1 /*-
2 * Copyright (c) 1989, 1991, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/6.4/sys/ufs/ffs/ffs_vfsops.c 179071 2008-05-17 12:46:24Z kib $");
34
35 #include "opt_mac.h"
36 #include "opt_quota.h"
37 #include "opt_ufs.h"
38 #include "opt_ffs.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/proc.h>
44 #include <sys/kernel.h>
45 #include <sys/mac.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
54
55 #include <ufs/ufs/extattr.h>
56 #include <ufs/ufs/quota.h>
57 #include <ufs/ufs/ufsmount.h>
58 #include <ufs/ufs/inode.h>
59 #include <ufs/ufs/ufs_extern.h>
60
61 #include <ufs/ffs/fs.h>
62 #include <ufs/ffs/ffs_extern.h>
63
64 #include <vm/vm.h>
65 #include <vm/uma.h>
66 #include <vm/vm_page.h>
67
68 #include <geom/geom.h>
69 #include <geom/geom_vfs.h>
70
71 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
72
73 static int ffs_sbupdate(struct ufsmount *, int, int);
74 static int ffs_reload(struct mount *, struct thread *);
75 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *);
76 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *,
77 ufs2_daddr_t);
78 static void ffs_oldfscompat_write(struct fs *, struct ufsmount *);
79 static void ffs_ifree(struct ufsmount *ump, struct inode *ip);
80 static vfs_init_t ffs_init;
81 static vfs_uninit_t ffs_uninit;
82 static vfs_extattrctl_t ffs_extattrctl;
83 static vfs_cmount_t ffs_cmount;
84 static vfs_unmount_t ffs_unmount;
85 static vfs_mount_t ffs_mount;
86 static vfs_statfs_t ffs_statfs;
87 static vfs_fhtovp_t ffs_fhtovp;
88 static vfs_vptofh_t ffs_vptofh;
89 static vfs_sync_t ffs_sync;
90
91 static struct vfsops ufs_vfsops = {
92 .vfs_extattrctl = ffs_extattrctl,
93 .vfs_fhtovp = ffs_fhtovp,
94 .vfs_init = ffs_init,
95 .vfs_mount = ffs_mount,
96 .vfs_cmount = ffs_cmount,
97 .vfs_quotactl = ufs_quotactl,
98 .vfs_root = ufs_root,
99 .vfs_statfs = ffs_statfs,
100 .vfs_sync = ffs_sync,
101 .vfs_uninit = ffs_uninit,
102 .vfs_unmount = ffs_unmount,
103 .vfs_vget = ffs_vget,
104 .vfs_vptofh = ffs_vptofh,
105 };
106
107 VFS_SET(ufs_vfsops, ufs, 0);
108 MODULE_VERSION(ufs, 1);
109
110 static b_strategy_t ffs_geom_strategy;
111 static b_write_t ffs_bufwrite;
112
113 static struct buf_ops ffs_ops = {
114 .bop_name = "FFS",
115 .bop_write = ffs_bufwrite,
116 .bop_strategy = ffs_geom_strategy,
117 .bop_sync = bufsync,
118 #ifdef NO_FFS_SNAPSHOT
119 .bop_bdflush = bufbdflush,
120 #else
121 .bop_bdflush = ffs_bdflush,
122 #endif
123 };
124
125 static const char *ffs_opts[] = { "acls", "async", "atime", "clusterr",
126 "clusterw", "exec", "export", "force", "from", "multilabel",
127 "snapshot", "suid", "suiddir", "symfollow", "sync",
128 "update", "union", NULL };
129
130 static int
131 ffs_mount(struct mount *mp, struct thread *td)
132 {
133 struct vnode *devvp;
134 struct ufsmount *ump = 0;
135 struct fs *fs;
136 int error, flags;
137 mode_t accessmode;
138 struct nameidata ndp;
139 char *fspec;
140
141 if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
142 return (EINVAL);
143 if (uma_inode == NULL) {
144 uma_inode = uma_zcreate("FFS inode",
145 sizeof(struct inode), NULL, NULL, NULL, NULL,
146 UMA_ALIGN_PTR, 0);
147 uma_ufs1 = uma_zcreate("FFS1 dinode",
148 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
149 UMA_ALIGN_PTR, 0);
150 uma_ufs2 = uma_zcreate("FFS2 dinode",
151 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
152 UMA_ALIGN_PTR, 0);
153 }
154
155 fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
156 if (error)
157 return (error);
158
159 /*
160 * If updating, check whether changing from read-only to
161 * read/write; if there is no device name, that's all we do.
162 */
163 if (mp->mnt_flag & MNT_UPDATE) {
164 ump = VFSTOUFS(mp);
165 fs = ump->um_fs;
166 devvp = ump->um_devvp;
167 if (fs->fs_ronly == 0 &&
168 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
169 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
170 return (error);
171 /*
172 * Flush any dirty data.
173 */
174 if ((error = ffs_sync(mp, MNT_WAIT, td)) != 0) {
175 vn_finished_write(mp);
176 return (error);
177 }
178 /*
179 * Check for and optionally get rid of files open
180 * for writing.
181 */
182 flags = WRITECLOSE;
183 if (mp->mnt_flag & MNT_FORCE)
184 flags |= FORCECLOSE;
185 if (mp->mnt_flag & MNT_SOFTDEP) {
186 error = softdep_flushfiles(mp, flags, td);
187 } else {
188 error = ffs_flushfiles(mp, flags, td);
189 }
190 if (error) {
191 vn_finished_write(mp);
192 return (error);
193 }
194 if (fs->fs_pendingblocks != 0 ||
195 fs->fs_pendinginodes != 0) {
196 printf("%s: %s: blocks %jd files %d\n",
197 fs->fs_fsmnt, "update error",
198 (intmax_t)fs->fs_pendingblocks,
199 fs->fs_pendinginodes);
200 fs->fs_pendingblocks = 0;
201 fs->fs_pendinginodes = 0;
202 }
203 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
204 fs->fs_clean = 1;
205 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
206 fs->fs_ronly = 0;
207 fs->fs_clean = 0;
208 vn_finished_write(mp);
209 return (error);
210 }
211 vn_finished_write(mp);
212 DROP_GIANT();
213 g_topology_lock();
214 g_access(ump->um_cp, 0, -1, 0);
215 g_topology_unlock();
216 PICKUP_GIANT();
217 fs->fs_ronly = 1;
218 MNT_ILOCK(mp);
219 mp->mnt_flag |= MNT_RDONLY;
220 MNT_IUNLOCK(mp);
221 }
222 if ((mp->mnt_flag & MNT_RELOAD) &&
223 (error = ffs_reload(mp, td)) != 0)
224 return (error);
225 if (fs->fs_ronly &&
226 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
227 /*
228 * If upgrade to read-write by non-root, then verify
229 * that user has necessary permissions on the device.
230 */
231 if (suser(td)) {
232 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
233 if ((error = VOP_ACCESS(devvp, VREAD | VWRITE,
234 td->td_ucred, td)) != 0) {
235 VOP_UNLOCK(devvp, 0, td);
236 return (error);
237 }
238 VOP_UNLOCK(devvp, 0, td);
239 }
240 fs->fs_flags &= ~FS_UNCLEAN;
241 if (fs->fs_clean == 0) {
242 fs->fs_flags |= FS_UNCLEAN;
243 if ((mp->mnt_flag & MNT_FORCE) ||
244 ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
245 (fs->fs_flags & FS_DOSOFTDEP))) {
246 printf("WARNING: %s was not %s\n",
247 fs->fs_fsmnt, "properly dismounted");
248 } else {
249 printf(
250 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n",
251 fs->fs_fsmnt);
252 return (EPERM);
253 }
254 }
255 DROP_GIANT();
256 g_topology_lock();
257 /*
258 * If we're the root device, we may not have an E count
259 * yet, get it now.
