefix the supplied path,
1873 * newpath, with the zone's root path, with two exceptions. The first
1874 * is where we have been explicitly directed to avoid doing so; this
1875 * will be the case following a failed remount, where the path supplied
1876 * will be a saved version which must now be restored. The second
1877 * exception is where newpath is not a pathname but a descriptive name,
1880 if (zone
== global_zone
|| (flag
& VFSSP_VERBATIM
) || *newpath
!= '/') {
1881 ref
= refstr_alloc(newpath
);
1886 * Truncate the trailing '/' in the zoneroot, and merge
1887 * in the zone's rootpath with the "newpath" (resource
1888 * or mountpoint) passed in.
1890 * The size of the required buffer is thus the size of
1891 * the buffer required for the passed-in newpath
1892 * (strlen(newpath) + 1), plus the size of the buffer
1893 * required to hold zone_rootpath (zone_rootpathlen)
1894 * minus one for one of the now-superfluous NUL
1895 * terminations, minus one for the trailing '/'.
1899 * (strlen(newpath) + 1) + zone_rootpathlen - 1 - 1
1901 * Which is what we have below.
1904 len
= strlen(newpath
) + zone
->zone_rootpathlen
- 1;
1905 sp
= kmem_alloc(len
, KM_SLEEP
);
1908 * Copy everything including the trailing slash, which
1909 * we then overwrite with the NUL character.
1912 (void) strcpy(sp
, zone
->zone_rootpath
);
1913 sp
[zone
->zone_rootpathlen
- 2] = '\0';
1914 (void) strcat(sp
, newpath
);
1916 ref
= refstr_alloc(sp
);
1921 if (have_list_lock
) {
1922 vfs_mnttab_modtimeupd();
1928 * Record a mounted resource name in a vfs structure.
1929 * If vfsp is already mounted, caller must hold the vfs lock.
1932 vfs_setresource(struct vfs
*vfsp
, const char *resource
, uint32_t flag
)
1934 if (resource
== NULL
|| resource
[0] == '\0')
1935 resource
= VFS_NORESOURCE
;
1936 vfs_setpath(vfsp
, &vfsp
->vfs_resource
, resource
, flag
);
1940 * Record a mount point name in a vfs structure.
1941 * If vfsp is already mounted, caller must hold the vfs lock.
1944 vfs_setmntpoint(struct vfs
*vfsp
, const char *mntpt
, uint32_t flag
)
1946 if (mntpt
== NULL
|| mntpt
[0] == '\0')
1947 mntpt
= VFS_NOMNTPT
;
1948 vfs_setpath(vfsp
, &vfsp
->vfs_mntpt
, mntpt
, flag
);
1951 /* Returns the vfs_resource. Caller must call refstr_rele() when finished. */
1954 vfs_getresource(const struct vfs
*vfsp
)
1958 vfs_list_read_lock();
1959 resource
= vfsp
->vfs_resource
;
1960 refstr_hold(resource
);
1966 /* Returns the vfs_mntpt. Caller must call refstr_rele() when finished. */
1969 vfs_getmntpoint(const struct vfs
*vfsp
)
1973 vfs_list_read_lock();
1974 mntpt
= vfsp
->vfs_mntpt
;
1982 * Create an empty options table with enough empty slots to hold all
1983 * The options in the options string passed as an argument.
1984 * Potentially prepend another options table.
1986 * Note: caller is responsible for locking the vfs list, if needed,
1990 vfs_createopttbl_extend(mntopts_t
*mops
, const char *opts
,
1991 const mntopts_t
*mtmpl
)
1993 const char *s
= opts
;
1996 if (opts
== NULL
|| *opts
== '\0') {
2002 * Count number of options in the string
2004 for (s
= strchr(s
, ','); s
!= NULL
; s
= strchr(s
, ',')) {
2009 vfs_copyopttbl_extend(mtmpl
, mops
, count
);
2013 * Create an empty options table with enough empty slots to hold all
2014 * The options in the options string passed as an argument.
2016 * This function is *not* for general use by filesystems.
2018 * Note: caller is responsible for locking the vfs list, if needed,
2022 vfs_createopttbl(mntopts_t
*mops
, const char *opts
)
2024 vfs_createopttbl_extend(mops
, opts
, NULL
);
2029 * Swap two mount options tables
2032 vfs_swapopttbl_nolock(mntopts_t
*optbl1
, mntopts_t
*optbl2
)
2037 tmpcnt
= optbl2
->mo_count
;
2038 tmplist
= optbl2
->mo_list
;
2039 optbl2
->mo_count
= optbl1
->mo_count
;
2040 optbl2
->mo_list
= optbl1
->mo_list
;
2041 optbl1
->mo_count
= tmpcnt
;
2042 optbl1
->mo_list
= tmplist
;
2046 vfs_swapopttbl(mntopts_t
*optbl1
, mntopts_t
*optbl2
)
2049 vfs_swapopttbl_nolock(optbl1
, optbl2
);
2050 vfs_mnttab_modtimeupd();
2055 vfs_copycancelopt_extend(char **const moc
, int extend
)
2062 for (; moc
[i
] != NULL
; i
++)
2063 /* count number of options to cancel */;
2066 if (i
+ extend
== 0)
2069 result
= kmem_alloc((i
+ extend
+ 1) * sizeof (char *), KM_SLEEP
);
2071 for (j
= 0; j
< i
; j
++) {
2072 result
[j
] = kmem_alloc(strlen(moc
[j
]) + 1, KM_SLEEP
);
2073 (void) strcpy(result
[j
], moc
[j
]);
2075 for (; j
<= i
+ extend
; j
++)
2082 vfs_copyopt(const mntopt_t
*s
, mntopt_t
*d
)
2086 d
->mo_flags
= s
->mo_flags
;
2087 d
->mo_data
= s
->mo_data
;
2090 dp
= kmem_alloc(strlen(sp
) + 1, KM_SLEEP
);
2091 (void) strcpy(dp
, sp
);
2094 d
->mo_name
= NULL
; /* should never happen */
2097 d
->mo_cancel
= vfs_copycancelopt_extend(s
->mo_cancel
, 0);
2101 dp
= kmem_alloc(strlen(sp
) + 1, KM_SLEEP
);
2102 (void) strcpy(dp
, sp
);
2110 * Copy a mount options table, possibly allocating some spare
2111 * slots at the end. It is permissible to copy_extend the NULL table.
2114 vfs_copyopttbl_extend(const mntopts_t
*smo
, mntopts_t
*dmo
, int extra
)
2120 * Clear out any existing stuff in the options table being initialized
2122 vfs_freeopttbl(dmo
);
2123 count
= (smo
== NULL
) ? 0 : smo
->mo_count
;
2124 if ((count
+ extra
) == 0) /* nothing to do */
2126 dmo
->mo_count
= count
+ extra
;
2127 motbl
= kmem_zalloc((count
+ extra
) * sizeof (mntopt_t
), KM_SLEEP
);
2128 dmo
->mo_list
= motbl
;
2129 for (i
= 0; i
< count
; i
++) {
2130 vfs_copyopt(&smo
->mo_list
[i
], &motbl
[i
]);
2132 for (i
= count
; i
< count
+ extra
; i
++) {
2133 motbl
[i
].mo_flags
= MO_EMPTY
;
2138 * Copy a mount options table.
2140 * This function is *not* for general use by filesystems.
2142 * Note: caller is responsible for locking the vfs list, if needed,
2143 * to protect smo and dmo.
2146 vfs_copyopttbl(const mntopts_t
*smo
, mntopts_t
*dmo
)
2148 vfs_copyopttbl_extend(smo
, dmo
, 0);
2152 vfs_mergecancelopts(const mntopt_t
*mop1
, const mntopt_t
*mop2
)
2157 char **sp1
, **sp2
, **dp
;
2160 * First we count both lists of cancel options.
2161 * If either is NULL or has no elements, we return a copy of
2164 if (mop1
->mo_cancel
!= NULL
) {
2165 for (; mop1
->mo_cancel
[c1
] != NULL
; c1
++)
2166 /* count cancel options in mop1 */;
2170 return (vfs_copycancelopt_extend(mop2
->mo_cancel
, 0));
2172 if (mop2
->mo_cancel
!= NULL
) {
2173 for (; mop2
->mo_cancel
[c2
] != NULL
; c2
++)
2174 /* count cancel options in mop2 */;
2177 result
= vfs_copycancelopt_extend(mop1
->mo_cancel
, c2
);
2183 * When we get here, we've got two sets of cancel options;
2184 * we need to merge the two sets. We know that the result
2185 * array has "c1+c2+1" entries and in the end we might shrink
2187 * Result now has a copy of the c1 entries from mop1; we'll
2188 * now lookup all the entries of mop2 in mop1 and copy it if
2190 * This operation is O(n^2) but it's only called once per
2191 * filesystem per duplicate option. This is a situation
2192 * which doesn't arise with the filesystems in ON and
2197 for (sp2
= mop2
->mo_cancel
; *sp2
!= NULL
; sp2
++) {
2198 for (sp1
= mop1
->mo_cancel
; *sp1
!= NULL
; sp1
++) {
2199 if (strcmp(*sp1
, *sp2
) == 0)
2204 * Option *sp2 not found in mop1, so copy it.
2205 * The calls to vfs_copycancelopt_extend()
2206 * guarantee that there's enough room.
2208 *dp
= kmem_alloc(strlen(*sp2
) + 1, KM_SLEEP
);
2209 (void) strcpy(*dp
++, *sp2
);
2212 if (dp
!= &result
[c1
+c2
]) {
2213 size_t bytes
= (dp
- result
+ 1) * sizeof (char *);
2214 char **nres
= kmem_alloc(bytes
, KM_SLEEP
);
2216 bcopy(result
, nres
, bytes
);
2217 kmem_free(result
, (c1
+ c2
+ 1) * sizeof (char *));
2224 * Merge two mount option tables (outer and inner) into one. This is very
2225 * similar to "merging" global variables and automatic variables in C.
2227 * This isn't (and doesn't have to be) fast.
