1 /* $NetBSD: ffs_wapbl.c,v 1.25 2013/10/25 11:35:55 martin Exp $ */ 2 3 /*- 4 * Copyright (c) 2003,2006,2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Wasabi Systems, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: ffs_wapbl.c,v 1.25 2013/10/25 11:35:55 martin Exp $"); 34 35 #define WAPBL_INTERNAL 36 37 #if defined(_KERNEL_OPT) 38 #include "opt_ffs.h" 39 #endif 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/kernel.h> 44 #include <sys/vnode.h> 45 #include <sys/mount.h> 46 #include <sys/file.h> 47 #include <sys/disk.h> 48 #include <sys/ioctl.h> 49 #include <sys/errno.h> 50 #include <sys/kauth.h> 51 #include <sys/wapbl.h> 52 53 #include <ufs/ufs/inode.h> 54 #include <ufs/ufs/quota.h> 55 #include <ufs/ufs/ufsmount.h> 56 #include <ufs/ufs/ufs_bswap.h> 57 #include <ufs/ufs/ufs_extern.h> 58 #include <ufs/ufs/ufs_wapbl.h> 59 60 #include <ufs/ffs/fs.h> 61 #include <ufs/ffs/ffs_extern.h> 62 63 #undef WAPBL_DEBUG 64 #ifdef WAPBL_DEBUG 65 int ffs_wapbl_debug = 1; 66 #define DPRINTF(fmt, args...) \ 67 do { \ 68 if (ffs_wapbl_debug) \ 69 printf("%s:%d "fmt, __func__ , __LINE__, ##args); \ 70 } while (/* CONSTCOND */0) 71 #else 72 #define DPRINTF(fmt, args...) \ 73 do { \ 74 /* nothing */ \ 75 } while (/* CONSTCOND */0) 76 #endif 77 78 static int ffs_superblock_layout(struct fs *); 79 static int wapbl_log_position(struct mount *, struct fs *, struct vnode *, 80 daddr_t *, size_t *, size_t *, uint64_t *); 81 static int wapbl_create_infs_log(struct mount *, struct fs *, struct vnode *, 82 daddr_t *, size_t *, uint64_t *); 83 static void wapbl_find_log_start(struct mount *, struct vnode *, off_t, 84 daddr_t *, daddr_t *, size_t *); 85 static int wapbl_remove_log(struct mount *); 86 static int wapbl_allocate_log_file(struct mount *, struct vnode *, 87 daddr_t *, size_t *, uint64_t *); 88 89 /* 90 * Return the super block layout format - UFS1 or UFS2. 91 * WAPBL only works with UFS2 layout (which is still available 92 * with FFSv1). 93 * 94 * XXX Should this be in ufs/ffs/fs.h? Same style of check is 95 * also used in ffs_alloc.c in a few places. 96 */ 97 static int 98 ffs_superblock_layout(struct fs *fs) 99 { 100 if ((fs->fs_magic == FS_UFS1_MAGIC) && 101 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) 102 return 1; 103 else 104 return 2; 105 } 106 107 /* 108 * This function is invoked after a log is replayed to 109 * disk to perform logical cleanup actions as described by 110 * the log 111 */ 112 void 113 ffs_wapbl_replay_finish(struct mount *mp) 114 { 115 struct wapbl_replay *wr = mp->mnt_wapbl_replay; 116 int i; 117 int error; 118 119 if (!wr) 120 return; 121 122 KDASSERT((mp->mnt_flag & MNT_RDONLY) == 0); 123 124 for (i = 0; i < wr->wr_inodescnt; i++) { 125 struct vnode *vp; 126 struct inode *ip; 127 error = VFS_VGET(mp, wr->wr_inodes[i].wr_inumber, &vp); 128 if (error) { 129 printf("ffs_wapbl_replay_finish: " 130 "unable to cleanup inode %" PRIu32 "\n", 131 wr->wr_inodes[i].wr_inumber); 132 continue; 133 } 134 ip = VTOI(vp); 135 KDASSERT(wr->wr_inodes[i].wr_inumber == ip->i_number); 136 #ifdef WAPBL_DEBUG 137 printf("ffs_wapbl_replay_finish: " 138 "cleaning inode %" PRIu64 " size=%" PRIu64 " mode=%o nlink=%d\n", 139 ip->i_number, ip->i_size, ip->i_mode, ip->i_nlink); 140 #endif 141 KASSERT(ip->i_nlink == 0); 142 143 /* 144 * The journal may have left partially allocated inodes in mode 145 * zero. This may occur if a crash occurs betweeen the node 146 * allocation in ffs_nodeallocg and when the node is properly 147 * initialized in ufs_makeinode. If so, just dallocate them. 148 */ 149 if (ip->i_mode == 0) { 150 UFS_WAPBL_BEGIN(mp); 151 ffs_vfree(vp, ip->i_number, wr->wr_inodes[i].wr_imode); 152 UFS_WAPBL_END(mp); 153 } 154 vput(vp); 155 } 156 wapbl_replay_stop(wr); 157 wapbl_replay_free(wr); 158 mp->mnt_wapbl_replay = NULL; 159 } 160 161 /* Callback for wapbl */ 162 void 163 ffs_wapbl_sync_metadata(struct mount *mp, daddr_t *deallocblks, 164 int *dealloclens, int dealloccnt) 165 { 166 struct ufsmount *ump = VFSTOUFS(mp); 167 struct fs *fs = ump->um_fs; 168 int i, error __diagused; 169 170 #ifdef WAPBL_DEBUG_INODES 171 ufs_wapbl_verify_inodes(mp, "ffs_wapbl_sync_metadata"); 172 #endif 173 174 for (i = 0; i< dealloccnt; i++) { 175 /* 176 * blkfree errors are unreported, might silently fail 177 * if it cannot read the cylinder group block 178 */ 179 ffs_blkfree(fs, ump->um_devvp, 180 FFS_DBTOFSB(fs, deallocblks[i]), dealloclens[i], -1); 181 } 182 183 fs->fs_fmod = 0; 184 fs->fs_time = time_second; 185 error = ffs_cgupdate(ump, 0); 186 KASSERT(error == 0); 187 } 188 189 void 190 ffs_wapbl_abort_sync_metadata(struct mount *mp, daddr_t *deallocblks, 191 int *dealloclens, int dealloccnt) 192 { 193 struct ufsmount *ump = VFSTOUFS(mp); 194 struct fs *fs = ump->um_fs; 195 int i; 196 197 for (i = 0; i < dealloccnt; i++) { 198 /* 199 * Since the above blkfree may have failed, this blkalloc might 200 * fail as well, so don't check its error. Note that if the 201 * blkfree succeeded above, then this shouldn't fail because 202 * the buffer will be locked in the current transaction. 203 */ 204 ffs_blkalloc_ump(ump, FFS_DBTOFSB(fs, deallocblks[i]), 205 dealloclens[i]); 206 } 207 } 208 209 static int 210 wapbl_remove_log(struct mount *mp) 211 { 212 struct ufsmount *ump = VFSTOUFS(mp); 213 struct fs *fs = ump->um_fs; 214 struct vnode *vp; 215 struct inode *ip; 216 ino_t log_ino; 217 int error; 218 219 /* If super block layout is too old to support WAPBL, return */ 220 if (ffs_superblock_layout(fs) < 2) 221 return 0; 222 223 /* If all the log locators are 0, just clean up */ 224 if (fs->fs_journallocs[0] == 0 && 225 fs->fs_journallocs[1] == 0 && 226 fs->fs_journallocs[2] == 0 && 227 fs->fs_journallocs[3] == 0) { 228 DPRINTF("empty locators, just clear\n"); 229 goto done; 230 } 231 232 switch (fs->fs_journal_location) { 233 case UFS_WAPBL_JOURNALLOC_NONE: 234 /* nothing! */ 235 DPRINTF("no log\n"); 236 break; 237 238 case UFS_WAPBL_JOURNALLOC_IN_FILESYSTEM: 239 log_ino = fs->fs_journallocs[UFS_WAPBL_INFS_INO]; 240 DPRINTF("in-fs log, ino = %" PRId64 "\n",log_ino); 241 242 /* if no existing log inode, just clear all fields and bail */ 243 if (log_ino == 0) 244 goto done; 245 error = VFS_VGET(mp, log_ino, &vp); 