260 */
261 if (ump->um_cp->ace == 0)
262 error = g_access(ump->um_cp, 0, 1, 1);
263 else
264 error = g_access(ump->um_cp, 0, 1, 0);
265 g_topology_unlock();
266 PICKUP_GIANT();
267 if (error)
268 return (error);
269 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
270 return (error);
271 fs->fs_ronly = 0;
272 MNT_ILOCK(mp);
273 mp->mnt_flag &= ~MNT_RDONLY;
274 MNT_IUNLOCK(mp);
275 fs->fs_clean = 0;
276 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
277 vn_finished_write(mp);
278 return (error);
279 }
280 /* check to see if we need to start softdep */
281 if ((fs->fs_flags & FS_DOSOFTDEP) &&
282 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
283 vn_finished_write(mp);
284 return (error);
285 }
286 if (fs->fs_snapinum[0] != 0)
287 ffs_snapshot_mount(mp);
288 vn_finished_write(mp);
289 }
290 /*
291 * Soft updates is incompatible with "async",
292 * so if we are doing softupdates stop the user
293 * from setting the async flag in an update.
294 * Softdep_mount() clears it in an initial mount
295 * or ro->rw remount.
296 */
297 if (mp->mnt_flag & MNT_SOFTDEP) {
298 /* XXX: Reset too late ? */
299 MNT_ILOCK(mp);
300 mp->mnt_flag &= ~MNT_ASYNC;
301 MNT_IUNLOCK(mp);
302 }
303 /*
304 * Keep MNT_ACLS flag if it is stored in superblock.
305 */
306 if ((fs->fs_flags & FS_ACLS) != 0) {
307 /* XXX: Set too late ? */
308 MNT_ILOCK(mp);
309 mp->mnt_flag |= MNT_ACLS;
310 MNT_IUNLOCK(mp);
311 }
312
313 /*
314 * If this is a snapshot request, take the snapshot.
315 */
316 if (mp->mnt_flag & MNT_SNAPSHOT)
317 return (ffs_snapshot(mp, fspec));
318 }
319
320 /*
321 * Not an update, or updating the name: look up the name
322 * and verify that it refers to a sensible disk device.
323 */
324 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
325 if ((error = namei(&ndp)) != 0)
326 return (error);
327 NDFREE(&ndp, NDF_ONLY_PNBUF);
328 devvp = ndp.ni_vp;
329 if (!vn_isdisk(devvp, &error)) {
330 vput(devvp);
331 return (error);
332 }
333
334 /*
335 * If mount by non-root, then verify that user has necessary
336 * permissions on the device.
337 */
338 if (suser(td)) {
339 accessmode = VREAD;
340 if ((mp->mnt_flag & MNT_RDONLY) == 0)
341 accessmode |= VWRITE;
342 if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){
343 vput(devvp);
344 return (error);
345 }
346 }
347
348 if (mp->mnt_flag & MNT_UPDATE) {
349 /*
350 * Update only
351 *
352 * If it's not the same vnode, or at least the same device
353 * then it's not correct.
354 */
355
356 if (devvp->v_rdev != ump->um_devvp->v_rdev)
357 error = EINVAL; /* needs translation */
358 vput(devvp);
359 if (error)
360 return (error);
361 } else {
362 /*
363 * New mount
364 *
365 * We need the name for the mount point (also used for
366 * "last mounted on") copied in. If an error occurs,
367 * the mount point is discarded by the upper level code.
368 * Note that vfs_mount() populates f_mntonname for us.
369 */
370 if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
371 vrele(devvp);
372 return (error);
373 }
374 }
375 vfs_mountedfrom(mp, fspec);
376 return (0);
377 }
378
379 /*
380 * Compatibility with old mount system call.
381 */
382
383 static int
384 ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td)
385 {
386 struct ufs_args args;
387 int error;
388
389 if (data == NULL)
390 return (EINVAL);
391 error = copyin(data, &args, sizeof args);
392 if (error)
393 return (error);
394
395 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
396 ma = mount_arg(ma, "export", &args.export, sizeof args.export);
397 error = kernel_mount(ma, flags);
398
399 return (error);
400 }
401
402 /*
403 * Reload all incore data for a filesystem (used after running fsck on
404 * the root filesystem and finding things to fix). The filesystem must
405 * be mounted read-only.
406 *
407 * Things to do to update the mount:
408 * 1) invalidate all cached meta-data.
409 * 2) re-read superblock from disk.
410 * 3) re-read summary information from disk.
411 * 4) invalidate all inactive vnodes.
412 * 5) invalidate all cached file data.
413 * 6) re-read inode data for all active vnodes.
414 */
415 static int
416 ffs_reload(struct mount *mp, struct thread *td)
417 {
418 struct vnode *vp, *mvp, *devvp;
419 struct inode *ip;
420 void *space;
421 struct buf *bp;
422 struct fs *fs, *newfs;
423 struct ufsmount *ump;
424 ufs2_daddr_t sblockloc;
425 int i, blks, size, error;
426 int32_t *lp;
427
428 if ((mp->mnt_flag & MNT_RDONLY) == 0)
429 return (EINVAL);
430 ump = VFSTOUFS(mp);
431 /*
432 * Step 1: invalidate all cached meta-data.
433 */
434 devvp = VFSTOUFS(mp)->um_devvp;
435 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
436 if (vinvalbuf(devvp, 0, td, 0, 0) != 0)
437 panic("ffs_reload: dirty1");
438 VOP_UNLOCK(devvp, 0, td);
439
440 /*
441 * Step 2: re-read superblock from disk.
442 */
443 fs = VFSTOUFS(mp)->um_fs;
444 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
445 NOCRED, &bp)) != 0)
446 return (error);
447 newfs = (struct fs *)bp->b_data;
448 if ((newfs->fs_magic != FS_UFS1_MAGIC &&
449 newfs->fs_magic != FS_UFS2_MAGIC) ||
450 newfs->fs_bsize > MAXBSIZE ||
451 newfs->fs_bsize < sizeof(struct fs)) {
452 brelse(bp);
453 return (EIO); /* XXX needs translation */
454 }
455 /*
456 * Copy pointer fields back into superblock before copying in XXX
457 * new superblock. These should really be in the ufsmount. XXX
458 * Note that important parameters (eg fs_ncg) are unchanged.
459 */
460 newfs->fs_csp = fs->fs_csp;
461 newfs->fs_maxcluster = fs->fs_maxcluster;
462 newfs->fs_contigdirs = fs->fs_contigdirs;
463 newfs->fs_active = fs->fs_active;
464 /* The file system is still read-only. */
465 newfs->fs_ronly = 1;
466 sblockloc = fs->fs_sblockloc;
467 bcopy(newfs, fs, (u_int)fs->fs_sbsize);
468 brelse(bp);
469 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
470 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
471 UFS_LOCK(ump);
472 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
473 printf("%s: reload pending error: blocks %jd files %d\n",
474 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
475 fs->fs_pendinginodes);
476 fs->fs_pendingblocks = 0;
477 fs->fs_pendinginodes = 0;
478 }
479 UFS_UNLOCK(ump);
480
481 /*
482 * Step 3: re-read summary information from disk.