2229 * This function is *not* for general use by filesystems.
2231 * Note: caller is responsible for locking the vfs list, if needed,
2232 * to protect omo, imo & dmo.
2235 vfs_mergeopttbl(const mntopts_t
*omo
, const mntopts_t
*imo
, mntopts_t
*dmo
)
2238 mntopt_t
*mop
, *motbl
;
2242 * First determine how much space we need to allocate.
2244 count
= omo
->mo_count
;
2245 for (i
= 0; i
< imo
->mo_count
; i
++) {
2246 if (imo
->mo_list
[i
].mo_flags
& MO_EMPTY
)
2248 if (vfs_hasopt(omo
, imo
->mo_list
[i
].mo_name
) == NULL
)
2251 ASSERT(count
>= omo
->mo_count
&&
2252 count
<= omo
->mo_count
+ imo
->mo_count
);
2253 motbl
= kmem_alloc(count
* sizeof (mntopt_t
), KM_SLEEP
);
2254 for (i
= 0; i
< omo
->mo_count
; i
++)
2255 vfs_copyopt(&omo
->mo_list
[i
], &motbl
[i
]);
2256 freeidx
= omo
->mo_count
;
2257 for (i
= 0; i
< imo
->mo_count
; i
++) {
2258 if (imo
->mo_list
[i
].mo_flags
& MO_EMPTY
)
2260 if ((mop
= vfs_hasopt(omo
, imo
->mo_list
[i
].mo_name
)) != NULL
) {
2262 uint_t index
= mop
- omo
->mo_list
;
2264 newcanp
= vfs_mergecancelopts(mop
, &motbl
[index
]);
2266 vfs_freeopt(&motbl
[index
]);
2267 vfs_copyopt(&imo
->mo_list
[i
], &motbl
[index
]);
2269 vfs_freecancelopt(motbl
[index
].mo_cancel
);
2270 motbl
[index
].mo_cancel
= newcanp
;
2273 * If it's a new option, just copy it over to the first
2276 vfs_copyopt(&imo
->mo_list
[i
], &motbl
[freeidx
++]);
2279 dmo
->mo_count
= count
;
2280 dmo
->mo_list
= motbl
;
2284 * Functions to set and clear mount options in a mount options table.
2288 * Clear a mount option, if it exists.
2290 * The update_mnttab arg indicates whether mops is part of a vfs that is on
2294 vfs_clearmntopt_nolock(mntopts_t
*mops
, const char *opt
, int update_mnttab
)
2299 ASSERT(!update_mnttab
|| RW_WRITE_HELD(&vfslist
));
2301 count
= mops
->mo_count
;
2302 for (i
= 0; i
< count
; i
++) {
2303 mop
= &mops
->mo_list
[i
];
2305 if (mop
->mo_flags
& MO_EMPTY
)
2307 if (strcmp(opt
, mop
->mo_name
))
2309 mop
->mo_flags
&= ~MO_SET
;
2310 if (mop
->mo_arg
!= NULL
) {
2311 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2315 vfs_mnttab_modtimeupd();
2321 vfs_clearmntopt(struct vfs
*vfsp
, const char *opt
)
2325 if (VFS_ON_LIST(vfsp
)) {
2329 vfs_clearmntopt_nolock(&vfsp
->vfs_mntopts
, opt
, gotlock
);
2336 * Set a mount option on. If it's not found in the table, it's silently
2337 * ignored. If the option has MO_IGNORE set, it is still set unless the
2338 * VFS_NOFORCEOPT bit is set in the flags. Also, VFS_DISPLAY/VFS_NODISPLAY flag
2339 * bits can be used to toggle the MO_NODISPLAY bit for the option.
2340 * If the VFS_CREATEOPT flag bit is set then the first option slot with
2341 * MO_EMPTY set is created as the option passed in.
2343 * The update_mnttab arg indicates whether mops is part of a vfs that is on
2347 vfs_setmntopt_nolock(mntopts_t
*mops
, const char *opt
,
2348 const char *arg
, int flags
, int update_mnttab
)
2354 ASSERT(!update_mnttab
|| RW_WRITE_HELD(&vfslist
));
2356 if (flags
& VFS_CREATEOPT
) {
2357 if (vfs_hasopt(mops
, opt
) != NULL
) {
2358 flags
&= ~VFS_CREATEOPT
;
2361 count
= mops
->mo_count
;
2362 for (i
= 0; i
< count
; i
++) {
2363 mop
= &mops
->mo_list
[i
];
2365 if (mop
->mo_flags
& MO_EMPTY
) {
2366 if ((flags
& VFS_CREATEOPT
) == 0)
2368 sp
= kmem_alloc(strlen(opt
) + 1, KM_SLEEP
);
2369 (void) strcpy(sp
, opt
);
2372 mop
->mo_flags
= MO_HASVALUE
;
2375 } else if (strcmp(opt
, mop
->mo_name
)) {
2378 if ((mop
->mo_flags
& MO_IGNORE
) && (flags
& VFS_NOFORCEOPT
))
2380 if (arg
!= NULL
&& (mop
->mo_flags
& MO_HASVALUE
) != 0) {
2381 sp
= kmem_alloc(strlen(arg
) + 1, KM_SLEEP
);
2382 (void) strcpy(sp
, arg
);
2386 if (mop
->mo_arg
!= NULL
)
2387 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2389 if (flags
& VFS_DISPLAY
)
2390 mop
->mo_flags
&= ~MO_NODISPLAY
;
2391 if (flags
& VFS_NODISPLAY
)
2392 mop
->mo_flags
|= MO_NODISPLAY
;
2393 mop
->mo_flags
|= MO_SET
;
2394 if (mop
->mo_cancel
!= NULL
) {
2397 for (cp
= mop
->mo_cancel
; *cp
!= NULL
; cp
++)
2398 vfs_clearmntopt_nolock(mops
, *cp
, 0);
2401 vfs_mnttab_modtimeupd();
2407 vfs_setmntopt(struct vfs
*vfsp
, const char *opt
, const char *arg
, int flags
)
2411 if (VFS_ON_LIST(vfsp
)) {
2415 vfs_setmntopt_nolock(&vfsp
->vfs_mntopts
, opt
, arg
, flags
, gotlock
);
2422 * Add a "tag" option to a mounted file system's options list.
2424 * Note: caller is responsible for locking the vfs list, if needed,
2428 vfs_addtag(mntopts_t
*mops
, const char *tag
)
2431 mntopt_t
*mop
, *motbl
;
2433 count
= mops
->mo_count
+ 1;
2434 motbl
= kmem_zalloc(count
* sizeof (mntopt_t
), KM_SLEEP
);
2435 if (mops
->mo_count
) {
2436 size_t len
= (count
- 1) * sizeof (mntopt_t
);
2438 bcopy(mops
->mo_list
, motbl
, len
);
2439 kmem_free(mops
->mo_list
, len
);
2441 mops
->mo_count
= count
;
2442 mops
->mo_list
= motbl
;
2443 mop
= &motbl
[count
- 1];
2444 mop
->mo_flags
= MO_TAG
;
2445 mop
->mo_name
= kmem_alloc(strlen(tag
) + 1, KM_SLEEP
);
2446 (void) strcpy(mop
->mo_name
, tag
);
2451 * Allow users to set arbitrary "tags" in a vfs's mount options.
2452 * Broader use within the kernel is discouraged.
2455 vfs_settag(uint_t major
, uint_t minor
, const char *mntpt
, const char *tag
,
2462 dev_t dev
= makedevice(major
, minor
);
2464 char *buf
= kmem_alloc(MAX_MNTOPT_STR
, KM_SLEEP
);
2467 * Find the desired mounted file system
2472 if (vfsp
->vfs_dev
== dev
&&
2473 strcmp(mntpt
, refstr_value(vfsp
->vfs_mntpt
)) == 0) {
2477 vfsp
= vfsp
->vfs_next
;
2478 } while (vfsp
!= rootvfs
);
2484 err
= secpolicy_fs_config(cr
, vfsp
);
2488 mops
= &vfsp
->vfs_mntopts
;
2490 * Add tag if it doesn't already exist
2492 if ((mop
= vfs_hasopt(mops
, tag
)) == NULL
) {
2495 (void) vfs_buildoptionstr(mops
, buf
, MAX_MNTOPT_STR
);
2497 if (len
+ strlen(tag
) + 2 > MAX_MNTOPT_STR
) {
2501 mop
= vfs_addtag(mops
, tag
);
2503 if ((mop
->mo_flags
& MO_TAG
) == 0) {
2507 vfs_setmntopt_nolock(mops
, tag
, NULL
, 0, 1);
2510 kmem_free(buf
, MAX_MNTOPT_STR
);
2515 * Allow users to remove arbitrary "tags" in a vfs's mount options.
2516 * Broader use within the kernel is discouraged.
2519 vfs_clrtag(uint_t major
, uint_t minor
, const char *mntpt
, const char *tag
,
2525 dev_t dev
= makedevice(major
, minor
);
2529 * Find the desired mounted file system
2534 if (vfsp
->vfs_dev
== dev
&&
2535 strcmp(mntpt
, refstr_value(vfsp
->vfs_mntpt
)) == 0) {
2539 vfsp
= vfsp
->vfs_next
;
2540 } while (vfsp
!= rootvfs
);
2546 err
= secpolicy_fs_config(cr
, vfsp
);
2550 if ((mop
= vfs_hasopt(&vfsp
->vfs_mntopts
, tag
)) == NULL
) {
2554 if ((mop
->mo_flags
& MO_TAG
) == 0) {
2558 vfs_clearmntopt_nolock(&vfsp
->vfs_mntopts
, tag
, 1);
2565 * Function to parse an option string and fill in a mount options table.
2566 * Unknown options are silently ignored. The input option string is modified
2567 * by replacing separators with nulls. If the create flag is set, options
2568 * not found in the table are just added on the fly. The table must have
2569 * an option slot marked MO_EMPTY to add an option on the fly.
2571 * This function is *not* for general use by filesystems.
