1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 #include "opt_ufs.h" 34 #include "opt_quota.h" 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/bio.h> 39 #include <sys/buf.h> 40 #include <sys/malloc.h> 41 #include <sys/mount.h> 42 #include <sys/proc.h> 43 #include <sys/racct.h> 44 #include <sys/random.h> 45 #include <sys/resourcevar.h> 46 #include <sys/rwlock.h> 47 #include <sys/stat.h> 48 #include <sys/vmmeter.h> 49 #include <sys/vnode.h> 50 51 #include <vm/vm.h> 52 #include <vm/vm_extern.h> 53 #include <vm/vm_object.h> 54 55 #include <ufs/ufs/extattr.h> 56 #include <ufs/ufs/quota.h> 57 #include <ufs/ufs/ufsmount.h> 58 #include <ufs/ufs/inode.h> 59 #include <ufs/ufs/dir.h> 60 #ifdef UFS_DIRHASH 61 #include <ufs/ufs/dirhash.h> 62 #endif 63 #include <ufs/ufs/ufs_extern.h> 64 65 #include <ufs/ffs/fs.h> 66 #include <ufs/ffs/ffs_extern.h> 67 68 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t, 69 ufs2_daddr_t, int, ufs2_daddr_t *); 70 71 static void 72 ffs_inode_bwrite(struct vnode *vp, struct buf *bp, int flags) 73 { 74 if ((flags & IO_SYNC) != 0) 75 bwrite(bp); 76 else if (DOINGASYNC(vp)) 77 bdwrite(bp); 78 else 79 bawrite(bp); 80 } 81 82 /* 83 * Update the access, modified, and inode change times as specified by the 84 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode 85 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by 86 * the timestamp update). The IN_LAZYMOD flag is set to force a write 87 * later if not now. The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs 88 * is currently being suspended (or is suspended) and vnode has been accessed. 89 * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to 90 * reflect the presumably successful write, and if waitfor is set, then wait 91 * for the write to complete. 92 */ 93 int 94 ffs_update(struct vnode *vp, int waitfor) 95 { 96 struct fs *fs; 97 struct buf *bp; 98 struct inode *ip; 99 daddr_t bn; 100 int flags, error; 101 102 ASSERT_VOP_ELOCKED(vp, "ffs_update"); 103 ufs_itimes(vp); 104 ip = VTOI(vp); 105 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0) 106 return (0); 107 ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED); 108 /* 109 * The IN_SIZEMOD and IN_IBLKDATA flags indicate changes to the 110 * file size and block pointer fields in the inode. When these 111 * fields have been changed, the fsync() and fsyncdata() system 112 * calls must write the inode to ensure their semantics that the 113 * file is on stable store. 114 * 115 * The IN_SIZEMOD and IN_IBLKDATA flags cannot be cleared until 116 * a synchronous write of the inode is done. If they are cleared 117 * on an asynchronous write, then the inode may not yet have been 118 * written to the disk when an fsync() or fsyncdata() call is done. 119 * Absent these flags, these calls would not know that they needed 120 * to write the inode. Thus, these flags only can be cleared on 121 * synchronous writes of the inode. Since the inode will be locked 122 * for the duration of the I/O that writes it to disk, no fsync() 123 * or fsyncdata() will be able to run before the on-disk inode 124 * is complete. 125 */ 126 if (waitfor) 127 ip->i_flag &= ~(IN_SIZEMOD | IN_IBLKDATA); 128 fs = ITOFS(ip); 129 if (fs->fs_ronly) 130 return (0); 131 /* 132 * If we are updating a snapshot and another process is currently 133 * writing the buffer containing the inode for this snapshot then 134 * a deadlock can occur when it tries to check the snapshot to see 135 * if that block needs to be copied. Thus when updating a snapshot 136 * we check to see if the buffer is already locked, and if it is 137 * we drop the snapshot lock until the buffer has been written 138 * and is available to us. We have to grab a reference to the 139 * snapshot vnode to prevent it from being removed while we are 140 * waiting for the buffer. 141 */ 142 loop: 143 flags = 0; 144 if (IS_SNAPSHOT(ip)) 145 flags = GB_LOCK_NOWAIT; 146 bn = fsbtodb(fs, ino_to_fsba(fs, ip->i_number)); 147 error = ffs_breadz(VFSTOUFS(vp->v_mount), ITODEVVP(ip), bn, bn, 148 (int) fs->fs_bsize, NULL, NULL, 0, NOCRED, flags, NULL, &bp); 149 if (error != 0) { 150 /* 151 * If EBUSY was returned without GB_LOCK_NOWAIT (which 152 * requests trylock for buffer lock), it is for some 153 * other reason and we should not handle it specially. 