483 */
484 blks = howmany(fs->fs_cssize, fs->fs_fsize);
485 space = fs->fs_csp;
486 for (i = 0; i < blks; i += fs->fs_frag) {
487 size = fs->fs_bsize;
488 if (i + fs->fs_frag > blks)
489 size = (blks - i) * fs->fs_fsize;
490 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
491 NOCRED, &bp);
492 if (error)
493 return (error);
494 bcopy(bp->b_data, space, (u_int)size);
495 space = (char *)space + size;
496 brelse(bp);
497 }
498 /*
499 * We no longer know anything about clusters per cylinder group.
500 */
501 if (fs->fs_contigsumsize > 0) {
502 lp = fs->fs_maxcluster;
503 for (i = 0; i < fs->fs_ncg; i++)
504 *lp++ = fs->fs_contigsumsize;
505 }
506
507 loop:
508 MNT_ILOCK(mp);
509 MNT_VNODE_FOREACH(vp, mp, mvp) {
510 VI_LOCK(vp);
511 if (vp->v_iflag & VI_DOOMED) {
512 VI_UNLOCK(vp);
513 continue;
514 }
515 MNT_IUNLOCK(mp);
516 /*
517 * Step 4: invalidate all cached file data.
518 */
519 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
520 MNT_VNODE_FOREACH_ABORT(mp, mvp);
521 goto loop;
522 }
523 if (vinvalbuf(vp, 0, td, 0, 0))
524 panic("ffs_reload: dirty2");
525 /*
526 * Step 5: re-read inode data for all active vnodes.
527 */
528 ip = VTOI(vp);
529 error =
530 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
531 (int)fs->fs_bsize, NOCRED, &bp);
532 if (error) {
533 VOP_UNLOCK(vp, 0, td);
534 vrele(vp);
535 MNT_VNODE_FOREACH_ABORT(mp, mvp);
536 return (error);
537 }
538 ffs_load_inode(bp, ip, fs, ip->i_number);
539 ip->i_effnlink = ip->i_nlink;
540 brelse(bp);
541 VOP_UNLOCK(vp, 0, td);
542 vrele(vp);
543 MNT_ILOCK(mp);
544 }
545 MNT_IUNLOCK(mp);
546 return (0);
547 }
548
549 /*
550 * Possible superblock locations ordered from most to least likely.
551 */
552 static int sblock_try[] = SBLOCKSEARCH;
553
554 /*
555 * Common code for mount and mountroot
556 */
557 static int
558 ffs_mountfs(devvp, mp, td)
559 struct vnode *devvp;
560 struct mount *mp;
561 struct thread *td;
562 {
563 struct ufsmount *ump;
564 struct buf *bp;
565 struct fs *fs;
566 struct cdev *dev;
567 void *space;
568 ufs2_daddr_t sblockloc;
569 int error, i, blks, size, ronly;
570 int32_t *lp;
571 struct ucred *cred;
572 struct g_consumer *cp;
573 struct mount *nmp;
574
575 dev = devvp->v_rdev;
576 cred = td ? td->td_ucred : NOCRED;
577
578 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
579 DROP_GIANT();
580 g_topology_lock();
581 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
582
583 /*
584 * If we are a root mount, drop the E flag so fsck can do its magic.
585 * We will pick it up again when we remount R/W.
586 */
587 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS))
588 error = g_access(cp, 0, 0, -1);
589 g_topology_unlock();
590 PICKUP_GIANT();
591 VOP_UNLOCK(devvp, 0, td);
592 if (error)
593 return (error);
594 if (devvp->v_rdev->si_iosize_max != 0)
595 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
596 if (mp->mnt_iosize_max > MAXPHYS)
597 mp->mnt_iosize_max = MAXPHYS;
598
599 devvp->v_bufobj.bo_private = cp;
600 devvp->v_bufobj.bo_ops = &ffs_ops;
601
602 bp = NULL;
603 ump = NULL;
604 fs = NULL;
605 sblockloc = 0;
606 /*
607 * Try reading the superblock in each of its possible locations.
608 */
609 for (i = 0; sblock_try[i] != -1; i++) {
610 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
611 error = EINVAL;
612 printf("Invalid sectorsize %d for superblock size %d\n",
613 cp->provider->sectorsize, SBLOCKSIZE);
614 goto out;
615 }
616 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
617 cred, &bp)) != 0)
618 goto out;
619 fs = (struct fs *)bp->b_data;
620 sblockloc = sblock_try[i];
621 if ((fs->fs_magic == FS_UFS1_MAGIC ||
622 (fs->fs_magic == FS_UFS2_MAGIC &&
623 (fs->fs_sblockloc == sblockloc ||
624 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
625 fs->fs_bsize <= MAXBSIZE &&
626 fs->fs_bsize >= sizeof(struct fs))
627 break;
628 brelse(bp);
629 bp = NULL;
630 }
631 if (sblock_try[i] == -1) {
632 error = EINVAL; /* XXX needs translation */
633 goto out;
634 }
635 fs->fs_fmod = 0;
636 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */
637 fs->fs_flags &= ~FS_UNCLEAN;
638 if (fs->fs_clean == 0) {
639 fs->fs_flags |= FS_UNCLEAN;
640 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
641 ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
642 (fs->fs_flags & FS_DOSOFTDEP))) {
643 printf(
644 "WARNING: %s was not properly dismounted\n",
645 fs->fs_fsmnt);
646 } else {
647 printf(
648 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n",
649 fs->fs_fsmnt);
650 error = EPERM;
651 goto out;
652 }
653 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
654 (mp->mnt_flag & MNT_FORCE)) {
655 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
656 (intmax_t)fs->fs_pendingblocks,
657 fs->fs_pendinginodes);
658 fs->fs_pendingblocks = 0;
659 fs->fs_pendinginodes = 0;
660 }
661 }
662 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
663 printf("%s: mount pending error: blocks %jd files %d\n",
664 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
665 fs->fs_pendinginodes);
666 fs->fs_pendingblocks = 0;
667 fs->fs_pendinginodes = 0;
668 }
669 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
670 ump->um_cp = cp;
671 ump->um_bo = &devvp->v_bufobj;
672 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
673 if (fs->fs_magic == FS_UFS1_MAGIC) {
674 ump->um_fstype = UFS1;
675 ump->um_balloc = ffs_balloc_ufs1;
676 } else {
677 ump->um_fstype = UFS2;
678 ump->um_balloc = ffs_balloc_ufs2;
679 }
680 ump->um_blkatoff = ffs_blkatoff;
681 ump->um_truncate = ffs_truncate;
682 ump->um_update = ffs_update;
683 ump->um_valloc = ffs_valloc;
684 ump->um_vfree = ffs_vfree;
685 ump->um_ifree = ffs_ifree;
686 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
687 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
688 if (fs->fs_sbsize < SBLOCKSIZE)
689 bp->b_flags |= B_INVAL | B_NOCACHE;
690 brelse(bp);
691 bp = NULL;
692 fs = ump->um_fs;
693 ffs_oldfscompat_read(fs, ump, sblockloc);
694 fs->fs_ronly = ronly;
695 size = fs->fs_cssize;
696 blks = howmany(size, fs->fs_fsize);
697 if (fs->fs_contigsumsize > 0)
698 size += fs->fs_ncg * sizeof(int32_t);
