38
39#include <sys/param.h>
40#include <sys/kernel.h>
41#include <sys/systm.h>
42#include <sys/conf.h>
43#include <sys/bio.h>
44#include <sys/buf.h>
45#include <sys/proc.h>
46#include <sys/namei.h>
47#include <sys/sched.h>
48#include <sys/stat.h>
49#include <sys/malloc.h>
50#include <sys/mount.h>
51#include <sys/resource.h>
52#include <sys/resourcevar.h>
53#include <sys/vnode.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#define KERNCRED thread0.td_ucred
65#define DEBUG 1
66
67TAILQ_HEAD(snaphead, inode);
68
69struct snapdata {
70 struct snaphead sn_head;
71 daddr_t sn_listsize;
72 daddr_t *sn_blklist;
73 struct lock sn_lock;
74};
75
76static int cgaccount(int, struct vnode *, struct buf *, int);
77static int expunge_ufs1(struct vnode *, struct inode *, struct fs *,
78 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
79 ufs_lbn_t, int), int);
80static int indiracct_ufs1(struct vnode *, struct vnode *, int,
81 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
82 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
83 ufs_lbn_t, int), int);
84static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
85 struct fs *, ufs_lbn_t, int);
86static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
87 struct fs *, ufs_lbn_t, int);
88static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
89 struct fs *, ufs_lbn_t, int);
90static int expunge_ufs2(struct vnode *, struct inode *, struct fs *,
91 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
92 ufs_lbn_t, int), int);
93static int indiracct_ufs2(struct vnode *, struct vnode *, int,
94 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
95 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
96 ufs_lbn_t, int), int);
97static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
98 struct fs *, ufs_lbn_t, int);
99static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
100 struct fs *, ufs_lbn_t, int);
101static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
102 struct fs *, ufs_lbn_t, int);
103static int ffs_copyonwrite(struct vnode *, struct buf *);
104static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t);
105
106/*
107 * To ensure the consistency of snapshots across crashes, we must
108 * synchronously write out copied blocks before allowing the
109 * originals to be modified. Because of the rather severe speed
110 * penalty that this imposes, the following flag allows this
111 * crash persistence to be disabled.
112 */
113int dopersistence = 0;
114
115#ifdef DEBUG
116#include <sys/sysctl.h>
117SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, "");
118static int snapdebug = 0;
119SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, "");
120int collectsnapstats = 0;
121SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats,
122 0, "");
123#endif /* DEBUG */
124
125/*
126 * Create a snapshot file and initialize it for the filesystem.
127 */
128int
129ffs_snapshot(mp, snapfile)
130 struct mount *mp;
131 char *snapfile;
132{
133 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist;
134 int error, cg, snaploc;
135 int i, size, len, loc;
136 int flag = mp->mnt_flag;
137 struct timespec starttime = {0, 0}, endtime;
138 char saved_nice = 0;
139 long redo = 0, snaplistsize = 0;
140 int32_t *lp;
141 void *space;
142 struct fs *copy_fs = NULL, *fs = VFSTOUFS(mp)->um_fs;
143 struct thread *td = curthread;
144 struct inode *ip, *xp;
145 struct buf *bp, *nbp, *ibp, *sbp = NULL;
146 struct nameidata nd;
147 struct mount *wrtmp;
148 struct vattr vat;
149 struct vnode *vp, *xvp, *nvp, *devvp;
150 struct uio auio;
151 struct iovec aiov;
152 struct snapdata *sn;
153
154 /*
155 * XXX: make sure we don't go to out1 before we setup sn
156 */
157 sn = (void *)0xdeadbeef;
158
159 /*
160 * Need to serialize access to snapshot code per filesystem.
161 */
162 /*
163 * Assign a snapshot slot in the superblock.
164 */
165 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
166 if (fs->fs_snapinum[snaploc] == 0)
167 break;
168 if (snaploc == FSMAXSNAP)
169 return (ENOSPC);
170 /*
171 * Create the snapshot file.
172 */
173restart:
174 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_USERSPACE, snapfile, td);
175 if ((error = namei(&nd)) != 0)
176 return (error);
177 if (nd.ni_vp != NULL) {
178 vput(nd.ni_vp);
179 error = EEXIST;
180 }
181 if (nd.ni_dvp->v_mount != mp)
182 error = EXDEV;
183 if (error) {
184 NDFREE(&nd, NDF_ONLY_PNBUF);
185 if (nd.ni_dvp == nd.ni_vp)
186 vrele(nd.ni_dvp);
187 else
188 vput(nd.ni_dvp);
189 return (error);
190 }
191 VATTR_NULL(&vat);
192 vat.va_type = VREG;
193 vat.va_mode = S_IRUSR;
194 vat.va_vaflags |= VA_EXCLUSIVE;
195 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp))
196 wrtmp = NULL;
197 if (wrtmp != mp)
198 panic("ffs_snapshot: mount mismatch");
199 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) {
200 NDFREE(&nd, NDF_ONLY_PNBUF);
201 vput(nd.ni_dvp);
202 if ((error = vn_start_write(NULL, &wrtmp,
203 V_XSLEEP | PCATCH)) != 0)
204 return (error);
205 goto restart;
206 }
207 VOP_LEASE(nd.ni_dvp, td, KERNCRED, LEASE_WRITE);
208 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat);
209 vput(nd.ni_dvp);
210 if (error) {
211 NDFREE(&nd, NDF_ONLY_PNBUF);
212 vn_finished_write(wrtmp);
213 return (error);
214 }
215 vp = nd.ni_vp;
216 ip = VTOI(vp);
217 devvp = ip->i_devvp;
218 /*
219 * Allocate and copy the last block contents so as to be able
220 * to set size to that of the filesystem.
221 */
222 numblks = howmany(fs->fs_size, fs->fs_frag);
223 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
224 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
225 if (error)
226 goto out;
227 ip->i_size = lblktosize(fs, (off_t)numblks);
228 DIP_SET(ip, i_size, ip->i_size);
229 ip->i_flag |= IN_CHANGE | IN_UPDATE;
230 if ((error = readblock(vp, bp, numblks - 1)) != 0)
231 goto out;
232 bawrite(bp);
233 /*
234 * Preallocate critical data structures so that we can copy
235 * them in without further allocation after we suspend all
236 * operations on the filesystem. We would like to just release
237 * the allocated buffers without writing them since they will
238 * be filled in below once we are ready to go, but this upsets
239 * the soft update code, so we go ahead and write the new buffers.
240 *
241 * Allocate all indirect blocks and mark all of them as not
242 * needing to be copied.
243 */
244 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
245 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
246 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp);
247 if (error)
248 goto out;
249 bawrite(ibp);
250 }
251 /*
252 * Allocate copies for the superblock and its summary information.
253 */
254 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED,
255 0, &nbp);
256 if (error)
257 goto out;
258 bawrite(nbp);
259 blkno = fragstoblks(fs, fs->fs_csaddr);
260 len = howmany(fs->fs_cssize, fs->fs_bsize);
261 for (loc = 0; loc < len; loc++) {
262 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)),
263 fs->fs_bsize, KERNCRED, 0, &nbp);
264 if (error)
265 goto out;
266 bawrite(nbp);
267 }
268 /*
269 * Allocate all cylinder group blocks.
270 */
271 for (cg = 0; cg < fs->fs_ncg; cg++) {
272 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
273 fs->fs_bsize, KERNCRED, 0, &nbp);
274 if (error)
275 goto out;
276 bawrite(nbp);
277 }
278 /*
279 * Copy all the cylinder group maps. Although the
280 * filesystem is still active, we hope that only a few
281 * cylinder groups will change between now and when we
282 * suspend operations. Thus, we will be able to quickly
283 * touch up the few cylinder groups that changed during
284 * the suspension period.
285 */
286 len = howmany(fs->fs_ncg, NBBY);
287 MALLOC(fs->fs_active, int *, len, M_DEVBUF, M_WAITOK);
288 bzero(fs->fs_active, len);
289 for (cg = 0; cg < fs->fs_ncg; cg++) {
290 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
291 fs->fs_bsize, KERNCRED, 0, &nbp);
292 if (error)
293 goto out;
294 error = cgaccount(cg, vp, nbp, 1);
295 bawrite(nbp);
296 if (error)
297 goto out;
298 }
299 /*
300 * Change inode to snapshot type file.
301 */
302 ip->i_flags |= SF_SNAPSHOT;
303 DIP_SET(ip, i_flags, ip->i_flags);
304 ip->i_flag |= IN_CHANGE | IN_UPDATE;
305 /*
306 * Ensure that the snapshot is completely on disk.
307 * Since we have marked it as a snapshot it is safe to
308 * unlock it as no process will be allowed to write to it.
309 */
310 if ((error = VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td)) != 0)
311 goto out;
312 VOP_UNLOCK(vp, 0, td);
313 /*
314 * All allocations are done, so we can now snapshot the system.
315 *
316 * Recind nice scheduling while running with the filesystem suspended.
317 */
318 if (td->td_proc->p_nice > 0) {
319 PROC_LOCK(td->td_proc);
320 mtx_lock_spin(&sched_lock);
321 saved_nice = td->td_proc->p_nice;
322 sched_nice(td->td_proc, 0);
323 mtx_unlock_spin(&sched_lock);
324 PROC_UNLOCK(td->td_proc);
325 }
326 /*
327 * Suspend operation on filesystem.
328 */
329 for (;;) {
330 vn_finished_write(wrtmp);
331 if ((error = vfs_write_suspend(vp->v_mount)) != 0) {
332 vn_start_write(NULL, &wrtmp, V_WAIT);
333 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
334 goto out;
335 }
336 if (mp->mnt_kern_flag & MNTK_SUSPENDED)
337 break;
338 vn_start_write(NULL, &wrtmp, V_WAIT);
339 }
340 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
341 if (collectsnapstats)
342 nanotime(&starttime);
343 /*
344 * First, copy all the cylinder group maps that have changed.
345 */
346 for (cg = 0; cg < fs->fs_ncg; cg++) {
347 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0)
348 continue;
349 redo++;
350 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
351 fs->fs_bsize, KERNCRED, 0, &nbp);
352 if (error)
353 goto out1;
354 error = cgaccount(cg, vp, nbp, 2);
355 bawrite(nbp);
356 if (error)
357 goto out1;
358 }
359 /*
360 * Grab a copy of the superblock and its summary information.
361 * We delay writing it until the suspension is released below.
362 */
363 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize,
364 KERNCRED, &sbp);
365 if (error) {
366 brelse(sbp);
367 sbp = NULL;
368 goto out1;
369 }
370 loc = blkoff(fs, fs->fs_sblockloc);
371 copy_fs = (struct fs *)(sbp->b_data + loc);
372 bcopy(fs, copy_fs, fs->fs_sbsize);
373 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
374 copy_fs->fs_clean = 1;
375 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
376 if (fs->fs_sbsize < size)
377 bzero(&sbp->b_data[loc + fs->fs_sbsize], size - fs->fs_sbsize);
378 size = blkroundup(fs, fs->fs_cssize);
379 if (fs->fs_contigsumsize > 0)
380 size += fs->fs_ncg * sizeof(int32_t);
381 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
382 copy_fs->fs_csp = space;
383 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize);
384 space = (char *)space + fs->fs_cssize;
385 loc = howmany(fs->fs_cssize, fs->fs_fsize);
386 i = fs->fs_frag - loc % fs->fs_frag;
387 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
388 if (len > 0) {
389 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
390 len, KERNCRED, &bp)) != 0) {
391 brelse(bp);
392 free(copy_fs->fs_csp, M_UFSMNT);
393 bawrite(sbp);
394 sbp = NULL;
395 goto out1;
396 }
397 bcopy(bp->b_data, space, (u_int)len);
398 space = (char *)space + len;
399 bp->b_flags |= B_INVAL | B_NOCACHE;
400 brelse(bp);
401 }
402 if (fs->fs_contigsumsize > 0) {
403 copy_fs->fs_maxcluster = lp = space;
404 for (i = 0; i < fs->fs_ncg; i++)
405 *lp++ = fs->fs_contigsumsize;
406 }
407 /*
408 * We must check for active files that have been unlinked
409 * (e.g., with a zero link count). We have to expunge all
410 * trace of these files from the snapshot so that they are
411 * not reclaimed prematurely by fsck or unnecessarily dumped.
412 * We turn off the MNTK_SUSPENDED flag to avoid a panic from
413 * spec_strategy about writing on a suspended filesystem.
414 * Note that we skip unlinked snapshot files as they will
415 * be handled separately below.
416 *
417 * We also calculate the needed size for the snapshot list.
418 */
419 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
420 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
421 mp->mnt_kern_flag &= ~MNTK_SUSPENDED;
422 MNT_ILOCK(mp);
423loop:
424 MNT_VNODE_FOREACH(xvp, mp, nvp) {
425 VI_LOCK(xvp);
426 MNT_IUNLOCK(mp);
427 if ((xvp->v_iflag & VI_XLOCK) ||
428 xvp->v_usecount == 0 || xvp->v_type == VNON ||
429 (VTOI(xvp)->i_flags & SF_SNAPSHOT)) {
430 VI_UNLOCK(xvp);
431 MNT_ILOCK(mp);
432 continue;
433 }
434 /*
435 * We can skip parent directory vnode because it must have
436 * this snapshot file in it.
437 */
438 if (xvp == nd.ni_dvp) {
439 VI_UNLOCK(xvp);
440 MNT_ILOCK(mp);
441 continue;
442 }
443 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK, td) != 0) {
444 MNT_ILOCK(mp);
445 goto loop;
446 }
447 if (snapdebug)
448 vprint("ffs_snapshot: busy vnode", xvp);
449 if (VOP_GETATTR(xvp, &vat, td->td_ucred, td) == 0 &&
450 vat.va_nlink > 0) {
451 VOP_UNLOCK(xvp, 0, td);
452 MNT_ILOCK(mp);
453 continue;
454 }
455 xp = VTOI(xvp);
456 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
457 VOP_UNLOCK(xvp, 0, td);
458 MNT_ILOCK(mp);
459 continue;
460 }
461 /*
462 * If there is a fragment, clear it here.
463 */
464 blkno = 0;
465 loc = howmany(xp->i_size, fs->fs_bsize) - 1;
466 if (loc < NDADDR) {
467 len = fragroundup(fs, blkoff(fs, xp->i_size));
468 if (len < fs->fs_bsize) {
469 ffs_blkfree(copy_fs, vp, DIP(xp, i_db[loc]),
470 len, xp->i_number);
471 blkno = DIP(xp, i_db[loc]);
472 DIP_SET(xp, i_db[loc], 0);
473 }
474 }
475 snaplistsize += 1;
476 if (xp->i_ump->um_fstype == UFS1)
477 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
478 BLK_NOCOPY);
479 else
480 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
481 BLK_NOCOPY);
482 if (blkno)
483 DIP_SET(xp, i_db[loc], blkno);
484 if (!error)
485 error = ffs_freefile(copy_fs, vp, xp->i_number,
486 xp->i_mode);
487 VOP_UNLOCK(xvp, 0, td);
488 if (error) {
489 free(copy_fs->fs_csp, M_UFSMNT);
490 bawrite(sbp);
491 sbp = NULL;
492 goto out1;
493 }
494 MNT_ILOCK(mp);
495 }
496 MNT_IUNLOCK(mp);
497 /*
498 * If there already exist snapshots on this filesystem, grab a
499 * reference to their shared lock. If this is the first snapshot
500 * on this filesystem, we need to allocate a lock for the snapshots
501 * to share. In either case, acquire the snapshot lock and give
502 * up our original private lock.
503 */
504 VI_LOCK(devvp);
505 sn = devvp->v_rdev->si_snapdata;
506 if (sn != NULL) {
507 xp = TAILQ_FIRST(&sn->sn_head);
508 VI_UNLOCK(devvp);
509 VI_LOCK(vp);
510 vp->v_vnlock = &sn->sn_lock;
511 } else {
512 VI_UNLOCK(devvp);
513 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO);
514 TAILQ_INIT(&sn->sn_head);
515 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT,
516 LK_CANRECURSE | LK_NOPAUSE);
517 VI_LOCK(vp);
518 vp->v_vnlock = &sn->sn_lock;
519 devvp->v_rdev->si_snapdata = sn;
520 xp = NULL;
521 }
522 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
523 transferlockers(&vp->v_lock, vp->v_vnlock);
524 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td);
525 /*
526 * If this is the first snapshot on this filesystem, then we need
527 * to allocate the space for the list of preallocated snapshot blocks.
528 * This list will be refined below, but this preliminary one will
529 * keep us out of deadlock until the full one is ready.
530 */
531 if (xp == NULL) {
532 MALLOC(snapblklist, daddr_t *, snaplistsize * sizeof(daddr_t),
533 M_UFSMNT, M_WAITOK);
534 blkp = &snapblklist[1];
535 *blkp++ = lblkno(fs, fs->fs_sblockloc);
536 blkno = fragstoblks(fs, fs->fs_csaddr);
537 for (cg = 0; cg < fs->fs_ncg; cg++) {
538 if (fragstoblks(fs, cgtod(fs, cg) > blkno))
539 break;
540 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
541 }
542 len = howmany(fs->fs_cssize, fs->fs_bsize);
543 for (loc = 0; loc < len; loc++)
544 *blkp++ = blkno + loc;
545 for (; cg < fs->fs_ncg; cg++)
546 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
547 snapblklist[0] = blkp - snapblklist;
548 VI_LOCK(devvp);
549 if (sn->sn_blklist != NULL)
550 panic("ffs_snapshot: non-empty list");
551 sn->sn_blklist = snapblklist;
552 sn->sn_listsize = blkp - snapblklist;
553 VI_UNLOCK(devvp);
554 }
555 /*
556 * Record snapshot inode. Since this is the newest snapshot,
557 * it must be placed at the end of the list.
