1 /*-
2 * Copyright (c) 1989, 1991, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_quota.h"
36 #include "opt_ufs.h"
37 #include "opt_ffs.h"
38 #include "opt_ddb.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/kernel.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/ioccom.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
55 #include <sys/rwlock.h>
56
57 #include <security/mac/mac_framework.h>
58
59 #include <ufs/ufs/extattr.h>
60 #include <ufs/ufs/gjournal.h>
61 #include <ufs/ufs/quota.h>
62 #include <ufs/ufs/ufsmount.h>
63 #include <ufs/ufs/inode.h>
64 #include <ufs/ufs/ufs_extern.h>
65
66 #include <ufs/ffs/fs.h>
67 #include <ufs/ffs/ffs_extern.h>
68
69 #include <vm/vm.h>
70 #include <vm/uma.h>
71 #include <vm/vm_page.h>
72
73 #include <geom/geom.h>
74 #include <geom/geom_vfs.h>
75
76 #include <ddb/ddb.h>
77
78 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
79
80 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *);
81 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *,
82 ufs2_daddr_t);
83 static void ffs_ifree(struct ufsmount *ump, struct inode *ip);
84 static int ffs_sync_lazy(struct mount *mp);
85
86 static vfs_init_t ffs_init;
87 static vfs_uninit_t ffs_uninit;
88 static vfs_extattrctl_t ffs_extattrctl;
89 static vfs_cmount_t ffs_cmount;
90 static vfs_unmount_t ffs_unmount;
91 static vfs_mount_t ffs_mount;
92 static vfs_statfs_t ffs_statfs;
93 static vfs_fhtovp_t ffs_fhtovp;
94 static vfs_sync_t ffs_sync;
95
96 static struct vfsops ufs_vfsops = {
97 .vfs_extattrctl = ffs_extattrctl,
98 .vfs_fhtovp = ffs_fhtovp,
99 .vfs_init = ffs_init,
100 .vfs_mount = ffs_mount,
101 .vfs_cmount = ffs_cmount,
102 .vfs_quotactl = ufs_quotactl,
103 .vfs_root = ufs_root,
104 .vfs_statfs = ffs_statfs,
105 .vfs_sync = ffs_sync,
106 .vfs_uninit = ffs_uninit,
107 .vfs_unmount = ffs_unmount,
108 .vfs_vget = ffs_vget,
109 .vfs_susp_clean = process_deferred_inactive,
110 };
111
112 VFS_SET(ufs_vfsops, ufs, 0);
113 MODULE_VERSION(ufs, 1);
114
115 static b_strategy_t ffs_geom_strategy;
116 static b_write_t ffs_bufwrite;
117
118 static struct buf_ops ffs_ops = {
119 .bop_name = "FFS",
120 .bop_write = ffs_bufwrite,
121 .bop_strategy = ffs_geom_strategy,
122 .bop_sync = bufsync,
123 #ifdef NO_FFS_SNAPSHOT
124 .bop_bdflush = bufbdflush,
125 #else
126 .bop_bdflush = ffs_bdflush,
127 #endif
128 };
129
130 /*
131 * Note that userquota and groupquota options are not currently used
132 * by UFS/FFS code and generally mount(8) does not pass those options
133 * from userland, but they can be passed by loader(8) via
134 * vfs.root.mountfrom.options.
135 */
136 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
137 "noclusterw", "noexec", "export", "force", "from", "groupquota",
138 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
139 "nosymfollow", "sync", "union", "userquota", NULL };
140
141 static int
ffs_mount(struct mount * mp)142 ffs_mount(struct mount *mp)
143 {
144 struct vnode *devvp;
145 struct thread *td;
146 struct ufsmount *ump = NULL;
147 struct fs *fs;
148 pid_t fsckpid = 0;
149 int error, flags;
150 uint64_t mntorflags;
151 accmode_t accmode;
152 struct nameidata ndp;
153 char *fspec;
154
155 td = curthread;
156 if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
157 return (EINVAL);
158 if (uma_inode == NULL) {
159 uma_inode = uma_zcreate("FFS inode",
160 sizeof(struct inode), NULL, NULL, NULL, NULL,
161 UMA_ALIGN_PTR, 0);
162 uma_ufs1 = uma_zcreate("FFS1 dinode",
163 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
164 UMA_ALIGN_PTR, 0);
165 uma_ufs2 = uma_zcreate("FFS2 dinode",
166 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
167 UMA_ALIGN_PTR, 0);
168 }
169
170 vfs_deleteopt(mp->mnt_optnew, "groupquota");
171 vfs_deleteopt(mp->mnt_optnew, "userquota");
172
173 fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
174 if (error)
175 return (error);
176
177 mntorflags = 0;
178 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
179 mntorflags |= MNT_ACLS;
180
181 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
182 mntorflags |= MNT_SNAPSHOT;
183 /*
184 * Once we have set the MNT_SNAPSHOT flag, do not
185 * persist "snapshot" in the options list.
186 */
187 vfs_deleteopt(mp->mnt_optnew, "snapshot");
188 vfs_deleteopt(mp->mnt_opt, "snapshot");
189 }
190
191 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
192 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
193 /*
194 * Once we have set the restricted PID, do not
195 * persist "fsckpid" in the options list.
196 */
197 vfs_deleteopt(mp->mnt_optnew, "fsckpid");
198 vfs_deleteopt(mp->mnt_opt, "fsckpid");
199 if (mp->mnt_flag & MNT_UPDATE) {
200 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
201 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
202 vfs_mount_error(mp,
203 "Checker enable: Must be read-only");
204 return (EINVAL);
205 }
206 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
207 vfs_mount_error(mp,
208 "Checker enable: Must be read-only");
209 return (EINVAL);
210 }
211 /* Set to -1 if we are done */
212 if (fsckpid == 0)
213 fsckpid = -1;
214 }
215
216 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
217 if (mntorflags & MNT_ACLS) {
218 vfs_mount_error(mp,
219 "\"acls\" and \"nfsv4acls\" options "
220 "are mutually exclusive");
221 return (EINVAL);
222 }
223 mntorflags |= MNT_NFS4ACLS;
224 }
225
226 MNT_ILOCK(mp);
227 mp->mnt_flag |= mntorflags;
228 MNT_IUNLOCK(mp);
229 /*
230 * If updating, check whether changing from read-only to
231 * read/write; if there is no device name, that's all we do.
232 */
233 if (mp->mnt_flag & MNT_UPDATE) {
234 ump = VFSTOUFS(mp);
235 fs = ump->um_fs;
236 devvp = ump->um_devvp;
237 if (fsckpid == -1 && ump->um_fsckpid > 0) {
238 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
239 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
240 return (error);
241 DROP_GIANT();
242 g_topology_lock();
243 /*
244 * Return to normal read-only mode.
245 */
246 error = g_access(ump->um_cp, 0, -1, 0);
247 g_topology_unlock();
248 PICKUP_GIANT();
249 ump->um_fsckpid = 0;
250 }
251 if (fs->fs_ronly == 0 &&
252 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
253 /*
254 * Flush any dirty data and suspend filesystem.
255 */
256 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
257 return (error);
258 error = vfs_write_suspend_umnt(mp);
259 if (error != 0)
260 return (error);
261 /*
262 * Check for and optionally get rid of files open
263 * for writing.
264 */
265 flags = WRITECLOSE;
266 if (mp->mnt_flag & MNT_FORCE)
267 flags |= FORCECLOSE;
268 if (MOUNTEDSOFTDEP(mp)) {
269 error = softdep_flushfiles(mp, flags, td);
270 } else {
271 error = ffs_flushfiles(mp, flags, td);
272 }
273 if (error) {
274 vfs_write_resume(mp, 0);
275 return (error);
276 }
277 if (fs->fs_pendingblocks != 0 ||
278 fs->fs_pendinginodes != 0) {
279 printf("WARNING: %s Update error: blocks %jd "
280 "files %d\n", fs->fs_fsmnt,
281 (intmax_t)fs->fs_pendingblocks,
282 fs->fs_pendinginodes);
283 fs->fs_pendingblocks = 0;
284 fs->fs_pendinginodes = 0;
285 }
286 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
287 fs->fs_clean = 1;
288 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
289 fs->fs_ronly = 0;
290 fs->fs_clean = 0;
291 vfs_write_resume(mp, 0);
292 return (error);
293 }
294 if (MOUNTEDSOFTDEP(mp))
295 softdep_unmount(mp);
296 DROP_GIANT();
297 g_topology_lock();
298 /*
299 * Drop our write and exclusive access.
300 */
301 g_access(ump->um_cp, 0, -1, -1);
302 g_topology_unlock();
303 PICKUP_GIANT();
304 fs->fs_ronly = 1;
305 MNT_ILOCK(mp);
306 mp->mnt_flag |= MNT_RDONLY;
307 MNT_IUNLOCK(mp);
308 /*
309 * Allow the writers to note that filesystem
310 * is ro now.
311 */
312 vfs_write_resume(mp, 0);
313 }
314 if ((mp->mnt_flag & MNT_RELOAD) &&
315 (error = ffs_reload(mp, td, 0)) != 0)
316 return (error);
317 if (fs->fs_ronly &&
318 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
319 /*
320 * If we are running a checker, do not allow upgrade.