699 size += fs->fs_ncg * sizeof(u_int8_t);
700 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
701 fs->fs_csp = space;
702 for (i = 0; i < blks; i += fs->fs_frag) {
703 size = fs->fs_bsize;
704 if (i + fs->fs_frag > blks)
705 size = (blks - i) * fs->fs_fsize;
706 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
707 cred, &bp)) != 0) {
708 free(fs->fs_csp, M_UFSMNT);
709 goto out;
710 }
711 bcopy(bp->b_data, space, (u_int)size);
712 space = (char *)space + size;
713 brelse(bp);
714 bp = NULL;
715 }
716 if (fs->fs_contigsumsize > 0) {
717 fs->fs_maxcluster = lp = space;
718 for (i = 0; i < fs->fs_ncg; i++)
719 *lp++ = fs->fs_contigsumsize;
720 space = lp;
721 }
722 size = fs->fs_ncg * sizeof(u_int8_t);
723 fs->fs_contigdirs = (u_int8_t *)space;
724 bzero(fs->fs_contigdirs, size);
725 fs->fs_active = NULL;
726 mp->mnt_data = (qaddr_t)ump;
727 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
728 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
729 nmp = NULL;
730 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
731 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
732 if (nmp)
733 vfs_rel(nmp);
734 vfs_getnewfsid(mp);
735 }
736 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
737 MNT_ILOCK(mp);
738 mp->mnt_flag |= MNT_LOCAL;
739 MNT_IUNLOCK(mp);
740 if ((fs->fs_flags & FS_MULTILABEL) != 0) {
741 #ifdef MAC
742 MNT_ILOCK(mp);
743 mp->mnt_flag |= MNT_MULTILABEL;
744 MNT_IUNLOCK(mp);
745 #else
746 printf(
747 "WARNING: %s: multilabel flag on fs but no MAC support\n",
748 mp->mnt_stat.f_mntonname);
749 #endif
750 }
751 if ((fs->fs_flags & FS_ACLS) != 0) {
752 #ifdef UFS_ACL
753 MNT_ILOCK(mp);
754 mp->mnt_flag |= MNT_ACLS;
755 MNT_IUNLOCK(mp);
756 #else
757 printf(
758 "WARNING: %s: ACLs flag on fs but no ACLs support\n",
759 mp->mnt_stat.f_mntonname);
760 #endif
761 }
762 ump->um_mountp = mp;
763 ump->um_dev = dev;
764 ump->um_devvp = devvp;
765 ump->um_nindir = fs->fs_nindir;
766 ump->um_bptrtodb = fs->fs_fsbtodb;
767 ump->um_seqinc = fs->fs_frag;
768 for (i = 0; i < MAXQUOTAS; i++)
769 ump->um_quotas[i] = NULLVP;
770 #ifdef UFS_EXTATTR
771 ufs_extattr_uepm_init(&ump->um_extattr);
772 #endif
773 /*
774 * Set FS local "last mounted on" information (NULL pad)
775 */
776 bzero(fs->fs_fsmnt, MAXMNTLEN);
777 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
778
779 if( mp->mnt_flag & MNT_ROOTFS) {
780 /*
781 * Root mount; update timestamp in mount structure.
782 * this will be used by the common root mount code
783 * to update the system clock.
784 */
785 mp->mnt_time = fs->fs_time;
786 }
787
788 if (ronly == 0) {
789 if ((fs->fs_flags & FS_DOSOFTDEP) &&
790 (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
791 free(fs->fs_csp, M_UFSMNT);
792 goto out;
793 }
794 if (fs->fs_snapinum[0] != 0)
795 ffs_snapshot_mount(mp);
796 fs->fs_fmod = 1;
797 fs->fs_clean = 0;
798 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
799 }
800 /*
801 * Initialize filesystem stat information in mount struct.
802 */
803 #ifdef UFS_EXTATTR
804 #ifdef UFS_EXTATTR_AUTOSTART
805 /*
806 *
807 * Auto-starting does the following:
808 * - check for /.attribute in the fs, and extattr_start if so
809 * - for each file in .attribute, enable that file with
810 * an attribute of the same name.
811 * Not clear how to report errors -- probably eat them.
812 * This would all happen while the filesystem was busy/not
813 * available, so would effectively be "atomic".
814 */
815 (void) ufs_extattr_autostart(mp, td);
816 #endif /* !UFS_EXTATTR_AUTOSTART */
817 #endif /* !UFS_EXTATTR */
818 MNT_ILOCK(mp);
819 mp->mnt_kern_flag |= MNTK_MPSAFE;
820 MNT_IUNLOCK(mp);
821 return (0);
822 out:
823 if (bp)
824 brelse(bp);
825 if (cp != NULL) {
826 DROP_GIANT();
827 g_topology_lock();
828 g_vfs_close(cp, td);
829 g_topology_unlock();
830 PICKUP_GIANT();
831 }
832 if (ump) {
833 mtx_destroy(UFS_MTX(ump));
834 free(ump->um_fs, M_UFSMNT);
835 free(ump, M_UFSMNT);
836 mp->mnt_data = (qaddr_t)0;
837 }
838 return (error);
839 }
840
841 #include <sys/sysctl.h>
842 static int bigcgs = 0;
843 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
844
845 /*
846 * Sanity checks for loading old filesystem superblocks.
847 * See ffs_oldfscompat_write below for unwound actions.
848 *
849 * XXX - Parts get retired eventually.
850 * Unfortunately new bits get added.
851 */
852 static void
853 ffs_oldfscompat_read(fs, ump, sblockloc)
854 struct fs *fs;
855 struct ufsmount *ump;
856 ufs2_daddr_t sblockloc;
857 {
858 off_t maxfilesize;
859
860 /*
861 * If not yet done, update fs_flags location and value of fs_sblockloc.
862 */
863 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
864 fs->fs_flags = fs->fs_old_flags;
865 fs->fs_old_flags |= FS_FLAGS_UPDATED;
866 fs->fs_sblockloc = sblockloc;
867 }
868 /*
869 * If not yet done, update UFS1 superblock with new wider fields.
870 */
871 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
872 fs->fs_maxbsize = fs->fs_bsize;
873 fs->fs_time = fs->fs_old_time;
874 fs->fs_size = fs->fs_old_size;
875 fs->fs_dsize = fs->fs_old_dsize;
876 fs->fs_csaddr = fs->fs_old_csaddr;
877 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
878 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
879 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
880 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
881 }
882 if (fs->fs_magic == FS_UFS1_MAGIC &&
883 fs->fs_old_inodefmt < FS_44INODEFMT) {
884 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
885 fs->fs_qbmask = ~fs->fs_bmask;
886 fs->fs_qfmask = ~fs->fs_fmask;
887 }
888 if (fs->fs_magic == FS_UFS1_MAGIC) {
889 ump->um_savedmaxfilesize = fs->fs_maxfilesize;
890 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
891 if (fs->fs_maxfilesize > maxfilesize)
892 fs->fs_maxfilesize = maxfilesize;
893 }
894 /* Compatibility for old filesystems */
895 if (fs->fs_avgfilesize <= 0)
896 fs->fs_avgfilesize = AVFILESIZ;
897 if (fs->fs_avgfpdir <= 0)
898 fs->fs_avgfpdir = AFPDIR;
899 if (bigcgs) {
900 fs->fs_save_cgsize = fs->fs_cgsize;
901 fs->fs_cgsize = fs->fs_bsize;
902 }
903 }
904
905 /*
906 * Unwinding superblock updates for old filesystems.