154 */ 155 if (error != EBUSY || (flags & GB_LOCK_NOWAIT) == 0) 156 return (error); 157 158 /* 159 * Wait for our inode block to become available. 160 * 161 * Hold a reference to the vnode to protect against 162 * ffs_snapgone(). Since we hold a reference, it can only 163 * get reclaimed (VIRF_DOOMED flag) in a forcible downgrade 164 * or unmount. For an unmount, the entire filesystem will be 165 * gone, so we cannot attempt to touch anything associated 166 * with it while the vnode is unlocked; all we can do is 167 * pause briefly and try again. If when we relock the vnode 168 * we discover that it has been reclaimed, updating it is no 169 * longer necessary and we can just return an error. 170 */ 171 vref(vp); 172 VOP_UNLOCK(vp); 173 pause("ffsupd", 1); 174 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 175 vrele(vp); 176 if (!IS_UFS(vp)) 177 return (ENOENT); 178 179 /* 180 * Recalculate flags, because the vnode was relocked and 181 * could no longer be a snapshot. 182 */ 183 goto loop; 184 } 185 if (DOINGSOFTDEP(vp)) 186 softdep_update_inodeblock(ip, bp, waitfor); 187 else if (ip->i_effnlink != ip->i_nlink) 188 panic("ffs_update: bad link cnt"); 189 if (I_IS_UFS1(ip)) { 190 *((struct ufs1_dinode *)bp->b_data + 191 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1; 192 /* 193 * XXX: FIX? The entropy here is desirable, 194 * but the harvesting may be expensive 195 */ 196 random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), RANDOM_FS_ATIME); 197 } else { 198 ffs_update_dinode_ckhash(fs, ip->i_din2); 199 *((struct ufs2_dinode *)bp->b_data + 200 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2; 201 /* 202 * XXX: FIX? The entropy here is desirable, 203 * but the harvesting may be expensive 204 */ 205 random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), RANDOM_FS_ATIME); 206 } 207 if (waitfor) { 208 error = bwrite(bp); 209 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error)) 210 error = 0; 211 } else if (vm_page_count_severe() || buf_dirty_count_severe()) { 212 bawrite(bp); 213 error = 0; 214 } else { 215 if (bp->b_bufsize == fs->fs_bsize) 216 bp->b_flags |= B_CLUSTEROK; 217 bdwrite(bp); 218 error = 0; 219 } 220 return (error); 221 } 222 223 #define SINGLE 0 /* index of single indirect block */ 224 #define DOUBLE 1 /* index of double indirect block */ 225 #define TRIPLE 2 /* index of triple indirect block */ 226 /* 227 * Truncate the inode ip to at most length size, freeing the 228 * disk blocks. 229 */ 230 int 231 ffs_truncate(struct vnode *vp, 232 off_t length, 233 int flags, 234 struct ucred *cred) 235 { 236 struct inode *ip; 237 ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR]; 238 ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR]; 239 #ifdef INVARIANTS 240 ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR]; 241 #endif 242 ufs2_daddr_t count, blocksreleased = 0, blkno; 243 struct bufobj *bo __diagused; 244 struct fs *fs; 245 struct buf *bp; 246 struct ufsmount *ump; 247 int softdeptrunc, journaltrunc; 248 int needextclean, extblocks; 249 int offset, size, level, nblocks; 250 int i, error, allerror, indiroff, waitforupdate; 251 uint64_t key; 252 off_t osize; 253 254 ip = VTOI(vp); 255 ump = VFSTOUFS(vp->v_mount); 256 fs = ump->um_fs; 257 bo = &vp->v_bufobj; 258 259 ASSERT_VOP_LOCKED(vp, "ffs_truncate"); 260 261 if (length < 0) 262 return (EINVAL); 263 if (length > fs->fs_maxfilesize) 264 return (EFBIG); 265 #ifdef QUOTA 266 error = getinoquota(ip); 267 if (error) 268 return (error); 269 #endif 270 /* 271 * Historically clients did not have to specify which data 272 * they were truncating. So, if not specified, we assume 273 * traditional behavior, e.g., just the normal data. 274 */ 275 if ((flags & (IO_EXT | IO_NORMAL)) == 0) 276 flags |= IO_NORMAL; 277 if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp)) 278 flags |= IO_SYNC; 279 waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp); 280 /* 281 * If we are truncating the extended-attributes, and cannot 282 * do it with soft updates, then do it slowly here. If we are 283 * truncating both the extended attributes and the file contents 284 * (e.g., the file is being unlinked), then pick it off with 285 * soft updates below. 