246 if (error != 0) { 247 printf("ffs_wapbl: vget failed %d\n", 248 error); 249 /* clear out log info on error */ 250 goto done; 251 } 252 ip = VTOI(vp); 253 KASSERT(log_ino == ip->i_number); 254 if ((ip->i_flags & SF_LOG) == 0) { 255 printf("ffs_wapbl: try to clear non-log inode " 256 "%" PRId64 "\n", log_ino); 257 vput(vp); 258 /* clear out log info on error */ 259 goto done; 260 } 261 262 /* 263 * remove the log inode by setting its link count back 264 * to zero and bail. 265 */ 266 ip->i_nlink = 0; 267 DIP_ASSIGN(ip, nlink, 0); 268 vput(vp); 269 270 case UFS_WAPBL_JOURNALLOC_END_PARTITION: 271 DPRINTF("end-of-partition log\n"); 272 /* no extra work required */ 273 break; 274 275 default: 276 printf("ffs_wapbl: unknown journal type %d\n", 277 fs->fs_journal_location); 278 break; 279 } 280 281 282 done: 283 /* Clear out all previous knowledge of journal */ 284 fs->fs_journal_version = 0; 285 fs->fs_journal_location = 0; 286 fs->fs_journal_flags = 0; 287 fs->fs_journallocs[0] = 0; 288 fs->fs_journallocs[1] = 0; 289 fs->fs_journallocs[2] = 0; 290 fs->fs_journallocs[3] = 0; 291 (void) ffs_sbupdate(ump, MNT_WAIT); 292 293 return 0; 294 } 295 296 int 297 ffs_wapbl_start(struct mount *mp) 298 { 299 struct ufsmount *ump = VFSTOUFS(mp); 300 struct fs *fs = ump->um_fs; 301 struct vnode *devvp = ump->um_devvp; 302 daddr_t off; 303 size_t count; 304 size_t blksize; 305 uint64_t extradata; 306 int error; 307 308 if (mp->mnt_wapbl == NULL) { 309 if (fs->fs_journal_flags & UFS_WAPBL_FLAGS_CLEAR_LOG) { 310 /* Clear out any existing journal file */ 311 error = wapbl_remove_log(mp); 312 if (error != 0) 313 return error; 314 } 315 316 if (mp->mnt_flag & MNT_LOG) { 317 KDASSERT(fs->fs_ronly == 0); 318 319 /* WAPBL needs UFS2 format super block */ 320 if (ffs_superblock_layout(fs) < 2) { 321 printf("%s fs superblock in old format, " 322 "not journaling\n", 323 VFSTOUFS(mp)->um_fs->fs_fsmnt); 324 mp->mnt_flag &= ~MNT_LOG; 325 return EINVAL; 326 } 327 328 error = wapbl_log_position(mp, fs, devvp, &off, 329 &count, &blksize, &extradata); 330 if (error) 331 return error; 332 333 error = wapbl_start(&mp->mnt_wapbl, mp, devvp, off, 334 count, blksize, mp->mnt_wapbl_replay, 335 ffs_wapbl_sync_metadata, 336 ffs_wapbl_abort_sync_metadata); 337 if (error) 338 return error; 339 340 mp->mnt_wapbl_op = &wapbl_ops; 341 342 #ifdef WAPBL_DEBUG 343 printf("%s: enabling logging\n", fs->fs_fsmnt); 344 #endif 345 346 if ((fs->fs_flags & FS_DOWAPBL) == 0) { 347 UFS_WAPBL_BEGIN(mp); 348 fs->fs_flags |= FS_DOWAPBL; 349 error = ffs_sbupdate(ump, MNT_WAIT); 350 if (error) { 351 UFS_WAPBL_END(mp); 352 ffs_wapbl_stop(mp, MNT_FORCE); 353 return error; 354 } 355 UFS_WAPBL_END(mp); 356 error = wapbl_flush(mp->mnt_wapbl, 1); 357 if (error) { 358 ffs_wapbl_stop(mp, MNT_FORCE); 359 return error; 360 } 361 } 362 } else if (fs->fs_flags & FS_DOWAPBL) { 363 fs->fs_fmod = 1; 364 fs->fs_flags &= ~FS_DOWAPBL; 365 } 366 } 367 368 /* 369 * It is recommended that you finish replay with logging enabled. 370 * However, even if logging is not enabled, the remaining log 371 * replay should be safely recoverable with an fsck, so perform 372 * it anyway. 