907 * See ffs_oldfscompat_read above for details.
908 *
909 * XXX - Parts get retired eventually.
910 * Unfortunately new bits get added.
911 */
912 static void
913 ffs_oldfscompat_write(fs, ump)
914 struct fs *fs;
915 struct ufsmount *ump;
916 {
917
918 /*
919 * Copy back UFS2 updated fields that UFS1 inspects.
920 */
921 if (fs->fs_magic == FS_UFS1_MAGIC) {
922 fs->fs_old_time = fs->fs_time;
923 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
924 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
925 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
926 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
927 fs->fs_maxfilesize = ump->um_savedmaxfilesize;
928 }
929 if (bigcgs) {
930 fs->fs_cgsize = fs->fs_save_cgsize;
931 fs->fs_save_cgsize = 0;
932 }
933 }
934
935 /*
936 * unmount system call
937 */
938 static int
939 ffs_unmount(mp, mntflags, td)
940 struct mount *mp;
941 int mntflags;
942 struct thread *td;
943 {
944 struct ufsmount *ump = VFSTOUFS(mp);
945 struct fs *fs;
946 int error, flags;
947
948 flags = 0;
949 if (mntflags & MNT_FORCE) {
950 flags |= FORCECLOSE;
951 }
952 #ifdef UFS_EXTATTR
953 if ((error = ufs_extattr_stop(mp, td))) {
954 if (error != EOPNOTSUPP)
955 printf("ffs_unmount: ufs_extattr_stop returned %d\n",
956 error);
957 } else {
958 ufs_extattr_uepm_destroy(&ump->um_extattr);
959 }
960 #endif
961 if (mp->mnt_flag & MNT_SOFTDEP) {
962 if ((error = softdep_flushfiles(mp, flags, td)) != 0)
963 return (error);
964 } else {
965 if ((error = ffs_flushfiles(mp, flags, td)) != 0)
966 return (error);
967 }
968 fs = ump->um_fs;
969 UFS_LOCK(ump);
970 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
971 printf("%s: unmount pending error: blocks %jd files %d\n",
972 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
973 fs->fs_pendinginodes);
974 fs->fs_pendingblocks = 0;
975 fs->fs_pendinginodes = 0;
976 }
977 UFS_UNLOCK(ump);
978 if (fs->fs_ronly == 0) {
979 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
980 error = ffs_sbupdate(ump, MNT_WAIT, 0);
981 if (error) {
982 fs->fs_clean = 0;
983 return (error);
984 }
985 }
986 DROP_GIANT();
987 g_topology_lock();
988 g_vfs_close(ump->um_cp, td);
989 g_topology_unlock();
990 PICKUP_GIANT();
991 vrele(ump->um_devvp);
992 mtx_destroy(UFS_MTX(ump));
993 free(fs->fs_csp, M_UFSMNT);
994 free(fs, M_UFSMNT);
995 free(ump, M_UFSMNT);
996 mp->mnt_data = (qaddr_t)0;
997 MNT_ILOCK(mp);
998 mp->mnt_flag &= ~MNT_LOCAL;
999 MNT_IUNLOCK(mp);
1000 return (error);
1001 }
1002
1003 /*
1004 * Flush out all the files in a filesystem.
1005 */
1006 int
1007 ffs_flushfiles(mp, flags, td)
1008 struct mount *mp;
1009 int flags;
1010 struct thread *td;
1011 {
1012 struct ufsmount *ump;
1013 int error;
1014
1015 ump = VFSTOUFS(mp);
1016 #ifdef QUOTA
1017 if (mp->mnt_flag & MNT_QUOTA) {
1018 int i;
1019 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1020 if (error)
1021 return (error);
1022 for (i = 0; i < MAXQUOTAS; i++) {
1023 quotaoff(td, mp, i);
1024 }
1025 /*
1026 * Here we fall through to vflush again to ensure
1027 * that we have gotten rid of all the system vnodes.
1028 */
1029 }
1030 #endif
1031 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1032 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1033 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1034 return (error);
1035 ffs_snapshot_unmount(mp);
1036 flags |= FORCECLOSE;
1037 /*
1038 * Here we fall through to vflush again to ensure
1039 * that we have gotten rid of all the system vnodes.
1040 */
1041 }
1042 /*
1043 * Flush all the files.
1044 */
1045 if ((error = vflush(mp, 0, flags, td)) != 0)
1046 return (error);
1047 /*
1048 * Flush filesystem metadata.
1049 */
1050 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td);
1051 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1052 VOP_UNLOCK(ump->um_devvp, 0, td);
1053 return (error);
1054 }
1055
1056 /*
1057 * Get filesystem statistics.
1058 */
1059 static int
1060 ffs_statfs(mp, sbp, td)
1061 struct mount *mp;
1062 struct statfs *sbp;
1063 struct thread *td;
1064 {
1065 struct ufsmount *ump;
1066 struct fs *fs;
1067
1068 ump = VFSTOUFS(mp);
1069 fs = ump->um_fs;
1070 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1071 panic("ffs_statfs");
1072 sbp->f_version = STATFS_VERSION;
1073 sbp->f_bsize = fs->fs_fsize;
1074 sbp->f_iosize = fs->fs_bsize;
1075 sbp->f_blocks = fs->fs_dsize;
1076 UFS_LOCK(ump);
1077 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1078 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1079 sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1080 dbtofsb(fs, fs->fs_pendingblocks);
1081 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO;
1082 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1083 UFS_UNLOCK(ump);
1084 sbp->f_namemax = NAME_MAX;
1085 return (0);
1086 }
1087
1088 /*
1089 * Go through the disk queues to initiate sandbagged IO;
1090 * go through the inodes to write those that have been modified;
1091 * initiate the writing of the super block if it has been modified.
1092 *
1093 * Note: we are always called with the filesystem marked `MPBUSY'.
1094 */
1095 static int
1096 ffs_sync(mp, waitfor, td)
1097 struct mount *mp;
1098 int waitfor;
1099 struct thread *td;
1100 {
1101 struct vnode *mvp, *vp, *devvp;
1102 struct inode *ip;
1103 struct ufsmount *ump = VFSTOUFS(mp);
1104 struct fs *fs;
1105 int error, count, wait, lockreq, allerror = 0;
1106 int suspend;
1107 int suspended;
1108 int secondary_writes;
1109 int secondary_accwrites;
1110 int softdep_deps;
1111 int softdep_accdeps;
1112 struct bufobj *bo;
1113
1114 fs = ump->um_fs;
1115 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */
1116 printf("fs = %s\n", fs->fs_fsmnt);
1117 panic("ffs_sync: rofs mod");
1118 }
1119 /*
1120 * Write back each (modified) inode.
