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 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include "opt_quota.h" 40 41 #include <sys/param.h> 42 #include <sys/kernel.h> 43 #include <sys/systm.h> 44 #include <sys/conf.h> 45 #include <sys/bio.h> 46 #include <sys/buf.h> 47 #include <sys/fcntl.h> 48 #include <sys/proc.h> 49 #include <sys/namei.h> 50 #include <sys/sched.h> 51 #include <sys/stat.h> 52 #include <sys/malloc.h> 53 #include <sys/mount.h> 54 #include <sys/resource.h> 55 #include <sys/resourcevar.h> 56 #include <sys/vnode.h> 57 58 #include <geom/geom.h> 59 60 #include <ufs/ufs/extattr.h> 61 #include <ufs/ufs/quota.h> 62 #include <ufs/ufs/ufsmount.h> 63 #include <ufs/ufs/inode.h> 64 #include <ufs/ufs/ufs_extern.h> 65 66 #include <ufs/ffs/fs.h> 67 #include <ufs/ffs/ffs_extern.h> 68 69 #define KERNCRED thread0.td_ucred 70 #define DEBUG 1 71 72 #include "opt_ffs.h" 73 74 #ifdef NO_FFS_SNAPSHOT 75 int 76 ffs_snapshot(mp, snapfile) 77 struct mount *mp; 78 char *snapfile; 79 { 80 return (EINVAL); 81 } 82 83 int 84 ffs_snapblkfree(fs, devvp, bno, size, inum) 85 struct fs *fs; 86 struct vnode *devvp; 87 ufs2_daddr_t bno; 88 long size; 89 ino_t inum; 90 { 91 return (EINVAL); 92 } 93 94 void 95 ffs_snapremove(vp) 96 struct vnode *vp; 97 { 98 } 99 100 void 101 ffs_snapshot_mount(mp) 102 struct mount *mp; 103 { 104 } 105 106 void 107 ffs_snapshot_unmount(mp) 108 struct mount *mp; 109 { 110 } 111 112 void 113 ffs_snapgone(ip) 114 struct inode *ip; 115 { 116 } 117 118 int 119 ffs_copyonwrite(devvp, bp) 120 struct vnode *devvp; 121 struct buf *bp; 122 { 123 return (EINVAL); 124 } 125 126 #else 127 128 TAILQ_HEAD(snaphead, inode); 129 130 struct snapdata { 131 LIST_ENTRY(snapdata) sn_link; 132 struct snaphead sn_head; 133 daddr_t sn_listsize; 134 daddr_t *sn_blklist; 135 struct lock sn_lock; 136 }; 137 138 LIST_HEAD(, snapdata) snapfree; 139 static struct mtx snapfree_lock; 140 MTX_SYSINIT(ffs_snapfree, &snapfree_lock, "snapdata free list", MTX_DEF); 141 142 static int cgaccount(int, struct vnode *, struct buf *, int); 143 static int expunge_ufs1(struct vnode *, struct inode *, struct fs *, 144 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, 145 ufs_lbn_t, int), int); 146 static int indiracct_ufs1(struct vnode *, struct vnode *, int, 147 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, 148 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, 149 ufs_lbn_t, int), int); 150 static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 151 struct fs *, ufs_lbn_t, int); 152 static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 153 struct fs *, ufs_lbn_t, int); 154 static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 155 struct fs *, ufs_lbn_t, int); 156 static int expunge_ufs2(struct vnode *, struct inode *, struct fs *, 157 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, 158 ufs_lbn_t, int), int); 159 static int indiracct_ufs2(struct vnode *, struct vnode *, int, 160 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, 161 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, 162 ufs_lbn_t, int), int); 163 static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 164 struct fs *, ufs_lbn_t, int); 165 static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 166 struct fs *, ufs_lbn_t, int); 167 static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 168 struct fs *, ufs_lbn_t, int); 169 static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t); 170 static void try_free_snapdata(struct vnode *devvp); 171 static struct snapdata *ffs_snapdata_acquire(struct vnode *devvp); 172 static int ffs_bp_snapblk(struct vnode *, struct buf *); 173 174 /* 175 * To ensure the consistency of snapshots across crashes, we must 176 * synchronously write out copied blocks before allowing the 177 * originals to be modified. Because of the rather severe speed 178 * penalty that this imposes, the following flag allows this 179 * crash persistence to be disabled. 180 */ 181 int dopersistence = 0; 182 183 #ifdef DEBUG 184 #include <sys/sysctl.h> 185 SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, ""); 186 static int snapdebug = 0; 187 SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, ""); 188 int collectsnapstats = 0; 189 SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats, 190 0, ""); 191 #endif /* DEBUG */ 192 193 /* 194 * Create a snapshot file and initialize it for the filesystem. 195 */ 196 int 197 ffs_snapshot(mp, snapfile) 198 struct mount *mp; 199 char *snapfile; 200 { 201 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist; 202 int error, cg, snaploc; 203 int i, size, len, loc; 204 int flag; 205 struct timespec starttime = {0, 0}, endtime; 206 char saved_nice = 0; 207 long redo = 0, snaplistsize = 0; 208 int32_t *lp; 209 void *space; 210 struct fs *copy_fs = NULL, *fs; 211 struct thread *td = curthread; 212 struct inode *ip, *xp; 213 struct buf *bp, *nbp, *ibp, *sbp = NULL; 214 struct nameidata nd; 215 struct mount *wrtmp; 216 struct vattr vat; 217 struct vnode *vp, *xvp, *mvp, *devvp; 218 struct uio auio; 219 struct iovec aiov; 220 struct snapdata *sn; 221 struct ufsmount *ump; 222 223 ump = VFSTOUFS(mp); 224 fs = ump->um_fs; 225 sn = NULL; 226 MNT_ILOCK(mp); 227 flag = mp->mnt_flag; 228 MNT_IUNLOCK(mp); 229 230 /* 231 * Need to serialize access to snapshot code per filesystem. 232 */ 233 /* 234 * Assign a snapshot slot in the superblock. 235 */ 236 UFS_LOCK(ump); 237 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 238 if (fs->fs_snapinum[snaploc] == 0) 239 break; 240 UFS_UNLOCK(ump); 241 if (snaploc == FSMAXSNAP) 242 return (ENOSPC); 243 /* 244 * Create the snapshot file. 245 */ 246 restart: 247 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_SYSSPACE, snapfile, td); 248 if ((error = namei(&nd)) != 0) 249 return (error); 250 if (nd.ni_vp != NULL) { 251 vput(nd.ni_vp); 252 error = EEXIST; 253 } 254 if (nd.ni_dvp->v_mount != mp) 255 error = EXDEV; 256 if (error) { 257 NDFREE(&nd, NDF_ONLY_PNBUF); 258 if (nd.ni_dvp == nd.ni_vp) 259 vrele(nd.ni_dvp); 260 else 261 vput(nd.ni_dvp); 262 return (error); 263 } 264 VATTR_NULL(&vat); 265 vat.va_type = VREG; 266 vat.va_mode = S_IRUSR; 267 vat.va_vaflags |= VA_EXCLUSIVE; 268 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp)) 269 wrtmp = NULL; 270 if (wrtmp != mp) 271 panic("ffs_snapshot: mount mismatch"); 272 vfs_rel(wrtmp); 273 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) { 274 NDFREE(&nd, NDF_ONLY_PNBUF); 275 vput(nd.ni_dvp); 276 if ((error = vn_start_write(NULL, &wrtmp, 277 V_XSLEEP | PCATCH)) != 0) 278 return (error); 279 goto restart; 280 } 281 VOP_LEASE(nd.ni_dvp, td, KERNCRED, LEASE_WRITE); 282 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat); 283 VOP_UNLOCK(nd.ni_dvp, 0); 284 if (error) { 285 NDFREE(&nd, NDF_ONLY_PNBUF); 286 vn_finished_write(wrtmp); 287 vrele(nd.ni_dvp); 288 return (error); 289 } 290 vp = nd.ni_vp; 291 vp->v_vflag |= VV_SYSTEM; 292 ip = VTOI(vp); 293 devvp = ip->i_devvp; 294 /* 295 * Allocate and copy the last block contents so as to be able 296 * to set size to that of the filesystem. 297 */ 298 numblks = howmany(fs->fs_size, fs->fs_frag); 299 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)), 300 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 301 if (error) 302 goto out; 303 ip->i_size = lblktosize(fs, (off_t)numblks); 304 DIP_SET(ip, i_size, ip->i_size); 305 ip->i_flag |= IN_CHANGE | IN_UPDATE; 306 error = readblock(vp, bp, numblks - 1); 307 bawrite(bp); 308 if (error != 0) 309 goto out; 310 /* 311 * Preallocate critical data structures so that we can copy 312 * them in without further allocation after we suspend all 313 * operations on the filesystem. We would like to just release 314 * the allocated buffers without writing them since they will 315 * be filled in below once we are ready to go, but this upsets 316 * the soft update code, so we go ahead and write the new buffers. 317 * 318 * Allocate all indirect blocks and mark all of them as not 319 * needing to be copied. 320 */ 321 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 322 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 323 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp); 324 if (error) 325 goto out; 326 bawrite(ibp); 327 } 328 /* 329 * Allocate copies for the superblock and its summary information. 330 */ 331 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED, 332 0, &nbp); 333 if (error) 334 goto out; 335 bawrite(nbp); 336 blkno = fragstoblks(fs, fs->fs_csaddr); 337 len = howmany(fs->fs_cssize, fs->fs_bsize); 338 for (loc = 0; loc < len; loc++) { 339 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)), 340 fs->fs_bsize, KERNCRED, 0, &nbp); 341 if (error) 342 goto out; 343 bawrite(nbp); 344 } 345 /* 346 * Allocate all cylinder group blocks. 347 */ 348 for (cg = 0; cg < fs->fs_ncg; cg++) { 349 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 350 fs->fs_bsize, KERNCRED, 0, &nbp); 351 if (error) 352 goto out; 353 bawrite(nbp); 354 if (cg % 10 == 0) 355 ffs_syncvnode(vp, MNT_WAIT); 356 } 357 /* 358 * Copy all the cylinder group maps. Although the 359 * filesystem is still active, we hope that only a few 360 * cylinder groups will change between now and when we 361 * suspend operations. Thus, we will be able to quickly 362 * touch up the few cylinder groups that changed during 363 * the suspension period. 