321 */
322 if (ump->um_fsckpid > 0) {
323 vfs_mount_error(mp,
324 "Active checker, cannot upgrade to write");
325 return (EINVAL);
326 }
327 /*
328 * If upgrade to read-write by non-root, then verify
329 * that user has necessary permissions on the device.
330 */
331 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
332 error = VOP_ACCESS(devvp, VREAD | VWRITE,
333 td->td_ucred, td);
334 if (error)
335 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
336 if (error) {
337 VOP_UNLOCK(devvp, 0);
338 return (error);
339 }
340 VOP_UNLOCK(devvp, 0);
341 fs->fs_flags &= ~FS_UNCLEAN;
342 if (fs->fs_clean == 0) {
343 fs->fs_flags |= FS_UNCLEAN;
344 if ((mp->mnt_flag & MNT_FORCE) ||
345 ((fs->fs_flags &
346 (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
347 (fs->fs_flags & FS_DOSOFTDEP))) {
348 printf("WARNING: %s was not properly "
349 "dismounted\n", fs->fs_fsmnt);
350 } else {
351 vfs_mount_error(mp,
352 "R/W mount of %s denied. %s.%s",
353 fs->fs_fsmnt,
354 "Filesystem is not clean - run fsck",
355 (fs->fs_flags & FS_SUJ) == 0 ? "" :
356 " Forced mount will invalidate"
357 " journal contents");
358 return (EPERM);
359 }
360 }
361 DROP_GIANT();
362 g_topology_lock();
363 /*
364 * Request exclusive write access.
365 */
366 error = g_access(ump->um_cp, 0, 1, 1);
367 g_topology_unlock();
368 PICKUP_GIANT();
369 if (error)
370 return (error);
371 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
372 return (error);
373 fs->fs_ronly = 0;
374 MNT_ILOCK(mp);
375 mp->mnt_flag &= ~MNT_RDONLY;
376 MNT_IUNLOCK(mp);
377 fs->fs_mtime = time_second;
378 /* check to see if we need to start softdep */
379 if ((fs->fs_flags & FS_DOSOFTDEP) &&
380 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
381 vn_finished_write(mp);
382 return (error);
383 }
384 fs->fs_clean = 0;
385 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
386 vn_finished_write(mp);
387 return (error);
388 }
389 if (fs->fs_snapinum[0] != 0)
390 ffs_snapshot_mount(mp);
391 vn_finished_write(mp);
392 }
393 /*
394 * Soft updates is incompatible with "async",
395 * so if we are doing softupdates stop the user
396 * from setting the async flag in an update.
397 * Softdep_mount() clears it in an initial mount
398 * or ro->rw remount.
399 */
400 if (MOUNTEDSOFTDEP(mp)) {
401 /* XXX: Reset too late ? */
402 MNT_ILOCK(mp);
403 mp->mnt_flag &= ~MNT_ASYNC;
404 MNT_IUNLOCK(mp);
405 }
406 /*
407 * Keep MNT_ACLS flag if it is stored in superblock.
408 */
409 if ((fs->fs_flags & FS_ACLS) != 0) {
410 /* XXX: Set too late ? */
411 MNT_ILOCK(mp);
412 mp->mnt_flag |= MNT_ACLS;
413 MNT_IUNLOCK(mp);
414 }
415
416 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
417 /* XXX: Set too late ? */
418 MNT_ILOCK(mp);
419 mp->mnt_flag |= MNT_NFS4ACLS;
420 MNT_IUNLOCK(mp);
421 }
422 /*
423 * If this is a request from fsck to clean up the filesystem,
424 * then allow the specified pid to proceed.
425 */
426 if (fsckpid > 0) {
427 if (ump->um_fsckpid != 0) {
428 vfs_mount_error(mp,
429 "Active checker already running on %s",
430 fs->fs_fsmnt);
431 return (EINVAL);
432 }
433 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
434 ("soft updates enabled on read-only file system"));
435 DROP_GIANT();
436 g_topology_lock();
437 /*
438 * Request write access.
439 */
440 error = g_access(ump->um_cp, 0, 1, 0);
441 g_topology_unlock();
442 PICKUP_GIANT();
443 if (error) {
444 vfs_mount_error(mp,
445 "Checker activation failed on %s",
446 fs->fs_fsmnt);
447 return (error);
448 }
449 ump->um_fsckpid = fsckpid;
450 if (fs->fs_snapinum[0] != 0)
451 ffs_snapshot_mount(mp);
452 fs->fs_mtime = time_second;
453 fs->fs_fmod = 1;
454 fs->fs_clean = 0;
455 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
456 }
457
458 /*
459 * If this is a snapshot request, take the snapshot.
460 */
461 if (mp->mnt_flag & MNT_SNAPSHOT)
462 return (ffs_snapshot(mp, fspec));
463 }
464
465 /*
466 * Not an update, or updating the name: look up the name
467 * and verify that it refers to a sensible disk device.
468 */
469 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
470 if ((error = namei(&ndp)) != 0)
471 return (error);
472 NDFREE(&ndp, NDF_ONLY_PNBUF);
473 devvp = ndp.ni_vp;
474 if (!vn_isdisk(devvp, &error)) {
475 vput(devvp);
476 return (error);
477 }
478
479 /*
480 * If mount by non-root, then verify that user has necessary
481 * permissions on the device.
482 */
483 accmode = VREAD;
484 if ((mp->mnt_flag & MNT_RDONLY) == 0)
485 accmode |= VWRITE;
486 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
487 if (error)
488 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
489 if (error) {
490 vput(devvp);
491 return (error);
492 }
493
494 if (mp->mnt_flag & MNT_UPDATE) {
495 /*
496 * Update only
497 *
498 * If it's not the same vnode, or at least the same device
499 * then it's not correct.
500 */
501
502 if (devvp->v_rdev != ump->um_devvp->v_rdev)
503 error = EINVAL; /* needs translation */
504 vput(devvp);
505 if (error)
506 return (error);
507 } else {
508 /*
509 * New mount
510 *
511 * We need the name for the mount point (also used for
512 * "last mounted on") copied in. If an error occurs,
513 * the mount point is discarded by the upper level code.
514 * Note that vfs_mount() populates f_mntonname for us.
515 */
516 if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
517 vrele(devvp);
518 return (error);
519 }
520 if (fsckpid > 0) {
521 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
522 ("soft updates enabled on read-only file system"));
523 ump = VFSTOUFS(mp);
524 fs = ump->um_fs;
525 DROP_GIANT();
526 g_topology_lock();
527 /*
528 * Request write access.
529 */
530 error = g_access(ump->um_cp, 0, 1, 0);
531 g_topology_unlock();
532 PICKUP_GIANT();
533 if (error) {
534 printf("WARNING: %s: Checker activation "
535 "failed\n", fs->fs_fsmnt);
536 } else {
537 ump->um_fsckpid = fsckpid;
538 if (fs->fs_snapinum[0] != 0)
539 ffs_snapshot_mount(mp);
540 fs->fs_mtime = time_second;
541 fs->fs_clean = 0;
542 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
543 }
544 }
545 }
546 vfs_mountedfrom(mp, fspec);
547 return (0);
548 }
549
550 /*
551 * Compatibility with old mount system call.
552 */
553
554 static int
ffs_cmount(struct mntarg * ma,void * data,uint64_t flags)555 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
556 {
557 struct ufs_args args;
558 struct export_args exp;
559 int error;
560
561 if (data == NULL)
562 return (EINVAL);
563 error = copyin(data, &args, sizeof args);
564 if (error)
565 return (error);
566 vfs_oexport_conv(&args.export, &exp);
567
568 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
569 ma = mount_arg(ma, "export", &exp, sizeof(exp));
570 error = kernel_mount(ma, flags);
571
572 return (error);
573 }
574
575 /*
576 * Reload all incore data for a filesystem (used after running fsck on
577 * the root filesystem and finding things to fix). If the 'force' flag
578 * is 0, the filesystem must be mounted read-only.
579 *
580 * Things to do to update the mount:
581 * 1) invalidate all cached meta-data.
582 * 2) re-read superblock from disk.
583 * 3) re-read summary information from disk.
584 * 4) invalidate all inactive vnodes.
585 * 5) invalidate all cached file data.
586 * 6) re-read inode data for all active vnodes.
587 */
588 int
ffs_reload(struct mount * mp,struct thread * td,int force)589 ffs_reload(struct mount *mp, struct thread *td, int force)
590 {
591 struct vnode *vp, *mvp, *devvp;
592 struct inode *ip;
593 void *space;
594 struct buf *bp;
595 struct fs *fs, *newfs;
596 struct ufsmount *ump;
597 ufs2_daddr_t sblockloc;
598 int i, blks, size, error;
599 int32_t *lp;
600
601 ump = VFSTOUFS(mp);
602
603 MNT_ILOCK(mp);
604 if ((mp->mnt_flag & MNT_RDONLY) == 0 && force == 0) {
605 MNT_IUNLOCK(mp);
606 return (EINVAL);
607 }
608 MNT_IUNLOCK(mp);
609
610 /*
611 * Step 1: invalidate all cached meta-data.
