1/*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright 2000 Marshall Kirk McKusick. All Rights Reserved. 5 * 6 * Further information about snapshots can be obtained from: 7 * 8 * Marshall Kirk McKusick http://www.mckusick.com/softdep/ 9 * 1614 Oxford Street mckusick@mckusick.com 10 * Berkeley, CA 94709-1608 +1-510-843-9542 11 * USA 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 23 * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY 24 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 25 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 26 * DISCLAIMED. IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR 27 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)ffs_snapshot.c 8.11 (McKusick) 7/23/00 36 */ 37 38#include <sys/cdefs.h> 39__FBSDID("$FreeBSD$"); 40 41#include "opt_quota.h" 42 43#include <sys/param.h> 44#include <sys/kernel.h> 45#include <sys/systm.h> 46#include <sys/conf.h> 47#include <sys/bio.h> 48#include <sys/buf.h> 49#include <sys/fcntl.h> 50#include <sys/proc.h> 51#include <sys/namei.h> 52#include <sys/sched.h> 53#include <sys/stat.h> 54#include <sys/malloc.h> 55#include <sys/mount.h> 56#include <sys/resource.h> 57#include <sys/resourcevar.h> 58#include <sys/rwlock.h> 59#include <sys/vnode.h> 60 61#include <vm/vm.h> 62#include <vm/vm_extern.h> 63 64#include <geom/geom.h> 65 66#include <ufs/ufs/extattr.h> 67#include <ufs/ufs/quota.h> 68#include <ufs/ufs/ufsmount.h> 69#include <ufs/ufs/inode.h> 70#include <ufs/ufs/ufs_extern.h> 71 72#include <ufs/ffs/fs.h> 73#include <ufs/ffs/ffs_extern.h> 74 75#define KERNCRED thread0.td_ucred 76#define DEBUG 1 77 78#include "opt_ffs.h" 79 80#ifdef NO_FFS_SNAPSHOT 81int 82ffs_snapshot(mp, snapfile) 83 struct mount *mp; 84 char *snapfile; 85{ 86 return (EINVAL); 87} 88 89int 90ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd) 91 struct fs *fs; 92 struct vnode *devvp; 93 ufs2_daddr_t bno; 94 long size; 95 ino_t inum; 96 enum vtype vtype; 97 struct workhead *wkhd; 98{ 99 return (EINVAL); 100} 101 102void 103ffs_snapremove(vp) 104 struct vnode *vp; 105{ 106} 107 108void 109ffs_snapshot_mount(mp) 110 struct mount *mp; 111{ 112} 113 114void 115ffs_snapshot_unmount(mp) 116 struct mount *mp; 117{ 118} 119 120void 121ffs_snapgone(ip) 122 struct inode *ip; 123{ 124} 125 126int 127ffs_copyonwrite(devvp, bp) 128 struct vnode *devvp; 129 struct buf *bp; 130{ 131 return (EINVAL); 132} 133 134void 135ffs_sync_snap(mp, waitfor) 136 struct mount *mp; 137 int waitfor; 138{ 139} 140 141#else 142FEATURE(ffs_snapshot, "FFS snapshot support"); 143 144LIST_HEAD(, snapdata) snapfree; 145static struct mtx snapfree_lock; 146MTX_SYSINIT(ffs_snapfree, &snapfree_lock, "snapdata free list", MTX_DEF); 147 148static int cgaccount(int, struct vnode *, struct buf *, int); 149static int expunge_ufs1(struct vnode *, struct inode *, struct fs *, 150 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, 151 ufs_lbn_t, int), int, int); 152static int indiracct_ufs1(struct vnode *, struct vnode *, int, 153 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, 154 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, 155 ufs_lbn_t, int), int); 156static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 157 struct fs *, ufs_lbn_t, int); 158static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 159 struct fs *, ufs_lbn_t, int); 160static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 161 struct fs *, ufs_lbn_t, int); 162static int expunge_ufs2(struct vnode *, struct inode *, struct fs *, 163 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, 164 ufs_lbn_t, int), int, int); 165static int indiracct_ufs2(struct vnode *, struct vnode *, int, 166 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, 167 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, 168 ufs_lbn_t, int), int); 169static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 170 struct fs *, ufs_lbn_t, int); 171static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 172 struct fs *, ufs_lbn_t, int); 173static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 174 struct fs *, ufs_lbn_t, int); 175static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t); 176static void try_free_snapdata(struct vnode *devvp); 177static struct snapdata *ffs_snapdata_acquire(struct vnode *devvp); 178static int ffs_bp_snapblk(struct vnode *, struct buf *); 179 180/* 181 * To ensure the consistency of snapshots across crashes, we must 182 * synchronously write out copied blocks before allowing the 183 * originals to be modified. Because of the rather severe speed 184 * penalty that this imposes, the code normally only ensures 185 * persistence for the filesystem metadata contained within a 186 * snapshot. Setting the following flag allows this crash 187 * persistence to be enabled for file contents. 188 */ 189int dopersistence = 0; 190 191#ifdef DEBUG 192#include <sys/sysctl.h> 193SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, ""); 194static int snapdebug = 0; 195SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, ""); 196int collectsnapstats = 0; 197SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats, 198 0, ""); 199#endif /* DEBUG */ 200 201/* 202 * Create a snapshot file and initialize it for the filesystem. 203 */ 204int 205ffs_snapshot(mp, snapfile) 206 struct mount *mp; 207 char *snapfile; 208{ 209 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist; 210 int error, cg, snaploc; 211 int i, size, len, loc; 212 ufs2_daddr_t blockno; 213 uint64_t flag; 214 struct timespec starttime = {0, 0}, endtime; 215 char saved_nice = 0; 216 long redo = 0, snaplistsize = 0; 217 int32_t *lp; 218 void *space; 219 struct fs *copy_fs = NULL, *fs; 220 struct thread *td = curthread; 221 struct inode *ip, *xp; 222 struct buf *bp, *nbp, *ibp; 223 struct nameidata nd; 224 struct mount *wrtmp; 225 struct vattr vat; 226 struct vnode *vp, *xvp, *mvp, *devvp; 227 struct uio auio; 228 struct iovec aiov; 229 struct snapdata *sn; 230 struct ufsmount *ump; 231 232 ump = VFSTOUFS(mp); 233 fs = ump->um_fs; 234 sn = NULL; 235 /* 236 * At the moment, journaled soft updates cannot support 237 * taking snapshots. 238 */ 239 if (MOUNTEDSUJ(mp)) { 240 vfs_mount_error(mp, "%s: Snapshots are not yet supported when " 241 "running with journaled soft updates", fs->fs_fsmnt); 242 return (EOPNOTSUPP); 243 } 244 MNT_ILOCK(mp); 245 flag = mp->mnt_flag; 246 MNT_IUNLOCK(mp); 247 /* 248 * Need to serialize access to snapshot code per filesystem. 249 */ 250 /* 251 * Assign a snapshot slot in the superblock. 252 */ 253 UFS_LOCK(ump); 254 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 255 if (fs->fs_snapinum[snaploc] == 0) 256 break; 257 UFS_UNLOCK(ump); 258 if (snaploc == FSMAXSNAP) 259 return (ENOSPC); 260 /* 261 * Create the snapshot file. 262 */ 263restart: 264 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF | NOCACHE, UIO_SYSSPACE, 265 snapfile, td); 266 if ((error = namei(&nd)) != 0) 267 return (error); 268 if (nd.ni_vp != NULL) { 269 vput(nd.ni_vp); 270 error = EEXIST; 271 } 272 if (nd.ni_dvp->v_mount != mp) 273 error = EXDEV; 274 if (error) { 275 NDFREE(&nd, NDF_ONLY_PNBUF); 276 if (nd.ni_dvp == nd.ni_vp) 277 vrele(nd.ni_dvp); 278 else 279 vput(nd.ni_dvp); 280 return (error); 281 } 282 VATTR_NULL(&vat); 283 vat.va_type = VREG; 284 vat.va_mode = S_IRUSR; 285 vat.va_vaflags |= VA_EXCLUSIVE; 286 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp)) 287 wrtmp = NULL; 288 if (wrtmp != mp) 289 panic("ffs_snapshot: mount mismatch"); 290 vfs_rel(wrtmp); 291 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) { 292 NDFREE(&nd, NDF_ONLY_PNBUF); 293 vput(nd.ni_dvp); 294 if ((error = vn_start_write(NULL, &wrtmp, 295 V_XSLEEP | PCATCH)) != 0) 296 return (error); 297 goto restart; 298 } 299 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat); 300 VOP_UNLOCK(nd.ni_dvp, 0); 301 if (error) { 302 NDFREE(&nd, NDF_ONLY_PNBUF); 303 vn_finished_write(wrtmp); 304 vrele(nd.ni_dvp); 305 return (error); 306 } 307 vp = nd.ni_vp; 308 vnode_create_vobject(nd.ni_vp, fs->fs_size, td); 309 vp->v_vflag |= VV_SYSTEM; 310 ip = VTOI(vp); 311 devvp = ITODEVVP(ip); 312 /* 313 * Calculate the size of the filesystem then allocate the block 314 * immediately following the last block of the filesystem that 315 * will contain the snapshot list. This operation allows us to 316 * set the size of the snapshot. 317 */ 318 numblks = howmany(fs->fs_size, fs->fs_frag); 319 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)numblks), 320 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 321 if (error) 322 goto out; 323 bawrite(bp); 324 ip->i_size = lblktosize(fs, (off_t)(numblks + 1)); 325 vnode_pager_setsize(vp, ip->i_size); 326 DIP_SET(ip, i_size, ip->i_size); 327 ip->i_flag |= IN_SIZEMOD | IN_CHANGE | IN_UPDATE; 328 /* 329 * Preallocate critical data structures so that we can copy 330 * them in without further allocation after we suspend all 331 * operations on the filesystem. We would like to just release 332 * the allocated buffers without writing them since they will 333 * be filled in below once we are ready to go, but this upsets 334 * the soft update code, so we go ahead and write the new buffers. 335 * 336 * Allocate all indirect blocks and mark all of them as not 337 * needing to be copied. 338 */ 339 for (blkno = UFS_NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 340 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 341 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp); 342 if (error) 343 goto out; 344 bawrite(ibp); 345 } 346 /* 347 * Allocate copies for the superblock and its summary information. 348 */ 349 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED, 350 0, &nbp); 351 if (error) 352 goto out; 353 bawrite(nbp); 354 blkno = fragstoblks(fs, fs->fs_csaddr); 355 len = howmany(fs->fs_cssize, fs->fs_bsize); 356 for (loc = 0; loc < len; loc++) { 357 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)), 358 fs->fs_bsize, KERNCRED, 0, &nbp); 359 if (error) 360 goto out; 361 bawrite(nbp); 362 } 363 /* 364 * Allocate all cylinder group blocks. 365 */ 366 for (cg = 0; cg < fs->fs_ncg; cg++) { 367 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 368 fs->fs_bsize, KERNCRED, 0, &nbp); 369 if (error) 370 goto out; 371 bawrite(nbp); 372 if (cg % 10 == 0) { 373 error = ffs_syncvnode(vp, MNT_WAIT, 0); 374 if (error != 0) 375 goto out; 376 } 377 } 378 /* 379 * Copy all the cylinder group maps. Although the 380 * filesystem is still active, we hope that only a few 381 * cylinder groups will change between now and when we 382 * suspend operations. Thus, we will be able to quickly 383 * touch up the few cylinder groups that changed during 384 * the suspension period. 