2573 * Note: caller is responsible for locking the vfs list, if needed,
2577 vfs_parsemntopts(mntopts_t
*mops
, char *osp
, int create
)
2579 char *s
= osp
, *p
, *nextop
, *valp
, *cp
, *ep
;
2580 int setflg
= VFS_NOFORCEOPT
;
2584 while (*s
!= '\0') {
2585 p
= strchr(s
, ','); /* find next option */
2590 cp
= p
; /* save location of comma */
2591 *p
++ = '\0'; /* mark end and point to next option */
2594 p
= strchr(s
, '='); /* look for value */
2596 valp
= NULL
; /* no value supplied */
2598 ep
= p
; /* save location of equals */
2599 *p
++ = '\0'; /* end option and point to value */
2603 * set option into options table
2606 setflg
|= VFS_CREATEOPT
;
2607 vfs_setmntopt_nolock(mops
, s
, valp
, setflg
, 0);
2609 *cp
= ','; /* restore the comma */
2611 *ep
= '='; /* restore the equals */
2617 * Function to inquire if an option exists in a mount options table.
2618 * Returns a pointer to the option if it exists, else NULL.
2620 * This function is *not* for general use by filesystems.
2622 * Note: caller is responsible for locking the vfs list, if needed,
2626 vfs_hasopt(const mntopts_t
*mops
, const char *opt
)
2631 count
= mops
->mo_count
;
2632 for (i
= 0; i
< count
; i
++) {
2633 mop
= &mops
->mo_list
[i
];
2635 if (mop
->mo_flags
& MO_EMPTY
)
2637 if (strcmp(opt
, mop
->mo_name
) == 0)
2644 * Function to inquire if an option is set in a mount options table.
2645 * Returns non-zero if set and fills in the arg pointer with a pointer to
2646 * the argument string or NULL if there is no argument string.
2649 vfs_optionisset_nolock(const mntopts_t
*mops
, const char *opt
, char **argp
)
2654 count
= mops
->mo_count
;
2655 for (i
= 0; i
< count
; i
++) {
2656 mop
= &mops
->mo_list
[i
];
2658 if (mop
->mo_flags
& MO_EMPTY
)
2660 if (strcmp(opt
, mop
->mo_name
))
2662 if ((mop
->mo_flags
& MO_SET
) == 0)
2664 if (argp
!= NULL
&& (mop
->mo_flags
& MO_HASVALUE
) != 0)
2665 *argp
= mop
->mo_arg
;
2673 vfs_optionisset(const struct vfs
*vfsp
, const char *opt
, char **argp
)
2677 vfs_list_read_lock();
2678 ret
= vfs_optionisset_nolock(&vfsp
->vfs_mntopts
, opt
, argp
);
2685 * Construct a comma separated string of the options set in the given
2686 * mount table, return the string in the given buffer. Return non-zero if
2687 * the buffer would overflow.
2689 * This function is *not* for general use by filesystems.
2691 * Note: caller is responsible for locking the vfs list, if needed,
2695 vfs_buildoptionstr(const mntopts_t
*mp
, char *buf
, int len
)
2702 for (i
= 0; i
< mp
->mo_count
; i
++) {
2705 mop
= &mp
->mo_list
[i
];
2706 if (mop
->mo_flags
& MO_SET
) {
2707 int optlen
, comma
= 0;
2711 optlen
= strlen(mop
->mo_name
);
2712 if (strlen(buf
) + comma
+ optlen
+ 1 > len
)
2716 (void) strcpy(cp
, mop
->mo_name
);
2719 * Append option value if there is one
2721 if (mop
->mo_arg
!= NULL
) {
2724 arglen
= strlen(mop
->mo_arg
);
2725 if (strlen(buf
) + arglen
+ 2 > len
)
2728 (void) strcpy(cp
, mop
->mo_arg
);
2739 vfs_freecancelopt(char **moc
)
2745 for (cp
= moc
; *cp
!= NULL
; cp
++) {
2746 kmem_free(*cp
, strlen(*cp
) + 1);
2749 kmem_free(moc
, (ccnt
+ 1) * sizeof (char *));
2754 vfs_freeopt(mntopt_t
*mop
)
2756 if (mop
->mo_name
!= NULL
)
2757 kmem_free(mop
->mo_name
, strlen(mop
->mo_name
) + 1);
2759 vfs_freecancelopt(mop
->mo_cancel
);
2761 if (mop
->mo_arg
!= NULL
)
2762 kmem_free(mop
->mo_arg
, strlen(mop
->mo_arg
) + 1);
2766 * Free a mount options table
2768 * This function is *not* for general use by filesystems.
2770 * Note: caller is responsible for locking the vfs list, if needed,
2774 vfs_freeopttbl(mntopts_t
*mp
)
2778 count
= mp
->mo_count
;
2779 for (i
= 0; i
< count
; i
++) {
2780 vfs_freeopt(&mp
->mo_list
[i
]);
2783 kmem_free(mp
->mo_list
, sizeof (mntopt_t
) * count
);
2792 vfs_mntdummyread(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cred
,
2793 caller_context_t
*ct
)
2800 vfs_mntdummywrite(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cred
,
2801 caller_context_t
*ct
)
2807 * The dummy vnode is currently used only by file events notification
2808 * module which is just interested in the timestamps.
2812 vfs_mntdummygetattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2813 caller_context_t
*ct
)
2815 bzero(vap
, sizeof (vattr_t
));
2816 vap
->va_type
= VREG
;
2818 vap
->va_ctime
= vfs_mnttab_ctime
;
2820 * it is ok to just copy mtime as the time will be monotonically
2823 vap
->va_mtime
= vfs_mnttab_mtime
;
2824 vap
->va_atime
= vap
->va_mtime
;
2829 vfs_mnttabvp_setup(void)
2832 vnodeops_t
*vfs_mntdummyvnops
;
2833 const fs_operation_def_t mnt_dummyvnodeops_template
[] = {
2834 VOPNAME_READ
, { .vop_read
= vfs_mntdummyread
},
2835 VOPNAME_WRITE
, { .vop_write
= vfs_mntdummywrite
},
2836 VOPNAME_GETATTR
, { .vop_getattr
= vfs_mntdummygetattr
},
2837 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
2841 if (vn_make_ops("mnttab", mnt_dummyvnodeops_template
,
2842 &vfs_mntdummyvnops
) != 0) {
2843 cmn_err(CE_WARN
, "vfs_mnttabvp_setup: vn_make_ops failed");
2844 /* Shouldn't happen, but not bad enough to panic */
2849 * A global dummy vnode is allocated to represent mntfs files.
2850 * The mntfs file (/etc/mnttab) can be monitored for file events
2851 * and receive an event when mnttab changes. Dummy VOP calls
2852 * will be made on this vnode. The file events notification module
2853 * intercepts this vnode and delivers relevant events.
2855 tvp
= vn_alloc(KM_SLEEP
);
2856 tvp
->v_flag
= VNOMOUNT
|VNOMAP
|VNOSWAP
|VNOCACHE
;
2857 vn_setops(tvp
, vfs_mntdummyvnops
);
2860 * The mnt dummy ops do not reference v_data.
2861 * No other module intercepting this vnode should either.
2862 * Just set it to point to itself.
2864 tvp
->v_data
= (caddr_t
)tvp
;
2865 tvp
->v_vfsp
= rootvfs
;
2866 vfs_mntdummyvp
= tvp
;
2870 * performs fake read/write ops
2873 vfs_mnttab_rwop(int rw
)
2879 if (vfs_mntdummyvp
== NULL
)
2882 bzero(&uio
, sizeof (uio
));
2883 bzero(&iov
, sizeof (iov
));
2888 uio
.uio_loffset
= 0;
2889 uio
.uio_segflg
= UIO_SYSSPACE
;
2892 (void) VOP_WRITE(vfs_mntdummyvp
, &uio
, 0, kcred
, NULL
);
2894 (void) VOP_READ(vfs_mntdummyvp
, &uio
, 0, kcred
, NULL
);
2899 * Generate a write operation.
2902 vfs_mnttab_writeop(void)
2908 * Generate a read operation.
2911 vfs_mnttab_readop(void)
2917 * Free any mnttab information recorded in the vfs struct.
2918 * The vfs must not be on the vfs list.
2921 vfs_freemnttab(struct vfs
*vfsp
)
2923 ASSERT(!VFS_ON_LIST(vfsp
));
2926 * Free device and mount point information
2928 if (vfsp
->vfs_mntpt
!= NULL
) {
2929 refstr_rele(vfsp
->vfs_mntpt
);
2930 vfsp
->vfs_mntpt
= NULL
;
2932 if (vfsp
->vfs_resource
!= NULL
) {
2933 refstr_rele(vfsp
->vfs_resource
);
2934 vfsp
->vfs_resource
= NULL
;
2937 * Now free mount options information
2939 vfs_freeopttbl(&vfsp
->vfs_mntopts
);
2943 * Return the last mnttab modification time
2946 vfs_mnttab_modtime(timespec_t
*ts
)
2948 ASSERT(RW_LOCK_HELD(&vfslist
));
2949 *ts
= vfs_mnttab_mtime
;
2953 * See if mnttab is changed
2956 vfs_mnttab_poll(timespec_t
*old
, struct pollhead
**phpp
)
2960 *phpp
= (struct pollhead
*)NULL
;
2963 * Note: don't grab vfs list lock before accessing vfs_mnttab_mtime.
2964 * Can lead to deadlock against vfs_mnttab_modtimeupd(). It is safe
2965 * to not grab the vfs list lock because tv_sec is monotonically
2969 changed
= (old
->tv_nsec
!= vfs_mnttab_mtime
.tv_nsec
) ||
2970 (old
->tv_sec
!= vfs_mnttab_mtime
.tv_sec
);
2972 *phpp
= &vfs_pollhd
;
2976 /* Provide a unique and monotonically-increasing timestamp. */
2978 vfs_mono_time(timespec_t
*ts
)
2980 static volatile hrtime_t hrt
; /* The saved time. */
2981 hrtime_t newhrt
, oldhrt
; /* For effecting the CAS. */
2985 * Try gethrestime() first, but be prepared to fabricate a sensible
2986 * answer at the first sign of any trouble.