612 */
613 devvp = VFSTOUFS(mp)->um_devvp;
614 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
615 if (vinvalbuf(devvp, 0, 0, 0) != 0)
616 panic("ffs_reload: dirty1");
617 VOP_UNLOCK(devvp, 0);
618
619 /*
620 * Step 2: re-read superblock from disk.
621 */
622 fs = VFSTOUFS(mp)->um_fs;
623 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
624 NOCRED, &bp)) != 0)
625 return (error);
626 newfs = (struct fs *)bp->b_data;
627 if ((newfs->fs_magic != FS_UFS1_MAGIC &&
628 newfs->fs_magic != FS_UFS2_MAGIC) ||
629 newfs->fs_bsize > MAXBSIZE ||
630 newfs->fs_bsize < sizeof(struct fs)) {
631 brelse(bp);
632 return (EIO); /* XXX needs translation */
633 }
634 /*
635 * Copy pointer fields back into superblock before copying in XXX
636 * new superblock. These should really be in the ufsmount. XXX
637 * Note that important parameters (eg fs_ncg) are unchanged.
638 */
639 newfs->fs_csp = fs->fs_csp;
640 newfs->fs_maxcluster = fs->fs_maxcluster;
641 newfs->fs_contigdirs = fs->fs_contigdirs;
642 newfs->fs_active = fs->fs_active;
643 newfs->fs_ronly = fs->fs_ronly;
644 sblockloc = fs->fs_sblockloc;
645 bcopy(newfs, fs, (u_int)fs->fs_sbsize);
646 brelse(bp);
647 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
648 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
649 UFS_LOCK(ump);
650 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
651 printf("WARNING: %s: reload pending error: blocks %jd "
652 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
653 fs->fs_pendinginodes);
654 fs->fs_pendingblocks = 0;
655 fs->fs_pendinginodes = 0;
656 }
657 UFS_UNLOCK(ump);
658
659 /*
660 * Step 3: re-read summary information from disk.
661 */
662 size = fs->fs_cssize;
663 blks = howmany(size, fs->fs_fsize);
664 if (fs->fs_contigsumsize > 0)
665 size += fs->fs_ncg * sizeof(int32_t);
666 size += fs->fs_ncg * sizeof(u_int8_t);
667 free(fs->fs_csp, M_UFSMNT);
668 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
669 fs->fs_csp = space;
670 for (i = 0; i < blks; i += fs->fs_frag) {
671 size = fs->fs_bsize;
672 if (i + fs->fs_frag > blks)
673 size = (blks - i) * fs->fs_fsize;
674 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
675 NOCRED, &bp);
676 if (error)
677 return (error);
678 bcopy(bp->b_data, space, (u_int)size);
679 space = (char *)space + size;
680 brelse(bp);
681 }
682 /*
683 * We no longer know anything about clusters per cylinder group.
684 */
685 if (fs->fs_contigsumsize > 0) {
686 fs->fs_maxcluster = lp = space;
687 for (i = 0; i < fs->fs_ncg; i++)
688 *lp++ = fs->fs_contigsumsize;
689 space = lp;
690 }
691 size = fs->fs_ncg * sizeof(u_int8_t);
692 fs->fs_contigdirs = (u_int8_t *)space;
693 bzero(fs->fs_contigdirs, size);
694
695 loop:
696 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
697 /*
698 * Skip syncer vnode.
699 */
700 if (vp->v_type == VNON) {
701 VI_UNLOCK(vp);
702 continue;
703 }
704 /*
705 * Step 4: invalidate all cached file data.
706 */
707 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
708 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
709 goto loop;
710 }
711 if (vinvalbuf(vp, 0, 0, 0))
712 panic("ffs_reload: dirty2");
713 /*
714 * Step 5: re-read inode data for all active vnodes.
715 */
716 ip = VTOI(vp);
717 error =
718 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
719 (int)fs->fs_bsize, NOCRED, &bp);
720 if (error) {
721 VOP_UNLOCK(vp, 0);
722 vrele(vp);
723 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
724 return (error);
725 }
726 ffs_load_inode(bp, ip, fs, ip->i_number);
727 ip->i_effnlink = ip->i_nlink;
728 brelse(bp);
729 VOP_UNLOCK(vp, 0);
730 vrele(vp);
731 }
732 return (0);
733 }
734
735 /*
736 * Possible superblock locations ordered from most to least likely.
737 */
738 static int sblock_try[] = SBLOCKSEARCH;
739
740 /*
741 * Common code for mount and mountroot
742 */
743 static int
ffs_mountfs(devvp,mp,td)744 ffs_mountfs(devvp, mp, td)
745 struct vnode *devvp;
746 struct mount *mp;
747 struct thread *td;
748 {
749 struct ufsmount *ump;
750 struct buf *bp;
751 struct fs *fs;
752 struct cdev *dev;
753 void *space;
754 ufs2_daddr_t sblockloc;
755 int error, i, blks, size, ronly;
756 int32_t *lp;
757 struct ucred *cred;
758 struct g_consumer *cp;
759 struct mount *nmp;
760
761 bp = NULL;
762 ump = NULL;
763 cred = td ? td->td_ucred : NOCRED;
764 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
765
766 dev = devvp->v_rdev;
767 dev_ref(dev);
768 DROP_GIANT();
769 g_topology_lock();
770 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
771 g_topology_unlock();
772 PICKUP_GIANT();
773 VOP_UNLOCK(devvp, 0);
774 if (error)
775 goto out;
776 if (devvp->v_rdev->si_iosize_max != 0)
777 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
778 if (mp->mnt_iosize_max > MAXPHYS)
779 mp->mnt_iosize_max = MAXPHYS;
780
781 devvp->v_bufobj.bo_ops = &ffs_ops;
782
783 fs = NULL;
784 sblockloc = 0;
785 /*
786 * Try reading the superblock in each of its possible locations.
787 */
788 for (i = 0; sblock_try[i] != -1; i++) {
789 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
790 error = EINVAL;
791 vfs_mount_error(mp,
792 "Invalid sectorsize %d for superblock size %d",
793 cp->provider->sectorsize, SBLOCKSIZE);
794 goto out;
795 }
796 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
797 cred, &bp)) != 0)
798 goto out;
799 fs = (struct fs *)bp->b_data;
800 sblockloc = sblock_try[i];
801 if ((fs->fs_magic == FS_UFS1_MAGIC ||
802 (fs->fs_magic == FS_UFS2_MAGIC &&
803 (fs->fs_sblockloc == sblockloc ||
804 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
805 fs->fs_bsize <= MAXBSIZE &&
806 fs->fs_bsize >= sizeof(struct fs))
807 break;
808 brelse(bp);
809 bp = NULL;
810 }
811 if (sblock_try[i] == -1) {
812 error = EINVAL; /* XXX needs translation */
813 goto out;
814 }
815 fs->fs_fmod = 0;
816 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */
817 fs->fs_flags &= ~FS_UNCLEAN;
818 if (fs->fs_clean == 0) {
819 fs->fs_flags |= FS_UNCLEAN;
820 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
821 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
822 (fs->fs_flags & FS_DOSOFTDEP))) {
823 printf("WARNING: %s was not properly dismounted\n",
824 fs->fs_fsmnt);
825 } else {
826 vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
827 fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
828 (fs->fs_flags & FS_SUJ) == 0 ? "" :
829 " Forced mount will invalidate journal contents");
830 error = EPERM;
831 goto out;
832 }
833 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
834 (mp->mnt_flag & MNT_FORCE)) {
835 printf("WARNING: %s: lost blocks %jd files %d\n",
836 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
837 fs->fs_pendinginodes);
838 fs->fs_pendingblocks = 0;
839 fs->fs_pendinginodes = 0;
840 }
841 }
842 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
843 printf("WARNING: %s: mount pending error: blocks %jd "
844 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
845 fs->fs_pendinginodes);
846 fs->fs_pendingblocks = 0;
847 fs->fs_pendinginodes = 0;
848 }
849 if ((fs->fs_flags & FS_GJOURNAL) != 0) {
850 #ifdef UFS_GJOURNAL
851 /*
852 * Get journal provider name.