373 */ 374 if ((fs->fs_ronly == 0) && mp->mnt_wapbl_replay) { 375 int saveflag = mp->mnt_flag & MNT_RDONLY; 376 /* 377 * Make sure MNT_RDONLY is not set so that the inode 378 * cleanup in ufs_inactive will actually do its work. 379 */ 380 mp->mnt_flag &= ~MNT_RDONLY; 381 ffs_wapbl_replay_finish(mp); 382 mp->mnt_flag |= saveflag; 383 KASSERT(fs->fs_ronly == 0); 384 } 385 386 return 0; 387 } 388 389 int 390 ffs_wapbl_stop(struct mount *mp, int force) 391 { 392 struct ufsmount *ump = VFSTOUFS(mp); 393 struct fs *fs = ump->um_fs; 394 int error; 395 396 if (mp->mnt_wapbl) { 397 KDASSERT(fs->fs_ronly == 0); 398 399 /* 400 * Make sure turning off FS_DOWAPBL is only removed 401 * as the only change in the final flush since otherwise 402 * a transaction may reorder writes. 403 */ 404 error = wapbl_flush(mp->mnt_wapbl, 1); 405 if (error && !force) 406 return error; 407 if (error && force) 408 goto forceout; 409 error = UFS_WAPBL_BEGIN(mp); 410 if (error && !force) 411 return error; 412 if (error && force) 413 goto forceout; 414 KASSERT(fs->fs_flags & FS_DOWAPBL); 415 416 fs->fs_flags &= ~FS_DOWAPBL; 417 error = ffs_sbupdate(ump, MNT_WAIT); 418 KASSERT(error == 0); /* XXX a bit drastic! */ 419 UFS_WAPBL_END(mp); 420 forceout: 421 error = wapbl_stop(mp->mnt_wapbl, force); 422 if (error) { 423 KASSERT(!force); 424 fs->fs_flags |= FS_DOWAPBL; 425 return error; 426 } 427 fs->fs_flags &= ~FS_DOWAPBL; /* Repeat in case of forced error */ 428 mp->mnt_wapbl = NULL; 429 430 #ifdef WAPBL_DEBUG 431 printf("%s: disabled logging\n", fs->fs_fsmnt); 432 #endif 433 } 434 435 return 0; 436 } 437 438 int 439 ffs_wapbl_replay_start(struct mount *mp, struct fs *fs, struct vnode *devvp) 440 { 441 int error; 442 daddr_t off; 443 size_t count; 444 size_t blksize; 445 uint64_t extradata; 446 447 /* 448 * WAPBL needs UFS2 format super block, if we got here with a 449 * UFS1 format super block something is amiss... 450 */ 451 if (ffs_superblock_layout(fs) < 2) 452 return EINVAL; 453 454 error = wapbl_log_position(mp, fs, devvp, &off, &count, &blksize, 455 &extradata); 456 457 if (error) 458 return error; 459 460 error = wapbl_replay_start(&mp->mnt_wapbl_replay, devvp, off, 461 count, blksize); 462 if (error) 463 return error; 464 465 mp->mnt_wapbl_op = &wapbl_ops; 466 467 return 0; 468 } 469 470 /* 471 * If the superblock doesn't already have a recorded journal location 472 * then we allocate the journal in one of two positions: 473 * 474 * - At the end of the partition after the filesystem if there's 475 * enough space. "Enough space" is defined as >= 1MB of journal 476 * per 1GB of filesystem or 64MB, whichever is smaller. 477 * 478 * - Inside the filesystem. We try to allocate a contiguous journal 479 * based on the total filesystem size - the target is 1MB of journal 480 * per 1GB of filesystem, up to a maximum journal size of 64MB. As 481 * a worst case allowing for fragmentation, we'll allocate a journal 482 * 1/4 of the desired size but never smaller than 1MB. 483 * 484 * XXX In the future if we allow for non-contiguous journal files we 485 * can tighten the above restrictions. 