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 75int 76ffs_snapshot(mp, snapfile) 77 struct mount *mp; 78 char *snapfile; 79{ 80 return (EINVAL); 81} 82 83int 84ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd) 85 struct fs *fs; 86 struct vnode *devvp; 87 ufs2_daddr_t bno; 88 long size; 89 ino_t inum; 90 enum vtype vtype; 91 struct workhead *wkhd; 92{ 93 return (EINVAL); 94} 95 96void 97ffs_snapremove(vp) 98 struct vnode *vp; 99{ 100} 101 102void 103ffs_snapshot_mount(mp) 104 struct mount *mp; 105{ 106} 107 108void 109ffs_snapshot_unmount(mp) 110 struct mount *mp; 111{ 112} 113 114void 115ffs_snapgone(ip) 116 struct inode *ip; 117{ 118} 119 120int 121ffs_copyonwrite(devvp, bp) 122 struct vnode *devvp; 123 struct buf *bp; 124{ 125 return (EINVAL); 126} 127 128void 129ffs_sync_snap(mp, waitfor) 130 struct mount *mp; 131 int waitfor; 132{ 133} 134 135#else 136FEATURE(ffs_snapshot, "FFS snapshot support"); 137 138LIST_HEAD(, snapdata) snapfree; 139static struct mtx snapfree_lock; 140MTX_SYSINIT(ffs_snapfree, &snapfree_lock, "snapdata free list", MTX_DEF); 141 142static int cgaccount(int, struct vnode *, struct buf *, int); 143static 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, int); 146static 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); 150static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 151 struct fs *, ufs_lbn_t, int); 152static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 153 struct fs *, ufs_lbn_t, int); 154static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 155 struct fs *, ufs_lbn_t, int); 156static 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, int); 159static 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); 163static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 164 struct fs *, ufs_lbn_t, int); 165static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 166 struct fs *, ufs_lbn_t, int); 167static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 168 struct fs *, ufs_lbn_t, int); 169static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t); 170static void try_free_snapdata(struct vnode *devvp); 171static struct snapdata *ffs_snapdata_acquire(struct vnode *devvp); 172static 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 code normally only ensures 179 * persistence for the filesystem metadata contained within a 180 * snapshot. Setting the following flag allows this crash 181 * persistence to be enabled for file contents. 182 */ 183int dopersistence = 0; 184 185#ifdef DEBUG 186#include <sys/sysctl.h> 187SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, ""); 188static int snapdebug = 0; 189SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, ""); 190int collectsnapstats = 0; 191SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats, 192 0, ""); 193#endif /* DEBUG */ 194 195/* 196 * Create a snapshot file and initialize it for the filesystem. 197 */ 198int 199ffs_snapshot(mp, snapfile) 200 struct mount *mp; 201 char *snapfile; 202{ 203 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist; 204 int error, cg, snaploc; 205 int i, size, len, loc; 206 ufs2_daddr_t blockno; 207 uint64_t flag; 208 struct timespec starttime = {0, 0}, endtime; 209 char saved_nice = 0; 210 long redo = 0, snaplistsize = 0; 211 int32_t *lp; 212 void *space; 213 struct fs *copy_fs = NULL, *fs; 214 struct thread *td = curthread; 215 struct inode *ip, *xp; 216 struct buf *bp, *nbp, *ibp; 217 struct nameidata nd; 218 struct mount *wrtmp; 219 struct vattr vat; 220 struct vnode *vp, *xvp, *mvp, *devvp; 221 struct uio auio; 222 struct iovec aiov; 223 struct snapdata *sn; 224 struct ufsmount *ump; 225 226 ump = VFSTOUFS(mp); 227 fs = ump->um_fs; 228 sn = NULL; 229 /* 230 * At the moment, journaled soft updates cannot support 231 * taking snapshots. 232 */ 233 if (MOUNTEDSUJ(mp)) { 234 vfs_mount_error(mp, "%s: Snapshots are not yet supported when " 235 "running with journaled soft updates", fs->fs_fsmnt); 236 return (EOPNOTSUPP); 237 } 238 MNT_ILOCK(mp); 239 flag = mp->mnt_flag; 240 MNT_IUNLOCK(mp); 241 /* 242 * Need to serialize access to snapshot code per filesystem. 243 */ 244 /* 245 * Assign a snapshot slot in the superblock. 246 */ 247 UFS_LOCK(ump); 248 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 249 if (fs->fs_snapinum[snaploc] == 0) 250 break; 251 UFS_UNLOCK(ump); 252 if (snaploc == FSMAXSNAP) 253 return (ENOSPC); 254 /* 255 * Create the snapshot file. 256 */ 257restart: 258 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_SYSSPACE, snapfile, td); 259 if ((error = namei(&nd)) != 0) 260 return (error); 261 if (nd.ni_vp != NULL) { 262 vput(nd.ni_vp); 263 error = EEXIST; 264 } 265 if (nd.ni_dvp->v_mount != mp) 266 error = EXDEV; 267 if (error) { 268 NDFREE(&nd, NDF_ONLY_PNBUF); 269 if (nd.ni_dvp == nd.ni_vp) 270 vrele(nd.ni_dvp); 271 else 272 vput(nd.ni_dvp); 273 return (error); 274 } 275 VATTR_NULL(&vat); 276 vat.va_type = VREG; 277 vat.va_mode = S_IRUSR; 278 vat.va_vaflags |= VA_EXCLUSIVE; 279 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp)) 280 wrtmp = NULL; 281 if (wrtmp != mp) 282 panic("ffs_snapshot: mount mismatch"); 283 vfs_rel(wrtmp); 284 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) { 285 NDFREE(&nd, NDF_ONLY_PNBUF); 286 vput(nd.ni_dvp); 287 if ((error = vn_start_write(NULL, &wrtmp, 288 V_XSLEEP | PCATCH)) != 0) 289 return (error); 290 goto restart; 291 } 292 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat); 293 VOP_UNLOCK(nd.ni_dvp, 0); 294 if (error) { 295 NDFREE(&nd, NDF_ONLY_PNBUF); 296 vn_finished_write(wrtmp); 297 vrele(nd.ni_dvp); 298 return (error); 299 } 300 vp = nd.ni_vp; 301 vp->v_vflag |= VV_SYSTEM; 302 ip = VTOI(vp); 303 devvp = ip->i_devvp; 304 /* 305 * Allocate and copy the last block contents so as to be able 306 * to set size to that of the filesystem. 307 */ 308 numblks = howmany(fs->fs_size, fs->fs_frag); 309 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)), 310 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 311 if (error) 312 goto out; 313 ip->i_size = lblktosize(fs, (off_t)numblks); 314 DIP_SET(ip, i_size, ip->i_size); 315 ip->i_flag |= IN_CHANGE | IN_UPDATE; 316 error = readblock(vp, bp, numblks - 1); 317 bawrite(bp); 318 if (error != 0) 319 goto out; 320 /* 321 * Preallocate critical data structures so that we can copy 322 * them in without further allocation after we suspend all 323 * operations on the filesystem. We would like to just release 324 * the allocated buffers without writing them since they will 325 * be filled in below once we are ready to go, but this upsets 326 * the soft update code, so we go ahead and write the new buffers. 327 * 328 * Allocate all indirect blocks and mark all of them as not 329 * needing to be copied. 330 */ 331 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 332 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 333 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp); 334 if (error) 335 goto out; 336 bawrite(ibp); 337 } 338 /* 339 * Allocate copies for the superblock and its summary information. 340 */ 341 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED, 342 0, &nbp); 343 if (error) 344 goto out; 345 bawrite(nbp); 346 blkno = fragstoblks(fs, fs->fs_csaddr); 347 len = howmany(fs->fs_cssize, fs->fs_bsize); 348 for (loc = 0; loc < len; loc++) { 349 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)), 350 fs->fs_bsize, KERNCRED, 0, &nbp); 351 if (error) 352 goto out; 353 bawrite(nbp); 354 } 355 /* 356 * Allocate all cylinder group blocks. 357 */ 358 for (cg = 0; cg < fs->fs_ncg; cg++) { 359 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 360 fs->fs_bsize, KERNCRED, 0, &nbp); 361 if (error) 362 goto out; 363 bawrite(nbp); 364 if (cg % 10 == 0) 365 ffs_syncvnode(vp, MNT_WAIT, 0); 366 } 367 /* 368 * Copy all the cylinder group maps. Although the 369 * filesystem is still active, we hope that only a few 370 * cylinder groups will change between now and when we 371 * suspend operations. Thus, we will be able to quickly 372 * touch up the few cylinder groups that changed during 373 * the suspension period. 374 */ 375 len = howmany(fs->fs_ncg, NBBY); 376 space = malloc(len, M_DEVBUF, M_WAITOK|M_ZERO); 377 UFS_LOCK(ump); 378 fs->fs_active = space; 379 UFS_UNLOCK(ump); 380 for (cg = 0; cg < fs->fs_ncg; cg++) { 381 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 382 fs->fs_bsize, KERNCRED, 0, &nbp); 383 if (error) 384 goto out; 385 error = cgaccount(cg, vp, nbp, 1); 386 bawrite(nbp); 387 if (cg % 10 == 0) 388 ffs_syncvnode(vp, MNT_WAIT, 0); 389 if (error) 390 goto out; 391 } 392 /* 393 * Change inode to snapshot type file. 394 */ 395 ip->i_flags |= SF_SNAPSHOT; 396 DIP_SET(ip, i_flags, ip->i_flags); 397 ip->i_flag |= IN_CHANGE | IN_UPDATE; 398 /* 399 * Ensure that the snapshot is completely on disk. 400 * Since we have marked it as a snapshot it is safe to 401 * unlock it as no process will be allowed to write to it. 402 */ 403 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 404 goto out; 405 VOP_UNLOCK(vp, 0); 406 /* 407 * All allocations are done, so we can now snapshot the system. 408 * 409 * Recind nice scheduling while running with the filesystem suspended. 410 */ 411 if (td->td_proc->p_nice > 0) { 412 struct proc *p; 413 414 p = td->td_proc; 415 PROC_LOCK(p); 416 saved_nice = p->p_nice; 417 sched_nice(p, 0); 418 PROC_UNLOCK(p); 419 } 420 /* 421 * Suspend operation on filesystem. 