853 */
854 size = 1024;
855 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
856 if (g_io_getattr("GJOURNAL::provider", cp, &size,
857 mp->mnt_gjprovider) == 0) {
858 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
859 M_UFSMNT, M_WAITOK);
860 MNT_ILOCK(mp);
861 mp->mnt_flag |= MNT_GJOURNAL;
862 MNT_IUNLOCK(mp);
863 } else {
864 printf("WARNING: %s: GJOURNAL flag on fs "
865 "but no gjournal provider below\n",
866 mp->mnt_stat.f_mntonname);
867 free(mp->mnt_gjprovider, M_UFSMNT);
868 mp->mnt_gjprovider = NULL;
869 }
870 #else
871 printf("WARNING: %s: GJOURNAL flag on fs but no "
872 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
873 #endif
874 } else {
875 mp->mnt_gjprovider = NULL;
876 }
877 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
878 ump->um_cp = cp;
879 ump->um_bo = &devvp->v_bufobj;
880 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
881 if (fs->fs_magic == FS_UFS1_MAGIC) {
882 ump->um_fstype = UFS1;
883 ump->um_balloc = ffs_balloc_ufs1;
884 } else {
885 ump->um_fstype = UFS2;
886 ump->um_balloc = ffs_balloc_ufs2;
887 }
888 ump->um_blkatoff = ffs_blkatoff;
889 ump->um_truncate = ffs_truncate;
890 ump->um_update = ffs_update;
891 ump->um_valloc = ffs_valloc;
892 ump->um_vfree = ffs_vfree;
893 ump->um_ifree = ffs_ifree;
894 ump->um_rdonly = ffs_rdonly;
895 ump->um_snapgone = ffs_snapgone;
896 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
897 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
898 if (fs->fs_sbsize < SBLOCKSIZE)
899 bp->b_flags |= B_INVAL | B_NOCACHE;
900 brelse(bp);
901 bp = NULL;
902 fs = ump->um_fs;
903 ffs_oldfscompat_read(fs, ump, sblockloc);
904 fs->fs_ronly = ronly;
905 size = fs->fs_cssize;
906 blks = howmany(size, fs->fs_fsize);
907 if (fs->fs_contigsumsize > 0)
908 size += fs->fs_ncg * sizeof(int32_t);
909 size += fs->fs_ncg * sizeof(u_int8_t);
910 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
911 fs->fs_csp = space;
912 for (i = 0; i < blks; i += fs->fs_frag) {
913 size = fs->fs_bsize;
914 if (i + fs->fs_frag > blks)
915 size = (blks - i) * fs->fs_fsize;
916 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
917 cred, &bp)) != 0) {
918 free(fs->fs_csp, M_UFSMNT);
919 goto out;
920 }
921 bcopy(bp->b_data, space, (u_int)size);
922 space = (char *)space + size;
923 brelse(bp);
924 bp = NULL;
925 }
926 if (fs->fs_contigsumsize > 0) {
927 fs->fs_maxcluster = lp = space;
928 for (i = 0; i < fs->fs_ncg; i++)
929 *lp++ = fs->fs_contigsumsize;
930 space = lp;
931 }
932 size = fs->fs_ncg * sizeof(u_int8_t);
933 fs->fs_contigdirs = (u_int8_t *)space;
934 bzero(fs->fs_contigdirs, size);
935 fs->fs_active = NULL;
936 mp->mnt_data = ump;
937 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
938 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
939 nmp = NULL;
940 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
941 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
942 if (nmp)
943 vfs_rel(nmp);
944 vfs_getnewfsid(mp);
945 }
946 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
947 MNT_ILOCK(mp);
948 mp->mnt_flag |= MNT_LOCAL;
949 MNT_IUNLOCK(mp);
950 if ((fs->fs_flags & FS_MULTILABEL) != 0) {
951 #ifdef MAC
952 MNT_ILOCK(mp);
953 mp->mnt_flag |= MNT_MULTILABEL;
954 MNT_IUNLOCK(mp);
955 #else
956 printf("WARNING: %s: multilabel flag on fs but "
957 "no MAC support\n", mp->mnt_stat.f_mntonname);
958 #endif
959 }
960 if ((fs->fs_flags & FS_ACLS) != 0) {
961 #ifdef UFS_ACL
962 MNT_ILOCK(mp);
963
964 if (mp->mnt_flag & MNT_NFS4ACLS)
965 printf("WARNING: %s: ACLs flag on fs conflicts with "
966 "\"nfsv4acls\" mount option; option ignored\n",
967 mp->mnt_stat.f_mntonname);
968 mp->mnt_flag &= ~MNT_NFS4ACLS;
969 mp->mnt_flag |= MNT_ACLS;
970
971 MNT_IUNLOCK(mp);
972 #else
973 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
974 mp->mnt_stat.f_mntonname);
975 #endif
976 }
977 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
978 #ifdef UFS_ACL
979 MNT_ILOCK(mp);
980
981 if (mp->mnt_flag & MNT_ACLS)
982 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
983 "with \"acls\" mount option; option ignored\n",
984 mp->mnt_stat.f_mntonname);
985 mp->mnt_flag &= ~MNT_ACLS;
986 mp->mnt_flag |= MNT_NFS4ACLS;
987
988 MNT_IUNLOCK(mp);
989 #else
990 printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
991 "ACLs support\n", mp->mnt_stat.f_mntonname);
992 #endif
993 }
994 if ((fs->fs_flags & FS_TRIM) != 0) {
995 size = sizeof(int);
996 if (g_io_getattr("GEOM::candelete", cp, &size,
997 &ump->um_candelete) == 0) {
998 if (!ump->um_candelete)
999 printf("WARNING: %s: TRIM flag on fs but disk "
1000 "does not support TRIM\n",
1001 mp->mnt_stat.f_mntonname);
1002 } else {
1003 printf("WARNING: %s: TRIM flag on fs but disk does "
1004 "not confirm that it supports TRIM\n",
1005 mp->mnt_stat.f_mntonname);
1006 ump->um_candelete = 0;
1007 }
1008 }
1009
1010 ump->um_mountp = mp;
1011 ump->um_dev = dev;
1012 ump->um_devvp = devvp;
1013 ump->um_nindir = fs->fs_nindir;
1014 ump->um_bptrtodb = fs->fs_fsbtodb;
1015 ump->um_seqinc = fs->fs_frag;
1016 for (i = 0; i < MAXQUOTAS; i++)
1017 ump->um_quotas[i] = NULLVP;
1018 #ifdef UFS_EXTATTR
1019 ufs_extattr_uepm_init(&ump->um_extattr);
1020 #endif
1021 /*
1022 * Set FS local "last mounted on" information (NULL pad)
1023 */
1024 bzero(fs->fs_fsmnt, MAXMNTLEN);
1025 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1026 mp->mnt_stat.f_iosize = fs->fs_bsize;
1027
1028 if (mp->mnt_flag & MNT_ROOTFS) {
1029 /*
1030 * Root mount; update timestamp in mount structure.
1031 * this will be used by the common root mount code
1032 * to update the system clock.
1033 */
1034 mp->mnt_time = fs->fs_time;
1035 }
1036
1037 if (ronly == 0) {
1038 fs->fs_mtime = time_second;
1039 if ((fs->fs_flags & FS_DOSOFTDEP) &&
1040 (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1041 free(fs->fs_csp, M_UFSMNT);
1042 ffs_flushfiles(mp, FORCECLOSE, td);
1043 goto out;
1044 }
1045 if (devvp->v_type == VCHR && devvp->v_rdev != NULL)
1046 devvp->v_rdev->si_mountpt = mp;
1047 if (fs->fs_snapinum[0] != 0)
1048 ffs_snapshot_mount(mp);
1049 fs->fs_fmod = 1;
1050 fs->fs_clean = 0;
1051 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
1052 }
1053 /*
1054 * Initialize filesystem stat information in mount struct.
1055 */
1056 MNT_ILOCK(mp);
1057 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1058 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1059 MNT_IUNLOCK(mp);
1060 #ifdef UFS_EXTATTR
1061 #ifdef UFS_EXTATTR_AUTOSTART
1062 /*
1063 *
1064 * Auto-starting does the following:
1065 * - check for /.attribute in the fs, and extattr_start if so
1066 * - for each file in .attribute, enable that file with
1067 * an attribute of the same name.
1068 * Not clear how to report errors -- probably eat them.
1069 * This would all happen while the filesystem was busy/not
1070 * available, so would effectively be "atomic".
1071 */
1072 (void) ufs_extattr_autostart(mp, td);
1073 #endif /* !UFS_EXTATTR_AUTOSTART */
1074 #endif /* !UFS_EXTATTR */
1075 return (0);
1076 out:
1077 if (bp)
1078 brelse(bp);
1079 if (cp != NULL) {
1080 DROP_GIANT();
1081 g_topology_lock();
1082 g_vfs_close(cp);
1083 g_topology_unlock();
1084 PICKUP_GIANT();
1085 }
1086 if (ump) {
1087 mtx_destroy(UFS_MTX(ump));
1088 if (mp->mnt_gjprovider != NULL) {
1089 free(mp->mnt_gjprovider, M_UFSMNT);
1090 mp->mnt_gjprovider = NULL;
1091 }
1092 free(ump->um_fs, M_UFSMNT);
1093 free(ump, M_UFSMNT);
1094 mp->mnt_data = NULL;
1095 }
1096 dev_rel(dev);
1097 return (error);
1098 }
1099
1100 #include <sys/sysctl.h>
1101 static int bigcgs = 0;
1102 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1103
1104 /*
1105 * Sanity checks for loading old filesystem superblocks.
1106 * See ffs_oldfscompat_write below for unwound actions.
1107 *
1108 * XXX - Parts get retired eventually.
1109 * Unfortunately new bits get added.
1110 */
1111 static void
ffs_oldfscompat_read(fs,ump,sblockloc)1112 ffs_oldfscompat_read(fs, ump, sblockloc)
1113 struct fs *fs;
1114 struct ufsmount *ump;
1115 ufs2_daddr_t sblockloc;
1116 {
1117 off_t maxfilesize;
1118
1119 /*
1120 * If not yet done, update fs_flags location and value of fs_sblockloc.