364 */ 365 len = howmany(fs->fs_ncg, NBBY); 366 space = malloc(len, M_DEVBUF, M_WAITOK|M_ZERO); 367 UFS_LOCK(ump); 368 fs->fs_active = space; 369 UFS_UNLOCK(ump); 370 for (cg = 0; cg < fs->fs_ncg; cg++) { 371 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 372 fs->fs_bsize, KERNCRED, 0, &nbp); 373 if (error) 374 goto out; 375 error = cgaccount(cg, vp, nbp, 1); 376 bawrite(nbp); 377 if (cg % 10 == 0) 378 ffs_syncvnode(vp, MNT_WAIT); 379 if (error) 380 goto out; 381 } 382 /* 383 * Change inode to snapshot type file. 384 */ 385 ip->i_flags |= SF_SNAPSHOT; 386 DIP_SET(ip, i_flags, ip->i_flags); 387 ip->i_flag |= IN_CHANGE | IN_UPDATE; 388 /* 389 * Ensure that the snapshot is completely on disk. 390 * Since we have marked it as a snapshot it is safe to 391 * unlock it as no process will be allowed to write to it. 392 */ 393 if ((error = ffs_syncvnode(vp, MNT_WAIT)) != 0) 394 goto out; 395 VOP_UNLOCK(vp, 0); 396 /* 397 * All allocations are done, so we can now snapshot the system. 398 * 399 * Recind nice scheduling while running with the filesystem suspended. 400 */ 401 if (td->td_proc->p_nice > 0) { 402 struct proc *p; 403 404 p = td->td_proc; 405 PROC_LOCK(p); 406 saved_nice = p->p_nice; 407 sched_nice(p, 0); 408 PROC_UNLOCK(p); 409 } 410 /* 411 * Suspend operation on filesystem. 412 */ 413 for (;;) { 414 vn_finished_write(wrtmp); 415 if ((error = vfs_write_suspend(vp->v_mount)) != 0) { 416 vn_start_write(NULL, &wrtmp, V_WAIT); 417 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 418 goto out; 419 } 420 if (mp->mnt_kern_flag & MNTK_SUSPENDED) 421 break; 422 vn_start_write(NULL, &wrtmp, V_WAIT); 423 } 424 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 425 if (ip->i_effnlink == 0) { 426 error = ENOENT; /* Snapshot file unlinked */ 427 goto out1; 428 } 429 if (collectsnapstats) 430 nanotime(&starttime); 431 432 /* The last block might have changed. Copy it again to be sure. */ 433 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)), 434 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 435 if (error != 0) 436 goto out1; 437 error = readblock(vp, bp, numblks - 1); 438 bp->b_flags |= B_VALIDSUSPWRT; 439 bawrite(bp); 440 if (error != 0) 441 goto out1; 442 /* 443 * First, copy all the cylinder group maps that have changed. 444 */ 445 for (cg = 0; cg < fs->fs_ncg; cg++) { 446 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0) 447 continue; 448 redo++; 449 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 450 fs->fs_bsize, KERNCRED, 0, &nbp); 451 if (error) 452 goto out1; 453 error = cgaccount(cg, vp, nbp, 2); 454 bawrite(nbp); 455 if (error) 456 goto out1; 457 } 458 /* 459 * Grab a copy of the superblock and its summary information. 460 * We delay writing it until the suspension is released below. 461 */ 462 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize, 463 KERNCRED, &sbp); 464 if (error) { 465 brelse(sbp); 466 sbp = NULL; 467 goto out1; 468 } 469 loc = blkoff(fs, fs->fs_sblockloc); 470 copy_fs = (struct fs *)(sbp->b_data + loc); 471 bcopy(fs, copy_fs, fs->fs_sbsize); 472 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 473 copy_fs->fs_clean = 1; 474 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE; 475 if (fs->fs_sbsize < size) 476 bzero(&sbp->b_data[loc + fs->fs_sbsize], size - fs->fs_sbsize); 477 size = blkroundup(fs, fs->fs_cssize); 478 if (fs->fs_contigsumsize > 0) 479 size += fs->fs_ncg * sizeof(int32_t); 480 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 481 copy_fs->fs_csp = space; 482 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize); 483 space = (char *)space + fs->fs_cssize; 484 loc = howmany(fs->fs_cssize, fs->fs_fsize); 485 i = fs->fs_frag - loc % fs->fs_frag; 486 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize; 487 if (len > 0) { 488 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc), 489 len, KERNCRED, &bp)) != 0) { 490 brelse(bp); 491 free(copy_fs->fs_csp, M_UFSMNT); 492 bawrite(sbp); 493 sbp = NULL; 494 goto out1; 495 } 496 bcopy(bp->b_data, space, (u_int)len); 497 space = (char *)space + len; 498 bp->b_flags |= B_INVAL | B_NOCACHE; 499 brelse(bp); 500 } 501 if (fs->fs_contigsumsize > 0) { 502 copy_fs->fs_maxcluster = lp = space; 503 for (i = 0; i < fs->fs_ncg; i++) 504 *lp++ = fs->fs_contigsumsize; 505 } 506 /* 507 * We must check for active files that have been unlinked 508 * (e.g., with a zero link count). We have to expunge all 509 * trace of these files from the snapshot so that they are 510 * not reclaimed prematurely by fsck or unnecessarily dumped. 511 * We turn off the MNTK_SUSPENDED flag to avoid a panic from 512 * spec_strategy about writing on a suspended filesystem. 513 * Note that we skip unlinked snapshot files as they will 514 * be handled separately below. 515 * 516 * We also calculate the needed size for the snapshot list. 517 */ 518 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) + 519 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */; 520 MNT_ILOCK(mp); 521 mp->mnt_kern_flag &= ~MNTK_SUSPENDED; 522 loop: 523 MNT_VNODE_FOREACH(xvp, mp, mvp) { 524 VI_LOCK(xvp); 525 MNT_IUNLOCK(mp); 526 if ((xvp->v_iflag & VI_DOOMED) || 527 (xvp->v_usecount == 0 && 528 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) || 529 xvp->v_type == VNON || 530 (VTOI(xvp)->i_flags & SF_SNAPSHOT)) { 531 VI_UNLOCK(xvp); 532 MNT_ILOCK(mp); 533 continue; 534 } 535 /* 536 * We can skip parent directory vnode because it must have 537 * this snapshot file in it. 538 */ 539 if (xvp == nd.ni_dvp) { 540 VI_UNLOCK(xvp); 541 MNT_ILOCK(mp); 542 continue; 543 } 544 vholdl(xvp); 545 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK) != 0) { 546 MNT_ILOCK(mp); 547 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 548 vdrop(xvp); 549 goto loop; 550 } 551 VI_LOCK(xvp); 552 if (xvp->v_usecount == 0 && 553 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) { 554 VI_UNLOCK(xvp); 555 VOP_UNLOCK(xvp, 0); 556 vdrop(xvp); 557 MNT_ILOCK(mp); 558 continue; 559 } 560 VI_UNLOCK(xvp); 561 if (snapdebug) 562 vprint("ffs_snapshot: busy vnode", xvp); 563 if (VOP_GETATTR(xvp, &vat, td->td_ucred) == 0 && 564 vat.va_nlink > 0) { 565 VOP_UNLOCK(xvp, 0); 566 vdrop(xvp); 567 MNT_ILOCK(mp); 568 continue; 569 } 570 xp = VTOI(xvp); 571 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) { 572 VOP_UNLOCK(xvp, 0); 573 vdrop(xvp); 574 MNT_ILOCK(mp); 575 continue; 576 } 577 /* 578 * If there is a fragment, clear it here. 579 */ 580 blkno = 0; 581 loc = howmany(xp->i_size, fs->fs_bsize) - 1; 582 if (loc < NDADDR) { 583 len = fragroundup(fs, blkoff(fs, xp->i_size)); 584 if (len != 0 && len < fs->fs_bsize) { 585 ffs_blkfree(ump, copy_fs, vp, 586 DIP(xp, i_db[loc]), len, xp->i_number); 587 blkno = DIP(xp, i_db[loc]); 588 DIP_SET(xp, i_db[loc], 0); 589 } 590 } 591 snaplistsize += 1; 592 if (xp->i_ump->um_fstype == UFS1) 593 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, 594 BLK_NOCOPY); 595 else 596 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, 597 BLK_NOCOPY); 598 if (blkno) 599 DIP_SET(xp, i_db[loc], blkno); 600 if (!error) 601 error = ffs_freefile(ump, copy_fs, vp, xp->i_number, 602 xp->i_mode); 603 VOP_UNLOCK(xvp, 0); 604 vdrop(xvp); 605 if (error) { 606 free(copy_fs->fs_csp, M_UFSMNT); 607 bawrite(sbp); 608 sbp = NULL; 609 MNT_VNODE_FOREACH_ABORT(mp, mvp); 610 goto out1; 611 } 612 MNT_ILOCK(mp); 613 } 614 MNT_IUNLOCK(mp); 615 /* 616 * Acquire a lock on the snapdata structure, creating it if necessary. 617 */ 618 sn = ffs_snapdata_acquire(devvp); 619 /* 620 * Change vnode to use shared snapshot lock instead of the original 621 * private lock. 622 */ 623 vp->v_vnlock = &sn->sn_lock; 624 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 625 xp = TAILQ_FIRST(&sn->sn_head); 626 /* 627 * If this is the first snapshot on this filesystem, then we need 628 * to allocate the space for the list of preallocated snapshot blocks. 629 * This list will be refined below, but this preliminary one will 630 * keep us out of deadlock until the full one is ready. 631 */ 632 if (xp == NULL) { 633 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 634 M_UFSMNT, M_WAITOK); 635 blkp = &snapblklist[1]; 636 *blkp++ = lblkno(fs, fs->fs_sblockloc); 637 blkno = fragstoblks(fs, fs->fs_csaddr); 638 for (cg = 0; cg < fs->fs_ncg; cg++) { 639 if (fragstoblks(fs, cgtod(fs, cg) > blkno)) 640 break; 641 *blkp++ = fragstoblks(fs, cgtod(fs, cg)); 642 } 643 len = howmany(fs->fs_cssize, fs->fs_bsize); 644 for (loc = 0; loc < len; loc++) 645 *blkp++ = blkno + loc; 646 for (; cg < fs->fs_ncg; cg++) 647 *blkp++ = fragstoblks(fs, cgtod(fs, cg)); 648 snapblklist[0] = blkp - snapblklist; 649 VI_LOCK(devvp); 650 if (sn->sn_blklist != NULL) 651 panic("ffs_snapshot: non-empty list"); 652 sn->sn_blklist = snapblklist; 653 sn->sn_listsize = blkp - snapblklist; 654 VI_UNLOCK(devvp); 655 } 656 /* 657 * Record snapshot inode. Since this is the newest snapshot, 658 * it must be placed at the end of the list. 659 */ 660 VI_LOCK(devvp); 661 fs->fs_snapinum[snaploc] = ip->i_number; 662 if (ip->i_nextsnap.tqe_prev != 0) 663 panic("ffs_snapshot: %d already on list", ip->i_number); 664 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); 665 devvp->v_vflag |= VV_COPYONWRITE; 666 VI_UNLOCK(devvp); 667 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp"); 668 out1: 669 KASSERT((sn != NULL && sbp != NULL && error == 0) || 670 (sn == NULL && sbp == NULL && error != 0), 671 ("email phk@ and mckusick@")); 672 /* 673 * Resume operation on filesystem. 