558 */
559 VI_LOCK(devvp);
560 fs->fs_snapinum[snaploc] = ip->i_number;
561 if (ip->i_nextsnap.tqe_prev != 0)
562 panic("ffs_snapshot: %d already on list", ip->i_number);
563 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
564 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
565 devvp->v_vflag |= VV_COPYONWRITE;
566 VI_UNLOCK(devvp);
567 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp");
568 vp->v_vflag |= VV_SYSTEM;
569out1:
570 KASSERT(sn != (void *)0xdeadbeef, ("email phk@ and mckusick@"));
571 /*
572 * Resume operation on filesystem.
573 */
574 vfs_write_resume(vp->v_mount);
575 vn_start_write(NULL, &wrtmp, V_WAIT);
576 if (collectsnapstats && starttime.tv_sec > 0) {
577 nanotime(&endtime);
578 timespecsub(&endtime, &starttime);
579 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n",
580 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec,
581 endtime.tv_nsec / 1000000, redo, fs->fs_ncg);
582 }
583 if (sbp == NULL)
584 goto out;
585 /*
586 * Copy allocation information from all the snapshots in
587 * this snapshot and then expunge them from its view.
588 */
589 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) {
590 if (xp == ip)
591 break;
592 if (xp->i_ump->um_fstype == UFS1)
593 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1,
594 BLK_SNAP);
595 else
596 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2,
597 BLK_SNAP);
598 if (error) {
599 fs->fs_snapinum[snaploc] = 0;
600 goto done;
601 }
602 }
603 /*
604 * Allocate space for the full list of preallocated snapshot blocks.
605 */
606 MALLOC(snapblklist, daddr_t *, snaplistsize * sizeof(daddr_t),
607 M_UFSMNT, M_WAITOK);
608 ip->i_snapblklist = &snapblklist[1];
609 /*
610 * Expunge the blocks used by the snapshots from the set of
611 * blocks marked as used in the snapshot bitmaps. Also, collect
612 * the list of allocated blocks in i_snapblklist.
613 */
614 if (ip->i_ump->um_fstype == UFS1)
615 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, BLK_SNAP);
616 else
617 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, BLK_SNAP);
618 if (error) {
619 fs->fs_snapinum[snaploc] = 0;
620 FREE(snapblklist, M_UFSMNT);
621 goto done;
622 }
623 if (snaplistsize < ip->i_snapblklist - snapblklist)
624 panic("ffs_snapshot: list too small");
625 snaplistsize = ip->i_snapblklist - snapblklist;
626 snapblklist[0] = snaplistsize;
627 ip->i_snapblklist = 0;
628 /*
629 * Write out the list of allocated blocks to the end of the snapshot.
630 */
631 auio.uio_iov = &aiov;
632 auio.uio_iovcnt = 1;
633 aiov.iov_base = (void *)snapblklist;
634 aiov.iov_len = snaplistsize * sizeof(daddr_t);
635 auio.uio_resid = aiov.iov_len;;
636 auio.uio_offset = ip->i_size;
637 auio.uio_segflg = UIO_SYSSPACE;
638 auio.uio_rw = UIO_WRITE;
639 auio.uio_td = td;
640 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
641 fs->fs_snapinum[snaploc] = 0;
642 FREE(snapblklist, M_UFSMNT);
643 goto done;
644 }
645 /*
646 * Write the superblock and its summary information
647 * to the snapshot.
648 */
649 blkno = fragstoblks(fs, fs->fs_csaddr);
650 len = howmany(fs->fs_cssize, fs->fs_bsize);
651 space = copy_fs->fs_csp;
652 for (loc = 0; loc < len; loc++) {
653 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp);
654 if (error) {
655 brelse(nbp);
656 fs->fs_snapinum[snaploc] = 0;
657 FREE(snapblklist, M_UFSMNT);
658 goto done;
659 }
660 bcopy(space, nbp->b_data, fs->fs_bsize);
661 space = (char *)space + fs->fs_bsize;
662 bawrite(nbp);
663 }
664 /*
665 * As this is the newest list, it is the most inclusive, so
666 * should replace the previous list.
667 */
668 VI_LOCK(devvp);
669 space = sn->sn_blklist;
670 sn->sn_blklist = snapblklist;
671 sn->sn_listsize = snaplistsize;
672 VI_UNLOCK(devvp);
673 if (space != NULL)
674 FREE(space, M_UFSMNT);
675done:
676 FREE(copy_fs->fs_csp, M_UFSMNT);
677 bawrite(sbp);
678out:
679 if (saved_nice > 0) {
680 PROC_LOCK(td->td_proc);
681 mtx_lock_spin(&sched_lock);
682 sched_nice(td->td_proc, saved_nice);
683 mtx_unlock_spin(&sched_lock);
684 PROC_UNLOCK(td->td_proc);
685 }
686 if (fs->fs_active != 0) {
687 FREE(fs->fs_active, M_DEVBUF);
688 fs->fs_active = 0;
689 }
690 mp->mnt_flag = flag;
691 if (error)
692 (void) UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td);
693 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
694 if (error)
695 vput(vp);
696 else
697 VOP_UNLOCK(vp, 0, td);
698 vn_finished_write(wrtmp);
699 return (error);
700}
701
702/*
703 * Copy a cylinder group map. All the unallocated blocks are marked
704 * BLK_NOCOPY so that the snapshot knows that it need not copy them
705 * if they are later written. If passno is one, then this is a first
706 * pass, so only setting needs to be done. If passno is 2, then this
707 * is a revision to a previous pass which must be undone as the
708 * replacement pass is done.
709 */
710static int
711cgaccount(cg, vp, nbp, passno)
712 int cg;
713 struct vnode *vp;
714 struct buf *nbp;
715 int passno;
716{
717 struct buf *bp, *ibp;
718 struct inode *ip;
719 struct cg *cgp;
720 struct fs *fs;
721 ufs2_daddr_t base, numblks;
722 int error, len, loc, indiroff;
723
724 ip = VTOI(vp);
725 fs = ip->i_fs;
726 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
727 (int)fs->fs_cgsize, KERNCRED, &bp);
728 if (error) {
729 brelse(bp);
730 return (error);
731 }
732 cgp = (struct cg *)bp->b_data;
733 if (!cg_chkmagic(cgp)) {
734 brelse(bp);
735 return (EIO);
736 }
737 atomic_set_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
738 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize);
739 if (fs->fs_cgsize < fs->fs_bsize)
740 bzero(&nbp->b_data[fs->fs_cgsize],
741 fs->fs_bsize - fs->fs_cgsize);
742 if (passno == 2)
743 nbp->b_flags |= B_VALIDSUSPWRT;
744 numblks = howmany(fs->fs_size, fs->fs_frag);
745 len = howmany(fs->fs_fpg, fs->fs_frag);
746 base = cg * fs->fs_fpg / fs->fs_frag;
747 if (base + len >= numblks)
748 len = numblks - base - 1;
749 loc = 0;
750 if (base < NDADDR) {
751 for ( ; loc < NDADDR; loc++) {
752 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
753 DIP_SET(ip, i_db[loc], BLK_NOCOPY);
754 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY)
755 DIP_SET(ip, i_db[loc], 0);
756 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY)
757 panic("ffs_snapshot: lost direct block");
758 }
759 }
760 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)),
761 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
762 if (error) {
763 brelse(bp);
764 return (error);
765 }
766 indiroff = (base + loc - NDADDR) % NINDIR(fs);
767 for ( ; loc < len; loc++, indiroff++) {
768 if (indiroff >= NINDIR(fs)) {
769 if (passno == 2)
770 ibp->b_flags |= B_VALIDSUSPWRT;
771 bawrite(ibp);
772 error = UFS_BALLOC(vp,
773 lblktosize(fs, (off_t)(base + loc)),
774 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
775 if (error) {
776 brelse(bp);
777 return (error);
778 }
779 indiroff = 0;
780 }
781 if (ip->i_ump->um_fstype == UFS1) {
782 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
783 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
784 BLK_NOCOPY;
785 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data))
786 [indiroff] == BLK_NOCOPY)
787 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0;
788 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data))
789 [indiroff] == BLK_NOCOPY)
790 panic("ffs_snapshot: lost indirect block");
791 continue;
792 }
793 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
794 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
795 else if (passno == 2 &&
796 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
797 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0;
798 else if (passno == 1 &&
799 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
800 panic("ffs_snapshot: lost indirect block");
801 }
802 bqrelse(bp);
803 if (passno == 2)
804 ibp->b_flags |= B_VALIDSUSPWRT;
805 bdwrite(ibp);
806 return (0);
807}
808
809/*
810 * Before expunging a snapshot inode, note all the
811 * blocks that it claims with BLK_SNAP so that fsck will
812 * be able to account for those blocks properly and so
813 * that this snapshot knows that it need not copy them
814 * if the other snapshot holding them is freed. This code
815 * is reproduced once each for UFS1 and UFS2.
816 */
817static int
818expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype)
819 struct vnode *snapvp;
820 struct inode *cancelip;
821 struct fs *fs;
822 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
823 struct fs *, ufs_lbn_t, int);
824 int expungetype;
825{
826 int i, error, indiroff;
827 ufs_lbn_t lbn, rlbn;
828 ufs2_daddr_t len, blkno, numblks, blksperindir;
829 struct ufs1_dinode *dip;
830 struct thread *td = curthread;
831 struct buf *bp;
832
833 /*
834 * Prepare to expunge the inode. If its inode block has not
835 * yet been copied, then allocate and fill the copy.
836 */
837 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
838 blkno = 0;
839 if (lbn < NDADDR) {
840 blkno = VTOI(snapvp)->i_din1->di_db[lbn];
841 } else {
842 td->td_pflags |= TDP_COWINPROGRESS;
843 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
844 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
845 td->td_pflags &= ~TDP_COWINPROGRESS;
846 if (error)
847 return (error);
848 indiroff = (lbn - NDADDR) % NINDIR(fs);
849 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
850 bqrelse(bp);
851 }
852 if (blkno != 0) {
853 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
854 return (error);
855 } else {
856 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
857 fs->fs_bsize, KERNCRED, 0, &bp);
858 if (error)
859 return (error);
860 if ((error = readblock(snapvp, bp, lbn)) != 0)
861 return (error);
862 }
863 /*
864 * Set a snapshot inode to be a zero length file, regular files
865 * to be completely unallocated.
866 */
867 dip = (struct ufs1_dinode *)bp->b_data +
868 ino_to_fsbo(fs, cancelip->i_number);
869 if (expungetype == BLK_NOCOPY)
870 dip->di_mode = 0;
871 dip->di_size = 0;
872 dip->di_blocks = 0;
873 dip->di_flags &= ~SF_SNAPSHOT;
874 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t));
875 bdwrite(bp);
876 /*
877 * Now go through and expunge all the blocks in the file
878 * using the function requested.
879 */
880 numblks = howmany(cancelip->i_size, fs->fs_bsize);
881 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0],
882 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype)))
883 return (error);
884 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0],
885 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype)))
886 return (error);
887 blksperindir = 1;
888 lbn = -NDADDR;
889 len = numblks - NDADDR;
890 rlbn = NDADDR;
891 for (i = 0; len > 0 && i < NIADDR; i++) {
892 error = indiracct_ufs1(snapvp, ITOV(cancelip), i,
893 cancelip->i_din1->di_ib[i], lbn, rlbn, len,
894 blksperindir, fs, acctfunc, expungetype);
895 if (error)
896 return (error);
897 blksperindir *= NINDIR(fs);
898 lbn -= blksperindir + 1;
899 len -= blksperindir;
900 rlbn += blksperindir;
901 }
902 return (0);
903}
904
905/*
906 * Descend an indirect block chain for vnode cancelvp accounting for all
907 * its indirect blocks in snapvp.
908 */
909static int
910indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
911 blksperindir, fs, acctfunc, expungetype)
912 struct vnode *snapvp;
913 struct vnode *cancelvp;
914 int level;
915 ufs1_daddr_t blkno;
916 ufs_lbn_t lbn;
917 ufs_lbn_t rlbn;
918 ufs_lbn_t remblks;
919 ufs_lbn_t blksperindir;
920 struct fs *fs;
921 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
922 struct fs *, ufs_lbn_t, int);
923 int expungetype;
924{
925 int error, num, i;
926 ufs_lbn_t subblksperindir;
927 struct indir indirs[NIADDR + 2];
928 ufs1_daddr_t last, *bap;
929 struct buf *bp;
930
931 if (blkno == 0) {
932 if (expungetype == BLK_NOCOPY)
933 return (0);
934 panic("indiracct_ufs1: missing indir");
935 }
936 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
937 return (error);
938 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
939 panic("indiracct_ufs1: botched params");
940 /*
941 * We have to expand bread here since it will deadlock looking
942 * up the block number for any blocks that are not in the cache.
943 */
944 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
945 bp->b_blkno = fsbtodb(fs, blkno);
946 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
947 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
948 brelse(bp);
949 return (error);
950 }
951 /*
952 * Account for the block pointers in this indirect block.
953 */
954 last = howmany(remblks, blksperindir);
955 if (last > NINDIR(fs))
956 last = NINDIR(fs);
957 MALLOC(bap, ufs1_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK);
958 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
959 bqrelse(bp);
960 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
961 level == 0 ? rlbn : -1, expungetype);
962 if (error || level == 0)
963 goto out;
964 /*
965 * Account for the block pointers in each of the indirect blocks
966 * in the levels below us.
967 */
968 subblksperindir = blksperindir / NINDIR(fs);
969 for (lbn++, level--, i = 0; i < last; i++) {
970 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn,
971 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
972 if (error)
973 goto out;
974 rlbn += blksperindir;
975 lbn -= blksperindir;
976 remblks -= blksperindir;
977 }
978out:
979 FREE(bap, M_DEVBUF);
980 return (error);
981}
982
983/*
984 * Do both snap accounting and map accounting.
985 */
986static int
987fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)
988 struct vnode *vp;
989 ufs1_daddr_t *oldblkp, *lastblkp;
990 struct fs *fs;
991 ufs_lbn_t lblkno;
992 int exptype; /* BLK_SNAP or BLK_NOCOPY */
993{
994 int error;
995
996 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
997 return (error);
998 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype));
999}
1000
1001/*
1002 * Identify a set of blocks allocated in a snapshot inode.
1003 */
1004static int
1005snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1006 struct vnode *vp;
1007 ufs1_daddr_t *oldblkp, *lastblkp;
1008 struct fs *fs;
1009 ufs_lbn_t lblkno;
1010 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1011{
1012 struct inode *ip = VTOI(vp);
1013 ufs1_daddr_t blkno, *blkp;
1014 ufs_lbn_t lbn;
1015 struct buf *ibp;
1016 int error;
1017
1018 for ( ; oldblkp < lastblkp; oldblkp++) {
1019 blkno = *oldblkp;
1020 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1021 continue;
1022 lbn = fragstoblks(fs, blkno);
1023 if (lbn < NDADDR) {
1024 blkp = &ip->i_din1->di_db[lbn];
1025 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1026 } else {
1027 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1028 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1029 if (error)
1030 return (error);
1031 blkp = &((ufs1_daddr_t *)(ibp->b_data))
1032 [(lbn - NDADDR) % NINDIR(fs)];
1033 }
1034 /*
1035 * If we are expunging a snapshot vnode and we
1036 * find a block marked BLK_NOCOPY, then it is
1037 * one that has been allocated to this snapshot after
1038 * we took our current snapshot and can be ignored.
1039 */
1040 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1041 if (lbn >= NDADDR)
1042 brelse(ibp);
1043 } else {
1044 if (*blkp != 0)
1045 panic("snapacct_ufs1: bad block");
1046 *blkp = expungetype;
1047 if (lbn >= NDADDR)
1048 bdwrite(ibp);
1049 }
1050 }
1051 return (0);
1052}
1053
1054/*
1055 * Account for a set of blocks allocated in a snapshot inode.
1056 */
1057static int
1058mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1059 struct vnode *vp;
1060 ufs1_daddr_t *oldblkp, *lastblkp;
1061 struct fs *fs;
1062 ufs_lbn_t lblkno;
1063 int expungetype;
1064{
1065 ufs1_daddr_t blkno;
1066 struct inode *ip;
1067 ino_t inum;
1068 int acctit;
1069
1070 ip = VTOI(vp);
1071 inum = ip->i_number;
1072 if (lblkno == -1)
1073 acctit = 0;
1074 else
1075 acctit = 1;
1076 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1077 blkno = *oldblkp;
1078 if (blkno == 0 || blkno == BLK_NOCOPY)
1079 continue;
1080 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1081 *ip->i_snapblklist++ = lblkno;
1082 if (blkno == BLK_SNAP)
1083 blkno = blkstofrags(fs, lblkno);
1084 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
1085 }
1086 return (0);
1087}
1088
1089/*
1090 * Before expunging a snapshot inode, note all the
1091 * blocks that it claims with BLK_SNAP so that fsck will
1092 * be able to account for those blocks properly and so
1093 * that this snapshot knows that it need not copy them
1094 * if the other snapshot holding them is freed. This code
1095 * is reproduced once each for UFS1 and UFS2.