2988 gethrestime(&newts
);
2989 newhrt
= ts2hrt(&newts
);
2994 if (atomic_cas_64((uint64_t *)&hrt
, oldhrt
, newhrt
) == oldhrt
)
3001 * Update the mnttab modification time and wake up any waiters for
3005 vfs_mnttab_modtimeupd()
3007 hrtime_t oldhrt
, newhrt
;
3009 ASSERT(RW_WRITE_HELD(&vfslist
));
3010 oldhrt
= ts2hrt(&vfs_mnttab_mtime
);
3011 gethrestime(&vfs_mnttab_mtime
);
3012 newhrt
= ts2hrt(&vfs_mnttab_mtime
);
3013 if (oldhrt
== (hrtime_t
)0)
3014 vfs_mnttab_ctime
= vfs_mnttab_mtime
;
3016 * Attempt to provide unique mtime (like uniqtime but not).
3018 if (newhrt
== oldhrt
) {
3020 hrt2ts(newhrt
, &vfs_mnttab_mtime
);
3022 pollwakeup(&vfs_pollhd
, (short)POLLRDBAND
);
3023 vfs_mnttab_writeop();
3027 dounmount(struct vfs
*vfsp
, int flag
, cred_t
*cr
)
3031 extern void teardown_vopstats(vfs_t
*);
3034 * Get covered vnode. This will be NULL if the vfs is not linked
3035 * into the file system name space (i.e., domount() with MNT_NOSPICE).
3037 coveredvp
= vfsp
->vfs_vnodecovered
;
3038 ASSERT(coveredvp
== NULL
|| vn_vfswlock_held(coveredvp
));
3041 * Purge all dnlc entries for this vfs.
3043 (void) dnlc_purge_vfsp(vfsp
, 0);
3045 /* For forcible umount, skip VFS_SYNC() since it may hang */
3046 if ((flag
& MS_FORCE
) == 0)
3047 (void) VFS_SYNC(vfsp
, 0, cr
);
3050 * Lock the vfs to maintain fs status quo during unmount. This
3051 * has to be done after the sync because ufs_update tries to acquire
3054 vfs_lock_wait(vfsp
);
3056 if (error
= VFS_UNMOUNT(vfsp
, flag
, cr
)) {
3058 if (coveredvp
!= NULL
)
3059 vn_vfsunlock(coveredvp
);
3060 } else if (coveredvp
!= NULL
) {
3061 teardown_vopstats(vfsp
);
3063 * vfs_remove() will do a VN_RELE(vfsp->vfs_vnodecovered)
3064 * when it frees vfsp so we do a VN_HOLD() so we can
3065 * continue to use coveredvp afterwards.
3069 vn_vfsunlock(coveredvp
);
3072 teardown_vopstats(vfsp
);
3074 * Release the reference to vfs that is not linked
3075 * into the name space.
3085 * Vfs_unmountall() is called by uadmin() to unmount all
3086 * mounted file systems (except the root file system) during shutdown.
3087 * It follows the existing locking protocol when traversing the vfs list
3088 * to sync and unmount vfses. Even though there should be no
3089 * other thread running while the system is shutting down, it is prudent
3090 * to still follow the locking protocol.
3093 vfs_unmountall(void)
3096 struct vfs
*prev_vfsp
= NULL
;
3100 * Toss all dnlc entries now so that the per-vfs sync
3101 * and unmount operations don't have to slog through
3102 * a bunch of uninteresting vnodes over and over again.
3107 for (vfsp
= rootvfs
->vfs_prev
; vfsp
!= rootvfs
; vfsp
= prev_vfsp
) {
3108 prev_vfsp
= vfsp
->vfs_prev
;
3110 if (vfs_lock(vfsp
) != 0)
3112 error
= vn_vfswlock(vfsp
->vfs_vnodecovered
);
3119 (void) VFS_SYNC(vfsp
, SYNC_CLOSE
, CRED());
3120 (void) dounmount(vfsp
, 0, CRED());
3123 * Since we dropped the vfslist lock above we must
3124 * verify that next_vfsp still exists, else start over.
3127 for (vfsp
= rootvfs
->vfs_prev
;
3128 vfsp
!= rootvfs
; vfsp
= vfsp
->vfs_prev
)
3129 if (vfsp
== prev_vfsp
)
3131 if (vfsp
== rootvfs
&& prev_vfsp
!= rootvfs
)
3132 prev_vfsp
= rootvfs
->vfs_prev
;
3138 * Called to add an entry to the end of the vfs mount in progress list
3141 vfs_addmip(dev_t dev
, struct vfs
*vfsp
)
3145 mipp
= (struct ipmnt
*)kmem_alloc(sizeof (struct ipmnt
), KM_SLEEP
);
3146 mipp
->mip_next
= NULL
;
3147 mipp
->mip_dev
= dev
;
3148 mipp
->mip_vfsp
= vfsp
;
3149 mutex_enter(&vfs_miplist_mutex
);
3150 if (vfs_miplist_end
!= NULL
)
3151 vfs_miplist_end
->mip_next
= mipp
;
3154 vfs_miplist_end
= mipp
;
3155 mutex_exit(&vfs_miplist_mutex
);
3159 * Called to remove an entry from the mount in progress list
3160 * Either because the mount completed or it failed.
3163 vfs_delmip(struct vfs
*vfsp
)
3165 struct ipmnt
*mipp
, *mipprev
;
3167 mutex_enter(&vfs_miplist_mutex
);
3169 for (mipp
= vfs_miplist
;
3170 mipp
&& mipp
->mip_vfsp
!= vfsp
; mipp
= mipp
->mip_next
) {
3174 return; /* shouldn't happen */
3175 if (mipp
== vfs_miplist_end
)
3176 vfs_miplist_end
= mipprev
;
3177 if (mipprev
== NULL
)
3178 vfs_miplist
= mipp
->mip_next
;
3180 mipprev
->mip_next
= mipp
->mip_next
;
3181 mutex_exit(&vfs_miplist_mutex
);
3182 kmem_free(mipp
, sizeof (struct ipmnt
));
3186 * vfs_add is called by a specific filesystem's mount routine to add
3187 * the new vfs into the vfs list/hash and to cover the mounted-on vnode.
3188 * The vfs should already have been locked by the caller.
3190 * coveredvp is NULL if this is the root.
3193 vfs_add(vnode_t
*coveredvp
, struct vfs
*vfsp
, int mflag
)
3197 ASSERT(vfs_lock_held(vfsp
));
3199 newflag
= vfsp
->vfs_flag
;
3200 if (mflag
& MS_RDONLY
)
3201 newflag
|= VFS_RDONLY
;
3203 newflag
&= ~VFS_RDONLY
;
3204 if (mflag
& MS_NOSUID
)
3205 newflag
|= (VFS_NOSETUID
|VFS_NODEVICES
);
3207 newflag
&= ~(VFS_NOSETUID
|VFS_NODEVICES
);
3208 if (mflag
& MS_NOMNTTAB
)
3209 newflag
|= VFS_NOMNTTAB
;
3211 newflag
&= ~VFS_NOMNTTAB
;
3213 if (coveredvp
!= NULL
) {
3214 ASSERT(vn_vfswlock_held(coveredvp
));
3215 coveredvp
->v_vfsmountedhere
= vfsp
;
3218 vfsp
->vfs_vnodecovered
= coveredvp
;
3219 vfsp
->vfs_flag
= newflag
;
3225 * Remove a vfs from the vfs list, null out the pointer from the
3226 * covered vnode to the vfs (v_vfsmountedhere), and null out the pointer
3227 * from the vfs to the covered vnode (vfs_vnodecovered). Release the
3228 * reference to the vfs and to the covered vnode.
3230 * Called from dounmount after it's confirmed with the file system
3231 * that the unmount is legal.
3234 vfs_remove(struct vfs
*vfsp
)
3238 ASSERT(vfs_lock_held(vfsp
));
3241 * Can't unmount root. Should never happen because fs will
3244 if (vfsp
== rootvfs
)
3245 panic("vfs_remove: unmounting root");
3247 vfs_list_remove(vfsp
);
3250 * Unhook from the file system name space.
3252 vp
= vfsp
->vfs_vnodecovered
;
3253 ASSERT(vn_vfswlock_held(vp
));
3254 vp
->v_vfsmountedhere
= NULL
;
3255 vfsp
->vfs_vnodecovered
= NULL
;
3259 * Release lock and wakeup anybody waiting.
3266 * Lock a filesystem to prevent access to it while mounting,
3267 * unmounting and syncing. Return EBUSY immediately if lock
3268 * can't be acquired.
3271 vfs_lock(vfs_t
*vfsp
)
3273 vn_vfslocks_entry_t
*vpvfsentry
;
3275 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3276 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_WRITER
))
3279 vn_vfslocks_rele(vpvfsentry
);
3284 vfs_rlock(vfs_t
*vfsp
)
3286 vn_vfslocks_entry_t
*vpvfsentry
;
3288 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3290 if (rwst_tryenter(&vpvfsentry
->ve_lock
, RW_READER
))
3293 vn_vfslocks_rele(vpvfsentry
);
3298 vfs_lock_wait(vfs_t
*vfsp
)
3300 vn_vfslocks_entry_t
*vpvfsentry
;
3302 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3303 rwst_enter(&vpvfsentry
->ve_lock
, RW_WRITER
);
3307 vfs_rlock_wait(vfs_t
*vfsp
)
3309 vn_vfslocks_entry_t
*vpvfsentry
;
3311 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3312 rwst_enter(&vpvfsentry
->ve_lock
, RW_READER
);
3316 * Unlock a locked filesystem.
3319 vfs_unlock(vfs_t
*vfsp
)
3321 vn_vfslocks_entry_t
*vpvfsentry
;
3324 * vfs_unlock will mimic sema_v behaviour to fix 4748018.
3325 * And these changes should remain for the patch changes as it is.
