ffs_snapshot.c revision 107915
1/* 2 * Copyright 2000 Marshall Kirk McKusick. All Rights Reserved. 3 * 4 * Further information about snapshots can be obtained from: 5 * 6 * Marshall Kirk McKusick http://www.mckusick.com/softdep/ 7 * 1614 Oxford Street mckusick@mckusick.com 8 * Berkeley, CA 94709-1608 +1-510-843-9542 9 * USA 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY 22 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 23 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 24 * DISCLAIMED. IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR 25 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)ffs_snapshot.c 8.11 (McKusick) 7/23/00 34 * $FreeBSD: head/sys/ufs/ffs/ffs_snapshot.c 107915 2002-12-15 19:25:59Z mckusick $ 35 */ 36 37#include <sys/param.h> 38#include <sys/stdint.h> 39#include <sys/kernel.h> 40#include <sys/systm.h> 41#include <sys/conf.h> 42#include <sys/bio.h> 43#include <sys/buf.h> 44#include <sys/proc.h> 45#include <sys/namei.h> 46#include <sys/stat.h> 47#include <sys/malloc.h> 48#include <sys/mount.h> 49#include <sys/resource.h> 50#include <sys/resourcevar.h> 51#include <sys/vnode.h> 52 53#include <ufs/ufs/extattr.h> 54#include <ufs/ufs/quota.h> 55#include <ufs/ufs/ufsmount.h> 56#include <ufs/ufs/inode.h> 57#include <ufs/ufs/ufs_extern.h> 58 59#include <ufs/ffs/fs.h> 60#include <ufs/ffs/ffs_extern.h> 61 62#define KERNCRED thread0.td_ucred 63#define DEBUG 1 64 65static int cgaccount(int, struct vnode *, struct buf *, int); 66static int expunge_ufs1(struct vnode *, struct inode *, struct fs *, 67 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, 68 ufs_lbn_t, int), int); 69static int indiracct_ufs1(struct vnode *, struct vnode *, int, 70 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, 71 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, 72 ufs_lbn_t, int), int); 73static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 74 struct fs *, ufs_lbn_t, int); 75static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 76 struct fs *, ufs_lbn_t, int); 77static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 78 struct fs *, ufs_lbn_t, int); 79static int expunge_ufs2(struct vnode *, struct inode *, struct fs *, 80 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, 81 ufs_lbn_t, int), int); 82static int indiracct_ufs2(struct vnode *, struct vnode *, int, 83 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, 84 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, 85 ufs_lbn_t, int), int); 86static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 87 struct fs *, ufs_lbn_t, int); 88static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 89 struct fs *, ufs_lbn_t, int); 90static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 91 struct fs *, ufs_lbn_t, int); 92static int ffs_copyonwrite(struct vnode *, struct buf *); 93static int readblock(struct buf *, ufs2_daddr_t); 94 95/* 96 * To ensure the consistency of snapshots across crashes, we must 97 * synchronously write out copied blocks before allowing the 98 * originals to be modified. Because of the rather severe speed 99 * penalty that this imposes, the following flag allows this 100 * crash persistence to be disabled. 101 */ 102int dopersistence = 0; 103 104#ifdef DEBUG 105#include <sys/sysctl.h> 106SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, ""); 107int snapdebug = 0; 108SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, ""); 109int collectsnapstats = 0; 110SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats, 111 0, ""); 112#endif /* DEBUG */ 113 114/* 115 * Create a snapshot file and initialize it for the filesystem. 116 */ 117int 118ffs_snapshot(mp, snapfile) 119 struct mount *mp; 120 char *snapfile; 121{ 122 ufs2_daddr_t numblks, blkno; 123 int error, cg, snaploc; 124 int i, size, len, loc; 125 int flag = mp->mnt_flag; 126 struct timespec starttime = {0, 0}, endtime; 127 char saved_nice = 0; 128 long redo = 0, snaplistsize; 129 int32_t *lp; 130 void *space; 131 daddr_t *snapblklist; 132 struct fs *copy_fs = NULL, *fs = VFSTOUFS(mp)->um_fs; 133 struct snaphead *snaphead; 134 struct thread *td = curthread; 135 struct inode *ip, *xp; 136 struct buf *bp, *nbp, *ibp, *sbp = NULL; 137 struct nameidata nd; 138 struct mount *wrtmp; 139 struct vattr vat; 140 struct vnode *vp, *xvp, *nvp, *devvp; 141 struct uio auio; 142 struct iovec aiov; 143 144 /* 145 * Need to serialize access to snapshot code per filesystem. 146 */ 147 /* 148 * Assign a snapshot slot in the superblock. 149 */ 150 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 151 if (fs->fs_snapinum[snaploc] == 0) 152 break; 153 if (snaploc == FSMAXSNAP) 154 return (ENOSPC); 155 /* 156 * Create the snapshot file. 157 */ 158restart: 159 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_USERSPACE, snapfile, td); 160 if ((error = namei(&nd)) != 0) 161 return (error); 162 if (nd.ni_vp != NULL) { 163 vput(nd.ni_vp); 164 error = EEXIST; 165 } 166 if (nd.ni_dvp->v_mount != mp) 167 error = EXDEV; 168 if (error) { 169 NDFREE(&nd, NDF_ONLY_PNBUF); 170 if (nd.ni_dvp == nd.ni_vp) 171 vrele(nd.ni_dvp); 172 else 173 vput(nd.ni_dvp); 174 return (error); 175 } 176 VATTR_NULL(&vat); 177 vat.va_type = VREG; 178 vat.va_mode = S_IRUSR; 179 vat.va_vaflags |= VA_EXCLUSIVE; 180 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp)) 181 wrtmp = NULL; 182 if (wrtmp != mp) 183 panic("ffs_snapshot: mount mismatch"); 184 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) { 185 NDFREE(&nd, NDF_ONLY_PNBUF); 186 vput(nd.ni_dvp); 187 if ((error = vn_start_write(NULL, &wrtmp, 188 V_XSLEEP | PCATCH)) != 0) 189 return (error); 190 goto restart; 191 } 192 VOP_LEASE(nd.ni_dvp, td, KERNCRED, LEASE_WRITE); 193 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat); 194 vput(nd.ni_dvp); 195 if (error) { 196 NDFREE(&nd, NDF_ONLY_PNBUF); 197 vn_finished_write(wrtmp); 198 return (error); 199 } 200 vp = nd.ni_vp; 201 ip = VTOI(vp); 202 devvp = ip->i_devvp; 203 /* 204 * Allocate and copy the last block contents so as to be able 205 * to set size to that of the filesystem. 206 */ 207 numblks = howmany(fs->fs_size, fs->fs_frag); 208 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)), 209 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 210 if (error) 211 goto out; 212 ip->i_size = lblktosize(fs, (off_t)numblks); 213 DIP(ip, i_size) = ip->i_size; 214 ip->i_flag |= IN_CHANGE | IN_UPDATE; 215 if ((error = readblock(bp, numblks - 1)) != 0) 216 goto out; 217 bawrite(bp); 218 /* 219 * Preallocate critical data structures so that we can copy 220 * them in without further allocation after we suspend all 221 * operations on the filesystem. We would like to just release 222 * the allocated buffers without writing them since they will 223 * be filled in below once we are ready to go, but this upsets 224 * the soft update code, so we go ahead and write the new buffers. 225 * 226 * Allocate all indirect blocks and mark all of them as not 227 * needing to be copied. 228 */ 229 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 230 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 231 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp); 232 if (error) 233 goto out; 234 bawrite(ibp); 235 } 236 /* 237 * Allocate copies for the superblock and its summary information. 238 */ 239 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED, 240 0, &nbp); 241 if (error) 242 goto out; 243 bawrite(nbp); 244 blkno = fragstoblks(fs, fs->fs_csaddr); 245 len = howmany(fs->fs_cssize, fs->fs_bsize); 246 for (loc = 0; loc < len; loc++) { 247 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)), 248 fs->fs_bsize, KERNCRED, 0, &nbp); 249 if (error) 250 goto out; 251 bawrite(nbp); 252 } 253 /* 254 * Allocate all cylinder group blocks. 