1121 */
1122 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1123 fs->fs_flags = fs->fs_old_flags;
1124 fs->fs_old_flags |= FS_FLAGS_UPDATED;
1125 fs->fs_sblockloc = sblockloc;
1126 }
1127 /*
1128 * If not yet done, update UFS1 superblock with new wider fields.
1129 */
1130 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1131 fs->fs_maxbsize = fs->fs_bsize;
1132 fs->fs_time = fs->fs_old_time;
1133 fs->fs_size = fs->fs_old_size;
1134 fs->fs_dsize = fs->fs_old_dsize;
1135 fs->fs_csaddr = fs->fs_old_csaddr;
1136 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1137 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1138 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1139 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1140 }
1141 if (fs->fs_magic == FS_UFS1_MAGIC &&
1142 fs->fs_old_inodefmt < FS_44INODEFMT) {
1143 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1144 fs->fs_qbmask = ~fs->fs_bmask;
1145 fs->fs_qfmask = ~fs->fs_fmask;
1146 }
1147 if (fs->fs_magic == FS_UFS1_MAGIC) {
1148 ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1149 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1150 if (fs->fs_maxfilesize > maxfilesize)
1151 fs->fs_maxfilesize = maxfilesize;
1152 }
1153 /* Compatibility for old filesystems */
1154 if (fs->fs_avgfilesize <= 0)
1155 fs->fs_avgfilesize = AVFILESIZ;
1156 if (fs->fs_avgfpdir <= 0)
1157 fs->fs_avgfpdir = AFPDIR;
1158 if (bigcgs) {
1159 fs->fs_save_cgsize = fs->fs_cgsize;
1160 fs->fs_cgsize = fs->fs_bsize;
1161 }
1162 }
1163
1164 /*
1165 * Unwinding superblock updates for old filesystems.
1166 * See ffs_oldfscompat_read above for details.
1167 *
1168 * XXX - Parts get retired eventually.
1169 * Unfortunately new bits get added.
1170 */
1171 void
ffs_oldfscompat_write(fs,ump)1172 ffs_oldfscompat_write(fs, ump)
1173 struct fs *fs;
1174 struct ufsmount *ump;
1175 {
1176
1177 /*
1178 * Copy back UFS2 updated fields that UFS1 inspects.
1179 */
1180 if (fs->fs_magic == FS_UFS1_MAGIC) {
1181 fs->fs_old_time = fs->fs_time;
1182 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1183 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1184 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1185 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1186 fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1187 }
1188 if (bigcgs) {
1189 fs->fs_cgsize = fs->fs_save_cgsize;
1190 fs->fs_save_cgsize = 0;
1191 }
1192 }
1193
1194 /*
1195 * unmount system call
1196 */
1197 static int
ffs_unmount(mp,mntflags)1198 ffs_unmount(mp, mntflags)
1199 struct mount *mp;
1200 int mntflags;
1201 {
1202 struct thread *td;
1203 struct ufsmount *ump = VFSTOUFS(mp);
1204 struct fs *fs;
1205 int error, flags, susp;
1206 #ifdef UFS_EXTATTR
1207 int e_restart;
1208 #endif
1209
1210 flags = 0;
1211 td = curthread;
1212 fs = ump->um_fs;
1213 susp = 0;
1214 if (mntflags & MNT_FORCE) {
1215 flags |= FORCECLOSE;
1216 susp = fs->fs_ronly == 0;
1217 }
1218 #ifdef UFS_EXTATTR
1219 if ((error = ufs_extattr_stop(mp, td))) {
1220 if (error != EOPNOTSUPP)
1221 printf("WARNING: unmount %s: ufs_extattr_stop "
1222 "returned errno %d\n", mp->mnt_stat.f_mntonname,
1223 error);
1224 e_restart = 0;
1225 } else {
1226 ufs_extattr_uepm_destroy(&ump->um_extattr);
1227 e_restart = 1;
1228 }
1229 #endif
1230 if (susp) {
1231 error = vfs_write_suspend_umnt(mp);
1232 if (error != 0)
1233 goto fail1;
1234 }
1235 if (MOUNTEDSOFTDEP(mp))
1236 error = softdep_flushfiles(mp, flags, td);
1237 else
1238 error = ffs_flushfiles(mp, flags, td);
1239 if (error != 0 && error != ENXIO)
1240 goto fail;
1241
1242 UFS_LOCK(ump);
1243 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1244 printf("WARNING: unmount %s: pending error: blocks %jd "
1245 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1246 fs->fs_pendinginodes);
1247 fs->fs_pendingblocks = 0;
1248 fs->fs_pendinginodes = 0;
1249 }
1250 UFS_UNLOCK(ump);
1251 if (MOUNTEDSOFTDEP(mp))
1252 softdep_unmount(mp);
1253 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1254 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1255 error = ffs_sbupdate(ump, MNT_WAIT, 0);
1256 if (error && error != ENXIO) {
1257 fs->fs_clean = 0;
1258 goto fail;
1259 }
1260 }
1261 if (susp)
1262 vfs_write_resume(mp, VR_START_WRITE);
1263 DROP_GIANT();
1264 g_topology_lock();
1265 if (ump->um_fsckpid > 0) {
1266 /*
1267 * Return to normal read-only mode.
1268 */
1269 error = g_access(ump->um_cp, 0, -1, 0);
1270 ump->um_fsckpid = 0;
1271 }
1272 g_vfs_close(ump->um_cp);
1273 g_topology_unlock();
1274 PICKUP_GIANT();
1275 if (ump->um_devvp->v_type == VCHR && ump->um_devvp->v_rdev != NULL)
1276 ump->um_devvp->v_rdev->si_mountpt = NULL;
1277 vrele(ump->um_devvp);
1278 dev_rel(ump->um_dev);
1279 mtx_destroy(UFS_MTX(ump));
1280 if (mp->mnt_gjprovider != NULL) {
1281 free(mp->mnt_gjprovider, M_UFSMNT);
1282 mp->mnt_gjprovider = NULL;
1283 }
1284 free(fs->fs_csp, M_UFSMNT);
1285 free(fs, M_UFSMNT);
1286 free(ump, M_UFSMNT);
1287 mp->mnt_data = NULL;
1288 MNT_ILOCK(mp);
1289 mp->mnt_flag &= ~MNT_LOCAL;
1290 MNT_IUNLOCK(mp);
1291 return (error);
1292
1293 fail:
1294 if (susp)
1295 vfs_write_resume(mp, VR_START_WRITE);
1296 fail1:
1297 #ifdef UFS_EXTATTR
1298 if (e_restart) {
1299 ufs_extattr_uepm_init(&ump->um_extattr);
1300 #ifdef UFS_EXTATTR_AUTOSTART
1301 (void) ufs_extattr_autostart(mp, td);
1302 #endif
1303 }
1304 #endif
1305
1306 return (error);
1307 }
1308
1309 /*
1310 * Flush out all the files in a filesystem.
1311 */
1312 int
ffs_flushfiles(mp,flags,td)1313 ffs_flushfiles(mp, flags, td)
1314 struct mount *mp;
1315 int flags;
1316 struct thread *td;
1317 {
1318 struct ufsmount *ump;
1319 int qerror, error;
1320
1321 ump = VFSTOUFS(mp);
1322 qerror = 0;
1323 #ifdef QUOTA
1324 if (mp->mnt_flag & MNT_QUOTA) {
1325 int i;
1326 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1327 if (error)
1328 return (error);
1329 for (i = 0; i < MAXQUOTAS; i++) {
1330 error = quotaoff(td, mp, i);
1331 if (error != 0) {
1332 if ((flags & EARLYFLUSH) == 0)
1333 return (error);
1334 else
1335 qerror = error;
1336 }
1337 }
1338
1339 /*
1340 * Here we fall through to vflush again to ensure that
1341 * we have gotten rid of all the system vnodes, unless
1342 * quotas must not be closed.
1343 */
1344 }
1345 #endif
1346 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1347 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1348 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1349 return (error);
1350 ffs_snapshot_unmount(mp);
1351 flags |= FORCECLOSE;
1352 /*
1353 * Here we fall through to vflush again to ensure
1354 * that we have gotten rid of all the system vnodes.
1355 */
1356 }
1357
1358 /*
1359 * Do not close system files if quotas were not closed, to be
1360 * able to sync the remaining dquots. The freeblks softupdate
1361 * workitems might hold a reference on a dquot, preventing
1362 * quotaoff() from completing. Next round of
1363 * softdep_flushworklist() iteration should process the
1364 * blockers, allowing the next run of quotaoff() to finally
1365 * flush held dquots.
1366 *
1367 * Otherwise, flush all the files.
1368 */
1369 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1370 return (error);
1371
1372 /*
1373 * Flush filesystem metadata.
1374 */
1375 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1376 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1377 VOP_UNLOCK(ump->um_devvp, 0);
1378 return (error);
1379 }
1380
1381 /*
1382 * Get filesystem statistics.
