1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1989, 1991, 1993, 1994
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 #include "opt_quota.h"
34 #include "opt_ufs.h"
35 #include "opt_ffs.h"
36 #include "opt_ddb.h"
37
38 #include <sys/param.h>
39 #include <sys/gsb_crc32.h>
40 #include <sys/systm.h>
41 #include <sys/namei.h>
42 #include <sys/priv.h>
43 #include <sys/proc.h>
44 #include <sys/taskqueue.h>
45 #include <sys/kernel.h>
46 #include <sys/ktr.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/fcntl.h>
53 #include <sys/ioccom.h>
54 #include <sys/malloc.h>
55 #include <sys/mutex.h>
56 #include <sys/rwlock.h>
57 #include <sys/sysctl.h>
58 #include <sys/vmmeter.h>
59
60 #include <security/mac/mac_framework.h>
61
62 #include <ufs/ufs/dir.h>
63 #include <ufs/ufs/extattr.h>
64 #include <ufs/ufs/gjournal.h>
65 #include <ufs/ufs/quota.h>
66 #include <ufs/ufs/ufsmount.h>
67 #include <ufs/ufs/inode.h>
68 #include <ufs/ufs/ufs_extern.h>
69
70 #include <ufs/ffs/fs.h>
71 #include <ufs/ffs/ffs_extern.h>
72
73 #include <vm/vm.h>
74 #include <vm/uma.h>
75 #include <vm/vm_page.h>
76
77 #include <geom/geom.h>
78 #include <geom/geom_vfs.h>
79
80 #include <ddb/ddb.h>
81
82 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
83 VFS_SMR_DECLARE;
84
85 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *);
86 static void ffs_ifree(struct ufsmount *ump, struct inode *ip);
87 static int ffs_sync_lazy(struct mount *mp);
88 static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size);
89 static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size);
90
91 static vfs_init_t ffs_init;
92 static vfs_uninit_t ffs_uninit;
93 static vfs_extattrctl_t ffs_extattrctl;
94 static vfs_cmount_t ffs_cmount;
95 static vfs_unmount_t ffs_unmount;
96 static vfs_mount_t ffs_mount;
97 static vfs_statfs_t ffs_statfs;
98 static vfs_fhtovp_t ffs_fhtovp;
99 static vfs_sync_t ffs_sync;
100
101 static struct vfsops ufs_vfsops = {
102 .vfs_extattrctl = ffs_extattrctl,
103 .vfs_fhtovp = ffs_fhtovp,
104 .vfs_init = ffs_init,
105 .vfs_mount = ffs_mount,
106 .vfs_cmount = ffs_cmount,
107 .vfs_quotactl = ufs_quotactl,
108 .vfs_root = vfs_cache_root,
109 .vfs_cachedroot = ufs_root,
110 .vfs_statfs = ffs_statfs,
111 .vfs_sync = ffs_sync,
112 .vfs_uninit = ffs_uninit,
113 .vfs_unmount = ffs_unmount,
114 .vfs_vget = ffs_vget,
115 .vfs_susp_clean = process_deferred_inactive,
116 };
117
118 VFS_SET(ufs_vfsops, ufs, VFCF_FILEREVINC);
119 MODULE_VERSION(ufs, 1);
120
121 static b_strategy_t ffs_geom_strategy;
122 static b_write_t ffs_bufwrite;
123
124 static struct buf_ops ffs_ops = {
125 .bop_name = "FFS",
126 .bop_write = ffs_bufwrite,
127 .bop_strategy = ffs_geom_strategy,
128 .bop_sync = bufsync,
129 #ifdef NO_FFS_SNAPSHOT
130 .bop_bdflush = bufbdflush,
131 #else
132 .bop_bdflush = ffs_bdflush,
133 #endif
134 };
135
136 /*
137 * Note that userquota and groupquota options are not currently used
138 * by UFS/FFS code and generally mount(8) does not pass those options
139 * from userland, but they can be passed by loader(8) via
140 * vfs.root.mountfrom.options.
141 */
142 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
143 "noclusterw", "noexec", "export", "force", "from", "groupquota",
144 "multilabel", "nfsv4acls", "snapshot", "nosuid", "suiddir",
145 "nosymfollow", "sync", "union", "userquota", "untrusted", NULL };
146
147 static int ffs_enxio_enable = 1;
148 SYSCTL_DECL(_vfs_ffs);
149 SYSCTL_INT(_vfs_ffs, OID_AUTO, enxio_enable, CTLFLAG_RWTUN,
150 &ffs_enxio_enable, 0,
151 "enable mapping of other disk I/O errors to ENXIO");
152
153 /*
154 * Return buffer with the contents of block "offset" from the beginning of
155 * directory "ip". If "res" is non-zero, fill it in with a pointer to the
156 * remaining space in the directory.
157 */
158 static int
ffs_blkatoff(struct vnode * vp,off_t offset,char ** res,struct buf ** bpp)159 ffs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp)
160 {
161 struct inode *ip;
162 struct fs *fs;
163 struct buf *bp;
164 ufs_lbn_t lbn;
165 int bsize, error;
166
167 ip = VTOI(vp);
168 fs = ITOFS(ip);
169 lbn = lblkno(fs, offset);
170 bsize = blksize(fs, ip, lbn);
171
172 *bpp = NULL;
173 error = bread(vp, lbn, bsize, NOCRED, &bp);
174 if (error) {
175 return (error);
176 }
177 if (res)
178 *res = (char *)bp->b_data + blkoff(fs, offset);
179 *bpp = bp;
180 return (0);
181 }
182
183 /*
184 * Load up the contents of an inode and copy the appropriate pieces
185 * to the incore copy.
186 */
187 static int
ffs_load_inode(struct buf * bp,struct inode * ip,struct fs * fs,ino_t ino)188 ffs_load_inode(struct buf *bp, struct inode *ip, struct fs *fs, ino_t ino)
189 {
190 struct ufs1_dinode *dip1;
191 struct ufs2_dinode *dip2;
192 time_t now;
193 int error;
194
195 now = time_second > fs->fs_time ? time_second : fs->fs_time;
196 if (I_IS_UFS1(ip)) {
197 dip1 = ip->i_din1;
198 *dip1 =
199 *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
200 ip->i_mode = dip1->di_mode;
201 ip->i_nlink = dip1->di_nlink;
202 ip->i_effnlink = dip1->di_nlink;
203 ip->i_size = dip1->di_size;
204 ip->i_flags = dip1->di_flags;
205 ip->i_gen = dip1->di_gen;
206 ip->i_uid = dip1->di_uid;
207 ip->i_gid = dip1->di_gid;
208 if (ffs_oldfscompat_inode_read(fs, ip->i_dp, now) &&
209 fs->fs_ronly == 0)
210 UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
211 return (0);
212 }
213 dip2 = ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
214 if ((error = ffs_verify_dinode_ckhash(fs, dip2)) != 0 &&
215 !ffs_fsfail_cleanup(ITOUMP(ip), error)) {
216 printf("%s: inode %jd: check-hash failed\n", fs->fs_fsmnt,
217 (intmax_t)ino);
218 return (error);
219 }
220 *ip->i_din2 = *dip2;
221 dip2 = ip->i_din2;
222 ip->i_mode = dip2->di_mode;
223 ip->i_nlink = dip2->di_nlink;
224 ip->i_effnlink = dip2->di_nlink;
225 ip->i_size = dip2->di_size;
226 ip->i_flags = dip2->di_flags;
227 ip->i_gen = dip2->di_gen;
228 ip->i_uid = dip2->di_uid;
229 ip->i_gid = dip2->di_gid;
230 if (ffs_oldfscompat_inode_read(fs, ip->i_dp, now) && fs->fs_ronly == 0)
231 UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
232 return (0);
233 }
234
235 /*
236 * Verify that a filesystem block number is a valid data block.
237 * This routine is only called on untrusted filesystems.
238 */
239 static int
ffs_check_blkno(struct mount * mp,ino_t inum,ufs2_daddr_t daddr,int blksize)240 ffs_check_blkno(struct mount *mp, ino_t inum, ufs2_daddr_t daddr, int blksize)
241 {
242 struct fs *fs;
243 struct ufsmount *ump;
244 ufs2_daddr_t end_daddr;
245 int cg, havemtx;
246
247 KASSERT((mp->mnt_flag & MNT_UNTRUSTED) != 0,
248 ("ffs_check_blkno called on a trusted file system"));
249 ump = VFSTOUFS(mp);
250 fs = ump->um_fs;
251 cg = dtog(fs, daddr);
252 end_daddr = daddr + numfrags(fs, blksize);
253 /*
254 * Verify that the block number is a valid data block. Also check
255 * that it does not point to an inode block or a superblock. Accept
256 * blocks that are unalloacted (0) or part of snapshot metadata
257 * (BLK_NOCOPY or BLK_SNAP).
258 *
259 * Thus, the block must be in a valid range for the filesystem and
260 * either in the space before a backup superblock (except the first
261 * cylinder group where that space is used by the bootstrap code) or
262 * after the inode blocks and before the end of the cylinder group.
263 */
264 if ((uint64_t)daddr <= BLK_SNAP ||
265 ((uint64_t)end_daddr <= fs->fs_size &&
266 ((cg > 0 && end_daddr <= cgsblock(fs, cg)) ||
267 (daddr >= cgdmin(fs, cg) &&
268 end_daddr <= cgbase(fs, cg) + fs->fs_fpg))))
269 return (0);
270 if ((havemtx = mtx_owned(UFS_MTX(ump))) == 0)
271 UFS_LOCK(ump);
272 if (ppsratecheck(&ump->um_last_integritymsg,
273 &ump->um_secs_integritymsg, 1)) {
274 UFS_UNLOCK(ump);
275 uprintf("\n%s: inode %jd, out-of-range indirect block "
276 "number %jd\n", mp->mnt_stat.f_mntonname, inum, daddr);
277 if (havemtx)
278 UFS_LOCK(ump);
279 } else if (!havemtx)
280 UFS_UNLOCK(ump);
281 return (EINTEGRITY);
282 }
283
284 /*
285 * On first ENXIO error, initiate an asynchronous forcible unmount.
286 * Used to unmount filesystems whose underlying media has gone away.
287 *
288 * Return true if a cleanup is in progress.
289 */
290 int
ffs_fsfail_cleanup(struct ufsmount * ump,int error)291 ffs_fsfail_cleanup(struct ufsmount *ump, int error)
292 {
293 int retval;
294
295 UFS_LOCK(ump);
296 retval = ffs_fsfail_cleanup_locked(ump, error);
297 UFS_UNLOCK(ump);
298 return (retval);
299 }
300
301 int
ffs_fsfail_cleanup_locked(struct ufsmount * ump,int error)302 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error)
303 {
304 mtx_assert(UFS_MTX(ump), MA_OWNED);
305 if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) {
306 ump->um_flags |= UM_FSFAIL_CLEANUP;
307 if (ump->um_mountp == rootvnode->v_mount)
308 panic("UFS: root fs would be forcibly unmounted");
309
310 /*
311 * Queue an async forced unmount.