385 */ 386 len = roundup2(howmany(fs->fs_ncg, NBBY), sizeof(int)); 387 space = malloc(len, M_DEVBUF, M_WAITOK | M_ZERO); 388 UFS_LOCK(ump); 389 fs->fs_active = space; 390 UFS_UNLOCK(ump); 391 for (cg = 0; cg < fs->fs_ncg; cg++) { 392 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 393 fs->fs_bsize, KERNCRED, 0, &nbp); 394 if (error) 395 goto out; 396 error = cgaccount(cg, vp, nbp, 1); 397 bawrite(nbp); 398 if (cg % 10 == 0) 399 ffs_syncvnode(vp, MNT_WAIT, 0); 400 if (error) 401 goto out; 402 } 403 /* 404 * Change inode to snapshot type file. 405 */ 406 ip->i_flags |= SF_SNAPSHOT; 407 DIP_SET(ip, i_flags, ip->i_flags); 408 ip->i_flag |= IN_CHANGE | IN_UPDATE; 409 /* 410 * Ensure that the snapshot is completely on disk. 411 * Since we have marked it as a snapshot it is safe to 412 * unlock it as no process will be allowed to write to it. 413 */ 414 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 415 goto out; 416 VOP_UNLOCK(vp, 0); 417 /* 418 * All allocations are done, so we can now snapshot the system. 419 * 420 * Recind nice scheduling while running with the filesystem suspended. 421 */ 422 if (td->td_proc->p_nice > 0) { 423 struct proc *p; 424 425 p = td->td_proc; 426 PROC_LOCK(p); 427 saved_nice = p->p_nice; 428 sched_nice(p, 0); 429 PROC_UNLOCK(p); 430 } 431 /* 432 * Suspend operation on filesystem. 433 */ 434 for (;;) { 435 vn_finished_write(wrtmp); 436 if ((error = vfs_write_suspend(vp->v_mount, 0)) != 0) { 437 vn_start_write(NULL, &wrtmp, V_WAIT); 438 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 439 goto out; 440 } 441 if (mp->mnt_kern_flag & MNTK_SUSPENDED) 442 break; 443 vn_start_write(NULL, &wrtmp, V_WAIT); 444 } 445 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 446 if (ip->i_effnlink == 0) { 447 error = ENOENT; /* Snapshot file unlinked */ 448 goto resumefs; 449 } 450 if (collectsnapstats) 451 nanotime(&starttime); 452 453 /* 454 * First, copy all the cylinder group maps that have changed. 455 */ 456 for (cg = 0; cg < fs->fs_ncg; cg++) { 457 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0) 458 continue; 459 redo++; 460 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 461 fs->fs_bsize, KERNCRED, 0, &nbp); 462 if (error) 463 goto resumefs; 464 error = cgaccount(cg, vp, nbp, 2); 465 bawrite(nbp); 466 if (error) 467 goto resumefs; 468 } 469 /* 470 * Grab a copy of the superblock and its summary information. 471 * We delay writing it until the suspension is released below. 472 */ 473 copy_fs = malloc((u_long)fs->fs_bsize, M_UFSMNT, M_WAITOK); 474 bcopy(fs, copy_fs, fs->fs_sbsize); 475 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 476 copy_fs->fs_clean = 1; 477 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE; 478 if (fs->fs_sbsize < size) 479 bzero(&((char *)copy_fs)[fs->fs_sbsize], 480 size - fs->fs_sbsize); 481 size = blkroundup(fs, fs->fs_cssize); 482 if (fs->fs_contigsumsize > 0) 483 size += fs->fs_ncg * sizeof(int32_t); 484 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 485 copy_fs->fs_csp = space; 486 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize); 487 space = (char *)space + fs->fs_cssize; 488 loc = howmany(fs->fs_cssize, fs->fs_fsize); 489 i = fs->fs_frag - loc % fs->fs_frag; 490 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize; 491 if (len > 0) { 492 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc), 493 len, KERNCRED, &bp)) != 0) { 494 brelse(bp); 495 goto resumefs; 496 } 497 bcopy(bp->b_data, space, (u_int)len); 498 space = (char *)space + len; 499 bp->b_flags |= B_INVAL | B_NOCACHE; 500 brelse(bp); 501 } 502 if (fs->fs_contigsumsize > 0) { 503 copy_fs->fs_maxcluster = lp = space; 504 for (i = 0; i < fs->fs_ncg; i++) 505 *lp++ = fs->fs_contigsumsize; 506 } 507 /* 508 * We must check for active files that have been unlinked 509 * (e.g., with a zero link count). We have to expunge all 510 * trace of these files from the snapshot so that they are 511 * not reclaimed prematurely by fsck or unnecessarily dumped. 512 * We turn off the MNTK_SUSPENDED flag to avoid a panic from 513 * spec_strategy about writing on a suspended filesystem. 514 * Note that we skip unlinked snapshot files as they will 515 * be handled separately below. 516 * 517 * We also calculate the size needed for the snapshot list. 518 * Initial number of entries is composed of: 519 * - one for each cylinder group map 520 * - one for each block used by superblock summary table 521 * - one for each snapshot inode block 522 * - one for the superblock 523 * - one for the snapshot list 524 * The direct block entries in the snapshot are always 525 * copied (see reason below). Note that the superblock and 526 * the first cylinder group will almost always be allocated 527 * in the direct blocks, but we add the slop for them in case 528 * they do not end up there. The snapshot list size may get 529 * expanded by one because of an update of an inode block for 530 * an unlinked but still open file when it is expunged. 531 * 532 * Because the direct block pointers are always copied, they 533 * are not added to the list. Instead ffs_copyonwrite() 534 * explicitly checks for them before checking the snapshot list. 535 */ 536 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) + 537 FSMAXSNAP + /* superblock */ 1 + /* snaplist */ 1; 538 MNT_ILOCK(mp); 539 mp->mnt_kern_flag &= ~MNTK_SUSPENDED; 540 MNT_IUNLOCK(mp); 541loop: 542 MNT_VNODE_FOREACH_ALL(xvp, mp, mvp) { 543 if ((xvp->v_usecount == 0 && 544 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) || 545 xvp->v_type == VNON || 546 IS_SNAPSHOT(VTOI(xvp))) { 547 VI_UNLOCK(xvp); 548 continue; 549 } 550 /* 551 * We can skip parent directory vnode because it must have 552 * this snapshot file in it. 553 */ 554 if (xvp == nd.ni_dvp) { 555 VI_UNLOCK(xvp); 556 continue; 557 } 558 vholdl(xvp); 559 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK) != 0) { 560 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 561 vdrop(xvp); 562 goto loop; 563 } 564 VI_LOCK(xvp); 565 if (xvp->v_usecount == 0 && 566 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) { 567 VI_UNLOCK(xvp); 568 VOP_UNLOCK(xvp, 0); 569 vdrop(xvp); 570 continue; 571 } 572 VI_UNLOCK(xvp); 573 if (snapdebug) 574 vn_printf(xvp, "ffs_snapshot: busy vnode "); 575 if (VOP_GETATTR(xvp, &vat, td->td_ucred) == 0 && 576 vat.va_nlink > 0) { 577 VOP_UNLOCK(xvp, 0); 578 vdrop(xvp); 579 continue; 580 } 581 xp = VTOI(xvp); 582 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) { 583 VOP_UNLOCK(xvp, 0); 584 vdrop(xvp); 585 continue; 586 } 587 /* 588 * If there is a fragment, clear it here. 589 */ 590 blkno = 0; 591 loc = howmany(xp->i_size, fs->fs_bsize) - 1; 592 if (loc < UFS_NDADDR) { 593 len = fragroundup(fs, blkoff(fs, xp->i_size)); 594 if (len != 0 && len < fs->fs_bsize) { 595 ffs_blkfree(ump, copy_fs, vp, 596 DIP(xp, i_db[loc]), len, xp->i_number, 597 xvp->v_type, NULL, SINGLETON_KEY); 598 blkno = DIP(xp, i_db[loc]); 599 DIP_SET(xp, i_db[loc], 0); 600 } 601 } 602 snaplistsize += 1; 603 if (I_IS_UFS1(xp)) 604 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, 605 BLK_NOCOPY, 1); 606 else 607 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, 608 BLK_NOCOPY, 1); 609 if (blkno) 610 DIP_SET(xp, i_db[loc], blkno); 611 if (!error) 612 error = ffs_freefile(ump, copy_fs, vp, xp->i_number, 613 xp->i_mode, NULL); 614 VOP_UNLOCK(xvp, 0); 615 vdrop(xvp); 616 if (error) { 617 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 618 goto resumefs; 619 } 620 } 621 /* 622 * Erase the journal file from the snapshot. 623 */ 624 if (fs->fs_flags & FS_SUJ) { 625 error = softdep_journal_lookup(mp, &xvp); 626 if (error) 627 goto resumefs; 628 xp = VTOI(xvp); 629 if (I_IS_UFS1(xp)) 630 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, 631 BLK_NOCOPY, 0); 632 else 633 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, 634 BLK_NOCOPY, 0); 635 vput(xvp); 636 } 637 /* 638 * Acquire a lock on the snapdata structure, creating it if necessary. 639 */ 640 sn = ffs_snapdata_acquire(devvp); 641 /* 642 * Change vnode to use shared snapshot lock instead of the original 643 * private lock. 644 */ 645 vp->v_vnlock = &sn->sn_lock; 646 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 647 xp = TAILQ_FIRST(&sn->sn_head); 648 /* 649 * If this is the first snapshot on this filesystem, then we need 650 * to allocate the space for the list of preallocated snapshot blocks. 651 * This list will be refined below, but this preliminary one will 652 * keep us out of deadlock until the full one is ready. 653 */ 654 if (xp == NULL) { 655 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 656 M_UFSMNT, M_WAITOK); 657 blkp = &snapblklist[1]; 658 *blkp++ = lblkno(fs, fs->fs_sblockloc); 659 blkno = fragstoblks(fs, fs->fs_csaddr); 660 for (cg = 0; cg < fs->fs_ncg; cg++) { 661 if (fragstoblks(fs, cgtod(fs, cg) > blkno)) 662 break; 663 *blkp++ = fragstoblks(fs, cgtod(fs, cg)); 664 } 665 len = howmany(fs->fs_cssize, fs->fs_bsize); 666 for (loc = 0; loc < len; loc++) 667 *blkp++ = blkno + loc; 668 for (; cg < fs->fs_ncg; cg++) 669 *blkp++ = fragstoblks(fs, cgtod(fs, cg)); 670 snapblklist[0] = blkp - snapblklist; 671 VI_LOCK(devvp); 672 if (sn->sn_blklist != NULL) 673 panic("ffs_snapshot: non-empty list"); 674 sn->sn_blklist = snapblklist; 675 sn->sn_listsize = blkp - snapblklist; 676 VI_UNLOCK(devvp); 677 } 678 /* 679 * Preallocate all the direct blocks in the snapshot inode so 680 * that we never have to write the inode itself to commit an 681 * update to the contents of the snapshot. Note that once 682 * created, the size of the snapshot will never change, so 683 * there will never be a need to write the inode except to 684 * update the non-integrity-critical time fields and 685 * allocated-block count. 