1121 */
1122 wait = 0;
1123 suspend = 0;
1124 suspended = 0;
1125 lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1126 if (waitfor == MNT_SUSPEND) {
1127 suspend = 1;
1128 waitfor = MNT_WAIT;
1129 }
1130 if (waitfor == MNT_WAIT) {
1131 wait = 1;
1132 lockreq = LK_EXCLUSIVE;
1133 }
1134 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1135 MNT_ILOCK(mp);
1136 loop:
1137 /* Grab snapshot of secondary write counts */
1138 secondary_writes = mp->mnt_secondary_writes;
1139 secondary_accwrites = mp->mnt_secondary_accwrites;
1140
1141 /* Grab snapshot of softdep dependency counts */
1142 MNT_IUNLOCK(mp);
1143 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1144 MNT_ILOCK(mp);
1145
1146 MNT_VNODE_FOREACH(vp, mp, mvp) {
1147 /*
1148 * Depend on the mntvnode_slock to keep things stable enough
1149 * for a quick test. Since there might be hundreds of
1150 * thousands of vnodes, we cannot afford even a subroutine
1151 * call unless there's a good chance that we have work to do.
1152 */
1153 VI_LOCK(vp);
1154 if (vp->v_iflag & VI_DOOMED) {
1155 VI_UNLOCK(vp);
1156 continue;
1157 }
1158 ip = VTOI(vp);
1159 if (vp->v_type == VNON || ((ip->i_flag &
1160 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1161 vp->v_bufobj.bo_dirty.bv_cnt == 0)) {
1162 VI_UNLOCK(vp);
1163 continue;
1164 }
1165 MNT_IUNLOCK(mp);
1166 if ((error = vget(vp, lockreq, td)) != 0) {
1167 MNT_ILOCK(mp);
1168 if (error == ENOENT || error == ENOLCK) {
1169 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
1170 goto loop;
1171 }
1172 continue;
1173 }
1174 if ((error = ffs_syncvnode(vp, waitfor)) != 0)
1175 allerror = error;
1176 vput(vp);
1177 MNT_ILOCK(mp);
1178 }
1179 MNT_IUNLOCK(mp);
1180 /*
1181 * Force stale filesystem control information to be flushed.
1182 */
1183 if (waitfor == MNT_WAIT) {
1184 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1185 allerror = error;
1186 /* Flushed work items may create new vnodes to clean */
1187 if (allerror == 0 && count) {
1188 MNT_ILOCK(mp);
1189 goto loop;
1190 }
1191 }
1192 #ifdef QUOTA
1193 qsync(mp);
1194 #endif
1195 devvp = ump->um_devvp;
1196 VI_LOCK(devvp);
1197 bo = &devvp->v_bufobj;
1198 if (waitfor != MNT_LAZY &&
1199 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
1200 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td);
1201 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0)
1202 allerror = error;
1203 VOP_UNLOCK(devvp, 0, td);
1204 if (allerror == 0 && waitfor == MNT_WAIT) {
1205 MNT_ILOCK(mp);
1206 goto loop;
1207 }
1208 } else if (suspend != 0) {
1209 if (softdep_check_suspend(mp,
1210 devvp,
1211 softdep_deps,
1212 softdep_accdeps,
1213 secondary_writes,
1214 secondary_accwrites) != 0)
1215 goto loop; /* More work needed */
1216 mtx_assert(MNT_MTX(mp), MA_OWNED);
1217 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1218 MNT_IUNLOCK(mp);
1219 suspended = 1;
1220 } else
1221 VI_UNLOCK(devvp);
1222 /*
1223 * Write back modified superblock.
1224 */
1225 if (fs->fs_fmod != 0 &&
1226 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1227 allerror = error;
1228 return (allerror);
1229 }
1230
1231 int
1232 ffs_vget(mp, ino, flags, vpp)
1233 struct mount *mp;
1234 ino_t ino;
1235 int flags;
1236 struct vnode **vpp;
1237 {
1238 struct fs *fs;
1239 struct inode *ip;
1240 struct ufsmount *ump;
1241 struct buf *bp;
1242 struct vnode *vp;
1243 struct cdev *dev;
1244 int error;
1245
1246 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1247 if (error || *vpp != NULL)
1248 return (error);
1249
1250 /*
1251 * We must promote to an exclusive lock for vnode creation. This
1252 * can happen if lookup is passed LOCKSHARED.
1253 */
1254 if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1255 flags &= ~LK_TYPE_MASK;
1256 flags |= LK_EXCLUSIVE;
1257 }
1258
1259 /*
1260 * We do not lock vnode creation as it is believed to be too
1261 * expensive for such rare case as simultaneous creation of vnode
1262 * for same ino by different processes. We just allow them to race
1263 * and check later to decide who wins. Let the race begin!
1264 */
1265
1266 ump = VFSTOUFS(mp);
1267 dev = ump->um_dev;
1268 fs = ump->um_fs;
1269
1270 /*
1271 * If this MALLOC() is performed after the getnewvnode()
1272 * it might block, leaving a vnode with a NULL v_data to be
1273 * found by ffs_sync() if a sync happens to fire right then,
1274 * which will cause a panic because ffs_sync() blindly
1275 * dereferences vp->v_data (as well it should).
1276 */
1277 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1278
1279 /* Allocate a new vnode/inode. */
1280 if (fs->fs_magic == FS_UFS1_MAGIC)
1281 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp);
1282 else
1283 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp);
1284 if (error) {
1285 *vpp = NULL;
1286 uma_zfree(uma_inode, ip);
1287 return (error);
1288 }
1289 /*
1290 * FFS supports recursive and shared locking.
1291 */
1292 vp->v_vnlock->lk_flags |= LK_CANRECURSE;
1293 vp->v_vnlock->lk_flags &= ~LK_NOSHARE;
1294 vp->v_data = ip;
1295 vp->v_bufobj.bo_bsize = fs->fs_bsize;
1296 ip->i_vnode = vp;
1297 ip->i_ump = ump;
1298 ip->i_fs = fs;
1299 ip->i_dev = dev;
1300 ip->i_number = ino;
1301 #ifdef QUOTA
1302 {
1303 int i;
1304 for (i = 0; i < MAXQUOTAS; i++)
1305 ip->i_dquot[i] = NODQUOT;
1306 }
1307 #endif
1308
1309 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1310 if (error || *vpp != NULL)
1311 return (error);
1312
1313 /* Read in the disk contents for the inode, copy into the inode. */
1314 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1315 (int)fs->fs_bsize, NOCRED, &bp);
1316 if (error) {
1317 /*
1318 * The inode does not contain anything useful, so it would
1319 * be misleading to leave it on its hash chain. With mode
1320 * still zero, it will be unlinked and returned to the free
1321 * list by vput().
1322 */
1323 brelse(bp);
1324 vput(vp);
1325 *vpp = NULL;
1326 return (error);
1327 }
1328 if (ip->i_ump->um_fstype == UFS1)
1329 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1330 else
1331 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1332 ffs_load_inode(bp, ip, fs, ino);
1333 if (DOINGSOFTDEP(vp))
1334 softdep_load_inodeblock(ip);
1335 else
1336 ip->i_effnlink = ip->i_nlink;
1337 bqrelse(bp);
1338
1339 /*
1340 * Initialize the vnode from the inode, check for aliases.
1341 * Note that the underlying vnode may have changed.
1342 */
1343 if (ip->i_ump->um_fstype == UFS1)
1344 error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1345 else
1346 error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1347 if (error) {
1348 vput(vp);
1349 *vpp = NULL;
1350 return (error);
1351 }
1352
1353 /*
1354 * Finish inode initialization.
1355 */
1356
1357 /*
1358 * Set up a generation number for this inode if it does not
1359 * already have one. This should only happen on old filesystems.