255 */ 256 for (cg = 0; cg < fs->fs_ncg; cg++) { 257 error = UFS_BALLOC(vp, (off_t)(cgtod(fs, cg)) << fs->fs_fshift, 258 fs->fs_bsize, KERNCRED, 0, &nbp); 259 if (error) 260 goto out; 261 bawrite(nbp); 262 } 263 /* 264 * Copy all the cylinder group maps. Although the 265 * filesystem is still active, we hope that only a few 266 * cylinder groups will change between now and when we 267 * suspend operations. Thus, we will be able to quickly 268 * touch up the few cylinder groups that changed during 269 * the suspension period. 270 */ 271 len = howmany(fs->fs_ncg, NBBY); 272 MALLOC(fs->fs_active, int *, len, M_DEVBUF, M_WAITOK); 273 bzero(fs->fs_active, len); 274 for (cg = 0; cg < fs->fs_ncg; cg++) { 275 error = UFS_BALLOC(vp, (off_t)(cgtod(fs, cg)) << fs->fs_fshift, 276 fs->fs_bsize, KERNCRED, 0, &nbp); 277 if (error) 278 goto out; 279 error = cgaccount(cg, vp, nbp, 1); 280 bawrite(nbp); 281 if (error) 282 goto out; 283 } 284 /* 285 * Change inode to snapshot type file. 286 */ 287 ip->i_flags |= SF_SNAPSHOT; 288 DIP(ip, i_flags) = ip->i_flags; 289 ip->i_flag |= IN_CHANGE | IN_UPDATE; 290 /* 291 * Ensure that the snapshot is completely on disk. 292 * Since we have marked it as a snapshot it is safe to 293 * unlock it as no process will be allowed to write to it. 294 */ 295 if ((error = VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td)) != 0) 296 goto out; 297 VOP_UNLOCK(vp, 0, td); 298 /* 299 * All allocations are done, so we can now snapshot the system. 300 * 301 * Recind nice scheduling while running with the filesystem suspended. 302 */ 303 if (td->td_ksegrp->kg_nice > 0) { 304 saved_nice = td->td_ksegrp->kg_nice; 305 td->td_ksegrp->kg_nice = 0; 306 } 307 /* 308 * Suspend operation on filesystem. 309 */ 310 for (;;) { 311 vn_finished_write(wrtmp); 312 if ((error = vfs_write_suspend(vp->v_mount)) != 0) { 313 vn_start_write(NULL, &wrtmp, V_WAIT); 314 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 315 goto out; 316 } 317 if (mp->mnt_kern_flag & MNTK_SUSPENDED) 318 break; 319 vn_start_write(NULL, &wrtmp, V_WAIT); 320 } 321 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 322 if (collectsnapstats) 323 nanotime(&starttime); 324 /* 325 * First, copy all the cylinder group maps that have changed. 326 */ 327 for (cg = 0; cg < fs->fs_ncg; cg++) { 328 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0) 329 continue; 330 redo++; 331 error = UFS_BALLOC(vp, (off_t)(cgtod(fs, cg)) << fs->fs_fshift, 332 fs->fs_bsize, KERNCRED, 0, &nbp); 333 if (error) 334 goto out1; 335 error = cgaccount(cg, vp, nbp, 2); 336 bawrite(nbp); 337 if (error) 338 goto out1; 339 } 340 /* 341 * Grab a copy of the superblock and its summary information. 342 * We delay writing it until the suspension is released below. 343 */ 344 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize, 345 KERNCRED, &sbp); 346 if (error) { 347 brelse(sbp); 348 sbp = NULL; 349 goto out1; 350 } 351 loc = blkoff(fs, fs->fs_sblockloc); 352 copy_fs = (struct fs *)(sbp->b_data + loc); 353 bcopy(fs, copy_fs, fs->fs_sbsize); 354 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 355 copy_fs->fs_clean = 1; 356 if (fs->fs_sbsize < SBLOCKSIZE) 357 bzero(&sbp->b_data[loc + fs->fs_sbsize], 358 SBLOCKSIZE - fs->fs_sbsize); 359 size = blkroundup(fs, fs->fs_cssize); 360 if (fs->fs_contigsumsize > 0) 361 size += fs->fs_ncg * sizeof(int32_t); 362 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 363 copy_fs->fs_csp = space; 364 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize); 365 (char *)space += fs->fs_cssize; 366 loc = howmany(fs->fs_cssize, fs->fs_fsize); 367 i = fs->fs_frag - loc % fs->fs_frag; 368 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize; 369 if (len > 0) { 370 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc), 371 len, KERNCRED, &bp)) != 0) { 372 brelse(bp); 373 free(copy_fs->fs_csp, M_UFSMNT); 374 bawrite(sbp); 375 sbp = NULL; 376 goto out1; 377 } 378 bcopy(bp->b_data, space, (u_int)len); 379 (char *)space += len; 380 bp->b_flags |= B_INVAL | B_NOCACHE; 381 brelse(bp); 382 } 383 if (fs->fs_contigsumsize > 0) { 384 copy_fs->fs_maxcluster = lp = space; 385 for (i = 0; i < fs->fs_ncg; i++) 386 *lp++ = fs->fs_contigsumsize; 387 } 388 /* 389 * We must check for active files that have been unlinked 390 * (e.g., with a zero link count). We have to expunge all 391 * trace of these files from the snapshot so that they are 392 * not reclaimed prematurely by fsck or unnecessarily dumped. 393 * We turn off the MNTK_SUSPENDED flag to avoid a panic from 394 * spec_strategy about writing on a suspended filesystem. 395 * Note that we skip unlinked snapshot files as they will 396 * be handled separately below. 397 */ 398 mp->mnt_kern_flag &= ~MNTK_SUSPENDED; 399 mtx_lock(&mntvnode_mtx); 400loop: 401 for (xvp = TAILQ_FIRST(&mp->mnt_nvnodelist); xvp; xvp = nvp) { 402 /* 403 * Make sure this vnode wasn't reclaimed in getnewvnode(). 404 * Start over if it has (it won't be on the list anymore). 405 */ 406 if (xvp->v_mount != mp) 407 goto loop; 408 nvp = TAILQ_NEXT(xvp, v_nmntvnodes); 409 mtx_unlock(&mntvnode_mtx); 410 mp_fixme("Unlocked GETATTR."); 411 if (vrefcnt(xvp) == 0 || xvp->v_type == VNON || 412 (VTOI(xvp)->i_flags & SF_SNAPSHOT) || 413 (VOP_GETATTR(xvp, &vat, td->td_proc->p_ucred, td) == 0 && 414 vat.va_nlink > 0)) { 415 mtx_lock(&mntvnode_mtx); 416 continue; 417 } 418 if (snapdebug) 419 vprint("ffs_snapshot: busy vnode", xvp); 420 if (vn_lock(xvp, LK_EXCLUSIVE, td) != 0) 421 goto loop; 422 xp = VTOI(xvp); 423 /* 424 * If there is a fragment, clear it here. 425 */ 426 blkno = 0; 427 loc = howmany(xp->i_size, fs->fs_bsize) - 1; 428 if (loc < NDADDR) { 429 len = fragroundup(fs, blkoff(fs, xp->i_size)); 430 if (len < fs->fs_bsize) { 431 ffs_blkfree(copy_fs, vp, DIP(xp, i_db[loc]), 432 len, xp->i_number); 433 blkno = DIP(xp, i_db[loc]); 434 DIP(xp, i_db[loc]) = 0; 435 } 436 } 437 if (xp->i_ump->um_fstype == UFS1) 438 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, 439 BLK_NOCOPY); 440 else 441 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, 442 BLK_NOCOPY); 443 if (blkno) 444 DIP(xp, i_db[loc]) = blkno; 445 if (!error) 446 error = ffs_freefile(copy_fs, vp, xp->i_number, 447 xp->i_mode); 448 VOP_UNLOCK(xvp, 0, td); 449 if (error) { 450 free(copy_fs->fs_csp, M_UFSMNT); 451 bawrite(sbp); 452 sbp = NULL; 453 goto out1; 454 } 455 mtx_lock(&mntvnode_mtx); 456 } 457 mtx_unlock(&mntvnode_mtx); 458 /* 459 * If there already exist snapshots on this filesystem, grab a 460 * reference to their shared lock. If this is the first snapshot 461 * on this filesystem, we need to allocate a lock for the snapshots 462 * to share. In either case, acquire the snapshot lock and give 463 * up our original private lock. 464 */ 465 VI_LOCK(devvp); 466 snaphead = &devvp->v_rdev->si_snapshots; 467 if ((xp = TAILQ_FIRST(snaphead)) != NULL) { 468 VI_LOCK(vp); 469 vp->v_vnlock = ITOV(xp)->v_vnlock; 470 VI_UNLOCK(devvp); 471 } else { 472 struct lock *lkp; 473 474 VI_UNLOCK(devvp); 475 MALLOC(lkp, struct lock *, sizeof(struct lock), M_UFSMNT, 476 M_WAITOK); 477 lockinit(lkp, PVFS, "snaplk", VLKTIMEOUT, 478 LK_CANRECURSE | LK_NOPAUSE); 479 VI_LOCK(vp); 480 vp->v_vnlock = lkp; 481 } 482 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td); 483 transferlockers(&vp->v_lock, vp->v_vnlock); 484 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td); 485 /* 486 * Record snapshot inode. Since this is the newest snapshot, 487 * it must be placed at the end of the list. 488 */ 489 VI_LOCK(devvp); 490 fs->fs_snapinum[snaploc] = ip->i_number; 491 if (ip->i_nextsnap.tqe_prev != 0) 492 panic("ffs_snapshot: %d already on list", ip->i_number); 493 TAILQ_INSERT_TAIL(snaphead, ip, i_nextsnap); 494 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite; 495 devvp->v_vflag |= VV_COPYONWRITE; 496 VI_UNLOCK(devvp); 497 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp"); 498 vp->v_vflag |= VV_SYSTEM; 499out1: 500 /* 501 * Resume operation on filesystem. 