286 */ 287 allerror = 0; 288 needextclean = 0; 289 softdeptrunc = 0; 290 journaltrunc = DOINGSUJ(vp); 291 journaltrunc = 0; /* XXX temp patch until bug found */ 292 if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0) 293 softdeptrunc = !softdep_slowdown(vp); 294 extblocks = 0; 295 if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) { 296 extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize)); 297 } 298 if ((flags & IO_EXT) && extblocks > 0) { 299 if (length != 0) 300 panic("ffs_truncate: partial trunc of extdata"); 301 if (softdeptrunc || journaltrunc) { 302 if ((flags & IO_NORMAL) == 0) 303 goto extclean; 304 needextclean = 1; 305 } else { 306 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 307 return (error); 308 #ifdef QUOTA 309 (void) chkdq(ip, -extblocks, NOCRED, FORCE); 310 #endif 311 vinvalbuf(vp, V_ALT, 0, 0); 312 vn_pages_remove(vp, 313 OFF_TO_IDX(lblktosize(fs, -extblocks)), 0); 314 osize = ip->i_din2->di_extsize; 315 ip->i_din2->di_blocks -= extblocks; 316 ip->i_din2->di_extsize = 0; 317 for (i = 0; i < UFS_NXADDR; i++) { 318 oldblks[i] = ip->i_din2->di_extb[i]; 319 ip->i_din2->di_extb[i] = 0; 320 } 321 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 322 if ((error = ffs_update(vp, waitforupdate))) 323 return (error); 324 for (i = 0; i < UFS_NXADDR; i++) { 325 if (oldblks[i] == 0) 326 continue; 327 ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i], 328 sblksize(fs, osize, i), ip->i_number, 329 vp->v_type, NULL, SINGLETON_KEY); 330 } 331 } 332 } 333 if ((flags & IO_NORMAL) == 0) 334 return (0); 335 if (vp->v_type == VLNK && ip->i_size < ump->um_maxsymlinklen) { 336 #ifdef INVARIANTS 337 if (length != 0) 338 panic("ffs_truncate: partial truncate of symlink"); 339 #endif 340 bzero(DIP(ip, i_shortlink), (uint64_t)ip->i_size); 341 ip->i_size = 0; 342 DIP_SET(ip, i_size, 0); 343 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 344 if (needextclean) 345 goto extclean; 346 return (ffs_update(vp, waitforupdate)); 347 } 348 if (ip->i_size == length) { 349 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 350 if (needextclean) 351 goto extclean; 352 return (ffs_update(vp, 0)); 353 } 354 if (fs->fs_ronly) 355 panic("ffs_truncate: read-only filesystem"); 356 if (IS_SNAPSHOT(ip)) 357 ffs_snapremove(vp); 358 cluster_init_vn(&ip->i_clusterw); 359 osize = ip->i_size; 360 /* 361 * Lengthen the size of the file. We must ensure that the 362 * last byte of the file is allocated. Since the smallest 363 * value of osize is 0, length will be at least 1. 364 */ 365 if (osize < length) { 366 vnode_pager_setsize(vp, length); 367 flags |= BA_CLRBUF; 368 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 369 if (error) { 370 vnode_pager_setsize(vp, osize); 371 return (error); 372 } 373 ip->i_size = length; 374 DIP_SET(ip, i_size, length); 375 if (bp->b_bufsize == fs->fs_bsize) 376 bp->b_flags |= B_CLUSTEROK; 377 ffs_inode_bwrite(vp, bp, flags); 378 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 379 return (ffs_update(vp, waitforupdate)); 380 } 381 /* 382 * Lookup block number for a given offset. Zero length files 383 * have no blocks, so return a blkno of -1. 384 */ 385 lbn = lblkno(fs, length - 1); 386 if (length == 0) { 387 blkno = -1; 388 } else if (lbn < UFS_NDADDR) { 389 blkno = DIP(ip, i_db[lbn]); 390 } else { 391 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, 392 cred, BA_METAONLY, &bp); 393 if (error) 394 return (error); 395 indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); 396 if (I_IS_UFS1(ip)) 397 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; 398 else 399 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; 400 /* 401 * If the block number is non-zero, then the indirect block 402 * must have been previously allocated and need not be written. 