1360 */
1361 if (ip->i_gen == 0) {
1362 ip->i_gen = arc4random() / 2 + 1;
1363 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1364 ip->i_flag |= IN_MODIFIED;
1365 DIP_SET(ip, i_gen, ip->i_gen);
1366 }
1367 }
1368 /*
1369 * Ensure that uid and gid are correct. This is a temporary
1370 * fix until fsck has been changed to do the update.
1371 */
1372 if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */
1373 fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */
1374 ip->i_uid = ip->i_din1->di_ouid; /* XXX */
1375 ip->i_gid = ip->i_din1->di_ogid; /* XXX */
1376 } /* XXX */
1377
1378 #ifdef MAC
1379 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1380 /*
1381 * If this vnode is already allocated, and we're running
1382 * multi-label, attempt to perform a label association
1383 * from the extended attributes on the inode.
1384 */
1385 error = mac_associate_vnode_extattr(mp, vp);
1386 if (error) {
1387 /* ufs_inactive will release ip->i_devvp ref. */
1388 vput(vp);
1389 *vpp = NULL;
1390 return (error);
1391 }
1392 }
1393 #endif
1394
1395 *vpp = vp;
1396 return (0);
1397 }
1398
1399 /*
1400 * File handle to vnode
1401 *
1402 * Have to be really careful about stale file handles:
1403 * - check that the inode number is valid
1404 * - call ffs_vget() to get the locked inode
1405 * - check for an unallocated inode (i_mode == 0)
1406 * - check that the given client host has export rights and return
1407 * those rights via. exflagsp and credanonp
1408 */
1409 static int
1410 ffs_fhtovp(mp, fhp, vpp)
1411 struct mount *mp;
1412 struct fid *fhp;
1413 struct vnode **vpp;
1414 {
1415 struct ufid *ufhp;
1416 struct fs *fs;
1417
1418 ufhp = (struct ufid *)fhp;
1419 fs = VFSTOUFS(mp)->um_fs;
1420 if (ufhp->ufid_ino < ROOTINO ||
1421 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1422 return (ESTALE);
1423 return (ufs_fhtovp(mp, ufhp, vpp));
1424 }
1425
1426 /*
1427 * Vnode pointer to File handle
1428 */
1429 /* ARGSUSED */
1430 static int
1431 ffs_vptofh(vp, fhp)
1432 struct vnode *vp;
1433 struct fid *fhp;
1434 {
1435 struct inode *ip;
1436 struct ufid *ufhp;
1437
1438 ip = VTOI(vp);
1439 ufhp = (struct ufid *)fhp;
1440 ufhp->ufid_len = sizeof(struct ufid);
1441 ufhp->ufid_ino = ip->i_number;
1442 ufhp->ufid_gen = ip->i_gen;
1443 return (0);
1444 }
1445
1446 /*
1447 * Initialize the filesystem.
1448 */
1449 static int
1450 ffs_init(vfsp)
1451 struct vfsconf *vfsp;
1452 {
1453
1454 softdep_initialize();
1455 return (ufs_init(vfsp));
1456 }
1457
1458 /*
1459 * Undo the work of ffs_init().
1460 */
1461 static int
1462 ffs_uninit(vfsp)
1463 struct vfsconf *vfsp;
1464 {
1465 int ret;
1466
1467 ret = ufs_uninit(vfsp);
1468 softdep_uninitialize();
1469 return (ret);
1470 }
1471
1472 /*
1473 * Write a superblock and associated information back to disk.
1474 */
1475 static int
1476 ffs_sbupdate(mp, waitfor, suspended)
1477 struct ufsmount *mp;
1478 int waitfor;
1479 int suspended;
1480 {
1481 struct fs *fs = mp->um_fs;
1482 struct buf *sbbp;
1483 struct buf *bp;
1484 int blks;
1485 void *space;
1486 int i, size, error, allerror = 0;
1487
1488 if (fs->fs_ronly == 1 &&
1489 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1490 (MNT_RDONLY | MNT_UPDATE))
1491 panic("ffs_sbupdate: write read-only filesystem");
1492 /*
1493 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1494 */
1495 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize,
1496 0, 0, 0);
1497 /*
1498 * First write back the summary information.
1499 */
1500 blks = howmany(fs->fs_cssize, fs->fs_fsize);
1501 space = fs->fs_csp;
1502 for (i = 0; i < blks; i += fs->fs_frag) {
1503 size = fs->fs_bsize;
1504 if (i + fs->fs_frag > blks)
1505 size = (blks - i) * fs->fs_fsize;
1506 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1507 size, 0, 0, 0);
1508 bcopy(space, bp->b_data, (u_int)size);
1509 space = (char *)space + size;
1510 if (suspended)
1511 bp->b_flags |= B_VALIDSUSPWRT;
1512 if (waitfor != MNT_WAIT)
1513 bawrite(bp);
1514 else if ((error = bwrite(bp)) != 0)
1515 allerror = error;
1516 }
1517 /*
1518 * Now write back the superblock itself. If any errors occurred
1519 * up to this point, then fail so that the superblock avoids
1520 * being written out as clean.
1521 */
1522 if (allerror) {
1523 brelse(sbbp);
1524 return (allerror);
1525 }
1526 bp = sbbp;
1527 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1528 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1529 printf("%s: correcting fs_sblockloc from %jd to %d\n",
1530 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1531 fs->fs_sblockloc = SBLOCK_UFS1;
1532 }
1533 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1534 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1535 printf("%s: correcting fs_sblockloc from %jd to %d\n",
1536 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1537 fs->fs_sblockloc = SBLOCK_UFS2;
1538 }
1539 fs->fs_fmod = 0;
1540 fs->fs_time = time_second;
1541 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1542 ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
1543 if (suspended)
1544 bp->b_flags |= B_VALIDSUSPWRT;
1545 if (waitfor != MNT_WAIT)
1546 bawrite(bp);
1547 else if ((error = bwrite(bp)) != 0)
1548 allerror = error;
1549 return (allerror);
1550 }
1551
1552 static int
1553 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1554 int attrnamespace, const char *attrname, struct thread *td)
1555 {
1556
1557 #ifdef UFS_EXTATTR
1558 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1559 attrname, td));
1560 #else
1561 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1562 attrname, td));
1563 #endif
1564 }
1565
1566 static void
1567 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1568 {
1569
1570 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1571 uma_zfree(uma_ufs1, ip->i_din1);
1572 else if (ip->i_din2 != NULL)
1573 uma_zfree(uma_ufs2, ip->i_din2);
1574 uma_zfree(uma_inode, ip);
1575 }
1576
1577 static int dobkgrdwrite = 1;
1578 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1579 "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1580
1581 /*
1582 * Complete a background write started from bwrite.
1583 */
1584 static void
1585 ffs_backgroundwritedone(struct buf *bp)
1586 {
1587 struct bufobj *bufobj;
1588 struct buf *origbp;
1589
1590 /*
1591 * Find the original buffer that we are writing.
1592 */
1593 bufobj = bp->b_bufobj;
1594 BO_LOCK(bufobj);
1595 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
1596 panic("backgroundwritedone: lost buffer");
1597 /* Grab an extra reference to be dropped by the bufdone() below. */
1598 bufobj_wrefl(bufobj);
1599 BO_UNLOCK(bufobj);
1600 /*
1601 * Process dependencies then return any unfinished ones.