502 */ 503 vfs_write_resume(vp->v_mount); 504 vn_start_write(NULL, &wrtmp, V_WAIT); 505 if (collectsnapstats && starttime.tv_sec > 0) { 506 nanotime(&endtime); 507 timespecsub(&endtime, &starttime); 508 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n", 509 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec, 510 endtime.tv_nsec / 1000000, redo, fs->fs_ncg); 511 } 512 if (sbp == NULL) 513 goto out; 514 /* 515 * Copy allocation information from all the snapshots in 516 * this snapshot and then expunge them from its view. 517 */ 518 snaphead = &devvp->v_rdev->si_snapshots; 519 TAILQ_FOREACH(xp, snaphead, i_nextsnap) { 520 if (xp == ip) 521 break; 522 if (xp->i_ump->um_fstype == UFS1) 523 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1, 524 BLK_SNAP); 525 else 526 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2, 527 BLK_SNAP); 528 if (error) { 529 fs->fs_snapinum[snaploc] = 0; 530 goto done; 531 } 532 } 533 /* 534 * Allocate the space for the list of preallocated snapshot blocks. 535 */ 536 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) + 537 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */; 538 MALLOC(snapblklist, daddr_t *, snaplistsize * sizeof(daddr_t), 539 M_UFSMNT, M_WAITOK); 540 ip->i_snapblklist = &snapblklist[1]; 541 /* 542 * Expunge the blocks used by the snapshots from the set of 543 * blocks marked as used in the snapshot bitmaps. Also, collect 544 * the list of allocated blocks in i_snapblklist. 545 */ 546 if (ip->i_ump->um_fstype == UFS1) 547 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, BLK_SNAP); 548 else 549 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, BLK_SNAP); 550 if (error) { 551 fs->fs_snapinum[snaploc] = 0; 552 FREE(snapblklist, M_UFSMNT); 553 goto done; 554 } 555 snaplistsize = ip->i_snapblklist - snapblklist; 556 snapblklist[0] = snaplistsize; 557 ip->i_snapblklist = 0; 558 /* 559 * Write out the list of allocated blocks to the end of the snapshot. 560 */ 561 auio.uio_iov = &aiov; 562 auio.uio_iovcnt = 1; 563 aiov.iov_base = (void *)snapblklist; 564 aiov.iov_len = snaplistsize * sizeof(daddr_t); 565 auio.uio_resid = aiov.iov_len;; 566 auio.uio_offset = ip->i_size; 567 auio.uio_segflg = UIO_SYSSPACE; 568 auio.uio_rw = UIO_WRITE; 569 auio.uio_td = td; 570 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 571 fs->fs_snapinum[snaploc] = 0; 572 FREE(snapblklist, M_UFSMNT); 573 goto done; 574 } 575 /* 576 * Write the superblock and its summary information 577 * to the snapshot. 578 */ 579 blkno = fragstoblks(fs, fs->fs_csaddr); 580 len = howmany(fs->fs_cssize, fs->fs_bsize); 581 space = copy_fs->fs_csp; 582 for (loc = 0; loc < len; loc++) { 583 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp); 584 if (error) { 585 brelse(nbp); 586 fs->fs_snapinum[snaploc] = 0; 587 FREE(snapblklist, M_UFSMNT); 588 goto done; 589 } 590 bcopy(space, nbp->b_data, fs->fs_bsize); 591 space = (char *)space + fs->fs_bsize; 592 bawrite(nbp); 593 } 594 /* 595 * As this is the newest list, it is the most inclusive, so 596 * should replace the previous list. 597 */ 598 VI_LOCK(devvp); 599 FREE(devvp->v_rdev->si_snapblklist, M_UFSMNT); 600 devvp->v_rdev->si_snapblklist = snapblklist; 601 devvp->v_rdev->si_snaplistsize = snaplistsize; 602 VI_UNLOCK(devvp); 603done: 604 free(copy_fs->fs_csp, M_UFSMNT); 605 bawrite(sbp); 606out: 607 if (saved_nice > 0) 608 td->td_ksegrp->kg_nice = saved_nice; 609 if (fs->fs_active != 0) { 610 FREE(fs->fs_active, M_DEVBUF); 611 fs->fs_active = 0; 612 } 613 mp->mnt_flag = flag; 614 if (error) 615 (void) UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td); 616 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 617 if (error) 618 vput(vp); 619 else 620 VOP_UNLOCK(vp, 0, td); 621 vn_finished_write(wrtmp); 622 return (error); 623} 624 625/* 626 * Copy a cylinder group map. All the unallocated blocks are marked 627 * BLK_NOCOPY so that the snapshot knows that it need not copy them 628 * if they are later written. If passno is one, then this is a first 629 * pass, so only setting needs to be done. If passno is 2, then this 630 * is a revision to a previous pass which must be undone as the 631 * replacement pass is done. 632 */ 633static int 634cgaccount(cg, vp, nbp, passno) 635 int cg; 636 struct vnode *vp; 637 struct buf *nbp; 638 int passno; 639{ 640 struct buf *bp, *ibp; 641 struct inode *ip; 642 struct cg *cgp; 643 struct fs *fs; 644 ufs2_daddr_t base, numblks; 645 int error, len, loc, indiroff; 646 647 ip = VTOI(vp); 648 fs = ip->i_fs; 649 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 650 (int)fs->fs_cgsize, KERNCRED, &bp); 651 if (error) { 652 brelse(bp); 653 return (error); 654 } 655 cgp = (struct cg *)bp->b_data; 656 if (!cg_chkmagic(cgp)) { 657 brelse(bp); 658 return (EIO); 659 } 660 atomic_set_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg)); 661 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize); 662 if (fs->fs_cgsize < fs->fs_bsize) 663 bzero(&nbp->b_data[fs->fs_cgsize], 664 fs->fs_bsize - fs->fs_cgsize); 665 if (passno == 2) 666 nbp->b_flags |= B_VALIDSUSPWRT; 667 numblks = howmany(fs->fs_size, fs->fs_frag); 668 len = howmany(fs->fs_fpg, fs->fs_frag); 669 base = cg * fs->fs_fpg / fs->fs_frag; 670 if (base + len >= numblks) 671 len = numblks - base - 1; 672 loc = 0; 673 if (base < NDADDR) { 674 for ( ; loc < NDADDR; loc++) { 675 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 676 DIP(ip, i_db[loc]) = BLK_NOCOPY; 677 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY) 678 DIP(ip, i_db[loc]) = 0; 679 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY) 680 panic("ffs_snapshot: lost direct block"); 681 } 682 } 683 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)), 684 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 685 if (error) { 686 brelse(bp); 687 return (error); 688 } 689 indiroff = (base + loc - NDADDR) % NINDIR(fs); 690 for ( ; loc < len; loc++, indiroff++) { 691 if (indiroff >= NINDIR(fs)) { 692 if (passno == 2) 693 ibp->b_flags |= B_VALIDSUSPWRT; 694 bawrite(ibp); 695 error = UFS_BALLOC(vp, 696 lblktosize(fs, (off_t)(base + loc)), 697 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 698 if (error) { 699 brelse(bp); 700 return (error); 701 } 702 indiroff = 0; 703 } 704 if (ip->i_ump->um_fstype == UFS1) { 705 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 706 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 707 BLK_NOCOPY; 708 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data)) 709 [indiroff] == BLK_NOCOPY) 710 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0; 711 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data)) 712 [indiroff] == BLK_NOCOPY) 713 panic("ffs_snapshot: lost indirect block"); 714 continue; 715 } 716 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 717 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; 718 else if (passno == 2 && 719 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 720 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0; 721 else if (passno == 1 && 722 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 723 panic("ffs_snapshot: lost indirect block"); 724 } 725 bqrelse(bp); 726 if (passno == 2) 727 ibp->b_flags |= B_VALIDSUSPWRT; 728 bdwrite(ibp); 729 return (0); 730} 731 732/* 733 * Before expunging a snapshot inode, note all the 734 * blocks that it claims with BLK_SNAP so that fsck will 735 * be able to account for those blocks properly and so 736 * that this snapshot knows that it need not copy them 737 * if the other snapshot holding them is freed. This code 738 * is reproduced once each for UFS1 and UFS2. 