1096 */
1097static int
1098expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype)
1099 struct vnode *snapvp;
1100 struct inode *cancelip;
1101 struct fs *fs;
1102 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1103 struct fs *, ufs_lbn_t, int);
1104 int expungetype;
1105{
1106 int i, error, indiroff;
1107 ufs_lbn_t lbn, rlbn;
1108 ufs2_daddr_t len, blkno, numblks, blksperindir;
1109 struct ufs2_dinode *dip;
1110 struct thread *td = curthread;
1111 struct buf *bp;
1112
1113 /*
1114 * Prepare to expunge the inode. If its inode block has not
1115 * yet been copied, then allocate and fill the copy.
1116 */
1117 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1118 blkno = 0;
1119 if (lbn < NDADDR) {
1120 blkno = VTOI(snapvp)->i_din2->di_db[lbn];
1121 } else {
1122 td->td_pflags |= TDP_COWINPROGRESS;
1123 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
1124 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1125 td->td_pflags &= ~TDP_COWINPROGRESS;
1126 if (error)
1127 return (error);
1128 indiroff = (lbn - NDADDR) % NINDIR(fs);
1129 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
1130 bqrelse(bp);
1131 }
1132 if (blkno != 0) {
1133 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1134 return (error);
1135 } else {
1136 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
1137 fs->fs_bsize, KERNCRED, 0, &bp);
1138 if (error)
1139 return (error);
1140 if ((error = readblock(snapvp, bp, lbn)) != 0)
1141 return (error);
1142 }
1143 /*
1144 * Set a snapshot inode to be a zero length file, regular files
1145 * to be completely unallocated.
1146 */
1147 dip = (struct ufs2_dinode *)bp->b_data +
1148 ino_to_fsbo(fs, cancelip->i_number);
1149 if (expungetype == BLK_NOCOPY)
1150 dip->di_mode = 0;
1151 dip->di_size = 0;
1152 dip->di_blocks = 0;
1153 dip->di_flags &= ~SF_SNAPSHOT;
1154 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t));
1155 bdwrite(bp);
1156 /*
1157 * Now go through and expunge all the blocks in the file
1158 * using the function requested.
1159 */
1160 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1161 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0],
1162 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype)))
1163 return (error);
1164 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0],
1165 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype)))
1166 return (error);
1167 blksperindir = 1;
1168 lbn = -NDADDR;
1169 len = numblks - NDADDR;
1170 rlbn = NDADDR;
1171 for (i = 0; len > 0 && i < NIADDR; i++) {
1172 error = indiracct_ufs2(snapvp, ITOV(cancelip), i,
1173 cancelip->i_din2->di_ib[i], lbn, rlbn, len,
1174 blksperindir, fs, acctfunc, expungetype);
1175 if (error)
1176 return (error);
1177 blksperindir *= NINDIR(fs);
1178 lbn -= blksperindir + 1;
1179 len -= blksperindir;
1180 rlbn += blksperindir;
1181 }
1182 return (0);
1183}
1184
1185/*
1186 * Descend an indirect block chain for vnode cancelvp accounting for all
1187 * its indirect blocks in snapvp.
1188 */
1189static int
1190indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1191 blksperindir, fs, acctfunc, expungetype)
1192 struct vnode *snapvp;
1193 struct vnode *cancelvp;
1194 int level;
1195 ufs2_daddr_t blkno;
1196 ufs_lbn_t lbn;
1197 ufs_lbn_t rlbn;
1198 ufs_lbn_t remblks;
1199 ufs_lbn_t blksperindir;
1200 struct fs *fs;
1201 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1202 struct fs *, ufs_lbn_t, int);
1203 int expungetype;
1204{
1205 int error, num, i;
1206 ufs_lbn_t subblksperindir;
1207 struct indir indirs[NIADDR + 2];
1208 ufs2_daddr_t last, *bap;
1209 struct buf *bp;
1210
1211 if (blkno == 0) {
1212 if (expungetype == BLK_NOCOPY)
1213 return (0);
1214 panic("indiracct_ufs2: missing indir");
1215 }
1216 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1217 return (error);
1218 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1219 panic("indiracct_ufs2: botched params");
1220 /*
1221 * We have to expand bread here since it will deadlock looking
1222 * up the block number for any blocks that are not in the cache.
1223 */
1224 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1225 bp->b_blkno = fsbtodb(fs, blkno);
1226 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1227 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1228 brelse(bp);
1229 return (error);
1230 }
1231 /*
1232 * Account for the block pointers in this indirect block.
1233 */
1234 last = howmany(remblks, blksperindir);
1235 if (last > NINDIR(fs))
1236 last = NINDIR(fs);
1237 MALLOC(bap, ufs2_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK);
1238 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1239 bqrelse(bp);
1240 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1241 level == 0 ? rlbn : -1, expungetype);
1242 if (error || level == 0)
1243 goto out;
1244 /*
1245 * Account for the block pointers in each of the indirect blocks
1246 * in the levels below us.
1247 */
1248 subblksperindir = blksperindir / NINDIR(fs);
1249 for (lbn++, level--, i = 0; i < last; i++) {
1250 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn,
1251 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1252 if (error)
1253 goto out;
1254 rlbn += blksperindir;
1255 lbn -= blksperindir;
1256 remblks -= blksperindir;
1257 }
1258out:
1259 FREE(bap, M_DEVBUF);
1260 return (error);
1261}
1262
1263/*
1264 * Do both snap accounting and map accounting.
1265 */
1266static int
1267fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1268 struct vnode *vp;
1269 ufs2_daddr_t *oldblkp, *lastblkp;
1270 struct fs *fs;
1271 ufs_lbn_t lblkno;
1272 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1273{
1274 int error;
1275
1276 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1277 return (error);
1278 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1279}
1280
1281/*
1282 * Identify a set of blocks allocated in a snapshot inode.
1283 */
1284static int
1285snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1286 struct vnode *vp;
1287 ufs2_daddr_t *oldblkp, *lastblkp;
1288 struct fs *fs;
1289 ufs_lbn_t lblkno;
1290 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1291{
1292 struct inode *ip = VTOI(vp);
1293 ufs2_daddr_t blkno, *blkp;
1294 ufs_lbn_t lbn;
1295 struct buf *ibp;
1296 int error;
1297
1298 for ( ; oldblkp < lastblkp; oldblkp++) {
1299 blkno = *oldblkp;
1300 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1301 continue;
1302 lbn = fragstoblks(fs, blkno);
1303 if (lbn < NDADDR) {
1304 blkp = &ip->i_din2->di_db[lbn];
1305 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1306 } else {
1307 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1308 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1309 if (error)
1310 return (error);
1311 blkp = &((ufs2_daddr_t *)(ibp->b_data))
1312 [(lbn - NDADDR) % NINDIR(fs)];
1313 }
1314 /*
1315 * If we are expunging a snapshot vnode and we
1316 * find a block marked BLK_NOCOPY, then it is
1317 * one that has been allocated to this snapshot after
1318 * we took our current snapshot and can be ignored.
1319 */
1320 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1321 if (lbn >= NDADDR)
1322 brelse(ibp);
1323 } else {
1324 if (*blkp != 0)
1325 panic("snapacct_ufs2: bad block");
1326 *blkp = expungetype;
1327 if (lbn >= NDADDR)
1328 bdwrite(ibp);
1329 }
1330 }
1331 return (0);
1332}
1333
1334/*
1335 * Account for a set of blocks allocated in a snapshot inode.
1336 */
1337static int
1338mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1339 struct vnode *vp;
1340 ufs2_daddr_t *oldblkp, *lastblkp;
1341 struct fs *fs;
1342 ufs_lbn_t lblkno;
1343 int expungetype;
1344{
1345 ufs2_daddr_t blkno;
1346 struct inode *ip;
1347 ino_t inum;
1348 int acctit;
1349
1350 ip = VTOI(vp);
1351 inum = ip->i_number;
1352 if (lblkno == -1)
1353 acctit = 0;
1354 else
1355 acctit = 1;
1356 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1357 blkno = *oldblkp;
1358 if (blkno == 0 || blkno == BLK_NOCOPY)
1359 continue;
1360 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1361 *ip->i_snapblklist++ = lblkno;
1362 if (blkno == BLK_SNAP)
1363 blkno = blkstofrags(fs, lblkno);
1364 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
1365 }
1366 return (0);
1367}
1368
1369/*
1370 * Decrement extra reference on snapshot when last name is removed.
1371 * It will not be freed until the last open reference goes away.
1372 */
1373void
1374ffs_snapgone(ip)
1375 struct inode *ip;
1376{
1377 struct inode *xp;
1378 struct fs *fs;
1379 int snaploc;
1380 struct snapdata *sn;
1381
1382 /*
1383 * Find snapshot in incore list.
1384 */
1385 xp = NULL;
1386 sn = ip->i_devvp->v_rdev->si_snapdata;
1387 if (sn != NULL)
1388 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap)
1389 if (xp == ip)
1390 break;
1391 if (xp != NULL)
1392 vrele(ITOV(ip));
1393 else if (snapdebug)
1394 printf("ffs_snapgone: lost snapshot vnode %d\n",
1395 ip->i_number);
1396 /*
1397 * Delete snapshot inode from superblock. Keep list dense.
1398 */
1399 fs = ip->i_fs;
1400 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
1401 if (fs->fs_snapinum[snaploc] == ip->i_number)
1402 break;
1403 if (snaploc < FSMAXSNAP) {
1404 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
1405 if (fs->fs_snapinum[snaploc] == 0)
1406 break;
1407 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
1408 }
1409 fs->fs_snapinum[snaploc - 1] = 0;
1410 }
1411}
1412
1413/*
1414 * Prepare a snapshot file for being removed.
1415 */
1416void
1417ffs_snapremove(vp)
1418 struct vnode *vp;
1419{
1420 struct inode *ip;
1421 struct vnode *devvp;
1422 struct lock *lkp;
1423 struct buf *ibp;
1424 struct fs *fs;
1425 struct thread *td = curthread;
1426 ufs2_daddr_t numblks, blkno, dblk, *snapblklist;
1427 int error, loc, last;
1428 struct snapdata *sn;
1429
1430 ip = VTOI(vp);
1431 fs = ip->i_fs;
1432 devvp = ip->i_devvp;
1433 sn = devvp->v_rdev->si_snapdata;
1434 /*
1435 * If active, delete from incore list (this snapshot may
1436 * already have been in the process of being deleted, so
1437 * would not have been active).
1438 *
1439 * Clear copy-on-write flag if last snapshot.
1440 */
1441 if (ip->i_nextsnap.tqe_prev != 0) {
1442 VI_LOCK(devvp);
1443 lockmgr(&vp->v_lock, LK_INTERLOCK | LK_EXCLUSIVE,
1444 VI_MTX(devvp), td);
1445 VI_LOCK(devvp);
1446 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap);
1447 ip->i_nextsnap.tqe_prev = 0;
1448 lkp = vp->v_vnlock;
1449 vp->v_vnlock = &vp->v_lock;
1450 lockmgr(lkp, LK_RELEASE, NULL, td);
1451 if (TAILQ_FIRST(&sn->sn_head) != 0) {
1452 VI_UNLOCK(devvp);
1453 } else {
1454 devvp->v_rdev->si_copyonwrite = 0;
1455 snapblklist = sn->sn_blklist;
1456 sn->sn_blklist = 0;
1457 sn->sn_listsize = 0;
1458 devvp->v_rdev->si_snapdata = NULL;
1459 devvp->v_vflag &= ~VV_COPYONWRITE;
1460 lockmgr(lkp, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp), td);
1461 lockmgr(lkp, LK_RELEASE, NULL, td);
1462 lockdestroy(lkp);
1463 free(sn, M_UFSMNT);
1464 FREE(snapblklist, M_UFSMNT);
1465 }
1466 }
1467 /*
1468 * Clear all BLK_NOCOPY fields. Pass any block claims to other
1469 * snapshots that want them (see ffs_snapblkfree below).
1470 */
1471 for (blkno = 1; blkno < NDADDR; blkno++) {
1472 dblk = DIP(ip, i_db[blkno]);
1473 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1474 DIP_SET(ip, i_db[blkno], 0);
1475 else if ((dblk == blkstofrags(fs, blkno) &&
1476 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
1477 ip->i_number))) {
1478 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) -
1479 btodb(fs->fs_bsize));
1480 DIP_SET(ip, i_db[blkno], 0);
1481 }
1482 }
1483 numblks = howmany(ip->i_size, fs->fs_bsize);
1484 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1485 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
1486 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1487 if (error)
1488 continue;
1489 if (fs->fs_size - blkno > NINDIR(fs))
1490 last = NINDIR(fs);
1491 else
1492 last = fs->fs_size - blkno;
1493 for (loc = 0; loc < last; loc++) {
1494 if (ip->i_ump->um_fstype == UFS1) {
1495 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc];
1496 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1497 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1498 else if ((dblk == blkstofrags(fs, blkno) &&
1499 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1500 fs->fs_bsize, ip->i_number))) {
1501 ip->i_din1->di_blocks -=
1502 btodb(fs->fs_bsize);
1503 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1504 }
1505 continue;
1506 }
1507 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc];
1508 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1509 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1510 else if ((dblk == blkstofrags(fs, blkno) &&
1511 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1512 fs->fs_bsize, ip->i_number))) {
1513 ip->i_din2->di_blocks -= btodb(fs->fs_bsize);
1514 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1515 }
1516 }
1517 bawrite(ibp);
1518 }
1519 /*
1520 * Clear snapshot flag and drop reference.
1521 */
1522 ip->i_flags &= ~SF_SNAPSHOT;
1523 DIP_SET(ip, i_flags, ip->i_flags);
1524 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1525}
1526
1527/*
1528 * Notification that a block is being freed. Return zero if the free
1529 * should be allowed to proceed. Return non-zero if the snapshot file
1530 * wants to claim the block. The block will be claimed if it is an
1531 * uncopied part of one of the snapshots. It will be freed if it is
1532 * either a BLK_NOCOPY or has already been copied in all of the snapshots.
1533 * If a fragment is being freed, then all snapshots that care about
1534 * it must make a copy since a snapshot file can only claim full sized
1535 * blocks. Note that if more than one snapshot file maps the block,
1536 * we can pick one at random to claim it. Since none of the snapshots
1537 * can change, we are assurred that they will all see the same unmodified
1538 * image. When deleting a snapshot file (see ffs_snapremove above), we
1539 * must push any of these claimed blocks to one of the other snapshots
1540 * that maps it. These claimed blocks are easily identified as they will
1541 * have a block number equal to their logical block number within the
1542 * snapshot. A copied block can never have this property because they
1543 * must always have been allocated from a BLK_NOCOPY location.
1544 */
1545int
1546ffs_snapblkfree(fs, devvp, bno, size, inum)
1547 struct fs *fs;
1548 struct vnode *devvp;
1549 ufs2_daddr_t bno;
1550 long size;
1551 ino_t inum;
1552{
1553 struct buf *ibp, *cbp, *savedcbp = 0;
1554 struct thread *td = curthread;
1555 struct inode *ip;
1556 struct vnode *vp = NULL;
1557 ufs_lbn_t lbn;
1558 ufs2_daddr_t blkno;
1559 int indiroff = 0, snapshot_locked = 0, error = 0, claimedblk = 0;
1560 struct snapdata *sn;
1561
1562 lbn = fragstoblks(fs, bno);
1563retry:
1564 VI_LOCK(devvp);
1565 sn = devvp->v_rdev->si_snapdata;
| 38
39#include <sys/param.h>
40#include <sys/kernel.h>
41#include <sys/systm.h>
42#include <sys/conf.h>
43#include <sys/bio.h>
44#include <sys/buf.h>
45#include <sys/proc.h>
46#include <sys/namei.h>
47#include <sys/sched.h>
48#include <sys/stat.h>
49#include <sys/malloc.h>
50#include <sys/mount.h>
51#include <sys/resource.h>
52#include <sys/resourcevar.h>
53#include <sys/vnode.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#define KERNCRED thread0.td_ucred
65#define DEBUG 1
66
67TAILQ_HEAD(snaphead, inode);
68
69struct snapdata {
70 struct snaphead sn_head;
71 daddr_t sn_listsize;
72 daddr_t *sn_blklist;
73 struct lock sn_lock;
74};
75
76static int cgaccount(int, struct vnode *, struct buf *, int);
77static int expunge_ufs1(struct vnode *, struct inode *, struct fs *,
78 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
79 ufs_lbn_t, int), int);
80static int indiracct_ufs1(struct vnode *, struct vnode *, int,
81 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
82 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
83 ufs_lbn_t, int), int);
84static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
85 struct fs *, ufs_lbn_t, int);
86static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
87 struct fs *, ufs_lbn_t, int);
88static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
89 struct fs *, ufs_lbn_t, int);
90static int expunge_ufs2(struct vnode *, struct inode *, struct fs *,
91 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
92 ufs_lbn_t, int), int);
93static int indiracct_ufs2(struct vnode *, struct vnode *, int,
94 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
95 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
96 ufs_lbn_t, int), int);
97static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
98 struct fs *, ufs_lbn_t, int);
99static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
100 struct fs *, ufs_lbn_t, int);
101static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
102 struct fs *, ufs_lbn_t, int);
103static int ffs_copyonwrite(struct vnode *, struct buf *);
104static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t);
105
106/*
107 * To ensure the consistency of snapshots across crashes, we must
108 * synchronously write out copied blocks before allowing the
109 * originals to be modified. Because of the rather severe speed
110 * penalty that this imposes, the following flag allows this
111 * crash persistence to be disabled.
112 */
113int dopersistence = 0;
114
115#ifdef DEBUG
116#include <sys/sysctl.h>
117SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, "");
118static int snapdebug = 0;
119SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, "");
120int collectsnapstats = 0;
121SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats,
122 0, "");
123#endif /* DEBUG */
124
125/*
126 * Create a snapshot file and initialize it for the filesystem.