686 */ 687 for (blockno = 0; blockno < UFS_NDADDR; blockno++) { 688 if (DIP(ip, i_db[blockno]) != 0) 689 continue; 690 error = UFS_BALLOC(vp, lblktosize(fs, blockno), 691 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 692 if (error) 693 goto resumefs; 694 error = readblock(vp, bp, blockno); 695 bawrite(bp); 696 if (error != 0) 697 goto resumefs; 698 } 699 /* 700 * Record snapshot inode. Since this is the newest snapshot, 701 * it must be placed at the end of the list. 702 */ 703 VI_LOCK(devvp); 704 fs->fs_snapinum[snaploc] = ip->i_number; 705 if (ip->i_nextsnap.tqe_prev != 0) 706 panic("ffs_snapshot: %ju already on list", 707 (uintmax_t)ip->i_number); 708 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); 709 devvp->v_vflag |= VV_COPYONWRITE; 710 VI_UNLOCK(devvp); 711resumefs: 712 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp"); 713 if (error != 0 && copy_fs != NULL) { 714 free(copy_fs->fs_csp, M_UFSMNT); 715 free(copy_fs, M_UFSMNT); 716 copy_fs = NULL; 717 } 718 KASSERT(error != 0 || (sn != NULL && copy_fs != NULL), 719 ("missing snapshot setup parameters")); 720 /* 721 * Resume operation on filesystem. 722 */ 723 vfs_write_resume(vp->v_mount, VR_START_WRITE | VR_NO_SUSPCLR); 724 if (collectsnapstats && starttime.tv_sec > 0) { 725 nanotime(&endtime); 726 timespecsub(&endtime, &starttime, &endtime); 727 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n", 728 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec, 729 endtime.tv_nsec / 1000000, redo, fs->fs_ncg); 730 } 731 if (copy_fs == NULL) 732 goto out; 733 /* 734 * Copy allocation information from all the snapshots in 735 * this snapshot and then expunge them from its view. 736 */ 737 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) { 738 if (xp == ip) 739 break; 740 if (I_IS_UFS1(xp)) 741 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1, 742 BLK_SNAP, 0); 743 else 744 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2, 745 BLK_SNAP, 0); 746 if (error == 0 && xp->i_effnlink == 0) { 747 error = ffs_freefile(ump, 748 copy_fs, 749 vp, 750 xp->i_number, 751 xp->i_mode, NULL); 752 } 753 if (error) { 754 fs->fs_snapinum[snaploc] = 0; 755 goto done; 756 } 757 } 758 /* 759 * Allocate space for the full list of preallocated snapshot blocks. 760 */ 761 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 762 M_UFSMNT, M_WAITOK); 763 ip->i_snapblklist = &snapblklist[1]; 764 /* 765 * Expunge the blocks used by the snapshots from the set of 766 * blocks marked as used in the snapshot bitmaps. Also, collect 767 * the list of allocated blocks in i_snapblklist. 768 */ 769 if (I_IS_UFS1(ip)) 770 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, 771 BLK_SNAP, 0); 772 else 773 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, 774 BLK_SNAP, 0); 775 if (error) { 776 fs->fs_snapinum[snaploc] = 0; 777 free(snapblklist, M_UFSMNT); 778 goto done; 779 } 780 if (snaplistsize < ip->i_snapblklist - snapblklist) 781 panic("ffs_snapshot: list too small"); 782 snaplistsize = ip->i_snapblklist - snapblklist; 783 snapblklist[0] = snaplistsize; 784 ip->i_snapblklist = 0; 785 /* 786 * Write out the list of allocated blocks to the end of the snapshot. 787 */ 788 auio.uio_iov = &aiov; 789 auio.uio_iovcnt = 1; 790 aiov.iov_base = (void *)snapblklist; 791 aiov.iov_len = snaplistsize * sizeof(daddr_t); 792 auio.uio_resid = aiov.iov_len; 793 auio.uio_offset = lblktosize(fs, (off_t)numblks); 794 auio.uio_segflg = UIO_SYSSPACE; 795 auio.uio_rw = UIO_WRITE; 796 auio.uio_td = td; 797 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 798 fs->fs_snapinum[snaploc] = 0; 799 free(snapblklist, M_UFSMNT); 800 goto done; 801 } 802 /* 803 * Write the superblock and its summary information 804 * to the snapshot. 805 */ 806 blkno = fragstoblks(fs, fs->fs_csaddr); 807 len = howmany(fs->fs_cssize, fs->fs_bsize); 808 space = copy_fs->fs_csp; 809 for (loc = 0; loc < len; loc++) { 810 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp); 811 if (error) { 812 fs->fs_snapinum[snaploc] = 0; 813 free(snapblklist, M_UFSMNT); 814 goto done; 815 } 816 bcopy(space, nbp->b_data, fs->fs_bsize); 817 space = (char *)space + fs->fs_bsize; 818 bawrite(nbp); 819 } 820 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize, 821 KERNCRED, &nbp); 822 if (error) { 823 brelse(nbp); 824 } else { 825 loc = blkoff(fs, fs->fs_sblockloc); 826 bcopy((char *)copy_fs, &nbp->b_data[loc], (u_int)fs->fs_sbsize); 827 bawrite(nbp); 828 } 829 /* 830 * As this is the newest list, it is the most inclusive, so 831 * should replace the previous list. 832 */ 833 VI_LOCK(devvp); 834 space = sn->sn_blklist; 835 sn->sn_blklist = snapblklist; 836 sn->sn_listsize = snaplistsize; 837 VI_UNLOCK(devvp); 838 if (space != NULL) 839 free(space, M_UFSMNT); 840done: 841 free(copy_fs->fs_csp, M_UFSMNT); 842 free(copy_fs, M_UFSMNT); 843 copy_fs = NULL; 844out: 845 NDFREE(&nd, NDF_ONLY_PNBUF); 846 if (saved_nice > 0) { 847 struct proc *p; 848 849 p = td->td_proc; 850 PROC_LOCK(p); 851 sched_nice(td->td_proc, saved_nice); 852 PROC_UNLOCK(td->td_proc); 853 } 854 UFS_LOCK(ump); 855 if (fs->fs_active != 0) { 856 free(fs->fs_active, M_DEVBUF); 857 fs->fs_active = 0; 858 } 859 UFS_UNLOCK(ump); 860 MNT_ILOCK(mp); 861 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 862 MNT_IUNLOCK(mp); 863 if (error) 864 (void) ffs_truncate(vp, (off_t)0, 0, NOCRED); 865 (void) ffs_syncvnode(vp, MNT_WAIT, 0); 866 if (error) 867 vput(vp); 868 else 869 VOP_UNLOCK(vp, 0); 870 vrele(nd.ni_dvp); 871 vn_finished_write(wrtmp); 872 process_deferred_inactive(mp); 873 return (error); 874} 875 876/* 877 * Copy a cylinder group map. All the unallocated blocks are marked 878 * BLK_NOCOPY so that the snapshot knows that it need not copy them 879 * if they are later written. If passno is one, then this is a first 880 * pass, so only setting needs to be done. If passno is 2, then this 881 * is a revision to a previous pass which must be undone as the 882 * replacement pass is done. 883 */ 884static int 885cgaccount(cg, vp, nbp, passno) 886 int cg; 887 struct vnode *vp; 888 struct buf *nbp; 889 int passno; 890{ 891 struct buf *bp, *ibp; 892 struct inode *ip; 893 struct cg *cgp; 894 struct fs *fs; 895 ufs2_daddr_t base, numblks; 896 int error, len, loc, indiroff; 897 898 ip = VTOI(vp); 899 fs = ITOFS(ip); 900 if ((error = ffs_getcg(fs, ITODEVVP(ip), cg, &bp, &cgp)) != 0) 901 return (error); 902 UFS_LOCK(ITOUMP(ip)); 903 ACTIVESET(fs, cg); 904 /* 905 * Recomputation of summary information might not have been performed 906 * at mount time. Sync up summary information for current cylinder 907 * group while data is in memory to ensure that result of background 908 * fsck is slightly more consistent. 909 */ 910 fs->fs_cs(fs, cg) = cgp->cg_cs; 911 UFS_UNLOCK(ITOUMP(ip)); 912 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize); 913 if (fs->fs_cgsize < fs->fs_bsize) 914 bzero(&nbp->b_data[fs->fs_cgsize], 915 fs->fs_bsize - fs->fs_cgsize); 916 cgp = (struct cg *)nbp->b_data; 917 bqrelse(bp); 918 if (passno == 2) 919 nbp->b_flags |= B_VALIDSUSPWRT; 920 numblks = howmany(fs->fs_size, fs->fs_frag); 921 len = howmany(fs->fs_fpg, fs->fs_frag); 922 base = cgbase(fs, cg) / fs->fs_frag; 923 if (base + len >= numblks) 924 len = numblks - base - 1; 925 loc = 0; 926 if (base < UFS_NDADDR) { 927 for ( ; loc < UFS_NDADDR; loc++) { 928 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 929 DIP_SET(ip, i_db[loc], BLK_NOCOPY); 930 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY) 931 DIP_SET(ip, i_db[loc], 0); 932 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY) 933 panic("ffs_snapshot: lost direct block"); 934 } 935 } 936 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)), 937 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 938 if (error) { 939 goto out; 940 } 941 indiroff = (base + loc - UFS_NDADDR) % NINDIR(fs); 942 for ( ; loc < len; loc++, indiroff++) { 943 if (indiroff >= NINDIR(fs)) { 944 if (passno == 2) 945 ibp->b_flags |= B_VALIDSUSPWRT; 946 bawrite(ibp); 947 error = UFS_BALLOC(vp, 948 lblktosize(fs, (off_t)(base + loc)), 949 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 950 if (error) { 951 goto out; 952 } 953 indiroff = 0; 954 } 955 if (I_IS_UFS1(ip)) { 956 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 957 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 958 BLK_NOCOPY; 959 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data)) 960 [indiroff] == BLK_NOCOPY) 961 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0; 962 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data)) 963 [indiroff] == BLK_NOCOPY) 964 panic("ffs_snapshot: lost indirect block"); 965 continue; 966 } 967 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 968 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; 969 else if (passno == 2 && 970 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 971 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0; 972 else if (passno == 1 && 973 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 974 panic("ffs_snapshot: lost indirect block"); 975 } 976 if (passno == 2) 977 ibp->b_flags |= B_VALIDSUSPWRT; 978 bdwrite(ibp); 979out: 980 /* 981 * We have to calculate the crc32c here rather than just setting the 982 * BX_CYLGRP b_xflags because the allocation of the block for the 983 * the cylinder group map will always be a full size block (fs_bsize) 984 * even though the cylinder group may be smaller (fs_cgsize). The 985 * crc32c must be computed only over fs_cgsize whereas the BX_CYLGRP 986 * flag causes it to be computed over the size of the buffer. 987 */ 988 if ((fs->fs_metackhash & CK_CYLGRP) != 0) { 989 ((struct cg *)nbp->b_data)->cg_ckhash = 0; 990 ((struct cg *)nbp->b_data)->cg_ckhash = 991 calculate_crc32c(~0L, nbp->b_data, fs->fs_cgsize); 992 } 993 return (error); 994} 995 996/* 997 * Before expunging a snapshot inode, note all the 998 * blocks that it claims with BLK_SNAP so that fsck will 999 * be able to account for those blocks properly and so 1000 * that this snapshot knows that it need not copy them 1001 * if the other snapshot holding them is freed. This code 1002 * is reproduced once each for UFS1 and UFS2. 