422 */ 423 for (;;) { 424 vn_finished_write(wrtmp); 425 if ((error = vfs_write_suspend(vp->v_mount)) != 0) { 426 vn_start_write(NULL, &wrtmp, V_WAIT); 427 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 428 goto out; 429 } 430 if (mp->mnt_kern_flag & MNTK_SUSPENDED) 431 break; 432 vn_start_write(NULL, &wrtmp, V_WAIT); 433 } 434 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 435 if (ip->i_effnlink == 0) { 436 error = ENOENT; /* Snapshot file unlinked */ 437 goto out1; 438 } 439 if (collectsnapstats) 440 nanotime(&starttime); 441 442 /* The last block might have changed. Copy it again to be sure. */ 443 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)), 444 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 445 if (error != 0) 446 goto out1; 447 error = readblock(vp, bp, numblks - 1); 448 bp->b_flags |= B_VALIDSUSPWRT; 449 bawrite(bp); 450 if (error != 0) 451 goto out1; 452 /* 453 * First, copy all the cylinder group maps that have changed. 454 */ 455 for (cg = 0; cg < fs->fs_ncg; cg++) { 456 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0) 457 continue; 458 redo++; 459 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), 460 fs->fs_bsize, KERNCRED, 0, &nbp); 461 if (error) 462 goto out1; 463 error = cgaccount(cg, vp, nbp, 2); 464 bawrite(nbp); 465 if (error) 466 goto out1; 467 } 468 /* 469 * Grab a copy of the superblock and its summary information. 470 * We delay writing it until the suspension is released below. 471 */ 472 copy_fs = malloc((u_long)fs->fs_bsize, M_UFSMNT, M_WAITOK); 473 bcopy(fs, copy_fs, fs->fs_sbsize); 474 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 475 copy_fs->fs_clean = 1; 476 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE; 477 if (fs->fs_sbsize < size) 478 bzero(&((char *)copy_fs)[fs->fs_sbsize], 479 size - fs->fs_sbsize); 480 size = blkroundup(fs, fs->fs_cssize); 481 if (fs->fs_contigsumsize > 0) 482 size += fs->fs_ncg * sizeof(int32_t); 483 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 484 copy_fs->fs_csp = space; 485 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize); 486 space = (char *)space + fs->fs_cssize; 487 loc = howmany(fs->fs_cssize, fs->fs_fsize); 488 i = fs->fs_frag - loc % fs->fs_frag; 489 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize; 490 if (len > 0) { 491 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc), 492 len, KERNCRED, &bp)) != 0) { 493 brelse(bp); 494 free(copy_fs->fs_csp, M_UFSMNT); 495 free(copy_fs, M_UFSMNT); 496 copy_fs = NULL; 497 goto out1; 498 } 499 bcopy(bp->b_data, space, (u_int)len); 500 space = (char *)space + len; 501 bp->b_flags |= B_INVAL | B_NOCACHE; 502 brelse(bp); 503 } 504 if (fs->fs_contigsumsize > 0) { 505 copy_fs->fs_maxcluster = lp = space; 506 for (i = 0; i < fs->fs_ncg; i++) 507 *lp++ = fs->fs_contigsumsize; 508 } 509 /* 510 * We must check for active files that have been unlinked 511 * (e.g., with a zero link count). We have to expunge all 512 * trace of these files from the snapshot so that they are 513 * not reclaimed prematurely by fsck or unnecessarily dumped. 514 * We turn off the MNTK_SUSPENDED flag to avoid a panic from 515 * spec_strategy about writing on a suspended filesystem. 516 * Note that we skip unlinked snapshot files as they will 517 * be handled separately below. 518 * 519 * We also calculate the needed size for the snapshot list. 520 */ 521 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) + 522 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */; 523 MNT_ILOCK(mp); 524 mp->mnt_kern_flag &= ~MNTK_SUSPENDED; 525 MNT_IUNLOCK(mp); 526loop: 527 MNT_VNODE_FOREACH_ALL(xvp, mp, mvp) { 528 if ((xvp->v_usecount == 0 && 529 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) || 530 xvp->v_type == VNON || 531 IS_SNAPSHOT(VTOI(xvp))) { 532 VI_UNLOCK(xvp); 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 continue; 542 } 543 vholdl(xvp); 544 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK) != 0) { 545 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 546 vdrop(xvp); 547 goto loop; 548 } 549 VI_LOCK(xvp); 550 if (xvp->v_usecount == 0 && 551 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) { 552 VI_UNLOCK(xvp); 553 VOP_UNLOCK(xvp, 0); 554 vdrop(xvp); 555 continue; 556 } 557 VI_UNLOCK(xvp); 558 if (snapdebug) 559 vprint("ffs_snapshot: busy vnode", xvp); 560 if (VOP_GETATTR(xvp, &vat, td->td_ucred) == 0 && 561 vat.va_nlink > 0) { 562 VOP_UNLOCK(xvp, 0); 563 vdrop(xvp); 564 continue; 565 } 566 xp = VTOI(xvp); 567 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) { 568 VOP_UNLOCK(xvp, 0); 569 vdrop(xvp); 570 continue; 571 } 572 /* 573 * If there is a fragment, clear it here. 574 */ 575 blkno = 0; 576 loc = howmany(xp->i_size, fs->fs_bsize) - 1; 577 if (loc < NDADDR) { 578 len = fragroundup(fs, blkoff(fs, xp->i_size)); 579 if (len != 0 && len < fs->fs_bsize) { 580 ffs_blkfree(ump, copy_fs, vp, 581 DIP(xp, i_db[loc]), len, xp->i_number, 582 xvp->v_type, NULL); 583 blkno = DIP(xp, i_db[loc]); 584 DIP_SET(xp, i_db[loc], 0); 585 } 586 } 587 snaplistsize += 1; 588 if (xp->i_ump->um_fstype == UFS1) 589 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, 590 BLK_NOCOPY, 1); 591 else 592 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, 593 BLK_NOCOPY, 1); 594 if (blkno) 595 DIP_SET(xp, i_db[loc], blkno); 596 if (!error) 597 error = ffs_freefile(ump, copy_fs, vp, xp->i_number, 598 xp->i_mode, NULL); 599 VOP_UNLOCK(xvp, 0); 600 vdrop(xvp); 601 if (error) { 602 free(copy_fs->fs_csp, M_UFSMNT); 603 free(copy_fs, M_UFSMNT); 604 copy_fs = NULL; 605 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 606 goto out1; 607 } 608 } 609 /* 610 * Erase the journal file from the snapshot. 611 */ 612 if (fs->fs_flags & FS_SUJ) { 613 error = softdep_journal_lookup(mp, &xvp); 614 if (error) { 615 free(copy_fs->fs_csp, M_UFSMNT); 616 free(copy_fs, M_UFSMNT); 617 copy_fs = NULL; 618 goto out1; 619 } 620 xp = VTOI(xvp); 621 if (xp->i_ump->um_fstype == UFS1) 622 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, 623 BLK_NOCOPY, 0); 624 else 625 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, 626 BLK_NOCOPY, 0); 627 vput(xvp); 628 } 629 /* 630 * Acquire a lock on the snapdata structure, creating it if necessary. 631 */ 632 sn = ffs_snapdata_acquire(devvp); 633 /* 634 * Change vnode to use shared snapshot lock instead of the original 635 * private lock. 636 */ 637 vp->v_vnlock = &sn->sn_lock; 638 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 639 xp = TAILQ_FIRST(&sn->sn_head); 640 /* 641 * If this is the first snapshot on this filesystem, then we need 642 * to allocate the space for the list of preallocated snapshot blocks. 643 * This list will be refined below, but this preliminary one will 644 * keep us out of deadlock until the full one is ready. 645 */ 646 if (xp == NULL) { 647 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 648 M_UFSMNT, M_WAITOK); 649 blkp = &snapblklist[1]; 650 *blkp++ = lblkno(fs, fs->fs_sblockloc); 651 blkno = fragstoblks(fs, fs->fs_csaddr); 652 for (cg = 0; cg < fs->fs_ncg; cg++) { 653 if (fragstoblks(fs, cgtod(fs, cg) > blkno)) 654 break; 655 *blkp++ = fragstoblks(fs, cgtod(fs, cg)); 656 } 657 len = howmany(fs->fs_cssize, fs->fs_bsize); 658 for (loc = 0; loc < len; loc++) 659 *blkp++ = blkno + loc; 660 for (; cg < fs->fs_ncg; cg++) 661 *blkp++ = fragstoblks(fs, cgtod(fs, cg)); 662 snapblklist[0] = blkp - snapblklist; 663 VI_LOCK(devvp); 664 if (sn->sn_blklist != NULL) 665 panic("ffs_snapshot: non-empty list"); 666 sn->sn_blklist = snapblklist; 667 sn->sn_listsize = blkp - snapblklist; 668 VI_UNLOCK(devvp); 669 } 670 /* 671 * Record snapshot inode. Since this is the newest snapshot, 672 * it must be placed at the end of the list. 673 */ 674 VI_LOCK(devvp); 675 fs->fs_snapinum[snaploc] = ip->i_number; 676 if (ip->i_nextsnap.tqe_prev != 0) 677 panic("ffs_snapshot: %d already on list", ip->i_number); 678 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); 679 devvp->v_vflag |= VV_COPYONWRITE; 680 VI_UNLOCK(devvp); 681 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp"); 682out1: 683 KASSERT((sn != NULL && copy_fs != NULL && error == 0) || 684 (sn == NULL && copy_fs == NULL && error != 0), 685 ("email phk@ and mckusick@")); 686 /* 687 * Resume operation on filesystem. 688 */ 689 vfs_write_resume_flags(vp->v_mount, VR_START_WRITE | VR_NO_SUSPCLR); 690 if (collectsnapstats && starttime.tv_sec > 0) { 691 nanotime(&endtime); 692 timespecsub(&endtime, &starttime); 693 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n", 694 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec, 695 endtime.tv_nsec / 1000000, redo, fs->fs_ncg); 696 } 697 if (copy_fs == NULL) 698 goto out; 699 /* 700 * Copy allocation information from all the snapshots in 701 * this snapshot and then expunge them from its view. 702 */ 703 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) { 704 if (xp == ip) 705 break; 706 if (xp->i_ump->um_fstype == UFS1) 707 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1, 708 BLK_SNAP, 0); 709 else 710 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2, 711 BLK_SNAP, 0); 712 if (error == 0 && xp->i_effnlink == 0) { 713 error = ffs_freefile(ump, 714 copy_fs, 715 vp, 716 xp->i_number, 717 xp->i_mode, NULL); 718 } 719 if (error) { 720 fs->fs_snapinum[snaploc] = 0; 721 goto done; 722 } 723 } 724 /* 725 * Allocate space for the full list of preallocated snapshot blocks. 726 */ 727 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 728 M_UFSMNT, M_WAITOK); 729 ip->i_snapblklist = &snapblklist[1]; 730 /* 731 * Expunge the blocks used by the snapshots from the set of 732 * blocks marked as used in the snapshot bitmaps. Also, collect 733 * the list of allocated blocks in i_snapblklist. 734 */ 735 if (ip->i_ump->um_fstype == UFS1) 736 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, 737 BLK_SNAP, 0); 738 else 739 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, 740 BLK_SNAP, 0); 741 if (error) { 742 fs->fs_snapinum[snaploc] = 0; 743 free(snapblklist, M_UFSMNT); 744 goto done; 745 } 746 if (snaplistsize < ip->i_snapblklist - snapblklist) 747 panic("ffs_snapshot: list too small"); 748 snaplistsize = ip->i_snapblklist - snapblklist; 749 snapblklist[0] = snaplistsize; 750 ip->i_snapblklist = 0; 751 /* 752 * Write out the list of allocated blocks to the end of the snapshot. 