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