1383 */
1384 static int
ffs_statfs(mp,sbp)1385 ffs_statfs(mp, sbp)
1386 struct mount *mp;
1387 struct statfs *sbp;
1388 {
1389 struct ufsmount *ump;
1390 struct fs *fs;
1391
1392 ump = VFSTOUFS(mp);
1393 fs = ump->um_fs;
1394 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1395 panic("ffs_statfs");
1396 sbp->f_version = STATFS_VERSION;
1397 sbp->f_bsize = fs->fs_fsize;
1398 sbp->f_iosize = fs->fs_bsize;
1399 sbp->f_blocks = fs->fs_dsize;
1400 UFS_LOCK(ump);
1401 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1402 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1403 sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1404 dbtofsb(fs, fs->fs_pendingblocks);
1405 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO;
1406 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1407 UFS_UNLOCK(ump);
1408 sbp->f_namemax = NAME_MAX;
1409 return (0);
1410 }
1411
1412 static bool
sync_doupdate(struct inode * ip)1413 sync_doupdate(struct inode *ip)
1414 {
1415
1416 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1417 IN_UPDATE)) != 0);
1418 }
1419
1420 /*
1421 * For a lazy sync, we only care about access times, quotas and the
1422 * superblock. Other filesystem changes are already converted to
1423 * cylinder group blocks or inode blocks updates and are written to
1424 * disk by syncer.
1425 */
1426 static int
ffs_sync_lazy(mp)1427 ffs_sync_lazy(mp)
1428 struct mount *mp;
1429 {
1430 struct vnode *mvp, *vp;
1431 struct inode *ip;
1432 struct thread *td;
1433 int allerror, error;
1434
1435 allerror = 0;
1436 td = curthread;
1437 if ((mp->mnt_flag & MNT_NOATIME) != 0)
1438 goto qupdate;
1439 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1440 if (vp->v_type == VNON) {
1441 VI_UNLOCK(vp);
1442 continue;
1443 }
1444 ip = VTOI(vp);
1445
1446 /*
1447 * The IN_ACCESS flag is converted to IN_MODIFIED by
1448 * ufs_close() and ufs_getattr() by the calls to
1449 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1450 * Test also all the other timestamp flags too, to pick up
1451 * any other cases that could be missed.
1452 */
1453 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1454 VI_UNLOCK(vp);
1455 continue;
1456 }
1457 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1458 td)) != 0)
1459 continue;
1460 if (sync_doupdate(ip))
1461 error = ffs_update(vp, 0);
1462 if (error != 0)
1463 allerror = error;
1464 vput(vp);
1465 }
1466
1467 qupdate:
1468 #ifdef QUOTA
1469 qsync(mp);
1470 #endif
1471
1472 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1473 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1474 allerror = error;
1475 return (allerror);
1476 }
1477
1478 /*
1479 * Go through the disk queues to initiate sandbagged IO;
1480 * go through the inodes to write those that have been modified;
1481 * initiate the writing of the super block if it has been modified.
1482 *
1483 * Note: we are always called with the filesystem marked busy using
1484 * vfs_busy().
1485 */
1486 static int
ffs_sync(mp,waitfor)1487 ffs_sync(mp, waitfor)
1488 struct mount *mp;
1489 int waitfor;
1490 {
1491 struct vnode *mvp, *vp, *devvp;
1492 struct thread *td;
1493 struct inode *ip;
1494 struct ufsmount *ump = VFSTOUFS(mp);
1495 struct fs *fs;
1496 int error, count, lockreq, allerror = 0;
1497 int suspend;
1498 int suspended;
1499 int secondary_writes;
1500 int secondary_accwrites;
1501 int softdep_deps;
1502 int softdep_accdeps;
1503 struct bufobj *bo;
1504
1505 suspend = 0;
1506 suspended = 0;
1507 td = curthread;
1508 fs = ump->um_fs;
1509 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1510 panic("%s: ffs_sync: modification on read-only filesystem",
1511 fs->fs_fsmnt);
1512 if (waitfor == MNT_LAZY) {
1513 if (!rebooting)
1514 return (ffs_sync_lazy(mp));
1515 waitfor = MNT_NOWAIT;
1516 }
1517
1518 /*
1519 * Write back each (modified) inode.
1520 */
1521 lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1522 if (waitfor == MNT_SUSPEND) {
1523 suspend = 1;
1524 waitfor = MNT_WAIT;
1525 }
1526 if (waitfor == MNT_WAIT)
1527 lockreq = LK_EXCLUSIVE;
1528 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1529 loop:
1530 /* Grab snapshot of secondary write counts */
1531 MNT_ILOCK(mp);
1532 secondary_writes = mp->mnt_secondary_writes;
1533 secondary_accwrites = mp->mnt_secondary_accwrites;
1534 MNT_IUNLOCK(mp);
1535
1536 /* Grab snapshot of softdep dependency counts */
1537 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1538
1539 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1540 /*
1541 * Depend on the vnode interlock to keep things stable enough
1542 * for a quick test. Since there might be hundreds of
1543 * thousands of vnodes, we cannot afford even a subroutine
1544 * call unless there's a good chance that we have work to do.
1545 */
1546 if (vp->v_type == VNON) {
1547 VI_UNLOCK(vp);
1548 continue;
1549 }
1550 ip = VTOI(vp);
1551 if ((ip->i_flag &
1552 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1553 vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1554 VI_UNLOCK(vp);
1555 continue;
1556 }
1557 if ((error = vget(vp, lockreq, td)) != 0) {
1558 if (error == ENOENT || error == ENOLCK) {
1559 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1560 goto loop;
1561 }
1562 continue;
1563 }
1564 if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1565 allerror = error;
1566 vput(vp);
1567 }
1568 /*
1569 * Force stale filesystem control information to be flushed.
1570 */
1571 if (waitfor == MNT_WAIT || rebooting) {
1572 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1573 allerror = error;
1574 /* Flushed work items may create new vnodes to clean */
1575 if (allerror == 0 && count)
1576 goto loop;
1577 }
1578 #ifdef QUOTA
1579 qsync(mp);
1580 #endif
1581
1582 devvp = ump->um_devvp;
1583 bo = &devvp->v_bufobj;
1584 BO_LOCK(bo);
1585 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1586 BO_UNLOCK(bo);
1587 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1588 error = VOP_FSYNC(devvp, waitfor, td);
1589 VOP_UNLOCK(devvp, 0);
1590 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1591 error = ffs_sbupdate(ump, waitfor, 0);
1592 if (error != 0)
1593 allerror = error;
1594 if (allerror == 0 && waitfor == MNT_WAIT)
1595 goto loop;
1596 } else if (suspend != 0) {
1597 if (softdep_check_suspend(mp,
1598 devvp,
1599 softdep_deps,
1600 softdep_accdeps,
1601 secondary_writes,
1602 secondary_accwrites) != 0) {
1603 MNT_IUNLOCK(mp);
1604 goto loop; /* More work needed */
1605 }
1606 mtx_assert(MNT_MTX(mp), MA_OWNED);
1607 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1608 MNT_IUNLOCK(mp);
1609 suspended = 1;
1610 } else
1611 BO_UNLOCK(bo);
1612 /*
1613 * Write back modified superblock.
1614 */
1615 if (fs->fs_fmod != 0 &&
1616 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1617 allerror = error;
1618 return (allerror);
1619 }
1620
1621 int
ffs_vget(mp,ino,flags,vpp)1622 ffs_vget(mp, ino, flags, vpp)
1623 struct mount *mp;
1624 ino_t ino;
1625 int flags;
1626 struct vnode **vpp;
1627 {
1628 return (ffs_vgetf(mp, ino, flags, vpp, 0));
1629 }
1630
1631 int
ffs_vgetf(mp,ino,flags,vpp,ffs_flags)1632 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1633 struct mount *mp;
1634 ino_t ino;
1635 int flags;
1636 struct vnode **vpp;
1637 int ffs_flags;
1638 {
1639 struct fs *fs;
1640 struct inode *ip;
1641 struct ufsmount *ump;
1642 struct buf *bp;
1643 struct vnode *vp;
1644 struct cdev *dev;
1645 int error;
1646
1647 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1648 if (error || *vpp != NULL)
1649 return (error);
1650
1651 /*
1652 * We must promote to an exclusive lock for vnode creation. This
1653 * can happen if lookup is passed LOCKSHARED.
1654 */
1655 if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1656 flags &= ~LK_TYPE_MASK;
1657 flags |= LK_EXCLUSIVE;
1658 }
1659
1660 /*
1661 * We do not lock vnode creation as it is believed to be too
1662 * expensive for such rare case as simultaneous creation of vnode
1663 * for same ino by different processes. We just allow them to race
1664 * and check later to decide who wins. Let the race begin!
1665 */
1666
1667 ump = VFSTOUFS(mp);
1668 dev = ump->um_dev;
1669 fs = ump->um_fs;
1670 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1671
1672 /* Allocate a new vnode/inode. */
1673 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1674 &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1675 if (error) {
1676 *vpp = NULL;
1677 uma_zfree(uma_inode, ip);
1678 return (error);
1679 }
1680 /*
1681 * FFS supports recursive locking.