127 */
128int
129ffs_snapshot(mp, snapfile)
130 struct mount *mp;
131 char *snapfile;
132{
133 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist;
134 int error, cg, snaploc;
135 int i, size, len, loc;
136 int flag = mp->mnt_flag;
137 struct timespec starttime = {0, 0}, endtime;
138 char saved_nice = 0;
139 long redo = 0, snaplistsize = 0;
140 int32_t *lp;
141 void *space;
142 struct fs *copy_fs = NULL, *fs = VFSTOUFS(mp)->um_fs;
143 struct thread *td = curthread;
144 struct inode *ip, *xp;
145 struct buf *bp, *nbp, *ibp, *sbp = NULL;
146 struct nameidata nd;
147 struct mount *wrtmp;
148 struct vattr vat;
149 struct vnode *vp, *xvp, *nvp, *devvp;
150 struct uio auio;
151 struct iovec aiov;
152 struct snapdata *sn;
153
154 /*
155 * XXX: make sure we don't go to out1 before we setup sn
156 */
157 sn = (void *)0xdeadbeef;
158
159 /*
160 * Need to serialize access to snapshot code per filesystem.
161 */
162 /*
163 * Assign a snapshot slot in the superblock.
164 */
165 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
166 if (fs->fs_snapinum[snaploc] == 0)
167 break;
168 if (snaploc == FSMAXSNAP)
169 return (ENOSPC);
170 /*
171 * Create the snapshot file.
172 */
173restart:
174 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_USERSPACE, snapfile, td);
175 if ((error = namei(&nd)) != 0)
176 return (error);
177 if (nd.ni_vp != NULL) {
178 vput(nd.ni_vp);
179 error = EEXIST;
180 }
181 if (nd.ni_dvp->v_mount != mp)
182 error = EXDEV;
183 if (error) {
184 NDFREE(&nd, NDF_ONLY_PNBUF);
185 if (nd.ni_dvp == nd.ni_vp)
186 vrele(nd.ni_dvp);
187 else
188 vput(nd.ni_dvp);
189 return (error);
190 }
191 VATTR_NULL(&vat);
192 vat.va_type = VREG;
193 vat.va_mode = S_IRUSR;
194 vat.va_vaflags |= VA_EXCLUSIVE;
195 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp))
196 wrtmp = NULL;
197 if (wrtmp != mp)
198 panic("ffs_snapshot: mount mismatch");
199 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) {
200 NDFREE(&nd, NDF_ONLY_PNBUF);
201 vput(nd.ni_dvp);
202 if ((error = vn_start_write(NULL, &wrtmp,
203 V_XSLEEP | PCATCH)) != 0)
204 return (error);
205 goto restart;
206 }
207 VOP_LEASE(nd.ni_dvp, td, KERNCRED, LEASE_WRITE);
208 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat);
209 vput(nd.ni_dvp);
210 if (error) {
211 NDFREE(&nd, NDF_ONLY_PNBUF);
212 vn_finished_write(wrtmp);
213 return (error);
214 }
215 vp = nd.ni_vp;
216 ip = VTOI(vp);
217 devvp = ip->i_devvp;
218 /*
219 * Allocate and copy the last block contents so as to be able
220 * to set size to that of the filesystem.
221 */
222 numblks = howmany(fs->fs_size, fs->fs_frag);
223 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
224 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
225 if (error)
226 goto out;
227 ip->i_size = lblktosize(fs, (off_t)numblks);
228 DIP_SET(ip, i_size, ip->i_size);
229 ip->i_flag |= IN_CHANGE | IN_UPDATE;
230 if ((error = readblock(vp, bp, numblks - 1)) != 0)
231 goto out;
232 bawrite(bp);
233 /*
234 * Preallocate critical data structures so that we can copy
235 * them in without further allocation after we suspend all
236 * operations on the filesystem. We would like to just release
237 * the allocated buffers without writing them since they will
238 * be filled in below once we are ready to go, but this upsets
239 * the soft update code, so we go ahead and write the new buffers.
240 *
241 * Allocate all indirect blocks and mark all of them as not
242 * needing to be copied.
243 */
244 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
245 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
246 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp);
247 if (error)
248 goto out;
249 bawrite(ibp);
250 }
251 /*
252 * Allocate copies for the superblock and its summary information.
253 */
254 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED,
255 0, &nbp);
256 if (error)
257 goto out;
258 bawrite(nbp);
259 blkno = fragstoblks(fs, fs->fs_csaddr);
260 len = howmany(fs->fs_cssize, fs->fs_bsize);
261 for (loc = 0; loc < len; loc++) {
262 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)),
263 fs->fs_bsize, KERNCRED, 0, &nbp);
264 if (error)
265 goto out;
266 bawrite(nbp);
267 }
268 /*
269 * Allocate all cylinder group blocks.
270 */
271 for (cg = 0; cg < fs->fs_ncg; cg++) {
272 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
273 fs->fs_bsize, KERNCRED, 0, &nbp);
274 if (error)
275 goto out;
276 bawrite(nbp);
277 }
278 /*
279 * Copy all the cylinder group maps. Although the
280 * filesystem is still active, we hope that only a few
281 * cylinder groups will change between now and when we
282 * suspend operations. Thus, we will be able to quickly
283 * touch up the few cylinder groups that changed during
284 * the suspension period.
285 */
286 len = howmany(fs->fs_ncg, NBBY);
287 MALLOC(fs->fs_active, int *, len, M_DEVBUF, M_WAITOK);
288 bzero(fs->fs_active, len);
289 for (cg = 0; cg < fs->fs_ncg; cg++) {
290 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
291 fs->fs_bsize, KERNCRED, 0, &nbp);
292 if (error)
293 goto out;
294 error = cgaccount(cg, vp, nbp, 1);
295 bawrite(nbp);
296 if (error)
297 goto out;
298 }
299 /*
300 * Change inode to snapshot type file.
301 */
302 ip->i_flags |= SF_SNAPSHOT;
303 DIP_SET(ip, i_flags, ip->i_flags);
304 ip->i_flag |= IN_CHANGE | IN_UPDATE;
305 /*
306 * Ensure that the snapshot is completely on disk.
307 * Since we have marked it as a snapshot it is safe to
308 * unlock it as no process will be allowed to write to it.
309 */
310 if ((error = VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td)) != 0)
311 goto out;
312 VOP_UNLOCK(vp, 0, td);
313 /*
314 * All allocations are done, so we can now snapshot the system.
315 *
316 * Recind nice scheduling while running with the filesystem suspended.
317 */
318 if (td->td_proc->p_nice > 0) {
319 PROC_LOCK(td->td_proc);
320 mtx_lock_spin(&sched_lock);
321 saved_nice = td->td_proc->p_nice;
322 sched_nice(td->td_proc, 0);
323 mtx_unlock_spin(&sched_lock);
324 PROC_UNLOCK(td->td_proc);
325 }
326 /*
327 * Suspend operation on filesystem.
328 */
329 for (;;) {
330 vn_finished_write(wrtmp);
331 if ((error = vfs_write_suspend(vp->v_mount)) != 0) {
332 vn_start_write(NULL, &wrtmp, V_WAIT);
333 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
334 goto out;
335 }
336 if (mp->mnt_kern_flag & MNTK_SUSPENDED)
337 break;
338 vn_start_write(NULL, &wrtmp, V_WAIT);
339 }
340 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
341 if (collectsnapstats)
342 nanotime(&starttime);
343 /*
344 * First, copy all the cylinder group maps that have changed.
345 */
346 for (cg = 0; cg < fs->fs_ncg; cg++) {
347 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0)
348 continue;
349 redo++;
350 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
351 fs->fs_bsize, KERNCRED, 0, &nbp);
352 if (error)
353 goto out1;
354 error = cgaccount(cg, vp, nbp, 2);
355 bawrite(nbp);
356 if (error)
357 goto out1;
358 }
359 /*
360 * Grab a copy of the superblock and its summary information.
361 * We delay writing it until the suspension is released below.
362 */
363 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize,
364 KERNCRED, &sbp);
365 if (error) {
366 brelse(sbp);
367 sbp = NULL;
368 goto out1;
369 }
370 loc = blkoff(fs, fs->fs_sblockloc);
371 copy_fs = (struct fs *)(sbp->b_data + loc);
372 bcopy(fs, copy_fs, fs->fs_sbsize);
373 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
374 copy_fs->fs_clean = 1;
375 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
376 if (fs->fs_sbsize < size)
377 bzero(&sbp->b_data[loc + fs->fs_sbsize], size - fs->fs_sbsize);
378 size = blkroundup(fs, fs->fs_cssize);
379 if (fs->fs_contigsumsize > 0)
380 size += fs->fs_ncg * sizeof(int32_t);
381 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
382 copy_fs->fs_csp = space;
383 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize);
384 space = (char *)space + fs->fs_cssize;
385 loc = howmany(fs->fs_cssize, fs->fs_fsize);
386 i = fs->fs_frag - loc % fs->fs_frag;
387 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
388 if (len > 0) {
389 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
390 len, KERNCRED, &bp)) != 0) {
391 brelse(bp);
392 free(copy_fs->fs_csp, M_UFSMNT);
393 bawrite(sbp);
394 sbp = NULL;
395 goto out1;
396 }
397 bcopy(bp->b_data, space, (u_int)len);
398 space = (char *)space + len;
399 bp->b_flags |= B_INVAL | B_NOCACHE;
400 brelse(bp);
401 }
402 if (fs->fs_contigsumsize > 0) {
403 copy_fs->fs_maxcluster = lp = space;
404 for (i = 0; i < fs->fs_ncg; i++)
405 *lp++ = fs->fs_contigsumsize;
406 }
407 /*
408 * We must check for active files that have been unlinked
409 * (e.g., with a zero link count). We have to expunge all
410 * trace of these files from the snapshot so that they are
411 * not reclaimed prematurely by fsck or unnecessarily dumped.
412 * We turn off the MNTK_SUSPENDED flag to avoid a panic from
413 * spec_strategy about writing on a suspended filesystem.
414 * Note that we skip unlinked snapshot files as they will
415 * be handled separately below.
416 *
417 * We also calculate the needed size for the snapshot list.
418 */
419 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
420 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
421 mp->mnt_kern_flag &= ~MNTK_SUSPENDED;
422 MNT_ILOCK(mp);
423loop:
424 MNT_VNODE_FOREACH(xvp, mp, nvp) {
425 VI_LOCK(xvp);
426 MNT_IUNLOCK(mp);
427 if ((xvp->v_iflag & VI_XLOCK) ||
428 xvp->v_usecount == 0 || xvp->v_type == VNON ||
429 (VTOI(xvp)->i_flags & SF_SNAPSHOT)) {
430 VI_UNLOCK(xvp);
431 MNT_ILOCK(mp);
432 continue;
433 }
434 /*
435 * We can skip parent directory vnode because it must have
436 * this snapshot file in it.
437 */
438 if (xvp == nd.ni_dvp) {
439 VI_UNLOCK(xvp);
440 MNT_ILOCK(mp);
441 continue;
442 }
443 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK, td) != 0) {
444 MNT_ILOCK(mp);
445 goto loop;
446 }
447 if (snapdebug)
448 vprint("ffs_snapshot: busy vnode", xvp);
449 if (VOP_GETATTR(xvp, &vat, td->td_ucred, td) == 0 &&
450 vat.va_nlink > 0) {
451 VOP_UNLOCK(xvp, 0, td);
452 MNT_ILOCK(mp);
453 continue;
454 }
455 xp = VTOI(xvp);
456 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
457 VOP_UNLOCK(xvp, 0, td);
458 MNT_ILOCK(mp);
459 continue;
460 }
461 /*
462 * If there is a fragment, clear it here.
463 */
464 blkno = 0;
465 loc = howmany(xp->i_size, fs->fs_bsize) - 1;
466 if (loc < NDADDR) {
467 len = fragroundup(fs, blkoff(fs, xp->i_size));
468 if (len < fs->fs_bsize) {
469 ffs_blkfree(copy_fs, vp, DIP(xp, i_db[loc]),
470 len, xp->i_number);
471 blkno = DIP(xp, i_db[loc]);
472 DIP_SET(xp, i_db[loc], 0);
473 }
474 }
475 snaplistsize += 1;
476 if (xp->i_ump->um_fstype == UFS1)
477 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
478 BLK_NOCOPY);
479 else
480 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
481 BLK_NOCOPY);
482 if (blkno)
483 DIP_SET(xp, i_db[loc], blkno);
484 if (!error)
485 error = ffs_freefile(copy_fs, vp, xp->i_number,
486 xp->i_mode);
487 VOP_UNLOCK(xvp, 0, td);
488 if (error) {
489 free(copy_fs->fs_csp, M_UFSMNT);
490 bawrite(sbp);
491 sbp = NULL;
492 goto out1;
493 }
494 MNT_ILOCK(mp);
495 }
496 MNT_IUNLOCK(mp);
497 /*
498 * If there already exist snapshots on this filesystem, grab a
499 * reference to their shared lock. If this is the first snapshot
500 * on this filesystem, we need to allocate a lock for the snapshots
501 * to share. In either case, acquire the snapshot lock and give
502 * up our original private lock.
503 */
504 VI_LOCK(devvp);
505 sn = devvp->v_rdev->si_snapdata;
506 if (sn != NULL) {
507 xp = TAILQ_FIRST(&sn->sn_head);
508 VI_UNLOCK(devvp);
509 VI_LOCK(vp);
510 vp->v_vnlock = &sn->sn_lock;
511 } else {
512 VI_UNLOCK(devvp);
513 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO);
514 TAILQ_INIT(&sn->sn_head);
515 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT,
516 LK_CANRECURSE | LK_NOPAUSE);
517 VI_LOCK(vp);
518 vp->v_vnlock = &sn->sn_lock;
519 devvp->v_rdev->si_snapdata = sn;
520 xp = NULL;
521 }
522 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
523 transferlockers(&vp->v_lock, vp->v_vnlock);
524 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td);
525 /*
526 * If this is the first snapshot on this filesystem, then we need
527 * to allocate the space for the list of preallocated snapshot blocks.
528 * This list will be refined below, but this preliminary one will
529 * keep us out of deadlock until the full one is ready.
530 */
531 if (xp == NULL) {
532 MALLOC(snapblklist, daddr_t *, snaplistsize * sizeof(daddr_t),
533 M_UFSMNT, M_WAITOK);
534 blkp = &snapblklist[1];
535 *blkp++ = lblkno(fs, fs->fs_sblockloc);
536 blkno = fragstoblks(fs, fs->fs_csaddr);
537 for (cg = 0; cg < fs->fs_ncg; cg++) {
538 if (fragstoblks(fs, cgtod(fs, cg) > blkno))
539 break;
540 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
541 }
542 len = howmany(fs->fs_cssize, fs->fs_bsize);
543 for (loc = 0; loc < len; loc++)
544 *blkp++ = blkno + loc;
545 for (; cg < fs->fs_ncg; cg++)
546 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
547 snapblklist[0] = blkp - snapblklist;
548 VI_LOCK(devvp);
549 if (sn->sn_blklist != NULL)
550 panic("ffs_snapshot: non-empty list");
551 sn->sn_blklist = snapblklist;
552 sn->sn_listsize = blkp - snapblklist;
553 VI_UNLOCK(devvp);
554 }
555 /*
556 * Record snapshot inode. Since this is the newest snapshot,
557 * it must be placed at the end of the list.
558 */
559 VI_LOCK(devvp);
560 fs->fs_snapinum[snaploc] = ip->i_number;
561 if (ip->i_nextsnap.tqe_prev != 0)
562 panic("ffs_snapshot: %d already on list", ip->i_number);
563 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
564 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
565 devvp->v_vflag |= VV_COPYONWRITE;
566 VI_UNLOCK(devvp);
567 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp");
568 vp->v_vflag |= VV_SYSTEM;
569out1:
570 KASSERT(sn != (void *)0xdeadbeef, ("email phk@ and mckusick@"));
571 /*
572 * Resume operation on filesystem.
573 */
574 vfs_write_resume(vp->v_mount);
575 vn_start_write(NULL, &wrtmp, V_WAIT);
576 if (collectsnapstats && starttime.tv_sec > 0) {
577 nanotime(&endtime);
578 timespecsub(&endtime, &starttime);
579 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n",
580 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec,
581 endtime.tv_nsec / 1000000, redo, fs->fs_ncg);
582 }
583 if (sbp == NULL)
584 goto out;
585 /*
586 * Copy allocation information from all the snapshots in
587 * this snapshot and then expunge them from its view.
588 */
589 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) {
590 if (xp == ip)
591 break;
592 if (xp->i_ump->um_fstype == UFS1)
593 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1,
594 BLK_SNAP);
595 else
596 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2,
597 BLK_SNAP);
598 if (error) {
599 fs->fs_snapinum[snaploc] = 0;
600 goto done;
601 }
602 }
603 /*
604 * Allocate space for the full list of preallocated snapshot blocks.
605 */
606 MALLOC(snapblklist, daddr_t *, snaplistsize * sizeof(daddr_t),
607 M_UFSMNT, M_WAITOK);
608 ip->i_snapblklist = &snapblklist[1];
609 /*
610 * Expunge the blocks used by the snapshots from the set of
611 * blocks marked as used in the snapshot bitmaps. Also, collect
612 * the list of allocated blocks in i_snapblklist.