753 */ 754 auio.uio_iov = &aiov; 755 auio.uio_iovcnt = 1; 756 aiov.iov_base = (void *)snapblklist; 757 aiov.iov_len = snaplistsize * sizeof(daddr_t); 758 auio.uio_resid = aiov.iov_len; 759 auio.uio_offset = ip->i_size; 760 auio.uio_segflg = UIO_SYSSPACE; 761 auio.uio_rw = UIO_WRITE; 762 auio.uio_td = td; 763 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 764 fs->fs_snapinum[snaploc] = 0; 765 free(snapblklist, M_UFSMNT); 766 goto done; 767 } 768 /* 769 * Write the superblock and its summary information 770 * to the snapshot. 771 */ 772 blkno = fragstoblks(fs, fs->fs_csaddr); 773 len = howmany(fs->fs_cssize, fs->fs_bsize); 774 space = copy_fs->fs_csp; 775 for (loc = 0; loc < len; loc++) { 776 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp); 777 if (error) { 778 brelse(nbp); 779 fs->fs_snapinum[snaploc] = 0; 780 free(snapblklist, M_UFSMNT); 781 goto done; 782 } 783 bcopy(space, nbp->b_data, fs->fs_bsize); 784 space = (char *)space + fs->fs_bsize; 785 bawrite(nbp); 786 } 787 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize, 788 KERNCRED, &nbp); 789 if (error) { 790 brelse(nbp); 791 } else { 792 loc = blkoff(fs, fs->fs_sblockloc); 793 bcopy((char *)copy_fs, &nbp->b_data[loc], (u_int)fs->fs_sbsize); 794 bawrite(nbp); 795 } 796 /* 797 * As this is the newest list, it is the most inclusive, so 798 * should replace the previous list. 799 */ 800 VI_LOCK(devvp); 801 space = sn->sn_blklist; 802 sn->sn_blklist = snapblklist; 803 sn->sn_listsize = snaplistsize; 804 VI_UNLOCK(devvp); 805 if (space != NULL) 806 free(space, M_UFSMNT); 807 /* 808 * Preallocate all the direct blocks in the snapshot inode so 809 * that we never have to write the inode itself to commit an 810 * update to the contents of the snapshot. Note that once 811 * created, the size of the snapshot will never change, so 812 * there will never be a need to write the inode except to 813 * update the non-integrity-critical time fields and 814 * allocated-block count. 815 */ 816 for (blockno = 0; blockno < NDADDR; blockno++) { 817 if (DIP(ip, i_db[blockno]) != 0) 818 continue; 819 error = UFS_BALLOC(vp, lblktosize(fs, blockno), 820 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); 821 if (error) 822 break; 823 error = readblock(vp, bp, blockno); 824 bawrite(bp); 825 if (error != 0) 826 break; 827 } 828done: 829 free(copy_fs->fs_csp, M_UFSMNT); 830 free(copy_fs, M_UFSMNT); 831 copy_fs = NULL; 832out: 833 NDFREE(&nd, NDF_ONLY_PNBUF); 834 if (saved_nice > 0) { 835 struct proc *p; 836 837 p = td->td_proc; 838 PROC_LOCK(p); 839 sched_nice(td->td_proc, saved_nice); 840 PROC_UNLOCK(td->td_proc); 841 } 842 UFS_LOCK(ump); 843 if (fs->fs_active != 0) { 844 free(fs->fs_active, M_DEVBUF); 845 fs->fs_active = 0; 846 } 847 UFS_UNLOCK(ump); 848 MNT_ILOCK(mp); 849 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 850 MNT_IUNLOCK(mp); 851 if (error) 852 (void) ffs_truncate(vp, (off_t)0, 0, NOCRED, td); 853 (void) ffs_syncvnode(vp, MNT_WAIT, 0); 854 if (error) 855 vput(vp); 856 else 857 VOP_UNLOCK(vp, 0); 858 vrele(nd.ni_dvp); 859 vn_finished_write(wrtmp); 860 process_deferred_inactive(mp); 861 return (error); 862} 863 864/* 865 * Copy a cylinder group map. All the unallocated blocks are marked 866 * BLK_NOCOPY so that the snapshot knows that it need not copy them 867 * if they are later written. If passno is one, then this is a first 868 * pass, so only setting needs to be done. If passno is 2, then this 869 * is a revision to a previous pass which must be undone as the 870 * replacement pass is done. 871 */ 872static int 873cgaccount(cg, vp, nbp, passno) 874 int cg; 875 struct vnode *vp; 876 struct buf *nbp; 877 int passno; 878{ 879 struct buf *bp, *ibp; 880 struct inode *ip; 881 struct cg *cgp; 882 struct fs *fs; 883 ufs2_daddr_t base, numblks; 884 int error, len, loc, indiroff; 885 886 ip = VTOI(vp); 887 fs = ip->i_fs; 888 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 889 (int)fs->fs_cgsize, KERNCRED, &bp); 890 if (error) { 891 brelse(bp); 892 return (error); 893 } 894 cgp = (struct cg *)bp->b_data; 895 if (!cg_chkmagic(cgp)) { 896 brelse(bp); 897 return (EIO); 898 } 899 UFS_LOCK(ip->i_ump); 900 ACTIVESET(fs, cg); 901 /* 902 * Recomputation of summary information might not have been performed 903 * at mount time. Sync up summary information for current cylinder 904 * group while data is in memory to ensure that result of background 905 * fsck is slightly more consistent. 906 */ 907 fs->fs_cs(fs, cg) = cgp->cg_cs; 908 UFS_UNLOCK(ip->i_ump); 909 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize); 910 if (fs->fs_cgsize < fs->fs_bsize) 911 bzero(&nbp->b_data[fs->fs_cgsize], 912 fs->fs_bsize - fs->fs_cgsize); 913 cgp = (struct cg *)nbp->b_data; 914 bqrelse(bp); 915 if (passno == 2) 916 nbp->b_flags |= B_VALIDSUSPWRT; 917 numblks = howmany(fs->fs_size, fs->fs_frag); 918 len = howmany(fs->fs_fpg, fs->fs_frag); 919 base = cgbase(fs, cg) / fs->fs_frag; 920 if (base + len >= numblks) 921 len = numblks - base - 1; 922 loc = 0; 923 if (base < NDADDR) { 924 for ( ; loc < NDADDR; loc++) { 925 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 926 DIP_SET(ip, i_db[loc], BLK_NOCOPY); 927 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY) 928 DIP_SET(ip, i_db[loc], 0); 929 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY) 930 panic("ffs_snapshot: lost direct block"); 931 } 932 } 933 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)), 934 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 935 if (error) { 936 return (error); 937 } 938 indiroff = (base + loc - NDADDR) % NINDIR(fs); 939 for ( ; loc < len; loc++, indiroff++) { 940 if (indiroff >= NINDIR(fs)) { 941 if (passno == 2) 942 ibp->b_flags |= B_VALIDSUSPWRT; 943 bawrite(ibp); 944 error = UFS_BALLOC(vp, 945 lblktosize(fs, (off_t)(base + loc)), 946 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 947 if (error) { 948 return (error); 949 } 950 indiroff = 0; 951 } 952 if (ip->i_ump->um_fstype == UFS1) { 953 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 954 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 955 BLK_NOCOPY; 956 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data)) 957 [indiroff] == BLK_NOCOPY) 958 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0; 959 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data)) 960 [indiroff] == BLK_NOCOPY) 961 panic("ffs_snapshot: lost indirect block"); 962 continue; 963 } 964 if (ffs_isblock(fs, cg_blksfree(cgp), loc)) 965 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; 966 else if (passno == 2 && 967 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 968 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0; 969 else if (passno == 1 && 970 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) 971 panic("ffs_snapshot: lost indirect block"); 972 } 973 if (passno == 2) 974 ibp->b_flags |= B_VALIDSUSPWRT; 975 bdwrite(ibp); 976 return (0); 977} 978 979/* 980 * Before expunging a snapshot inode, note all the 981 * blocks that it claims with BLK_SNAP so that fsck will 982 * be able to account for those blocks properly and so 983 * that this snapshot knows that it need not copy them 984 * if the other snapshot holding them is freed. This code 985 * is reproduced once each for UFS1 and UFS2. 986 */ 987static int 988expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype, clearmode) 989 struct vnode *snapvp; 990 struct inode *cancelip; 991 struct fs *fs; 992 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 993 struct fs *, ufs_lbn_t, int); 994 int expungetype; 995 int clearmode; 996{ 997 int i, error, indiroff; 998 ufs_lbn_t lbn, rlbn; 999 ufs2_daddr_t len, blkno, numblks, blksperindir; 1000 struct ufs1_dinode *dip; 1001 struct thread *td = curthread; 1002 struct buf *bp; 1003 1004 /* 1005 * Prepare to expunge the inode. If its inode block has not 1006 * yet been copied, then allocate and fill the copy. 1007 */ 1008 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 1009 blkno = 0; 1010 if (lbn < NDADDR) { 1011 blkno = VTOI(snapvp)->i_din1->di_db[lbn]; 1012 } else { 1013 if (DOINGSOFTDEP(snapvp)) 1014 softdep_prealloc(snapvp, MNT_WAIT); 1015 td->td_pflags |= TDP_COWINPROGRESS; 1016 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), 1017 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 1018 td->td_pflags &= ~TDP_COWINPROGRESS; 1019 if (error) 1020 return (error); 1021 indiroff = (lbn - NDADDR) % NINDIR(fs); 1022 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; 1023 bqrelse(bp); 1024 } 1025 if (blkno != 0) { 1026 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 1027 return (error); 1028 } else { 1029 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), 1030 fs->fs_bsize, KERNCRED, 0, &bp); 1031 if (error) 1032 return (error); 1033 if ((error = readblock(snapvp, bp, lbn)) != 0) 1034 return (error); 1035 } 1036 /* 1037 * Set a snapshot inode to be a zero length file, regular files 1038 * or unlinked snapshots to be completely unallocated. 1039 */ 1040 dip = (struct ufs1_dinode *)bp->b_data + 1041 ino_to_fsbo(fs, cancelip->i_number); 1042 if (clearmode || cancelip->i_effnlink == 0) 1043 dip->di_mode = 0; 1044 dip->di_size = 0; 1045 dip->di_blocks = 0; 1046 dip->di_flags &= ~SF_SNAPSHOT; 1047 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t)); 1048 bdwrite(bp); 1049 /* 1050 * Now go through and expunge all the blocks in the file 1051 * using the function requested. 