312 */
313 vfs_ref(ump->um_mountp);
314 dounmount(ump->um_mountp,
315 MNT_FORCE | MNT_RECURSE | MNT_DEFERRED, curthread);
316 printf("UFS: forcibly unmounting %s from %s\n",
317 ump->um_mountp->mnt_stat.f_mntfromname,
318 ump->um_mountp->mnt_stat.f_mntonname);
319 }
320 return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0);
321 }
322
323 /*
324 * Wrapper used during ENXIO cleanup to allocate empty buffers when
325 * the kernel is unable to read the real one. They are needed so that
326 * the soft updates code can use them to unwind its dependencies.
327 */
328 int
ffs_breadz(struct ufsmount * ump,struct vnode * vp,daddr_t lblkno,daddr_t dblkno,int size,daddr_t * rablkno,int * rabsize,int cnt,struct ucred * cred,int flags,void (* ckhashfunc)(struct buf *),struct buf ** bpp)329 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno,
330 daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt,
331 struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *),
332 struct buf **bpp)
333 {
334 int error;
335
336 flags |= GB_CVTENXIO;
337 error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt,
338 cred, flags, ckhashfunc, bpp);
339 if (error != 0 && ffs_fsfail_cleanup(ump, error)) {
340 error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp);
341 KASSERT(error == 0, ("getblkx failed"));
342 vfs_bio_bzero_buf(*bpp, 0, size);
343 }
344 return (error);
345 }
346
347 static int
ffs_mount(struct mount * mp)348 ffs_mount(struct mount *mp)
349 {
350 struct vnode *devvp, *odevvp;
351 struct thread *td;
352 struct ufsmount *ump = NULL;
353 struct fs *fs;
354 int error, flags;
355 int error1 __diagused;
356 uint64_t mntorflags, saved_mnt_flag;
357 accmode_t accmode;
358 struct nameidata ndp;
359 char *fspec;
360 bool mounted_softdep;
361
362 td = curthread;
363 if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
364 return (EINVAL);
365 if (uma_inode == NULL) {
366 uma_inode = uma_zcreate("FFS inode",
367 sizeof(struct inode), NULL, NULL, NULL, NULL,
368 UMA_ALIGN_PTR, 0);
369 uma_ufs1 = uma_zcreate("FFS1 dinode",
370 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
371 UMA_ALIGN_PTR, 0);
372 uma_ufs2 = uma_zcreate("FFS2 dinode",
373 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
374 UMA_ALIGN_PTR, 0);
375 VFS_SMR_ZONE_SET(uma_inode);
376 }
377
378 vfs_deleteopt(mp->mnt_optnew, "groupquota");
379 vfs_deleteopt(mp->mnt_optnew, "userquota");
380
381 fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
382 if (error)
383 return (error);
384
385 mntorflags = 0;
386 if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0)
387 mntorflags |= MNT_UNTRUSTED;
388
389 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
390 mntorflags |= MNT_ACLS;
391
392 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
393 mntorflags |= MNT_SNAPSHOT;
394 /*
395 * Once we have set the MNT_SNAPSHOT flag, do not
396 * persist "snapshot" in the options list.
397 */
398 vfs_deleteopt(mp->mnt_optnew, "snapshot");
399 vfs_deleteopt(mp->mnt_opt, "snapshot");
400 }
401
402 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
403 if (mntorflags & MNT_ACLS) {
404 vfs_mount_error(mp,
405 "\"acls\" and \"nfsv4acls\" options "
406 "are mutually exclusive");
407 return (EINVAL);
408 }
409 mntorflags |= MNT_NFS4ACLS;
410 }
411
412 MNT_ILOCK(mp);
413 mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
414 mp->mnt_flag |= mntorflags;
415 MNT_IUNLOCK(mp);
416
417 /*
418 * If this is a snapshot request, take the snapshot.
419 */
420 if (mp->mnt_flag & MNT_SNAPSHOT) {
421 if ((mp->mnt_flag & MNT_UPDATE) == 0)
422 return (EINVAL);
423 return (ffs_snapshot(mp, fspec));
424 }
425
426 /*
427 * Must not call namei() while owning busy ref.
428 */
429 if (mp->mnt_flag & MNT_UPDATE)
430 vfs_unbusy(mp);
431
432 /*
433 * Not an update, or updating the name: look up the name
434 * and verify that it refers to a sensible disk device.
435 */
436 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec);
437 error = namei(&ndp);
438 if ((mp->mnt_flag & MNT_UPDATE) != 0) {
439 /*
440 * Unmount does not start if MNT_UPDATE is set. Mount
441 * update busies mp before setting MNT_UPDATE. We
442 * must be able to retain our busy ref successfully,
443 * without sleep.
444 */
445 error1 = vfs_busy(mp, MBF_NOWAIT);
446 MPASS(error1 == 0);
447 }
448 if (error != 0)
449 return (error);
450 NDFREE_PNBUF(&ndp);
451 if (!vn_isdisk_error(ndp.ni_vp, &error)) {
452 vput(ndp.ni_vp);
453 return (error);
454 }
455
456 /*
457 * If mount by non-root, then verify that user has necessary
458 * permissions on the device.
459 */
460 accmode = VREAD;
461 if ((mp->mnt_flag & MNT_RDONLY) == 0)
462 accmode |= VWRITE;
463 error = VOP_ACCESS(ndp.ni_vp, accmode, td->td_ucred, td);
464 if (error)
465 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
466 if (error) {
467 vput(ndp.ni_vp);
468 return (error);
469 }
470
471 /*
472 * New mount
473 *
474 * We need the name for the mount point (also used for
475 * "last mounted on") copied in. If an error occurs,
476 * the mount point is discarded by the upper level code.
477 * Note that vfs_mount_alloc() populates f_mntonname for us.
478 */
479 if ((mp->mnt_flag & MNT_UPDATE) == 0) {
480 if ((error = ffs_mountfs(ndp.ni_vp, mp, td)) != 0) {
481 vrele(ndp.ni_vp);
482 return (error);
483 }
484 } else {
485 /*
486 * When updating, check whether changing from read-only to
487 * read/write; if there is no device name, that's all we do.
488 */
489 ump = VFSTOUFS(mp);
490 fs = ump->um_fs;
491 odevvp = ump->um_odevvp;
492 devvp = ump->um_devvp;
493
494 /*
495 * If it's not the same vnode, or at least the same device
496 * then it's not correct.
497 */
498 if (ndp.ni_vp->v_rdev != ump->um_odevvp->v_rdev)
499 error = EINVAL; /* needs translation */
500 vput(ndp.ni_vp);
501 if (error)
502 return (error);
503 if (fs->fs_ronly == 0 &&
504 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
505 /*
506 * Flush any dirty data and suspend filesystem.
507 */
508 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
509 return (error);
510 error = vfs_write_suspend_umnt(mp);
511 if (error != 0)
512 return (error);
513
514 fs->fs_ronly = 1;
515 if (MOUNTEDSOFTDEP(mp)) {
516 MNT_ILOCK(mp);
517 mp->mnt_flag &= ~MNT_SOFTDEP;
518 MNT_IUNLOCK(mp);
519 mounted_softdep = true;
520 } else
521 mounted_softdep = false;
522
523 /*
524 * Check for and optionally get rid of files open
525 * for writing.
526 */
527 flags = WRITECLOSE;
528 if (mp->mnt_flag & MNT_FORCE)
529 flags |= FORCECLOSE;
530 if (mounted_softdep) {
531 error = softdep_flushfiles(mp, flags, td);
532 } else {
533 error = ffs_flushfiles(mp, flags, td);
534 }
535 if (error) {
536 fs->fs_ronly = 0;
537 if (mounted_softdep) {
538 MNT_ILOCK(mp);
539 mp->mnt_flag |= MNT_SOFTDEP;
540 MNT_IUNLOCK(mp);
541 }
542 vfs_write_resume(mp, 0);
543 return (error);
544 }
545
546 if (fs->fs_pendingblocks != 0 ||
547 fs->fs_pendinginodes != 0) {
548 printf("WARNING: %s Update error: blocks %jd "
549 "files %d\n", fs->fs_fsmnt,
550 (intmax_t)fs->fs_pendingblocks,
551 fs->fs_pendinginodes);
552 fs->fs_pendingblocks = 0;
553 fs->fs_pendinginodes = 0;
554 }
555 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
556 fs->fs_clean = 1;
557 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
558 fs->fs_ronly = 0;
559 fs->fs_clean = 0;
560 if (mounted_softdep) {
561 MNT_ILOCK(mp);
562 mp->mnt_flag |= MNT_SOFTDEP;
563 MNT_IUNLOCK(mp);
564 }
565 vfs_write_resume(mp, 0);
566 return (error);
567 }
568 if (mounted_softdep)
569 softdep_unmount(mp);
570 g_topology_lock();
571 /*
572 * Drop our write and exclusive access.
573 */
574 g_access(ump->um_cp, 0, -1, -1);
575 g_topology_unlock();
576 MNT_ILOCK(mp);
577 mp->mnt_flag |= MNT_RDONLY;
578 MNT_IUNLOCK(mp);
579 /*
580 * Allow the writers to note that filesystem
581 * is ro now.
582 */
583 vfs_write_resume(mp, 0);
584 }
585 if ((mp->mnt_flag & MNT_RELOAD) &&
586 (error = ffs_reload(mp, 0)) != 0) {
587 return (error);
588 } else {
589 /* ffs_reload replaces the superblock structure */
590 fs = ump->um_fs;
591 }
592 if (fs->fs_ronly &&
593 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
594 /*
595 * If upgrade to read-write by non-root, then verify
596 * that user has necessary permissions on the device.
597 */
598 vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY);
599 error = VOP_ACCESS(odevvp, VREAD | VWRITE,
600 td->td_ucred, td);
601 if (error)
602 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
603 VOP_UNLOCK(odevvp);
604 if (error) {
605 return (error);
606 }
607 fs->fs_flags &= ~FS_UNCLEAN;
608 if (fs->fs_clean == 0) {
609 fs->fs_flags |= FS_UNCLEAN;
610 if ((mp->mnt_flag & MNT_FORCE) ||
611 ((fs->fs_flags &
612 (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
613 (fs->fs_flags & FS_DOSOFTDEP))) {
614 printf("WARNING: %s was not properly "
615 "dismounted\n",
616 mp->mnt_stat.f_mntonname);
617 } else {
618 vfs_mount_error(mp,
619 "R/W mount of %s denied. %s.%s",
620 mp->mnt_stat.f_mntonname,
621 "Filesystem is not clean - run fsck",
622 (fs->fs_flags & FS_SUJ) == 0 ? "" :
623 " Forced mount will invalidate"
624 " journal contents");
625 return (EPERM);
626 }
627 }
628 g_topology_lock();
629 /*
630 * Request exclusive write access.