674 */ 675 vfs_write_resume(vp->v_mount); 676 vn_start_write(NULL, &wrtmp, V_WAIT); 677 if (collectsnapstats && starttime.tv_sec > 0) { 678 nanotime(&endtime); 679 timespecsub(&endtime, &starttime); 680 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n", 681 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec, 682 endtime.tv_nsec / 1000000, redo, fs->fs_ncg); 683 } 684 if (sbp == NULL) 685 goto out; 686 /* 687 * Copy allocation information from all the snapshots in 688 * this snapshot and then expunge them from its view. 689 */ 690 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) { 691 if (xp == ip) 692 break; 693 if (xp->i_ump->um_fstype == UFS1) 694 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1, 695 BLK_SNAP); 696 else 697 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2, 698 BLK_SNAP); 699 if (error == 0 && xp->i_effnlink == 0) { 700 error = ffs_freefile(ump, 701 copy_fs, 702 vp, 703 xp->i_number, 704 xp->i_mode); 705 } 706 if (error) { 707 fs->fs_snapinum[snaploc] = 0; 708 goto done; 709 } 710 } 711 /* 712 * Allocate space for the full list of preallocated snapshot blocks. 713 */ 714 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 715 M_UFSMNT, M_WAITOK); 716 ip->i_snapblklist = &snapblklist[1]; 717 /* 718 * Expunge the blocks used by the snapshots from the set of 719 * blocks marked as used in the snapshot bitmaps. Also, collect 720 * the list of allocated blocks in i_snapblklist. 721 */ 722 if (ip->i_ump->um_fstype == UFS1) 723 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, BLK_SNAP); 724 else 725 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, BLK_SNAP); 726 if (error) { 727 fs->fs_snapinum[snaploc] = 0; 728 free(snapblklist, M_UFSMNT); 729 goto done; 730 } 731 if (snaplistsize < ip->i_snapblklist - snapblklist) 732 panic("ffs_snapshot: list too small"); 733 snaplistsize = ip->i_snapblklist - snapblklist; 734 snapblklist[0] = snaplistsize; 735 ip->i_snapblklist = 0; 736 /* 737 * Write out the list of allocated blocks to the end of the snapshot. 738 */ 739 auio.uio_iov = &aiov; 740 auio.uio_iovcnt = 1; 741 aiov.iov_base = (void *)snapblklist; 742 aiov.iov_len = snaplistsize * sizeof(daddr_t); 743 auio.uio_resid = aiov.iov_len;; 744 auio.uio_offset = ip->i_size; 745 auio.uio_segflg = UIO_SYSSPACE; 746 auio.uio_rw = UIO_WRITE; 747 auio.uio_td = td; 748 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 749 fs->fs_snapinum[snaploc] = 0; 750 free(snapblklist, M_UFSMNT); 751 goto done; 752 } 753 /* 754 * Write the superblock and its summary information 755 * to the snapshot. 756 */ 757 blkno = fragstoblks(fs, fs->fs_csaddr); 758 len = howmany(fs->fs_cssize, fs->fs_bsize); 759 space = copy_fs->fs_csp; 760 for (loc = 0; loc < len; loc++) { 761 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp); 762 if (error) { 763 brelse(nbp); 764 fs->fs_snapinum[snaploc] = 0; 765 free(snapblklist, M_UFSMNT); 766 goto done; 767 } 768 bcopy(space, nbp->b_data, fs->fs_bsize); 769 space = (char *)space + fs->fs_bsize; 770 bawrite(nbp); 771 } 772 /* 773 * As this is the newest list, it is the most inclusive, so 774 * should replace the previous list. 775 */ 776 VI_LOCK(devvp); 777 space = sn->sn_blklist; 778 sn->sn_blklist = snapblklist; 779 sn->sn_listsize = snaplistsize; 780 VI_UNLOCK(devvp); 781 if (space != NULL) 782 free(space, M_UFSMNT); 783 /* 784 * If another process is currently writing the buffer containing 785 * the inode for this snapshot then a deadlock can occur. Drop 786 * the snapshot lock until the buffer has been written. 787 */ 788 VREF(vp); /* Protect against ffs_snapgone() */ 789 VOP_UNLOCK(vp, 0); 790 (void) bread(ip->i_devvp, 791 fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 792 (int) fs->fs_bsize, NOCRED, &nbp); 793 brelse(nbp); 794 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 795 if (ip->i_effnlink == 0) 796 error = ENOENT; /* Snapshot file unlinked */ 797 else 798 vrele(vp); /* Drop extra reference */ 799 done: 800 free(copy_fs->fs_csp, M_UFSMNT); 801 bawrite(sbp); 802 out: 803 NDFREE(&nd, NDF_ONLY_PNBUF); 804 if (saved_nice > 0) { 805 struct proc *p; 806 807 p = td->td_proc; 808 PROC_LOCK(p); 809 sched_nice(td->td_proc, saved_nice); 810 PROC_UNLOCK(td->td_proc); 811 } 812 UFS_LOCK(ump); 813 if (fs->fs_active != 0) { 814 free(fs->fs_active, M_DEVBUF); 815 fs->fs_active = 0; 816 } 817 UFS_UNLOCK(ump); 818 MNT_ILOCK(mp); 819 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 820 MNT_IUNLOCK(mp); 821 if (error) 822 (void) ffs_truncate(vp, (off_t)0, 0, NOCRED, td); 823 (void) ffs_syncvnode(vp, MNT_WAIT); 824 if (error) 825 vput(vp); 826 else 827 VOP_UNLOCK(vp, 0); 828 vrele(nd.ni_dvp); 829 vn_finished_write(wrtmp); 830 process_deferred_inactive(mp); 831 return (error); 832 } 833 834 /* 835 * Copy a cylinder group map. All the unallocated blocks are marked 836 * BLK_NOCOPY so that the snapshot knows that it need not copy them 837 * if they are later written. If passno is one, then this is a first 838 * pass, so only setting needs to be done. If passno is 2, then this 839 * is a revision to a previous pass which must be undone as the 840 * replacement pass is done. 841 */ 842 static int 843 cgaccount(cg, vp, nbp, passno) 844 int cg; 845 struct vnode *vp; 846 struct buf *nbp; 847 int passno; 848 { 849 struct buf *bp, *ibp; 850 struct inode *ip; 851 struct cg *cgp; 852 struct fs *fs; 853 ufs2_daddr_t base, numblks; 854 int error, len, loc, indiroff; 855 856 ip = VTOI(vp); 857 fs = ip->i_fs; 858 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 859 (int)fs->fs_cgsize, KERNCRED, &bp); 860 if (error) { 861 brelse(bp); 862 return (error); 863 } 864 cgp = (struct cg *)bp->b_data; 865 if (!cg_chkmagic(cgp)) { 866 brelse(bp); 867 return (EIO); 868 } 869 UFS_LOCK(ip->i_ump); 870 ACTIVESET(fs, cg); 871 /* 872 * Recomputation of summary information might not have been performed 873 * at mount time. Sync up summary information for current cylinder 874 * group while data is in memory to ensure that result of background 875 * fsck is slightly more consistent. 876 */ 877 fs->fs_cs(fs, cg) = cgp->cg_cs; 878 UFS_UNLOCK(ip->i_ump); 879 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize); 880 if (fs->fs_cgsize < fs->fs_bsize) 881 bzero(&nbp->b_data[fs->fs_cgsize], 882 fs->fs_bsize - fs->fs_cgsize); 883 cgp = (struct cg *)nbp->b_data; 884 bqrelse(bp); 885 if (passno == 2) 886 nbp->b_flags |= B_VALIDSUSPWRT; 887 numblks = howmany(fs->fs_size, fs->fs_frag); 888 len = howmany(fs->fs_fpg, fs->fs_frag); 889 base = cgbase(fs, cg) / fs->fs_frag; 890 if (base + len >= numblks) 891 len = numblks - base - 1; 892 loc = 0; 893 if (base < NDADDR) { 894 for ( ; loc < NDADDR; loc++) { 895 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 896 DIP_SET(ip, i_db[loc], BLK_NOCOPY); 897 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY) 898 DIP_SET(ip, i_db[loc], 0); 899 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY) 900 panic("ffs_snapshot: lost direct block"); 901 } 902 } 903 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)), 904 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 905 if (error) { 906 return (error); 907 } 908 indiroff = (base + loc - NDADDR) % NINDIR(fs); 909 for ( ; loc < len; loc++, indiroff++) { 910 if (indiroff >= NINDIR(fs)) { 911 if (passno == 2) 912 ibp->b_flags |= B_VALIDSUSPWRT; 913 bawrite(ibp); 914 error = UFS_BALLOC(vp, 915 lblktosize(fs, (off_t)(base + loc)), 916 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 917 if (error) { 918 return (error); 919 } 920 indiroff = 0; 921 } 922 if (ip->i_ump->um_fstype == UFS1) { 923 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 924 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 925 BLK_NOCOPY; 926 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data)) 927 [indiroff] == BLK_NOCOPY) 928 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0; 929 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data)) 930 [indiroff] == BLK_NOCOPY) 931 panic("ffs_snapshot: lost indirect block"); 932 continue; 933 } 934 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 935 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; 936 else if (passno == 2 && 937 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 938 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0; 939 else if (passno == 1 && 940 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 941 panic("ffs_snapshot: lost indirect block"); 942 } 943 if (passno == 2) 944 ibp->b_flags |= B_VALIDSUSPWRT; 945 bdwrite(ibp); 946 return (0); 947 } 948 949 /* 950 * Before expunging a snapshot inode, note all the 951 * blocks that it claims with BLK_SNAP so that fsck will 952 * be able to account for those blocks properly and so 953 * that this snapshot knows that it need not copy them 954 * if the other snapshot holding them is freed. This code 955 * is reproduced once each for UFS1 and UFS2. 956 */ 957 static int 958 expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype) 959 struct vnode *snapvp; 960 struct inode *cancelip; 961 struct fs *fs; 962 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 963 struct fs *, ufs_lbn_t, int); 964 int expungetype; 965 { 966 int i, error, indiroff; 967 ufs_lbn_t lbn, rlbn; 968 ufs2_daddr_t len, blkno, numblks, blksperindir; 969 struct ufs1_dinode *dip; 970 struct thread *td = curthread; 971 struct buf *bp; 972 973 /* 974 * Prepare to expunge the inode. If its inode block has not 975 * yet been copied, then allocate and fill the copy. 976 */ 977 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 978 blkno = 0; 979 if (lbn < NDADDR) { 980 blkno = VTOI(snapvp)->i_din1->di_db[lbn]; 981 } else { 982 td->td_pflags |= TDP_COWINPROGRESS; 983 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), 984 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 985 td->td_pflags &= ~TDP_COWINPROGRESS; 986 if (error) 987 return (error); 988 indiroff = (lbn - NDADDR) % NINDIR(fs); 989 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; 990 bqrelse(bp); 991 } 992 if (blkno != 0) { 993 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 994 return (error); 995 } else { 996 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), 997 fs->fs_bsize, KERNCRED, 0, &bp); 998 if (error) 999 return (error); 1000 if ((error = readblock(snapvp, bp, lbn)) != 0) 1001 return (error); 1002 } 1003 /* 1004 * Set a snapshot inode to be a zero length file, regular files 1005 * or unlinked snapshots to be completely unallocated. 