1052 */ 1053 numblks = howmany(cancelip->i_size, fs->fs_bsize); 1054 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0], 1055 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype))) 1056 return (error); 1057 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0], 1058 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype))) 1059 return (error); 1060 blksperindir = 1; 1061 lbn = -NDADDR; 1062 len = numblks - NDADDR; 1063 rlbn = NDADDR; 1064 for (i = 0; len > 0 && i < NIADDR; i++) { 1065 error = indiracct_ufs1(snapvp, ITOV(cancelip), i, 1066 cancelip->i_din1->di_ib[i], lbn, rlbn, len, 1067 blksperindir, fs, acctfunc, expungetype); 1068 if (error) 1069 return (error); 1070 blksperindir *= NINDIR(fs); 1071 lbn -= blksperindir + 1; 1072 len -= blksperindir; 1073 rlbn += blksperindir; 1074 } 1075 return (0); 1076} 1077 1078/* 1079 * Descend an indirect block chain for vnode cancelvp accounting for all 1080 * its indirect blocks in snapvp. 1081 */ 1082static int 1083indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 1084 blksperindir, fs, acctfunc, expungetype) 1085 struct vnode *snapvp; 1086 struct vnode *cancelvp; 1087 int level; 1088 ufs1_daddr_t blkno; 1089 ufs_lbn_t lbn; 1090 ufs_lbn_t rlbn; 1091 ufs_lbn_t remblks; 1092 ufs_lbn_t blksperindir; 1093 struct fs *fs; 1094 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, 1095 struct fs *, ufs_lbn_t, int); 1096 int expungetype; 1097{ 1098 int error, num, i; 1099 ufs_lbn_t subblksperindir; 1100 struct indir indirs[NIADDR + 2]; 1101 ufs1_daddr_t last, *bap; 1102 struct buf *bp; 1103 1104 if (blkno == 0) { 1105 if (expungetype == BLK_NOCOPY) 1106 return (0); 1107 panic("indiracct_ufs1: missing indir"); 1108 } 1109 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 1110 return (error); 1111 if (lbn != indirs[num - 1 - level].in_lbn || num < 2) 1112 panic("indiracct_ufs1: botched params"); 1113 /* 1114 * We have to expand bread here since it will deadlock looking 1115 * up the block number for any blocks that are not in the cache. 1116 */ 1117 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); 1118 bp->b_blkno = fsbtodb(fs, blkno); 1119 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && 1120 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { 1121 brelse(bp); 1122 return (error); 1123 } 1124 /* 1125 * Account for the block pointers in this indirect block. 1126 */ 1127 last = howmany(remblks, blksperindir); 1128 if (last > NINDIR(fs)) 1129 last = NINDIR(fs); 1130 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); 1131 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 1132 bqrelse(bp); 1133 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 1134 level == 0 ? rlbn : -1, expungetype); 1135 if (error || level == 0) 1136 goto out; 1137 /* 1138 * Account for the block pointers in each of the indirect blocks 1139 * in the levels below us. 1140 */ 1141 subblksperindir = blksperindir / NINDIR(fs); 1142 for (lbn++, level--, i = 0; i < last; i++) { 1143 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn, 1144 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 1145 if (error) 1146 goto out; 1147 rlbn += blksperindir; 1148 lbn -= blksperindir; 1149 remblks -= blksperindir; 1150 } 1151out: 1152 free(bap, M_DEVBUF); 1153 return (error); 1154} 1155 1156/* 1157 * Do both snap accounting and map accounting. 1158 */ 1159static int 1160fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype) 1161 struct vnode *vp; 1162 ufs1_daddr_t *oldblkp, *lastblkp; 1163 struct fs *fs; 1164 ufs_lbn_t lblkno; 1165 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 1166{ 1167 int error; 1168 1169 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 1170 return (error); 1171 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 1172} 1173 1174/* 1175 * Identify a set of blocks allocated in a snapshot inode. 1176 */ 1177static int 1178snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1179 struct vnode *vp; 1180 ufs1_daddr_t *oldblkp, *lastblkp; 1181 struct fs *fs; 1182 ufs_lbn_t lblkno; 1183 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 1184{ 1185 struct inode *ip = VTOI(vp); 1186 ufs1_daddr_t blkno, *blkp; 1187 ufs_lbn_t lbn; 1188 struct buf *ibp; 1189 int error; 1190 1191 for ( ; oldblkp < lastblkp; oldblkp++) { 1192 blkno = *oldblkp; 1193 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 1194 continue; 1195 lbn = fragstoblks(fs, blkno); 1196 if (lbn < NDADDR) { 1197 blkp = &ip->i_din1->di_db[lbn]; 1198 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1199 } else { 1200 error = ffs_balloc_ufs1(vp, lblktosize(fs, (off_t)lbn), 1201 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1202 if (error) 1203 return (error); 1204 blkp = &((ufs1_daddr_t *)(ibp->b_data)) 1205 [(lbn - NDADDR) % NINDIR(fs)]; 1206 } 1207 /* 1208 * If we are expunging a snapshot vnode and we 1209 * find a block marked BLK_NOCOPY, then it is 1210 * one that has been allocated to this snapshot after 1211 * we took our current snapshot and can be ignored. 1212 */ 1213 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 1214 if (lbn >= NDADDR) 1215 brelse(ibp); 1216 } else { 1217 if (*blkp != 0) 1218 panic("snapacct_ufs1: bad block"); 1219 *blkp = expungetype; 1220 if (lbn >= NDADDR) 1221 bdwrite(ibp); 1222 } 1223 } 1224 return (0); 1225} 1226 1227/* 1228 * Account for a set of blocks allocated in a snapshot inode. 1229 */ 1230static int 1231mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1232 struct vnode *vp; 1233 ufs1_daddr_t *oldblkp, *lastblkp; 1234 struct fs *fs; 1235 ufs_lbn_t lblkno; 1236 int expungetype; 1237{ 1238 ufs1_daddr_t blkno; 1239 struct inode *ip; 1240 ino_t inum; 1241 int acctit; 1242 1243 ip = VTOI(vp); 1244 inum = ip->i_number; 1245 if (lblkno == -1) 1246 acctit = 0; 1247 else 1248 acctit = 1; 1249 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 1250 blkno = *oldblkp; 1251 if (blkno == 0 || blkno == BLK_NOCOPY) 1252 continue; 1253 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP) 1254 *ip->i_snapblklist++ = lblkno; 1255 if (blkno == BLK_SNAP) 1256 blkno = blkstofrags(fs, lblkno); 1257 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum, 1258 vp->v_type, NULL); 1259 } 1260 return (0); 1261} 1262 1263/* 1264 * Before expunging a snapshot inode, note all the 1265 * blocks that it claims with BLK_SNAP so that fsck will 1266 * be able to account for those blocks properly and so 1267 * that this snapshot knows that it need not copy them 1268 * if the other snapshot holding them is freed. This code 1269 * is reproduced once each for UFS1 and UFS2. 1270 */ 1271static int 1272expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype, clearmode) 1273 struct vnode *snapvp; 1274 struct inode *cancelip; 1275 struct fs *fs; 1276 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1277 struct fs *, ufs_lbn_t, int); 1278 int expungetype; 1279 int clearmode; 1280{ 1281 int i, error, indiroff; 1282 ufs_lbn_t lbn, rlbn; 1283 ufs2_daddr_t len, blkno, numblks, blksperindir; 1284 struct ufs2_dinode *dip; 1285 struct thread *td = curthread; 1286 struct buf *bp; 1287 1288 /* 1289 * Prepare to expunge the inode. If its inode block has not 1290 * yet been copied, then allocate and fill the copy. 1291 */ 1292 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); 1293 blkno = 0; 1294 if (lbn < NDADDR) { 1295 blkno = VTOI(snapvp)->i_din2->di_db[lbn]; 1296 } else { 1297 if (DOINGSOFTDEP(snapvp)) 1298 softdep_prealloc(snapvp, MNT_WAIT); 1299 td->td_pflags |= TDP_COWINPROGRESS; 1300 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), 1301 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); 1302 td->td_pflags &= ~TDP_COWINPROGRESS; 1303 if (error) 1304 return (error); 1305 indiroff = (lbn - NDADDR) % NINDIR(fs); 1306 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; 1307 bqrelse(bp); 1308 } 1309 if (blkno != 0) { 1310 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) 1311 return (error); 1312 } else { 1313 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), 1314 fs->fs_bsize, KERNCRED, 0, &bp); 1315 if (error) 1316 return (error); 1317 if ((error = readblock(snapvp, bp, lbn)) != 0) 1318 return (error); 1319 } 1320 /* 1321 * Set a snapshot inode to be a zero length file, regular files 1322 * to be completely unallocated. 1323 */ 1324 dip = (struct ufs2_dinode *)bp->b_data + 1325 ino_to_fsbo(fs, cancelip->i_number); 1326 if (clearmode || cancelip->i_effnlink == 0) 1327 dip->di_mode = 0; 1328 dip->di_size = 0; 1329 dip->di_blocks = 0; 1330 dip->di_flags &= ~SF_SNAPSHOT; 1331 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t)); 1332 bdwrite(bp); 1333 /* 1334 * Now go through and expunge all the blocks in the file 1335 * using the function requested. 1336 */ 1337 numblks = howmany(cancelip->i_size, fs->fs_bsize); 1338 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0], 1339 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype))) 1340 return (error); 1341 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0], 1342 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype))) 1343 return (error); 1344 blksperindir = 1; 1345 lbn = -NDADDR; 1346 len = numblks - NDADDR; 1347 rlbn = NDADDR; 1348 for (i = 0; len > 0 && i < NIADDR; i++) { 1349 error = indiracct_ufs2(snapvp, ITOV(cancelip), i, 1350 cancelip->i_din2->di_ib[i], lbn, rlbn, len, 1351 blksperindir, fs, acctfunc, expungetype); 1352 if (error) 1353 return (error); 1354 blksperindir *= NINDIR(fs); 1355 lbn -= blksperindir + 1; 1356 len -= blksperindir; 1357 rlbn += blksperindir; 1358 } 1359 return (0); 1360} 1361 1362/* 1363 * Descend an indirect block chain for vnode cancelvp accounting for all 1364 * its indirect blocks in snapvp. 