1003 */ 1004static int 1005expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype, clearmode) 1006 struct vnode *snapvp; 1007 struct inode *cancelip; 1008 struct fs *fs; 1009 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 1010 struct fs *, ufs_lbn_t, int); 1011 int expungetype; 1012 int clearmode; 1013{ 1014 int i, error, indiroff; 1015 ufs_lbn_t lbn, rlbn; 1016 ufs2_daddr_t len, blkno, numblks, blksperindir; 1017 struct ufs1_dinode *dip; 1018 struct thread *td = curthread; 1019 struct buf *bp; 1020 1021 /* 1022 * Prepare to expunge the inode. If its inode block has not 1023 * yet been copied, then allocate and fill the copy. 1024 */ 1025 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 1026 blkno = 0; 1027 if (lbn < UFS_NDADDR) { 1028 blkno = VTOI(snapvp)->i_din1->di_db[lbn]; 1029 } else { 1030 if (DOINGSOFTDEP(snapvp)) 1031 softdep_prealloc(snapvp, MNT_WAIT); 1032 td->td_pflags |= TDP_COWINPROGRESS; 1033 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), 1034 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 1035 td->td_pflags &= ~TDP_COWINPROGRESS; 1036 if (error) 1037 return (error); 1038 indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); 1039 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; 1040 bqrelse(bp); 1041 } 1042 if (blkno != 0) { 1043 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 1044 return (error); 1045 } else { 1046 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), 1047 fs->fs_bsize, KERNCRED, 0, &bp); 1048 if (error) 1049 return (error); 1050 if ((error = readblock(snapvp, bp, lbn)) != 0) 1051 return (error); 1052 } 1053 /* 1054 * Set a snapshot inode to be a zero length file, regular files 1055 * or unlinked snapshots to be completely unallocated. 1056 */ 1057 dip = (struct ufs1_dinode *)bp->b_data + 1058 ino_to_fsbo(fs, cancelip->i_number); 1059 if (clearmode || cancelip->i_effnlink == 0) 1060 dip->di_mode = 0; 1061 dip->di_size = 0; 1062 dip->di_blocks = 0; 1063 dip->di_flags &= ~SF_SNAPSHOT; 1064 bzero(&dip->di_db[0], (UFS_NDADDR + UFS_NIADDR) * sizeof(ufs1_daddr_t)); 1065 bdwrite(bp); 1066 /* 1067 * Now go through and expunge all the blocks in the file 1068 * using the function requested. 1069 */ 1070 numblks = howmany(cancelip->i_size, fs->fs_bsize); 1071 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0], 1072 &cancelip->i_din1->di_db[UFS_NDADDR], fs, 0, expungetype))) 1073 return (error); 1074 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0], 1075 &cancelip->i_din1->di_ib[UFS_NIADDR], fs, -1, expungetype))) 1076 return (error); 1077 blksperindir = 1; 1078 lbn = -UFS_NDADDR; 1079 len = numblks - UFS_NDADDR; 1080 rlbn = UFS_NDADDR; 1081 for (i = 0; len > 0 && i < UFS_NIADDR; i++) { 1082 error = indiracct_ufs1(snapvp, ITOV(cancelip), i, 1083 cancelip->i_din1->di_ib[i], lbn, rlbn, len, 1084 blksperindir, fs, acctfunc, expungetype); 1085 if (error) 1086 return (error); 1087 blksperindir *= NINDIR(fs); 1088 lbn -= blksperindir + 1; 1089 len -= blksperindir; 1090 rlbn += blksperindir; 1091 } 1092 return (0); 1093} 1094 1095/* 1096 * Descend an indirect block chain for vnode cancelvp accounting for all 1097 * its indirect blocks in snapvp. 1098 */ 1099static int 1100indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 1101 blksperindir, fs, acctfunc, expungetype) 1102 struct vnode *snapvp; 1103 struct vnode *cancelvp; 1104 int level; 1105 ufs1_daddr_t blkno; 1106 ufs_lbn_t lbn; 1107 ufs_lbn_t rlbn; 1108 ufs_lbn_t remblks; 1109 ufs_lbn_t blksperindir; 1110 struct fs *fs; 1111 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 1112 struct fs *, ufs_lbn_t, int); 1113 int expungetype; 1114{ 1115 int error, num, i; 1116 ufs_lbn_t subblksperindir; 1117 struct indir indirs[UFS_NIADDR + 2]; 1118 ufs1_daddr_t last, *bap; 1119 struct buf *bp; 1120 1121 if (blkno == 0) { 1122 if (expungetype == BLK_NOCOPY) 1123 return (0); 1124 panic("indiracct_ufs1: missing indir"); 1125 } 1126 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 1127 return (error); 1128 if (lbn != indirs[num - 1 - level].in_lbn || num < 2) 1129 panic("indiracct_ufs1: botched params"); 1130 /* 1131 * We have to expand bread here since it will deadlock looking 1132 * up the block number for any blocks that are not in the cache. 1133 */ 1134 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); 1135 bp->b_blkno = fsbtodb(fs, blkno); 1136 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && 1137 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { 1138 brelse(bp); 1139 return (error); 1140 } 1141 /* 1142 * Account for the block pointers in this indirect block. 1143 */ 1144 last = howmany(remblks, blksperindir); 1145 if (last > NINDIR(fs)) 1146 last = NINDIR(fs); 1147 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); 1148 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 1149 bqrelse(bp); 1150 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 1151 level == 0 ? rlbn : -1, expungetype); 1152 if (error || level == 0) 1153 goto out; 1154 /* 1155 * Account for the block pointers in each of the indirect blocks 1156 * in the levels below us. 1157 */ 1158 subblksperindir = blksperindir / NINDIR(fs); 1159 for (lbn++, level--, i = 0; i < last; i++) { 1160 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn, 1161 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 1162 if (error) 1163 goto out; 1164 rlbn += blksperindir; 1165 lbn -= blksperindir; 1166 remblks -= blksperindir; 1167 } 1168out: 1169 free(bap, M_DEVBUF); 1170 return (error); 1171} 1172 1173/* 1174 * Do both snap accounting and map accounting. 1175 */ 1176static int 1177fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype) 1178 struct vnode *vp; 1179 ufs1_daddr_t *oldblkp, *lastblkp; 1180 struct fs *fs; 1181 ufs_lbn_t lblkno; 1182 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 1183{ 1184 int error; 1185 1186 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 1187 return (error); 1188 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 1189} 1190 1191/* 1192 * Identify a set of blocks allocated in a snapshot inode. 1193 */ 1194static int 1195snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1196 struct vnode *vp; 1197 ufs1_daddr_t *oldblkp, *lastblkp; 1198 struct fs *fs; 1199 ufs_lbn_t lblkno; 1200 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 1201{ 1202 struct inode *ip = VTOI(vp); 1203 ufs1_daddr_t blkno, *blkp; 1204 ufs_lbn_t lbn; 1205 struct buf *ibp; 1206 int error; 1207 1208 for ( ; oldblkp < lastblkp; oldblkp++) { 1209 blkno = *oldblkp; 1210 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 1211 continue; 1212 lbn = fragstoblks(fs, blkno); 1213 if (lbn < UFS_NDADDR) { 1214 blkp = &ip->i_din1->di_db[lbn]; 1215 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1216 } else { 1217 error = ffs_balloc_ufs1(vp, lblktosize(fs, (off_t)lbn), 1218 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1219 if (error) 1220 return (error); 1221 blkp = &((ufs1_daddr_t *)(ibp->b_data)) 1222 [(lbn - UFS_NDADDR) % NINDIR(fs)]; 1223 } 1224 /* 1225 * If we are expunging a snapshot vnode and we 1226 * find a block marked BLK_NOCOPY, then it is 1227 * one that has been allocated to this snapshot after 1228 * we took our current snapshot and can be ignored. 1229 */ 1230 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 1231 if (lbn >= UFS_NDADDR) 1232 brelse(ibp); 1233 } else { 1234 if (*blkp != 0) 1235 panic("snapacct_ufs1: bad block"); 1236 *blkp = expungetype; 1237 if (lbn >= UFS_NDADDR) 1238 bdwrite(ibp); 1239 } 1240 } 1241 return (0); 1242} 1243 1244/* 1245 * Account for a set of blocks allocated in a snapshot inode. 1246 */ 1247static int 1248mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1249 struct vnode *vp; 1250 ufs1_daddr_t *oldblkp, *lastblkp; 1251 struct fs *fs; 1252 ufs_lbn_t lblkno; 1253 int expungetype; 1254{ 1255 ufs1_daddr_t blkno; 1256 struct inode *ip; 1257 ino_t inum; 1258 int acctit; 1259 1260 ip = VTOI(vp); 1261 inum = ip->i_number; 1262 if (lblkno == -1) 1263 acctit = 0; 1264 else 1265 acctit = 1; 1266 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 1267 blkno = *oldblkp; 1268 if (blkno == 0 || blkno == BLK_NOCOPY) 1269 continue; 1270 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP) 1271 *ip->i_snapblklist++ = lblkno; 1272 if (blkno == BLK_SNAP) 1273 blkno = blkstofrags(fs, lblkno); 1274 ffs_blkfree(ITOUMP(ip), fs, vp, blkno, fs->fs_bsize, inum, 1275 vp->v_type, NULL, SINGLETON_KEY); 1276 } 1277 return (0); 1278} 1279 1280/* 1281 * Before expunging a snapshot inode, note all the 1282 * blocks that it claims with BLK_SNAP so that fsck will 1283 * be able to account for those blocks properly and so 1284 * that this snapshot knows that it need not copy them 1285 * if the other snapshot holding them is freed. This code 1286 * is reproduced once each for UFS1 and UFS2. 1287 */ 1288static int 1289expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype, clearmode) 1290 struct vnode *snapvp; 1291 struct inode *cancelip; 1292 struct fs *fs; 1293 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1294 struct fs *, ufs_lbn_t, int); 1295 int expungetype; 1296 int clearmode; 1297{ 1298 int i, error, indiroff; 1299 ufs_lbn_t lbn, rlbn; 1300 ufs2_daddr_t len, blkno, numblks, blksperindir; 1301 struct ufs2_dinode *dip; 1302 struct thread *td = curthread; 1303 struct buf *bp; 1304 1305 /* 1306 * Prepare to expunge the inode. If its inode block has not 1307 * yet been copied, then allocate and fill the copy. 1308 */ 1309 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 1310 blkno = 0; 1311 if (lbn < UFS_NDADDR) { 1312 blkno = VTOI(snapvp)->i_din2->di_db[lbn]; 1313 } else { 1314 if (DOINGSOFTDEP(snapvp)) 1315 softdep_prealloc(snapvp, MNT_WAIT); 1316 td->td_pflags |= TDP_COWINPROGRESS; 1317 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), 1318 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 1319 td->td_pflags &= ~TDP_COWINPROGRESS; 1320 if (error) 1321 return (error); 1322 indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); 1323 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; 1324 bqrelse(bp); 1325 } 1326 if (blkno != 0) { 1327 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 1328 return (error); 1329 } else { 1330 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), 1331 fs->fs_bsize, KERNCRED, 0, &bp); 1332 if (error) 1333 return (error); 1334 if ((error = readblock(snapvp, bp, lbn)) != 0) 1335 return (error); 1336 } 1337 /* 1338 * Set a snapshot inode to be a zero length file, regular files 1339 * to be completely unallocated. 1340 */ 1341 dip = (struct ufs2_dinode *)bp->b_data + 1342 ino_to_fsbo(fs, cancelip->i_number); 1343 if (clearmode || cancelip->i_effnlink == 0) 1344 dip->di_mode = 0; 1345 dip->di_size = 0; 1346 dip->di_blocks = 0; 1347 dip->di_flags &= ~SF_SNAPSHOT; 1348 bzero(&dip->di_db[0], (UFS_NDADDR + UFS_NIADDR) * sizeof(ufs2_daddr_t)); 1349 bdwrite(bp); 1350 /* 1351 * Now go through and expunge all the blocks in the file 1352 * using the function requested. 