631 */
632 error = g_access(ump->um_cp, 0, 1, 1);
633 g_topology_unlock();
634 if (error)
635 return (error);
636 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
637 return (error);
638 error = vfs_write_suspend_umnt(mp);
639 if (error != 0)
640 return (error);
641 fs->fs_ronly = 0;
642 MNT_ILOCK(mp);
643 saved_mnt_flag = MNT_RDONLY;
644 if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag &
645 MNT_ASYNC) != 0)
646 saved_mnt_flag |= MNT_ASYNC;
647 mp->mnt_flag &= ~saved_mnt_flag;
648 MNT_IUNLOCK(mp);
649 fs->fs_mtime = time_second;
650 /* check to see if we need to start softdep */
651 if ((fs->fs_flags & FS_DOSOFTDEP) &&
652 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
653 fs->fs_ronly = 1;
654 MNT_ILOCK(mp);
655 mp->mnt_flag |= saved_mnt_flag;
656 MNT_IUNLOCK(mp);
657 vfs_write_resume(mp, 0);
658 return (error);
659 }
660 fs->fs_clean = 0;
661 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
662 fs->fs_ronly = 1;
663 if ((fs->fs_flags & FS_DOSOFTDEP) != 0)
664 softdep_unmount(mp);
665 MNT_ILOCK(mp);
666 mp->mnt_flag |= saved_mnt_flag;
667 MNT_IUNLOCK(mp);
668 vfs_write_resume(mp, 0);
669 return (error);
670 }
671 if (fs->fs_snapinum[0] != 0)
672 ffs_snapshot_mount(mp);
673 vfs_write_resume(mp, 0);
674 }
675 /*
676 * Soft updates is incompatible with "async",
677 * so if we are doing softupdates stop the user
678 * from setting the async flag in an update.
679 * Softdep_mount() clears it in an initial mount
680 * or ro->rw remount.
681 */
682 if (MOUNTEDSOFTDEP(mp)) {
683 /* XXX: Reset too late ? */
684 MNT_ILOCK(mp);
685 mp->mnt_flag &= ~MNT_ASYNC;
686 MNT_IUNLOCK(mp);
687 }
688 /*
689 * Keep MNT_ACLS flag if it is stored in superblock.
690 */
691 if ((fs->fs_flags & FS_ACLS) != 0) {
692 /* XXX: Set too late ? */
693 MNT_ILOCK(mp);
694 mp->mnt_flag |= MNT_ACLS;
695 MNT_IUNLOCK(mp);
696 }
697
698 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
699 /* XXX: Set too late ? */
700 MNT_ILOCK(mp);
701 mp->mnt_flag |= MNT_NFS4ACLS;
702 MNT_IUNLOCK(mp);
703 }
704
705 }
706
707 MNT_ILOCK(mp);
708 /*
709 * This is racy versus lookup, see ufs_fplookup_vexec for details.
710 */
711 if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0)
712 panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp);
713 if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0)
714 mp->mnt_kern_flag |= MNTK_FPLOOKUP;
715 MNT_IUNLOCK(mp);
716
717 vfs_mountedfrom(mp, fspec);
718 return (0);
719 }
720
721 /*
722 * Compatibility with old mount system call.
723 */
724
725 static int
ffs_cmount(struct mntarg * ma,void * data,uint64_t flags)726 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
727 {
728 struct ufs_args args;
729 int error;
730
731 if (data == NULL)
732 return (EINVAL);
733 error = copyin(data, &args, sizeof args);
734 if (error)
735 return (error);
736
737 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
738 ma = mount_arg(ma, "export", &args.export, sizeof(args.export));
739 error = kernel_mount(ma, flags);
740
741 return (error);
742 }
743
744 /*
745 * Reload all incore data for a filesystem (used after running fsck on
746 * the root filesystem and finding things to fix). If the 'force' flag
747 * is 0, the filesystem must be mounted read-only.
748 *
749 * Things to do to update the mount:
750 * 1) invalidate all cached meta-data.
751 * 2) re-read superblock from disk.
752 * 3) If requested, clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags
753 * to allow secondary writers.
754 * 4) invalidate all cached file data.
755 * 5) re-read inode data for all active vnodes.
756 */
757 int
ffs_reload(struct mount * mp,int flags)758 ffs_reload(struct mount *mp, int flags)
759 {
760 struct vnode *vp, *mvp, *devvp;
761 struct inode *ip;
762 struct buf *bp;
763 struct fs *fs, *newfs;
764 struct ufsmount *ump;
765 int error;
766
767 ump = VFSTOUFS(mp);
768
769 MNT_ILOCK(mp);
770 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
771 MNT_IUNLOCK(mp);
772 return (EINVAL);
773 }
774 MNT_IUNLOCK(mp);
775
776 /*
777 * Step 1: invalidate all cached meta-data.
778 */
779 devvp = VFSTOUFS(mp)->um_devvp;
780 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
781 if (vinvalbuf(devvp, 0, 0, 0) != 0)
782 panic("ffs_reload: dirty1");
783 VOP_UNLOCK(devvp);
784
785 /*
786 * Step 2: re-read superblock from disk.
787 */
788 if ((error = ffs_sbget(devvp, &newfs, UFS_STDSB, 0, M_UFSMNT,
789 ffs_use_bread)) != 0)
790 return (error);
791 /*
792 * Replace our superblock with the new superblock. Preserve
793 * our read-only status.
794 */
795 fs = VFSTOUFS(mp)->um_fs;
796 newfs->fs_ronly = fs->fs_ronly;
797 free(fs->fs_csp, M_UFSMNT);
798 free(fs->fs_si, M_UFSMNT);
799 free(fs, M_UFSMNT);
800 fs = VFSTOUFS(mp)->um_fs = newfs;
801 ump->um_bsize = fs->fs_bsize;
802 ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
803 UFS_LOCK(ump);
804 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
805 printf("WARNING: %s: reload pending error: blocks %jd "
806 "files %d\n", mp->mnt_stat.f_mntonname,
807 (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes);
808 fs->fs_pendingblocks = 0;
809 fs->fs_pendinginodes = 0;
810 }
811 UFS_UNLOCK(ump);
812 /*
813 * Step 3: If requested, clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags
814 * to allow secondary writers.
815 */
816 if ((flags & FFSR_UNSUSPEND) != 0) {
817 MNT_ILOCK(mp);
818 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
819 wakeup(&mp->mnt_flag);
820 MNT_IUNLOCK(mp);
821 }
822
823 loop:
824 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
825 /*
826 * Skip syncer vnode.
827 */
828 if (vp->v_type == VNON) {
829 VI_UNLOCK(vp);
830 continue;
831 }
832 /*
833 * Step 4: invalidate all cached file data.
834 */
835 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
836 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
837 goto loop;
838 }
839 if (vinvalbuf(vp, 0, 0, 0))
840 panic("ffs_reload: dirty2");
841 /*
842 * Step 5: re-read inode data for all active vnodes.
843 */
844 ip = VTOI(vp);
845 error =
846 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
847 (int)fs->fs_bsize, NOCRED, &bp);
848 if (error) {
849 vput(vp);
850 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
851 return (error);
852 }
853 if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) {
854 brelse(bp);
855 vput(vp);
856 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
857 return (error);
858 }
859 ip->i_effnlink = ip->i_nlink;
860 brelse(bp);
861 vput(vp);
862 }
863 return (0);
864 }
865
866 /*
867 * Common code for mount and mountroot
868 */
869 static int
ffs_mountfs(struct vnode * odevvp,struct mount * mp,struct thread * td)870 ffs_mountfs(struct vnode *odevvp, struct mount *mp, struct thread *td)
871 {
872 struct ufsmount *ump;
873 struct fs *fs;
874 struct cdev *dev;
875 int error, i, len, ronly;
876 struct ucred *cred;
877 struct g_consumer *cp;
878 struct mount *nmp;
879 struct vnode *devvp;
880 int candelete, canspeedup;
881
882 fs = NULL;
883 ump = NULL;
884 cred = td ? td->td_ucred : NOCRED;
885 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
886
887 devvp = mntfs_allocvp(mp, odevvp);
888 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
889 dev = devvp->v_rdev;
890 KASSERT(dev->si_snapdata == NULL, ("non-NULL snapshot data"));
891 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
892 (uintptr_t)mp) == 0) {
893 mntfs_freevp(devvp);
894 return (EBUSY);
895 }
896 g_topology_lock();
897 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
898 g_topology_unlock();
899 if (error != 0) {
900 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
901 mntfs_freevp(devvp);
902 return (error);
903 }
904 dev_ref(dev);
905 devvp->v_bufobj.bo_ops = &ffs_ops;
906 BO_LOCK(&odevvp->v_bufobj);
907 odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
908 BO_UNLOCK(&odevvp->v_bufobj);
909 VOP_UNLOCK(devvp);
910 if (dev->si_iosize_max != 0)
911 mp->mnt_iosize_max = dev->si_iosize_max;
912 if (mp->mnt_iosize_max > maxphys)
913 mp->mnt_iosize_max = maxphys;
914 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
915 error = EINVAL;
916 vfs_mount_error(mp,
917 "Invalid sectorsize %d for superblock size %d",
918 cp->provider->sectorsize, SBLOCKSIZE);
919 goto out;
920 }
921 /* fetch the superblock and summary information */
922 if ((mp->mnt_flag & (MNT_ROOTFS | MNT_FORCE)) != 0)
923 error = ffs_sbsearch(devvp, &fs, 0, M_UFSMNT, ffs_use_bread);
924 else
925 error = ffs_sbget(devvp, &fs, UFS_STDSB, 0, M_UFSMNT,
926 ffs_use_bread);
927 if (error != 0)
928 goto out;
929 fs->fs_flags &= ~FS_UNCLEAN;
930 if (fs->fs_clean == 0) {
931 fs->fs_flags |= FS_UNCLEAN;
932 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
933 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
934 (fs->fs_flags & FS_DOSOFTDEP))) {
935 printf("WARNING: %s was not properly dismounted\n",
936 mp->mnt_stat.f_mntonname);
937 } else {
938 vfs_mount_error(mp, "R/W mount on %s denied. "
939 "Filesystem is not clean - run fsck.%s",
940 mp->mnt_stat.f_mntonname,
941 (fs->fs_flags & FS_SUJ) == 0 ? "" :
942 " Forced mount will invalidate journal contents");
943 error = EPERM;
944 goto out;
945 }
946 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
947 (mp->mnt_flag & MNT_FORCE)) {
948 printf("WARNING: %s: lost blocks %jd files %d\n",
949 mp->mnt_stat.f_mntonname,
950 (intmax_t)fs->fs_pendingblocks,
951 fs->fs_pendinginodes);
952 fs->fs_pendingblocks = 0;
953 fs->fs_pendinginodes = 0;
954 }
955 }
956 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
957 printf("WARNING: %s: mount pending error: blocks %jd "
958 "files %d\n", mp->mnt_stat.f_mntonname,
959 (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes);
960 fs->fs_pendingblocks = 0;
961 fs->fs_pendinginodes = 0;
962 }
963 if ((fs->fs_flags & FS_GJOURNAL) != 0) {
964 #ifdef UFS_GJOURNAL
965 /*
966 * Get journal provider name.