486 * 487 * XXX 488 * These seems like a lot of duplication both here and in some of 489 * the userland tools (fsck_ffs, dumpfs, tunefs) with similar 490 * "switch (fs_journal_location)" constructs. Can we centralise 491 * this sort of code somehow/somewhere? 492 */ 493 static int 494 wapbl_log_position(struct mount *mp, struct fs *fs, struct vnode *devvp, 495 daddr_t *startp, size_t *countp, size_t *blksizep, uint64_t *extradatap) 496 { 497 struct ufsmount *ump = VFSTOUFS(mp); 498 daddr_t logstart, logend, desired_logsize; 499 uint64_t numsecs; 500 unsigned secsize; 501 int error, location; 502 503 if (fs->fs_journal_version == UFS_WAPBL_VERSION) { 504 switch (fs->fs_journal_location) { 505 case UFS_WAPBL_JOURNALLOC_END_PARTITION: 506 DPRINTF("found existing end-of-partition log\n"); 507 *startp = fs->fs_journallocs[UFS_WAPBL_EPART_ADDR]; 508 *countp = fs->fs_journallocs[UFS_WAPBL_EPART_COUNT]; 509 *blksizep = fs->fs_journallocs[UFS_WAPBL_EPART_BLKSZ]; 510 DPRINTF(" start = %" PRId64 ", size = %zu, " 511 "blksize = %zu\n", *startp, *countp, *blksizep); 512 return 0; 513 514 case UFS_WAPBL_JOURNALLOC_IN_FILESYSTEM: 515 DPRINTF("found existing in-filesystem log\n"); 516 *startp = fs->fs_journallocs[UFS_WAPBL_INFS_ADDR]; 517 *countp = fs->fs_journallocs[UFS_WAPBL_INFS_COUNT]; 518 *blksizep = fs->fs_journallocs[UFS_WAPBL_INFS_BLKSZ]; 519 DPRINTF(" start = %" PRId64 ", size = %zu, " 520 "blksize = %zu\n", *startp, *countp, *blksizep); 521 return 0; 522 523 default: 524 printf("ffs_wapbl: unknown journal type %d\n", 525 fs->fs_journal_location); 526 return EINVAL; 527 } 528 } 529 530 desired_logsize = 531 ffs_lfragtosize(fs, fs->fs_size) / UFS_WAPBL_JOURNAL_SCALE; 532 DPRINTF("desired log size = %" PRId64 " kB\n", desired_logsize / 1024); 533 desired_logsize = max(desired_logsize, UFS_WAPBL_MIN_JOURNAL_SIZE); 534 desired_logsize = min(desired_logsize, UFS_WAPBL_MAX_JOURNAL_SIZE); 535 DPRINTF("adjusted desired log size = %" PRId64 " kB\n", 536 desired_logsize / 1024); 537 538 /* Is there space after after filesystem on partition for log? */ 539 logstart = FFS_FSBTODB(fs, fs->fs_size); 540 error = getdisksize(devvp, &numsecs, &secsize); 541 if (error) 542 return error; 543 KDASSERT(secsize != 0); 544 logend = btodb(numsecs * secsize); 545 546 if (dbtob(logend - logstart) >= desired_logsize) { 547 DPRINTF("enough space, use end-of-partition log\n"); 548 549 location = UFS_WAPBL_JOURNALLOC_END_PARTITION; 550 *blksizep = secsize; 551 552 *startp = logstart; 553 *countp = (logend - logstart); 554 *extradatap = 0; 555 556 /* convert to physical block numbers */ 557 *startp = dbtob(*startp) / secsize; 558 *countp = dbtob(*countp) / secsize; 559 560 fs->fs_journallocs[UFS_WAPBL_EPART_ADDR] = *startp; 561 fs->fs_journallocs[UFS_WAPBL_EPART_COUNT] = *countp; 562 fs->fs_journallocs[UFS_WAPBL_EPART_BLKSZ] = *blksizep; 563 fs->fs_journallocs[UFS_WAPBL_EPART_UNUSED] = *extradatap; 564 } else { 565 DPRINTF("end-of-partition has only %" PRId64 " free\n", 566 logend - logstart); 567 568 location = UFS_WAPBL_JOURNALLOC_IN_FILESYSTEM; 