1353 */ 1354 numblks = howmany(cancelip->i_size, fs->fs_bsize); 1355 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0], 1356 &cancelip->i_din2->di_db[UFS_NDADDR], fs, 0, expungetype))) 1357 return (error); 1358 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0], 1359 &cancelip->i_din2->di_ib[UFS_NIADDR], fs, -1, expungetype))) 1360 return (error); 1361 blksperindir = 1; 1362 lbn = -UFS_NDADDR; 1363 len = numblks - UFS_NDADDR; 1364 rlbn = UFS_NDADDR; 1365 for (i = 0; len > 0 && i < UFS_NIADDR; i++) { 1366 error = indiracct_ufs2(snapvp, ITOV(cancelip), i, 1367 cancelip->i_din2->di_ib[i], lbn, rlbn, len, 1368 blksperindir, fs, acctfunc, expungetype); 1369 if (error) 1370 return (error); 1371 blksperindir *= NINDIR(fs); 1372 lbn -= blksperindir + 1; 1373 len -= blksperindir; 1374 rlbn += blksperindir; 1375 } 1376 return (0); 1377} 1378 1379/* 1380 * Descend an indirect block chain for vnode cancelvp accounting for all 1381 * its indirect blocks in snapvp. 1382 */ 1383static int 1384indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 1385 blksperindir, fs, acctfunc, expungetype) 1386 struct vnode *snapvp; 1387 struct vnode *cancelvp; 1388 int level; 1389 ufs2_daddr_t blkno; 1390 ufs_lbn_t lbn; 1391 ufs_lbn_t rlbn; 1392 ufs_lbn_t remblks; 1393 ufs_lbn_t blksperindir; 1394 struct fs *fs; 1395 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1396 struct fs *, ufs_lbn_t, int); 1397 int expungetype; 1398{ 1399 int error, num, i; 1400 ufs_lbn_t subblksperindir; 1401 struct indir indirs[UFS_NIADDR + 2]; 1402 ufs2_daddr_t last, *bap; 1403 struct buf *bp; 1404 1405 if (blkno == 0) { 1406 if (expungetype == BLK_NOCOPY) 1407 return (0); 1408 panic("indiracct_ufs2: missing indir"); 1409 } 1410 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 1411 return (error); 1412 if (lbn != indirs[num - 1 - level].in_lbn || num < 2) 1413 panic("indiracct_ufs2: botched params"); 1414 /* 1415 * We have to expand bread here since it will deadlock looking 1416 * up the block number for any blocks that are not in the cache. 1417 */ 1418 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); 1419 bp->b_blkno = fsbtodb(fs, blkno); 1420 if ((bp->b_flags & B_CACHE) == 0 && 1421 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { 1422 brelse(bp); 1423 return (error); 1424 } 1425 /* 1426 * Account for the block pointers in this indirect block. 1427 */ 1428 last = howmany(remblks, blksperindir); 1429 if (last > NINDIR(fs)) 1430 last = NINDIR(fs); 1431 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); 1432 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 1433 bqrelse(bp); 1434 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 1435 level == 0 ? rlbn : -1, expungetype); 1436 if (error || level == 0) 1437 goto out; 1438 /* 1439 * Account for the block pointers in each of the indirect blocks 1440 * in the levels below us. 1441 */ 1442 subblksperindir = blksperindir / NINDIR(fs); 1443 for (lbn++, level--, i = 0; i < last; i++) { 1444 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn, 1445 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 1446 if (error) 1447 goto out; 1448 rlbn += blksperindir; 1449 lbn -= blksperindir; 1450 remblks -= blksperindir; 1451 } 1452out: 1453 free(bap, M_DEVBUF); 1454 return (error); 1455} 1456 1457/* 1458 * Do both snap accounting and map accounting. 1459 */ 1460static int 1461fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype) 1462 struct vnode *vp; 1463 ufs2_daddr_t *oldblkp, *lastblkp; 1464 struct fs *fs; 1465 ufs_lbn_t lblkno; 1466 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 1467{ 1468 int error; 1469 1470 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 1471 return (error); 1472 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 1473} 1474 1475/* 1476 * Identify a set of blocks allocated in a snapshot inode. 1477 */ 1478static int 1479snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1480 struct vnode *vp; 1481 ufs2_daddr_t *oldblkp, *lastblkp; 1482 struct fs *fs; 1483 ufs_lbn_t lblkno; 1484 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 1485{ 1486 struct inode *ip = VTOI(vp); 1487 ufs2_daddr_t blkno, *blkp; 1488 ufs_lbn_t lbn; 1489 struct buf *ibp; 1490 int error; 1491 1492 for ( ; oldblkp < lastblkp; oldblkp++) { 1493 blkno = *oldblkp; 1494 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 1495 continue; 1496 lbn = fragstoblks(fs, blkno); 1497 if (lbn < UFS_NDADDR) { 1498 blkp = &ip->i_din2->di_db[lbn]; 1499 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1500 } else { 1501 error = ffs_balloc_ufs2(vp, lblktosize(fs, (off_t)lbn), 1502 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1503 if (error) 1504 return (error); 1505 blkp = &((ufs2_daddr_t *)(ibp->b_data)) 1506 [(lbn - UFS_NDADDR) % NINDIR(fs)]; 1507 } 1508 /* 1509 * If we are expunging a snapshot vnode and we 1510 * find a block marked BLK_NOCOPY, then it is 1511 * one that has been allocated to this snapshot after 1512 * we took our current snapshot and can be ignored. 1513 */ 1514 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 1515 if (lbn >= UFS_NDADDR) 1516 brelse(ibp); 1517 } else { 1518 if (*blkp != 0) 1519 panic("snapacct_ufs2: bad block"); 1520 *blkp = expungetype; 1521 if (lbn >= UFS_NDADDR) 1522 bdwrite(ibp); 1523 } 1524 } 1525 return (0); 1526} 1527 1528/* 1529 * Account for a set of blocks allocated in a snapshot inode. 1530 */ 1531static int 1532mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1533 struct vnode *vp; 1534 ufs2_daddr_t *oldblkp, *lastblkp; 1535 struct fs *fs; 1536 ufs_lbn_t lblkno; 1537 int expungetype; 1538{ 1539 ufs2_daddr_t blkno; 1540 struct inode *ip; 1541 ino_t inum; 1542 int acctit; 1543 1544 ip = VTOI(vp); 1545 inum = ip->i_number; 1546 if (lblkno == -1) 1547 acctit = 0; 1548 else 1549 acctit = 1; 1550 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 1551 blkno = *oldblkp; 1552 if (blkno == 0 || blkno == BLK_NOCOPY) 1553 continue; 1554 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP && 1555 lblkno >= UFS_NDADDR) 1556 *ip->i_snapblklist++ = lblkno; 1557 if (blkno == BLK_SNAP) 1558 blkno = blkstofrags(fs, lblkno); 1559 ffs_blkfree(ITOUMP(ip), fs, vp, blkno, fs->fs_bsize, inum, 1560 vp->v_type, NULL, SINGLETON_KEY); 1561 } 1562 return (0); 1563} 1564 1565/* 1566 * Decrement extra reference on snapshot when last name is removed. 1567 * It will not be freed until the last open reference goes away. 1568 */ 1569void 1570ffs_snapgone(ip) 1571 struct inode *ip; 1572{ 1573 struct inode *xp; 1574 struct fs *fs; 1575 int snaploc; 1576 struct snapdata *sn; 1577 struct ufsmount *ump; 1578 1579 /* 1580 * Find snapshot in incore list. 1581 */ 1582 xp = NULL; 1583 sn = ITODEVVP(ip)->v_rdev->si_snapdata; 1584 if (sn != NULL) 1585 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) 1586 if (xp == ip) 1587 break; 1588 if (xp != NULL) 1589 vrele(ITOV(ip)); 1590 else if (snapdebug) 1591 printf("ffs_snapgone: lost snapshot vnode %ju\n", 1592 (uintmax_t)ip->i_number); 1593 /* 1594 * Delete snapshot inode from superblock. Keep list dense. 1595 */ 1596 ump = ITOUMP(ip); 1597 fs = ump->um_fs; 1598 UFS_LOCK(ump); 1599 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 1600 if (fs->fs_snapinum[snaploc] == ip->i_number) 1601 break; 1602 if (snaploc < FSMAXSNAP) { 1603 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) { 1604 if (fs->fs_snapinum[snaploc] == 0) 1605 break; 1606 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc]; 1607 } 1608 fs->fs_snapinum[snaploc - 1] = 0; 1609 } 1610 UFS_UNLOCK(ump); 1611} 1612 1613/* 1614 * Prepare a snapshot file for being removed. 1615 */ 1616void 1617ffs_snapremove(vp) 1618 struct vnode *vp; 1619{ 1620 struct inode *ip; 1621 struct vnode *devvp; 1622 struct buf *ibp; 1623 struct fs *fs; 1624 ufs2_daddr_t numblks, blkno, dblk; 1625 int error, i, last, loc; 1626 struct snapdata *sn; 1627 1628 ip = VTOI(vp); 1629 fs = ITOFS(ip); 1630 devvp = ITODEVVP(ip); 1631 /* 1632 * If active, delete from incore list (this snapshot may 1633 * already have been in the process of being deleted, so 1634 * would not have been active). 1635 * 1636 * Clear copy-on-write flag if last snapshot. 1637 */ 1638 VI_LOCK(devvp); 1639 if (ip->i_nextsnap.tqe_prev != 0) { 1640 sn = devvp->v_rdev->si_snapdata; 1641 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap); 1642 ip->i_nextsnap.tqe_prev = 0; 1643 VI_UNLOCK(devvp); 1644 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); 1645 for (i = 0; i < sn->sn_lock.lk_recurse; i++) 1646 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); 1647 KASSERT(vp->v_vnlock == &sn->sn_lock, 1648 ("ffs_snapremove: lost lock mutation")); 1649 vp->v_vnlock = &vp->v_lock; 1650 VI_LOCK(devvp); 1651 while (sn->sn_lock.lk_recurse > 0) 1652 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 1653 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 1654 try_free_snapdata(devvp); 1655 } else 1656 VI_UNLOCK(devvp); 1657 /* 1658 * Clear all BLK_NOCOPY fields. Pass any block claims to other 1659 * snapshots that want them (see ffs_snapblkfree below). 1660 */ 1661 for (blkno = 1; blkno < UFS_NDADDR; blkno++) { 1662 dblk = DIP(ip, i_db[blkno]); 1663 if (dblk == 0) 1664 continue; 1665 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1666 DIP_SET(ip, i_db[blkno], 0); 1667 else if ((dblk == blkstofrags(fs, blkno) && 1668 ffs_snapblkfree(fs, ITODEVVP(ip), dblk, fs->fs_bsize, 1669 ip->i_number, vp->v_type, NULL))) { 1670 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - 1671 btodb(fs->fs_bsize)); 1672 DIP_SET(ip, i_db[blkno], 0); 1673 } 1674 } 1675 numblks = howmany(ip->i_size, fs->fs_bsize); 1676 for (blkno = UFS_NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 1677 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 1678 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1679 if (error) 1680 continue; 1681 if (fs->fs_size - blkno > NINDIR(fs)) 1682 last = NINDIR(fs); 1683 else 1684 last = fs->fs_size - blkno; 1685 for (loc = 0; loc < last; loc++) { 1686 if (I_IS_UFS1(ip)) { 1687 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc]; 1688 if (dblk == 0) 1689 continue; 1690 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1691 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1692 else if ((dblk == blkstofrags(fs, blkno) && 1693 ffs_snapblkfree(fs, ITODEVVP(ip), dblk, 1694 fs->fs_bsize, ip->i_number, vp->v_type, 1695 NULL))) { 1696 ip->i_din1->di_blocks -= 1697 btodb(fs->fs_bsize); 1698 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1699 } 1700 continue; 1701 } 1702 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc]; 1703 if (dblk == 0) 1704 continue; 1705 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1706 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1707 else if ((dblk == blkstofrags(fs, blkno) && 1708 ffs_snapblkfree(fs, ITODEVVP(ip), dblk, 1709 fs->fs_bsize, ip->i_number, vp->v_type, NULL))) { 1710 ip->i_din2->di_blocks -= btodb(fs->fs_bsize); 1711 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1712 } 1713 } 1714 bawrite(ibp); 1715 } 1716 /* 1717 * Clear snapshot flag and drop reference. 