739 */ 740static int 741expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype) 742 struct vnode *snapvp; 743 struct inode *cancelip; 744 struct fs *fs; 745 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 746 struct fs *, ufs_lbn_t, int); 747 int expungetype; 748{ 749 int i, error, indiroff; 750 ufs_lbn_t lbn, rlbn; 751 ufs2_daddr_t len, blkno, numblks, blksperindir; 752 struct ufs1_dinode *dip; 753 struct thread *td = curthread; 754 struct buf *bp; 755 756 /* 757 * Prepare to expunge the inode. If its inode block has not 758 * yet been copied, then allocate and fill the copy. 759 */ 760 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 761 blkno = 0; 762 if (lbn < NDADDR) { 763 blkno = VTOI(snapvp)->i_din1->di_db[lbn]; 764 } else { 765 td->td_proc->p_flag |= P_COWINPROGRESS; 766 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn), 767 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 768 td->td_proc->p_flag &= ~P_COWINPROGRESS; 769 if (error) 770 return (error); 771 indiroff = (lbn - NDADDR) % NINDIR(fs); 772 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; 773 bqrelse(bp); 774 } 775 if (blkno != 0) { 776 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 777 return (error); 778 } else { 779 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn), 780 fs->fs_bsize, KERNCRED, 0, &bp); 781 if (error) 782 return (error); 783 if ((error = readblock(bp, lbn)) != 0) 784 return (error); 785 } 786 /* 787 * Set a snapshot inode to be a zero length file, regular files 788 * to be completely unallocated. 789 */ 790 dip = (struct ufs1_dinode *)bp->b_data + 791 ino_to_fsbo(fs, cancelip->i_number); 792 if (expungetype == BLK_NOCOPY) 793 dip->di_mode = 0; 794 dip->di_size = 0; 795 dip->di_blocks = 0; 796 dip->di_flags &= ~SF_SNAPSHOT; 797 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t)); 798 bdwrite(bp); 799 /* 800 * Now go through and expunge all the blocks in the file 801 * using the function requested. 802 */ 803 numblks = howmany(cancelip->i_size, fs->fs_bsize); 804 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0], 805 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype))) 806 return (error); 807 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0], 808 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype))) 809 return (error); 810 blksperindir = 1; 811 lbn = -NDADDR; 812 len = numblks - NDADDR; 813 rlbn = NDADDR; 814 for (i = 0; len > 0 && i < NIADDR; i++) { 815 error = indiracct_ufs1(snapvp, ITOV(cancelip), i, 816 cancelip->i_din1->di_ib[i], lbn, rlbn, len, 817 blksperindir, fs, acctfunc, expungetype); 818 if (error) 819 return (error); 820 blksperindir *= NINDIR(fs); 821 lbn -= blksperindir + 1; 822 len -= blksperindir; 823 rlbn += blksperindir; 824 } 825 return (0); 826} 827 828/* 829 * Descend an indirect block chain for vnode cancelvp accounting for all 830 * its indirect blocks in snapvp. 831 */ 832static int 833indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 834 blksperindir, fs, acctfunc, expungetype) 835 struct vnode *snapvp; 836 struct vnode *cancelvp; 837 int level; 838 ufs1_daddr_t blkno; 839 ufs_lbn_t lbn; 840 ufs_lbn_t rlbn; 841 ufs_lbn_t remblks; 842 ufs_lbn_t blksperindir; 843 struct fs *fs; 844 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 845 struct fs *, ufs_lbn_t, int); 846 int expungetype; 847{ 848 int error, num, i; 849 ufs_lbn_t subblksperindir; 850 struct indir indirs[NIADDR + 2]; 851 ufs1_daddr_t last, *bap; 852 struct buf *bp; 853 854 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 855 return (error); 856 if (lbn != indirs[num - 1 - level].in_lbn || blkno == 0 || num < 2) 857 panic("indiracct: botched params"); 858 /* 859 * We have to expand bread here since it will deadlock looking 860 * up the block number for any blocks that are not in the cache. 861 */ 862 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0); 863 bp->b_blkno = fsbtodb(fs, blkno); 864 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && 865 (error = readblock(bp, fragstoblks(fs, blkno)))) { 866 brelse(bp); 867 return (error); 868 } 869 /* 870 * Account for the block pointers in this indirect block. 871 */ 872 last = howmany(remblks, blksperindir); 873 if (last > NINDIR(fs)) 874 last = NINDIR(fs); 875 MALLOC(bap, ufs1_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK); 876 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 877 bqrelse(bp); 878 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 879 level == 0 ? rlbn : -1, expungetype); 880 if (error || level == 0) 881 goto out; 882 /* 883 * Account for the block pointers in each of the indirect blocks 884 * in the levels below us. 885 */ 886 subblksperindir = blksperindir / NINDIR(fs); 887 for (lbn++, level--, i = 0; i < last; i++) { 888 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn, 889 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 890 if (error) 891 goto out; 892 rlbn += blksperindir; 893 lbn -= blksperindir; 894 remblks -= blksperindir; 895 } 896out: 897 FREE(bap, M_DEVBUF); 898 return (error); 899} 900 901/* 902 * Do both snap accounting and map accounting. 903 */ 904static int 905fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype) 906 struct vnode *vp; 907 ufs1_daddr_t *oldblkp, *lastblkp; 908 struct fs *fs; 909 ufs_lbn_t lblkno; 910 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 911{ 912 int error; 913 914 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 915 return (error); 916 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 917} 918 919/* 920 * Identify a set of blocks allocated in a snapshot inode. 921 */ 922static int 923snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 924 struct vnode *vp; 925 ufs1_daddr_t *oldblkp, *lastblkp; 926 struct fs *fs; 927 ufs_lbn_t lblkno; 928 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 929{ 930 struct inode *ip = VTOI(vp); 931 ufs1_daddr_t blkno, *blkp; 932 ufs_lbn_t lbn; 933 struct buf *ibp; 934 int error; 935 936 for ( ; oldblkp < lastblkp; oldblkp++) { 937 blkno = *oldblkp; 938 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 939 continue; 940 lbn = fragstoblks(fs, blkno); 941 if (lbn < NDADDR) { 942 blkp = &ip->i_din1->di_db[lbn]; 943 ip->i_flag |= IN_CHANGE | IN_UPDATE; 944 } else { 945 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 946 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 947 if (error) 948 return (error); 949 blkp = &((ufs1_daddr_t *)(ibp->b_data)) 950 [(lbn - NDADDR) % NINDIR(fs)]; 951 } 952 /* 953 * If we are expunging a snapshot vnode and we 954 * find a block marked BLK_NOCOPY, then it is 955 * one that has been allocated to this snapshot after 956 * we took our current snapshot and can be ignored. 957 */ 958 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 959 if (lbn >= NDADDR) 960 brelse(ibp); 961 } else { 962 if (*blkp != 0) 963 panic("snapacct: bad block"); 964 *blkp = expungetype; 965 if (lbn >= NDADDR) 966 bdwrite(ibp); 967 } 968 } 969 return (0); 970} 971 972/* 973 * Account for a set of blocks allocated in a snapshot inode. 974 */ 975static int 976mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 977 struct vnode *vp; 978 ufs1_daddr_t *oldblkp, *lastblkp; 979 struct fs *fs; 980 ufs_lbn_t lblkno; 981 int expungetype; 982{ 983 ufs1_daddr_t blkno; 984 struct inode *ip; 985 ino_t inum; 986 987 /* 988 * We only care about the leaf block numbers, not the 989 * meta-block numbers. 990 */ 991 if (lblkno == -1) 992 return (0); 993 ip = VTOI(vp); 994 inum = ip->i_number; 995 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 996 blkno = *oldblkp; 997 if (blkno == 0 || blkno == BLK_NOCOPY) 998 continue; 999 if (expungetype == BLK_SNAP && blkno != BLK_SNAP) 1000 *ip->i_snapblklist++ = lblkno; 1001 if (blkno == BLK_SNAP) 1002 blkno = blkstofrags(fs, lblkno); 1003 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum); 1004 } 1005 return (0); 1006} 1007 1008/* 1009 * Before expunging a snapshot inode, note all the 1010 * blocks that it claims with BLK_SNAP so that fsck will 1011 * be able to account for those blocks properly and so 1012 * that this snapshot knows that it need not copy them 1013 * if the other snapshot holding them is freed. This code 1014 * is reproduced once each for UFS1 and UFS2. 