1006 */ 1007 dip = (struct ufs1_dinode *)bp->b_data + 1008 ino_to_fsbo(fs, cancelip->i_number); 1009 if (expungetype == BLK_NOCOPY || cancelip->i_effnlink == 0) 1010 dip->di_mode = 0; 1011 dip->di_size = 0; 1012 dip->di_blocks = 0; 1013 dip->di_flags &= ~SF_SNAPSHOT; 1014 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t)); 1015 bdwrite(bp); 1016 /* 1017 * Now go through and expunge all the blocks in the file 1018 * using the function requested. 1019 */ 1020 numblks = howmany(cancelip->i_size, fs->fs_bsize); 1021 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0], 1022 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype))) 1023 return (error); 1024 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0], 1025 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype))) 1026 return (error); 1027 blksperindir = 1; 1028 lbn = -NDADDR; 1029 len = numblks - NDADDR; 1030 rlbn = NDADDR; 1031 for (i = 0; len > 0 && i < NIADDR; i++) { 1032 error = indiracct_ufs1(snapvp, ITOV(cancelip), i, 1033 cancelip->i_din1->di_ib[i], lbn, rlbn, len, 1034 blksperindir, fs, acctfunc, expungetype); 1035 if (error) 1036 return (error); 1037 blksperindir *= NINDIR(fs); 1038 lbn -= blksperindir + 1; 1039 len -= blksperindir; 1040 rlbn += blksperindir; 1041 } 1042 return (0); 1043 } 1044 1045 /* 1046 * Descend an indirect block chain for vnode cancelvp accounting for all 1047 * its indirect blocks in snapvp. 1048 */ 1049 static int 1050 indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 1051 blksperindir, fs, acctfunc, expungetype) 1052 struct vnode *snapvp; 1053 struct vnode *cancelvp; 1054 int level; 1055 ufs1_daddr_t blkno; 1056 ufs_lbn_t lbn; 1057 ufs_lbn_t rlbn; 1058 ufs_lbn_t remblks; 1059 ufs_lbn_t blksperindir; 1060 struct fs *fs; 1061 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 1062 struct fs *, ufs_lbn_t, int); 1063 int expungetype; 1064 { 1065 int error, num, i; 1066 ufs_lbn_t subblksperindir; 1067 struct indir indirs[NIADDR + 2]; 1068 ufs1_daddr_t last, *bap; 1069 struct buf *bp; 1070 1071 if (blkno == 0) { 1072 if (expungetype == BLK_NOCOPY) 1073 return (0); 1074 panic("indiracct_ufs1: missing indir"); 1075 } 1076 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 1077 return (error); 1078 if (lbn != indirs[num - 1 - level].in_lbn || num < 2) 1079 panic("indiracct_ufs1: botched params"); 1080 /* 1081 * We have to expand bread here since it will deadlock looking 1082 * up the block number for any blocks that are not in the cache. 1083 */ 1084 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); 1085 bp->b_blkno = fsbtodb(fs, blkno); 1086 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && 1087 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { 1088 brelse(bp); 1089 return (error); 1090 } 1091 /* 1092 * Account for the block pointers in this indirect block. 1093 */ 1094 last = howmany(remblks, blksperindir); 1095 if (last > NINDIR(fs)) 1096 last = NINDIR(fs); 1097 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); 1098 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 1099 bqrelse(bp); 1100 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 1101 level == 0 ? rlbn : -1, expungetype); 1102 if (error || level == 0) 1103 goto out; 1104 /* 1105 * Account for the block pointers in each of the indirect blocks 1106 * in the levels below us. 1107 */ 1108 subblksperindir = blksperindir / NINDIR(fs); 1109 for (lbn++, level--, i = 0; i < last; i++) { 1110 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn, 1111 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 1112 if (error) 1113 goto out; 1114 rlbn += blksperindir; 1115 lbn -= blksperindir; 1116 remblks -= blksperindir; 1117 } 1118 out: 1119 free(bap, M_DEVBUF); 1120 return (error); 1121 } 1122 1123 /* 1124 * Do both snap accounting and map accounting. 1125 */ 1126 static int 1127 fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype) 1128 struct vnode *vp; 1129 ufs1_daddr_t *oldblkp, *lastblkp; 1130 struct fs *fs; 1131 ufs_lbn_t lblkno; 1132 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 1133 { 1134 int error; 1135 1136 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 1137 return (error); 1138 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 1139 } 1140 1141 /* 1142 * Identify a set of blocks allocated in a snapshot inode. 1143 */ 1144 static int 1145 snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1146 struct vnode *vp; 1147 ufs1_daddr_t *oldblkp, *lastblkp; 1148 struct fs *fs; 1149 ufs_lbn_t lblkno; 1150 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 1151 { 1152 struct inode *ip = VTOI(vp); 1153 ufs1_daddr_t blkno, *blkp; 1154 ufs_lbn_t lbn; 1155 struct buf *ibp; 1156 int error; 1157 1158 for ( ; oldblkp < lastblkp; oldblkp++) { 1159 blkno = *oldblkp; 1160 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 1161 continue; 1162 lbn = fragstoblks(fs, blkno); 1163 if (lbn < NDADDR) { 1164 blkp = &ip->i_din1->di_db[lbn]; 1165 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1166 } else { 1167 error = ffs_balloc_ufs1(vp, lblktosize(fs, (off_t)lbn), 1168 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1169 if (error) 1170 return (error); 1171 blkp = &((ufs1_daddr_t *)(ibp->b_data)) 1172 [(lbn - NDADDR) % NINDIR(fs)]; 1173 } 1174 /* 1175 * If we are expunging a snapshot vnode and we 1176 * find a block marked BLK_NOCOPY, then it is 1177 * one that has been allocated to this snapshot after 1178 * we took our current snapshot and can be ignored. 1179 */ 1180 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 1181 if (lbn >= NDADDR) 1182 brelse(ibp); 1183 } else { 1184 if (*blkp != 0) 1185 panic("snapacct_ufs1: bad block"); 1186 *blkp = expungetype; 1187 if (lbn >= NDADDR) 1188 bdwrite(ibp); 1189 } 1190 } 1191 return (0); 1192 } 1193 1194 /* 1195 * Account for a set of blocks allocated in a snapshot inode. 1196 */ 1197 static int 1198 mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1199 struct vnode *vp; 1200 ufs1_daddr_t *oldblkp, *lastblkp; 1201 struct fs *fs; 1202 ufs_lbn_t lblkno; 1203 int expungetype; 1204 { 1205 ufs1_daddr_t blkno; 1206 struct inode *ip; 1207 ino_t inum; 1208 int acctit; 1209 1210 ip = VTOI(vp); 1211 inum = ip->i_number; 1212 if (lblkno == -1) 1213 acctit = 0; 1214 else 1215 acctit = 1; 1216 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 1217 blkno = *oldblkp; 1218 if (blkno == 0 || blkno == BLK_NOCOPY) 1219 continue; 1220 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP) 1221 *ip->i_snapblklist++ = lblkno; 1222 if (blkno == BLK_SNAP) 1223 blkno = blkstofrags(fs, lblkno); 1224 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum); 1225 } 1226 return (0); 1227 } 1228 1229 /* 1230 * Before expunging a snapshot inode, note all the 1231 * blocks that it claims with BLK_SNAP so that fsck will 1232 * be able to account for those blocks properly and so 1233 * that this snapshot knows that it need not copy them 1234 * if the other snapshot holding them is freed. This code 1235 * is reproduced once each for UFS1 and UFS2. 1236 */ 1237 static int 1238 expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype) 1239 struct vnode *snapvp; 1240 struct inode *cancelip; 1241 struct fs *fs; 1242 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1243 struct fs *, ufs_lbn_t, int); 1244 int expungetype; 1245 { 1246 int i, error, indiroff; 1247 ufs_lbn_t lbn, rlbn; 1248 ufs2_daddr_t len, blkno, numblks, blksperindir; 1249 struct ufs2_dinode *dip; 1250 struct thread *td = curthread; 1251 struct buf *bp; 1252 1253 /* 1254 * Prepare to expunge the inode. If its inode block has not 1255 * yet been copied, then allocate and fill the copy. 1256 */ 1257 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 1258 blkno = 0; 1259 if (lbn < NDADDR) { 1260 blkno = VTOI(snapvp)->i_din2->di_db[lbn]; 1261 } else { 1262 td->td_pflags |= TDP_COWINPROGRESS; 1263 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), 1264 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 1265 td->td_pflags &= ~TDP_COWINPROGRESS; 1266 if (error) 1267 return (error); 1268 indiroff = (lbn - NDADDR) % NINDIR(fs); 1269 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; 1270 bqrelse(bp); 1271 } 1272 if (blkno != 0) { 1273 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 1274 return (error); 1275 } else { 1276 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), 1277 fs->fs_bsize, KERNCRED, 0, &bp); 1278 if (error) 1279 return (error); 1280 if ((error = readblock(snapvp, bp, lbn)) != 0) 1281 return (error); 1282 } 1283 /* 1284 * Set a snapshot inode to be a zero length file, regular files 1285 * to be completely unallocated. 1286 */ 1287 dip = (struct ufs2_dinode *)bp->b_data + 1288 ino_to_fsbo(fs, cancelip->i_number); 1289 if (expungetype == BLK_NOCOPY) 1290 dip->di_mode = 0; 1291 dip->di_size = 0; 1292 dip->di_blocks = 0; 1293 dip->di_flags &= ~SF_SNAPSHOT; 1294 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t)); 1295 bdwrite(bp); 1296 /* 1297 * Now go through and expunge all the blocks in the file 1298 * using the function requested. 