967 */
968 len = 1024;
969 mp->mnt_gjprovider = malloc((uint64_t)len, M_UFSMNT, M_WAITOK);
970 if (g_io_getattr("GJOURNAL::provider", cp, &len,
971 mp->mnt_gjprovider) == 0) {
972 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
973 M_UFSMNT, M_WAITOK);
974 MNT_ILOCK(mp);
975 mp->mnt_flag |= MNT_GJOURNAL;
976 MNT_IUNLOCK(mp);
977 } else {
978 if ((mp->mnt_flag & MNT_RDONLY) == 0)
979 printf("WARNING: %s: GJOURNAL flag on fs "
980 "but no gjournal provider below\n",
981 mp->mnt_stat.f_mntonname);
982 free(mp->mnt_gjprovider, M_UFSMNT);
983 mp->mnt_gjprovider = NULL;
984 }
985 #else
986 printf("WARNING: %s: GJOURNAL flag on fs but no "
987 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
988 #endif
989 } else {
990 mp->mnt_gjprovider = NULL;
991 }
992 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
993 ump->um_cp = cp;
994 ump->um_bo = &devvp->v_bufobj;
995 ump->um_fs = fs;
996 if (fs->fs_magic == FS_UFS1_MAGIC) {
997 ump->um_fstype = UFS1;
998 ump->um_balloc = ffs_balloc_ufs1;
999 } else {
1000 ump->um_fstype = UFS2;
1001 ump->um_balloc = ffs_balloc_ufs2;
1002 }
1003 ump->um_blkatoff = ffs_blkatoff;
1004 ump->um_truncate = ffs_truncate;
1005 ump->um_update = ffs_update;
1006 ump->um_valloc = ffs_valloc;
1007 ump->um_vfree = ffs_vfree;
1008 ump->um_ifree = ffs_ifree;
1009 ump->um_rdonly = ffs_rdonly;
1010 ump->um_snapgone = ffs_snapgone;
1011 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0)
1012 ump->um_check_blkno = ffs_check_blkno;
1013 else
1014 ump->um_check_blkno = NULL;
1015 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
1016 fs->fs_ronly = ronly;
1017 fs->fs_active = NULL;
1018 mp->mnt_data = ump;
1019 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
1020 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
1021 nmp = NULL;
1022 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
1023 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
1024 if (nmp)
1025 vfs_rel(nmp);
1026 vfs_getnewfsid(mp);
1027 }
1028 ump->um_bsize = fs->fs_bsize;
1029 ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
1030 MNT_ILOCK(mp);
1031 mp->mnt_flag |= MNT_LOCAL;
1032 MNT_IUNLOCK(mp);
1033 if ((fs->fs_flags & FS_MULTILABEL) != 0) {
1034 #ifdef MAC
1035 MNT_ILOCK(mp);
1036 mp->mnt_flag |= MNT_MULTILABEL;
1037 MNT_IUNLOCK(mp);
1038 #else
1039 printf("WARNING: %s: multilabel flag on fs but "
1040 "no MAC support\n", mp->mnt_stat.f_mntonname);
1041 #endif
1042 }
1043 if ((fs->fs_flags & FS_ACLS) != 0) {
1044 #ifdef UFS_ACL
1045 MNT_ILOCK(mp);
1046
1047 if (mp->mnt_flag & MNT_NFS4ACLS)
1048 printf("WARNING: %s: ACLs flag on fs conflicts with "
1049 "\"nfsv4acls\" mount option; option ignored\n",
1050 mp->mnt_stat.f_mntonname);
1051 mp->mnt_flag &= ~MNT_NFS4ACLS;
1052 mp->mnt_flag |= MNT_ACLS;
1053
1054 MNT_IUNLOCK(mp);
1055 #else
1056 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
1057 mp->mnt_stat.f_mntonname);
1058 #endif
1059 }
1060 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1061 #ifdef UFS_ACL
1062 MNT_ILOCK(mp);
1063
1064 if (mp->mnt_flag & MNT_ACLS)
1065 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1066 "with \"acls\" mount option; option ignored\n",
1067 mp->mnt_stat.f_mntonname);
1068 mp->mnt_flag &= ~MNT_ACLS;
1069 mp->mnt_flag |= MNT_NFS4ACLS;
1070
1071 MNT_IUNLOCK(mp);
1072 #else
1073 printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1074 "ACLs support\n", mp->mnt_stat.f_mntonname);
1075 #endif
1076 }
1077 if ((fs->fs_flags & FS_TRIM) != 0) {
1078 len = sizeof(int);
1079 if (g_io_getattr("GEOM::candelete", cp, &len,
1080 &candelete) == 0) {
1081 if (candelete)
1082 ump->um_flags |= UM_CANDELETE;
1083 else
1084 printf("WARNING: %s: TRIM flag on fs but disk "
1085 "does not support TRIM\n",
1086 mp->mnt_stat.f_mntonname);
1087 } else {
1088 printf("WARNING: %s: TRIM flag on fs but disk does "
1089 "not confirm that it supports TRIM\n",
1090 mp->mnt_stat.f_mntonname);
1091 }
1092 if (((ump->um_flags) & UM_CANDELETE) != 0) {
1093 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1094 taskqueue_thread_enqueue, &ump->um_trim_tq);
1095 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1096 "%s trim", mp->mnt_stat.f_mntonname);
1097 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM,
1098 &ump->um_trimlisthashsize);
1099 }
1100 }
1101
1102 len = sizeof(int);
1103 if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) {
1104 if (canspeedup)
1105 ump->um_flags |= UM_CANSPEEDUP;
1106 }
1107
1108 ump->um_mountp = mp;
1109 ump->um_dev = dev;
1110 ump->um_devvp = devvp;
1111 ump->um_odevvp = odevvp;
1112 ump->um_nindir = fs->fs_nindir;
1113 ump->um_bptrtodb = fs->fs_fsbtodb;
1114 ump->um_seqinc = fs->fs_frag;
1115 for (i = 0; i < MAXQUOTAS; i++)
1116 ump->um_quotas[i] = NULL;
1117 #ifdef UFS_EXTATTR
1118 ufs_extattr_uepm_init(&ump->um_extattr);
1119 #endif
1120 /*
1121 * Set FS local "last mounted on" information (NULL pad)
1122 */
1123 bzero(fs->fs_fsmnt, MAXMNTLEN);
1124 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1125 mp->mnt_stat.f_iosize = fs->fs_bsize;
1126
1127 if (mp->mnt_flag & MNT_ROOTFS) {
1128 /*
1129 * Root mount; update timestamp in mount structure.
1130 * this will be used by the common root mount code
1131 * to update the system clock.
1132 */
1133 mp->mnt_time = fs->fs_time;
1134 }
1135
1136 if (ronly == 0) {
1137 fs->fs_mtime = time_second;
1138 if ((fs->fs_flags & FS_DOSOFTDEP) &&
1139 (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1140 ffs_flushfiles(mp, FORCECLOSE, td);
1141 goto out;
1142 }
1143 if (fs->fs_snapinum[0] != 0)
1144 ffs_snapshot_mount(mp);
1145 fs->fs_fmod = 1;
1146 fs->fs_clean = 0;
1147 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
1148 }
1149 /*
1150 * Initialize filesystem state information in mount struct.
1151 */
1152 MNT_ILOCK(mp);
1153 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1154 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1155 MNT_IUNLOCK(mp);
1156 #ifdef UFS_EXTATTR
1157 #ifdef UFS_EXTATTR_AUTOSTART
1158 /*
1159 *
1160 * Auto-starting does the following:
1161 * - check for /.attribute in the fs, and extattr_start if so
1162 * - for each file in .attribute, enable that file with
1163 * an attribute of the same name.
1164 * Not clear how to report errors -- probably eat them.
1165 * This would all happen while the filesystem was busy/not
1166 * available, so would effectively be "atomic".
1167 */
1168 (void) ufs_extattr_autostart(mp, td);
1169 #endif /* !UFS_EXTATTR_AUTOSTART */
1170 #endif /* !UFS_EXTATTR */
1171 return (0);
1172 out:
1173 if (fs != NULL) {
1174 free(fs->fs_csp, M_UFSMNT);
1175 free(fs->fs_si, M_UFSMNT);
1176 free(fs, M_UFSMNT);
1177 }
1178 if (cp != NULL) {
1179 g_topology_lock();
1180 g_vfs_close(cp);
1181 g_topology_unlock();
1182 }
1183 if (ump != NULL) {
1184 mtx_destroy(UFS_MTX(ump));
1185 if (mp->mnt_gjprovider != NULL) {
1186 free(mp->mnt_gjprovider, M_UFSMNT);
1187 mp->mnt_gjprovider = NULL;
1188 }
1189 MPASS(ump->um_softdep == NULL);
1190 free(ump, M_UFSMNT);
1191 mp->mnt_data = NULL;
1192 }
1193 BO_LOCK(&odevvp->v_bufobj);
1194 odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1195 BO_UNLOCK(&odevvp->v_bufobj);
1196 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1197 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1198 mntfs_freevp(devvp);
1199 dev_rel(dev);
1200 return (error);
1201 }
1202
1203 /*
1204 * A read function for use by filesystem-layer routines.