1015 */ 1016static int 1017expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype) 1018 struct vnode *snapvp; 1019 struct inode *cancelip; 1020 struct fs *fs; 1021 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1022 struct fs *, ufs_lbn_t, int); 1023 int expungetype; 1024{ 1025 int i, error, indiroff; 1026 ufs_lbn_t lbn, rlbn; 1027 ufs2_daddr_t len, blkno, numblks, blksperindir; 1028 struct ufs2_dinode *dip; 1029 struct thread *td = curthread; 1030 struct buf *bp; 1031 1032 /* 1033 * Prepare to expunge the inode. If its inode block has not 1034 * yet been copied, then allocate and fill the copy. 1035 */ 1036 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 1037 blkno = 0; 1038 if (lbn < NDADDR) { 1039 blkno = VTOI(snapvp)->i_din2->di_db[lbn]; 1040 } else { 1041 td->td_proc->p_flag |= P_COWINPROGRESS; 1042 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn), 1043 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 1044 td->td_proc->p_flag &= ~P_COWINPROGRESS; 1045 if (error) 1046 return (error); 1047 indiroff = (lbn - NDADDR) % NINDIR(fs); 1048 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; 1049 bqrelse(bp); 1050 } 1051 if (blkno != 0) { 1052 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 1053 return (error); 1054 } else { 1055 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn), 1056 fs->fs_bsize, KERNCRED, 0, &bp); 1057 if (error) 1058 return (error); 1059 if ((error = readblock(bp, lbn)) != 0) 1060 return (error); 1061 } 1062 /* 1063 * Set a snapshot inode to be a zero length file, regular files 1064 * to be completely unallocated. 1065 */ 1066 dip = (struct ufs2_dinode *)bp->b_data + 1067 ino_to_fsbo(fs, cancelip->i_number); 1068 if (expungetype == BLK_NOCOPY) 1069 dip->di_mode = 0; 1070 dip->di_size = 0; 1071 dip->di_blocks = 0; 1072 dip->di_flags &= ~SF_SNAPSHOT; 1073 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t)); 1074 bdwrite(bp); 1075 /* 1076 * Now go through and expunge all the blocks in the file 1077 * using the function requested. 1078 */ 1079 numblks = howmany(cancelip->i_size, fs->fs_bsize); 1080 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0], 1081 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype))) 1082 return (error); 1083 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0], 1084 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype))) 1085 return (error); 1086 blksperindir = 1; 1087 lbn = -NDADDR; 1088 len = numblks - NDADDR; 1089 rlbn = NDADDR; 1090 for (i = 0; len > 0 && i < NIADDR; i++) { 1091 error = indiracct_ufs2(snapvp, ITOV(cancelip), i, 1092 cancelip->i_din2->di_ib[i], lbn, rlbn, len, 1093 blksperindir, fs, acctfunc, expungetype); 1094 if (error) 1095 return (error); 1096 blksperindir *= NINDIR(fs); 1097 lbn -= blksperindir + 1; 1098 len -= blksperindir; 1099 rlbn += blksperindir; 1100 } 1101 return (0); 1102} 1103 1104/* 1105 * Descend an indirect block chain for vnode cancelvp accounting for all 1106 * its indirect blocks in snapvp. 1107 */ 1108static int 1109indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 1110 blksperindir, fs, acctfunc, expungetype) 1111 struct vnode *snapvp; 1112 struct vnode *cancelvp; 1113 int level; 1114 ufs2_daddr_t blkno; 1115 ufs_lbn_t lbn; 1116 ufs_lbn_t rlbn; 1117 ufs_lbn_t remblks; 1118 ufs_lbn_t blksperindir; 1119 struct fs *fs; 1120 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1121 struct fs *, ufs_lbn_t, int); 1122 int expungetype; 1123{ 1124 int error, num, i; 1125 ufs_lbn_t subblksperindir; 1126 struct indir indirs[NIADDR + 2]; 1127 ufs2_daddr_t last, *bap; 1128 struct buf *bp; 1129 1130 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 1131 return (error); 1132 if (lbn != indirs[num - 1 - level].in_lbn || blkno == 0 || num < 2) 1133 panic("indiracct: botched params"); 1134 /* 1135 * We have to expand bread here since it will deadlock looking 1136 * up the block number for any blocks that are not in the cache. 1137 */ 1138 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0); 1139 bp->b_blkno = fsbtodb(fs, blkno); 1140 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && 1141 (error = readblock(bp, fragstoblks(fs, blkno)))) { 1142 brelse(bp); 1143 return (error); 1144 } 1145 /* 1146 * Account for the block pointers in this indirect block. 1147 */ 1148 last = howmany(remblks, blksperindir); 1149 if (last > NINDIR(fs)) 1150 last = NINDIR(fs); 1151 MALLOC(bap, ufs2_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK); 1152 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 1153 bqrelse(bp); 1154 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 1155 level == 0 ? rlbn : -1, expungetype); 1156 if (error || level == 0) 1157 goto out; 1158 /* 1159 * Account for the block pointers in each of the indirect blocks 1160 * in the levels below us. 1161 */ 1162 subblksperindir = blksperindir / NINDIR(fs); 1163 for (lbn++, level--, i = 0; i < last; i++) { 1164 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn, 1165 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 1166 if (error) 1167 goto out; 1168 rlbn += blksperindir; 1169 lbn -= blksperindir; 1170 remblks -= blksperindir; 1171 } 1172out: 1173 FREE(bap, M_DEVBUF); 1174 return (error); 1175} 1176 1177/* 1178 * Do both snap accounting and map accounting. 1179 */ 1180static int 1181fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype) 1182 struct vnode *vp; 1183 ufs2_daddr_t *oldblkp, *lastblkp; 1184 struct fs *fs; 1185 ufs_lbn_t lblkno; 1186 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 1187{ 1188 int error; 1189 1190 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 1191 return (error); 1192 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 1193} 1194 1195/* 1196 * Identify a set of blocks allocated in a snapshot inode. 1197 */ 1198static int 1199snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1200 struct vnode *vp; 1201 ufs2_daddr_t *oldblkp, *lastblkp; 1202 struct fs *fs; 1203 ufs_lbn_t lblkno; 1204 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 1205{ 1206 struct inode *ip = VTOI(vp); 1207 ufs2_daddr_t blkno, *blkp; 1208 ufs_lbn_t lbn; 1209 struct buf *ibp; 1210 int error; 1211 1212 for ( ; oldblkp < lastblkp; oldblkp++) { 1213 blkno = *oldblkp; 1214 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 1215 continue; 1216 lbn = fragstoblks(fs, blkno); 1217 if (lbn < NDADDR) { 1218 blkp = &ip->i_din2->di_db[lbn]; 1219 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1220 } else { 1221 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1222 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1223 if (error) 1224 return (error); 1225 blkp = &((ufs2_daddr_t *)(ibp->b_data)) 1226 [(lbn - NDADDR) % NINDIR(fs)]; 1227 } 1228 /* 1229 * If we are expunging a snapshot vnode and we 1230 * find a block marked BLK_NOCOPY, then it is 1231 * one that has been allocated to this snapshot after 1232 * we took our current snapshot and can be ignored. 1233 */ 1234 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 1235 if (lbn >= NDADDR) 1236 brelse(ibp); 1237 } else { 1238 if (*blkp != 0) 1239 panic("snapacct: bad block"); 1240 *blkp = expungetype; 1241 if (lbn >= NDADDR) 1242 bdwrite(ibp); 1243 } 1244 } 1245 return (0); 1246} 1247 1248/* 1249 * Account for a set of blocks allocated in a snapshot inode. 1250 */ 1251static int 1252mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1253 struct vnode *vp; 1254 ufs2_daddr_t *oldblkp, *lastblkp; 1255 struct fs *fs; 1256 ufs_lbn_t lblkno; 1257 int expungetype; 1258{ 1259 ufs2_daddr_t blkno; 1260 struct inode *ip; 1261 ino_t inum; 1262 1263 /* 1264 * We only care about the leaf block numbers, not the 1265 * meta-block numbers. 1266 */ 1267 if (lblkno == -1) 1268 return (0); 1269 ip = VTOI(vp); 1270 inum = ip->i_number; 1271 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 1272 blkno = *oldblkp; 1273 if (blkno == 0 || blkno == BLK_NOCOPY) 1274 continue; 1275 if (expungetype == BLK_SNAP && blkno != BLK_SNAP) 1276 *ip->i_snapblklist++ = lblkno; 1277 if (blkno == BLK_SNAP) 1278 blkno = blkstofrags(fs, lblkno); 1279 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum); 1280 } 1281 return (0); 1282} 1283 1284/* 1285 * Decrement extra reference on snapshot when last name is removed. 