1299 */ 1300 numblks = howmany(cancelip->i_size, fs->fs_bsize); 1301 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0], 1302 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype))) 1303 return (error); 1304 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0], 1305 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype))) 1306 return (error); 1307 blksperindir = 1; 1308 lbn = -NDADDR; 1309 len = numblks - NDADDR; 1310 rlbn = NDADDR; 1311 for (i = 0; len > 0 && i < NIADDR; i++) { 1312 error = indiracct_ufs2(snapvp, ITOV(cancelip), i, 1313 cancelip->i_din2->di_ib[i], lbn, rlbn, len, 1314 blksperindir, fs, acctfunc, expungetype); 1315 if (error) 1316 return (error); 1317 blksperindir *= NINDIR(fs); 1318 lbn -= blksperindir + 1; 1319 len -= blksperindir; 1320 rlbn += blksperindir; 1321 } 1322 return (0); 1323 } 1324 1325 /* 1326 * Descend an indirect block chain for vnode cancelvp accounting for all 1327 * its indirect blocks in snapvp. 1328 */ 1329 static int 1330 indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 1331 blksperindir, fs, acctfunc, expungetype) 1332 struct vnode *snapvp; 1333 struct vnode *cancelvp; 1334 int level; 1335 ufs2_daddr_t blkno; 1336 ufs_lbn_t lbn; 1337 ufs_lbn_t rlbn; 1338 ufs_lbn_t remblks; 1339 ufs_lbn_t blksperindir; 1340 struct fs *fs; 1341 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1342 struct fs *, ufs_lbn_t, int); 1343 int expungetype; 1344 { 1345 int error, num, i; 1346 ufs_lbn_t subblksperindir; 1347 struct indir indirs[NIADDR + 2]; 1348 ufs2_daddr_t last, *bap; 1349 struct buf *bp; 1350 1351 if (blkno == 0) { 1352 if (expungetype == BLK_NOCOPY) 1353 return (0); 1354 panic("indiracct_ufs2: missing indir"); 1355 } 1356 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 1357 return (error); 1358 if (lbn != indirs[num - 1 - level].in_lbn || num < 2) 1359 panic("indiracct_ufs2: botched params"); 1360 /* 1361 * We have to expand bread here since it will deadlock looking 1362 * up the block number for any blocks that are not in the cache. 1363 */ 1364 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); 1365 bp->b_blkno = fsbtodb(fs, blkno); 1366 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && 1367 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { 1368 brelse(bp); 1369 return (error); 1370 } 1371 /* 1372 * Account for the block pointers in this indirect block. 1373 */ 1374 last = howmany(remblks, blksperindir); 1375 if (last > NINDIR(fs)) 1376 last = NINDIR(fs); 1377 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); 1378 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 1379 bqrelse(bp); 1380 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 1381 level == 0 ? rlbn : -1, expungetype); 1382 if (error || level == 0) 1383 goto out; 1384 /* 1385 * Account for the block pointers in each of the indirect blocks 1386 * in the levels below us. 1387 */ 1388 subblksperindir = blksperindir / NINDIR(fs); 1389 for (lbn++, level--, i = 0; i < last; i++) { 1390 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn, 1391 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 1392 if (error) 1393 goto out; 1394 rlbn += blksperindir; 1395 lbn -= blksperindir; 1396 remblks -= blksperindir; 1397 } 1398 out: 1399 free(bap, M_DEVBUF); 1400 return (error); 1401 } 1402 1403 /* 1404 * Do both snap accounting and map accounting. 1405 */ 1406 static int 1407 fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype) 1408 struct vnode *vp; 1409 ufs2_daddr_t *oldblkp, *lastblkp; 1410 struct fs *fs; 1411 ufs_lbn_t lblkno; 1412 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 1413 { 1414 int error; 1415 1416 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 1417 return (error); 1418 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 1419 } 1420 1421 /* 1422 * Identify a set of blocks allocated in a snapshot inode. 1423 */ 1424 static int 1425 snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1426 struct vnode *vp; 1427 ufs2_daddr_t *oldblkp, *lastblkp; 1428 struct fs *fs; 1429 ufs_lbn_t lblkno; 1430 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 1431 { 1432 struct inode *ip = VTOI(vp); 1433 ufs2_daddr_t blkno, *blkp; 1434 ufs_lbn_t lbn; 1435 struct buf *ibp; 1436 int error; 1437 1438 for ( ; oldblkp < lastblkp; oldblkp++) { 1439 blkno = *oldblkp; 1440 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 1441 continue; 1442 lbn = fragstoblks(fs, blkno); 1443 if (lbn < NDADDR) { 1444 blkp = &ip->i_din2->di_db[lbn]; 1445 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1446 } else { 1447 error = ffs_balloc_ufs2(vp, lblktosize(fs, (off_t)lbn), 1448 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1449 if (error) 1450 return (error); 1451 blkp = &((ufs2_daddr_t *)(ibp->b_data)) 1452 [(lbn - NDADDR) % NINDIR(fs)]; 1453 } 1454 /* 1455 * If we are expunging a snapshot vnode and we 1456 * find a block marked BLK_NOCOPY, then it is 1457 * one that has been allocated to this snapshot after 1458 * we took our current snapshot and can be ignored. 1459 */ 1460 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 1461 if (lbn >= NDADDR) 1462 brelse(ibp); 1463 } else { 1464 if (*blkp != 0) 1465 panic("snapacct_ufs2: bad block"); 1466 *blkp = expungetype; 1467 if (lbn >= NDADDR) 1468 bdwrite(ibp); 1469 } 1470 } 1471 return (0); 1472 } 1473 1474 /* 1475 * Account for a set of blocks allocated in a snapshot inode. 1476 */ 1477 static int 1478 mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1479 struct vnode *vp; 1480 ufs2_daddr_t *oldblkp, *lastblkp; 1481 struct fs *fs; 1482 ufs_lbn_t lblkno; 1483 int expungetype; 1484 { 1485 ufs2_daddr_t blkno; 1486 struct inode *ip; 1487 ino_t inum; 1488 int acctit; 1489 1490 ip = VTOI(vp); 1491 inum = ip->i_number; 1492 if (lblkno == -1) 1493 acctit = 0; 1494 else 1495 acctit = 1; 1496 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 1497 blkno = *oldblkp; 1498 if (blkno == 0 || blkno == BLK_NOCOPY) 1499 continue; 1500 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP) 1501 *ip->i_snapblklist++ = lblkno; 1502 if (blkno == BLK_SNAP) 1503 blkno = blkstofrags(fs, lblkno); 1504 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum); 1505 } 1506 return (0); 1507 } 1508 1509 /* 1510 * Decrement extra reference on snapshot when last name is removed. 1511 * It will not be freed until the last open reference goes away. 1512 */ 1513 void 1514 ffs_snapgone(ip) 1515 struct inode *ip; 1516 { 1517 struct inode *xp; 1518 struct fs *fs; 1519 int snaploc; 1520 struct snapdata *sn; 1521 struct ufsmount *ump; 1522 1523 /* 1524 * Find snapshot in incore list. 1525 */ 1526 xp = NULL; 1527 sn = ip->i_devvp->v_rdev->si_snapdata; 1528 if (sn != NULL) 1529 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) 1530 if (xp == ip) 1531 break; 1532 if (xp != NULL) 1533 vrele(ITOV(ip)); 1534 else if (snapdebug) 1535 printf("ffs_snapgone: lost snapshot vnode %d\n", 1536 ip->i_number); 1537 /* 1538 * Delete snapshot inode from superblock. Keep list dense. 1539 */ 1540 fs = ip->i_fs; 1541 ump = ip->i_ump; 1542 UFS_LOCK(ump); 1543 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 1544 if (fs->fs_snapinum[snaploc] == ip->i_number) 1545 break; 1546 if (snaploc < FSMAXSNAP) { 1547 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) { 1548 if (fs->fs_snapinum[snaploc] == 0) 1549 break; 1550 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc]; 1551 } 1552 fs->fs_snapinum[snaploc - 1] = 0; 1553 } 1554 UFS_UNLOCK(ump); 1555 } 1556 1557 /* 1558 * Prepare a snapshot file for being removed. 1559 */ 1560 void 1561 ffs_snapremove(vp) 1562 struct vnode *vp; 1563 { 1564 struct inode *ip; 1565 struct vnode *devvp; 1566 struct buf *ibp; 1567 struct fs *fs; 1568 ufs2_daddr_t numblks, blkno, dblk; 1569 int error, loc, last; 1570 struct snapdata *sn; 1571 1572 ip = VTOI(vp); 1573 fs = ip->i_fs; 1574 devvp = ip->i_devvp; 1575 /* 1576 * If active, delete from incore list (this snapshot may 1577 * already have been in the process of being deleted, so 1578 * would not have been active). 1579 * 1580 * Clear copy-on-write flag if last snapshot. 1581 */ 1582 VI_LOCK(devvp); 1583 if (ip->i_nextsnap.tqe_prev != 0) { 1584 sn = devvp->v_rdev->si_snapdata; 1585 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap); 1586 ip->i_nextsnap.tqe_prev = 0; 1587 VI_UNLOCK(devvp); 1588 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); 1589 KASSERT(vp->v_vnlock == &sn->sn_lock, 1590 ("ffs_snapremove: lost lock mutation")); 1591 vp->v_vnlock = &vp->v_lock; 1592 VI_LOCK(devvp); 1593 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 1594 try_free_snapdata(devvp); 1595 } else 1596 VI_UNLOCK(devvp); 1597 /* 1598 * Clear all BLK_NOCOPY fields. Pass any block claims to other 1599 * snapshots that want them (see ffs_snapblkfree below). 