3331 * ve_refcount needs to be dropped twice here.
3332 * 1. To release refernce after a call to vfs_locks_getlock()
3333 * 2. To release the reference from the locking routines like
3334 * vfs_rlock_wait/vfs_wlock_wait/vfs_wlock etc,.
3337 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3338 vn_vfslocks_rele(vpvfsentry
);
3340 rwst_exit(&vpvfsentry
->ve_lock
);
3341 vn_vfslocks_rele(vpvfsentry
);
3345 * Utility routine that allows a filesystem to construct its
3346 * fsid in "the usual way" - by munging some underlying dev_t and
3347 * the filesystem type number into the 64-bit fsid. Note that
3348 * this implicitly relies on dev_t persistence to make filesystem
3351 * There's nothing to prevent an individual fs from constructing its
3352 * fsid in a different way, and indeed they should.
3354 * Since we want fsids to be 32-bit quantities (so that they can be
3355 * exported identically by either 32-bit or 64-bit APIs, as well as
3356 * the fact that fsid's are "known" to NFS), we compress the device
3357 * number given down to 32-bits, and panic if that isn't possible.
3360 vfs_make_fsid(fsid_t
*fsi
, dev_t dev
, int val
)
3362 if (!cmpldev((dev32_t
*)&fsi
->val
[0], dev
))
3363 panic("device number too big for fsid!");
3368 vfs_lock_held(vfs_t
*vfsp
)
3371 vn_vfslocks_entry_t
*vpvfsentry
;
3374 * vfs_lock_held will mimic sema_held behaviour
3375 * if panicstr is set. And these changes should remain
3376 * for the patch changes as it is.
3381 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3382 held
= rwst_lock_held(&vpvfsentry
->ve_lock
, RW_WRITER
);
3384 vn_vfslocks_rele(vpvfsentry
);
3389 vfs_lock_owner(vfs_t
*vfsp
)
3391 struct _kthread
*owner
;
3392 vn_vfslocks_entry_t
*vpvfsentry
;
3395 * vfs_wlock_held will mimic sema_held behaviour
3396 * if panicstr is set. And these changes should remain
3397 * for the patch changes as it is.
3402 vpvfsentry
= vn_vfslocks_getlock(vfsp
);
3403 owner
= rwst_owner(&vpvfsentry
->ve_lock
);
3405 vn_vfslocks_rele(vpvfsentry
);
3412 * Rather than manipulate the vfslist lock directly, we abstract into lock
3413 * and unlock routines to allow the locking implementation to be changed for
3416 * Whenever the vfs list is modified through its hash links, the overall list
3417 * lock must be obtained before locking the relevant hash bucket. But to see
3418 * whether a given vfs is on the list, it suffices to obtain the lock for the
3419 * hash bucket without getting the overall list lock. (See getvfs() below.)
3425 rw_enter(&vfslist
, RW_WRITER
);
3429 vfs_list_read_lock()
3431 rw_enter(&vfslist
, RW_READER
);
3441 * Low level worker routines for adding entries to and removing entries from
3446 vfs_hash_add(struct vfs
*vfsp
, int insert_at_head
)
3452 ASSERT(RW_WRITE_HELD(&vfslist
));
3454 dev
= expldev(vfsp
->vfs_fsid
.val
[0]);
3455 vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3457 mutex_enter(&rvfs_list
[vhno
].rvfs_lock
);
3460 * Link into the hash table, inserting it at the end, so that LOFS
3461 * with the same fsid as UFS (or other) file systems will not hide the
3464 if (insert_at_head
) {
3465 vfsp
->vfs_hash
= rvfs_list
[vhno
].rvfs_head
;
3466 rvfs_list
[vhno
].rvfs_head
= vfsp
;
3468 for (hp
= &rvfs_list
[vhno
].rvfs_head
; *hp
!= NULL
;
3469 hp
= &(*hp
)->vfs_hash
)
3472 * hp now contains the address of the pointer to update
3473 * to effect the insertion.
3475 vfsp
->vfs_hash
= NULL
;
3479 rvfs_list
[vhno
].rvfs_len
++;
3480 mutex_exit(&rvfs_list
[vhno
].rvfs_lock
);
3485 vfs_hash_remove(struct vfs
*vfsp
)
3491 ASSERT(RW_WRITE_HELD(&vfslist
));
3493 dev
= expldev(vfsp
->vfs_fsid
.val
[0]);
3494 vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3496 mutex_enter(&rvfs_list
[vhno
].rvfs_lock
);
3501 if (rvfs_list
[vhno
].rvfs_head
== vfsp
) {
3502 rvfs_list
[vhno
].rvfs_head
= vfsp
->vfs_hash
;
3503 rvfs_list
[vhno
].rvfs_len
--;
3506 for (tvfsp
= rvfs_list
[vhno
].rvfs_head
; tvfsp
!= NULL
;
3507 tvfsp
= tvfsp
->vfs_hash
) {
3508 if (tvfsp
->vfs_hash
== vfsp
) {
3509 tvfsp
->vfs_hash
= vfsp
->vfs_hash
;
3510 rvfs_list
[vhno
].rvfs_len
--;
3514 cmn_err(CE_WARN
, "vfs_list_remove: vfs not found in hash");
3518 mutex_exit(&rvfs_list
[vhno
].rvfs_lock
);
3523 vfs_list_add(struct vfs
*vfsp
)
3528 * Typically, the vfs_t will have been created on behalf of the file
3529 * system in vfs_init, where it will have been provided with a
3530 * vfs_impl_t. This, however, might be lacking if the vfs_t was created
3531 * by an unbundled file system. We therefore check for such an example
3532 * before stamping the vfs_t with its creation time for the benefit of
3535 if (vfsp
->vfs_implp
== NULL
)
3536 vfsimpl_setup(vfsp
);
3537 vfs_mono_time(&vfsp
->vfs_hrctime
);
3540 * The zone that owns the mount is the one that performed the mount.
3541 * Note that this isn't necessarily the same as the zone mounted into.
3542 * The corresponding zone_rele_ref() will be done when the vfs_t
3545 vfsp
->vfs_zone
= curproc
->p_zone
;
3546 zone_init_ref(&vfsp
->vfs_implp
->vi_zone_ref
);
3547 zone_hold_ref(vfsp
->vfs_zone
, &vfsp
->vfs_implp
->vi_zone_ref
,
3551 * Find the zone mounted into, and put this mount on its vfs list.
3553 zone
= zone_find_by_path(refstr_value(vfsp
->vfs_mntpt
));
3554 ASSERT(zone
!= NULL
);
3556 * Special casing for the root vfs. This structure is allocated
3557 * statically and hooked onto rootvfs at link time. During the
3558 * vfs_mountroot call at system startup time, the root file system's
3559 * VFS_MOUNTROOT routine will call vfs_add with this root vfs struct
3560 * as argument. The code below must detect and handle this special
3561 * case. The only apparent justification for this special casing is
3562 * to ensure that the root file system appears at the head of the
3565 * XXX: I'm assuming that it's ok to do normal list locking when
3566 * adding the entry for the root file system (this used to be
3567 * done with no locks held).
3571 * Link into the vfs list proper.
3573 if (vfsp
== &root
) {
3575 * Assert: This vfs is already on the list as its first entry.
3576 * Thus, there's nothing to do.
3578 ASSERT(rootvfs
== vfsp
);
3580 * Add it to the head of the global zone's vfslist.
3582 ASSERT(zone
== global_zone
);
3583 ASSERT(zone
->zone_vfslist
== NULL
);
3584 zone
->zone_vfslist
= vfsp
;
3587 * Link to end of list using vfs_prev (as rootvfs is now a
3588 * doubly linked circular list) so list is in mount order for
3591 rootvfs
->vfs_prev
->vfs_next
= vfsp
;
3592 vfsp
->vfs_prev
= rootvfs
->vfs_prev
;
3593 rootvfs
->vfs_prev
= vfsp
;
3594 vfsp
->vfs_next
= rootvfs
;
3597 * Do it again for the zone-private list (which may be NULL).
3599 if (zone
->zone_vfslist
== NULL
) {
3600 ASSERT(zone
!= global_zone
);
3601 zone
->zone_vfslist
= vfsp
;
3603 zone
->zone_vfslist
->vfs_zone_prev
->vfs_zone_next
= vfsp
;
3604 vfsp
->vfs_zone_prev
= zone
->zone_vfslist
->vfs_zone_prev
;
3605 zone
->zone_vfslist
->vfs_zone_prev
= vfsp
;
3606 vfsp
->vfs_zone_next
= zone
->zone_vfslist
;
3611 * Link into the hash table, inserting it at the end, so that LOFS
3612 * with the same fsid as UFS (or other) file systems will not hide
3615 vfs_hash_add(vfsp
, 0);
3618 * update the mnttab modification time
3620 vfs_mnttab_modtimeupd();
3626 vfs_list_remove(struct vfs
*vfsp
)
3630 zone
= zone_find_by_path(refstr_value(vfsp
->vfs_mntpt
));
3631 ASSERT(zone
!= NULL
);
3633 * Callers are responsible for preventing attempts to unmount the
3636 ASSERT(vfsp
!= rootvfs
);
3643 vfs_hash_remove(vfsp
);
3646 * Remove from vfs list.
3648 vfsp
->vfs_prev
->vfs_next
= vfsp
->vfs_next
;
3649 vfsp
->vfs_next
->vfs_prev
= vfsp
->vfs_prev
;
3650 vfsp
->vfs_next
= vfsp
->vfs_prev
= NULL
;
3653 * Remove from zone-specific vfs list.