613 */
614 if (ip->i_ump->um_fstype == UFS1)
615 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, BLK_SNAP);
616 else
617 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, BLK_SNAP);
618 if (error) {
619 fs->fs_snapinum[snaploc] = 0;
620 FREE(snapblklist, M_UFSMNT);
621 goto done;
622 }
623 if (snaplistsize < ip->i_snapblklist - snapblklist)
624 panic("ffs_snapshot: list too small");
625 snaplistsize = ip->i_snapblklist - snapblklist;
626 snapblklist[0] = snaplistsize;
627 ip->i_snapblklist = 0;
628 /*
629 * Write out the list of allocated blocks to the end of the snapshot.
630 */
631 auio.uio_iov = &aiov;
632 auio.uio_iovcnt = 1;
633 aiov.iov_base = (void *)snapblklist;
634 aiov.iov_len = snaplistsize * sizeof(daddr_t);
635 auio.uio_resid = aiov.iov_len;;
636 auio.uio_offset = ip->i_size;
637 auio.uio_segflg = UIO_SYSSPACE;
638 auio.uio_rw = UIO_WRITE;
639 auio.uio_td = td;
640 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
641 fs->fs_snapinum[snaploc] = 0;
642 FREE(snapblklist, M_UFSMNT);
643 goto done;
644 }
645 /*
646 * Write the superblock and its summary information
647 * to the snapshot.
648 */
649 blkno = fragstoblks(fs, fs->fs_csaddr);
650 len = howmany(fs->fs_cssize, fs->fs_bsize);
651 space = copy_fs->fs_csp;
652 for (loc = 0; loc < len; loc++) {
653 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp);
654 if (error) {
655 brelse(nbp);
656 fs->fs_snapinum[snaploc] = 0;
657 FREE(snapblklist, M_UFSMNT);
658 goto done;
659 }
660 bcopy(space, nbp->b_data, fs->fs_bsize);
661 space = (char *)space + fs->fs_bsize;
662 bawrite(nbp);
663 }
664 /*
665 * As this is the newest list, it is the most inclusive, so
666 * should replace the previous list.
667 */
668 VI_LOCK(devvp);
669 space = sn->sn_blklist;
670 sn->sn_blklist = snapblklist;
671 sn->sn_listsize = snaplistsize;
672 VI_UNLOCK(devvp);
673 if (space != NULL)
674 FREE(space, M_UFSMNT);
675done:
676 FREE(copy_fs->fs_csp, M_UFSMNT);
677 bawrite(sbp);
678out:
679 if (saved_nice > 0) {
680 PROC_LOCK(td->td_proc);
681 mtx_lock_spin(&sched_lock);
682 sched_nice(td->td_proc, saved_nice);
683 mtx_unlock_spin(&sched_lock);
684 PROC_UNLOCK(td->td_proc);
685 }
686 if (fs->fs_active != 0) {
687 FREE(fs->fs_active, M_DEVBUF);
688 fs->fs_active = 0;
689 }
690 mp->mnt_flag = flag;
691 if (error)
692 (void) UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td);
693 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
694 if (error)
695 vput(vp);
696 else
697 VOP_UNLOCK(vp, 0, td);
698 vn_finished_write(wrtmp);
699 return (error);
700}
701
702/*
703 * Copy a cylinder group map. All the unallocated blocks are marked
704 * BLK_NOCOPY so that the snapshot knows that it need not copy them
705 * if they are later written. If passno is one, then this is a first
706 * pass, so only setting needs to be done. If passno is 2, then this
707 * is a revision to a previous pass which must be undone as the
708 * replacement pass is done.
709 */
710static int
711cgaccount(cg, vp, nbp, passno)
712 int cg;
713 struct vnode *vp;
714 struct buf *nbp;
715 int passno;
716{
717 struct buf *bp, *ibp;
718 struct inode *ip;
719 struct cg *cgp;
720 struct fs *fs;
721 ufs2_daddr_t base, numblks;
722 int error, len, loc, indiroff;
723
724 ip = VTOI(vp);
725 fs = ip->i_fs;
726 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
727 (int)fs->fs_cgsize, KERNCRED, &bp);
728 if (error) {
729 brelse(bp);
730 return (error);
731 }
732 cgp = (struct cg *)bp->b_data;
733 if (!cg_chkmagic(cgp)) {
734 brelse(bp);
735 return (EIO);
736 }
737 atomic_set_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
738 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize);
739 if (fs->fs_cgsize < fs->fs_bsize)
740 bzero(&nbp->b_data[fs->fs_cgsize],
741 fs->fs_bsize - fs->fs_cgsize);
742 if (passno == 2)
743 nbp->b_flags |= B_VALIDSUSPWRT;
744 numblks = howmany(fs->fs_size, fs->fs_frag);
745 len = howmany(fs->fs_fpg, fs->fs_frag);
746 base = cg * fs->fs_fpg / fs->fs_frag;
747 if (base + len >= numblks)
748 len = numblks - base - 1;
749 loc = 0;
750 if (base < NDADDR) {
751 for ( ; loc < NDADDR; loc++) {
752 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
753 DIP_SET(ip, i_db[loc], BLK_NOCOPY);
754 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY)
755 DIP_SET(ip, i_db[loc], 0);
756 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY)
757 panic("ffs_snapshot: lost direct block");
758 }
759 }
760 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)),
761 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
762 if (error) {
763 brelse(bp);
764 return (error);
765 }
766 indiroff = (base + loc - NDADDR) % NINDIR(fs);
767 for ( ; loc < len; loc++, indiroff++) {
768 if (indiroff >= NINDIR(fs)) {
769 if (passno == 2)
770 ibp->b_flags |= B_VALIDSUSPWRT;
771 bawrite(ibp);
772 error = UFS_BALLOC(vp,
773 lblktosize(fs, (off_t)(base + loc)),
774 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
775 if (error) {
776 brelse(bp);
777 return (error);
778 }
779 indiroff = 0;
780 }
781 if (ip->i_ump->um_fstype == UFS1) {
782 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
783 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
784 BLK_NOCOPY;
785 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data))
786 [indiroff] == BLK_NOCOPY)
787 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0;
788 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data))
789 [indiroff] == BLK_NOCOPY)
790 panic("ffs_snapshot: lost indirect block");
791 continue;
792 }
793 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
794 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
795 else if (passno == 2 &&
796 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
797 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0;
798 else if (passno == 1 &&
799 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
800 panic("ffs_snapshot: lost indirect block");
801 }
802 bqrelse(bp);
803 if (passno == 2)
804 ibp->b_flags |= B_VALIDSUSPWRT;
805 bdwrite(ibp);
806 return (0);
807}
808
809/*
810 * Before expunging a snapshot inode, note all the
811 * blocks that it claims with BLK_SNAP so that fsck will
812 * be able to account for those blocks properly and so
813 * that this snapshot knows that it need not copy them
814 * if the other snapshot holding them is freed. This code
815 * is reproduced once each for UFS1 and UFS2.
816 */
817static int
818expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype)
819 struct vnode *snapvp;
820 struct inode *cancelip;
821 struct fs *fs;
822 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
823 struct fs *, ufs_lbn_t, int);
824 int expungetype;
825{
826 int i, error, indiroff;
827 ufs_lbn_t lbn, rlbn;
828 ufs2_daddr_t len, blkno, numblks, blksperindir;
829 struct ufs1_dinode *dip;
830 struct thread *td = curthread;
831 struct buf *bp;
832
833 /*
834 * Prepare to expunge the inode. If its inode block has not
835 * yet been copied, then allocate and fill the copy.
836 */
837 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
838 blkno = 0;
839 if (lbn < NDADDR) {
840 blkno = VTOI(snapvp)->i_din1->di_db[lbn];
841 } else {
842 td->td_pflags |= TDP_COWINPROGRESS;
843 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
844 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
845 td->td_pflags &= ~TDP_COWINPROGRESS;
846 if (error)
847 return (error);
848 indiroff = (lbn - NDADDR) % NINDIR(fs);
849 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
850 bqrelse(bp);
851 }
852 if (blkno != 0) {
853 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
854 return (error);
855 } else {
856 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
857 fs->fs_bsize, KERNCRED, 0, &bp);
858 if (error)
859 return (error);
860 if ((error = readblock(snapvp, bp, lbn)) != 0)
861 return (error);
862 }
863 /*
864 * Set a snapshot inode to be a zero length file, regular files
865 * to be completely unallocated.
866 */
867 dip = (struct ufs1_dinode *)bp->b_data +
868 ino_to_fsbo(fs, cancelip->i_number);
869 if (expungetype == BLK_NOCOPY)
870 dip->di_mode = 0;
871 dip->di_size = 0;
872 dip->di_blocks = 0;
873 dip->di_flags &= ~SF_SNAPSHOT;
874 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t));
875 bdwrite(bp);
876 /*
877 * Now go through and expunge all the blocks in the file
878 * using the function requested.
879 */
880 numblks = howmany(cancelip->i_size, fs->fs_bsize);
881 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0],
882 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype)))
883 return (error);
884 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0],
885 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype)))
886 return (error);
887 blksperindir = 1;
888 lbn = -NDADDR;
889 len = numblks - NDADDR;
890 rlbn = NDADDR;
891 for (i = 0; len > 0 && i < NIADDR; i++) {
892 error = indiracct_ufs1(snapvp, ITOV(cancelip), i,
893 cancelip->i_din1->di_ib[i], lbn, rlbn, len,
894 blksperindir, fs, acctfunc, expungetype);
895 if (error)
896 return (error);
897 blksperindir *= NINDIR(fs);
898 lbn -= blksperindir + 1;
899 len -= blksperindir;
900 rlbn += blksperindir;
901 }
902 return (0);
903}
904
905/*
906 * Descend an indirect block chain for vnode cancelvp accounting for all
907 * its indirect blocks in snapvp.
908 */
909static int
910indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
911 blksperindir, fs, acctfunc, expungetype)
912 struct vnode *snapvp;
913 struct vnode *cancelvp;
914 int level;
915 ufs1_daddr_t blkno;
916 ufs_lbn_t lbn;
917 ufs_lbn_t rlbn;
918 ufs_lbn_t remblks;
919 ufs_lbn_t blksperindir;
920 struct fs *fs;
921 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
922 struct fs *, ufs_lbn_t, int);
923 int expungetype;
924{
925 int error, num, i;
926 ufs_lbn_t subblksperindir;
927 struct indir indirs[NIADDR + 2];
928 ufs1_daddr_t last, *bap;
929 struct buf *bp;
930
931 if (blkno == 0) {
932 if (expungetype == BLK_NOCOPY)
933 return (0);
934 panic("indiracct_ufs1: missing indir");
935 }
936 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
937 return (error);
938 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
939 panic("indiracct_ufs1: botched params");
940 /*
941 * We have to expand bread here since it will deadlock looking
942 * up the block number for any blocks that are not in the cache.
943 */
944 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
945 bp->b_blkno = fsbtodb(fs, blkno);
946 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
947 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
948 brelse(bp);
949 return (error);
950 }
951 /*
952 * Account for the block pointers in this indirect block.
953 */
954 last = howmany(remblks, blksperindir);
955 if (last > NINDIR(fs))
956 last = NINDIR(fs);
957 MALLOC(bap, ufs1_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK);
958 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
959 bqrelse(bp);
960 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
961 level == 0 ? rlbn : -1, expungetype);
962 if (error || level == 0)
963 goto out;
964 /*
965 * Account for the block pointers in each of the indirect blocks
966 * in the levels below us.
967 */
968 subblksperindir = blksperindir / NINDIR(fs);
969 for (lbn++, level--, i = 0; i < last; i++) {
970 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn,
971 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
972 if (error)
973 goto out;
974 rlbn += blksperindir;
975 lbn -= blksperindir;
976 remblks -= blksperindir;
977 }
978out:
979 FREE(bap, M_DEVBUF);
980 return (error);
981}
982
983/*
984 * Do both snap accounting and map accounting.
985 */
986static int
987fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)
988 struct vnode *vp;
989 ufs1_daddr_t *oldblkp, *lastblkp;
990 struct fs *fs;
991 ufs_lbn_t lblkno;
992 int exptype; /* BLK_SNAP or BLK_NOCOPY */
993{
994 int error;
995
996 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
997 return (error);
998 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype));
999}
1000
1001/*
1002 * Identify a set of blocks allocated in a snapshot inode.
1003 */
1004static int
1005snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1006 struct vnode *vp;
1007 ufs1_daddr_t *oldblkp, *lastblkp;
1008 struct fs *fs;
1009 ufs_lbn_t lblkno;
1010 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1011{
1012 struct inode *ip = VTOI(vp);
1013 ufs1_daddr_t blkno, *blkp;
1014 ufs_lbn_t lbn;
1015 struct buf *ibp;
1016 int error;
1017
1018 for ( ; oldblkp < lastblkp; oldblkp++) {
1019 blkno = *oldblkp;
1020 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1021 continue;
1022 lbn = fragstoblks(fs, blkno);
1023 if (lbn < NDADDR) {
1024 blkp = &ip->i_din1->di_db[lbn];
1025 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1026 } else {
1027 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1028 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1029 if (error)
1030 return (error);
1031 blkp = &((ufs1_daddr_t *)(ibp->b_data))
1032 [(lbn - NDADDR) % NINDIR(fs)];
1033 }
1034 /*
1035 * If we are expunging a snapshot vnode and we
1036 * find a block marked BLK_NOCOPY, then it is
1037 * one that has been allocated to this snapshot after
1038 * we took our current snapshot and can be ignored.
1039 */
1040 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1041 if (lbn >= NDADDR)
1042 brelse(ibp);
1043 } else {
1044 if (*blkp != 0)
1045 panic("snapacct_ufs1: bad block");
1046 *blkp = expungetype;
1047 if (lbn >= NDADDR)
1048 bdwrite(ibp);
1049 }
1050 }
1051 return (0);
1052}
1053
1054/*
1055 * Account for a set of blocks allocated in a snapshot inode.
1056 */
1057static int
1058mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1059 struct vnode *vp;
1060 ufs1_daddr_t *oldblkp, *lastblkp;
1061 struct fs *fs;
1062 ufs_lbn_t lblkno;
1063 int expungetype;
1064{
1065 ufs1_daddr_t blkno;
1066 struct inode *ip;
1067 ino_t inum;
1068 int acctit;
1069
1070 ip = VTOI(vp);
1071 inum = ip->i_number;
1072 if (lblkno == -1)
1073 acctit = 0;
1074 else
1075 acctit = 1;
1076 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1077 blkno = *oldblkp;
1078 if (blkno == 0 || blkno == BLK_NOCOPY)
1079 continue;
1080 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1081 *ip->i_snapblklist++ = lblkno;
1082 if (blkno == BLK_SNAP)
1083 blkno = blkstofrags(fs, lblkno);
1084 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
1085 }
1086 return (0);
1087}
1088
1089/*
1090 * Before expunging a snapshot inode, note all the
1091 * blocks that it claims with BLK_SNAP so that fsck will
1092 * be able to account for those blocks properly and so
1093 * that this snapshot knows that it need not copy them
1094 * if the other snapshot holding them is freed. This code
1095 * is reproduced once each for UFS1 and UFS2.
1096 */
1097static int
1098expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype)
1099 struct vnode *snapvp;
1100 struct inode *cancelip;
1101 struct fs *fs;
1102 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1103 struct fs *, ufs_lbn_t, int);
1104 int expungetype;
1105{
1106 int i, error, indiroff;
1107 ufs_lbn_t lbn, rlbn;
1108 ufs2_daddr_t len, blkno, numblks, blksperindir;
1109 struct ufs2_dinode *dip;
1110 struct thread *td = curthread;
1111 struct buf *bp;
1112
1113 /*
1114 * Prepare to expunge the inode. If its inode block has not
1115 * yet been copied, then allocate and fill the copy.
1116 */
1117 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1118 blkno = 0;
1119 if (lbn < NDADDR) {
1120 blkno = VTOI(snapvp)->i_din2->di_db[lbn];
1121 } else {
1122 td->td_pflags |= TDP_COWINPROGRESS;
1123 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
1124 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1125 td->td_pflags &= ~TDP_COWINPROGRESS;
1126 if (error)
1127 return (error);
1128 indiroff = (lbn - NDADDR) % NINDIR(fs);
1129 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
1130 bqrelse(bp);
1131 }
1132 if (blkno != 0) {
1133 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1134 return (error);
1135 } else {
1136 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
1137 fs->fs_bsize, KERNCRED, 0, &bp);
1138 if (error)
1139 return (error);
1140 if ((error = readblock(snapvp, bp, lbn)) != 0)
1141 return (error);
1142 }
1143 /*
1144 * Set a snapshot inode to be a zero length file, regular files
1145 * to be completely unallocated.
1146 */
1147 dip = (struct ufs2_dinode *)bp->b_data +
1148 ino_to_fsbo(fs, cancelip->i_number);
1149 if (expungetype == BLK_NOCOPY)
1150 dip->di_mode = 0;
1151 dip->di_size = 0;
1152 dip->di_blocks = 0;
1153 dip->di_flags &= ~SF_SNAPSHOT;
1154 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t));
1155 bdwrite(bp);
1156 /*
1157 * Now go through and expunge all the blocks in the file
1158 * using the function requested.
1159 */
1160 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1161 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0],
1162 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype)))
1163 return (error);
1164 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0],
1165 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype)))
1166 return (error);
1167 blksperindir = 1;
1168 lbn = -NDADDR;
1169 len = numblks - NDADDR;
1170 rlbn = NDADDR;
1171 for (i = 0; len > 0 && i < NIADDR; i++) {
1172 error = indiracct_ufs2(snapvp, ITOV(cancelip), i,
1173 cancelip->i_din2->di_ib[i], lbn, rlbn, len,
1174 blksperindir, fs, acctfunc, expungetype);
1175 if (error)
1176 return (error);
1177 blksperindir *= NINDIR(fs);
1178 lbn -= blksperindir + 1;
1179 len -= blksperindir;
1180 rlbn += blksperindir;
1181 }
1182 return (0);
1183}
1184
1185/*
1186 * Descend an indirect block chain for vnode cancelvp accounting for all
1187 * its indirect blocks in snapvp.