1600 */ 1601 for (blkno = 1; blkno < NDADDR; blkno++) { 1602 dblk = DIP(ip, i_db[blkno]); 1603 if (dblk == 0) 1604 continue; 1605 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1606 DIP_SET(ip, i_db[blkno], 0); 1607 else if ((dblk == blkstofrags(fs, blkno) && 1608 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize, 1609 ip->i_number))) { 1610 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - 1611 btodb(fs->fs_bsize)); 1612 DIP_SET(ip, i_db[blkno], 0); 1613 } 1614 } 1615 numblks = howmany(ip->i_size, fs->fs_bsize); 1616 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 1617 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 1618 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1619 if (error) 1620 continue; 1621 if (fs->fs_size - blkno > NINDIR(fs)) 1622 last = NINDIR(fs); 1623 else 1624 last = fs->fs_size - blkno; 1625 for (loc = 0; loc < last; loc++) { 1626 if (ip->i_ump->um_fstype == UFS1) { 1627 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc]; 1628 if (dblk == 0) 1629 continue; 1630 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1631 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1632 else if ((dblk == blkstofrags(fs, blkno) && 1633 ffs_snapblkfree(fs, ip->i_devvp, dblk, 1634 fs->fs_bsize, ip->i_number))) { 1635 ip->i_din1->di_blocks -= 1636 btodb(fs->fs_bsize); 1637 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1638 } 1639 continue; 1640 } 1641 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc]; 1642 if (dblk == 0) 1643 continue; 1644 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1645 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1646 else if ((dblk == blkstofrags(fs, blkno) && 1647 ffs_snapblkfree(fs, ip->i_devvp, dblk, 1648 fs->fs_bsize, ip->i_number))) { 1649 ip->i_din2->di_blocks -= btodb(fs->fs_bsize); 1650 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1651 } 1652 } 1653 bawrite(ibp); 1654 } 1655 /* 1656 * Clear snapshot flag and drop reference. 1657 */ 1658 ip->i_flags &= ~SF_SNAPSHOT; 1659 DIP_SET(ip, i_flags, ip->i_flags); 1660 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1661 #ifdef QUOTA 1662 /* 1663 * Reenable disk quotas for ex-snapshot file. 1664 */ 1665 if (!getinoquota(ip)) 1666 (void) chkdq(ip, DIP(ip, i_blocks), KERNCRED, FORCE); 1667 #endif 1668 } 1669 1670 /* 1671 * Notification that a block is being freed. Return zero if the free 1672 * should be allowed to proceed. Return non-zero if the snapshot file 1673 * wants to claim the block. The block will be claimed if it is an 1674 * uncopied part of one of the snapshots. It will be freed if it is 1675 * either a BLK_NOCOPY or has already been copied in all of the snapshots. 1676 * If a fragment is being freed, then all snapshots that care about 1677 * it must make a copy since a snapshot file can only claim full sized 1678 * blocks. Note that if more than one snapshot file maps the block, 1679 * we can pick one at random to claim it. Since none of the snapshots 1680 * can change, we are assurred that they will all see the same unmodified 1681 * image. When deleting a snapshot file (see ffs_snapremove above), we 1682 * must push any of these claimed blocks to one of the other snapshots 1683 * that maps it. These claimed blocks are easily identified as they will 1684 * have a block number equal to their logical block number within the 1685 * snapshot. A copied block can never have this property because they 1686 * must always have been allocated from a BLK_NOCOPY location. 1687 */ 1688 int 1689 ffs_snapblkfree(fs, devvp, bno, size, inum) 1690 struct fs *fs; 1691 struct vnode *devvp; 1692 ufs2_daddr_t bno; 1693 long size; 1694 ino_t inum; 1695 { 1696 struct buf *ibp, *cbp, *savedcbp = 0; 1697 struct thread *td = curthread; 1698 struct inode *ip; 1699 struct vnode *vp = NULL; 1700 ufs_lbn_t lbn; 1701 ufs2_daddr_t blkno; 1702 int indiroff = 0, error = 0, claimedblk = 0; 1703 struct snapdata *sn; 1704 1705 lbn = fragstoblks(fs, bno); 1706 retry: 1707 VI_LOCK(devvp); 1708 sn = devvp->v_rdev->si_snapdata; 1709 if (sn == NULL) { 1710 VI_UNLOCK(devvp); 1711 return (0); 1712 } 1713 if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 1714 VI_MTX(devvp)) != 0) 1715 goto retry; 1716 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 1717 vp = ITOV(ip); 1718 /* 1719 * Lookup block being written. 1720 */ 1721 if (lbn < NDADDR) { 1722 blkno = DIP(ip, i_db[lbn]); 1723 } else { 1724 td->td_pflags |= TDP_COWINPROGRESS; 1725 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1726 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1727 td->td_pflags &= ~TDP_COWINPROGRESS; 1728 if (error) 1729 break; 1730 indiroff = (lbn - NDADDR) % NINDIR(fs); 1731 if (ip->i_ump->um_fstype == UFS1) 1732 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 1733 else 1734 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 1735 } 1736 /* 1737 * Check to see if block needs to be copied. 1738 */ 1739 if (blkno == 0) { 1740 /* 1741 * A block that we map is being freed. If it has not 1742 * been claimed yet, we will claim or copy it (below). 1743 */ 1744 claimedblk = 1; 1745 } else if (blkno == BLK_SNAP) { 1746 /* 1747 * No previous snapshot claimed the block, 1748 * so it will be freed and become a BLK_NOCOPY 1749 * (don't care) for us. 1750 */ 1751 if (claimedblk) 1752 panic("snapblkfree: inconsistent block type"); 1753 if (lbn < NDADDR) { 1754 DIP_SET(ip, i_db[lbn], BLK_NOCOPY); 1755 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1756 } else if (ip->i_ump->um_fstype == UFS1) { 1757 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 1758 BLK_NOCOPY; 1759 bdwrite(ibp); 1760 } else { 1761 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 1762 BLK_NOCOPY; 1763 bdwrite(ibp); 1764 } 1765 continue; 1766 } else /* BLK_NOCOPY or default */ { 1767 /* 1768 * If the snapshot has already copied the block 1769 * (default), or does not care about the block, 1770 * it is not needed. 1771 */ 1772 if (lbn >= NDADDR) 1773 bqrelse(ibp); 1774 continue; 1775 } 1776 /* 1777 * If this is a full size block, we will just grab it 1778 * and assign it to the snapshot inode. Otherwise we 1779 * will proceed to copy it. See explanation for this 1780 * routine as to why only a single snapshot needs to 1781 * claim this block. 1782 */ 1783 if (size == fs->fs_bsize) { 1784 #ifdef DEBUG 1785 if (snapdebug) 1786 printf("%s %d lbn %jd from inum %d\n", 1787 "Grabonremove: snapino", ip->i_number, 1788 (intmax_t)lbn, inum); 1789 #endif 1790 if (lbn < NDADDR) { 1791 DIP_SET(ip, i_db[lbn], bno); 1792 } else if (ip->i_ump->um_fstype == UFS1) { 1793 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno; 1794 bdwrite(ibp); 1795 } else { 1796 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno; 1797 bdwrite(ibp); 1798 } 1799 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size)); 1800 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1801 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 1802 return (1); 1803 } 1804 if (lbn >= NDADDR) 1805 bqrelse(ibp); 1806 /* 1807 * Allocate the block into which to do the copy. Note that this 1808 * allocation will never require any additional allocations for 1809 * the snapshot inode. 1810 */ 1811 td->td_pflags |= TDP_COWINPROGRESS; 1812 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1813 fs->fs_bsize, KERNCRED, 0, &cbp); 1814 td->td_pflags &= ~TDP_COWINPROGRESS; 1815 if (error) 1816 break; 1817 #ifdef DEBUG 1818 if (snapdebug) 1819 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n", 1820 "Copyonremove: snapino ", ip->i_number, 1821 (intmax_t)lbn, "for inum", inum, size, 1822 (intmax_t)cbp->b_blkno); 1823 #endif 1824 /* 1825 * If we have already read the old block contents, then 1826 * simply copy them to the new block. Note that we need 1827 * to synchronously write snapshots that have not been 1828 * unlinked, and hence will be visible after a crash, 1829 * to ensure their integrity. 1830 */ 1831 if (savedcbp != 0) { 1832 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 1833 bawrite(cbp); 1834 if (dopersistence && ip->i_effnlink > 0) 1835 (void) ffs_syncvnode(vp, MNT_WAIT); 1836 continue; 1837 } 1838 /* 1839 * Otherwise, read the old block contents into the buffer. 1840 */ 1841 if ((error = readblock(vp, cbp, lbn)) != 0) { 1842 bzero(cbp->b_data, fs->fs_bsize); 1843 bawrite(cbp); 1844 if (dopersistence && ip->i_effnlink > 0) 1845 (void) ffs_syncvnode(vp, MNT_WAIT); 1846 break; 1847 } 1848 savedcbp = cbp; 1849 } 1850 /* 1851 * Note that we need to synchronously write snapshots that 1852 * have not been unlinked, and hence will be visible after 1853 * a crash, to ensure their integrity. 1854 */ 1855 if (savedcbp) { 1856 vp = savedcbp->b_vp; 1857 bawrite(savedcbp); 1858 if (dopersistence && VTOI(vp)->i_effnlink > 0) 1859 (void) ffs_syncvnode(vp, MNT_WAIT); 1860 } 1861 /* 1862 * If we have been unable to allocate a block in which to do 1863 * the copy, then return non-zero so that the fragment will 1864 * not be freed. Although space will be lost, the snapshot 1865 * will stay consistent. 1866 */ 1867 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 1868 return (error); 1869 } 1870 1871 /* 1872 * Associate snapshot files when mounting. 1873 */ 1874 void 1875 ffs_snapshot_mount(mp) 1876 struct mount *mp; 1877 { 1878 struct ufsmount *ump = VFSTOUFS(mp); 1879 struct vnode *devvp = ump->um_devvp; 1880 struct fs *fs = ump->um_fs; 1881 struct thread *td = curthread; 1882 struct snapdata *sn; 1883 struct vnode *vp; 1884 struct vnode *lastvp; 1885 struct inode *ip; 1886 struct uio auio; 1887 struct iovec aiov; 1888 void *snapblklist; 1889 char *reason; 1890 daddr_t snaplistsize; 1891 int error, snaploc, loc; 1892 1893 /* 1894 * XXX The following needs to be set before ffs_truncate or 1895 * VOP_READ can be called. 1896 */ 1897 mp->mnt_stat.f_iosize = fs->fs_bsize; 1898 /* 1899 * Process each snapshot listed in the superblock. 1900 */ 1901 vp = NULL; 1902 lastvp = NULL; 1903 sn = NULL; 1904 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) { 1905 if (fs->fs_snapinum[snaploc] == 0) 1906 break; 1907 if ((error = ffs_vget(mp, fs->fs_snapinum[snaploc], 1908 LK_EXCLUSIVE, &vp)) != 0){ 1909 printf("ffs_snapshot_mount: vget failed %d\n", error); 1910 continue; 1911 } 1912 ip = VTOI(vp); 1913 if ((ip->i_flags & SF_SNAPSHOT) == 0 || ip->i_size == 1914 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) { 1915 if ((ip->i_flags & SF_SNAPSHOT) == 0) { 1916 reason = "non-snapshot"; 1917 } else { 1918 reason = "old format snapshot"; 1919 (void)ffs_truncate(vp, (off_t)0, 0, NOCRED, td); 1920 (void)ffs_syncvnode(vp, MNT_WAIT); 1921 } 1922 printf("ffs_snapshot_mount: %s inode %d\n", 1923 reason, fs->fs_snapinum[snaploc]); 1924 vput(vp); 1925 vp = NULL; 1926 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) { 1927 if (fs->fs_snapinum[loc] == 0) 1928 break; 1929 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc]; 1930 } 1931 fs->fs_snapinum[loc - 1] = 0; 1932 snaploc--; 1933 continue; 1934 } 1935 /* 1936 * Acquire a lock on the snapdata structure, creating it if 1937 * necessary. 