1286 * It will not be freed until the last open reference goes away. 1287 */ 1288void 1289ffs_snapgone(ip) 1290 struct inode *ip; 1291{ 1292 struct inode *xp; 1293 struct fs *fs; 1294 int snaploc; 1295 1296 /* 1297 * Find snapshot in incore list. 1298 */ 1299 TAILQ_FOREACH(xp, &ip->i_devvp->v_rdev->si_snapshots, i_nextsnap) 1300 if (xp == ip) 1301 break; 1302 if (xp != NULL) 1303 vrele(ITOV(ip)); 1304 else if (snapdebug) 1305 printf("ffs_snapgone: lost snapshot vnode %d\n", 1306 ip->i_number); 1307 /* 1308 * Delete snapshot inode from superblock. Keep list dense. 1309 */ 1310 fs = ip->i_fs; 1311 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 1312 if (fs->fs_snapinum[snaploc] == ip->i_number) 1313 break; 1314 if (snaploc < FSMAXSNAP) { 1315 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) { 1316 if (fs->fs_snapinum[snaploc] == 0) 1317 break; 1318 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc]; 1319 } 1320 fs->fs_snapinum[snaploc - 1] = 0; 1321 } 1322} 1323 1324/* 1325 * Prepare a snapshot file for being removed. 1326 */ 1327void 1328ffs_snapremove(vp) 1329 struct vnode *vp; 1330{ 1331 struct inode *ip; 1332 struct vnode *devvp; 1333 struct lock *lkp; 1334 struct buf *ibp; 1335 struct fs *fs; 1336 struct thread *td = curthread; 1337 ufs2_daddr_t numblks, blkno, dblk, *snapblklist; 1338 int error, loc, last; 1339 1340 ip = VTOI(vp); 1341 fs = ip->i_fs; 1342 devvp = ip->i_devvp; 1343 /* 1344 * If active, delete from incore list (this snapshot may 1345 * already have been in the process of being deleted, so 1346 * would not have been active). 1347 * 1348 * Clear copy-on-write flag if last snapshot. 1349 */ 1350 if (ip->i_nextsnap.tqe_prev != 0) { 1351 VI_LOCK(devvp); 1352 lockmgr(&vp->v_lock, LK_INTERLOCK | LK_EXCLUSIVE, 1353 VI_MTX(devvp), td); 1354 VI_LOCK(devvp); 1355 TAILQ_REMOVE(&devvp->v_rdev->si_snapshots, ip, i_nextsnap); 1356 ip->i_nextsnap.tqe_prev = 0; 1357 lkp = vp->v_vnlock; 1358 vp->v_vnlock = &vp->v_lock; 1359 lockmgr(lkp, LK_RELEASE, NULL, td); 1360 if (TAILQ_FIRST(&devvp->v_rdev->si_snapshots) != 0) { 1361 VI_UNLOCK(devvp); 1362 } else { 1363 snapblklist = devvp->v_rdev->si_snapblklist; 1364 devvp->v_rdev->si_snapblklist = 0; 1365 devvp->v_rdev->si_snaplistsize = 0; 1366 devvp->v_rdev->si_copyonwrite = 0; 1367 devvp->v_vflag &= ~VV_COPYONWRITE; 1368 lockmgr(lkp, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp), td); 1369 lockmgr(lkp, LK_RELEASE, NULL, td); 1370 lockdestroy(lkp); 1371 FREE(lkp, M_UFSMNT); 1372 FREE(snapblklist, M_UFSMNT); 1373 } 1374 } 1375 /* 1376 * Clear all BLK_NOCOPY fields. Pass any block claims to other 1377 * snapshots that want them (see ffs_snapblkfree below). 1378 */ 1379 for (blkno = 1; blkno < NDADDR; blkno++) { 1380 dblk = DIP(ip, i_db[blkno]); 1381 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1382 DIP(ip, i_db[blkno]) = 0; 1383 else if ((dblk == blkstofrags(fs, blkno) && 1384 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize, 1385 ip->i_number))) { 1386 DIP(ip, i_blocks) -= btodb(fs->fs_bsize); 1387 DIP(ip, i_db[blkno]) = 0; 1388 } 1389 } 1390 numblks = howmany(ip->i_size, fs->fs_bsize); 1391 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 1392 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 1393 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1394 if (error) 1395 continue; 1396 if (fs->fs_size - blkno > NINDIR(fs)) 1397 last = NINDIR(fs); 1398 else 1399 last = fs->fs_size - blkno; 1400 for (loc = 0; loc < last; loc++) { 1401 if (ip->i_ump->um_fstype == UFS1) { 1402 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc]; 1403 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1404 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1405 else if ((dblk == blkstofrags(fs, blkno) && 1406 ffs_snapblkfree(fs, ip->i_devvp, dblk, 1407 fs->fs_bsize, ip->i_number))) { 1408 ip->i_din1->di_blocks -= 1409 btodb(fs->fs_bsize); 1410 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1411 } 1412 continue; 1413 } 1414 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc]; 1415 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1416 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1417 else if ((dblk == blkstofrags(fs, blkno) && 1418 ffs_snapblkfree(fs, ip->i_devvp, dblk, 1419 fs->fs_bsize, ip->i_number))) { 1420 ip->i_din2->di_blocks -= btodb(fs->fs_bsize); 1421 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1422 } 1423 } 1424 bawrite(ibp); 1425 } 1426 /* 1427 * Clear snapshot flag and drop reference. 1428 */ 1429 ip->i_flags &= ~SF_SNAPSHOT; 1430 DIP(ip, i_flags) = ip->i_flags; 1431 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1432} 1433 1434/* 1435 * Notification that a block is being freed. Return zero if the free 1436 * should be allowed to proceed. Return non-zero if the snapshot file 1437 * wants to claim the block. The block will be claimed if it is an 1438 * uncopied part of one of the snapshots. It will be freed if it is 1439 * either a BLK_NOCOPY or has already been copied in all of the snapshots. 1440 * If a fragment is being freed, then all snapshots that care about 1441 * it must make a copy since a snapshot file can only claim full sized 1442 * blocks. Note that if more than one snapshot file maps the block, 1443 * we can pick one at random to claim it. Since none of the snapshots 1444 * can change, we are assurred that they will all see the same unmodified 1445 * image. When deleting a snapshot file (see ffs_snapremove above), we 1446 * must push any of these claimed blocks to one of the other snapshots 1447 * that maps it. These claimed blocks are easily identified as they will 1448 * have a block number equal to their logical block number within the 1449 * snapshot. A copied block can never have this property because they 1450 * must always have been allocated from a BLK_NOCOPY location. 1451 */ 1452int 1453ffs_snapblkfree(fs, devvp, bno, size, inum) 1454 struct fs *fs; 1455 struct vnode *devvp; 1456 ufs2_daddr_t bno; 1457 long size; 1458 ino_t inum; 1459{ 1460 struct buf *ibp, *cbp, *savedcbp = 0; 1461 struct thread *td = curthread; 1462 struct inode *ip; 1463 struct vnode *vp = NULL; 1464 ufs_lbn_t lbn; 1465 ufs2_daddr_t blkno; 1466 int indiroff = 0, snapshot_locked = 0, error = 0, claimedblk = 0; 1467 struct snaphead *snaphead; 1468 1469 lbn = fragstoblks(fs, bno); 1470retry: 1471 VI_LOCK(devvp); 1472 snaphead = &devvp->v_rdev->si_snapshots; 1473 TAILQ_FOREACH(ip, snaphead, i_nextsnap) { 1474 vp = ITOV(ip); 1475 /* 1476 * Lookup block being written. 1477 */ 1478 if (lbn < NDADDR) { 1479 blkno = DIP(ip, i_db[lbn]); 1480 } else { 1481 if (snapshot_locked == 0 && 1482 lockmgr(vp->v_vnlock, 1483 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 1484 VI_MTX(devvp), td) != 0) 1485 goto retry; 1486 snapshot_locked = 1; 1487 td->td_proc->p_flag |= P_COWINPROGRESS; 1488 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1489 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1490 td->td_proc->p_flag &= ~P_COWINPROGRESS; 1491 if (error) 1492 break; 1493 indiroff = (lbn - NDADDR) % NINDIR(fs); 1494 if (ip->i_ump->um_fstype == UFS1) 1495 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 1496 else 1497 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 1498 } 1499 /* 1500 * Check to see if block needs to be copied. 1501 */ 1502 if (blkno == 0) { 1503 /* 1504 * A block that we map is being freed. If it has not 1505 * been claimed yet, we will claim or copy it (below). 1506 */ 1507 claimedblk = 1; 1508 } else if (blkno == BLK_SNAP) { 1509 /* 1510 * No previous snapshot claimed the block, 1511 * so it will be freed and become a BLK_NOCOPY 1512 * (don't care) for us. 1513 */ 1514 if (claimedblk) 1515 panic("snapblkfree: inconsistent block type"); 1516 if (snapshot_locked == 0 && 1517 lockmgr(vp->v_vnlock, 1518 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT, 1519 VI_MTX(devvp), td) != 0) { 1520 if (lbn >= NDADDR) 1521 bqrelse(ibp); 1522 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td); 1523 goto retry; 1524 } 1525 snapshot_locked = 1; 1526 if (lbn < NDADDR) { 1527 DIP(ip, i_db[lbn]) = BLK_NOCOPY; 1528 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1529 } else if (ip->i_ump->um_fstype == UFS1) { 1530 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 1531 BLK_NOCOPY; 1532 bdwrite(ibp); 1533 } else { 1534 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 1535 BLK_NOCOPY; 1536 bdwrite(ibp); 1537 } 1538 continue; 1539 } else /* BLK_NOCOPY or default */ { 1540 /* 1541 * If the snapshot has already copied the block 1542 * (default), or does not care about the block, 1543 * it is not needed. 