569 *blksizep = secsize; 570 571 error = wapbl_create_infs_log(mp, fs, devvp, 572 startp, countp, extradatap); 573 ffs_sync(mp, MNT_WAIT, FSCRED); 574 575 /* convert to physical block numbers */ 576 *startp = dbtob(*startp) / secsize; 577 *countp = dbtob(*countp) / secsize; 578 579 fs->fs_journallocs[UFS_WAPBL_INFS_ADDR] = *startp; 580 fs->fs_journallocs[UFS_WAPBL_INFS_COUNT] = *countp; 581 fs->fs_journallocs[UFS_WAPBL_INFS_BLKSZ] = *blksizep; 582 fs->fs_journallocs[UFS_WAPBL_INFS_INO] = *extradatap; 583 } 584 585 if (error == 0) { 586 /* update superblock with log location */ 587 fs->fs_journal_version = UFS_WAPBL_VERSION; 588 fs->fs_journal_location = location; 589 fs->fs_journal_flags = 0; 590 591 error = ffs_sbupdate(ump, MNT_WAIT); 592 } 593 594 return error; 595 } 596 597 /* 598 * Try to create a journal log inside the filesystem. 599 */ 600 static int 601 wapbl_create_infs_log(struct mount *mp, struct fs *fs, struct vnode *devvp, 602 daddr_t *startp, size_t *countp, uint64_t *extradatap) 603 { 604 struct vnode *vp, *rvp; 605 struct inode *ip; 606 int error; 607 608 if ((error = VFS_ROOT(mp, &rvp)) != 0) 609 return error; 610 611 error = UFS_VALLOC(rvp, 0 | S_IFREG, NOCRED, &vp); 612 if (mp->mnt_flag & MNT_UPDATE) { 613 vput(rvp); 614 } else { 615 VOP_UNLOCK(rvp); 616 vgone(rvp); 617 } 618 if (error != 0) 619 return error; 620 621 vp->v_type = VREG; 622 ip = VTOI(vp); 623 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 624 ip->i_mode = 0 | IFREG; 625 DIP_ASSIGN(ip, mode, ip->i_mode); 626 ip->i_flags = SF_LOG; 627 DIP_ASSIGN(ip, flags, ip->i_flags); 628 ip->i_nlink = 1; 629 DIP_ASSIGN(ip, nlink, 1); 630 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 631 632 if ((error = wapbl_allocate_log_file(mp, vp, 633 startp, countp, extradatap)) != 0) { 634 /* 635 * If we couldn't allocate the space for the log file, 636 * remove the inode by setting its link count back to 637 * zero and bail. 638 */ 639 ip->i_nlink = 0; 640 DIP_ASSIGN(ip, nlink, 0); 641 VOP_UNLOCK(vp); 642 vgone(vp); 643 644 return error; 645 } 646 647 /* 648 * Now that we have the place-holder inode for the journal, 649 * we don't need the vnode ever again. 650 */ 651 VOP_UNLOCK(vp); 652 vgone(vp); 653 654 return 0; 655 } 656 657 int 658 wapbl_allocate_log_file(struct mount *mp, struct vnode *vp, 659 daddr_t *startp, size_t *countp, uint64_t *extradatap) 660 { 661 struct ufsmount *ump = VFSTOUFS(mp); 662 struct fs *fs = ump->um_fs; 663 daddr_t addr, indir_addr; 664 off_t logsize; 665 size_t size; 666 int error; 667 668 logsize = 0; 669 /* check if there's a suggested log size */ 670 if (fs->fs_journal_flags & UFS_WAPBL_FLAGS_CREATE_LOG && 671 fs->fs_journal_location == UFS_WAPBL_JOURNALLOC_IN_FILESYSTEM) 672 logsize = fs->fs_journallocs[UFS_WAPBL_INFS_COUNT]; 673 674 if (vp->v_size > 0) { 675 printf("%s: file size (%" PRId64 ") non zero\n", __func__, 676 vp->v_size); 677 return EEXIST; 678 } 679 wapbl_find_log_start(mp, vp, logsize, &addr, &indir_addr, &size); 680 if (addr == 0) { 681 printf("%s: log not allocated, largest extent is " 682 "%" PRId64 "MB\n", __func__, 683 ffs_lblktosize(fs, size) / (1024 * 1024)); 684 return ENOSPC; 685 } 686 687 logsize = ffs_lblktosize(fs, size); /* final log size */ 688 689 VTOI(vp)->i_ffs_first_data_blk = addr; 690 VTOI(vp)->i_ffs_first_indir_blk = indir_addr; 691 692 error = GOP_ALLOC(vp, 0, logsize, B_CONTIG, FSCRED); 693 if (error) { 694 printf("%s: GOP_ALLOC error %d\n", __func__, error); 695 return error; 696 } 697 698 *startp = FFS_FSBTODB(fs, addr); 699 *countp = btodb(logsize); 700 *extradatap = VTOI(vp)->i_number; 701 702 return 0; 703 } 704 705 /* 706 * Find a suitable location for the journal in the filesystem. 707 * 708 * Our strategy here is to look for a contiguous block of free space 709 * at least "logfile" MB in size (plus room for any indirect blocks). 710 * We start at the middle of the filesystem and check each cylinder 711 * group working outwards. If "logfile" MB is not available as a 712 * single contigous chunk, then return the address and size of the 713 * largest chunk found. 714 * 715 * XXX 716 * At what stage does the search fail? Is if the largest space we could 717 * find is less than a quarter the requested space reasonable? If the 718 * search fails entirely, return a block address if "0" it indicate this. 719 */ 720 static void 721 wapbl_find_log_start(struct mount *mp, struct vnode *vp, off_t logsize, 722 daddr_t *addr, daddr_t *indir_addr, size_t *size) 723 { 724 struct ufsmount *ump = VFSTOUFS(mp); 725 struct fs *fs = ump->um_fs; 726 struct vnode *devvp = ump->um_devvp; 727 struct cg *cgp; 728 struct buf *bp; 729 uint8_t *blksfree; 730 daddr_t blkno, best_addr, start_addr; 731 daddr_t desired_blks, min_desired_blks; 732 daddr_t freeblks, best_blks; 733 int bpcg, cg, error, fixedsize, indir_blks, n, s; 734 const int needswap = UFS_FSNEEDSWAP(fs); 735 736 if (logsize == 0) { 737 fixedsize = 0; /* We can adjust the size if tight */ 738 logsize = ffs_lfragtosize(fs, fs->fs_dsize) / 739 UFS_WAPBL_JOURNAL_SCALE; 740 DPRINTF("suggested log size = %" PRId64 "\n", logsize); 741 logsize = max(logsize, UFS_WAPBL_MIN_JOURNAL_SIZE); 742 logsize = min(logsize, UFS_WAPBL_MAX_JOURNAL_SIZE); 743 DPRINTF("adjusted log size = %" PRId64 "\n", logsize); 744 } else { 745 fixedsize = 1; 746 DPRINTF("fixed log size = %" PRId64 "\n", logsize); 747 } 748 749 desired_blks = logsize / fs->fs_bsize; 750 DPRINTF("desired blocks = %" PRId64 "\n", desired_blks); 751 752 /* add in number of indirect blocks needed */ 753 indir_blks = 0; 754 if (desired_blks >= UFS_NDADDR) { 755 struct indir indirs[UFS_NIADDR + 2]; 756 int num; 757 758 error = ufs_getlbns(vp, desired_blks, indirs, &num); 759 if (error) { 760 printf("%s: ufs_getlbns failed, error %d!\n", 761 __func__, error); 762 goto bad; 763 } 764 765 switch (num) { 766 case 2: 767 indir_blks = 1; /* 1st level indirect */ 768 break; 769 case 3: 770 indir_blks = 1 + /* 1st level indirect */ 771 1 + /* 2nd level indirect */ 772 indirs[1].in_off + 1; /* extra 1st level indirect */ 773 break; 774 default: 775 printf("%s: unexpected numlevels %d from ufs_getlbns\n", 776 __func__, num); 777 *size = 0; 778 goto bad; 779 } 780 desired_blks += indir_blks; 781 } 782 DPRINTF("desired blocks = %" PRId64 " (including indirect)\n", 783 desired_blks); 784 785 /* 786 * If a specific size wasn't requested, allow for a smaller log 787 * if we're really tight for space... 