1205 */
1206 static int
ffs_use_bread(void * devfd,off_t loc,void ** bufp,int size)1207 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size)
1208 {
1209 struct buf *bp;
1210 int error;
1211
1212 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp));
1213 *bufp = malloc(size, M_UFSMNT, M_WAITOK);
1214 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED,
1215 &bp)) != 0)
1216 return (error);
1217 bcopy(bp->b_data, *bufp, size);
1218 bp->b_flags |= B_INVAL | B_NOCACHE;
1219 brelse(bp);
1220 return (0);
1221 }
1222
1223 /*
1224 * unmount system call
1225 */
1226 static int
ffs_unmount(struct mount * mp,int mntflags)1227 ffs_unmount(struct mount *mp, int mntflags)
1228 {
1229 struct thread *td;
1230 struct ufsmount *ump = VFSTOUFS(mp);
1231 struct fs *fs;
1232 int error, flags, susp;
1233 #ifdef UFS_EXTATTR
1234 int e_restart;
1235 #endif
1236
1237 flags = 0;
1238 td = curthread;
1239 fs = ump->um_fs;
1240 if (mntflags & MNT_FORCE)
1241 flags |= FORCECLOSE;
1242 susp = fs->fs_ronly == 0;
1243 #ifdef UFS_EXTATTR
1244 if ((error = ufs_extattr_stop(mp, td))) {
1245 if (error != EOPNOTSUPP)
1246 printf("WARNING: unmount %s: ufs_extattr_stop "
1247 "returned errno %d\n", mp->mnt_stat.f_mntonname,
1248 error);
1249 e_restart = 0;
1250 } else {
1251 ufs_extattr_uepm_destroy(&ump->um_extattr);
1252 e_restart = 1;
1253 }
1254 #endif
1255 if (susp) {
1256 error = vfs_write_suspend_umnt(mp);
1257 if (error != 0)
1258 goto fail1;
1259 }
1260 if (MOUNTEDSOFTDEP(mp))
1261 error = softdep_flushfiles(mp, flags, td);
1262 else
1263 error = ffs_flushfiles(mp, flags, td);
1264 if (error != 0 && !ffs_fsfail_cleanup(ump, error))
1265 goto fail;
1266
1267 UFS_LOCK(ump);
1268 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1269 printf("WARNING: unmount %s: pending error: blocks %jd "
1270 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1271 fs->fs_pendinginodes);
1272 fs->fs_pendingblocks = 0;
1273 fs->fs_pendinginodes = 0;
1274 }
1275 UFS_UNLOCK(ump);
1276 if (MOUNTEDSOFTDEP(mp))
1277 softdep_unmount(mp);
1278 MPASS(ump->um_softdep == NULL);
1279 if (fs->fs_ronly == 0) {
1280 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1281 error = ffs_sbupdate(ump, MNT_WAIT, 0);
1282 if (ffs_fsfail_cleanup(ump, error))
1283 error = 0;
1284 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) {
1285 fs->fs_clean = 0;
1286 goto fail;
1287 }
1288 }
1289 if (susp)
1290 vfs_write_resume(mp, VR_START_WRITE);
1291 if (ump->um_trim_tq != NULL) {
1292 MPASS(ump->um_trim_inflight == 0);
1293 taskqueue_free(ump->um_trim_tq);
1294 free (ump->um_trimhash, M_TRIM);
1295 }
1296 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1297 g_topology_lock();
1298 g_vfs_close(ump->um_cp);
1299 g_topology_unlock();
1300 BO_LOCK(&ump->um_odevvp->v_bufobj);
1301 ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1302 BO_UNLOCK(&ump->um_odevvp->v_bufobj);
1303 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1304 mntfs_freevp(ump->um_devvp);
1305 vrele(ump->um_odevvp);
1306 dev_rel(ump->um_dev);
1307 mtx_destroy(UFS_MTX(ump));
1308 if (mp->mnt_gjprovider != NULL) {
1309 free(mp->mnt_gjprovider, M_UFSMNT);
1310 mp->mnt_gjprovider = NULL;
1311 }
1312 free(fs->fs_csp, M_UFSMNT);
1313 free(fs->fs_si, M_UFSMNT);
1314 free(fs, M_UFSMNT);
1315 free(ump, M_UFSMNT);
1316 mp->mnt_data = NULL;
1317 if (td->td_su == mp) {
1318 td->td_su = NULL;
1319 vfs_rel(mp);
1320 }
1321 return (error);
1322
1323 fail:
1324 if (susp)
1325 vfs_write_resume(mp, VR_START_WRITE);
1326 fail1:
1327 #ifdef UFS_EXTATTR
1328 if (e_restart) {
1329 ufs_extattr_uepm_init(&ump->um_extattr);
1330 #ifdef UFS_EXTATTR_AUTOSTART
1331 (void) ufs_extattr_autostart(mp, td);
1332 #endif
1333 }
1334 #endif
1335
1336 return (error);
1337 }
1338
1339 /*
1340 * Flush out all the files in a filesystem.
1341 */
1342 int
ffs_flushfiles(struct mount * mp,int flags,struct thread * td)1343 ffs_flushfiles(struct mount *mp, int flags, struct thread *td)
1344 {
1345 struct ufsmount *ump;
1346 int qerror, error;
1347
1348 ump = VFSTOUFS(mp);
1349 qerror = 0;
1350 #ifdef QUOTA
1351 if (mp->mnt_flag & MNT_QUOTA) {
1352 int i;
1353 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1354 if (error)
1355 return (error);
1356 for (i = 0; i < MAXQUOTAS; i++) {
1357 error = quotaoff(td, mp, i);
1358 if (error != 0) {
1359 if ((flags & EARLYFLUSH) == 0)
1360 return (error);
1361 else
1362 qerror = error;
1363 }
1364 }
1365
1366 /*
1367 * Here we fall through to vflush again to ensure that
1368 * we have gotten rid of all the system vnodes, unless
1369 * quotas must not be closed.
1370 */
1371 }
1372 #endif
1373 /* devvp is not locked there */
1374 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1375 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1376 return (error);
1377 ffs_snapshot_unmount(mp);
1378 flags |= FORCECLOSE;
1379 /*
1380 * Here we fall through to vflush again to ensure
1381 * that we have gotten rid of all the system vnodes.
1382 */
1383 }
1384
1385 /*
1386 * Do not close system files if quotas were not closed, to be
1387 * able to sync the remaining dquots. The freeblks softupdate
1388 * workitems might hold a reference on a dquot, preventing
1389 * quotaoff() from completing. Next round of
1390 * softdep_flushworklist() iteration should process the
1391 * blockers, allowing the next run of quotaoff() to finally
1392 * flush held dquots.
1393 *
1394 * Otherwise, flush all the files.
1395 */
1396 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1397 return (error);
1398
1399 /*
1400 * If this is a forcible unmount and there were any files that
1401 * were unlinked but still open, then vflush() will have
1402 * truncated and freed those files, which might have started
1403 * some trim work. Wait here for any trims to complete
1404 * and process the blkfrees which follow the trims.
1405 * This may create more dirty devvp buffers and softdep deps.
1406 */
1407 if (ump->um_trim_tq != NULL) {
1408 while (ump->um_trim_inflight != 0)
1409 pause("ufsutr", hz);
1410 taskqueue_drain_all(ump->um_trim_tq);
1411 }
1412
1413 /*
1414 * Flush filesystem metadata.
1415 */
1416 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1417 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1418 VOP_UNLOCK(ump->um_devvp);
1419 return (error);
1420 }
1421
1422 /*
1423 * Get filesystem statistics.
1424 */
1425 static int
ffs_statfs(struct mount * mp,struct statfs * sbp)1426 ffs_statfs(struct mount *mp, struct statfs *sbp)
1427 {
1428 struct ufsmount *ump;
1429 struct fs *fs;
1430
1431 ump = VFSTOUFS(mp);
1432 fs = ump->um_fs;
1433 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1434 panic("ffs_statfs");
1435 sbp->f_version = STATFS_VERSION;
1436 sbp->f_bsize = fs->fs_fsize;
1437 sbp->f_iosize = fs->fs_bsize;
1438 sbp->f_blocks = fs->fs_dsize;
1439 UFS_LOCK(ump);
1440 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1441 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1442 sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1443 dbtofsb(fs, fs->fs_pendingblocks);
1444 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO;
1445 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1446 UFS_UNLOCK(ump);
1447 sbp->f_namemax = UFS_MAXNAMLEN;
1448 return (0);
1449 }
1450
1451 static bool
sync_doupdate(struct inode * ip)1452 sync_doupdate(struct inode *ip)
1453 {
1454
1455 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1456 IN_UPDATE)) != 0);
1457 }
1458
1459 static int
ffs_sync_lazy_filter(struct vnode * vp,void * arg __unused)1460 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused)
1461 {
1462 struct inode *ip;
1463
1464 /*
1465 * Flags are safe to access because ->v_data invalidation
1466 * is held off by listmtx.
1467 */
1468 if (vp->v_type == VNON)
1469 return (false);
1470 ip = VTOI(vp);
1471 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0)
1472 return (false);
1473 return (true);
1474 }
1475
1476 /*
1477 * For a lazy sync, we only care about access times, quotas and the
1478 * superblock. Other filesystem changes are already converted to
1479 * cylinder group blocks or inode blocks updates and are written to
1480 * disk by syncer.
1481 */
1482 static int
ffs_sync_lazy(struct mount * mp)1483 ffs_sync_lazy(struct mount *mp)
1484 {
1485 struct vnode *mvp, *vp;
1486 struct inode *ip;
1487 int allerror, error;
1488
1489 allerror = 0;
1490 if ((mp->mnt_flag & MNT_NOATIME) != 0) {
1491 #ifdef QUOTA
1492 qsync(mp);
1493 #endif
1494 goto sbupdate;
1495 }
1496 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) {
1497 if (vp->v_type == VNON) {
1498 VI_UNLOCK(vp);
1499 continue;
1500 }
1501 ip = VTOI(vp);
1502
1503 /*
1504 * The IN_ACCESS flag is converted to IN_MODIFIED by
1505 * ufs_close() and ufs_getattr() by the calls to
1506 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1507 * Test also all the other timestamp flags too, to pick up
1508 * any other cases that could be missed.
1509 */
1510 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1511 VI_UNLOCK(vp);
1512 continue;
1513 }
1514 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0)
1515 continue;
1516 #ifdef QUOTA
1517 qsyncvp(vp);
1518 #endif
1519 if (sync_doupdate(ip))
1520 error = ffs_update(vp, 0);
1521 if (error != 0)
1522 allerror = error;
1523 vput(vp);
1524 }
1525 sbupdate:
1526 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1527 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1528 allerror = error;
1529 return (allerror);
1530 }
1531
1532 /*
1533 * Go through the disk queues to initiate sandbagged IO;
1534 * go through the inodes to write those that have been modified;
1535 * initiate the writing of the super block if it has been modified.
1536 *
1537 * Note: we are always called with the filesystem marked busy using
1538 * vfs_busy().