1188 */
1189static int
1190indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1191 blksperindir, fs, acctfunc, expungetype)
1192 struct vnode *snapvp;
1193 struct vnode *cancelvp;
1194 int level;
1195 ufs2_daddr_t blkno;
1196 ufs_lbn_t lbn;
1197 ufs_lbn_t rlbn;
1198 ufs_lbn_t remblks;
1199 ufs_lbn_t blksperindir;
1200 struct fs *fs;
1201 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1202 struct fs *, ufs_lbn_t, int);
1203 int expungetype;
1204{
1205 int error, num, i;
1206 ufs_lbn_t subblksperindir;
1207 struct indir indirs[NIADDR + 2];
1208 ufs2_daddr_t last, *bap;
1209 struct buf *bp;
1210
1211 if (blkno == 0) {
1212 if (expungetype == BLK_NOCOPY)
1213 return (0);
1214 panic("indiracct_ufs2: missing indir");
1215 }
1216 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1217 return (error);
1218 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1219 panic("indiracct_ufs2: botched params");
1220 /*
1221 * We have to expand bread here since it will deadlock looking
1222 * up the block number for any blocks that are not in the cache.
1223 */
1224 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1225 bp->b_blkno = fsbtodb(fs, blkno);
1226 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1227 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1228 brelse(bp);
1229 return (error);
1230 }
1231 /*
1232 * Account for the block pointers in this indirect block.
1233 */
1234 last = howmany(remblks, blksperindir);
1235 if (last > NINDIR(fs))
1236 last = NINDIR(fs);
1237 MALLOC(bap, ufs2_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK);
1238 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1239 bqrelse(bp);
1240 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1241 level == 0 ? rlbn : -1, expungetype);
1242 if (error || level == 0)
1243 goto out;
1244 /*
1245 * Account for the block pointers in each of the indirect blocks
1246 * in the levels below us.
1247 */
1248 subblksperindir = blksperindir / NINDIR(fs);
1249 for (lbn++, level--, i = 0; i < last; i++) {
1250 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn,
1251 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1252 if (error)
1253 goto out;
1254 rlbn += blksperindir;
1255 lbn -= blksperindir;
1256 remblks -= blksperindir;
1257 }
1258out:
1259 FREE(bap, M_DEVBUF);
1260 return (error);
1261}
1262
1263/*
1264 * Do both snap accounting and map accounting.
1265 */
1266static int
1267fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1268 struct vnode *vp;
1269 ufs2_daddr_t *oldblkp, *lastblkp;
1270 struct fs *fs;
1271 ufs_lbn_t lblkno;
1272 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1273{
1274 int error;
1275
1276 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1277 return (error);
1278 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1279}
1280
1281/*
1282 * Identify a set of blocks allocated in a snapshot inode.
1283 */
1284static int
1285snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1286 struct vnode *vp;
1287 ufs2_daddr_t *oldblkp, *lastblkp;
1288 struct fs *fs;
1289 ufs_lbn_t lblkno;
1290 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1291{
1292 struct inode *ip = VTOI(vp);
1293 ufs2_daddr_t blkno, *blkp;
1294 ufs_lbn_t lbn;
1295 struct buf *ibp;
1296 int error;
1297
1298 for ( ; oldblkp < lastblkp; oldblkp++) {
1299 blkno = *oldblkp;
1300 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1301 continue;
1302 lbn = fragstoblks(fs, blkno);
1303 if (lbn < NDADDR) {
1304 blkp = &ip->i_din2->di_db[lbn];
1305 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1306 } else {
1307 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1308 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1309 if (error)
1310 return (error);
1311 blkp = &((ufs2_daddr_t *)(ibp->b_data))
1312 [(lbn - NDADDR) % NINDIR(fs)];
1313 }
1314 /*
1315 * If we are expunging a snapshot vnode and we
1316 * find a block marked BLK_NOCOPY, then it is
1317 * one that has been allocated to this snapshot after
1318 * we took our current snapshot and can be ignored.
1319 */
1320 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1321 if (lbn >= NDADDR)
1322 brelse(ibp);
1323 } else {
1324 if (*blkp != 0)
1325 panic("snapacct_ufs2: bad block");
1326 *blkp = expungetype;
1327 if (lbn >= NDADDR)
1328 bdwrite(ibp);
1329 }
1330 }
1331 return (0);
1332}
1333
1334/*
1335 * Account for a set of blocks allocated in a snapshot inode.
1336 */
1337static int
1338mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1339 struct vnode *vp;
1340 ufs2_daddr_t *oldblkp, *lastblkp;
1341 struct fs *fs;
1342 ufs_lbn_t lblkno;
1343 int expungetype;
1344{
1345 ufs2_daddr_t blkno;
1346 struct inode *ip;
1347 ino_t inum;
1348 int acctit;
1349
1350 ip = VTOI(vp);
1351 inum = ip->i_number;
1352 if (lblkno == -1)
1353 acctit = 0;
1354 else
1355 acctit = 1;
1356 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1357 blkno = *oldblkp;
1358 if (blkno == 0 || blkno == BLK_NOCOPY)
1359 continue;
1360 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1361 *ip->i_snapblklist++ = lblkno;
1362 if (blkno == BLK_SNAP)
1363 blkno = blkstofrags(fs, lblkno);
1364 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
1365 }
1366 return (0);
1367}
1368
1369/*
1370 * Decrement extra reference on snapshot when last name is removed.
1371 * It will not be freed until the last open reference goes away.
1372 */
1373void
1374ffs_snapgone(ip)
1375 struct inode *ip;
1376{
1377 struct inode *xp;
1378 struct fs *fs;
1379 int snaploc;
1380 struct snapdata *sn;
1381
1382 /*
1383 * Find snapshot in incore list.
1384 */
1385 xp = NULL;
1386 sn = ip->i_devvp->v_rdev->si_snapdata;
1387 if (sn != NULL)
1388 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap)
1389 if (xp == ip)
1390 break;
1391 if (xp != NULL)
1392 vrele(ITOV(ip));
1393 else if (snapdebug)
1394 printf("ffs_snapgone: lost snapshot vnode %d\n",
1395 ip->i_number);
1396 /*
1397 * Delete snapshot inode from superblock. Keep list dense.
1398 */
1399 fs = ip->i_fs;
1400 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
1401 if (fs->fs_snapinum[snaploc] == ip->i_number)
1402 break;
1403 if (snaploc < FSMAXSNAP) {
1404 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
1405 if (fs->fs_snapinum[snaploc] == 0)
1406 break;
1407 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
1408 }
1409 fs->fs_snapinum[snaploc - 1] = 0;
1410 }
1411}
1412
1413/*
1414 * Prepare a snapshot file for being removed.
1415 */
1416void
1417ffs_snapremove(vp)
1418 struct vnode *vp;
1419{
1420 struct inode *ip;
1421 struct vnode *devvp;
1422 struct lock *lkp;
1423 struct buf *ibp;
1424 struct fs *fs;
1425 struct thread *td = curthread;
1426 ufs2_daddr_t numblks, blkno, dblk, *snapblklist;
1427 int error, loc, last;
1428 struct snapdata *sn;
1429
1430 ip = VTOI(vp);
1431 fs = ip->i_fs;
1432 devvp = ip->i_devvp;
1433 sn = devvp->v_rdev->si_snapdata;
1434 /*
1435 * If active, delete from incore list (this snapshot may
1436 * already have been in the process of being deleted, so
1437 * would not have been active).
1438 *
1439 * Clear copy-on-write flag if last snapshot.
1440 */
1441 if (ip->i_nextsnap.tqe_prev != 0) {
1442 VI_LOCK(devvp);
1443 lockmgr(&vp->v_lock, LK_INTERLOCK | LK_EXCLUSIVE,
1444 VI_MTX(devvp), td);
1445 VI_LOCK(devvp);
1446 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap);
1447 ip->i_nextsnap.tqe_prev = 0;
1448 lkp = vp->v_vnlock;
1449 vp->v_vnlock = &vp->v_lock;
1450 lockmgr(lkp, LK_RELEASE, NULL, td);
1451 if (TAILQ_FIRST(&sn->sn_head) != 0) {
1452 VI_UNLOCK(devvp);
1453 } else {
1454 devvp->v_rdev->si_copyonwrite = 0;
1455 snapblklist = sn->sn_blklist;
1456 sn->sn_blklist = 0;
1457 sn->sn_listsize = 0;
1458 devvp->v_rdev->si_snapdata = NULL;
1459 devvp->v_vflag &= ~VV_COPYONWRITE;
1460 lockmgr(lkp, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp), td);
1461 lockmgr(lkp, LK_RELEASE, NULL, td);
1462 lockdestroy(lkp);
1463 free(sn, M_UFSMNT);
1464 FREE(snapblklist, M_UFSMNT);
1465 }
1466 }
1467 /*
1468 * Clear all BLK_NOCOPY fields. Pass any block claims to other
1469 * snapshots that want them (see ffs_snapblkfree below).
1470 */
1471 for (blkno = 1; blkno < NDADDR; blkno++) {
1472 dblk = DIP(ip, i_db[blkno]);
1473 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1474 DIP_SET(ip, i_db[blkno], 0);
1475 else if ((dblk == blkstofrags(fs, blkno) &&
1476 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
1477 ip->i_number))) {
1478 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) -
1479 btodb(fs->fs_bsize));
1480 DIP_SET(ip, i_db[blkno], 0);
1481 }
1482 }
1483 numblks = howmany(ip->i_size, fs->fs_bsize);
1484 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1485 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
1486 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1487 if (error)
1488 continue;
1489 if (fs->fs_size - blkno > NINDIR(fs))
1490 last = NINDIR(fs);
1491 else
1492 last = fs->fs_size - blkno;
1493 for (loc = 0; loc < last; loc++) {
1494 if (ip->i_ump->um_fstype == UFS1) {
1495 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc];
1496 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1497 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1498 else if ((dblk == blkstofrags(fs, blkno) &&
1499 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1500 fs->fs_bsize, ip->i_number))) {
1501 ip->i_din1->di_blocks -=
1502 btodb(fs->fs_bsize);
1503 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1504 }
1505 continue;
1506 }
1507 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc];
1508 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1509 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1510 else if ((dblk == blkstofrags(fs, blkno) &&
1511 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1512 fs->fs_bsize, ip->i_number))) {
1513 ip->i_din2->di_blocks -= btodb(fs->fs_bsize);
1514 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1515 }
1516 }
1517 bawrite(ibp);
1518 }
1519 /*
1520 * Clear snapshot flag and drop reference.
1521 */
1522 ip->i_flags &= ~SF_SNAPSHOT;
1523 DIP_SET(ip, i_flags, ip->i_flags);
1524 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1525}
1526
1527/*
1528 * Notification that a block is being freed. Return zero if the free
1529 * should be allowed to proceed. Return non-zero if the snapshot file
1530 * wants to claim the block. The block will be claimed if it is an
1531 * uncopied part of one of the snapshots. It will be freed if it is
1532 * either a BLK_NOCOPY or has already been copied in all of the snapshots.
1533 * If a fragment is being freed, then all snapshots that care about
1534 * it must make a copy since a snapshot file can only claim full sized
1535 * blocks. Note that if more than one snapshot file maps the block,
1536 * we can pick one at random to claim it. Since none of the snapshots
1537 * can change, we are assurred that they will all see the same unmodified
1538 * image. When deleting a snapshot file (see ffs_snapremove above), we
1539 * must push any of these claimed blocks to one of the other snapshots
1540 * that maps it. These claimed blocks are easily identified as they will
1541 * have a block number equal to their logical block number within the
1542 * snapshot. A copied block can never have this property because they
1543 * must always have been allocated from a BLK_NOCOPY location.
1544 */
1545int
1546ffs_snapblkfree(fs, devvp, bno, size, inum)
1547 struct fs *fs;
1548 struct vnode *devvp;
1549 ufs2_daddr_t bno;
1550 long size;
1551 ino_t inum;
1552{
1553 struct buf *ibp, *cbp, *savedcbp = 0;
1554 struct thread *td = curthread;
1555 struct inode *ip;
1556 struct vnode *vp = NULL;
1557 ufs_lbn_t lbn;
1558 ufs2_daddr_t blkno;
1559 int indiroff = 0, snapshot_locked = 0, error = 0, claimedblk = 0;
1560 struct snapdata *sn;
1561
1562 lbn = fragstoblks(fs, bno);
1563retry:
1564 VI_LOCK(devvp);
1565 sn = devvp->v_rdev->si_snapdata;
|
1566 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
1567 vp = ITOV(ip);
1568 /*
1569 * Lookup block being written.
1570 */
1571 if (lbn < NDADDR) {
1572 blkno = DIP(ip, i_db[lbn]);
1573 } else {
1574 if (snapshot_locked == 0 &&
1575 lockmgr(vp->v_vnlock,
1576 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
1577 VI_MTX(devvp), td) != 0)
1578 goto retry;
1579 snapshot_locked = 1;
1580 td->td_pflags |= TDP_COWINPROGRESS;
1581 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1582 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1583 td->td_pflags &= ~TDP_COWINPROGRESS;
1584 if (error)
1585 break;
1586 indiroff = (lbn - NDADDR) % NINDIR(fs);
1587 if (ip->i_ump->um_fstype == UFS1)
1588 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
1589 else
1590 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
1591 }
1592 /*
1593 * Check to see if block needs to be copied.
1594 */
1595 if (blkno == 0) {
1596 /*
1597 * A block that we map is being freed. If it has not
1598 * been claimed yet, we will claim or copy it (below).
1599 */
1600 claimedblk = 1;
1601 } else if (blkno == BLK_SNAP) {
1602 /*
1603 * No previous snapshot claimed the block,
1604 * so it will be freed and become a BLK_NOCOPY
1605 * (don't care) for us.
1606 */
1607 if (claimedblk)
1608 panic("snapblkfree: inconsistent block type");
1609 if (snapshot_locked == 0 &&
1610 lockmgr(vp->v_vnlock,
1611 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
1612 VI_MTX(devvp), td) != 0) {
1613 if (lbn >= NDADDR)
1614 bqrelse(ibp);
1615 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td);
1616 goto retry;
1617 }
1618 snapshot_locked = 1;
1619 if (lbn < NDADDR) {
1620 DIP_SET(ip, i_db[lbn], BLK_NOCOPY);
1621 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1622 } else if (ip->i_ump->um_fstype == UFS1) {
1623 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
1624 BLK_NOCOPY;
1625 bdwrite(ibp);
1626 } else {
1627 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] =
1628 BLK_NOCOPY;
1629 bdwrite(ibp);
1630 }
1631 continue;
1632 } else /* BLK_NOCOPY or default */ {
1633 /*
1634 * If the snapshot has already copied the block
1635 * (default), or does not care about the block,
1636 * it is not needed.
1637 */
1638 if (lbn >= NDADDR)
1639 bqrelse(ibp);
1640 continue;
1641 }
1642 /*
1643 * If this is a full size block, we will just grab it
1644 * and assign it to the snapshot inode. Otherwise we
1645 * will proceed to copy it. See explanation for this
1646 * routine as to why only a single snapshot needs to
1647 * claim this block.
1648 */
1649 if (snapshot_locked == 0 &&
1650 lockmgr(vp->v_vnlock,
1651 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
1652 VI_MTX(devvp), td) != 0) {
1653 if (lbn >= NDADDR)
1654 bqrelse(ibp);
1655 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td);
1656 goto retry;
1657 }
1658 snapshot_locked = 1;
1659 if (size == fs->fs_bsize) {
1660#ifdef DEBUG
1661 if (snapdebug)
1662 printf("%s %d lbn %jd from inum %d\n",
1663 "Grabonremove: snapino", ip->i_number,
1664 (intmax_t)lbn, inum);
1665#endif
1666 if (lbn < NDADDR) {
1667 DIP_SET(ip, i_db[lbn], bno);
1668 } else if (ip->i_ump->um_fstype == UFS1) {
1669 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno;
1670 bdwrite(ibp);
1671 } else {
1672 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno;
1673 bdwrite(ibp);
1674 }
1675 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size));
1676 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1677 VOP_UNLOCK(vp, 0, td);
1678 return (1);
1679 }
1680 if (lbn >= NDADDR)
1681 bqrelse(ibp);
1682 /*
1683 * Allocate the block into which to do the copy. Note that this
1684 * allocation will never require any additional allocations for
1685 * the snapshot inode.
1686 */
1687 td->td_pflags |= TDP_COWINPROGRESS;
1688 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1689 fs->fs_bsize, KERNCRED, 0, &cbp);
1690 td->td_pflags &= ~TDP_COWINPROGRESS;
1691 if (error)
1692 break;
1693#ifdef DEBUG
1694 if (snapdebug)
1695 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n",
1696 "Copyonremove: snapino ", ip->i_number,
1697 (intmax_t)lbn, "for inum", inum, size,
1698 (intmax_t)cbp->b_blkno);
1699#endif
1700 /*
1701 * If we have already read the old block contents, then
1702 * simply copy them to the new block. Note that we need
1703 * to synchronously write snapshots that have not been
1704 * unlinked, and hence will be visible after a crash,
1705 * to ensure their integrity.