3655 if (zone
->zone_vfslist
== vfsp
)
3656 zone
->zone_vfslist
= vfsp
->vfs_zone_next
;
3658 if (vfsp
->vfs_zone_next
== vfsp
) {
3659 ASSERT(vfsp
->vfs_zone_prev
== vfsp
);
3660 ASSERT(zone
->zone_vfslist
== vfsp
);
3661 zone
->zone_vfslist
= NULL
;
3664 vfsp
->vfs_zone_prev
->vfs_zone_next
= vfsp
->vfs_zone_next
;
3665 vfsp
->vfs_zone_next
->vfs_zone_prev
= vfsp
->vfs_zone_prev
;
3666 vfsp
->vfs_zone_next
= vfsp
->vfs_zone_prev
= NULL
;
3669 * update the mnttab modification time
3671 vfs_mnttab_modtimeupd();
3677 getvfs(fsid_t
*fsid
)
3680 int val0
= fsid
->val
[0];
3681 int val1
= fsid
->val
[1];
3682 dev_t dev
= expldev(val0
);
3683 int vhno
= VFSHASH(getmajor(dev
), getminor(dev
));
3684 kmutex_t
*hmp
= &rvfs_list
[vhno
].rvfs_lock
;
3687 for (vfsp
= rvfs_list
[vhno
].rvfs_head
; vfsp
; vfsp
= vfsp
->vfs_hash
) {
3688 if (vfsp
->vfs_fsid
.val
[0] == val0
&&
3689 vfsp
->vfs_fsid
.val
[1] == val1
) {
3700 * Search the vfs mount in progress list for a specified device/vfs entry.
3701 * Returns 0 if the first entry in the list that the device matches has the
3702 * given vfs pointer as well. If the device matches but a different vfs
3703 * pointer is encountered in the list before the given vfs pointer then
3708 vfs_devmounting(dev_t dev
, struct vfs
*vfsp
)
3713 mutex_enter(&vfs_miplist_mutex
);
3714 for (mipp
= vfs_miplist
; mipp
!= NULL
; mipp
= mipp
->mip_next
) {
3715 if (mipp
->mip_dev
== dev
) {
3716 if (mipp
->mip_vfsp
!= vfsp
)
3721 mutex_exit(&vfs_miplist_mutex
);
3726 * Search the vfs list for a specified device. Returns 1, if entry is found
3727 * or 0 if no suitable entry is found.
3731 vfs_devismounted(dev_t dev
)
3736 vfs_list_read_lock();
3740 if (vfsp
->vfs_dev
== dev
) {
3744 vfsp
= vfsp
->vfs_next
;
3745 } while (vfsp
!= rootvfs
);
3752 * Search the vfs list for a specified device. Returns a pointer to it
3753 * or NULL if no suitable entry is found. The caller of this routine
3754 * is responsible for releasing the returned vfs pointer.
3757 vfs_dev2vfsp(dev_t dev
)
3762 vfs_list_read_lock();
3767 * The following could be made more efficient by making
3768 * the entire loop use vfs_zone_next if the call is from
3769 * a zone. The only callers, however, ustat(2) and
3770 * umount2(2), don't seem to justify the added
3771 * complexity at present.
3773 if (vfsp
->vfs_dev
== dev
&&
3774 ZONE_PATH_VISIBLE(refstr_value(vfsp
->vfs_mntpt
),
3780 vfsp
= vfsp
->vfs_next
;
3781 } while (vfsp
!= rootvfs
);
3783 return (found
? vfsp
: NULL
);
3787 * Search the vfs list for a specified mntpoint. Returns a pointer to it
3788 * or NULL if no suitable entry is found. The caller of this routine
3789 * is responsible for releasing the returned vfs pointer.
3791 * Note that if multiple mntpoints match, the last one matching is
3792 * returned in an attempt to return the "top" mount when overlay
3793 * mounts are covering the same mount point. This is accomplished by starting
3794 * at the end of the list and working our way backwards, stopping at the first
3798 vfs_mntpoint2vfsp(const char *mp
)
3801 struct vfs
*retvfsp
= NULL
;
3802 zone_t
*zone
= curproc
->p_zone
;
3805 vfs_list_read_lock();
3806 if (getzoneid() == GLOBAL_ZONEID
) {
3808 * The global zone may see filesystems in any zone.
3810 vfsp
= rootvfs
->vfs_prev
;
3812 if (strcmp(refstr_value(vfsp
->vfs_mntpt
), mp
) == 0) {
3816 vfsp
= vfsp
->vfs_prev
;
3817 } while (vfsp
!= rootvfs
->vfs_prev
);
3818 } else if ((list
= zone
->zone_vfslist
) != NULL
) {
3821 vfsp
= list
->vfs_zone_prev
;
3823 mntpt
= refstr_value(vfsp
->vfs_mntpt
);
3824 mntpt
= ZONE_PATH_TRANSLATE(mntpt
, zone
);
3825 if (strcmp(mntpt
, mp
) == 0) {
3829 vfsp
= vfsp
->vfs_zone_prev
;
3830 } while (vfsp
!= list
->vfs_zone_prev
);
3839 * Search the vfs list for a specified vfsops.
3840 * if vfs entry is found then return 1, else 0.
3843 vfs_opsinuse(vfsops_t
*ops
)
3848 vfs_list_read_lock();
3852 if (vfs_getops(vfsp
) == ops
) {
3856 vfsp
= vfsp
->vfs_next
;
3857 } while (vfsp
!= rootvfs
);
3863 * Allocate an entry in vfssw for a file system type
3866 allocate_vfssw(const char *type
)
3870 if (type
[0] == '\0' || strlen(type
) + 1 > _ST_FSTYPSZ
) {
3872 * The vfssw table uses the empty string to identify an
3873 * available entry; we cannot add any type which has
3874 * a leading NUL. The string length is limited to
3875 * the size of the st_fstype array in struct stat.
3880 ASSERT(VFSSW_WRITE_LOCKED());
3881 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++)
3882 if (!ALLOCATED_VFSSW(vswp
)) {
3883 vswp
->vsw_name
= kmem_alloc(strlen(type
) + 1, KM_SLEEP
);
3884 (void) strcpy(vswp
->vsw_name
, type
);
3885 ASSERT(vswp
->vsw_count
== 0);
3886 vswp
->vsw_count
= 1;
3887 mutex_init(&vswp
->vsw_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
3894 * Impose additional layer of translation between vfstype names
3895 * and module names in the filesystem.
3898 vfs_to_modname(const char *vfstype
)
3900 if (strcmp(vfstype
, "proc") == 0) {
3902 } else if (strcmp(vfstype
, "fd") == 0) {
3904 } else if (strncmp(vfstype
, "nfs", 3) == 0) {
3912 * Find a vfssw entry given a file system type name.
3913 * Try to autoload the filesystem if it's not found.
3914 * If it's installed, return the vfssw locked to prevent unloading.
3917 vfs_getvfssw(const char *type
)
3920 const char *modname
;
3923 vswp
= vfs_getvfsswbyname(type
);
3924 modname
= vfs_to_modname(type
);
3926 if (rootdir
== NULL
) {
3928 * If we haven't yet loaded the root file system, then our
3929 * _init won't be called until later. Allocate vfssw entry,
3930 * because mod_installfs won't be called.
3935 if ((vswp
= vfs_getvfsswbyname(type
)) == NULL
) {
3936 if ((vswp
= allocate_vfssw(type
)) == NULL
) {
3944 if (!VFS_INSTALLED(vswp
)) {
3946 (void) modloadonly("fs", modname
);
3953 * Try to load the filesystem. Before calling modload(), we drop
3954 * our lock on the VFS switch table, and pick it up after the
3955 * module is loaded. However, there is a potential race: the
3956 * module could be unloaded after the call to modload() completes
3957 * but before we pick up the lock and drive on. Therefore,
3958 * we keep reloading the module until we've loaded the module
3959 * _and_ we have the lock on the VFS switch table.
3961 while (vswp
== NULL
|| !VFS_INSTALLED(vswp
)) {
3963 if (modload("fs", modname
) == -1)
3967 if ((vswp
= vfs_getvfsswbyname(type
)) == NULL
)
3976 * Find a vfssw entry given a file system type name.
3979 vfs_getvfsswbyname(const char *type
)
3983 ASSERT(VFSSW_LOCKED());
3984 if (type
== NULL
|| *type
== '\0')
3987 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
3988 if (strcmp(type
, vswp
->vsw_name
) == 0) {
3998 * Find a vfssw entry given a set of vfsops.
4001 vfs_getvfsswbyvfsops(vfsops_t
*vfsops
)
4006 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
4007 if (ALLOCATED_VFSSW(vswp
) && &vswp
->vsw_vfsops
== vfsops
) {
4019 * Reference a vfssw entry.
4022 vfs_refvfssw(struct vfssw
*vswp
)
4025 mutex_enter(&vswp
->vsw_lock
);
4027 mutex_exit(&vswp
->vsw_lock
);
4031 * Unreference a vfssw entry.
4034 vfs_unrefvfssw(struct vfssw
*vswp
)
4037 mutex_enter(&vswp
->vsw_lock
);
4039 mutex_exit(&vswp
->vsw_lock
);
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 the uadmin A_SHUTDOWN code. It should only
4053 * be used after some higher-level mechanism has quiesced the system so that
4054 * new writes are not being initiated while we are waiting for completion.
4056 * To ensure finite running time, our algorithm uses sync_triesleft (a progress
4057 * counter used by the vfs_syncall() loop below). It is declared above so
4058 * it can be found easily in the debugger.
4060 * The sync_triesleft counter is updated by vfs_syncall() itself. If we make
4061 * sync_retries consecutive calls to bio_busy() and page_busy() without
4062 * decreasing either the number of dirty buffers or dirty pages below the
4063 * lowest count we have seen so far, we give up and return from vfs_syncall().
4065 * Each loop iteration ends with a call to delay() one second to allow time for
4066 * i/o completion and to permit the user time to read our progress messages.