1539 */
1540 static int
ffs_sync(struct mount * mp,int waitfor)1541 ffs_sync(struct mount *mp, int waitfor)
1542 {
1543 struct vnode *mvp, *vp, *devvp;
1544 struct thread *td;
1545 struct inode *ip;
1546 struct ufsmount *ump = VFSTOUFS(mp);
1547 struct fs *fs;
1548 int error, count, lockreq, allerror = 0;
1549 int suspend;
1550 int suspended;
1551 int secondary_writes;
1552 int secondary_accwrites;
1553 int softdep_deps;
1554 int softdep_accdeps;
1555 struct bufobj *bo;
1556
1557 suspend = 0;
1558 suspended = 0;
1559 td = curthread;
1560 fs = ump->um_fs;
1561 if (fs->fs_fmod != 0 && fs->fs_ronly != 0)
1562 panic("%s: ffs_sync: modification on read-only filesystem",
1563 fs->fs_fsmnt);
1564 if (waitfor == MNT_LAZY) {
1565 if (!rebooting)
1566 return (ffs_sync_lazy(mp));
1567 waitfor = MNT_NOWAIT;
1568 }
1569
1570 /*
1571 * Write back each (modified) inode.
1572 */
1573 lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1574 if (waitfor == MNT_SUSPEND) {
1575 suspend = 1;
1576 waitfor = MNT_WAIT;
1577 }
1578 if (waitfor == MNT_WAIT)
1579 lockreq = LK_EXCLUSIVE;
1580 lockreq |= LK_INTERLOCK;
1581 loop:
1582 /* Grab snapshot of secondary write counts */
1583 MNT_ILOCK(mp);
1584 secondary_writes = mp->mnt_secondary_writes;
1585 secondary_accwrites = mp->mnt_secondary_accwrites;
1586 MNT_IUNLOCK(mp);
1587
1588 /* Grab snapshot of softdep dependency counts */
1589 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1590
1591 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1592 /*
1593 * Depend on the vnode interlock to keep things stable enough
1594 * for a quick test. Since there might be hundreds of
1595 * thousands of vnodes, we cannot afford even a subroutine
1596 * call unless there's a good chance that we have work to do.
1597 */
1598 if (vp->v_type == VNON) {
1599 VI_UNLOCK(vp);
1600 continue;
1601 }
1602 ip = VTOI(vp);
1603 if ((ip->i_flag &
1604 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1605 vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1606 VI_UNLOCK(vp);
1607 continue;
1608 }
1609 if ((error = vget(vp, lockreq)) != 0) {
1610 if (error == ENOENT) {
1611 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1612 goto loop;
1613 }
1614 continue;
1615 }
1616 #ifdef QUOTA
1617 qsyncvp(vp);
1618 #endif
1619 for (;;) {
1620 error = ffs_syncvnode(vp, waitfor, 0);
1621 if (error == ERELOOKUP)
1622 continue;
1623 if (error != 0)
1624 allerror = error;
1625 break;
1626 }
1627 vput(vp);
1628 }
1629 /*
1630 * Force stale filesystem control information to be flushed.
1631 */
1632 if (waitfor == MNT_WAIT || rebooting) {
1633 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1634 allerror = error;
1635 if (ffs_fsfail_cleanup(ump, allerror))
1636 allerror = 0;
1637 /* Flushed work items may create new vnodes to clean */
1638 if (allerror == 0 && count)
1639 goto loop;
1640 }
1641
1642 devvp = ump->um_devvp;
1643 bo = &devvp->v_bufobj;
1644 BO_LOCK(bo);
1645 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1646 BO_UNLOCK(bo);
1647 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1648 error = VOP_FSYNC(devvp, waitfor, td);
1649 VOP_UNLOCK(devvp);
1650 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1651 error = ffs_sbupdate(ump, waitfor, 0);
1652 if (error != 0)
1653 allerror = error;
1654 if (ffs_fsfail_cleanup(ump, allerror))
1655 allerror = 0;
1656 if (allerror == 0 && waitfor == MNT_WAIT)
1657 goto loop;
1658 } else if (suspend != 0) {
1659 if (softdep_check_suspend(mp,
1660 devvp,
1661 softdep_deps,
1662 softdep_accdeps,
1663 secondary_writes,
1664 secondary_accwrites) != 0) {
1665 MNT_IUNLOCK(mp);
1666 goto loop; /* More work needed */
1667 }
1668 mtx_assert(MNT_MTX(mp), MA_OWNED);
1669 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1670 MNT_IUNLOCK(mp);
1671 suspended = 1;
1672 } else
1673 BO_UNLOCK(bo);
1674 /*
1675 * Write back modified superblock.
1676 */
1677 if (fs->fs_fmod != 0 &&
1678 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1679 allerror = error;
1680 if (ffs_fsfail_cleanup(ump, allerror))
1681 allerror = 0;
1682 return (allerror);
1683 }
1684
1685 int
ffs_vget(struct mount * mp,ino_t ino,int flags,struct vnode ** vpp)1686 ffs_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
1687 {
1688 return (ffs_vgetf(mp, ino, flags, vpp, 0));
1689 }
1690
1691 int
ffs_vgetf(struct mount * mp,ino_t ino,int flags,struct vnode ** vpp,int ffs_flags)1692 ffs_vgetf(struct mount *mp,
1693 ino_t ino,
1694 int flags,
1695 struct vnode **vpp,
1696 int ffs_flags)
1697 {
1698 struct fs *fs;
1699 struct inode *ip;
1700 struct ufsmount *ump;
1701 struct buf *bp;
1702 struct vnode *vp;
1703 daddr_t dbn;
1704 int error;
1705
1706 MPASS((ffs_flags & (FFSV_REPLACE | FFSV_REPLACE_DOOMED)) == 0 ||
1707 (flags & LK_EXCLUSIVE) != 0);
1708
1709 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1710 if (error != 0)
1711 return (error);
1712 if (*vpp != NULL) {
1713 if ((ffs_flags & FFSV_REPLACE) == 0 ||
1714 ((ffs_flags & FFSV_REPLACE_DOOMED) == 0 ||
1715 !VN_IS_DOOMED(*vpp)))
1716 return (0);
1717 vgone(*vpp);
1718 vput(*vpp);
1719 }
1720
1721 /*
1722 * We must promote to an exclusive lock for vnode creation. This
1723 * can happen if lookup is passed LOCKSHARED.
1724 */
1725 if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1726 flags &= ~LK_TYPE_MASK;
1727 flags |= LK_EXCLUSIVE;
1728 }
1729
1730 /*
1731 * We do not lock vnode creation as it is believed to be too
1732 * expensive for such rare case as simultaneous creation of vnode
1733 * for same ino by different processes. We just allow them to race
1734 * and check later to decide who wins. Let the race begin!
1735 */
1736
1737 ump = VFSTOUFS(mp);
1738 fs = ump->um_fs;
1739 ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO);
1740
1741 /* Allocate a new vnode/inode. */
1742 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1743 &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1744 if (error) {
1745 *vpp = NULL;
1746 uma_zfree_smr(uma_inode, ip);
1747 return (error);
1748 }
1749 /*
1750 * FFS supports recursive locking.
1751 */
1752 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
1753 VN_LOCK_AREC(vp);
1754 vp->v_data = ip;
1755 vp->v_bufobj.bo_bsize = fs->fs_bsize;
1756 ip->i_vnode = vp;
1757 ip->i_ump = ump;
1758 ip->i_number = ino;
1759 ip->i_ea_refs = 0;
1760 ip->i_nextclustercg = -1;
1761 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
1762 ip->i_mode = 0; /* ensure error cases below throw away vnode */
1763 cluster_init_vn(&ip->i_clusterw);
1764 #ifdef DIAGNOSTIC
1765 ufs_init_trackers(ip);
1766 #endif
1767 #ifdef QUOTA
1768 {
1769 int i;
1770 for (i = 0; i < MAXQUOTAS; i++)
1771 ip->i_dquot[i] = NODQUOT;
1772 }
1773 #endif
1774
1775 if (ffs_flags & FFSV_FORCEINSMQ)
1776 vp->v_vflag |= VV_FORCEINSMQ;
1777 error = insmntque(vp, mp);
1778 if (error != 0) {
1779 uma_zfree_smr(uma_inode, ip);
1780 *vpp = NULL;
1781 return (error);
1782 }
1783 vp->v_vflag &= ~VV_FORCEINSMQ;
1784 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1785 if (error != 0)
1786 return (error);
1787 if (*vpp != NULL) {
1788 /*
1789 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set)
1790 * operate on empty inode, which must not be found by
1791 * other threads until fully filled. Vnode for empty
1792 * inode must be not re-inserted on the hash by other
1793 * thread, after removal by us at the beginning.
1794 */
1795 MPASS((ffs_flags & FFSV_REPLACE) == 0);
1796 return (0);
1797 }
1798 if (I_IS_UFS1(ip))
1799 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1800 else
1801 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1802
1803 if ((ffs_flags & FFSV_NEWINODE) != 0) {
1804 /* New inode, just zero out its contents. */
1805 if (I_IS_UFS1(ip))
1806 memset(ip->i_din1, 0, sizeof(struct ufs1_dinode));
1807 else
1808 memset(ip->i_din2, 0, sizeof(struct ufs2_dinode));
1809 } else {
1810 /* Read the disk contents for the inode, copy into the inode. */
1811 dbn = fsbtodb(fs, ino_to_fsba(fs, ino));
1812 error = ffs_breadz(ump, ump->um_devvp, dbn, dbn,
1813 (int)fs->fs_bsize, NULL, NULL, 0, NOCRED, 0, NULL, &bp);
1814 if (error != 0) {
1815 /*
1816 * The inode does not contain anything useful, so it
1817 * would be misleading to leave it on its hash chain.
1818 * With mode still zero, it will be unlinked and
1819 * returned to the free list by vput().
1820 */
1821 vgone(vp);
1822 vput(vp);
1823 *vpp = NULL;
1824 return (error);
1825 }
1826 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) {
1827 bqrelse(bp);
1828 vgone(vp);
1829 vput(vp);
1830 *vpp = NULL;
1831 return (error);
1832 }
1833 bqrelse(bp);
1834 }
1835 if (DOINGSOFTDEP(vp) && (!fs->fs_ronly ||
1836 (ffs_flags & FFSV_FORCEINODEDEP) != 0))
1837 softdep_load_inodeblock(ip);
1838 else
1839 ip->i_effnlink = ip->i_nlink;
1840
1841 /*
1842 * Initialize the vnode from the inode, check for aliases.
1843 * Note that the underlying vnode may have changed.
1844 */
1845 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
1846 &vp);
1847 if (error) {
1848 vgone(vp);
1849 vput(vp);
1850 *vpp = NULL;
1851 return (error);
1852 }
1853
1854 /*
1855 * Finish inode initialization.