1365 */ 1366static int 1367indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, 1368 blksperindir, fs, acctfunc, expungetype) 1369 struct vnode *snapvp; 1370 struct vnode *cancelvp; 1371 int level; 1372 ufs2_daddr_t blkno; 1373 ufs_lbn_t lbn; 1374 ufs_lbn_t rlbn; 1375 ufs_lbn_t remblks; 1376 ufs_lbn_t blksperindir; 1377 struct fs *fs; 1378 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, 1379 struct fs *, ufs_lbn_t, int); 1380 int expungetype; 1381{ 1382 int error, num, i; 1383 ufs_lbn_t subblksperindir; 1384 struct indir indirs[NIADDR + 2]; 1385 ufs2_daddr_t last, *bap; 1386 struct buf *bp; 1387 1388 if (blkno == 0) { 1389 if (expungetype == BLK_NOCOPY) 1390 return (0); 1391 panic("indiracct_ufs2: missing indir"); 1392 } 1393 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) 1394 return (error); 1395 if (lbn != indirs[num - 1 - level].in_lbn || num < 2) 1396 panic("indiracct_ufs2: botched params"); 1397 /* 1398 * We have to expand bread here since it will deadlock looking 1399 * up the block number for any blocks that are not in the cache. 1400 */ 1401 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); 1402 bp->b_blkno = fsbtodb(fs, blkno); 1403 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && 1404 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { 1405 brelse(bp); 1406 return (error); 1407 } 1408 /* 1409 * Account for the block pointers in this indirect block. 1410 */ 1411 last = howmany(remblks, blksperindir); 1412 if (last > NINDIR(fs)) 1413 last = NINDIR(fs); 1414 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); 1415 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); 1416 bqrelse(bp); 1417 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, 1418 level == 0 ? rlbn : -1, expungetype); 1419 if (error || level == 0) 1420 goto out; 1421 /* 1422 * Account for the block pointers in each of the indirect blocks 1423 * in the levels below us. 1424 */ 1425 subblksperindir = blksperindir / NINDIR(fs); 1426 for (lbn++, level--, i = 0; i < last; i++) { 1427 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn, 1428 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); 1429 if (error) 1430 goto out; 1431 rlbn += blksperindir; 1432 lbn -= blksperindir; 1433 remblks -= blksperindir; 1434 } 1435out: 1436 free(bap, M_DEVBUF); 1437 return (error); 1438} 1439 1440/* 1441 * Do both snap accounting and map accounting. 1442 */ 1443static int 1444fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype) 1445 struct vnode *vp; 1446 ufs2_daddr_t *oldblkp, *lastblkp; 1447 struct fs *fs; 1448 ufs_lbn_t lblkno; 1449 int exptype; /* BLK_SNAP or BLK_NOCOPY */ 1450{ 1451 int error; 1452 1453 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype))) 1454 return (error); 1455 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)); 1456} 1457 1458/* 1459 * Identify a set of blocks allocated in a snapshot inode. 1460 */ 1461static int 1462snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1463 struct vnode *vp; 1464 ufs2_daddr_t *oldblkp, *lastblkp; 1465 struct fs *fs; 1466 ufs_lbn_t lblkno; 1467 int expungetype; /* BLK_SNAP or BLK_NOCOPY */ 1468{ 1469 struct inode *ip = VTOI(vp); 1470 ufs2_daddr_t blkno, *blkp; 1471 ufs_lbn_t lbn; 1472 struct buf *ibp; 1473 int error; 1474 1475 for ( ; oldblkp < lastblkp; oldblkp++) { 1476 blkno = *oldblkp; 1477 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) 1478 continue; 1479 lbn = fragstoblks(fs, blkno); 1480 if (lbn < NDADDR) { 1481 blkp = &ip->i_din2->di_db[lbn]; 1482 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1483 } else { 1484 error = ffs_balloc_ufs2(vp, lblktosize(fs, (off_t)lbn), 1485 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1486 if (error) 1487 return (error); 1488 blkp = &((ufs2_daddr_t *)(ibp->b_data)) 1489 [(lbn - NDADDR) % NINDIR(fs)]; 1490 } 1491 /* 1492 * If we are expunging a snapshot vnode and we 1493 * find a block marked BLK_NOCOPY, then it is 1494 * one that has been allocated to this snapshot after 1495 * we took our current snapshot and can be ignored. 1496 */ 1497 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { 1498 if (lbn >= NDADDR) 1499 brelse(ibp); 1500 } else { 1501 if (*blkp != 0) 1502 panic("snapacct_ufs2: bad block"); 1503 *blkp = expungetype; 1504 if (lbn >= NDADDR) 1505 bdwrite(ibp); 1506 } 1507 } 1508 return (0); 1509} 1510 1511/* 1512 * Account for a set of blocks allocated in a snapshot inode. 1513 */ 1514static int 1515mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) 1516 struct vnode *vp; 1517 ufs2_daddr_t *oldblkp, *lastblkp; 1518 struct fs *fs; 1519 ufs_lbn_t lblkno; 1520 int expungetype; 1521{ 1522 ufs2_daddr_t blkno; 1523 struct inode *ip; 1524 ino_t inum; 1525 int acctit; 1526 1527 ip = VTOI(vp); 1528 inum = ip->i_number; 1529 if (lblkno == -1) 1530 acctit = 0; 1531 else 1532 acctit = 1; 1533 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { 1534 blkno = *oldblkp; 1535 if (blkno == 0 || blkno == BLK_NOCOPY) 1536 continue; 1537 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP) 1538 *ip->i_snapblklist++ = lblkno; 1539 if (blkno == BLK_SNAP) 1540 blkno = blkstofrags(fs, lblkno); 1541 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum, 1542 vp->v_type, NULL); 1543 } 1544 return (0); 1545} 1546 1547/* 1548 * Decrement extra reference on snapshot when last name is removed. 1549 * It will not be freed until the last open reference goes away. 1550 */ 1551void 1552ffs_snapgone(ip) 1553 struct inode *ip; 1554{ 1555 struct inode *xp; 1556 struct fs *fs; 1557 int snaploc; 1558 struct snapdata *sn; 1559 struct ufsmount *ump; 1560 1561 /* 1562 * Find snapshot in incore list. 1563 */ 1564 xp = NULL; 1565 sn = ip->i_devvp->v_rdev->si_snapdata; 1566 if (sn != NULL) 1567 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) 1568 if (xp == ip) 1569 break; 1570 if (xp != NULL) 1571 vrele(ITOV(ip)); 1572 else if (snapdebug) 1573 printf("ffs_snapgone: lost snapshot vnode %d\n", 1574 ip->i_number); 1575 /* 1576 * Delete snapshot inode from superblock. Keep list dense. 1577 */ 1578 fs = ip->i_fs; 1579 ump = ip->i_ump; 1580 UFS_LOCK(ump); 1581 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) 1582 if (fs->fs_snapinum[snaploc] == ip->i_number) 1583 break; 1584 if (snaploc < FSMAXSNAP) { 1585 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) { 1586 if (fs->fs_snapinum[snaploc] == 0) 1587 break; 1588 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc]; 1589 } 1590 fs->fs_snapinum[snaploc - 1] = 0; 1591 } 1592 UFS_UNLOCK(ump); 1593} 1594 1595/* 1596 * Prepare a snapshot file for being removed. 1597 */ 1598void 1599ffs_snapremove(vp) 1600 struct vnode *vp; 1601{ 1602 struct inode *ip; 1603 struct vnode *devvp; 1604 struct buf *ibp; 1605 struct fs *fs; 1606 ufs2_daddr_t numblks, blkno, dblk; 1607 int error, loc, last; 1608 struct snapdata *sn; 1609 1610 ip = VTOI(vp); 1611 fs = ip->i_fs; 1612 devvp = ip->i_devvp; 1613 /* 1614 * If active, delete from incore list (this snapshot may 1615 * already have been in the process of being deleted, so 1616 * would not have been active). 1617 * 1618 * Clear copy-on-write flag if last snapshot. 1619 */ 1620 VI_LOCK(devvp); 1621 if (ip->i_nextsnap.tqe_prev != 0) { 1622 sn = devvp->v_rdev->si_snapdata; 1623 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap); 1624 ip->i_nextsnap.tqe_prev = 0; 1625 VI_UNLOCK(devvp); 1626 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); 1627 KASSERT(vp->v_vnlock == &sn->sn_lock, 1628 ("ffs_snapremove: lost lock mutation")); 1629 vp->v_vnlock = &vp->v_lock; 1630 VI_LOCK(devvp); 1631 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 1632 try_free_snapdata(devvp); 1633 } else 1634 VI_UNLOCK(devvp); 1635 /* 1636 * Clear all BLK_NOCOPY fields. Pass any block claims to other 1637 * snapshots that want them (see ffs_snapblkfree below). 1638 */ 1639 for (blkno = 1; blkno < NDADDR; blkno++) { 1640 dblk = DIP(ip, i_db[blkno]); 1641 if (dblk == 0) 1642 continue; 1643 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1644 DIP_SET(ip, i_db[blkno], 0); 1645 else if ((dblk == blkstofrags(fs, blkno) && 1646 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize, 1647 ip->i_number, vp->v_type, NULL))) { 1648 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - 1649 btodb(fs->fs_bsize)); 1650 DIP_SET(ip, i_db[blkno], 0); 1651 } 1652 } 1653 numblks = howmany(ip->i_size, fs->fs_bsize); 1654 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) { 1655 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), 1656 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1657 if (error) 1658 continue; 1659 if (fs->fs_size - blkno > NINDIR(fs)) 1660 last = NINDIR(fs); 1661 else 1662 last = fs->fs_size - blkno; 1663 for (loc = 0; loc < last; loc++) { 1664 if (ip->i_ump->um_fstype == UFS1) { 1665 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc]; 1666 if (dblk == 0) 1667 continue; 1668 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1669 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1670 else if ((dblk == blkstofrags(fs, blkno) && 1671 ffs_snapblkfree(fs, ip->i_devvp, dblk, 1672 fs->fs_bsize, ip->i_number, vp->v_type, 1673 NULL))) { 1674 ip->i_din1->di_blocks -= 1675 btodb(fs->fs_bsize); 1676 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; 1677 } 1678 continue; 1679 } 1680 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc]; 1681 if (dblk == 0) 1682 continue; 1683 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) 1684 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1685 else if ((dblk == blkstofrags(fs, blkno) && 1686 ffs_snapblkfree(fs, ip->i_devvp, dblk, 1687 fs->fs_bsize, ip->i_number, vp->v_type, NULL))) { 1688 ip->i_din2->di_blocks -= btodb(fs->fs_bsize); 1689 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; 1690 } 1691 } 1692 bawrite(ibp); 1693 } 1694 /* 1695 * Clear snapshot flag and drop reference. 1696 */ 1697 ip->i_flags &= ~SF_SNAPSHOT; 1698 DIP_SET(ip, i_flags, ip->i_flags); 1699 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1700 /* 1701 * The dirtied indirects must be written out before 1702 * softdep_setup_freeblocks() is called. Otherwise indir_trunc() 1703 * may find indirect pointers using the magic BLK_* values. 