4071 if (rootdir
== NULL
&& !modrootloaded
)
4072 return; /* no filesystems have been loaded yet */
4074 printf("syncing file systems...");
4077 sync_triesleft
= sync_retries
;
4079 old_bufcnt
= new_bufcnt
= INT_MAX
;
4080 old_pgcnt
= new_pgcnt
= ULONG_MAX
;
4082 while (sync_triesleft
> 0) {
4083 old_bufcnt
= MIN(old_bufcnt
, new_bufcnt
);
4084 old_pgcnt
= MIN(old_pgcnt
, new_pgcnt
);
4086 new_bufcnt
= bio_busy(B_TRUE
);
4087 new_pgcnt
= page_busy(B_TRUE
);
4089 if (new_bufcnt
== 0 && new_pgcnt
== 0)
4092 if (new_bufcnt
< old_bufcnt
|| new_pgcnt
< old_pgcnt
)
4093 sync_triesleft
= sync_retries
;
4098 printf(" [%d]", new_bufcnt
);
4100 printf(" %lu", new_pgcnt
);
4105 if (new_bufcnt
!= 0 || new_pgcnt
!= 0)
4106 printf(" done (not all i/o completed)\n");
4114 * Map VFS flags to statvfs flags. These shouldn't really be separate
4118 vf_to_stf(uint_t vf
)
4122 if (vf
& VFS_RDONLY
)
4124 if (vf
& VFS_NOSETUID
)
4126 if (vf
& VFS_NOTRUNC
)
4133 * Entries for (illegal) fstype 0.
4137 vfsstray_sync(struct vfs
*vfsp
, short arg
, struct cred
*cr
)
4139 cmn_err(CE_PANIC
, "stray vfs operation");
4144 * Entries for (illegal) fstype 0.
4149 cmn_err(CE_PANIC
, "stray vfs operation");
4154 * Support for dealing with forced UFS unmount and its interaction with
4155 * LOFS. Could be used by any filesystem.
4165 * We've gotta define the op for sync separately, since the compiler gets
4166 * confused if we mix and match ANSI and normal style prototypes when
4167 * a "short" argument is present and spits out a warning.
4171 vfs_EIO_sync(struct vfs
*vfsp
, short arg
, struct cred
*cr
)
4177 vfsops_t
*EIO_vfsops
;
4180 * Called from startup() to initialize all loaded vfs's
4187 extern int vopstats_enabled
;
4188 extern void vopstats_startup();
4190 static const fs_operation_def_t EIO_vfsops_template
[] = {
4191 VFSNAME_MOUNT
, { .error
= vfs_EIO
},
4192 VFSNAME_UNMOUNT
, { .error
= vfs_EIO
},
4193 VFSNAME_ROOT
, { .error
= vfs_EIO
},
4194 VFSNAME_STATVFS
, { .error
= vfs_EIO
},
4195 VFSNAME_SYNC
, { .vfs_sync
= vfs_EIO_sync
},
4196 VFSNAME_VGET
, { .error
= vfs_EIO
},
4197 VFSNAME_MOUNTROOT
, { .error
= vfs_EIO
},
4198 VFSNAME_FREEVFS
, { .error
= vfs_EIO
},
4199 VFSNAME_VNSTATE
, { .error
= vfs_EIO
},
4200 VFSNAME_SYNCFS
, { .error
= vfs_EIO
},
4204 static const fs_operation_def_t stray_vfsops_template
[] = {
4205 VFSNAME_MOUNT
, { .error
= vfsstray
},
4206 VFSNAME_UNMOUNT
, { .error
= vfsstray
},
4207 VFSNAME_ROOT
, { .error
= vfsstray
},
4208 VFSNAME_STATVFS
, { .error
= vfsstray
},
4209 VFSNAME_SYNC
, { .vfs_sync
= vfsstray_sync
},
4210 VFSNAME_VGET
, { .error
= vfsstray
},
4211 VFSNAME_MOUNTROOT
, { .error
= vfsstray
},
4212 VFSNAME_FREEVFS
, { .error
= vfsstray
},
4213 VFSNAME_VNSTATE
, { .error
= vfsstray
},
4214 VFSNAME_SYNCFS
, { .error
= vfsstray
},
4218 /* Create vfs cache */
4219 vfs_cache
= kmem_cache_create("vfs_cache", sizeof (struct vfs
),
4220 sizeof (uintptr_t), NULL
, NULL
, NULL
, NULL
, NULL
, 0);
4222 /* Initialize the vnode cache (file systems may use it during init). */
4225 /* Setup event monitor framework */
4228 /* Initialize the dummy stray file system type. */
4229 error
= vfs_setfsops(0, stray_vfsops_template
, NULL
);
4231 /* Initialize the dummy EIO file system. */
4232 error
= vfs_makefsops(EIO_vfsops_template
, &EIO_vfsops
);
4234 cmn_err(CE_WARN
, "vfsinit: bad EIO vfs ops template");
4235 /* Shouldn't happen, but not bad enough to panic */
4238 VFS_INIT(&EIO_vfs
, EIO_vfsops
, (caddr_t
)NULL
);
4241 * Default EIO_vfs.vfs_flag to VFS_UNMOUNTED so a lookup
4242 * on this vfs can immediately notice it's invalid.
4244 EIO_vfs
.vfs_flag
|= VFS_UNMOUNTED
;
4247 * Call the init routines of non-loadable filesystems only.
4248 * Filesystems which are loaded as separate modules will be
4249 * initialized by the module loading code instead.
4252 for (vswp
= &vfssw
[1]; vswp
< &vfssw
[nfstype
]; vswp
++) {
4254 if (vswp
->vsw_init
!= NULL
)
4255 (void) (*vswp
->vsw_init
)(vswp
- vfssw
, vswp
->vsw_name
);
4261 if (vopstats_enabled
) {
4262 /* EIO_vfs can collect stats, but we don't retrieve them */
4263 initialize_vopstats(&EIO_vfs
.vfs_vopstats
);
4264 EIO_vfs
.vfs_fstypevsp
= NULL
;
4265 EIO_vfs
.vfs_vskap
= NULL
;
4266 EIO_vfs
.vfs_flag
|= VFS_STATS
;
4271 reparse_point_init();
4275 vfs_alloc(int kmflag
)
4279 vfsp
= kmem_cache_alloc(vfs_cache
, kmflag
);
4282 * Do the simplest initialization here.
4283 * Everything else gets done in vfs_init()
4285 bzero(vfsp
, sizeof (vfs_t
));
4290 vfs_free(vfs_t
*vfsp
)
4293 * One would be tempted to assert that "vfsp->vfs_count == 0".
4294 * The problem is that this gets called out of domount() with
4295 * a partially initialized vfs and a vfs_count of 1. This is
4296 * also called from vfs_rele() with a vfs_count of 0. We can't
4297 * call VFS_RELE() from domount() if VFS_MOUNT() hasn't successfully
4298 * returned. This is because VFS_MOUNT() fully initializes the
4299 * vfs structure and its associated data. VFS_RELE() will call
4300 * VFS_FREEVFS() which may panic the system if the data structures
4301 * aren't fully initialized from a successful VFS_MOUNT()).
4304 /* If FEM was in use, make sure everything gets cleaned up */
4305 if (vfsp
->vfs_femhead
) {
4306 ASSERT(vfsp
->vfs_femhead
->femh_list
== NULL
);
4307 mutex_destroy(&vfsp
->vfs_femhead
->femh_lock
);
4308 kmem_free(vfsp
->vfs_femhead
, sizeof (*(vfsp
->vfs_femhead
)));
4309 vfsp
->vfs_femhead
= NULL
;
4312 if (vfsp
->vfs_implp
)
4313 vfsimpl_teardown(vfsp
);
4314 sema_destroy(&vfsp
->vfs_reflock
);
4315 kmem_cache_free(vfs_cache
, vfsp
);
4319 * Increments the vfs reference count by one atomically.
4322 vfs_hold(vfs_t
*vfsp
)
4324 atomic_inc_32(&vfsp
->vfs_count
);
4325 ASSERT(vfsp
->vfs_count
!= 0);
4329 * Decrements the vfs reference count by one atomically. When
4330 * vfs reference count becomes zero, it calls the file system
4331 * specific vfs_freevfs() to free up the resources.
4334 vfs_rele(vfs_t
*vfsp
)
4336 ASSERT(vfsp
->vfs_count
!= 0);
4337 if (atomic_dec_32_nv(&vfsp
->vfs_count
) == 0) {
4341 zone_rele_ref(&vfsp
->vfs_implp
->vi_zone_ref
,
4343 vfs_freemnttab(vfsp
);
4349 * Generic operations vector support.
4351 * This is used to build operations vectors for both the vfs and vnode.
4352 * It's normally called only when a file system is loaded.
4354 * There are many possible algorithms for this, including the following:
4356 * (1) scan the list of known operations; for each, see if the file system
4357 * includes an entry for it, and fill it in as appropriate.
4359 * (2) set up defaults for all known operations. scan the list of ops
4360 * supplied by the file system; for each which is both supplied and
4361 * known, fill it in.
4363 * (3) sort the lists of known ops & supplied ops; scan the list, filling
4364 * in entries as we go.
4366 * we choose (1) for simplicity, and because performance isn't critical here.
4367 * note that (2) could be sped up using a precomputed hash table on known ops.
4368 * (3) could be faster than either, but only if the lists were very large or
4369 * supplied in sorted order.
4374 fs_build_vector(void *vector
, int *unused_ops
,
4375 const fs_operation_trans_def_t
*translation
,
4376 const fs_operation_def_t
*operations
)
4378 int i
, num_trans
, num_ops
, used
;
4381 * Count the number of translations and the number of supplied
4386 const fs_operation_trans_def_t
*p
;
4388 for (num_trans
= 0, p
= translation
;
4395 const fs_operation_def_t
*p
;
4397 for (num_ops
= 0, p
= operations
;
4403 /* Walk through each operation known to our caller. There will be */
4404 /* one entry in the supplied "translation table" for each. */
4408 for (i
= 0; i
< num_trans
; i
++) {
4411 fs_generic_func_p result
;
4412 fs_generic_func_p
*location
;
4414 curname
= translation
[i
].name
;
4416 /* Look for a matching operation in the list supplied by the */
4421 for (j
= 0; j
< num_ops
; j
++) {
4422 if (strcmp(operations
[j
].name
, curname
) == 0) {
4430 * If the file system is using a "placeholder" for default
4431 * or error functions, grab the appropriate function out of
4432 * the translation table. If the file system didn't supply
4433 * this operation at all, use the default function.