1602 */
1603 if (!LIST_EMPTY(&bp->b_dep))
1604 buf_complete(bp);
1605 #ifdef SOFTUPDATES
1606 if (!LIST_EMPTY(&bp->b_dep))
1607 softdep_move_dependencies(bp, origbp);
1608 #endif
1609 /*
1610 * This buffer is marked B_NOCACHE so when it is released
1611 * by biodone it will be tossed.
1612 */
1613 bp->b_flags |= B_NOCACHE;
1614 bp->b_flags &= ~B_CACHE;
1615 bufdone(bp);
1616 BO_LOCK(bufobj);
1617 /*
1618 * Clear the BV_BKGRDINPROG flag in the original buffer
1619 * and awaken it if it is waiting for the write to complete.
1620 * If BV_BKGRDINPROG is not set in the original buffer it must
1621 * have been released and re-instantiated - which is not legal.
1622 */
1623 KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
1624 ("backgroundwritedone: lost buffer2"));
1625 origbp->b_vflags &= ~BV_BKGRDINPROG;
1626 if (origbp->b_vflags & BV_BKGRDWAIT) {
1627 origbp->b_vflags &= ~BV_BKGRDWAIT;
1628 wakeup(&origbp->b_xflags);
1629 }
1630 BO_UNLOCK(bufobj);
1631 }
1632
1633
1634 /*
1635 * Write, release buffer on completion. (Done by iodone
1636 * if async). Do not bother writing anything if the buffer
1637 * is invalid.
1638 *
1639 * Note that we set B_CACHE here, indicating that buffer is
1640 * fully valid and thus cacheable. This is true even of NFS
1641 * now so we set it generally. This could be set either here
1642 * or in biodone() since the I/O is synchronous. We put it
1643 * here.
1644 */
1645 static int
1646 ffs_bufwrite(struct buf *bp)
1647 {
1648 int oldflags, s;
1649 struct buf *newbp;
1650
1651 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1652 if (bp->b_flags & B_INVAL) {
1653 brelse(bp);
1654 return (0);
1655 }
1656
1657 oldflags = bp->b_flags;
1658
1659 if (BUF_REFCNT(bp) == 0)
1660 panic("bufwrite: buffer is not busy???");
1661 s = splbio();
1662 /*
1663 * If a background write is already in progress, delay
1664 * writing this block if it is asynchronous. Otherwise
1665 * wait for the background write to complete.
1666 */
1667 BO_LOCK(bp->b_bufobj);
1668 if (bp->b_vflags & BV_BKGRDINPROG) {
1669 if (bp->b_flags & B_ASYNC) {
1670 BO_UNLOCK(bp->b_bufobj);
1671 splx(s);
1672 bdwrite(bp);
1673 return (0);
1674 }
1675 bp->b_vflags |= BV_BKGRDWAIT;
1676 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0);
1677 if (bp->b_vflags & BV_BKGRDINPROG)
1678 panic("bufwrite: still writing");
1679 }
1680 BO_UNLOCK(bp->b_bufobj);
1681
1682 /* Mark the buffer clean */
1683 bundirty(bp);
1684
1685 /*
1686 * If this buffer is marked for background writing and we
1687 * do not have to wait for it, make a copy and write the
1688 * copy so as to leave this buffer ready for further use.
1689 *
1690 * This optimization eats a lot of memory. If we have a page
1691 * or buffer shortfall we can't do it.
1692 */
1693 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
1694 (bp->b_flags & B_ASYNC) &&
1695 !vm_page_count_severe() &&
1696 !buf_dirty_count_severe()) {
1697 KASSERT(bp->b_iodone == NULL,
1698 ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
1699
1700 /* get a new block */
1701 newbp = geteblk(bp->b_bufsize);
1702
1703 /*
1704 * set it to be identical to the old block. We have to
1705 * set b_lblkno and BKGRDMARKER before calling bgetvp()
1706 * to avoid confusing the splay tree and gbincore().
1707 */
1708 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
1709 newbp->b_lblkno = bp->b_lblkno;
1710 newbp->b_xflags |= BX_BKGRDMARKER;
1711 BO_LOCK(bp->b_bufobj);
1712 bp->b_vflags |= BV_BKGRDINPROG;
1713 bgetvp(bp->b_vp, newbp);
1714 BO_UNLOCK(bp->b_bufobj);
1715 newbp->b_bufobj = &bp->b_vp->v_bufobj;
1716 newbp->b_blkno = bp->b_blkno;
1717 newbp->b_offset = bp->b_offset;
1718 newbp->b_iodone = ffs_backgroundwritedone;
1719 newbp->b_flags |= B_ASYNC;
1720 newbp->b_flags &= ~B_INVAL;
1721
1722 #ifdef SOFTUPDATES
1723 /* move over the dependencies */
1724 if (!LIST_EMPTY(&bp->b_dep))
1725 softdep_move_dependencies(bp, newbp);
1726 #endif
1727
1728 /*
1729 * Initiate write on the copy, release the original to
1730 * the B_LOCKED queue so that it cannot go away until
1731 * the background write completes. If not locked it could go
1732 * away and then be reconstituted while it was being written.
1733 * If the reconstituted buffer were written, we could end up
1734 * with two background copies being written at the same time.
1735 */
1736 bqrelse(bp);
1737 bp = newbp;
1738 }
1739
1740 /* Let the normal bufwrite do the rest for us */
1741 return (bufwrite(bp));
1742 }
1743
1744
1745 static void
1746 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
1747 {
1748 struct vnode *vp;
1749 int error;
1750 struct buf *tbp;
1751
1752 vp = bo->__bo_vnode;
1753 if (bp->b_iocmd == BIO_WRITE) {
1754 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
1755 bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
1756 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1757 panic("ffs_geom_strategy: bad I/O");
1758 bp->b_flags &= ~B_VALIDSUSPWRT;
1759 if ((vp->v_vflag & VV_COPYONWRITE) &&
1760 vp->v_rdev->si_snapdata != NULL) {
1761 if ((bp->b_flags & B_CLUSTER) != 0) {
1762 runningbufwakeup(bp);
1763 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
1764 b_cluster.cluster_entry) {
1765 error = ffs_copyonwrite(vp, tbp);
1766 if (error != 0 &&
1767 error != EOPNOTSUPP) {
1768 bp->b_error = error;
1769 bp->b_ioflags |= BIO_ERROR;
1770 bufdone(bp);
1771 return;
1772 }
1773 }
1774 bp->b_runningbufspace = bp->b_bufsize;
1775 atomic_add_int(&runningbufspace,
1776 bp->b_runningbufspace);
1777 } else {
1778 error = ffs_copyonwrite(vp, bp);
1779 if (error != 0 && error != EOPNOTSUPP) {
1780 bp->b_error = error;
1781 bp->b_ioflags |= BIO_ERROR;
1782 bufdone(bp);
1783 return;
1784 }
1785 }
1786 }
1787 #ifdef SOFTUPDATES
1788 if ((bp->b_flags & B_CLUSTER) != 0) {
1789 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
1790 b_cluster.cluster_entry) {
1791 if (!LIST_EMPTY(&tbp->b_dep))
1792 buf_start(tbp);
1793 }
1794 } else {
1795 if (!LIST_EMPTY(&bp->b_dep))
1796 buf_start(bp);
1797 }
1798
1799 #endif
1800 }
1801 g_vfs_strategy(bo, bp);
1802 }
Cache object: d592d99e255ad65188c5154b5ace7a5a
|