788 */ 789 min_desired_blks = desired_blks; 790 if (!fixedsize) 791 min_desired_blks = desired_blks / 4; 792 793 /* Look at number of blocks per CG. If it's too small, bail early. */ 794 bpcg = ffs_fragstoblks(fs, fs->fs_fpg); 795 if (min_desired_blks > bpcg) { 796 printf("ffs_wapbl: cylinder group size of %" PRId64 " MB " 797 " is not big enough for journal\n", 798 ffs_lblktosize(fs, bpcg) / (1024 * 1024)); 799 goto bad; 800 } 801 802 /* 803 * Start with the middle cylinder group, and search outwards in 804 * both directions until we either find the requested log size 805 * or reach the start/end of the file system. If we reach the 806 * start/end without finding enough space for the full requested 807 * log size, use the largest extent found if it is large enough 808 * to satisfy the our minimum size. 809 * 810 * XXX 811 * Can we just use the cluster contigsum stuff (esp on UFS2) 812 * here to simplify this search code? 813 */ 814 best_addr = 0; 815 best_blks = 0; 816 for (cg = fs->fs_ncg / 2, s = 0, n = 1; 817 best_blks < desired_blks && cg >= 0 && cg < fs->fs_ncg; 818 s++, n = -n, cg += n * s) { 819 DPRINTF("check cg %d of %d\n", cg, fs->fs_ncg); 820 error = bread(devvp, FFS_FSBTODB(fs, cgtod(fs, cg)), 821 fs->fs_cgsize, FSCRED, 0, &bp); 822 if (error) { 823 continue; 824 } 825 cgp = (struct cg *)bp->b_data; 826 if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) { 827 brelse(bp, 0); 828 continue; 829 } 830 831 blksfree = cg_blksfree(cgp, needswap); 832 833 for (blkno = 0; blkno < bpcg;) { 834 /* look for next free block */ 835 /* XXX use scanc() and fragtbl[] here? */ 836 for (; blkno < bpcg - min_desired_blks; blkno++) 837 if (ffs_isblock(fs, blksfree, blkno)) 838 break; 839 840 /* past end of search space in this CG? */ 841 if (blkno >= bpcg - min_desired_blks) 842 break; 843 844 /* count how many free blocks in this extent */ 845 start_addr = blkno; 846 for (freeblks = 0; blkno < bpcg; blkno++, freeblks++) 847 if (!ffs_isblock(fs, blksfree, blkno)) 848 break; 849 850 if (freeblks > best_blks) { 851 best_blks = freeblks; 852 best_addr = ffs_blkstofrags(fs, start_addr) + 853 cgbase(fs, cg); 854 855 if (freeblks >= desired_blks) { 856 DPRINTF("found len %" PRId64 857 " at offset %" PRId64 " in gc\n", 858 freeblks, start_addr); 859 break; 860 } 861 } 862 } 863 brelse(bp, 0); 864 } 865 DPRINTF("best found len = %" PRId64 ", wanted %" PRId64 866 " at addr %" PRId64 "\n", best_blks, desired_blks, best_addr); 867 868 if (best_blks < min_desired_blks) { 869 *addr = 0; 870 *indir_addr = 0; 871 } else { 872 /* put indirect blocks at start, and data blocks after */ 873 *addr = best_addr + ffs_blkstofrags(fs, indir_blks); 874 *indir_addr = best_addr; 875 } 876 *size = min(desired_blks, best_blks) - indir_blks; 877 return; 878 879 bad: 880 *addr = 0; 881 *indir_addr = 0; 882 *size = 0; 883 return; 884 } 885