1718 */ 1719 ip->i_flags &= ~SF_SNAPSHOT; 1720 DIP_SET(ip, i_flags, ip->i_flags); 1721 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1722 /* 1723 * The dirtied indirects must be written out before 1724 * softdep_setup_freeblocks() is called. Otherwise indir_trunc() 1725 * may find indirect pointers using the magic BLK_* values. 1726 */ 1727 if (DOINGSOFTDEP(vp)) 1728 ffs_syncvnode(vp, MNT_WAIT, 0); 1729#ifdef QUOTA 1730 /* 1731 * Reenable disk quotas for ex-snapshot file. 1732 */ 1733 if (!getinoquota(ip)) 1734 (void) chkdq(ip, DIP(ip, i_blocks), KERNCRED, FORCE); 1735#endif 1736} 1737 1738/* 1739 * Notification that a block is being freed. Return zero if the free 1740 * should be allowed to proceed. Return non-zero if the snapshot file 1741 * wants to claim the block. The block will be claimed if it is an 1742 * uncopied part of one of the snapshots. It will be freed if it is 1743 * either a BLK_NOCOPY or has already been copied in all of the snapshots. 1744 * If a fragment is being freed, then all snapshots that care about 1745 * it must make a copy since a snapshot file can only claim full sized 1746 * blocks. Note that if more than one snapshot file maps the block, 1747 * we can pick one at random to claim it. Since none of the snapshots 1748 * can change, we are assurred that they will all see the same unmodified 1749 * image. When deleting a snapshot file (see ffs_snapremove above), we 1750 * must push any of these claimed blocks to one of the other snapshots 1751 * that maps it. These claimed blocks are easily identified as they will 1752 * have a block number equal to their logical block number within the 1753 * snapshot. A copied block can never have this property because they 1754 * must always have been allocated from a BLK_NOCOPY location. 1755 */ 1756int 1757ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd) 1758 struct fs *fs; 1759 struct vnode *devvp; 1760 ufs2_daddr_t bno; 1761 long size; 1762 ino_t inum; 1763 enum vtype vtype; 1764 struct workhead *wkhd; 1765{ 1766 struct buf *ibp, *cbp, *savedcbp = NULL; 1767 struct thread *td = curthread; 1768 struct inode *ip; 1769 struct vnode *vp = NULL; 1770 ufs_lbn_t lbn; 1771 ufs2_daddr_t blkno; 1772 int indiroff = 0, error = 0, claimedblk = 0; 1773 struct snapdata *sn; 1774 1775 lbn = fragstoblks(fs, bno); 1776retry: 1777 VI_LOCK(devvp); 1778 sn = devvp->v_rdev->si_snapdata; 1779 if (sn == NULL) { 1780 VI_UNLOCK(devvp); 1781 return (0); 1782 } 1783 if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 1784 VI_MTX(devvp)) != 0) 1785 goto retry; 1786 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 1787 vp = ITOV(ip); 1788 if (DOINGSOFTDEP(vp)) 1789 softdep_prealloc(vp, MNT_WAIT); 1790 /* 1791 * Lookup block being written. 1792 */ 1793 if (lbn < UFS_NDADDR) { 1794 blkno = DIP(ip, i_db[lbn]); 1795 } else { 1796 td->td_pflags |= TDP_COWINPROGRESS; 1797 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1798 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1799 td->td_pflags &= ~TDP_COWINPROGRESS; 1800 if (error) 1801 break; 1802 indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); 1803 if (I_IS_UFS1(ip)) 1804 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 1805 else 1806 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 1807 } 1808 /* 1809 * Check to see if block needs to be copied. 1810 */ 1811 if (blkno == 0) { 1812 /* 1813 * A block that we map is being freed. If it has not 1814 * been claimed yet, we will claim or copy it (below). 1815 */ 1816 claimedblk = 1; 1817 } else if (blkno == BLK_SNAP) { 1818 /* 1819 * No previous snapshot claimed the block, 1820 * so it will be freed and become a BLK_NOCOPY 1821 * (don't care) for us. 1822 */ 1823 if (claimedblk) 1824 panic("snapblkfree: inconsistent block type"); 1825 if (lbn < UFS_NDADDR) { 1826 DIP_SET(ip, i_db[lbn], BLK_NOCOPY); 1827 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1828 } else if (I_IS_UFS1(ip)) { 1829 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 1830 BLK_NOCOPY; 1831 bdwrite(ibp); 1832 } else { 1833 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 1834 BLK_NOCOPY; 1835 bdwrite(ibp); 1836 } 1837 continue; 1838 } else /* BLK_NOCOPY or default */ { 1839 /* 1840 * If the snapshot has already copied the block 1841 * (default), or does not care about the block, 1842 * it is not needed. 1843 */ 1844 if (lbn >= UFS_NDADDR) 1845 bqrelse(ibp); 1846 continue; 1847 } 1848 /* 1849 * If this is a full size block, we will just grab it 1850 * and assign it to the snapshot inode. Otherwise we 1851 * will proceed to copy it. See explanation for this 1852 * routine as to why only a single snapshot needs to 1853 * claim this block. 1854 */ 1855 if (size == fs->fs_bsize) { 1856#ifdef DEBUG 1857 if (snapdebug) 1858 printf("%s %ju lbn %jd from inum %ju\n", 1859 "Grabonremove: snapino", 1860 (uintmax_t)ip->i_number, 1861 (intmax_t)lbn, (uintmax_t)inum); 1862#endif 1863 /* 1864 * If journaling is tracking this write we must add 1865 * the work to the inode or indirect being written. 1866 */ 1867 if (wkhd != NULL) { 1868 if (lbn < UFS_NDADDR) 1869 softdep_inode_append(ip, 1870 curthread->td_ucred, wkhd); 1871 else 1872 softdep_buf_append(ibp, wkhd); 1873 } 1874 if (lbn < UFS_NDADDR) { 1875 DIP_SET(ip, i_db[lbn], bno); 1876 } else if (I_IS_UFS1(ip)) { 1877 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno; 1878 bdwrite(ibp); 1879 } else { 1880 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno; 1881 bdwrite(ibp); 1882 } 1883 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size)); 1884 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1885 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 1886 return (1); 1887 } 1888 if (lbn >= UFS_NDADDR) 1889 bqrelse(ibp); 1890 /* 1891 * Allocate the block into which to do the copy. Note that this 1892 * allocation will never require any additional allocations for 1893 * the snapshot inode. 1894 */ 1895 td->td_pflags |= TDP_COWINPROGRESS; 1896 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1897 fs->fs_bsize, KERNCRED, 0, &cbp); 1898 td->td_pflags &= ~TDP_COWINPROGRESS; 1899 if (error) 1900 break; 1901#ifdef DEBUG 1902 if (snapdebug) 1903 printf("%s%ju lbn %jd %s %ju size %ld to blkno %jd\n", 1904 "Copyonremove: snapino ", (uintmax_t)ip->i_number, 1905 (intmax_t)lbn, "for inum", (uintmax_t)inum, size, 1906 (intmax_t)cbp->b_blkno); 1907#endif 1908 /* 1909 * If we have already read the old block contents, then 1910 * simply copy them to the new block. Note that we need 1911 * to synchronously write snapshots that have not been 1912 * unlinked, and hence will be visible after a crash, 1913 * to ensure their integrity. At a minimum we ensure the 1914 * integrity of the filesystem metadata, but use the 1915 * dopersistence sysctl-setable flag to decide on the 1916 * persistence needed for file content data. 1917 */ 1918 if (savedcbp != NULL) { 1919 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 1920 bawrite(cbp); 1921 if ((vtype == VDIR || dopersistence) && 1922 ip->i_effnlink > 0) 1923 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 1924 continue; 1925 } 1926 /* 1927 * Otherwise, read the old block contents into the buffer. 1928 */ 1929 if ((error = readblock(vp, cbp, lbn)) != 0) { 1930 bzero(cbp->b_data, fs->fs_bsize); 1931 bawrite(cbp); 1932 if ((vtype == VDIR || dopersistence) && 1933 ip->i_effnlink > 0) 1934 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 1935 break; 1936 } 1937 savedcbp = cbp; 1938 } 1939 /* 1940 * Note that we need to synchronously write snapshots that 1941 * have not been unlinked, and hence will be visible after 1942 * a crash, to ensure their integrity. At a minimum we 1943 * ensure the integrity of the filesystem metadata, but 1944 * use the dopersistence sysctl-setable flag to decide on 1945 * the persistence needed for file content data. 1946 */ 1947 if (savedcbp) { 1948 vp = savedcbp->b_vp; 1949 bawrite(savedcbp); 1950 if ((vtype == VDIR || dopersistence) && 1951 VTOI(vp)->i_effnlink > 0) 1952 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 1953 } 1954 /* 1955 * If we have been unable to allocate a block in which to do 1956 * the copy, then return non-zero so that the fragment will 1957 * not be freed. Although space will be lost, the snapshot 1958 * will stay consistent. 1959 */ 1960 if (error != 0 && wkhd != NULL) 1961 softdep_freework(wkhd); 1962 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 1963 return (error); 1964} 1965 1966/* 1967 * Associate snapshot files when mounting. 1968 */ 1969void 1970ffs_snapshot_mount(mp) 1971 struct mount *mp; 1972{ 1973 struct ufsmount *ump = VFSTOUFS(mp); 1974 struct vnode *devvp = ump->um_devvp; 1975 struct fs *fs = ump->um_fs; 1976 struct thread *td = curthread; 1977 struct snapdata *sn; 1978 struct vnode *vp; 1979 struct vnode *lastvp; 1980 struct inode *ip; 1981 struct uio auio; 1982 struct iovec aiov; 1983 void *snapblklist; 1984 char *reason; 1985 daddr_t snaplistsize; 1986 int error, snaploc, loc; 1987 1988 /* 1989 * XXX The following needs to be set before ffs_truncate or 1990 * VOP_READ can be called. 1991 */ 1992 mp->mnt_stat.f_iosize = fs->fs_bsize; 1993 /* 1994 * Process each snapshot listed in the superblock. 