1704 */ 1705 if (DOINGSOFTDEP(vp)) 1706 ffs_syncvnode(vp, MNT_WAIT, 0); 1707#ifdef QUOTA 1708 /* 1709 * Reenable disk quotas for ex-snapshot file. 1710 */ 1711 if (!getinoquota(ip)) 1712 (void) chkdq(ip, DIP(ip, i_blocks), KERNCRED, FORCE); 1713#endif 1714} 1715 1716/* 1717 * Notification that a block is being freed. Return zero if the free 1718 * should be allowed to proceed. Return non-zero if the snapshot file 1719 * wants to claim the block. The block will be claimed if it is an 1720 * uncopied part of one of the snapshots. It will be freed if it is 1721 * either a BLK_NOCOPY or has already been copied in all of the snapshots. 1722 * If a fragment is being freed, then all snapshots that care about 1723 * it must make a copy since a snapshot file can only claim full sized 1724 * blocks. Note that if more than one snapshot file maps the block, 1725 * we can pick one at random to claim it. Since none of the snapshots 1726 * can change, we are assurred that they will all see the same unmodified 1727 * image. When deleting a snapshot file (see ffs_snapremove above), we 1728 * must push any of these claimed blocks to one of the other snapshots 1729 * that maps it. These claimed blocks are easily identified as they will 1730 * have a block number equal to their logical block number within the 1731 * snapshot. A copied block can never have this property because they 1732 * must always have been allocated from a BLK_NOCOPY location. 1733 */ 1734int 1735ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd) 1736 struct fs *fs; 1737 struct vnode *devvp; 1738 ufs2_daddr_t bno; 1739 long size; 1740 ino_t inum; 1741 enum vtype vtype; 1742 struct workhead *wkhd; 1743{ 1744 struct buf *ibp, *cbp, *savedcbp = 0; 1745 struct thread *td = curthread; 1746 struct inode *ip; 1747 struct vnode *vp = NULL; 1748 ufs_lbn_t lbn; 1749 ufs2_daddr_t blkno; 1750 int indiroff = 0, error = 0, claimedblk = 0; 1751 struct snapdata *sn; 1752 1753 lbn = fragstoblks(fs, bno); 1754retry: 1755 VI_LOCK(devvp); 1756 sn = devvp->v_rdev->si_snapdata; 1757 if (sn == NULL) { 1758 VI_UNLOCK(devvp); 1759 return (0); 1760 } 1761 if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 1762 VI_MTX(devvp)) != 0) 1763 goto retry; 1764 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 1765 vp = ITOV(ip); 1766 if (DOINGSOFTDEP(vp)) 1767 softdep_prealloc(vp, MNT_WAIT); 1768 /* 1769 * Lookup block being written. 1770 */ 1771 if (lbn < NDADDR) { 1772 blkno = DIP(ip, i_db[lbn]); 1773 } else { 1774 td->td_pflags |= TDP_COWINPROGRESS; 1775 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1776 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 1777 td->td_pflags &= ~TDP_COWINPROGRESS; 1778 if (error) 1779 break; 1780 indiroff = (lbn - NDADDR) % NINDIR(fs); 1781 if (ip->i_ump->um_fstype == UFS1) 1782 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 1783 else 1784 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 1785 } 1786 /* 1787 * Check to see if block needs to be copied. 1788 */ 1789 if (blkno == 0) { 1790 /* 1791 * A block that we map is being freed. If it has not 1792 * been claimed yet, we will claim or copy it (below). 1793 */ 1794 claimedblk = 1; 1795 } else if (blkno == BLK_SNAP) { 1796 /* 1797 * No previous snapshot claimed the block, 1798 * so it will be freed and become a BLK_NOCOPY 1799 * (don't care) for us. 1800 */ 1801 if (claimedblk) 1802 panic("snapblkfree: inconsistent block type"); 1803 if (lbn < NDADDR) { 1804 DIP_SET(ip, i_db[lbn], BLK_NOCOPY); 1805 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1806 } else if (ip->i_ump->um_fstype == UFS1) { 1807 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 1808 BLK_NOCOPY; 1809 bdwrite(ibp); 1810 } else { 1811 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 1812 BLK_NOCOPY; 1813 bdwrite(ibp); 1814 } 1815 continue; 1816 } else /* BLK_NOCOPY or default */ { 1817 /* 1818 * If the snapshot has already copied the block 1819 * (default), or does not care about the block, 1820 * it is not needed. 1821 */ 1822 if (lbn >= NDADDR) 1823 bqrelse(ibp); 1824 continue; 1825 } 1826 /* 1827 * If this is a full size block, we will just grab it 1828 * and assign it to the snapshot inode. Otherwise we 1829 * will proceed to copy it. See explanation for this 1830 * routine as to why only a single snapshot needs to 1831 * claim this block. 1832 */ 1833 if (size == fs->fs_bsize) { 1834#ifdef DEBUG 1835 if (snapdebug) 1836 printf("%s %d lbn %jd from inum %d\n", 1837 "Grabonremove: snapino", ip->i_number, 1838 (intmax_t)lbn, inum); 1839#endif 1840 /* 1841 * If journaling is tracking this write we must add 1842 * the work to the inode or indirect being written. 1843 */ 1844 if (wkhd != NULL) { 1845 if (lbn < NDADDR) 1846 softdep_inode_append(ip, 1847 curthread->td_ucred, wkhd); 1848 else 1849 softdep_buf_append(ibp, wkhd); 1850 } 1851 if (lbn < NDADDR) { 1852 DIP_SET(ip, i_db[lbn], bno); 1853 } else if (ip->i_ump->um_fstype == UFS1) { 1854 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno; 1855 bdwrite(ibp); 1856 } else { 1857 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno; 1858 bdwrite(ibp); 1859 } 1860 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size)); 1861 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1862 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 1863 return (1); 1864 } 1865 if (lbn >= NDADDR) 1866 bqrelse(ibp); 1867 /* 1868 * Allocate the block into which to do the copy. Note that this 1869 * allocation will never require any additional allocations for 1870 * the snapshot inode. 1871 */ 1872 td->td_pflags |= TDP_COWINPROGRESS; 1873 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 1874 fs->fs_bsize, KERNCRED, 0, &cbp); 1875 td->td_pflags &= ~TDP_COWINPROGRESS; 1876 if (error) 1877 break; 1878#ifdef DEBUG 1879 if (snapdebug) 1880 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n", 1881 "Copyonremove: snapino ", ip->i_number, 1882 (intmax_t)lbn, "for inum", inum, size, 1883 (intmax_t)cbp->b_blkno); 1884#endif 1885 /* 1886 * If we have already read the old block contents, then 1887 * simply copy them to the new block. Note that we need 1888 * to synchronously write snapshots that have not been 1889 * unlinked, and hence will be visible after a crash, 1890 * to ensure their integrity. At a minimum we ensure the 1891 * integrity of the filesystem metadata, but use the 1892 * dopersistence sysctl-setable flag to decide on the 1893 * persistence needed for file content data. 1894 */ 1895 if (savedcbp != 0) { 1896 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 1897 bawrite(cbp); 1898 if ((vtype == VDIR || dopersistence) && 1899 ip->i_effnlink > 0) 1900 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 1901 continue; 1902 } 1903 /* 1904 * Otherwise, read the old block contents into the buffer. 1905 */ 1906 if ((error = readblock(vp, cbp, lbn)) != 0) { 1907 bzero(cbp->b_data, fs->fs_bsize); 1908 bawrite(cbp); 1909 if ((vtype == VDIR || dopersistence) && 1910 ip->i_effnlink > 0) 1911 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 1912 break; 1913 } 1914 savedcbp = cbp; 1915 } 1916 /* 1917 * Note that we need to synchronously write snapshots that 1918 * have not been unlinked, and hence will be visible after 1919 * a crash, to ensure their integrity. At a minimum we 1920 * ensure the integrity of the filesystem metadata, but 1921 * use the dopersistence sysctl-setable flag to decide on 1922 * the persistence needed for file content data. 1923 */ 1924 if (savedcbp) { 1925 vp = savedcbp->b_vp; 1926 bawrite(savedcbp); 1927 if ((vtype == VDIR || dopersistence) && 1928 VTOI(vp)->i_effnlink > 0) 1929 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 1930 } 1931 /* 1932 * If we have been unable to allocate a block in which to do 1933 * the copy, then return non-zero so that the fragment will 1934 * not be freed. Although space will be lost, the snapshot 1935 * will stay consistent. 1936 */ 1937 if (error != 0 && wkhd != NULL) 1938 softdep_freework(wkhd); 1939 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 1940 return (error); 1941} 1942 1943/* 1944 * Associate snapshot files when mounting. 1945 */ 1946void 1947ffs_snapshot_mount(mp) 1948 struct mount *mp; 1949{ 1950 struct ufsmount *ump = VFSTOUFS(mp); 1951 struct vnode *devvp = ump->um_devvp; 1952 struct fs *fs = ump->um_fs; 1953 struct thread *td = curthread; 1954 struct snapdata *sn; 1955 struct vnode *vp; 1956 struct vnode *lastvp; 1957 struct inode *ip; 1958 struct uio auio; 1959 struct iovec aiov; 1960 void *snapblklist; 1961 char *reason; 1962 daddr_t snaplistsize; 1963 int error, snaploc, loc; 1964 1965 /* 1966 * XXX The following needs to be set before ffs_truncate or 1967 * VOP_READ can be called. 1968 */ 1969 mp->mnt_stat.f_iosize = fs->fs_bsize; 1970 /* 1971 * Process each snapshot listed in the superblock. 1972 */ 1973 vp = NULL; 1974 lastvp = NULL; 1975 sn = NULL; 1976 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) { 1977 if (fs->fs_snapinum[snaploc] == 0) 1978 break; 1979 if ((error = ffs_vget(mp, fs->fs_snapinum[snaploc], 1980 LK_EXCLUSIVE, &vp)) != 0){ 1981 printf("ffs_snapshot_mount: vget failed %d\n", error); 1982 continue; 1983 } 1984 ip = VTOI(vp); 1985 if (!IS_SNAPSHOT(ip) || ip->i_size == 1986 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) { 1987 if (!IS_SNAPSHOT(ip)) { 1988 reason = "non-snapshot"; 1989 } else { 1990 reason = "old format snapshot"; 1991 (void)ffs_truncate(vp, (off_t)0, 0, NOCRED, td); 1992 (void)ffs_syncvnode(vp, MNT_WAIT, 0); 1993 } 1994 printf("ffs_snapshot_mount: %s inode %d\n", 1995 reason, fs->fs_snapinum[snaploc]); 1996 vput(vp); 1997 vp = NULL; 1998 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) { 1999 if (fs->fs_snapinum[loc] == 0) 2000 break; 2001 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc]; 2002 } 2003 fs->fs_snapinum[loc - 1] = 0; 2004 snaploc--; 2005 continue; 2006 } 2007 /* 2008 * Acquire a lock on the snapdata structure, creating it if 2009 * necessary. 