1706 */
1707 if (savedcbp != 0) {
1708 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
1709 bawrite(cbp);
1710 if (dopersistence && ip->i_effnlink > 0)
1711 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1712 continue;
1713 }
1714 /*
1715 * Otherwise, read the old block contents into the buffer.
1716 */
1717 if ((error = readblock(vp, cbp, lbn)) != 0) {
1718 bzero(cbp->b_data, fs->fs_bsize);
1719 bawrite(cbp);
1720 if (dopersistence && ip->i_effnlink > 0)
1721 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1722 break;
1723 }
1724 savedcbp = cbp;
1725 }
1726 /*
1727 * Note that we need to synchronously write snapshots that
1728 * have not been unlinked, and hence will be visible after
1729 * a crash, to ensure their integrity.
1730 */
1731 if (savedcbp) {
1732 vp = savedcbp->b_vp;
1733 bawrite(savedcbp);
1734 if (dopersistence && VTOI(vp)->i_effnlink > 0)
1735 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1736 }
1737 /*
1738 * If we have been unable to allocate a block in which to do
1739 * the copy, then return non-zero so that the fragment will
1740 * not be freed. Although space will be lost, the snapshot
1741 * will stay consistent.
1742 */
1743 if (snapshot_locked)
1744 VOP_UNLOCK(vp, 0, td);
1745 else
1746 VI_UNLOCK(devvp);
1747 return (error);
1748}
1749
1750/*
1751 * Associate snapshot files when mounting.
1752 */
1753void
1754ffs_snapshot_mount(mp)
1755 struct mount *mp;
1756{
1757 struct ufsmount *ump = VFSTOUFS(mp);
1758 struct vnode *devvp = ump->um_devvp;
1759 struct fs *fs = ump->um_fs;
1760 struct thread *td = curthread;
1761 struct snapdata *sn;
1762 struct vnode *vp;
1763 struct inode *ip;
1764 struct uio auio;
1765 struct iovec aiov;
1766 void *snapblklist;
1767 char *reason;
1768 daddr_t snaplistsize;
1769 int error, snaploc, loc;
1770
1771 /*
1772 * XXX The following needs to be set before UFS_TRUNCATE or
1773 * VOP_READ can be called.
1774 */
1775 mp->mnt_stat.f_iosize = fs->fs_bsize;
1776 /*
1777 * Process each snapshot listed in the superblock.
1778 */
1779 vp = NULL;
1780 sn = devvp->v_rdev->si_snapdata;
1781 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
1782 if (fs->fs_snapinum[snaploc] == 0)
1783 break;
1784 if ((error = VFS_VGET(mp, fs->fs_snapinum[snaploc],
1785 LK_EXCLUSIVE, &vp)) != 0){
1786 printf("ffs_snapshot_mount: vget failed %d\n", error);
1787 continue;
1788 }
1789 ip = VTOI(vp);
1790 if ((ip->i_flags & SF_SNAPSHOT) == 0 || ip->i_size ==
1791 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) {
1792 if ((ip->i_flags & SF_SNAPSHOT) == 0) {
1793 reason = "non-snapshot";
1794 } else {
1795 reason = "old format snapshot";
1796 (void)UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td);
1797 (void)VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1798 }
1799 printf("ffs_snapshot_mount: %s inode %d\n",
1800 reason, fs->fs_snapinum[snaploc]);
1801 vput(vp);
1802 vp = NULL;
1803 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
1804 if (fs->fs_snapinum[loc] == 0)
1805 break;
1806 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
1807 }
1808 fs->fs_snapinum[loc - 1] = 0;
1809 snaploc--;
1810 continue;
1811 }
1812 /*
1813 * If there already exist snapshots on this filesystem, grab a
1814 * reference to their shared lock. If this is the first snapshot
1815 * on this filesystem, we need to allocate a lock for the
1816 * snapshots to share. In either case, acquire the snapshot
1817 * lock and give up our original private lock.
1818 */
1819 VI_LOCK(devvp);
1820 if (sn != NULL) {
1821
1822 VI_UNLOCK(devvp);
1823 VI_LOCK(vp);
1824 vp->v_vnlock = &sn->sn_lock;
1825 } else {
1826 VI_UNLOCK(devvp);
1827 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO);
1828 TAILQ_INIT(&sn->sn_head);
1829 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT,
1830 LK_CANRECURSE | LK_NOPAUSE);
1831 VI_LOCK(vp);
1832 vp->v_vnlock = &sn->sn_lock;
1833 devvp->v_rdev->si_snapdata = sn;
1834 }
1835 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
1836 transferlockers(&vp->v_lock, vp->v_vnlock);
1837 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td);
1838 /*
1839 * Link it onto the active snapshot list.
1840 */
1841 VI_LOCK(devvp);
1842 if (ip->i_nextsnap.tqe_prev != 0)
1843 panic("ffs_snapshot_mount: %d already on list",
1844 ip->i_number);
1845 else
1846 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
1847 vp->v_vflag |= VV_SYSTEM;
1848 VI_UNLOCK(devvp);
1849 VOP_UNLOCK(vp, 0, td);
1850 }
1851 /*
1852 * No usable snapshots found.
1853 */
1854 if (vp == NULL)
1855 return;
1856 /*
1857 * Allocate the space for the block hints list. We always want to
1858 * use the list from the newest snapshot.
1859 */
1860 auio.uio_iov = &aiov;
1861 auio.uio_iovcnt = 1;
1862 aiov.iov_base = (void *)&snaplistsize;
1863 aiov.iov_len = sizeof(snaplistsize);
1864 auio.uio_resid = aiov.iov_len;
1865 auio.uio_offset =
1866 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag));
1867 auio.uio_segflg = UIO_SYSSPACE;
1868 auio.uio_rw = UIO_READ;
1869 auio.uio_td = td;
1870 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1871 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
1872 printf("ffs_snapshot_mount: read_1 failed %d\n", error);
1873 VOP_UNLOCK(vp, 0, td);
1874 return;
1875 }
1876 MALLOC(snapblklist, void *, snaplistsize * sizeof(daddr_t),
1877 M_UFSMNT, M_WAITOK);
1878 auio.uio_iovcnt = 1;
1879 aiov.iov_base = snapblklist;
1880 aiov.iov_len = snaplistsize * sizeof (daddr_t);
1881 auio.uio_resid = aiov.iov_len;
1882 auio.uio_offset -= sizeof(snaplistsize);
1883 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
1884 printf("ffs_snapshot_mount: read_2 failed %d\n", error);
1885 VOP_UNLOCK(vp, 0, td);
1886 FREE(snapblklist, M_UFSMNT);
1887 return;
1888 }
1889 VOP_UNLOCK(vp, 0, td);
1890 VI_LOCK(devvp);
1891 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount");
1892 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
1893 sn->sn_listsize = snaplistsize;
1894 sn->sn_blklist = (daddr_t *)snapblklist;
1895 devvp->v_vflag |= VV_COPYONWRITE;
1896 VI_UNLOCK(devvp);
1897}
1898
1899/*
1900 * Disassociate snapshot files when unmounting.
1901 */
1902void
1903ffs_snapshot_unmount(mp)
1904 struct mount *mp;
1905{
1906 struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
1907 struct snapdata *sn;
1908 struct inode *xp;
1909 struct vnode *vp;
1910
1911 sn = devvp->v_rdev->si_snapdata;
1912 VI_LOCK(devvp);
1913 while ((xp = TAILQ_FIRST(&sn->sn_head)) != 0) {
1914 vp = ITOV(xp);
1915 vp->v_vnlock = &vp->v_lock;
1916 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap);
1917 xp->i_nextsnap.tqe_prev = 0;
1918 if (xp->i_effnlink > 0) {
1919 VI_UNLOCK(devvp);
1920 vrele(vp);
1921 VI_LOCK(devvp);
1922 }
1923 }
1924 if (sn->sn_blklist != NULL) {
1925 FREE(sn->sn_blklist, M_UFSMNT);
1926 sn->sn_blklist = NULL;
1927 sn->sn_listsize = 0;
1928 }
1929 lockdestroy(&sn->sn_lock);
1930 free(sn, M_UFSMNT);
1931 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount");
1932 devvp->v_rdev->si_copyonwrite = 0;
1933 devvp->v_rdev->si_snapdata = NULL;
1934 devvp->v_vflag &= ~VV_COPYONWRITE;
1935 VI_UNLOCK(devvp);
1936}
1937
1938/*
1939 * Check for need to copy block that is about to be written,
1940 * copying the block if necessary.
1941 */
1942static int
1943ffs_copyonwrite(devvp, bp)
1944 struct vnode *devvp;
1945 struct buf *bp;
1946{
1947 struct snapdata *sn;
1948 struct buf *ibp, *cbp, *savedcbp = 0;
1949 struct thread *td = curthread;
1950 struct fs *fs;
1951 struct inode *ip;
1952 struct vnode *vp = 0;
1953 ufs2_daddr_t lbn, blkno, *snapblklist;
1954 int lower, upper, mid, indiroff, snapshot_locked = 0, error = 0;
1955
1956 if (td->td_pflags & TDP_COWINPROGRESS)
1957 panic("ffs_copyonwrite: recursive call");
1958 /*
1959 * First check to see if it is in the preallocated list.
1960 * By doing this check we avoid several potential deadlocks.
1961 */
1962 VI_LOCK(devvp);
1963 sn = devvp->v_rdev->si_snapdata;
1964 ip = TAILQ_FIRST(&sn->sn_head);
1965 fs = ip->i_fs;
1966 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
1967 snapblklist = sn->sn_blklist;
1968 upper = sn->sn_listsize - 1;
1969 lower = 1;
1970 while (lower <= upper) {
1971 mid = (lower + upper) / 2;
1972 if (snapblklist[mid] == lbn)
1973 break;
1974 if (snapblklist[mid] < lbn)
1975 lower = mid + 1;
1976 else
1977 upper = mid - 1;
1978 }
1979 if (lower <= upper) {
1980 VI_UNLOCK(devvp);
1981 return (0);
1982 }
1983 /*
1984 * Not in the precomputed list, so check the snapshots.
1985 */
1986retry:
1987 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
1988 vp = ITOV(ip);
1989 /*
1990 * We ensure that everything of our own that needs to be
1991 * copied will be done at the time that ffs_snapshot is
1992 * called. Thus we can skip the check here which can
1993 * deadlock in doing the lookup in UFS_BALLOC.
1994 */
1995 if (bp->b_vp == vp)
1996 continue;
1997 /*
1998 * Check to see if block needs to be copied. We do not have
1999 * to hold the snapshot lock while doing this lookup as it
2000 * will never require any additional allocations for the
2001 * snapshot inode.
2002 */
2003 if (lbn < NDADDR) {
2004 blkno = DIP(ip, i_db[lbn]);
2005 } else {
2006 if (snapshot_locked == 0 &&
2007 lockmgr(vp->v_vnlock,
2008 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2009 VI_MTX(devvp), td) != 0) {
2010 VI_LOCK(devvp);
2011 goto retry;
2012 }
2013 snapshot_locked = 1;
2014 td->td_pflags |= TDP_COWINPROGRESS;
2015 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2016 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
2017 td->td_pflags &= ~TDP_COWINPROGRESS;
2018 if (error)
2019 break;
2020 indiroff = (lbn - NDADDR) % NINDIR(fs);
2021 if (ip->i_ump->um_fstype == UFS1)
2022 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
2023 else
2024 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
2025 bqrelse(ibp);
2026 }
2027#ifdef DIAGNOSTIC
2028 if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
2029 panic("ffs_copyonwrite: bad copy block");
2030#endif
2031 if (blkno != 0)
2032 continue;
2033 /*
2034 * Allocate the block into which to do the copy. Since
2035 * multiple processes may all try to copy the same block,
2036 * we have to recheck our need to do a copy if we sleep
2037 * waiting for the lock.
2038 *
2039 * Because all snapshots on a filesystem share a single
2040 * lock, we ensure that we will never be in competition
2041 * with another process to allocate a block.
2042 */
2043 if (snapshot_locked == 0 &&
2044 lockmgr(vp->v_vnlock,
2045 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2046 VI_MTX(devvp), td) != 0) {
2047 VI_LOCK(devvp);
2048 goto retry;
2049 }
2050 snapshot_locked = 1;
2051 td->td_pflags |= TDP_COWINPROGRESS;
2052 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2053 fs->fs_bsize, KERNCRED, 0, &cbp);
2054 td->td_pflags &= ~TDP_COWINPROGRESS;
2055 if (error)
2056 break;
2057#ifdef DEBUG
2058 if (snapdebug) {
2059 printf("Copyonwrite: snapino %d lbn %jd for ",
2060 ip->i_number, (intmax_t)lbn);
2061 if (bp->b_vp == devvp)
2062 printf("fs metadata");
2063 else
2064 printf("inum %d", VTOI(bp->b_vp)->i_number);
2065 printf(" lblkno %jd to blkno %jd\n",
2066 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno);
2067 }
2068#endif
2069 /*
2070 * If we have already read the old block contents, then
2071 * simply copy them to the new block. Note that we need
2072 * to synchronously write snapshots that have not been
2073 * unlinked, and hence will be visible after a crash,
2074 * to ensure their integrity.
2075 */
2076 if (savedcbp != 0) {
2077 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
2078 bawrite(cbp);
2079 if (dopersistence && ip->i_effnlink > 0)
2080 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2081 continue;
2082 }
2083 /*
2084 * Otherwise, read the old block contents into the buffer.
2085 */
2086 if ((error = readblock(vp, cbp, lbn)) != 0) {
2087 bzero(cbp->b_data, fs->fs_bsize);
2088 bawrite(cbp);
2089 if (dopersistence && ip->i_effnlink > 0)
2090 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2091 break;
2092 }
2093 savedcbp = cbp;
2094 }
2095 /*
2096 * Note that we need to synchronously write snapshots that
2097 * have not been unlinked, and hence will be visible after
2098 * a crash, to ensure their integrity.
2099 */
2100 if (savedcbp) {
2101 vp = savedcbp->b_vp;
2102 bawrite(savedcbp);
2103 if (dopersistence && VTOI(vp)->i_effnlink > 0)
2104 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2105 }
2106 if (snapshot_locked)
2107 VOP_UNLOCK(vp, 0, td);
2108 else
2109 VI_UNLOCK(devvp);
2110 return (error);
2111}
2112
2113/*
2114 * Read the specified block into the given buffer.
2115 * Much of this boiler-plate comes from bwrite().
2116 */
2117static int
2118readblock(vp, bp, lbn)
2119 struct vnode *vp;
2120 struct buf *bp;
2121 ufs2_daddr_t lbn;
2122{
2123 struct uio auio;
2124 struct iovec aiov;
2125 struct thread *td = curthread;
2126 struct inode *ip = VTOI(vp);
2127
2128 aiov.iov_base = bp->b_data;
2129 aiov.iov_len = bp->b_bcount;
2130 auio.uio_iov = &aiov;
2131 auio.uio_iovcnt = 1;
2132 auio.uio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn)));
2133 auio.uio_resid = bp->b_bcount;
2134 auio.uio_rw = UIO_READ;
2135 auio.uio_segflg = UIO_SYSSPACE;
2136 auio.uio_td = td;
2137 return (physio(ip->i_devvp->v_rdev, &auio, 0));
2138}
| 1570 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
1571 vp = ITOV(ip);
1572 /*
1573 * Lookup block being written.
1574 */
1575 if (lbn < NDADDR) {
1576 blkno = DIP(ip, i_db[lbn]);
1577 } else {
1578 if (snapshot_locked == 0 &&
1579 lockmgr(vp->v_vnlock,
1580 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
1581 VI_MTX(devvp), td) != 0)
1582 goto retry;
1583 snapshot_locked = 1;
1584 td->td_pflags |= TDP_COWINPROGRESS;
1585 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1586 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1587 td->td_pflags &= ~TDP_COWINPROGRESS;
1588 if (error)
1589 break;
1590 indiroff = (lbn - NDADDR) % NINDIR(fs);
1591 if (ip->i_ump->um_fstype == UFS1)
1592 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
1593 else
1594 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
1595 }
1596 /*
1597 * Check to see if block needs to be copied.
1598 */
1599 if (blkno == 0) {
1600 /*
1601 * A block that we map is being freed. If it has not
1602 * been claimed yet, we will claim or copy it (below).
1603 */
1604 claimedblk = 1;
1605 } else if (blkno == BLK_SNAP) {
1606 /*
1607 * No previous snapshot claimed the block,
1608 * so it will be freed and become a BLK_NOCOPY
1609 * (don't care) for us.
1610 */
1611 if (claimedblk)
1612 panic("snapblkfree: inconsistent block type");
1613 if (snapshot_locked == 0 &&
1614 lockmgr(vp->v_vnlock,
1615 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
1616 VI_MTX(devvp), td) != 0) {
1617 if (lbn >= NDADDR)
1618 bqrelse(ibp);
1619 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td);
1620 goto retry;
1621 }
1622 snapshot_locked = 1;
1623 if (lbn < NDADDR) {
1624 DIP_SET(ip, i_db[lbn], BLK_NOCOPY);
1625 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1626 } else if (ip->i_ump->um_fstype == UFS1) {
1627 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
1628 BLK_NOCOPY;
1629 bdwrite(ibp);
1630 } else {
1631 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] =
1632 BLK_NOCOPY;
1633 bdwrite(ibp);
1634 }
1635 continue;
1636 } else /* BLK_NOCOPY or default */ {
1637 /*
1638 * If the snapshot has already copied the block
1639 * (default), or does not care about the block,
1640 * it is not needed.