1938 */ 1939 sn = ffs_snapdata_acquire(devvp); 1940 /* 1941 * Change vnode to use shared snapshot lock instead of the 1942 * original private lock. 1943 */ 1944 vp->v_vnlock = &sn->sn_lock; 1945 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 1946 /* 1947 * Link it onto the active snapshot list. 1948 */ 1949 VI_LOCK(devvp); 1950 if (ip->i_nextsnap.tqe_prev != 0) 1951 panic("ffs_snapshot_mount: %d already on list", 1952 ip->i_number); 1953 else 1954 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); 1955 vp->v_vflag |= VV_SYSTEM; 1956 VI_UNLOCK(devvp); 1957 VOP_UNLOCK(vp, 0); 1958 lastvp = vp; 1959 } 1960 vp = lastvp; 1961 /* 1962 * No usable snapshots found. 1963 */ 1964 if (sn == NULL || vp == NULL) 1965 return; 1966 /* 1967 * Allocate the space for the block hints list. We always want to 1968 * use the list from the newest snapshot. 1969 */ 1970 auio.uio_iov = &aiov; 1971 auio.uio_iovcnt = 1; 1972 aiov.iov_base = (void *)&snaplistsize; 1973 aiov.iov_len = sizeof(snaplistsize); 1974 auio.uio_resid = aiov.iov_len; 1975 auio.uio_offset = 1976 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)); 1977 auio.uio_segflg = UIO_SYSSPACE; 1978 auio.uio_rw = UIO_READ; 1979 auio.uio_td = td; 1980 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1981 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 1982 printf("ffs_snapshot_mount: read_1 failed %d\n", error); 1983 VOP_UNLOCK(vp, 0); 1984 return; 1985 } 1986 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 1987 M_UFSMNT, M_WAITOK); 1988 auio.uio_iovcnt = 1; 1989 aiov.iov_base = snapblklist; 1990 aiov.iov_len = snaplistsize * sizeof (daddr_t); 1991 auio.uio_resid = aiov.iov_len; 1992 auio.uio_offset -= sizeof(snaplistsize); 1993 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 1994 printf("ffs_snapshot_mount: read_2 failed %d\n", error); 1995 VOP_UNLOCK(vp, 0); 1996 free(snapblklist, M_UFSMNT); 1997 return; 1998 } 1999 VOP_UNLOCK(vp, 0); 2000 VI_LOCK(devvp); 2001 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount"); 2002 sn->sn_listsize = snaplistsize; 2003 sn->sn_blklist = (daddr_t *)snapblklist; 2004 devvp->v_vflag |= VV_COPYONWRITE; 2005 VI_UNLOCK(devvp); 2006 } 2007 2008 /* 2009 * Disassociate snapshot files when unmounting. 2010 */ 2011 void 2012 ffs_snapshot_unmount(mp) 2013 struct mount *mp; 2014 { 2015 struct vnode *devvp = VFSTOUFS(mp)->um_devvp; 2016 struct snapdata *sn; 2017 struct inode *xp; 2018 struct vnode *vp; 2019 2020 VI_LOCK(devvp); 2021 sn = devvp->v_rdev->si_snapdata; 2022 while (sn != NULL && (xp = TAILQ_FIRST(&sn->sn_head)) != NULL) { 2023 vp = ITOV(xp); 2024 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap); 2025 xp->i_nextsnap.tqe_prev = 0; 2026 lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE, 2027 VI_MTX(devvp)); 2028 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); 2029 KASSERT(vp->v_vnlock == &sn->sn_lock, 2030 ("ffs_snapshot_unmount: lost lock mutation")); 2031 vp->v_vnlock = &vp->v_lock; 2032 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 2033 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2034 if (xp->i_effnlink > 0) 2035 vrele(vp); 2036 VI_LOCK(devvp); 2037 sn = devvp->v_rdev->si_snapdata; 2038 } 2039 try_free_snapdata(devvp); 2040 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount"); 2041 } 2042 2043 /* 2044 * Check the buffer block to be belong to device buffer that shall be 2045 * locked after snaplk. devvp shall be locked on entry, and will be 2046 * leaved locked upon exit. 2047 */ 2048 static int 2049 ffs_bp_snapblk(devvp, bp) 2050 struct vnode *devvp; 2051 struct buf *bp; 2052 { 2053 struct snapdata *sn; 2054 struct fs *fs; 2055 ufs2_daddr_t lbn, *snapblklist; 2056 int lower, upper, mid; 2057 2058 ASSERT_VI_LOCKED(devvp, "ffs_bp_snapblk"); 2059 KASSERT(devvp->v_type == VCHR, ("Not a device %p", devvp)); 2060 sn = devvp->v_rdev->si_snapdata; 2061 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) == NULL) 2062 return (0); 2063 fs = TAILQ_FIRST(&sn->sn_head)->i_fs; 2064 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); 2065 snapblklist = sn->sn_blklist; 2066 upper = sn->sn_listsize - 1; 2067 lower = 1; 2068 while (lower <= upper) { 2069 mid = (lower + upper) / 2; 2070 if (snapblklist[mid] == lbn) 2071 break; 2072 if (snapblklist[mid] < lbn) 2073 lower = mid + 1; 2074 else 2075 upper = mid - 1; 2076 } 2077 if (lower <= upper) 2078 return (1); 2079 return (0); 2080 } 2081 2082 void 2083 ffs_bdflush(bo, bp) 2084 struct bufobj *bo; 2085 struct buf *bp; 2086 { 2087 struct thread *td; 2088 struct vnode *vp, *devvp; 2089 struct buf *nbp; 2090 int bp_bdskip; 2091 2092 if (bo->bo_dirty.bv_cnt <= dirtybufthresh) 2093 return; 2094 2095 td = curthread; 2096 vp = bp->b_vp; 2097 devvp = bo->__bo_vnode; 2098 KASSERT(vp == devvp, ("devvp != vp %p %p", bo, bp)); 2099 2100 VI_LOCK(devvp); 2101 bp_bdskip = ffs_bp_snapblk(devvp, bp); 2102 if (bp_bdskip) 2103 bdwriteskip++; 2104 VI_UNLOCK(devvp); 2105 if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10 && !bp_bdskip) { 2106 (void) VOP_FSYNC(vp, MNT_NOWAIT, td); 2107 altbufferflushes++; 2108 } else { 2109 BO_LOCK(bo); 2110 /* 2111 * Try to find a buffer to flush. 2112 */ 2113 TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { 2114 if ((nbp->b_vflags & BV_BKGRDINPROG) || 2115 BUF_LOCK(nbp, 2116 LK_EXCLUSIVE | LK_NOWAIT, NULL)) 2117 continue; 2118 if (bp == nbp) 2119 panic("bdwrite: found ourselves"); 2120 BO_UNLOCK(bo); 2121 /* 2122 * Don't countdeps with the bo lock 2123 * held. 2124 */ 2125 if (buf_countdeps(nbp, 0)) { 2126 BO_LOCK(bo); 2127 BUF_UNLOCK(nbp); 2128 continue; 2129 } 2130 if (bp_bdskip) { 2131 VI_LOCK(devvp); 2132 if (!ffs_bp_snapblk(vp, nbp)) { 2133 if (BO_MTX(bo) != VI_MTX(vp)) { 2134 VI_UNLOCK(devvp); 2135 BO_LOCK(bo); 2136 } 2137 BUF_UNLOCK(nbp); 2138 continue; 2139 } 2140 VI_UNLOCK(devvp); 2141 } 2142 if (nbp->b_flags & B_CLUSTEROK) { 2143 vfs_bio_awrite(nbp); 2144 } else { 2145 bremfree(nbp); 2146 bawrite(nbp); 2147 } 2148 dirtybufferflushes++; 2149 break; 2150 } 2151 if (nbp == NULL) 2152 BO_UNLOCK(bo); 2153 } 2154 } 2155 2156 /* 2157 * Check for need to copy block that is about to be written, 2158 * copying the block if necessary. 2159 */ 2160 int 2161 ffs_copyonwrite(devvp, bp) 2162 struct vnode *devvp; 2163 struct buf *bp; 2164 { 2165 struct snapdata *sn; 2166 struct buf *ibp, *cbp, *savedcbp = 0; 2167 struct thread *td = curthread; 2168 struct fs *fs; 2169 struct inode *ip; 2170 struct vnode *vp = 0; 2171 ufs2_daddr_t lbn, blkno, *snapblklist; 2172 int lower, upper, mid, indiroff, error = 0; 2173 int launched_async_io, prev_norunningbuf; 2174 long saved_runningbufspace; 2175 2176 if (devvp != bp->b_vp && (VTOI(bp->b_vp)->i_flags & SF_SNAPSHOT) != 0) 2177 return (0); /* Update on a snapshot file */ 2178 if (td->td_pflags & TDP_COWINPROGRESS) 2179 panic("ffs_copyonwrite: recursive call"); 2180 /* 2181 * First check to see if it is in the preallocated list. 2182 * By doing this check we avoid several potential deadlocks. 2183 */ 2184 VI_LOCK(devvp); 2185 sn = devvp->v_rdev->si_snapdata; 2186 if (sn == NULL || 2187 TAILQ_EMPTY(&sn->sn_head)) { 2188 VI_UNLOCK(devvp); 2189 return (0); /* No snapshot */ 2190 } 2191 ip = TAILQ_FIRST(&sn->sn_head); 2192 fs = ip->i_fs; 2193 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); 2194 snapblklist = sn->sn_blklist; 2195 upper = sn->sn_listsize - 1; 2196 lower = 1; 2197 while (lower <= upper) { 2198 mid = (lower + upper) / 2; 2199 if (snapblklist[mid] == lbn) 2200 break; 2201 if (snapblklist[mid] < lbn) 2202 lower = mid + 1; 2203 else 2204 upper = mid - 1; 2205 } 2206 if (lower <= upper) { 2207 VI_UNLOCK(devvp); 2208 return (0); 2209 } 2210 launched_async_io = 0; 2211 prev_norunningbuf = td->td_pflags & TDP_NORUNNINGBUF; 2212 /* 2213 * Since I/O on bp isn't yet in progress and it may be blocked 2214 * for a long time waiting on snaplk, back it out of 2215 * runningbufspace, possibly waking other threads waiting for space. 2216 */ 2217 saved_runningbufspace = bp->b_runningbufspace; 2218 if (saved_runningbufspace != 0) 2219 runningbufwakeup(bp); 2220 /* 2221 * Not in the precomputed list, so check the snapshots. 2222 */ 2223 while (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 2224 VI_MTX(devvp)) != 0) { 2225 VI_LOCK(devvp); 2226 sn = devvp->v_rdev->si_snapdata; 2227 if (sn == NULL || 2228 TAILQ_EMPTY(&sn->sn_head)) { 2229 VI_UNLOCK(devvp); 2230 if (saved_runningbufspace != 0) { 2231 bp->b_runningbufspace = saved_runningbufspace; 2232 atomic_add_int(&runningbufspace, 2233 bp->b_runningbufspace); 2234 } 2235 return (0); /* Snapshot gone */ 2236 } 2237 } 2238 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 2239 vp = ITOV(ip); 2240 /* 2241 * We ensure that everything of our own that needs to be 2242 * copied will be done at the time that ffs_snapshot is 2243 * called. Thus we can skip the check here which can 2244 * deadlock in doing the lookup in UFS_BALLOC. 2245 */ 2246 if (bp->b_vp == vp) 2247 continue; 2248 /* 2249 * Check to see if block needs to be copied. We do not have 2250 * to hold the snapshot lock while doing this lookup as it 2251 * will never require any additional allocations for the 2252 * snapshot inode. 