1856 */
1857 if (vp->v_type != VFIFO) {
1858 /* FFS supports shared locking for all files except fifos. */
1859 VN_LOCK_ASHARE(vp);
1860 }
1861
1862 /*
1863 * Set up a generation number for this inode if it does not
1864 * already have one. This should only happen on old filesystems.
1865 */
1866 if (ip->i_gen == 0) {
1867 while (ip->i_gen == 0)
1868 ip->i_gen = arc4random();
1869 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1870 UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
1871 DIP_SET(ip, i_gen, ip->i_gen);
1872 }
1873 }
1874 #ifdef MAC
1875 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1876 /*
1877 * If this vnode is already allocated, and we're running
1878 * multi-label, attempt to perform a label association
1879 * from the extended attributes on the inode.
1880 */
1881 error = mac_vnode_associate_extattr(mp, vp);
1882 if (error) {
1883 /* ufs_inactive will release ip->i_devvp ref. */
1884 vgone(vp);
1885 vput(vp);
1886 *vpp = NULL;
1887 return (error);
1888 }
1889 }
1890 #endif
1891
1892 vn_set_state(vp, VSTATE_CONSTRUCTED);
1893 *vpp = vp;
1894 return (0);
1895 }
1896
1897 /*
1898 * File handle to vnode
1899 *
1900 * Have to be really careful about stale file handles:
1901 * - check that the inode number is valid
1902 * - for UFS2 check that the inode number is initialized
1903 * - call ffs_vget() to get the locked inode
1904 * - check for an unallocated inode (i_mode == 0)
1905 * - check that the given client host has export rights and return
1906 * those rights via. exflagsp and credanonp
1907 */
1908 static int
ffs_fhtovp(struct mount * mp,struct fid * fhp,int flags,struct vnode ** vpp)1909 ffs_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1910 {
1911 struct ufid *ufhp;
1912
1913 ufhp = (struct ufid *)fhp;
1914 return (ffs_inotovp(mp, ufhp->ufid_ino, ufhp->ufid_gen, flags,
1915 vpp, 0));
1916 }
1917
1918 /*
1919 * Return a vnode from a mounted filesystem for inode with specified
1920 * generation number. Return ESTALE if the inode with given generation
1921 * number no longer exists on that filesystem.
1922 */
1923 int
ffs_inotovp(struct mount * mp,ino_t ino,uint64_t gen,int lflags,struct vnode ** vpp,int ffs_flags)1924 ffs_inotovp(struct mount *mp,
1925 ino_t ino,
1926 uint64_t gen,
1927 int lflags,
1928 struct vnode **vpp,
1929 int ffs_flags)
1930 {
1931 struct ufsmount *ump;
1932 struct vnode *nvp;
1933 struct inode *ip;
1934 struct fs *fs;
1935 struct cg *cgp;
1936 struct buf *bp;
1937 uint64_t cg;
1938
1939 ump = VFSTOUFS(mp);
1940 fs = ump->um_fs;
1941 *vpp = NULL;
1942
1943 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
1944 return (ESTALE);
1945
1946 /*
1947 * Need to check if inode is initialized because UFS2 does lazy
1948 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
1949 */
1950 if (fs->fs_magic == FS_UFS2_MAGIC) {
1951 cg = ino_to_cg(fs, ino);
1952 if (ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp) != 0)
1953 return (ESTALE);
1954 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
1955 brelse(bp);
1956 return (ESTALE);
1957 }
1958 brelse(bp);
1959 }
1960
1961 if (ffs_vgetf(mp, ino, lflags, &nvp, ffs_flags) != 0)
1962 return (ESTALE);
1963
1964 ip = VTOI(nvp);
1965 if (ip->i_mode == 0 || ip->i_gen != gen || ip->i_effnlink <= 0) {
1966 if (ip->i_mode == 0)
1967 vgone(nvp);
1968 vput(nvp);
1969 return (ESTALE);
1970 }
1971
1972 vnode_create_vobject(nvp, DIP(ip, i_size), curthread);
1973 *vpp = nvp;
1974 return (0);
1975 }
1976
1977 /*
1978 * Initialize the filesystem.
1979 */
1980 static int
ffs_init(struct vfsconf * vfsp)1981 ffs_init(struct vfsconf *vfsp)
1982 {
1983
1984 ffs_susp_initialize();
1985 softdep_initialize();
1986 return (ufs_init(vfsp));
1987 }
1988
1989 /*
1990 * Undo the work of ffs_init().
1991 */
1992 static int
ffs_uninit(struct vfsconf * vfsp)1993 ffs_uninit(struct vfsconf *vfsp)
1994 {
1995 int ret;
1996
1997 ret = ufs_uninit(vfsp);
1998 softdep_uninitialize();
1999 ffs_susp_uninitialize();
2000 taskqueue_drain_all(taskqueue_thread);
2001 return (ret);
2002 }
2003
2004 /*
2005 * Structure used to pass information from ffs_sbupdate to its
2006 * helper routine ffs_use_bwrite.
2007 */
2008 struct devfd {
2009 struct ufsmount *ump;
2010 struct buf *sbbp;
2011 int waitfor;
2012 int suspended;
2013 int error;
2014 };
2015
2016 /*
2017 * Write a superblock and associated information back to disk.
2018 */
2019 int
ffs_sbupdate(struct ufsmount * ump,int waitfor,int suspended)2020 ffs_sbupdate(struct ufsmount *ump, int waitfor, int suspended)
2021 {
2022 struct fs *fs;
2023 struct buf *sbbp;
2024 struct devfd devfd;
2025
2026 fs = ump->um_fs;
2027 if (fs->fs_ronly == 1 &&
2028 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
2029 (MNT_RDONLY | MNT_UPDATE))
2030 panic("ffs_sbupdate: write read-only filesystem");
2031 /*
2032 * We use the superblock's buf to serialize calls to ffs_sbupdate().
2033 * Copy superblock to this buffer and have it written out.
2034 */
2035 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
2036 (int)fs->fs_sbsize, 0, 0, 0);
2037 UFS_LOCK(ump);
2038 fs->fs_fmod = 0;
2039 bcopy((caddr_t)fs, sbbp->b_data, (uint64_t)fs->fs_sbsize);
2040 UFS_UNLOCK(ump);
2041 fs = (struct fs *)sbbp->b_data;
2042 /*
2043 * Initialize info needed for write function.
2044 */
2045 devfd.ump = ump;
2046 devfd.sbbp = sbbp;
2047 devfd.waitfor = waitfor;
2048 devfd.suspended = suspended;
2049 devfd.error = 0;
2050 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite));
2051 }
2052
2053 /*
2054 * Write function for use by filesystem-layer routines.
2055 */
2056 static int
ffs_use_bwrite(void * devfd,off_t loc,void * buf,int size)2057 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size)
2058 {
2059 struct devfd *devfdp;
2060 struct ufsmount *ump;
2061 struct buf *bp;
2062 struct fs *fs;
2063 int error;
2064
2065 devfdp = devfd;
2066 ump = devfdp->ump;
2067 bp = devfdp->sbbp;
2068 fs = (struct fs *)bp->b_data;
2069 /*
2070 * Writing the superblock summary information.
2071 */
2072 if (loc != fs->fs_sblockloc) {
2073 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0);
2074 bcopy(buf, bp->b_data, (uint64_t)size);
2075 if (devfdp->suspended)
2076 bp->b_flags |= B_VALIDSUSPWRT;
2077 if (devfdp->waitfor != MNT_WAIT)
2078 bawrite(bp);
2079 else if ((error = bwrite(bp)) != 0)
2080 devfdp->error = error;
2081 return (0);
2082 }
2083 /*
2084 * Writing the superblock itself. We need to do special checks for it.
2085 * A negative error code is returned to indicate that a copy of the
2086 * superblock has been made and that the copy is discarded when the
2087 * I/O is done. So the caller should not attempt to restore the
2088 * fs_si field after the write is done. The caller will convert the
2089 * error code back to its usual positive value when returning it.
2090 */
2091 if (ffs_fsfail_cleanup(ump, devfdp->error))
2092 devfdp->error = 0;
2093 if (devfdp->error != 0) {
2094 brelse(bp);
2095 return (-devfdp->error - 1);
2096 }
2097 if (MOUNTEDSOFTDEP(ump->um_mountp))
2098 softdep_setup_sbupdate(ump, fs, bp);
2099 if (devfdp->suspended)
2100 bp->b_flags |= B_VALIDSUSPWRT;
2101 if (devfdp->waitfor != MNT_WAIT)
2102 bawrite(bp);
2103 else if ((error = bwrite(bp)) != 0)
2104 devfdp->error = error;
2105 return (-devfdp->error - 1);
2106 }
2107
2108 static int
ffs_extattrctl(struct mount * mp,int cmd,struct vnode * filename_vp,int attrnamespace,const char * attrname)2109 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
2110 int attrnamespace, const char *attrname)
2111 {
2112
2113 #ifdef UFS_EXTATTR
2114 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
2115 attrname));
2116 #else
2117 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
2118 attrname));
2119 #endif
2120 }
2121
2122 static void
ffs_ifree(struct ufsmount * ump,struct inode * ip)2123 ffs_ifree(struct ufsmount *ump, struct inode *ip)
2124 {
2125
2126 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
2127 uma_zfree(uma_ufs1, ip->i_din1);
2128 else if (ip->i_din2 != NULL)
2129 uma_zfree(uma_ufs2, ip->i_din2);
2130 uma_zfree_smr(uma_inode, ip);
2131 }
2132
2133 static int dobkgrdwrite = 1;
2134 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2135 "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2136
2137 /*
2138 * Complete a background write started from bwrite.
2139 */
2140 static void
ffs_backgroundwritedone(struct buf * bp)2141 ffs_backgroundwritedone(struct buf *bp)
2142 {
2143 struct bufobj *bufobj;
2144 struct buf *origbp;
2145
2146 #ifdef SOFTUPDATES
2147 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0)
2148 softdep_handle_error(bp);
2149 #endif
2150
2151 /*
2152 * Find the original buffer that we are writing.
2153 */
2154 bufobj = bp->b_bufobj;
2155 BO_LOCK(bufobj);
2156 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2157 panic("backgroundwritedone: lost buffer");
2158
2159 /*
2160 * We should mark the cylinder group buffer origbp as
2161 * dirty, to not lose the failed write.
2162 */
2163 if ((bp->b_ioflags & BIO_ERROR) != 0)
2164 origbp->b_vflags |= BV_BKGRDERR;
2165 BO_UNLOCK(bufobj);
2166 /*
2167 * Process dependencies then return any unfinished ones.