1682 */
1683 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1684 VN_LOCK_AREC(vp);
1685 vp->v_data = ip;
1686 vp->v_bufobj.bo_bsize = fs->fs_bsize;
1687 ip->i_vnode = vp;
1688 ip->i_ump = ump;
1689 ip->i_fs = fs;
1690 ip->i_dev = dev;
1691 ip->i_number = ino;
1692 ip->i_ea_refs = 0;
1693 ip->i_nextclustercg = -1;
1694 #ifdef QUOTA
1695 {
1696 int i;
1697 for (i = 0; i < MAXQUOTAS; i++)
1698 ip->i_dquot[i] = NODQUOT;
1699 }
1700 #endif
1701
1702 if (ffs_flags & FFSV_FORCEINSMQ)
1703 vp->v_vflag |= VV_FORCEINSMQ;
1704 error = insmntque(vp, mp);
1705 if (error != 0) {
1706 uma_zfree(uma_inode, ip);
1707 *vpp = NULL;
1708 return (error);
1709 }
1710 vp->v_vflag &= ~VV_FORCEINSMQ;
1711 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1712 if (error || *vpp != NULL)
1713 return (error);
1714
1715 /* Read in the disk contents for the inode, copy into the inode. */
1716 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1717 (int)fs->fs_bsize, NOCRED, &bp);
1718 if (error) {
1719 /*
1720 * The inode does not contain anything useful, so it would
1721 * be misleading to leave it on its hash chain. With mode
1722 * still zero, it will be unlinked and returned to the free
1723 * list by vput().
1724 */
1725 brelse(bp);
1726 vput(vp);
1727 *vpp = NULL;
1728 return (error);
1729 }
1730 if (ip->i_ump->um_fstype == UFS1)
1731 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1732 else
1733 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1734 ffs_load_inode(bp, ip, fs, ino);
1735 if (DOINGSOFTDEP(vp))
1736 softdep_load_inodeblock(ip);
1737 else
1738 ip->i_effnlink = ip->i_nlink;
1739 bqrelse(bp);
1740
1741 /*
1742 * Initialize the vnode from the inode, check for aliases.
1743 * Note that the underlying vnode may have changed.
1744 */
1745 if (ip->i_ump->um_fstype == UFS1)
1746 error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1747 else
1748 error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1749 if (error) {
1750 vput(vp);
1751 *vpp = NULL;
1752 return (error);
1753 }
1754
1755 /*
1756 * Finish inode initialization.
1757 */
1758 if (vp->v_type != VFIFO) {
1759 /* FFS supports shared locking for all files except fifos. */
1760 VN_LOCK_ASHARE(vp);
1761 }
1762
1763 /*
1764 * Set up a generation number for this inode if it does not
1765 * already have one. This should only happen on old filesystems.
1766 */
1767 if (ip->i_gen == 0) {
1768 ip->i_gen = arc4random() / 2 + 1;
1769 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1770 ip->i_flag |= IN_MODIFIED;
1771 DIP_SET(ip, i_gen, ip->i_gen);
1772 }
1773 }
1774 #ifdef MAC
1775 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1776 /*
1777 * If this vnode is already allocated, and we're running
1778 * multi-label, attempt to perform a label association
1779 * from the extended attributes on the inode.
1780 */
1781 error = mac_vnode_associate_extattr(mp, vp);
1782 if (error) {
1783 /* ufs_inactive will release ip->i_devvp ref. */
1784 vput(vp);
1785 *vpp = NULL;
1786 return (error);
1787 }
1788 }
1789 #endif
1790
1791 *vpp = vp;
1792 return (0);
1793 }
1794
1795 /*
1796 * File handle to vnode
1797 *
1798 * Have to be really careful about stale file handles:
1799 * - check that the inode number is valid
1800 * - call ffs_vget() to get the locked inode
1801 * - check for an unallocated inode (i_mode == 0)
1802 * - check that the given client host has export rights and return
1803 * those rights via. exflagsp and credanonp
1804 */
1805 static int
ffs_fhtovp(mp,fhp,flags,vpp)1806 ffs_fhtovp(mp, fhp, flags, vpp)
1807 struct mount *mp;
1808 struct fid *fhp;
1809 int flags;
1810 struct vnode **vpp;
1811 {
1812 struct ufid *ufhp;
1813 struct fs *fs;
1814
1815 ufhp = (struct ufid *)fhp;
1816 fs = VFSTOUFS(mp)->um_fs;
1817 if (ufhp->ufid_ino < ROOTINO ||
1818 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1819 return (ESTALE);
1820 return (ufs_fhtovp(mp, ufhp, flags, vpp));
1821 }
1822
1823 /*
1824 * Initialize the filesystem.
1825 */
1826 static int
ffs_init(vfsp)1827 ffs_init(vfsp)
1828 struct vfsconf *vfsp;
1829 {
1830
1831 ffs_susp_initialize();
1832 softdep_initialize();
1833 return (ufs_init(vfsp));
1834 }
1835
1836 /*
1837 * Undo the work of ffs_init().
1838 */
1839 static int
ffs_uninit(vfsp)1840 ffs_uninit(vfsp)
1841 struct vfsconf *vfsp;
1842 {
1843 int ret;
1844
1845 ret = ufs_uninit(vfsp);
1846 softdep_uninitialize();
1847 ffs_susp_uninitialize();
1848 return (ret);
1849 }
1850
1851 /*
1852 * Write a superblock and associated information back to disk.
1853 */
1854 int
ffs_sbupdate(ump,waitfor,suspended)1855 ffs_sbupdate(ump, waitfor, suspended)
1856 struct ufsmount *ump;
1857 int waitfor;
1858 int suspended;
1859 {
1860 struct fs *fs = ump->um_fs;
1861 struct buf *sbbp;
1862 struct buf *bp;
1863 int blks;
1864 void *space;
1865 int i, size, error, allerror = 0;
1866
1867 if (fs->fs_ronly == 1 &&
1868 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1869 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1870 panic("ffs_sbupdate: write read-only filesystem");
1871 /*
1872 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1873 */
1874 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1875 (int)fs->fs_sbsize, 0, 0, 0);
1876 /*
1877 * First write back the summary information.
1878 */
1879 blks = howmany(fs->fs_cssize, fs->fs_fsize);
1880 space = fs->fs_csp;
1881 for (i = 0; i < blks; i += fs->fs_frag) {
1882 size = fs->fs_bsize;
1883 if (i + fs->fs_frag > blks)
1884 size = (blks - i) * fs->fs_fsize;
1885 bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1886 size, 0, 0, 0);
1887 bcopy(space, bp->b_data, (u_int)size);
1888 space = (char *)space + size;
1889 if (suspended)
1890 bp->b_flags |= B_VALIDSUSPWRT;
1891 if (waitfor != MNT_WAIT)
1892 bawrite(bp);
1893 else if ((error = bwrite(bp)) != 0)
1894 allerror = error;
1895 }
1896 /*
1897 * Now write back the superblock itself. If any errors occurred
1898 * up to this point, then fail so that the superblock avoids
1899 * being written out as clean.
1900 */
1901 if (allerror) {
1902 brelse(sbbp);
1903 return (allerror);
1904 }
1905 bp = sbbp;
1906 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1907 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1908 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1909 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1910 fs->fs_sblockloc = SBLOCK_UFS1;
1911 }
1912 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1913 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1914 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1915 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1916 fs->fs_sblockloc = SBLOCK_UFS2;
1917 }
1918 fs->fs_fmod = 0;
1919 fs->fs_time = time_second;
1920 if (MOUNTEDSOFTDEP(ump->um_mountp))
1921 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1922 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1923 ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1924 if (suspended)
1925 bp->b_flags |= B_VALIDSUSPWRT;
1926 if (waitfor != MNT_WAIT)
1927 bawrite(bp);
1928 else if ((error = bwrite(bp)) != 0)
1929 allerror = error;
1930 return (allerror);
1931 }
1932
1933 static int
ffs_extattrctl(struct mount * mp,int cmd,struct vnode * filename_vp,int attrnamespace,const char * attrname)1934 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1935 int attrnamespace, const char *attrname)
1936 {
1937
1938 #ifdef UFS_EXTATTR
1939 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1940 attrname));
1941 #else
1942 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1943 attrname));
1944 #endif
1945 }
1946
1947 static void
ffs_ifree(struct ufsmount * ump,struct inode * ip)1948 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1949 {
1950
1951 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1952 uma_zfree(uma_ufs1, ip->i_din1);
1953 else if (ip->i_din2 != NULL)
1954 uma_zfree(uma_ufs2, ip->i_din2);
1955 uma_zfree(uma_inode, ip);
1956 }
1957
1958 static int dobkgrdwrite = 1;
1959 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1960 "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1961
1962 /*
1963 * Complete a background write started from bwrite.
1964 */
1965 static void
ffs_backgroundwritedone(struct buf * bp)1966 ffs_backgroundwritedone(struct buf *bp)
1967 {
1968 struct bufobj *bufobj;
1969 struct buf *origbp;
1970
1971 /*
1972 * Find the original buffer that we are writing.