2253 */ 2254 if (lbn < NDADDR) { 2255 blkno = DIP(ip, i_db[lbn]); 2256 } else { 2257 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; 2258 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 2259 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 2260 td->td_pflags &= ~TDP_COWINPROGRESS; 2261 if (error) 2262 break; 2263 indiroff = (lbn - NDADDR) % NINDIR(fs); 2264 if (ip->i_ump->um_fstype == UFS1) 2265 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 2266 else 2267 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 2268 bqrelse(ibp); 2269 } 2270 #ifdef INVARIANTS 2271 if (blkno == BLK_SNAP && bp->b_lblkno >= 0) 2272 panic("ffs_copyonwrite: bad copy block"); 2273 #endif 2274 if (blkno != 0) 2275 continue; 2276 /* 2277 * Allocate the block into which to do the copy. Since 2278 * multiple processes may all try to copy the same block, 2279 * we have to recheck our need to do a copy if we sleep 2280 * waiting for the lock. 2281 * 2282 * Because all snapshots on a filesystem share a single 2283 * lock, we ensure that we will never be in competition 2284 * with another process to allocate a block. 2285 */ 2286 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; 2287 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 2288 fs->fs_bsize, KERNCRED, 0, &cbp); 2289 td->td_pflags &= ~TDP_COWINPROGRESS; 2290 if (error) 2291 break; 2292 #ifdef DEBUG 2293 if (snapdebug) { 2294 printf("Copyonwrite: snapino %d lbn %jd for ", 2295 ip->i_number, (intmax_t)lbn); 2296 if (bp->b_vp == devvp) 2297 printf("fs metadata"); 2298 else 2299 printf("inum %d", VTOI(bp->b_vp)->i_number); 2300 printf(" lblkno %jd to blkno %jd\n", 2301 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno); 2302 } 2303 #endif 2304 /* 2305 * If we have already read the old block contents, then 2306 * simply copy them to the new block. Note that we need 2307 * to synchronously write snapshots that have not been 2308 * unlinked, and hence will be visible after a crash, 2309 * to ensure their integrity. 2310 */ 2311 if (savedcbp != 0) { 2312 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 2313 bawrite(cbp); 2314 if (dopersistence && ip->i_effnlink > 0) 2315 (void) ffs_syncvnode(vp, MNT_WAIT); 2316 else 2317 launched_async_io = 1; 2318 continue; 2319 } 2320 /* 2321 * Otherwise, read the old block contents into the buffer. 2322 */ 2323 if ((error = readblock(vp, cbp, lbn)) != 0) { 2324 bzero(cbp->b_data, fs->fs_bsize); 2325 bawrite(cbp); 2326 if (dopersistence && ip->i_effnlink > 0) 2327 (void) ffs_syncvnode(vp, MNT_WAIT); 2328 else 2329 launched_async_io = 1; 2330 break; 2331 } 2332 savedcbp = cbp; 2333 } 2334 /* 2335 * Note that we need to synchronously write snapshots that 2336 * have not been unlinked, and hence will be visible after 2337 * a crash, to ensure their integrity. 2338 */ 2339 if (savedcbp) { 2340 vp = savedcbp->b_vp; 2341 bawrite(savedcbp); 2342 if (dopersistence && VTOI(vp)->i_effnlink > 0) 2343 (void) ffs_syncvnode(vp, MNT_WAIT); 2344 else 2345 launched_async_io = 1; 2346 } 2347 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 2348 td->td_pflags = (td->td_pflags & ~TDP_NORUNNINGBUF) | 2349 prev_norunningbuf; 2350 if (launched_async_io && (td->td_pflags & TDP_NORUNNINGBUF) == 0) 2351 waitrunningbufspace(); 2352 /* 2353 * I/O on bp will now be started, so count it in runningbufspace. 2354 */ 2355 if (saved_runningbufspace != 0) { 2356 bp->b_runningbufspace = saved_runningbufspace; 2357 atomic_add_int(&runningbufspace, bp->b_runningbufspace); 2358 } 2359 return (error); 2360 } 2361 2362 /* 2363 * Read the specified block into the given buffer. 2364 * Much of this boiler-plate comes from bwrite(). 2365 */ 2366 static int 2367 readblock(vp, bp, lbn) 2368 struct vnode *vp; 2369 struct buf *bp; 2370 ufs2_daddr_t lbn; 2371 { 2372 struct inode *ip = VTOI(vp); 2373 struct bio *bip; 2374 2375 bip = g_alloc_bio(); 2376 bip->bio_cmd = BIO_READ; 2377 bip->bio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn))); 2378 bip->bio_data = bp->b_data; 2379 bip->bio_length = bp->b_bcount; 2380 bip->bio_done = NULL; 2381 2382 g_io_request(bip, ip->i_devvp->v_bufobj.bo_private); 2383 bp->b_error = biowait(bip, "snaprdb"); 2384 g_destroy_bio(bip); 2385 return (bp->b_error); 2386 } 2387 2388 #endif 2389 2390 /* 2391 * Process file deletes that were deferred by ufs_inactive() due to 2392 * the file system being suspended. Transfer IN_LAZYACCESS into 2393 * IN_MODIFIED for vnodes that were accessed during suspension. 2394 */ 2395 void 2396 process_deferred_inactive(struct mount *mp) 2397 { 2398 struct vnode *vp, *mvp; 2399 struct inode *ip; 2400 struct thread *td; 2401 int error; 2402 2403 td = curthread; 2404 (void) vn_start_secondary_write(NULL, &mp, V_WAIT); 2405 MNT_ILOCK(mp); 2406 loop: 2407 MNT_VNODE_FOREACH(vp, mp, mvp) { 2408 VI_LOCK(vp); 2409 /* 2410 * IN_LAZYACCESS is checked here without holding any 2411 * vnode lock, but this flag is set only while holding 2412 * vnode interlock. 2413 */ 2414 if (vp->v_type == VNON || (vp->v_iflag & VI_DOOMED) != 0 || 2415 ((VTOI(vp)->i_flag & IN_LAZYACCESS) == 0 && 2416 ((vp->v_iflag & VI_OWEINACT) == 0 || 2417 vp->v_usecount > 0))) { 2418 VI_UNLOCK(vp); 2419 continue; 2420 } 2421 MNT_IUNLOCK(mp); 2422 vholdl(vp); 2423 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2424 if (error != 0) { 2425 vdrop(vp); 2426 MNT_ILOCK(mp); 2427 if (error == ENOENT) 2428 continue; /* vnode recycled */ 2429 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 2430 goto loop; 2431 } 2432 ip = VTOI(vp); 2433 if ((ip->i_flag & IN_LAZYACCESS) != 0) { 2434 ip->i_flag &= ~IN_LAZYACCESS; 2435 ip->i_flag |= IN_MODIFIED; 2436 } 2437 VI_LOCK(vp); 2438 if ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0) { 2439 VI_UNLOCK(vp); 2440 VOP_UNLOCK(vp, 0); 2441 vdrop(vp); 2442 MNT_ILOCK(mp); 2443 continue; 2444 } 2445 2446 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2447 ("process_deferred_inactive: " 2448 "recursed on VI_DOINGINACT")); 2449 vp->v_iflag |= VI_DOINGINACT; 2450 vp->v_iflag &= ~VI_OWEINACT; 2451 VI_UNLOCK(vp); 2452 (void) VOP_INACTIVE(vp, td); 2453 VI_LOCK(vp); 2454 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2455 ("process_deferred_inactive: lost VI_DOINGINACT")); 2456 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, 2457 ("process_deferred_inactive: got VI_OWEINACT")); 2458 vp->v_iflag &= ~VI_DOINGINACT; 2459 VI_UNLOCK(vp); 2460 VOP_UNLOCK(vp, 0); 2461 vdrop(vp); 2462 MNT_ILOCK(mp); 2463 } 2464 MNT_IUNLOCK(mp); 2465 vn_finished_secondary_write(mp); 2466 } 2467 2468 #ifndef NO_FFS_SNAPSHOT 2469 2470 static struct snapdata * 2471 ffs_snapdata_alloc(void) 2472 { 2473 struct snapdata *sn; 2474 2475 /* 2476 * Fetch a snapdata from the free list if there is one available. 2477 */ 2478 mtx_lock(&snapfree_lock); 2479 sn = LIST_FIRST(&snapfree); 2480 if (sn != NULL) 2481 LIST_REMOVE(sn, sn_link); 2482 mtx_unlock(&snapfree_lock); 2483 if (sn != NULL) 2484 return (sn); 2485 /* 2486 * If there were no free snapdatas allocate one. 2487 */ 2488 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO); 2489 TAILQ_INIT(&sn->sn_head); 2490 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT, 2491 LK_CANRECURSE | LK_NOSHARE); 2492 return (sn); 2493 } 2494 2495 /* 2496 * The snapdata is never freed because we can not be certain that 2497 * there are no threads sleeping on the snap lock. Persisting 2498 * them permanently avoids costly synchronization in ffs_lock(). 2499 */ 2500 static void 2501 ffs_snapdata_free(struct snapdata *sn) 2502 { 2503 mtx_lock(&snapfree_lock); 2504 LIST_INSERT_HEAD(&snapfree, sn, sn_link); 2505 mtx_unlock(&snapfree_lock); 2506 } 2507 2508 /* Try to free snapdata associated with devvp */ 2509 static void 2510 try_free_snapdata(struct vnode *devvp) 2511 { 2512 struct snapdata *sn; 2513 ufs2_daddr_t *snapblklist; 2514 2515 ASSERT_VI_LOCKED(devvp, "try_free_snapdata"); 2516 sn = devvp->v_rdev->si_snapdata; 2517 2518 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) != NULL || 2519 (devvp->v_vflag & VV_COPYONWRITE) == 0) { 2520 VI_UNLOCK(devvp); 2521 return; 2522 } 2523 2524 devvp->v_rdev->si_snapdata = NULL; 2525 devvp->v_vflag &= ~VV_COPYONWRITE; 2526 lockmgr(&sn->sn_lock, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp)); 2527 snapblklist = sn->sn_blklist; 2528 sn->sn_blklist = NULL; 2529 sn->sn_listsize = 0; 2530 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2531 if (snapblklist != NULL) 2532 free(snapblklist, M_UFSMNT); 2533 ffs_snapdata_free(sn); 2534 } 2535 2536 static struct snapdata * 2537 ffs_snapdata_acquire(struct vnode *devvp) 2538 { 2539 struct snapdata *nsn; 2540 struct snapdata *sn; 2541 2542 /* 2543 * Allocate a free snapdata. This is done before acquiring the 2544 * devvp lock to avoid allocation while the devvp interlock is 2545 * held. 2546 */ 2547 nsn = ffs_snapdata_alloc(); 2548 /* 2549 * If there snapshots already exist on this filesystem grab a 2550 * reference to the shared lock. Otherwise this is the first 2551 * snapshot on this filesystem and we need to use our 2552 * pre-allocated snapdata. 2553 */ 2554 VI_LOCK(devvp); 2555 if (devvp->v_rdev->si_snapdata == NULL) { 2556 devvp->v_rdev->si_snapdata = nsn; 2557 nsn = NULL; 2558 } 2559 sn = devvp->v_rdev->si_snapdata; 2560 /* 2561 * Acquire the snapshot lock. 2562 */ 2563 lockmgr(&sn->sn_lock, 2564 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, VI_MTX(devvp)); 2565 /* 2566 * Free any unused snapdata. 2567 */ 2568 if (nsn != NULL) 2569 ffs_snapdata_free(nsn); 2570 2571 return (sn); 2572 } 2573 2574 #endif 2575