2168 */
2169 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2170 buf_complete(bp);
2171 #ifdef SOFTUPDATES
2172 if (!LIST_EMPTY(&bp->b_dep))
2173 softdep_move_dependencies(bp, origbp);
2174 #endif
2175 /*
2176 * This buffer is marked B_NOCACHE so when it is released
2177 * by biodone it will be tossed. Clear B_IOSTARTED in case of error.
2178 */
2179 bp->b_flags |= B_NOCACHE;
2180 bp->b_flags &= ~(B_CACHE | B_IOSTARTED);
2181 pbrelvp(bp);
2182
2183 /*
2184 * Prevent brelse() from trying to keep and re-dirtying bp on
2185 * errors. It causes b_bufobj dereference in
2186 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2187 * pbrelvp() above.
2188 */
2189 if ((bp->b_ioflags & BIO_ERROR) != 0)
2190 bp->b_flags |= B_INVAL;
2191 bufdone(bp);
2192 BO_LOCK(bufobj);
2193 /*
2194 * Clear the BV_BKGRDINPROG flag in the original buffer
2195 * and awaken it if it is waiting for the write to complete.
2196 * If BV_BKGRDINPROG is not set in the original buffer it must
2197 * have been released and re-instantiated - which is not legal.
2198 */
2199 KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2200 ("backgroundwritedone: lost buffer2"));
2201 origbp->b_vflags &= ~BV_BKGRDINPROG;
2202 if (origbp->b_vflags & BV_BKGRDWAIT) {
2203 origbp->b_vflags &= ~BV_BKGRDWAIT;
2204 wakeup(&origbp->b_xflags);
2205 }
2206 BO_UNLOCK(bufobj);
2207 }
2208
2209 /*
2210 * Write, release buffer on completion. (Done by iodone
2211 * if async). Do not bother writing anything if the buffer
2212 * is invalid.
2213 *
2214 * Note that we set B_CACHE here, indicating that buffer is
2215 * fully valid and thus cacheable. This is true even of NFS
2216 * now so we set it generally. This could be set either here
2217 * or in biodone() since the I/O is synchronous. We put it
2218 * here.
2219 */
2220 static int
ffs_bufwrite(struct buf * bp)2221 ffs_bufwrite(struct buf *bp)
2222 {
2223 struct buf *newbp;
2224 struct cg *cgp;
2225
2226 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2227 if (bp->b_flags & B_INVAL) {
2228 brelse(bp);
2229 return (0);
2230 }
2231
2232 if (!BUF_ISLOCKED(bp))
2233 panic("bufwrite: buffer is not busy???");
2234 /*
2235 * If a background write is already in progress, delay
2236 * writing this block if it is asynchronous. Otherwise
2237 * wait for the background write to complete.
2238 */
2239 BO_LOCK(bp->b_bufobj);
2240 if (bp->b_vflags & BV_BKGRDINPROG) {
2241 if (bp->b_flags & B_ASYNC) {
2242 BO_UNLOCK(bp->b_bufobj);
2243 bdwrite(bp);
2244 return (0);
2245 }
2246 bp->b_vflags |= BV_BKGRDWAIT;
2247 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2248 "bwrbg", 0);
2249 if (bp->b_vflags & BV_BKGRDINPROG)
2250 panic("bufwrite: still writing");
2251 }
2252 bp->b_vflags &= ~BV_BKGRDERR;
2253 BO_UNLOCK(bp->b_bufobj);
2254
2255 /*
2256 * If this buffer is marked for background writing and we
2257 * do not have to wait for it, make a copy and write the
2258 * copy so as to leave this buffer ready for further use.
2259 *
2260 * This optimization eats a lot of memory. If we have a page
2261 * or buffer shortfall we can't do it.
2262 */
2263 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2264 (bp->b_flags & B_ASYNC) &&
2265 !vm_page_count_severe() &&
2266 !buf_dirty_count_severe()) {
2267 KASSERT(bp->b_iodone == NULL,
2268 ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2269
2270 /* get a new block */
2271 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2272 if (newbp == NULL)
2273 goto normal_write;
2274
2275 KASSERT(buf_mapped(bp), ("Unmapped cg"));
2276 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2277 BO_LOCK(bp->b_bufobj);
2278 bp->b_vflags |= BV_BKGRDINPROG;
2279 BO_UNLOCK(bp->b_bufobj);
2280 newbp->b_xflags |=
2281 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER;
2282 newbp->b_lblkno = bp->b_lblkno;
2283 newbp->b_blkno = bp->b_blkno;
2284 newbp->b_offset = bp->b_offset;
2285 newbp->b_iodone = ffs_backgroundwritedone;
2286 newbp->b_flags |= B_ASYNC;
2287 newbp->b_flags &= ~B_INVAL;
2288 pbgetvp(bp->b_vp, newbp);
2289
2290 #ifdef SOFTUPDATES
2291 /*
2292 * Move over the dependencies. If there are rollbacks,
2293 * leave the parent buffer dirtied as it will need to
2294 * be written again.
2295 */
2296 if (LIST_EMPTY(&bp->b_dep) ||
2297 softdep_move_dependencies(bp, newbp) == 0)
2298 bundirty(bp);
2299 #else
2300 bundirty(bp);
2301 #endif
2302
2303 /*
2304 * Initiate write on the copy, release the original. The
2305 * BKGRDINPROG flag prevents it from going away until
2306 * the background write completes. We have to recalculate
2307 * its check hash in case the buffer gets freed and then
2308 * reconstituted from the buffer cache during a later read.
2309 */
2310 if ((bp->b_xflags & BX_CYLGRP) != 0) {
2311 cgp = (struct cg *)bp->b_data;
2312 cgp->cg_ckhash = 0;
2313 cgp->cg_ckhash =
2314 calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2315 }
2316 bqrelse(bp);
2317 bp = newbp;
2318 } else
2319 /* Mark the buffer clean */
2320 bundirty(bp);
2321
2322 /* Let the normal bufwrite do the rest for us */
2323 normal_write:
2324 /*
2325 * If we are writing a cylinder group, update its time.
2326 */
2327 if ((bp->b_xflags & BX_CYLGRP) != 0) {
2328 cgp = (struct cg *)bp->b_data;
2329 cgp->cg_old_time = cgp->cg_time = time_second;
2330 }
2331 return (bufwrite(bp));
2332 }
2333
2334 static void
ffs_geom_strategy(struct bufobj * bo,struct buf * bp)2335 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2336 {
2337 struct vnode *vp;
2338 struct buf *tbp;
2339 int error, nocopy;
2340
2341 /*
2342 * This is the bufobj strategy for the private VCHR vnodes
2343 * used by FFS to access the underlying storage device.
2344 * We override the default bufobj strategy and thus bypass
2345 * VOP_STRATEGY() for these vnodes.
2346 */
2347 vp = bo2vnode(bo);
2348 KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR ||
2349 bp->b_vp->v_rdev == NULL ||
2350 bp->b_vp->v_rdev->si_mountpt == NULL ||
2351 VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL ||
2352 vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp,
2353 ("ffs_geom_strategy() with wrong vp"));
2354 if (bp->b_iocmd == BIO_WRITE) {
2355 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2356 bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2357 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2358 panic("ffs_geom_strategy: bad I/O");
2359 nocopy = bp->b_flags & B_NOCOPY;
2360 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2361 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2362 vp->v_rdev->si_snapdata != NULL) {
2363 if ((bp->b_flags & B_CLUSTER) != 0) {
2364 runningbufwakeup(bp);
2365 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2366 b_cluster.cluster_entry) {
2367 error = ffs_copyonwrite(vp, tbp);
2368 if (error != 0 &&
2369 error != EOPNOTSUPP) {
2370 bp->b_error = error;
2371 bp->b_ioflags |= BIO_ERROR;
2372 bp->b_flags &= ~B_BARRIER;
2373 bufdone(bp);
2374 return;
2375 }
2376 }
2377 (void)runningbufclaim(bp, bp->b_bufsize);
2378 } else {
2379 error = ffs_copyonwrite(vp, bp);
2380 if (error != 0 && error != EOPNOTSUPP) {
2381 bp->b_error = error;
2382 bp->b_ioflags |= BIO_ERROR;
2383 bp->b_flags &= ~B_BARRIER;
2384 bufdone(bp);
2385 return;
2386 }
2387 }
2388 }
2389 #ifdef SOFTUPDATES
2390 if ((bp->b_flags & B_CLUSTER) != 0) {
2391 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2392 b_cluster.cluster_entry) {
2393 if (!LIST_EMPTY(&tbp->b_dep))
2394 buf_start(tbp);
2395 }
2396 } else {
2397 if (!LIST_EMPTY(&bp->b_dep))
2398 buf_start(bp);
2399 }
2400
2401 #endif
2402 /*
2403 * Check for metadata that needs check-hashes and update them.
2404 */
2405 switch (bp->b_xflags & BX_FSPRIV) {
2406 case BX_CYLGRP:
2407 ((struct cg *)bp->b_data)->cg_ckhash = 0;
2408 ((struct cg *)bp->b_data)->cg_ckhash =
2409 calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2410 break;
2411
2412 case BX_SUPERBLOCK:
2413 case BX_INODE:
2414 case BX_INDIR:
2415 case BX_DIR:
2416 printf("Check-hash write is unimplemented!!!\n");
2417 break;
2418
2419 case 0:
2420 break;
2421
2422 default:
2423 printf("multiple buffer types 0x%b\n",
2424 (bp->b_xflags & BX_FSPRIV), PRINT_UFS_BUF_XFLAGS);
2425 break;
2426 }
2427 }
2428 if (bp->b_iocmd != BIO_READ && ffs_enxio_enable)
2429 bp->b_xflags |= BX_CVTENXIO;
2430 g_vfs_strategy(bo, bp);
2431 }
2432
2433 int
ffs_own_mount(const struct mount * mp)2434 ffs_own_mount(const struct mount *mp)
2435 {
2436
2437 if (mp->mnt_op == &ufs_vfsops)
2438 return (1);
2439 return (0);
2440 }
2441
2442 #ifdef DDB
2443 #ifdef SOFTUPDATES
2444
2445 /* defined in ffs_softdep.c */
2446 extern void db_print_ffs(struct ufsmount *ump);
2447
DB_SHOW_COMMAND(ffs,db_show_ffs)2448 DB_SHOW_COMMAND(ffs, db_show_ffs)
2449 {
2450 struct mount *mp;
2451 struct ufsmount *ump;
2452
2453 if (have_addr) {
2454 ump = VFSTOUFS((struct mount *)addr);
2455 db_print_ffs(ump);
2456 return;
2457 }
2458
2459 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2460 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2461 db_print_ffs(VFSTOUFS(mp));
2462 }
2463 }
2464
2465 #endif /* SOFTUPDATES */
2466 #endif /* DDB */
2467