4437 result
= operations
[j
].func
.fs_generic
;
4438 if (result
== fs_default
) {
4439 result
= translation
[i
].defaultFunc
;
4440 } else if (result
== fs_error
) {
4441 result
= translation
[i
].errorFunc
;
4442 } else if (result
== NULL
) {
4443 /* Null values are PROHIBITED */
4447 result
= translation
[i
].defaultFunc
;
4450 /* Now store the function into the operations vector. */
4452 location
= (fs_generic_func_p
*)
4453 (((char *)vector
) + translation
[i
].offset
);
4458 *unused_ops
= num_ops
- used
;
4463 /* Placeholder functions, should never be called. */
4468 cmn_err(CE_PANIC
, "fs_error called");
4475 cmn_err(CE_PANIC
, "fs_default called");
4482 * Part of the implementation of booting off a mirrored root
4483 * involves a change of dev_t for the root device. To
4484 * accomplish this, first remove the existing hash table
4485 * entry for the root device, convert to the new dev_t,
4486 * then re-insert in the hash table at the head of the list.
4489 vfs_root_redev(vfs_t
*vfsp
, dev_t ndev
, int fstype
)
4493 vfs_hash_remove(vfsp
);
4495 vfsp
->vfs_dev
= ndev
;
4496 vfs_make_fsid(&vfsp
->vfs_fsid
, ndev
, fstype
);
4498 vfs_hash_add(vfsp
, 1);
4503 #else /* x86 NEWBOOT */
4506 extern int hvmboot_rootconf();
4509 extern ib_boot_prop_t
*iscsiboot_prop
;
4516 extern void pm_init();
4517 char *fstyp
, *fsmod
;
4520 getrootfs(&fstyp
, &fsmod
);
4524 * hvmboot_rootconf() is defined in the hvm_bootstrap misc module,
4525 * which lives in /platform/i86hvm, and hence is only available when
4526 * booted in an x86 hvm environment. If the hvm_bootstrap misc module
4527 * is not available then the modstub for this function will return 0.
4528 * If the hvm_bootstrap misc module is available it will be loaded
4529 * and hvmboot_rootconf() will be invoked.
4531 if (error
= hvmboot_rootconf())
4535 if (error
= clboot_rootconf())
4538 if (modload("fs", fsmod
) == -1)
4539 panic("Cannot _init %s module", fsmod
);
4542 vsw
= vfs_getvfsswbyname(fstyp
);
4545 cmn_err(CE_CONT
, "Cannot find %s filesystem\n", fstyp
);
4548 VFS_INIT(rootvfs
, &vsw
->vsw_vfsops
, 0);
4551 /* always mount readonly first */
4552 rootvfs
->vfs_flag
|= VFS_RDONLY
;
4556 if (netboot
&& iscsiboot_prop
) {
4557 cmn_err(CE_WARN
, "NFS boot and iSCSI boot"
4558 " shouldn't happen in the same time");
4562 if (netboot
|| iscsiboot_prop
) {
4565 cmn_err(CE_WARN
, "Cannot plumb network device %d", ret
);
4570 if ((ret
== 0) && iscsiboot_prop
) {
4571 ret
= modload("drv", "iscsi");
4572 /* -1 indicates fail */
4574 cmn_err(CE_WARN
, "Failed to load iscsi module");
4575 iscsi_boot_prop_free();
4578 if (!i_ddi_attach_pseudo_node("iscsi")) {
4580 "Failed to attach iscsi driver");
4581 iscsi_boot_prop_free();
4587 error
= VFS_MOUNTROOT(rootvfs
, ROOT_INIT
);
4588 vfs_unrefvfssw(vsw
);
4589 rootdev
= rootvfs
->vfs_dev
;
4592 cmn_err(CE_CONT
, "Cannot mount root on %s fstype %s\n",
4593 rootfs
.bo_name
, fstyp
);
4595 cmn_err(CE_CONT
, "?root on %s fstype %s\n",
4596 rootfs
.bo_name
, fstyp
);
4601 * XXX this is called by nfs only and should probably be removed
4602 * If booted with ASKNAME, prompt on the console for a filesystem
4603 * name and return it.
4606 getfsname(char *askfor
, char *name
, size_t namelen
)
4608 if (boothowto
& RB_ASKNAME
) {
4609 printf("%s name: ", askfor
);
4610 console_gets(name
, namelen
);
4615 * Init the root filesystem type (rootfs.bo_fstype) from the "fstype"
4618 * Filesystem types starting with the prefix "nfs" are diskless clients;
4619 * init the root filename name (rootfs.bo_name), too.
4621 * If we are booting via NFS we currently have these options:
4622 * nfs - dynamically choose NFS V2, V3, or V4 (default)
4623 * nfs2 - force NFS V2
4624 * nfs3 - force NFS V3
4625 * nfs4 - force NFS V4
4626 * Because we need to maintain backward compatibility with the naming
4627 * convention that the NFS V2 filesystem name is "nfs" (see vfs_conf.c)
4628 * we need to map "nfs" => "nfsdyn" and "nfs2" => "nfs". The dynamic
4629 * nfs module will map the type back to either "nfs", "nfs3", or "nfs4".
4630 * This is only for root filesystems, all other uses will expect
4631 * that "nfs" == NFS V2.
4634 getrootfs(char **fstypp
, char **fsmodp
)
4636 char *propstr
= NULL
;
4639 * Check fstype property; for diskless it should be one of "nfs",
4640 * "nfs2", "nfs3" or "nfs4".
4642 if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4643 DDI_PROP_DONTPASS
, "fstype", &propstr
)
4645 (void) strncpy(rootfs
.bo_fstype
, propstr
, BO_MAXFSNAME
);
4646 ddi_prop_free(propstr
);
4649 * if the boot property 'fstype' is not set, but 'zfs-bootfs' is set,
4650 * assume the type of this root filesystem is 'zfs'.
4652 } else if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4653 DDI_PROP_DONTPASS
, "zfs-bootfs", &propstr
)
4655 (void) strncpy(rootfs
.bo_fstype
, "zfs", BO_MAXFSNAME
);
4656 ddi_prop_free(propstr
);
4659 if (strncmp(rootfs
.bo_fstype
, "nfs", 3) != 0) {
4660 *fstypp
= *fsmodp
= rootfs
.bo_fstype
;
4666 if (strcmp(rootfs
.bo_fstype
, "nfs2") == 0)
4667 (void) strcpy(rootfs
.bo_fstype
, "nfs");
4668 else if (strcmp(rootfs
.bo_fstype
, "nfs") == 0)
4669 (void) strcpy(rootfs
.bo_fstype
, "nfsdyn");
4672 * check if path to network interface is specified in bootpath
4673 * or by a hypervisor domain configuration file.
4674 * XXPV - enable strlumb_get_netdev_path()
4676 if (ddi_prop_exists(DDI_DEV_T_ANY
, ddi_root_node(), DDI_PROP_DONTPASS
,
4678 (void) strcpy(rootfs
.bo_name
, "/xpvd/xnf@0");
4679 } else if (ddi_prop_lookup_string(DDI_DEV_T_ANY
, ddi_root_node(),
4680 DDI_PROP_DONTPASS
, "bootpath", &propstr
)
4682 (void) strncpy(rootfs
.bo_name
, propstr
, BO_MAXOBJNAME
);
4683 ddi_prop_free(propstr
);
4685 rootfs
.bo_name
[0] = '\0';
4687 *fstypp
= rootfs
.bo_fstype
;
4693 * VFS feature routines
4696 #define VFTINDEX(feature) (((feature) >> 32) & 0xFFFFFFFF)
4697 #define VFTBITS(feature) ((feature) & 0xFFFFFFFFLL)
4699 /* Register a feature in the vfs */
4701 vfs_set_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4703 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4704 if (vfsp
->vfs_implp
== NULL
)
4707 vfsp
->vfs_featureset
[VFTINDEX(feature
)] |= VFTBITS(feature
);
4711 vfs_clear_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4713 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4714 if (vfsp
->vfs_implp
== NULL
)
4716 vfsp
->vfs_featureset
[VFTINDEX(feature
)] &= VFTBITS(~feature
);
4720 * Query a vfs for a feature.
4721 * Returns 1 if feature is present, 0 if not
4724 vfs_has_feature(vfs_t
*vfsp
, vfs_feature_t feature
)
4728 /* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4729 if (vfsp
->vfs_implp
== NULL
)
4732 if (vfsp
->vfs_featureset
[VFTINDEX(feature
)] & VFTBITS(feature
))
4739 * Propagate feature set from one vfs to another
4742 vfs_propagate_features(vfs_t
*from
, vfs_t
*to
)
4746 if (to
->vfs_implp
== NULL
|| from
->vfs_implp
== NULL
)
4749 for (i
= 1; i
<= to
->vfs_featureset
[0]; i
++) {
4750 to
->vfs_featureset
[i
] = from
->vfs_featureset
[i
];
4754 #define LOFINODE_PATH "/dev/lofi/%d"
4757 * Return the vnode for the lofi node if there's a lofi mount in place.
4758 * Returns -1 when there's no lofi node, 0 on success, and > 0 on
4762 vfs_get_lofi(vfs_t
*vfsp
, vnode_t
**vpp
)
4768 if (vfsp
->vfs_lofi_id
== 0) {
4773 strsize
= snprintf(NULL
, 0, LOFINODE_PATH
, vfsp
->vfs_lofi_id
);
4774 path
= kmem_alloc(strsize
+ 1, KM_SLEEP
);
4775 (void) snprintf(path
, strsize
+ 1, LOFINODE_PATH
, vfsp
->vfs_lofi_id
);
4778 * We may be inside a zone, so we need to use the /dev path, but
4779 * it's created asynchronously, so we wait here.
4782 err
= lookupname(path
, UIO_SYSSPACE
, FOLLOW
, NULLVPP
, vpp
);
4787 if ((err
= delay_sig(hz
/ 8)) == EINTR
)
4794 kmem_free(path
, strsize
+ 1);