1544 */ 1545 if (lbn >= NDADDR) 1546 bqrelse(ibp); 1547 continue; 1548 } 1549 /* 1550 * If this is a full size block, we will just grab it 1551 * and assign it to the snapshot inode. Otherwise we 1552 * will proceed to copy it. See explanation for this 1553 * routine as to why only a single snapshot needs to 1554 * claim this block. 1555 */ 1556 if (snapshot_locked == 0 && 1557 lockmgr(vp->v_vnlock, 1558 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT, 1559 VI_MTX(devvp), td) != 0) { 1560 if (lbn >= NDADDR) 1561 bqrelse(ibp); 1562 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td); 1563 goto retry; 1564 } 1565 snapshot_locked = 1; 1566 if (size == fs->fs_bsize) { 1567#ifdef DEBUG 1568 if (snapdebug) 1569 printf("%s %d lbn %jd from inum %d\n", 1570 "Grabonremove: snapino", ip->i_number, 1571 (intmax_t)lbn, inum); 1572#endif 1573 if (lbn < NDADDR) { 1574 DIP(ip, i_db[lbn]) = bno; 1575 } else if (ip->i_ump->um_fstype == UFS1) { 1576 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno; 1577 bdwrite(ibp); 1578 } else { 1579 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno; 1580 bdwrite(ibp); 1581 } 1582 DIP(ip, i_blocks) += btodb(size); 1583 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1584 VOP_UNLOCK(vp, 0, td); 1585 return (1); 1586 } 1587 if (lbn >= NDADDR) 1588 bqrelse(ibp); 1589 /* 1590 * Allocate the block into which to do the copy. Note that this 1591 * allocation will never require any additional allocations for 1592 * the snapshot inode. 1593 */ 1594 td->td_proc->p_flag |= P_COWINPROGRESS; 1595 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1596 fs->fs_bsize, KERNCRED, 0, &cbp); 1597 td->td_proc->p_flag &= ~P_COWINPROGRESS; 1598 if (error) 1599 break; 1600#ifdef DEBUG 1601 if (snapdebug) 1602 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n", 1603 "Copyonremove: snapino ", ip->i_number, 1604 (intmax_t)lbn, "for inum", inum, size, 1605 (intmax_t)cbp->b_blkno); 1606#endif 1607 /* 1608 * If we have already read the old block contents, then 1609 * simply copy them to the new block. Note that we need 1610 * to synchronously write snapshots that have not been 1611 * unlinked, and hence will be visible after a crash, 1612 * to ensure their integrity. 1613 */ 1614 if (savedcbp != 0) { 1615 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 1616 bawrite(cbp); 1617 if (dopersistence && ip->i_effnlink > 0) 1618 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 1619 continue; 1620 } 1621 /* 1622 * Otherwise, read the old block contents into the buffer. 1623 */ 1624 if ((error = readblock(cbp, lbn)) != 0) { 1625 bzero(cbp->b_data, fs->fs_bsize); 1626 bawrite(cbp); 1627 if (dopersistence && ip->i_effnlink > 0) 1628 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 1629 break; 1630 } 1631 savedcbp = cbp; 1632 } 1633 /* 1634 * Note that we need to synchronously write snapshots that 1635 * have not been unlinked, and hence will be visible after 1636 * a crash, to ensure their integrity. 1637 */ 1638 if (savedcbp) { 1639 vp = savedcbp->b_vp; 1640 bawrite(savedcbp); 1641 if (dopersistence && VTOI(vp)->i_effnlink > 0) 1642 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 1643 } 1644 /* 1645 * If we have been unable to allocate a block in which to do 1646 * the copy, then return non-zero so that the fragment will 1647 * not be freed. Although space will be lost, the snapshot 1648 * will stay consistent. 1649 */ 1650 if (snapshot_locked) 1651 VOP_UNLOCK(vp, 0, td); 1652 else 1653 VI_UNLOCK(devvp); 1654 return (error); 1655} 1656 1657/* 1658 * Associate snapshot files when mounting. 1659 */ 1660void 1661ffs_snapshot_mount(mp) 1662 struct mount *mp; 1663{ 1664 struct ufsmount *ump = VFSTOUFS(mp); 1665 struct vnode *devvp = ump->um_devvp; 1666 struct fs *fs = ump->um_fs; 1667 struct thread *td = curthread; 1668 struct snaphead *snaphead; 1669 struct vnode *vp; 1670 struct inode *ip, *xp; 1671 struct uio auio; 1672 struct iovec aiov; 1673 void *snapblklist; 1674 char *reason; 1675 daddr_t snaplistsize; 1676 int error, snaploc, loc; 1677 1678 /* 1679 * XXX The following needs to be set before UFS_TRUNCATE or 1680 * VOP_READ can be called. 1681 */ 1682 mp->mnt_stat.f_iosize = fs->fs_bsize; 1683 /* 1684 * Process each snapshot listed in the superblock. 1685 */ 1686 vp = NULL; 1687 snaphead = &devvp->v_rdev->si_snapshots; 1688 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) { 1689 if (fs->fs_snapinum[snaploc] == 0) 1690 break; 1691 if ((error = VFS_VGET(mp, fs->fs_snapinum[snaploc], 1692 LK_EXCLUSIVE, &vp)) != 0){ 1693 printf("ffs_snapshot_mount: vget failed %d\n", error); 1694 continue; 1695 } 1696 ip = VTOI(vp); 1697 if ((ip->i_flags & SF_SNAPSHOT) == 0 || ip->i_size == 1698 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) { 1699 if ((ip->i_flags & SF_SNAPSHOT) == 0) { 1700 reason = "non-snapshot"; 1701 } else { 1702 reason = "old format snapshot"; 1703 (void)UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td); 1704 (void)VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 1705 } 1706 printf("ffs_snapshot_mount: %s inode %d\n", 1707 reason, fs->fs_snapinum[snaploc]); 1708 vput(vp); 1709 vp = NULL; 1710 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) { 1711 if (fs->fs_snapinum[loc] == 0) 1712 break; 1713 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc]; 1714 } 1715 fs->fs_snapinum[loc - 1] = 0; 1716 snaploc--; 1717 continue; 1718 } 1719 /* 1720 * If there already exist snapshots on this filesystem, grab a 1721 * reference to their shared lock. If this is the first snapshot 1722 * on this filesystem, we need to allocate a lock for the 1723 * snapshots to share. In either case, acquire the snapshot 1724 * lock and give up our original private lock. 1725 */ 1726 VI_LOCK(devvp); 1727 if ((xp = TAILQ_FIRST(snaphead)) != NULL) { 1728 VI_LOCK(vp); 1729 vp->v_vnlock = ITOV(xp)->v_vnlock; 1730 VI_UNLOCK(devvp); 1731 } else { 1732 struct lock *lkp; 1733 1734 VI_UNLOCK(devvp); 1735 MALLOC(lkp, struct lock *, sizeof(struct lock), 1736 M_UFSMNT, M_WAITOK); 1737 lockinit(lkp, PVFS, "snaplk", VLKTIMEOUT, 1738 LK_CANRECURSE | LK_NOPAUSE); 1739 VI_LOCK(vp); 1740 vp->v_vnlock = lkp; 1741 } 1742 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td); 1743 transferlockers(&vp->v_lock, vp->v_vnlock); 1744 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td); 1745 /* 1746 * Link it onto the active snapshot list. 1747 */ 1748 VI_LOCK(devvp); 1749 if (ip->i_nextsnap.tqe_prev != 0) 1750 panic("ffs_snapshot_mount: %d already on list", 1751 ip->i_number); 1752 else 1753 TAILQ_INSERT_TAIL(snaphead, ip, i_nextsnap); 1754 vp->v_vflag |= VV_SYSTEM; 1755 VI_UNLOCK(devvp); 1756 VOP_UNLOCK(vp, 0, td); 1757 } 1758 /* 1759 * No usable snapshots found. 1760 */ 1761 if (vp == NULL) 1762 return; 1763 /* 1764 * Allocate the space for the block hints list. We always want to 1765 * use the list from the newest snapshot. 1766 */ 1767 auio.uio_iov = &aiov; 1768 auio.uio_iovcnt = 1; 1769 aiov.iov_base = (void *)&snaplistsize; 1770 aiov.iov_len = sizeof(snaplistsize); 1771 auio.uio_resid = aiov.iov_len; 1772 auio.uio_offset = 1773 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)); 1774 auio.uio_segflg = UIO_SYSSPACE; 1775 auio.uio_rw = UIO_READ; 1776 auio.uio_td = td; 1777 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1778 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 1779 printf("ffs_snapshot_mount: read_1 failed %d\n", error); 1780 VOP_UNLOCK(vp, 0, td); 1781 return; 1782 } 1783 MALLOC(snapblklist, void *, snaplistsize * sizeof(daddr_t), 1784 M_UFSMNT, M_WAITOK); 1785 auio.uio_iovcnt = 1; 1786 aiov.iov_base = snapblklist; 1787 aiov.iov_len = snaplistsize * sizeof (daddr_t); 1788 auio.uio_resid = aiov.iov_len; 1789 auio.uio_offset -= sizeof(snaplistsize); 1790 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 1791 printf("ffs_snapshot_mount: read_2 failed %d\n", error); 1792 VOP_UNLOCK(vp, 0, td); 1793 FREE(snapblklist, M_UFSMNT); 1794 return; 1795 } 1796 VOP_UNLOCK(vp, 0, td); 1797 VI_LOCK(devvp); 1798 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount"); 1799 devvp->v_rdev->si_snaplistsize = snaplistsize; 1800 devvp->v_rdev->si_snapblklist = (daddr_t *)snapblklist; 1801 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite; 1802 devvp->v_vflag |= VV_COPYONWRITE; 1803 VI_UNLOCK(devvp); 1804} 1805 1806/* 1807 * Disassociate snapshot files when unmounting. 