1973 */
1974 bufobj = bp->b_bufobj;
1975 BO_LOCK(bufobj);
1976 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
1977 panic("backgroundwritedone: lost buffer");
1978
1979 /*
1980 * We should mark the cylinder group buffer origbp as
1981 * dirty, to not loose the failed write.
1982 */
1983 if ((bp->b_ioflags & BIO_ERROR) != 0)
1984 origbp->b_vflags |= BV_BKGRDERR;
1985 BO_UNLOCK(bufobj);
1986 /*
1987 * Process dependencies then return any unfinished ones.
1988 */
1989 pbrelvp(bp);
1990 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
1991 buf_complete(bp);
1992 #ifdef SOFTUPDATES
1993 if (!LIST_EMPTY(&bp->b_dep))
1994 softdep_move_dependencies(bp, origbp);
1995 #endif
1996 /*
1997 * This buffer is marked B_NOCACHE so when it is released
1998 * by biodone it will be tossed.
1999 */
2000 bp->b_flags |= B_NOCACHE;
2001 bp->b_flags &= ~B_CACHE;
2002
2003 /*
2004 * Prevent brelse() from trying to keep and re-dirtying bp on
2005 * errors. It causes b_bufobj dereference in
2006 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2007 * pbrelvp() above.
2008 */
2009 if ((bp->b_ioflags & BIO_ERROR) != 0)
2010 bp->b_flags |= B_INVAL;
2011 bufdone(bp);
2012 BO_LOCK(bufobj);
2013 /*
2014 * Clear the BV_BKGRDINPROG flag in the original buffer
2015 * and awaken it if it is waiting for the write to complete.
2016 * If BV_BKGRDINPROG is not set in the original buffer it must
2017 * have been released and re-instantiated - which is not legal.
2018 */
2019 KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2020 ("backgroundwritedone: lost buffer2"));
2021 origbp->b_vflags &= ~BV_BKGRDINPROG;
2022 if (origbp->b_vflags & BV_BKGRDWAIT) {
2023 origbp->b_vflags &= ~BV_BKGRDWAIT;
2024 wakeup(&origbp->b_xflags);
2025 }
2026 BO_UNLOCK(bufobj);
2027 }
2028
2029
2030 /*
2031 * Write, release buffer on completion. (Done by iodone
2032 * if async). Do not bother writing anything if the buffer
2033 * is invalid.
2034 *
2035 * Note that we set B_CACHE here, indicating that buffer is
2036 * fully valid and thus cacheable. This is true even of NFS
2037 * now so we set it generally. This could be set either here
2038 * or in biodone() since the I/O is synchronous. We put it
2039 * here.
2040 */
2041 static int
ffs_bufwrite(struct buf * bp)2042 ffs_bufwrite(struct buf *bp)
2043 {
2044 struct buf *newbp;
2045
2046 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2047 if (bp->b_flags & B_INVAL) {
2048 brelse(bp);
2049 return (0);
2050 }
2051
2052 if (!BUF_ISLOCKED(bp))
2053 panic("bufwrite: buffer is not busy???");
2054 /*
2055 * If a background write is already in progress, delay
2056 * writing this block if it is asynchronous. Otherwise
2057 * wait for the background write to complete.
2058 */
2059 BO_LOCK(bp->b_bufobj);
2060 if (bp->b_vflags & BV_BKGRDINPROG) {
2061 if (bp->b_flags & B_ASYNC) {
2062 BO_UNLOCK(bp->b_bufobj);
2063 bdwrite(bp);
2064 return (0);
2065 }
2066 bp->b_vflags |= BV_BKGRDWAIT;
2067 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2068 "bwrbg", 0);
2069 if (bp->b_vflags & BV_BKGRDINPROG)
2070 panic("bufwrite: still writing");
2071 }
2072 bp->b_vflags &= ~BV_BKGRDERR;
2073 BO_UNLOCK(bp->b_bufobj);
2074
2075 /*
2076 * If this buffer is marked for background writing and we
2077 * do not have to wait for it, make a copy and write the
2078 * copy so as to leave this buffer ready for further use.
2079 *
2080 * This optimization eats a lot of memory. If we have a page
2081 * or buffer shortfall we can't do it.
2082 */
2083 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2084 (bp->b_flags & B_ASYNC) &&
2085 !vm_page_count_severe() &&
2086 !buf_dirty_count_severe()) {
2087 KASSERT(bp->b_iodone == NULL,
2088 ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2089
2090 /* get a new block */
2091 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2092 if (newbp == NULL)
2093 goto normal_write;
2094
2095 KASSERT(buf_mapped(bp), ("Unmapped cg"));
2096 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2097 BO_LOCK(bp->b_bufobj);
2098 bp->b_vflags |= BV_BKGRDINPROG;
2099 BO_UNLOCK(bp->b_bufobj);
2100 newbp->b_xflags |= BX_BKGRDMARKER;
2101 newbp->b_lblkno = bp->b_lblkno;
2102 newbp->b_blkno = bp->b_blkno;
2103 newbp->b_offset = bp->b_offset;
2104 newbp->b_iodone = ffs_backgroundwritedone;
2105 newbp->b_flags |= B_ASYNC;
2106 newbp->b_flags &= ~B_INVAL;
2107 pbgetvp(bp->b_vp, newbp);
2108
2109 #ifdef SOFTUPDATES
2110 /*
2111 * Move over the dependencies. If there are rollbacks,
2112 * leave the parent buffer dirtied as it will need to
2113 * be written again.
2114 */
2115 if (LIST_EMPTY(&bp->b_dep) ||
2116 softdep_move_dependencies(bp, newbp) == 0)
2117 bundirty(bp);
2118 #else
2119 bundirty(bp);
2120 #endif
2121
2122 /*
2123 * Initiate write on the copy, release the original. The
2124 * BKGRDINPROG flag prevents it from going away until
2125 * the background write completes.
2126 */
2127 bqrelse(bp);
2128 bp = newbp;
2129 } else
2130 /* Mark the buffer clean */
2131 bundirty(bp);
2132
2133
2134 /* Let the normal bufwrite do the rest for us */
2135 normal_write:
2136 return (bufwrite(bp));
2137 }
2138
2139
2140 static void
ffs_geom_strategy(struct bufobj * bo,struct buf * bp)2141 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2142 {
2143 struct vnode *vp;
2144 int error;
2145 struct buf *tbp;
2146 int nocopy;
2147
2148 vp = bo->__bo_vnode;
2149 if (bp->b_iocmd == BIO_WRITE) {
2150 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2151 bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2152 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2153 panic("ffs_geom_strategy: bad I/O");
2154 nocopy = bp->b_flags & B_NOCOPY;
2155 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2156 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2157 vp->v_rdev->si_snapdata != NULL) {
2158 if ((bp->b_flags & B_CLUSTER) != 0) {
2159 runningbufwakeup(bp);
2160 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2161 b_cluster.cluster_entry) {
2162 error = ffs_copyonwrite(vp, tbp);
2163 if (error != 0 &&
2164 error != EOPNOTSUPP) {
2165 bp->b_error = error;
2166 bp->b_ioflags |= BIO_ERROR;
2167 bufdone(bp);
2168 return;
2169 }
2170 }
2171 bp->b_runningbufspace = bp->b_bufsize;
2172 atomic_add_long(&runningbufspace,
2173 bp->b_runningbufspace);
2174 } else {
2175 error = ffs_copyonwrite(vp, bp);
2176 if (error != 0 && error != EOPNOTSUPP) {
2177 bp->b_error = error;
2178 bp->b_ioflags |= BIO_ERROR;
2179 bufdone(bp);
2180 return;
2181 }
2182 }
2183 }
2184 #ifdef SOFTUPDATES
2185 if ((bp->b_flags & B_CLUSTER) != 0) {
2186 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2187 b_cluster.cluster_entry) {
2188 if (!LIST_EMPTY(&tbp->b_dep))
2189 buf_start(tbp);
2190 }
2191 } else {
2192 if (!LIST_EMPTY(&bp->b_dep))
2193 buf_start(bp);
2194 }
2195
2196 #endif
2197 }
2198 g_vfs_strategy(bo, bp);
2199 }
2200
2201 int
ffs_own_mount(const struct mount * mp)2202 ffs_own_mount(const struct mount *mp)
2203 {
2204
2205 if (mp->mnt_op == &ufs_vfsops)
2206 return (1);
2207 return (0);
2208 }
2209
2210 #ifdef DDB
2211 #ifdef SOFTUPDATES
2212
2213 /* defined in ffs_softdep.c */
2214 extern void db_print_ffs(struct ufsmount *ump);
2215
DB_SHOW_COMMAND(ffs,db_show_ffs)2216 DB_SHOW_COMMAND(ffs, db_show_ffs)
2217 {
2218 struct mount *mp;
2219 struct ufsmount *ump;
2220
2221 if (have_addr) {
2222 ump = VFSTOUFS((struct mount *)addr);
2223 db_print_ffs(ump);
2224 return;
2225 }
2226
2227 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2228 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2229 db_print_ffs(VFSTOUFS(mp));
2230 }
2231 }
2232
2233 #endif /* SOFTUPDATES */
2234 #endif /* DDB */
2235