1995 */ 1996 vp = NULL; 1997 lastvp = NULL; 1998 sn = NULL; 1999 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) { 2000 if (fs->fs_snapinum[snaploc] == 0) 2001 break; 2002 if ((error = ffs_vget(mp, fs->fs_snapinum[snaploc], 2003 LK_EXCLUSIVE, &vp)) != 0){ 2004 printf("ffs_snapshot_mount: vget failed %d\n", error); 2005 continue; 2006 } 2007 ip = VTOI(vp); 2008 if (vp->v_type != VREG) { 2009 reason = "non-file snapshot"; 2010 } else if (!IS_SNAPSHOT(ip)) { 2011 reason = "non-snapshot"; 2012 } else if (ip->i_size == 2013 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) { 2014 reason = "old format snapshot"; 2015 (void)ffs_truncate(vp, (off_t)0, 0, NOCRED); 2016 (void)ffs_syncvnode(vp, MNT_WAIT, 0); 2017 } else { 2018 reason = NULL; 2019 } 2020 if (reason != NULL) { 2021 printf("ffs_snapshot_mount: %s inode %d\n", 2022 reason, fs->fs_snapinum[snaploc]); 2023 vput(vp); 2024 vp = NULL; 2025 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) { 2026 if (fs->fs_snapinum[loc] == 0) 2027 break; 2028 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc]; 2029 } 2030 fs->fs_snapinum[loc - 1] = 0; 2031 snaploc--; 2032 continue; 2033 } 2034 /* 2035 * Acquire a lock on the snapdata structure, creating it if 2036 * necessary. 2037 */ 2038 sn = ffs_snapdata_acquire(devvp); 2039 /* 2040 * Change vnode to use shared snapshot lock instead of the 2041 * original private lock. 2042 */ 2043 vp->v_vnlock = &sn->sn_lock; 2044 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 2045 /* 2046 * Link it onto the active snapshot list. 2047 */ 2048 VI_LOCK(devvp); 2049 if (ip->i_nextsnap.tqe_prev != 0) 2050 panic("ffs_snapshot_mount: %ju already on list", 2051 (uintmax_t)ip->i_number); 2052 else 2053 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); 2054 vp->v_vflag |= VV_SYSTEM; 2055 VI_UNLOCK(devvp); 2056 VOP_UNLOCK(vp, 0); 2057 lastvp = vp; 2058 } 2059 vp = lastvp; 2060 /* 2061 * No usable snapshots found. 2062 */ 2063 if (sn == NULL || vp == NULL) 2064 return; 2065 /* 2066 * Allocate the space for the block hints list. We always want to 2067 * use the list from the newest snapshot. 2068 */ 2069 auio.uio_iov = &aiov; 2070 auio.uio_iovcnt = 1; 2071 aiov.iov_base = (void *)&snaplistsize; 2072 aiov.iov_len = sizeof(snaplistsize); 2073 auio.uio_resid = aiov.iov_len; 2074 auio.uio_offset = 2075 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)); 2076 auio.uio_segflg = UIO_SYSSPACE; 2077 auio.uio_rw = UIO_READ; 2078 auio.uio_td = td; 2079 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2080 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 2081 printf("ffs_snapshot_mount: read_1 failed %d\n", error); 2082 VOP_UNLOCK(vp, 0); 2083 return; 2084 } 2085 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 2086 M_UFSMNT, M_WAITOK); 2087 auio.uio_iovcnt = 1; 2088 aiov.iov_base = snapblklist; 2089 aiov.iov_len = snaplistsize * sizeof (daddr_t); 2090 auio.uio_resid = aiov.iov_len; 2091 auio.uio_offset -= sizeof(snaplistsize); 2092 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 2093 printf("ffs_snapshot_mount: read_2 failed %d\n", error); 2094 VOP_UNLOCK(vp, 0); 2095 free(snapblklist, M_UFSMNT); 2096 return; 2097 } 2098 VOP_UNLOCK(vp, 0); 2099 VI_LOCK(devvp); 2100 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount"); 2101 sn->sn_listsize = snaplistsize; 2102 sn->sn_blklist = (daddr_t *)snapblklist; 2103 devvp->v_vflag |= VV_COPYONWRITE; 2104 VI_UNLOCK(devvp); 2105} 2106 2107/* 2108 * Disassociate snapshot files when unmounting. 2109 */ 2110void 2111ffs_snapshot_unmount(mp) 2112 struct mount *mp; 2113{ 2114 struct vnode *devvp = VFSTOUFS(mp)->um_devvp; 2115 struct snapdata *sn; 2116 struct inode *xp; 2117 struct vnode *vp; 2118 2119 VI_LOCK(devvp); 2120 sn = devvp->v_rdev->si_snapdata; 2121 while (sn != NULL && (xp = TAILQ_FIRST(&sn->sn_head)) != NULL) { 2122 vp = ITOV(xp); 2123 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap); 2124 xp->i_nextsnap.tqe_prev = 0; 2125 lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE, 2126 VI_MTX(devvp)); 2127 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); 2128 KASSERT(vp->v_vnlock == &sn->sn_lock, 2129 ("ffs_snapshot_unmount: lost lock mutation")); 2130 vp->v_vnlock = &vp->v_lock; 2131 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 2132 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2133 if (xp->i_effnlink > 0) 2134 vrele(vp); 2135 VI_LOCK(devvp); 2136 sn = devvp->v_rdev->si_snapdata; 2137 } 2138 try_free_snapdata(devvp); 2139 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount"); 2140} 2141 2142/* 2143 * Check the buffer block to be belong to device buffer that shall be 2144 * locked after snaplk. devvp shall be locked on entry, and will be 2145 * leaved locked upon exit. 2146 */ 2147static int 2148ffs_bp_snapblk(devvp, bp) 2149 struct vnode *devvp; 2150 struct buf *bp; 2151{ 2152 struct snapdata *sn; 2153 struct fs *fs; 2154 ufs2_daddr_t lbn, *snapblklist; 2155 int lower, upper, mid; 2156 2157 ASSERT_VI_LOCKED(devvp, "ffs_bp_snapblk"); 2158 KASSERT(devvp->v_type == VCHR, ("Not a device %p", devvp)); 2159 sn = devvp->v_rdev->si_snapdata; 2160 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) == NULL) 2161 return (0); 2162 fs = ITOFS(TAILQ_FIRST(&sn->sn_head)); 2163 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); 2164 snapblklist = sn->sn_blklist; 2165 upper = sn->sn_listsize - 1; 2166 lower = 1; 2167 while (lower <= upper) { 2168 mid = (lower + upper) / 2; 2169 if (snapblklist[mid] == lbn) 2170 break; 2171 if (snapblklist[mid] < lbn) 2172 lower = mid + 1; 2173 else 2174 upper = mid - 1; 2175 } 2176 if (lower <= upper) 2177 return (1); 2178 return (0); 2179} 2180 2181void 2182ffs_bdflush(bo, bp) 2183 struct bufobj *bo; 2184 struct buf *bp; 2185{ 2186 struct thread *td; 2187 struct vnode *vp, *devvp; 2188 struct buf *nbp; 2189 int bp_bdskip; 2190 2191 if (bo->bo_dirty.bv_cnt <= dirtybufthresh) 2192 return; 2193 2194 td = curthread; 2195 vp = bp->b_vp; 2196 devvp = bo2vnode(bo); 2197 KASSERT(vp == devvp, ("devvp != vp %p %p", bo, bp)); 2198 2199 VI_LOCK(devvp); 2200 bp_bdskip = ffs_bp_snapblk(devvp, bp); 2201 if (bp_bdskip) 2202 bdwriteskip++; 2203 VI_UNLOCK(devvp); 2204 if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10 && !bp_bdskip) { 2205 (void) VOP_FSYNC(vp, MNT_NOWAIT, td); 2206 altbufferflushes++; 2207 } else { 2208 BO_LOCK(bo); 2209 /* 2210 * Try to find a buffer to flush. 2211 */ 2212 TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { 2213 if ((nbp->b_vflags & BV_BKGRDINPROG) || 2214 BUF_LOCK(nbp, 2215 LK_EXCLUSIVE | LK_NOWAIT, NULL)) 2216 continue; 2217 if (bp == nbp) 2218 panic("bdwrite: found ourselves"); 2219 BO_UNLOCK(bo); 2220 /* 2221 * Don't countdeps with the bo lock 2222 * held. 2223 */ 2224 if (buf_countdeps(nbp, 0)) { 2225 BO_LOCK(bo); 2226 BUF_UNLOCK(nbp); 2227 continue; 2228 } 2229 if (bp_bdskip) { 2230 VI_LOCK(devvp); 2231 if (!ffs_bp_snapblk(vp, nbp)) { 2232 VI_UNLOCK(devvp); 2233 BO_LOCK(bo); 2234 BUF_UNLOCK(nbp); 2235 continue; 2236 } 2237 VI_UNLOCK(devvp); 2238 } 2239 if (nbp->b_flags & B_CLUSTEROK) { 2240 vfs_bio_awrite(nbp); 2241 } else { 2242 bremfree(nbp); 2243 bawrite(nbp); 2244 } 2245 dirtybufferflushes++; 2246 break; 2247 } 2248 if (nbp == NULL) 2249 BO_UNLOCK(bo); 2250 } 2251} 2252 2253/* 2254 * Check for need to copy block that is about to be written, 2255 * copying the block if necessary. 2256 */ 2257int 2258ffs_copyonwrite(devvp, bp) 2259 struct vnode *devvp; 2260 struct buf *bp; 2261{ 2262 struct snapdata *sn; 2263 struct buf *ibp, *cbp, *savedcbp = NULL; 2264 struct thread *td = curthread; 2265 struct fs *fs; 2266 struct inode *ip; 2267 struct vnode *vp = NULL; 2268 ufs2_daddr_t lbn, blkno, *snapblklist; 2269 int lower, upper, mid, indiroff, error = 0; 2270 int launched_async_io, prev_norunningbuf; 2271 long saved_runningbufspace; 2272 2273 if (devvp != bp->b_vp && IS_SNAPSHOT(VTOI(bp->b_vp))) 2274 return (0); /* Update on a snapshot file */ 2275 if (td->td_pflags & TDP_COWINPROGRESS) 2276 panic("ffs_copyonwrite: recursive call"); 2277 /* 2278 * First check to see if it is in the preallocated list. 2279 * By doing this check we avoid several potential deadlocks. 2280 */ 2281 VI_LOCK(devvp); 2282 sn = devvp->v_rdev->si_snapdata; 2283 if (sn == NULL || 2284 TAILQ_EMPTY(&sn->sn_head)) { 2285 VI_UNLOCK(devvp); 2286 return (0); /* No snapshot */ 2287 } 2288 ip = TAILQ_FIRST(&sn->sn_head); 2289 fs = ITOFS(ip); 2290 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); 2291 if (lbn < UFS_NDADDR) { 2292 VI_UNLOCK(devvp); 2293 return (0); /* Direct blocks are always copied */ 2294 } 2295 snapblklist = sn->sn_blklist; 2296 upper = sn->sn_listsize - 1; 2297 lower = 1; 2298 while (lower <= upper) { 2299 mid = (lower + upper) / 2; 2300 if (snapblklist[mid] == lbn) 2301 break; 2302 if (snapblklist[mid] < lbn) 2303 lower = mid + 1; 2304 else 2305 upper = mid - 1; 2306 } 2307 if (lower <= upper) { 2308 VI_UNLOCK(devvp); 2309 return (0); 2310 } 2311 launched_async_io = 0; 2312 prev_norunningbuf = td->td_pflags & TDP_NORUNNINGBUF; 2313 /* 2314 * Since I/O on bp isn't yet in progress and it may be blocked 2315 * for a long time waiting on snaplk, back it out of 2316 * runningbufspace, possibly waking other threads waiting for space. 2317 */ 2318 saved_runningbufspace = bp->b_runningbufspace; 2319 if (saved_runningbufspace != 0) 2320 runningbufwakeup(bp); 2321 /* 2322 * Not in the precomputed list, so check the snapshots. 2323 */ 2324 while (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 2325 VI_MTX(devvp)) != 0) { 2326 VI_LOCK(devvp); 2327 sn = devvp->v_rdev->si_snapdata; 2328 if (sn == NULL || 2329 TAILQ_EMPTY(&sn->sn_head)) { 2330 VI_UNLOCK(devvp); 2331 if (saved_runningbufspace != 0) { 2332 bp->b_runningbufspace = saved_runningbufspace; 2333 atomic_add_long(&runningbufspace, 2334 bp->b_runningbufspace); 2335 } 2336 return (0); /* Snapshot gone */ 2337 } 2338 } 2339 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 2340 vp = ITOV(ip); 2341 if (DOINGSOFTDEP(vp)) 2342 softdep_prealloc(vp, MNT_WAIT); 2343 /* 2344 * We ensure that everything of our own that needs to be 2345 * copied will be done at the time that ffs_snapshot is 2346 * called. Thus we can skip the check here which can 2347 * deadlock in doing the lookup in UFS_BALLOC. 2348 */ 2349 if (bp->b_vp == vp) 2350 continue; 2351 /* 2352 * Check to see if block needs to be copied. We do not have 2353 * to hold the snapshot lock while doing this lookup as it 2354 * will never require any additional allocations for the 2355 * snapshot inode. 