1808 */ 1809void 1810ffs_snapshot_unmount(mp) 1811 struct mount *mp; 1812{ 1813 struct vnode *devvp = VFSTOUFS(mp)->um_devvp; 1814 struct snaphead *snaphead = &devvp->v_rdev->si_snapshots; 1815 struct lock *lkp = NULL; 1816 struct inode *xp; 1817 struct vnode *vp; 1818 1819 VI_LOCK(devvp); 1820 while ((xp = TAILQ_FIRST(snaphead)) != 0) { 1821 vp = ITOV(xp); 1822 lkp = vp->v_vnlock; 1823 vp->v_vnlock = &vp->v_lock; 1824 TAILQ_REMOVE(snaphead, xp, i_nextsnap); 1825 xp->i_nextsnap.tqe_prev = 0; 1826 if (xp->i_effnlink > 0) { 1827 VI_UNLOCK(devvp); 1828 vrele(vp); 1829 VI_LOCK(devvp); 1830 } 1831 } 1832 if (devvp->v_rdev->si_snapblklist != NULL) { 1833 FREE(devvp->v_rdev->si_snapblklist, M_UFSMNT); 1834 devvp->v_rdev->si_snapblklist = NULL; 1835 devvp->v_rdev->si_snaplistsize = 0; 1836 } 1837 if (lkp != NULL) { 1838 lockdestroy(lkp); 1839 FREE(lkp, M_UFSMNT); 1840 } 1841 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount"); 1842 devvp->v_rdev->si_copyonwrite = 0; 1843 devvp->v_vflag &= ~VV_COPYONWRITE; 1844 VI_UNLOCK(devvp); 1845} 1846 1847/* 1848 * Check for need to copy block that is about to be written, 1849 * copying the block if necessary. 1850 */ 1851static int 1852ffs_copyonwrite(devvp, bp) 1853 struct vnode *devvp; 1854 struct buf *bp; 1855{ 1856 struct snaphead *snaphead; 1857 struct buf *ibp, *cbp, *savedcbp = 0; 1858 struct thread *td = curthread; 1859 struct fs *fs; 1860 struct inode *ip; 1861 struct vnode *vp = 0; 1862 ufs2_daddr_t lbn, blkno, *snapblklist; 1863 int lower, upper, mid, indiroff, snapshot_locked = 0, error = 0; 1864 1865 if (td->td_proc->p_flag & P_COWINPROGRESS) 1866 panic("ffs_copyonwrite: recursive call"); 1867 /* 1868 * First check to see if it is in the preallocated list. 1869 * By doing this check we avoid several potential deadlocks. 1870 */ 1871 VI_LOCK(devvp); 1872 snaphead = &devvp->v_rdev->si_snapshots; 1873 ip = TAILQ_FIRST(snaphead); 1874 fs = ip->i_fs; 1875 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); 1876 snapblklist = devvp->v_rdev->si_snapblklist; 1877 upper = devvp->v_rdev->si_snaplistsize - 1; 1878 lower = 1; 1879 while (lower <= upper) { 1880 mid = (lower + upper) / 2; 1881 if (snapblklist[mid] == lbn) 1882 break; 1883 if (snapblklist[mid] < lbn) 1884 lower = mid + 1; 1885 else 1886 upper = mid - 1; 1887 } 1888 if (lower <= upper) { 1889 VI_UNLOCK(devvp); 1890 return (0); 1891 } 1892 /* 1893 * Not in the precomputed list, so check the snapshots. 1894 */ 1895retry: 1896 TAILQ_FOREACH(ip, snaphead, i_nextsnap) { 1897 vp = ITOV(ip); 1898 /* 1899 * We ensure that everything of our own that needs to be 1900 * copied will be done at the time that ffs_snapshot is 1901 * called. Thus we can skip the check here which can 1902 * deadlock in doing the lookup in UFS_BALLOC. 1903 */ 1904 if (bp->b_vp == vp) 1905 continue; 1906 /* 1907 * Check to see if block needs to be copied. We do not have 1908 * to hold the snapshot lock while doing this lookup as it 1909 * will never require any additional allocations for the 1910 * snapshot inode. 1911 */ 1912 if (lbn < NDADDR) { 1913 blkno = DIP(ip, i_db[lbn]); 1914 } else { 1915 if (snapshot_locked == 0 && 1916 lockmgr(vp->v_vnlock, 1917 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 1918 VI_MTX(devvp), td) != 0) { 1919 VI_LOCK(devvp); 1920 goto retry; 1921 } 1922 snapshot_locked = 1; 1923 td->td_proc->p_flag |= P_COWINPROGRESS; 1924 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1925 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1926 td->td_proc->p_flag &= ~P_COWINPROGRESS; 1927 if (error) 1928 break; 1929 indiroff = (lbn - NDADDR) % NINDIR(fs); 1930 if (ip->i_ump->um_fstype == UFS1) 1931 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 1932 else 1933 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 1934 bqrelse(ibp); 1935 } 1936#ifdef DIAGNOSTIC 1937 if (blkno == BLK_SNAP && bp->b_lblkno >= 0) 1938 panic("ffs_copyonwrite: bad copy block"); 1939#endif 1940 if (blkno != 0) 1941 continue; 1942 /* 1943 * Allocate the block into which to do the copy. Since 1944 * multiple processes may all try to copy the same block, 1945 * we have to recheck our need to do a copy if we sleep 1946 * waiting for the lock. 1947 * 1948 * Because all snapshots on a filesystem share a single 1949 * lock, we ensure that we will never be in competition 1950 * with another process to allocate a block. 1951 */ 1952 if (snapshot_locked == 0 && 1953 lockmgr(vp->v_vnlock, 1954 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 1955 VI_MTX(devvp), td) != 0) { 1956 VI_LOCK(devvp); 1957 goto retry; 1958 } 1959 snapshot_locked = 1; 1960 td->td_proc->p_flag |= P_COWINPROGRESS; 1961 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1962 fs->fs_bsize, KERNCRED, 0, &cbp); 1963 td->td_proc->p_flag &= ~P_COWINPROGRESS; 1964 if (error) 1965 break; 1966#ifdef DEBUG 1967 if (snapdebug) { 1968 printf("Copyonwrite: snapino %d lbn %jd for ", 1969 ip->i_number, (intmax_t)lbn); 1970 if (bp->b_vp == devvp) 1971 printf("fs metadata"); 1972 else 1973 printf("inum %d", VTOI(bp->b_vp)->i_number); 1974 printf(" lblkno %jd to blkno %jd\n", 1975 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno); 1976 } 1977#endif 1978 /* 1979 * If we have already read the old block contents, then 1980 * simply copy them to the new block. Note that we need 1981 * to synchronously write snapshots that have not been 1982 * unlinked, and hence will be visible after a crash, 1983 * to ensure their integrity. 1984 */ 1985 if (savedcbp != 0) { 1986 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 1987 bawrite(cbp); 1988 if (dopersistence && ip->i_effnlink > 0) 1989 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 1990 continue; 1991 } 1992 /* 1993 * Otherwise, read the old block contents into the buffer. 1994 */ 1995 if ((error = readblock(cbp, lbn)) != 0) { 1996 bzero(cbp->b_data, fs->fs_bsize); 1997 bawrite(cbp); 1998 if (dopersistence && ip->i_effnlink > 0) 1999 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 2000 break; 2001 } 2002 savedcbp = cbp; 2003 } 2004 /* 2005 * Note that we need to synchronously write snapshots that 2006 * have not been unlinked, and hence will be visible after 2007 * a crash, to ensure their integrity. 2008 */ 2009 if (savedcbp) { 2010 vp = savedcbp->b_vp; 2011 bawrite(savedcbp); 2012 if (dopersistence && VTOI(vp)->i_effnlink > 0) 2013 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td); 2014 } 2015 if (snapshot_locked) 2016 VOP_UNLOCK(vp, 0, td); 2017 else 2018 VI_UNLOCK(devvp); 2019 return (error); 2020} 2021 2022/* 2023 * Read the specified block into the given buffer. 2024 * Much of this boiler-plate comes from bwrite(). 2025 */ 2026static int 2027readblock(bp, lbn) 2028 struct buf *bp; 2029 ufs2_daddr_t lbn; 2030{ 2031 struct uio auio; 2032 struct iovec aiov; 2033 struct thread *td = curthread; 2034 struct inode *ip = VTOI(bp->b_vp); 2035 2036 aiov.iov_base = bp->b_data; 2037 aiov.iov_len = bp->b_bcount; 2038 auio.uio_iov = &aiov; 2039 auio.uio_iovcnt = 1; 2040 auio.uio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn))); 2041 auio.uio_resid = bp->b_bcount; 2042 auio.uio_rw = UIO_READ; 2043 auio.uio_segflg = UIO_SYSSPACE; 2044 auio.uio_td = td; 2045 return (physio(ip->i_devvp->v_rdev, &auio, 0)); 2046} 2047