1641 */
1642 if (lbn >= NDADDR)
1643 bqrelse(ibp);
1644 continue;
1645 }
1646 /*
1647 * If this is a full size block, we will just grab it
1648 * and assign it to the snapshot inode. Otherwise we
1649 * will proceed to copy it. See explanation for this
1650 * routine as to why only a single snapshot needs to
1651 * claim this block.
1652 */
1653 if (snapshot_locked == 0 &&
1654 lockmgr(vp->v_vnlock,
1655 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
1656 VI_MTX(devvp), td) != 0) {
1657 if (lbn >= NDADDR)
1658 bqrelse(ibp);
1659 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td);
1660 goto retry;
1661 }
1662 snapshot_locked = 1;
1663 if (size == fs->fs_bsize) {
1664#ifdef DEBUG
1665 if (snapdebug)
1666 printf("%s %d lbn %jd from inum %d\n",
1667 "Grabonremove: snapino", ip->i_number,
1668 (intmax_t)lbn, inum);
1669#endif
1670 if (lbn < NDADDR) {
1671 DIP_SET(ip, i_db[lbn], bno);
1672 } else if (ip->i_ump->um_fstype == UFS1) {
1673 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno;
1674 bdwrite(ibp);
1675 } else {
1676 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno;
1677 bdwrite(ibp);
1678 }
1679 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size));
1680 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1681 VOP_UNLOCK(vp, 0, td);
1682 return (1);
1683 }
1684 if (lbn >= NDADDR)
1685 bqrelse(ibp);
1686 /*
1687 * Allocate the block into which to do the copy. Note that this
1688 * allocation will never require any additional allocations for
1689 * the snapshot inode.
1690 */
1691 td->td_pflags |= TDP_COWINPROGRESS;
1692 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1693 fs->fs_bsize, KERNCRED, 0, &cbp);
1694 td->td_pflags &= ~TDP_COWINPROGRESS;
1695 if (error)
1696 break;
1697#ifdef DEBUG
1698 if (snapdebug)
1699 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n",
1700 "Copyonremove: snapino ", ip->i_number,
1701 (intmax_t)lbn, "for inum", inum, size,
1702 (intmax_t)cbp->b_blkno);
1703#endif
1704 /*
1705 * If we have already read the old block contents, then
1706 * simply copy them to the new block. Note that we need
1707 * to synchronously write snapshots that have not been
1708 * unlinked, and hence will be visible after a crash,
1709 * to ensure their integrity.
1710 */
1711 if (savedcbp != 0) {
1712 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
1713 bawrite(cbp);
1714 if (dopersistence && ip->i_effnlink > 0)
1715 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1716 continue;
1717 }
1718 /*
1719 * Otherwise, read the old block contents into the buffer.
1720 */
1721 if ((error = readblock(vp, cbp, lbn)) != 0) {
1722 bzero(cbp->b_data, fs->fs_bsize);
1723 bawrite(cbp);
1724 if (dopersistence && ip->i_effnlink > 0)
1725 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1726 break;
1727 }
1728 savedcbp = cbp;
1729 }
1730 /*
1731 * Note that we need to synchronously write snapshots that
1732 * have not been unlinked, and hence will be visible after
1733 * a crash, to ensure their integrity.
1734 */
1735 if (savedcbp) {
1736 vp = savedcbp->b_vp;
1737 bawrite(savedcbp);
1738 if (dopersistence && VTOI(vp)->i_effnlink > 0)
1739 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1740 }
1741 /*
1742 * If we have been unable to allocate a block in which to do
1743 * the copy, then return non-zero so that the fragment will
1744 * not be freed. Although space will be lost, the snapshot
1745 * will stay consistent.
1746 */
1747 if (snapshot_locked)
1748 VOP_UNLOCK(vp, 0, td);
1749 else
1750 VI_UNLOCK(devvp);
1751 return (error);
1752}
1753
1754/*
1755 * Associate snapshot files when mounting.
1756 */
1757void
1758ffs_snapshot_mount(mp)
1759 struct mount *mp;
1760{
1761 struct ufsmount *ump = VFSTOUFS(mp);
1762 struct vnode *devvp = ump->um_devvp;
1763 struct fs *fs = ump->um_fs;
1764 struct thread *td = curthread;
1765 struct snapdata *sn;
1766 struct vnode *vp;
1767 struct inode *ip;
1768 struct uio auio;
1769 struct iovec aiov;
1770 void *snapblklist;
1771 char *reason;
1772 daddr_t snaplistsize;
1773 int error, snaploc, loc;
1774
1775 /*
1776 * XXX The following needs to be set before UFS_TRUNCATE or
1777 * VOP_READ can be called.
1778 */
1779 mp->mnt_stat.f_iosize = fs->fs_bsize;
1780 /*
1781 * Process each snapshot listed in the superblock.
1782 */
1783 vp = NULL;
1784 sn = devvp->v_rdev->si_snapdata;
1785 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
1786 if (fs->fs_snapinum[snaploc] == 0)
1787 break;
1788 if ((error = VFS_VGET(mp, fs->fs_snapinum[snaploc],
1789 LK_EXCLUSIVE, &vp)) != 0){
1790 printf("ffs_snapshot_mount: vget failed %d\n", error);
1791 continue;
1792 }
1793 ip = VTOI(vp);
1794 if ((ip->i_flags & SF_SNAPSHOT) == 0 || ip->i_size ==
1795 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) {
1796 if ((ip->i_flags & SF_SNAPSHOT) == 0) {
1797 reason = "non-snapshot";
1798 } else {
1799 reason = "old format snapshot";
1800 (void)UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td);
1801 (void)VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1802 }
1803 printf("ffs_snapshot_mount: %s inode %d\n",
1804 reason, fs->fs_snapinum[snaploc]);
1805 vput(vp);
1806 vp = NULL;
1807 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
1808 if (fs->fs_snapinum[loc] == 0)
1809 break;
1810 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
1811 }
1812 fs->fs_snapinum[loc - 1] = 0;
1813 snaploc--;
1814 continue;
1815 }
1816 /*
1817 * If there already exist snapshots on this filesystem, grab a
1818 * reference to their shared lock. If this is the first snapshot
1819 * on this filesystem, we need to allocate a lock for the
1820 * snapshots to share. In either case, acquire the snapshot
1821 * lock and give up our original private lock.
1822 */
1823 VI_LOCK(devvp);
1824 if (sn != NULL) {
1825
1826 VI_UNLOCK(devvp);
1827 VI_LOCK(vp);
1828 vp->v_vnlock = &sn->sn_lock;
1829 } else {
1830 VI_UNLOCK(devvp);
1831 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO);
1832 TAILQ_INIT(&sn->sn_head);
1833 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT,
1834 LK_CANRECURSE | LK_NOPAUSE);
1835 VI_LOCK(vp);
1836 vp->v_vnlock = &sn->sn_lock;
1837 devvp->v_rdev->si_snapdata = sn;
1838 }
1839 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
1840 transferlockers(&vp->v_lock, vp->v_vnlock);
1841 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td);
1842 /*
1843 * Link it onto the active snapshot list.
1844 */
1845 VI_LOCK(devvp);
1846 if (ip->i_nextsnap.tqe_prev != 0)
1847 panic("ffs_snapshot_mount: %d already on list",
1848 ip->i_number);
1849 else
1850 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
1851 vp->v_vflag |= VV_SYSTEM;
1852 VI_UNLOCK(devvp);
1853 VOP_UNLOCK(vp, 0, td);
1854 }
1855 /*
1856 * No usable snapshots found.
1857 */
1858 if (vp == NULL)
1859 return;
1860 /*
1861 * Allocate the space for the block hints list. We always want to
1862 * use the list from the newest snapshot.
1863 */
1864 auio.uio_iov = &aiov;
1865 auio.uio_iovcnt = 1;
1866 aiov.iov_base = (void *)&snaplistsize;
1867 aiov.iov_len = sizeof(snaplistsize);
1868 auio.uio_resid = aiov.iov_len;
1869 auio.uio_offset =
1870 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag));
1871 auio.uio_segflg = UIO_SYSSPACE;
1872 auio.uio_rw = UIO_READ;
1873 auio.uio_td = td;
1874 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1875 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
1876 printf("ffs_snapshot_mount: read_1 failed %d\n", error);
1877 VOP_UNLOCK(vp, 0, td);
1878 return;
1879 }
1880 MALLOC(snapblklist, void *, snaplistsize * sizeof(daddr_t),
1881 M_UFSMNT, M_WAITOK);
1882 auio.uio_iovcnt = 1;
1883 aiov.iov_base = snapblklist;
1884 aiov.iov_len = snaplistsize * sizeof (daddr_t);
1885 auio.uio_resid = aiov.iov_len;
1886 auio.uio_offset -= sizeof(snaplistsize);
1887 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
1888 printf("ffs_snapshot_mount: read_2 failed %d\n", error);
1889 VOP_UNLOCK(vp, 0, td);
1890 FREE(snapblklist, M_UFSMNT);
1891 return;
1892 }
1893 VOP_UNLOCK(vp, 0, td);
1894 VI_LOCK(devvp);
1895 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount");
1896 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
1897 sn->sn_listsize = snaplistsize;
1898 sn->sn_blklist = (daddr_t *)snapblklist;
1899 devvp->v_vflag |= VV_COPYONWRITE;
1900 VI_UNLOCK(devvp);
1901}
1902
1903/*
1904 * Disassociate snapshot files when unmounting.
1905 */
1906void
1907ffs_snapshot_unmount(mp)
1908 struct mount *mp;
1909{
1910 struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
1911 struct snapdata *sn;
1912 struct inode *xp;
1913 struct vnode *vp;
1914
1915 sn = devvp->v_rdev->si_snapdata;
1916 VI_LOCK(devvp);
1917 while ((xp = TAILQ_FIRST(&sn->sn_head)) != 0) {
1918 vp = ITOV(xp);
1919 vp->v_vnlock = &vp->v_lock;
1920 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap);
1921 xp->i_nextsnap.tqe_prev = 0;
1922 if (xp->i_effnlink > 0) {
1923 VI_UNLOCK(devvp);
1924 vrele(vp);
1925 VI_LOCK(devvp);
1926 }
1927 }
1928 if (sn->sn_blklist != NULL) {
1929 FREE(sn->sn_blklist, M_UFSMNT);
1930 sn->sn_blklist = NULL;
1931 sn->sn_listsize = 0;
1932 }
1933 lockdestroy(&sn->sn_lock);
1934 free(sn, M_UFSMNT);
1935 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount");
1936 devvp->v_rdev->si_copyonwrite = 0;
1937 devvp->v_rdev->si_snapdata = NULL;
1938 devvp->v_vflag &= ~VV_COPYONWRITE;
1939 VI_UNLOCK(devvp);
1940}
1941
1942/*
1943 * Check for need to copy block that is about to be written,
1944 * copying the block if necessary.
1945 */
1946static int
1947ffs_copyonwrite(devvp, bp)
1948 struct vnode *devvp;
1949 struct buf *bp;
1950{
1951 struct snapdata *sn;
1952 struct buf *ibp, *cbp, *savedcbp = 0;
1953 struct thread *td = curthread;
1954 struct fs *fs;
1955 struct inode *ip;
1956 struct vnode *vp = 0;
1957 ufs2_daddr_t lbn, blkno, *snapblklist;
1958 int lower, upper, mid, indiroff, snapshot_locked = 0, error = 0;
1959
1960 if (td->td_pflags & TDP_COWINPROGRESS)
1961 panic("ffs_copyonwrite: recursive call");
1962 /*
1963 * First check to see if it is in the preallocated list.
1964 * By doing this check we avoid several potential deadlocks.
1965 */
1966 VI_LOCK(devvp);
1967 sn = devvp->v_rdev->si_snapdata;
1968 ip = TAILQ_FIRST(&sn->sn_head);
1969 fs = ip->i_fs;
1970 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
1971 snapblklist = sn->sn_blklist;
1972 upper = sn->sn_listsize - 1;
1973 lower = 1;
1974 while (lower <= upper) {
1975 mid = (lower + upper) / 2;
1976 if (snapblklist[mid] == lbn)
1977 break;
1978 if (snapblklist[mid] < lbn)
1979 lower = mid + 1;
1980 else
1981 upper = mid - 1;
1982 }
1983 if (lower <= upper) {
1984 VI_UNLOCK(devvp);
1985 return (0);
1986 }
1987 /*
1988 * Not in the precomputed list, so check the snapshots.
1989 */
1990retry:
1991 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
1992 vp = ITOV(ip);
1993 /*
1994 * We ensure that everything of our own that needs to be
1995 * copied will be done at the time that ffs_snapshot is
1996 * called. Thus we can skip the check here which can
1997 * deadlock in doing the lookup in UFS_BALLOC.
1998 */
1999 if (bp->b_vp == vp)
2000 continue;
2001 /*
2002 * Check to see if block needs to be copied. We do not have
2003 * to hold the snapshot lock while doing this lookup as it
2004 * will never require any additional allocations for the
2005 * snapshot inode.
2006 */
2007 if (lbn < NDADDR) {
2008 blkno = DIP(ip, i_db[lbn]);
2009 } else {
2010 if (snapshot_locked == 0 &&
2011 lockmgr(vp->v_vnlock,
2012 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2013 VI_MTX(devvp), td) != 0) {
2014 VI_LOCK(devvp);
2015 goto retry;
2016 }
2017 snapshot_locked = 1;
2018 td->td_pflags |= TDP_COWINPROGRESS;
2019 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2020 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
2021 td->td_pflags &= ~TDP_COWINPROGRESS;
2022 if (error)
2023 break;
2024 indiroff = (lbn - NDADDR) % NINDIR(fs);
2025 if (ip->i_ump->um_fstype == UFS1)
2026 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
2027 else
2028 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
2029 bqrelse(ibp);
2030 }
2031#ifdef DIAGNOSTIC
2032 if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
2033 panic("ffs_copyonwrite: bad copy block");
2034#endif
2035 if (blkno != 0)
2036 continue;
2037 /*
2038 * Allocate the block into which to do the copy. Since
2039 * multiple processes may all try to copy the same block,
2040 * we have to recheck our need to do a copy if we sleep
2041 * waiting for the lock.
2042 *
2043 * Because all snapshots on a filesystem share a single
2044 * lock, we ensure that we will never be in competition
2045 * with another process to allocate a block.
2046 */
2047 if (snapshot_locked == 0 &&
2048 lockmgr(vp->v_vnlock,
2049 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2050 VI_MTX(devvp), td) != 0) {
2051 VI_LOCK(devvp);
2052 goto retry;
2053 }
2054 snapshot_locked = 1;
2055 td->td_pflags |= TDP_COWINPROGRESS;
2056 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2057 fs->fs_bsize, KERNCRED, 0, &cbp);
2058 td->td_pflags &= ~TDP_COWINPROGRESS;
2059 if (error)
2060 break;
2061#ifdef DEBUG
2062 if (snapdebug) {
2063 printf("Copyonwrite: snapino %d lbn %jd for ",
2064 ip->i_number, (intmax_t)lbn);
2065 if (bp->b_vp == devvp)
2066 printf("fs metadata");
2067 else
2068 printf("inum %d", VTOI(bp->b_vp)->i_number);
2069 printf(" lblkno %jd to blkno %jd\n",
2070 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno);
2071 }
2072#endif
2073 /*
2074 * If we have already read the old block contents, then
2075 * simply copy them to the new block. Note that we need
2076 * to synchronously write snapshots that have not been
2077 * unlinked, and hence will be visible after a crash,
2078 * to ensure their integrity.
2079 */
2080 if (savedcbp != 0) {
2081 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
2082 bawrite(cbp);
2083 if (dopersistence && ip->i_effnlink > 0)
2084 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2085 continue;
2086 }
2087 /*
2088 * Otherwise, read the old block contents into the buffer.
2089 */
2090 if ((error = readblock(vp, cbp, lbn)) != 0) {
2091 bzero(cbp->b_data, fs->fs_bsize);
2092 bawrite(cbp);
2093 if (dopersistence && ip->i_effnlink > 0)
2094 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2095 break;
2096 }
2097 savedcbp = cbp;
2098 }
2099 /*
2100 * Note that we need to synchronously write snapshots that
2101 * have not been unlinked, and hence will be visible after
2102 * a crash, to ensure their integrity.
2103 */
2104 if (savedcbp) {
2105 vp = savedcbp->b_vp;
2106 bawrite(savedcbp);
2107 if (dopersistence && VTOI(vp)->i_effnlink > 0)
2108 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2109 }
2110 if (snapshot_locked)
2111 VOP_UNLOCK(vp, 0, td);
2112 else
2113 VI_UNLOCK(devvp);
2114 return (error);
2115}
2116
2117/*
2118 * Read the specified block into the given buffer.
2119 * Much of this boiler-plate comes from bwrite().
2120 */
2121static int
2122readblock(vp, bp, lbn)
2123 struct vnode *vp;
2124 struct buf *bp;
2125 ufs2_daddr_t lbn;
2126{
2127 struct uio auio;
2128 struct iovec aiov;
2129 struct thread *td = curthread;
2130 struct inode *ip = VTOI(vp);
2131
2132 aiov.iov_base = bp->b_data;
2133 aiov.iov_len = bp->b_bcount;
2134 auio.uio_iov = &aiov;
2135 auio.uio_iovcnt = 1;
2136 auio.uio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn)));
2137 auio.uio_resid = bp->b_bcount;
2138 auio.uio_rw = UIO_READ;
2139 auio.uio_segflg = UIO_SYSSPACE;
2140 auio.uio_td = td;
2141 return (physio(ip->i_devvp->v_rdev, &auio, 0));
2142}
|