2356 */ 2357 if (lbn < UFS_NDADDR) { 2358 blkno = DIP(ip, i_db[lbn]); 2359 } else { 2360 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; 2361 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 2362 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 2363 td->td_pflags &= ~TDP_COWINPROGRESS; 2364 if (error) 2365 break; 2366 indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); 2367 if (I_IS_UFS1(ip)) 2368 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 2369 else 2370 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 2371 bqrelse(ibp); 2372 } 2373#ifdef INVARIANTS 2374 if (blkno == BLK_SNAP && bp->b_lblkno >= 0) 2375 panic("ffs_copyonwrite: bad copy block"); 2376#endif 2377 if (blkno != 0) 2378 continue; 2379 /* 2380 * Allocate the block into which to do the copy. Since 2381 * multiple processes may all try to copy the same block, 2382 * we have to recheck our need to do a copy if we sleep 2383 * waiting for the lock. 2384 * 2385 * Because all snapshots on a filesystem share a single 2386 * lock, we ensure that we will never be in competition 2387 * with another process to allocate a block. 2388 */ 2389 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; 2390 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 2391 fs->fs_bsize, KERNCRED, 0, &cbp); 2392 td->td_pflags &= ~TDP_COWINPROGRESS; 2393 if (error) 2394 break; 2395#ifdef DEBUG 2396 if (snapdebug) { 2397 printf("Copyonwrite: snapino %ju lbn %jd for ", 2398 (uintmax_t)ip->i_number, (intmax_t)lbn); 2399 if (bp->b_vp == devvp) 2400 printf("fs metadata"); 2401 else 2402 printf("inum %ju", 2403 (uintmax_t)VTOI(bp->b_vp)->i_number); 2404 printf(" lblkno %jd to blkno %jd\n", 2405 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno); 2406 } 2407#endif 2408 /* 2409 * If we have already read the old block contents, then 2410 * simply copy them to the new block. Note that we need 2411 * to synchronously write snapshots that have not been 2412 * unlinked, and hence will be visible after a crash, 2413 * to ensure their integrity. At a minimum we ensure the 2414 * integrity of the filesystem metadata, but use the 2415 * dopersistence sysctl-setable flag to decide on the 2416 * persistence needed for file content data. 2417 */ 2418 if (savedcbp != NULL) { 2419 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 2420 bawrite(cbp); 2421 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || 2422 dopersistence) && ip->i_effnlink > 0) 2423 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 2424 else 2425 launched_async_io = 1; 2426 continue; 2427 } 2428 /* 2429 * Otherwise, read the old block contents into the buffer. 2430 */ 2431 if ((error = readblock(vp, cbp, lbn)) != 0) { 2432 bzero(cbp->b_data, fs->fs_bsize); 2433 bawrite(cbp); 2434 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || 2435 dopersistence) && ip->i_effnlink > 0) 2436 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 2437 else 2438 launched_async_io = 1; 2439 break; 2440 } 2441 savedcbp = cbp; 2442 } 2443 /* 2444 * Note that we need to synchronously write snapshots that 2445 * have not been unlinked, and hence will be visible after 2446 * a crash, to ensure their integrity. At a minimum we 2447 * ensure the integrity of the filesystem metadata, but 2448 * use the dopersistence sysctl-setable flag to decide on 2449 * the persistence needed for file content data. 2450 */ 2451 if (savedcbp) { 2452 vp = savedcbp->b_vp; 2453 bawrite(savedcbp); 2454 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || 2455 dopersistence) && VTOI(vp)->i_effnlink > 0) 2456 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 2457 else 2458 launched_async_io = 1; 2459 } 2460 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 2461 td->td_pflags = (td->td_pflags & ~TDP_NORUNNINGBUF) | 2462 prev_norunningbuf; 2463 if (launched_async_io && (td->td_pflags & TDP_NORUNNINGBUF) == 0) 2464 waitrunningbufspace(); 2465 /* 2466 * I/O on bp will now be started, so count it in runningbufspace. 2467 */ 2468 if (saved_runningbufspace != 0) { 2469 bp->b_runningbufspace = saved_runningbufspace; 2470 atomic_add_long(&runningbufspace, bp->b_runningbufspace); 2471 } 2472 return (error); 2473} 2474 2475/* 2476 * sync snapshots to force freework records waiting on snapshots to claim 2477 * blocks to free. 2478 */ 2479void 2480ffs_sync_snap(mp, waitfor) 2481 struct mount *mp; 2482 int waitfor; 2483{ 2484 struct snapdata *sn; 2485 struct vnode *devvp; 2486 struct vnode *vp; 2487 struct inode *ip; 2488 2489 devvp = VFSTOUFS(mp)->um_devvp; 2490 if ((devvp->v_vflag & VV_COPYONWRITE) == 0) 2491 return; 2492 for (;;) { 2493 VI_LOCK(devvp); 2494 sn = devvp->v_rdev->si_snapdata; 2495 if (sn == NULL) { 2496 VI_UNLOCK(devvp); 2497 return; 2498 } 2499 if (lockmgr(&sn->sn_lock, 2500 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 2501 VI_MTX(devvp)) == 0) 2502 break; 2503 } 2504 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 2505 vp = ITOV(ip); 2506 ffs_syncvnode(vp, waitfor, NO_INO_UPDT); 2507 } 2508 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2509} 2510 2511/* 2512 * Read the specified block into the given buffer. 2513 * Much of this boiler-plate comes from bwrite(). 2514 */ 2515static int 2516readblock(vp, bp, lbn) 2517 struct vnode *vp; 2518 struct buf *bp; 2519 ufs2_daddr_t lbn; 2520{ 2521 struct inode *ip; 2522 struct bio *bip; 2523 struct fs *fs; 2524 2525 ip = VTOI(vp); 2526 fs = ITOFS(ip); 2527 2528 bip = g_alloc_bio(); 2529 bip->bio_cmd = BIO_READ; 2530 bip->bio_offset = dbtob(fsbtodb(fs, blkstofrags(fs, lbn))); 2531 bip->bio_data = bp->b_data; 2532 bip->bio_length = bp->b_bcount; 2533 bip->bio_done = NULL; 2534 2535 g_io_request(bip, ITODEVVP(ip)->v_bufobj.bo_private); 2536 bp->b_error = biowait(bip, "snaprdb"); 2537 g_destroy_bio(bip); 2538 return (bp->b_error); 2539} 2540 2541#endif 2542 2543/* 2544 * Process file deletes that were deferred by ufs_inactive() due to 2545 * the file system being suspended. Transfer IN_LAZYACCESS into 2546 * IN_MODIFIED for vnodes that were accessed during suspension. 2547 */ 2548void 2549process_deferred_inactive(struct mount *mp) 2550{ 2551 struct vnode *vp, *mvp; 2552 struct inode *ip; 2553 struct thread *td; 2554 int error; 2555 2556 td = curthread; 2557 (void) vn_start_secondary_write(NULL, &mp, V_WAIT); 2558 loop: 2559 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 2560 /* 2561 * IN_LAZYACCESS is checked here without holding any 2562 * vnode lock, but this flag is set only while holding 2563 * vnode interlock. 2564 */ 2565 if (vp->v_type == VNON || 2566 ((VTOI(vp)->i_flag & IN_LAZYACCESS) == 0 && 2567 ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0))) { 2568 VI_UNLOCK(vp); 2569 continue; 2570 } 2571 vholdl(vp); 2572 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2573 if (error != 0) { 2574 vdrop(vp); 2575 if (error == ENOENT) 2576 continue; /* vnode recycled */ 2577 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 2578 goto loop; 2579 } 2580 ip = VTOI(vp); 2581 if ((ip->i_flag & IN_LAZYACCESS) != 0) { 2582 ip->i_flag &= ~IN_LAZYACCESS; 2583 ip->i_flag |= IN_MODIFIED; 2584 } 2585 VI_LOCK(vp); 2586 if ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0) { 2587 VI_UNLOCK(vp); 2588 VOP_UNLOCK(vp, 0); 2589 vdrop(vp); 2590 continue; 2591 } 2592 vinactive(vp, td); 2593 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, 2594 ("process_deferred_inactive: got VI_OWEINACT")); 2595 VI_UNLOCK(vp); 2596 VOP_UNLOCK(vp, 0); 2597 vdrop(vp); 2598 } 2599 vn_finished_secondary_write(mp); 2600} 2601 2602#ifndef NO_FFS_SNAPSHOT 2603 2604static struct snapdata * 2605ffs_snapdata_alloc(void) 2606{ 2607 struct snapdata *sn; 2608 2609 /* 2610 * Fetch a snapdata from the free list if there is one available. 2611 */ 2612 mtx_lock(&snapfree_lock); 2613 sn = LIST_FIRST(&snapfree); 2614 if (sn != NULL) 2615 LIST_REMOVE(sn, sn_link); 2616 mtx_unlock(&snapfree_lock); 2617 if (sn != NULL) 2618 return (sn); 2619 /* 2620 * If there were no free snapdatas allocate one. 2621 */ 2622 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO); 2623 TAILQ_INIT(&sn->sn_head); 2624 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT, 2625 LK_CANRECURSE | LK_NOSHARE); 2626 return (sn); 2627} 2628 2629/* 2630 * The snapdata is never freed because we can not be certain that 2631 * there are no threads sleeping on the snap lock. Persisting 2632 * them permanently avoids costly synchronization in ffs_lock(). 2633 */ 2634static void 2635ffs_snapdata_free(struct snapdata *sn) 2636{ 2637 mtx_lock(&snapfree_lock); 2638 LIST_INSERT_HEAD(&snapfree, sn, sn_link); 2639 mtx_unlock(&snapfree_lock); 2640} 2641 2642/* Try to free snapdata associated with devvp */ 2643static void 2644try_free_snapdata(struct vnode *devvp) 2645{ 2646 struct snapdata *sn; 2647 ufs2_daddr_t *snapblklist; 2648 2649 ASSERT_VI_LOCKED(devvp, "try_free_snapdata"); 2650 sn = devvp->v_rdev->si_snapdata; 2651 2652 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) != NULL || 2653 (devvp->v_vflag & VV_COPYONWRITE) == 0) { 2654 VI_UNLOCK(devvp); 2655 return; 2656 } 2657 2658 devvp->v_rdev->si_snapdata = NULL; 2659 devvp->v_vflag &= ~VV_COPYONWRITE; 2660 lockmgr(&sn->sn_lock, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp)); 2661 snapblklist = sn->sn_blklist; 2662 sn->sn_blklist = NULL; 2663 sn->sn_listsize = 0; 2664 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2665 if (snapblklist != NULL) 2666 free(snapblklist, M_UFSMNT); 2667 ffs_snapdata_free(sn); 2668} 2669 2670static struct snapdata * 2671ffs_snapdata_acquire(struct vnode *devvp) 2672{ 2673 struct snapdata *nsn, *sn; 2674 int error; 2675 2676 /* 2677 * Allocate a free snapdata. This is done before acquiring the 2678 * devvp lock to avoid allocation while the devvp interlock is 2679 * held. 2680 */ 2681 nsn = ffs_snapdata_alloc(); 2682 2683 for (;;) { 2684 VI_LOCK(devvp); 2685 sn = devvp->v_rdev->si_snapdata; 2686 if (sn == NULL) { 2687 /* 2688 * This is the first snapshot on this 2689 * filesystem and we use our pre-allocated 2690 * snapdata. Publish sn with the sn_lock 2691 * owned by us, to avoid the race. 2692 */ 2693 error = lockmgr(&nsn->sn_lock, LK_EXCLUSIVE | 2694 LK_NOWAIT, NULL); 2695 if (error != 0) 2696 panic("leaked sn, lockmgr error %d", error); 2697 sn = devvp->v_rdev->si_snapdata = nsn; 2698 VI_UNLOCK(devvp); 2699 nsn = NULL; 2700 break; 2701 } 2702 2703 /* 2704 * There is a snapshots which already exists on this 2705 * filesystem, grab a reference to the common lock. 2706 */ 2707 error = lockmgr(&sn->sn_lock, LK_INTERLOCK | 2708 LK_EXCLUSIVE | LK_SLEEPFAIL, VI_MTX(devvp)); 2709 if (error == 0) 2710 break; 2711 } 2712 2713 /* 2714 * Free any unused snapdata. 2715 */ 2716 if (nsn != NULL) 2717 ffs_snapdata_free(nsn); 2718 2719 return (sn); 2720} 2721 2722#endif 2723