2010 */ 2011 sn = ffs_snapdata_acquire(devvp); 2012 /* 2013 * Change vnode to use shared snapshot lock instead of the 2014 * original private lock. 2015 */ 2016 vp->v_vnlock = &sn->sn_lock; 2017 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 2018 /* 2019 * Link it onto the active snapshot list. 2020 */ 2021 VI_LOCK(devvp); 2022 if (ip->i_nextsnap.tqe_prev != 0) 2023 panic("ffs_snapshot_mount: %d already on list", 2024 ip->i_number); 2025 else 2026 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); 2027 vp->v_vflag |= VV_SYSTEM; 2028 VI_UNLOCK(devvp); 2029 VOP_UNLOCK(vp, 0); 2030 lastvp = vp; 2031 } 2032 vp = lastvp; 2033 /* 2034 * No usable snapshots found. 2035 */ 2036 if (sn == NULL || vp == NULL) 2037 return; 2038 /* 2039 * Allocate the space for the block hints list. We always want to 2040 * use the list from the newest snapshot. 2041 */ 2042 auio.uio_iov = &aiov; 2043 auio.uio_iovcnt = 1; 2044 aiov.iov_base = (void *)&snaplistsize; 2045 aiov.iov_len = sizeof(snaplistsize); 2046 auio.uio_resid = aiov.iov_len; 2047 auio.uio_offset = 2048 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)); 2049 auio.uio_segflg = UIO_SYSSPACE; 2050 auio.uio_rw = UIO_READ; 2051 auio.uio_td = td; 2052 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2053 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 2054 printf("ffs_snapshot_mount: read_1 failed %d\n", error); 2055 VOP_UNLOCK(vp, 0); 2056 return; 2057 } 2058 snapblklist = malloc(snaplistsize * sizeof(daddr_t), 2059 M_UFSMNT, M_WAITOK); 2060 auio.uio_iovcnt = 1; 2061 aiov.iov_base = snapblklist; 2062 aiov.iov_len = snaplistsize * sizeof (daddr_t); 2063 auio.uio_resid = aiov.iov_len; 2064 auio.uio_offset -= sizeof(snaplistsize); 2065 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { 2066 printf("ffs_snapshot_mount: read_2 failed %d\n", error); 2067 VOP_UNLOCK(vp, 0); 2068 free(snapblklist, M_UFSMNT); 2069 return; 2070 } 2071 VOP_UNLOCK(vp, 0); 2072 VI_LOCK(devvp); 2073 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount"); 2074 sn->sn_listsize = snaplistsize; 2075 sn->sn_blklist = (daddr_t *)snapblklist; 2076 devvp->v_vflag |= VV_COPYONWRITE; 2077 VI_UNLOCK(devvp); 2078} 2079 2080/* 2081 * Disassociate snapshot files when unmounting. 2082 */ 2083void 2084ffs_snapshot_unmount(mp) 2085 struct mount *mp; 2086{ 2087 struct vnode *devvp = VFSTOUFS(mp)->um_devvp; 2088 struct snapdata *sn; 2089 struct inode *xp; 2090 struct vnode *vp; 2091 2092 VI_LOCK(devvp); 2093 sn = devvp->v_rdev->si_snapdata; 2094 while (sn != NULL && (xp = TAILQ_FIRST(&sn->sn_head)) != NULL) { 2095 vp = ITOV(xp); 2096 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap); 2097 xp->i_nextsnap.tqe_prev = 0; 2098 lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE, 2099 VI_MTX(devvp)); 2100 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); 2101 KASSERT(vp->v_vnlock == &sn->sn_lock, 2102 ("ffs_snapshot_unmount: lost lock mutation")); 2103 vp->v_vnlock = &vp->v_lock; 2104 lockmgr(&vp->v_lock, LK_RELEASE, NULL); 2105 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2106 if (xp->i_effnlink > 0) 2107 vrele(vp); 2108 VI_LOCK(devvp); 2109 sn = devvp->v_rdev->si_snapdata; 2110 } 2111 try_free_snapdata(devvp); 2112 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount"); 2113} 2114 2115/* 2116 * Check the buffer block to be belong to device buffer that shall be 2117 * locked after snaplk. devvp shall be locked on entry, and will be 2118 * leaved locked upon exit. 2119 */ 2120static int 2121ffs_bp_snapblk(devvp, bp) 2122 struct vnode *devvp; 2123 struct buf *bp; 2124{ 2125 struct snapdata *sn; 2126 struct fs *fs; 2127 ufs2_daddr_t lbn, *snapblklist; 2128 int lower, upper, mid; 2129 2130 ASSERT_VI_LOCKED(devvp, "ffs_bp_snapblk"); 2131 KASSERT(devvp->v_type == VCHR, ("Not a device %p", devvp)); 2132 sn = devvp->v_rdev->si_snapdata; 2133 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) == NULL) 2134 return (0); 2135 fs = TAILQ_FIRST(&sn->sn_head)->i_fs; 2136 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); 2137 snapblklist = sn->sn_blklist; 2138 upper = sn->sn_listsize - 1; 2139 lower = 1; 2140 while (lower <= upper) { 2141 mid = (lower + upper) / 2; 2142 if (snapblklist[mid] == lbn) 2143 break; 2144 if (snapblklist[mid] < lbn) 2145 lower = mid + 1; 2146 else 2147 upper = mid - 1; 2148 } 2149 if (lower <= upper) 2150 return (1); 2151 return (0); 2152} 2153 2154void 2155ffs_bdflush(bo, bp) 2156 struct bufobj *bo; 2157 struct buf *bp; 2158{ 2159 struct thread *td; 2160 struct vnode *vp, *devvp; 2161 struct buf *nbp; 2162 int bp_bdskip; 2163 2164 if (bo->bo_dirty.bv_cnt <= dirtybufthresh) 2165 return; 2166 2167 td = curthread; 2168 vp = bp->b_vp; 2169 devvp = bo->__bo_vnode; 2170 KASSERT(vp == devvp, ("devvp != vp %p %p", bo, bp)); 2171 2172 VI_LOCK(devvp); 2173 bp_bdskip = ffs_bp_snapblk(devvp, bp); 2174 if (bp_bdskip) 2175 bdwriteskip++; 2176 VI_UNLOCK(devvp); 2177 if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10 && !bp_bdskip) { 2178 (void) VOP_FSYNC(vp, MNT_NOWAIT, td); 2179 altbufferflushes++; 2180 } else { 2181 BO_LOCK(bo); 2182 /* 2183 * Try to find a buffer to flush. 2184 */ 2185 TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { 2186 if ((nbp->b_vflags & BV_BKGRDINPROG) || 2187 BUF_LOCK(nbp, 2188 LK_EXCLUSIVE | LK_NOWAIT, NULL)) 2189 continue; 2190 if (bp == nbp) 2191 panic("bdwrite: found ourselves"); 2192 BO_UNLOCK(bo); 2193 /* 2194 * Don't countdeps with the bo lock 2195 * held. 2196 */ 2197 if (buf_countdeps(nbp, 0)) { 2198 BO_LOCK(bo); 2199 BUF_UNLOCK(nbp); 2200 continue; 2201 } 2202 if (bp_bdskip) { 2203 VI_LOCK(devvp); 2204 if (!ffs_bp_snapblk(vp, nbp)) { 2205 if (BO_MTX(bo) != VI_MTX(vp)) { 2206 VI_UNLOCK(devvp); 2207 BO_LOCK(bo); 2208 } 2209 BUF_UNLOCK(nbp); 2210 continue; 2211 } 2212 VI_UNLOCK(devvp); 2213 } 2214 if (nbp->b_flags & B_CLUSTEROK) { 2215 vfs_bio_awrite(nbp); 2216 } else { 2217 bremfree(nbp); 2218 bawrite(nbp); 2219 } 2220 dirtybufferflushes++; 2221 break; 2222 } 2223 if (nbp == NULL) 2224 BO_UNLOCK(bo); 2225 } 2226} 2227 2228/* 2229 * Check for need to copy block that is about to be written, 2230 * copying the block if necessary. 2231 */ 2232int 2233ffs_copyonwrite(devvp, bp) 2234 struct vnode *devvp; 2235 struct buf *bp; 2236{ 2237 struct snapdata *sn; 2238 struct buf *ibp, *cbp, *savedcbp = 0; 2239 struct thread *td = curthread; 2240 struct fs *fs; 2241 struct inode *ip; 2242 struct vnode *vp = 0; 2243 ufs2_daddr_t lbn, blkno, *snapblklist; 2244 int lower, upper, mid, indiroff, error = 0; 2245 int launched_async_io, prev_norunningbuf; 2246 long saved_runningbufspace; 2247 2248 if (devvp != bp->b_vp && IS_SNAPSHOT(VTOI(bp->b_vp))) 2249 return (0); /* Update on a snapshot file */ 2250 if (td->td_pflags & TDP_COWINPROGRESS) 2251 panic("ffs_copyonwrite: recursive call"); 2252 /* 2253 * First check to see if it is in the preallocated list. 2254 * By doing this check we avoid several potential deadlocks. 2255 */ 2256 VI_LOCK(devvp); 2257 sn = devvp->v_rdev->si_snapdata; 2258 if (sn == NULL || 2259 TAILQ_EMPTY(&sn->sn_head)) { 2260 VI_UNLOCK(devvp); 2261 return (0); /* No snapshot */ 2262 } 2263 ip = TAILQ_FIRST(&sn->sn_head); 2264 fs = ip->i_fs; 2265 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); 2266 snapblklist = sn->sn_blklist; 2267 upper = sn->sn_listsize - 1; 2268 lower = 1; 2269 while (lower <= upper) { 2270 mid = (lower + upper) / 2; 2271 if (snapblklist[mid] == lbn) 2272 break; 2273 if (snapblklist[mid] < lbn) 2274 lower = mid + 1; 2275 else 2276 upper = mid - 1; 2277 } 2278 if (lower <= upper) { 2279 VI_UNLOCK(devvp); 2280 return (0); 2281 } 2282 launched_async_io = 0; 2283 prev_norunningbuf = td->td_pflags & TDP_NORUNNINGBUF; 2284 /* 2285 * Since I/O on bp isn't yet in progress and it may be blocked 2286 * for a long time waiting on snaplk, back it out of 2287 * runningbufspace, possibly waking other threads waiting for space. 2288 */ 2289 saved_runningbufspace = bp->b_runningbufspace; 2290 if (saved_runningbufspace != 0) 2291 runningbufwakeup(bp); 2292 /* 2293 * Not in the precomputed list, so check the snapshots. 2294 */ 2295 while (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 2296 VI_MTX(devvp)) != 0) { 2297 VI_LOCK(devvp); 2298 sn = devvp->v_rdev->si_snapdata; 2299 if (sn == NULL || 2300 TAILQ_EMPTY(&sn->sn_head)) { 2301 VI_UNLOCK(devvp); 2302 if (saved_runningbufspace != 0) { 2303 bp->b_runningbufspace = saved_runningbufspace; 2304 atomic_add_long(&runningbufspace, 2305 bp->b_runningbufspace); 2306 } 2307 return (0); /* Snapshot gone */ 2308 } 2309 } 2310 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 2311 vp = ITOV(ip); 2312 if (DOINGSOFTDEP(vp)) 2313 softdep_prealloc(vp, MNT_WAIT); 2314 /* 2315 * We ensure that everything of our own that needs to be 2316 * copied will be done at the time that ffs_snapshot is 2317 * called. Thus we can skip the check here which can 2318 * deadlock in doing the lookup in UFS_BALLOC. 2319 */ 2320 if (bp->b_vp == vp) 2321 continue; 2322 /* 2323 * Check to see if block needs to be copied. We do not have 2324 * to hold the snapshot lock while doing this lookup as it 2325 * will never require any additional allocations for the 2326 * snapshot inode. 2327 */ 2328 if (lbn < NDADDR) { 2329 blkno = DIP(ip, i_db[lbn]); 2330 } else { 2331 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; 2332 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 2333 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); 2334 td->td_pflags &= ~TDP_COWINPROGRESS; 2335 if (error) 2336 break; 2337 indiroff = (lbn - NDADDR) % NINDIR(fs); 2338 if (ip->i_ump->um_fstype == UFS1) 2339 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; 2340 else 2341 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; 2342 bqrelse(ibp); 2343 } 2344#ifdef INVARIANTS 2345 if (blkno == BLK_SNAP && bp->b_lblkno >= 0) 2346 panic("ffs_copyonwrite: bad copy block"); 2347#endif 2348 if (blkno != 0) 2349 continue; 2350 /* 2351 * Allocate the block into which to do the copy. Since 2352 * multiple processes may all try to copy the same block, 2353 * we have to recheck our need to do a copy if we sleep 2354 * waiting for the lock. 2355 * 2356 * Because all snapshots on a filesystem share a single 2357 * lock, we ensure that we will never be in competition 2358 * with another process to allocate a block. 2359 */ 2360 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; 2361 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), 2362 fs->fs_bsize, KERNCRED, 0, &cbp); 2363 td->td_pflags &= ~TDP_COWINPROGRESS; 2364 if (error) 2365 break; 2366#ifdef DEBUG 2367 if (snapdebug) { 2368 printf("Copyonwrite: snapino %d lbn %jd for ", 2369 ip->i_number, (intmax_t)lbn); 2370 if (bp->b_vp == devvp) 2371 printf("fs metadata"); 2372 else 2373 printf("inum %d", VTOI(bp->b_vp)->i_number); 2374 printf(" lblkno %jd to blkno %jd\n", 2375 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno); 2376 } 2377#endif 2378 /* 2379 * If we have already read the old block contents, then 2380 * simply copy them to the new block. Note that we need 2381 * to synchronously write snapshots that have not been 2382 * unlinked, and hence will be visible after a crash, 2383 * to ensure their integrity. At a minimum we ensure the 2384 * integrity of the filesystem metadata, but use the 2385 * dopersistence sysctl-setable flag to decide on the 2386 * persistence needed for file content data. 2387 */ 2388 if (savedcbp != 0) { 2389 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); 2390 bawrite(cbp); 2391 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || 2392 dopersistence) && ip->i_effnlink > 0) 2393 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 2394 else 2395 launched_async_io = 1; 2396 continue; 2397 } 2398 /* 2399 * Otherwise, read the old block contents into the buffer. 2400 */ 2401 if ((error = readblock(vp, cbp, lbn)) != 0) { 2402 bzero(cbp->b_data, fs->fs_bsize); 2403 bawrite(cbp); 2404 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || 2405 dopersistence) && ip->i_effnlink > 0) 2406 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 2407 else 2408 launched_async_io = 1; 2409 break; 2410 } 2411 savedcbp = cbp; 2412 } 2413 /* 2414 * Note that we need to synchronously write snapshots that 2415 * have not been unlinked, and hence will be visible after 2416 * a crash, to ensure their integrity. At a minimum we 2417 * ensure the integrity of the filesystem metadata, but 2418 * use the dopersistence sysctl-setable flag to decide on 2419 * the persistence needed for file content data. 2420 */ 2421 if (savedcbp) { 2422 vp = savedcbp->b_vp; 2423 bawrite(savedcbp); 2424 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || 2425 dopersistence) && VTOI(vp)->i_effnlink > 0) 2426 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); 2427 else 2428 launched_async_io = 1; 2429 } 2430 lockmgr(vp->v_vnlock, LK_RELEASE, NULL); 2431 td->td_pflags = (td->td_pflags & ~TDP_NORUNNINGBUF) | 2432 prev_norunningbuf; 2433 if (launched_async_io && (td->td_pflags & TDP_NORUNNINGBUF) == 0) 2434 waitrunningbufspace(); 2435 /* 2436 * I/O on bp will now be started, so count it in runningbufspace. 2437 */ 2438 if (saved_runningbufspace != 0) { 2439 bp->b_runningbufspace = saved_runningbufspace; 2440 atomic_add_long(&runningbufspace, bp->b_runningbufspace); 2441 } 2442 return (error); 2443} 2444 2445/* 2446 * sync snapshots to force freework records waiting on snapshots to claim 2447 * blocks to free. 2448 */ 2449void 2450ffs_sync_snap(mp, waitfor) 2451 struct mount *mp; 2452 int waitfor; 2453{ 2454 struct snapdata *sn; 2455 struct vnode *devvp; 2456 struct vnode *vp; 2457 struct inode *ip; 2458 2459 devvp = VFSTOUFS(mp)->um_devvp; 2460 if ((devvp->v_vflag & VV_COPYONWRITE) == 0) 2461 return; 2462 for (;;) { 2463 VI_LOCK(devvp); 2464 sn = devvp->v_rdev->si_snapdata; 2465 if (sn == NULL) { 2466 VI_UNLOCK(devvp); 2467 return; 2468 } 2469 if (lockmgr(&sn->sn_lock, 2470 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, 2471 VI_MTX(devvp)) == 0) 2472 break; 2473 } 2474 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { 2475 vp = ITOV(ip); 2476 ffs_syncvnode(vp, waitfor, NO_INO_UPDT); 2477 } 2478 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2479} 2480 2481/* 2482 * Read the specified block into the given buffer. 2483 * Much of this boiler-plate comes from bwrite(). 2484 */ 2485static int 2486readblock(vp, bp, lbn) 2487 struct vnode *vp; 2488 struct buf *bp; 2489 ufs2_daddr_t lbn; 2490{ 2491 struct inode *ip = VTOI(vp); 2492 struct bio *bip; 2493 2494 bip = g_alloc_bio(); 2495 bip->bio_cmd = BIO_READ; 2496 bip->bio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn))); 2497 bip->bio_data = bp->b_data; 2498 bip->bio_length = bp->b_bcount; 2499 bip->bio_done = NULL; 2500 2501 g_io_request(bip, ip->i_devvp->v_bufobj.bo_private); 2502 bp->b_error = biowait(bip, "snaprdb"); 2503 g_destroy_bio(bip); 2504 return (bp->b_error); 2505} 2506 2507#endif 2508 2509/* 2510 * Process file deletes that were deferred by ufs_inactive() due to 2511 * the file system being suspended. Transfer IN_LAZYACCESS into 2512 * IN_MODIFIED for vnodes that were accessed during suspension. 2513 */ 2514void 2515process_deferred_inactive(struct mount *mp) 2516{ 2517 struct vnode *vp, *mvp; 2518 struct inode *ip; 2519 struct thread *td; 2520 int error; 2521 2522 td = curthread; 2523 (void) vn_start_secondary_write(NULL, &mp, V_WAIT); 2524 loop: 2525 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 2526 /* 2527 * IN_LAZYACCESS is checked here without holding any 2528 * vnode lock, but this flag is set only while holding 2529 * vnode interlock. 2530 */ 2531 if (vp->v_type == VNON || 2532 ((VTOI(vp)->i_flag & IN_LAZYACCESS) == 0 && 2533 ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0))) { 2534 VI_UNLOCK(vp); 2535 continue; 2536 } 2537 vholdl(vp); 2538 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2539 if (error != 0) { 2540 vdrop(vp); 2541 if (error == ENOENT) 2542 continue; /* vnode recycled */ 2543 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 2544 goto loop; 2545 } 2546 ip = VTOI(vp); 2547 if ((ip->i_flag & IN_LAZYACCESS) != 0) { 2548 ip->i_flag &= ~IN_LAZYACCESS; 2549 ip->i_flag |= IN_MODIFIED; 2550 } 2551 VI_LOCK(vp); 2552 if ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0) { 2553 VI_UNLOCK(vp); 2554 VOP_UNLOCK(vp, 0); 2555 vdrop(vp); 2556 continue; 2557 } 2558 vinactive(vp, td); 2559 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, 2560 ("process_deferred_inactive: got VI_OWEINACT")); 2561 VI_UNLOCK(vp); 2562 VOP_UNLOCK(vp, 0); 2563 vdrop(vp); 2564 } 2565 vn_finished_secondary_write(mp); 2566} 2567 2568#ifndef NO_FFS_SNAPSHOT 2569 2570static struct snapdata * 2571ffs_snapdata_alloc(void) 2572{ 2573 struct snapdata *sn; 2574 2575 /* 2576 * Fetch a snapdata from the free list if there is one available. 2577 */ 2578 mtx_lock(&snapfree_lock); 2579 sn = LIST_FIRST(&snapfree); 2580 if (sn != NULL) 2581 LIST_REMOVE(sn, sn_link); 2582 mtx_unlock(&snapfree_lock); 2583 if (sn != NULL) 2584 return (sn); 2585 /* 2586 * If there were no free snapdatas allocate one. 2587 */ 2588 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO); 2589 TAILQ_INIT(&sn->sn_head); 2590 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT, 2591 LK_CANRECURSE | LK_NOSHARE); 2592 return (sn); 2593} 2594 2595/* 2596 * The snapdata is never freed because we can not be certain that 2597 * there are no threads sleeping on the snap lock. Persisting 2598 * them permanently avoids costly synchronization in ffs_lock(). 2599 */ 2600static void 2601ffs_snapdata_free(struct snapdata *sn) 2602{ 2603 mtx_lock(&snapfree_lock); 2604 LIST_INSERT_HEAD(&snapfree, sn, sn_link); 2605 mtx_unlock(&snapfree_lock); 2606} 2607 2608/* Try to free snapdata associated with devvp */ 2609static void 2610try_free_snapdata(struct vnode *devvp) 2611{ 2612 struct snapdata *sn; 2613 ufs2_daddr_t *snapblklist; 2614 2615 ASSERT_VI_LOCKED(devvp, "try_free_snapdata"); 2616 sn = devvp->v_rdev->si_snapdata; 2617 2618 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) != NULL || 2619 (devvp->v_vflag & VV_COPYONWRITE) == 0) { 2620 VI_UNLOCK(devvp); 2621 return; 2622 } 2623 2624 devvp->v_rdev->si_snapdata = NULL; 2625 devvp->v_vflag &= ~VV_COPYONWRITE; 2626 lockmgr(&sn->sn_lock, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp)); 2627 snapblklist = sn->sn_blklist; 2628 sn->sn_blklist = NULL; 2629 sn->sn_listsize = 0; 2630 lockmgr(&sn->sn_lock, LK_RELEASE, NULL); 2631 if (snapblklist != NULL) 2632 free(snapblklist, M_UFSMNT); 2633 ffs_snapdata_free(sn); 2634} 2635 2636static struct snapdata * 2637ffs_snapdata_acquire(struct vnode *devvp) 2638{ 2639 struct snapdata *nsn; 2640 struct snapdata *sn; 2641 2642 /* 2643 * Allocate a free snapdata. This is done before acquiring the 2644 * devvp lock to avoid allocation while the devvp interlock is 2645 * held. 2646 */ 2647 nsn = ffs_snapdata_alloc(); 2648 /* 2649 * If there snapshots already exist on this filesystem grab a 2650 * reference to the shared lock. Otherwise this is the first 2651 * snapshot on this filesystem and we need to use our 2652 * pre-allocated snapdata. 2653 */ 2654 VI_LOCK(devvp); 2655 if (devvp->v_rdev->si_snapdata == NULL) { 2656 devvp->v_rdev->si_snapdata = nsn; 2657 nsn = NULL; 2658 } 2659 sn = devvp->v_rdev->si_snapdata; 2660 /* 2661 * Acquire the snapshot lock. 2662 */ 2663 lockmgr(&sn->sn_lock, 2664 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, VI_MTX(devvp)); 2665 /* 2666 * Free any unused snapdata. 2667 */ 2668 if (nsn != NULL) 2669 ffs_snapdata_free(nsn); 2670 2671 return (sn); 2672} 2673 2674#endif 2675