403 * If the block number is zero, then we may have allocated 404 * the indirect block and hence need to write it out. 405 */ 406 if (blkno != 0) 407 brelse(bp); 408 else if (flags & IO_SYNC) 409 bwrite(bp); 410 else 411 bdwrite(bp); 412 } 413 /* 414 * If the block number at the new end of the file is zero, 415 * then we must allocate it to ensure that the last block of 416 * the file is allocated. Soft updates does not handle this 417 * case, so here we have to clean up the soft updates data 418 * structures describing the allocation past the truncation 419 * point. Finding and deallocating those structures is a lot of 420 * work. Since partial truncation with a hole at the end occurs 421 * rarely, we solve the problem by syncing the file so that it 422 * will have no soft updates data structures left. 423 */ 424 if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 425 return (error); 426 if (blkno != 0 && DOINGSOFTDEP(vp)) { 427 if (softdeptrunc == 0 && journaltrunc == 0) { 428 /* 429 * If soft updates cannot handle this truncation, 430 * clean up soft dependency data structures and 431 * fall through to the synchronous truncation. 432 */ 433 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 434 return (error); 435 } else { 436 flags = IO_NORMAL | (needextclean ? IO_EXT: 0); 437 if (journaltrunc) 438 softdep_journal_freeblocks(ip, cred, length, 439 flags); 440 else 441 softdep_setup_freeblocks(ip, length, flags); 442 ASSERT_VOP_LOCKED(vp, "ffs_truncate1"); 443 if (journaltrunc == 0) { 444 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 445 error = ffs_update(vp, 0); 446 } 447 return (error); 448 } 449 } 450 /* 451 * Shorten the size of the file. If the last block of the 452 * shortened file is unallocated, we must allocate it. 453 * Additionally, if the file is not being truncated to a 454 * block boundary, the contents of the partial block 455 * following the end of the file must be zero'ed in 456 * case it ever becomes accessible again because of 457 * subsequent file growth. Directories however are not 458 * zero'ed as they should grow back initialized to empty. 459 */ 460 offset = blkoff(fs, length); 461 if (blkno != 0 && offset == 0) { 462 ip->i_size = length; 463 DIP_SET(ip, i_size, length); 464 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 465 #ifdef UFS_DIRHASH 466 if (vp->v_type == VDIR && ip->i_dirhash != NULL) 467 ufsdirhash_dirtrunc(ip, length); 468 #endif 469 } else { 470 lbn = lblkno(fs, length); 471 flags |= BA_CLRBUF; 472 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 473 if (error) 474 return (error); 475 ffs_inode_bwrite(vp, bp, flags); 476 477 /* 478 * When we are doing soft updates and the UFS_BALLOC 479 * above fills in a direct block hole with a full sized 480 * block that will be truncated down to a fragment below, 481 * we must flush out the block dependency with an FSYNC 482 * so that we do not get a soft updates inconsistency 483 * when we create the fragment below. 484 */ 485 if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR && 486 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize && 487 (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 488 return (error); 489 490 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 491 if (error) 492 return (error); 493 ip->i_size = length; 494 DIP_SET(ip, i_size, length); 495 #ifdef UFS_DIRHASH 496 if (vp->v_type == VDIR && ip->i_dirhash != NULL) 497 ufsdirhash_dirtrunc(ip, length); 498 #endif 499 size = blksize(fs, ip, lbn); 500 if (vp->v_type != VDIR && offset != 0) 501 bzero((char *)bp->b_data + offset, 502 (uint64_t)(size - offset)); 503 /* Kirk's code has reallocbuf(bp, size, 1) here */ 504 allocbuf(bp, size); 505 if (bp->b_bufsize == fs->fs_bsize) 506 bp->b_flags |= B_CLUSTEROK; 507 ffs_inode_bwrite(vp, bp, flags); 508 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 509 } 510 /* 511 * Calculate index into inode's block list of 512 * last direct and indirect blocks (if any) 513 * which we want to keep. Lastblock is -1 when 514 * the file is truncated to 0. 515 */ 516 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 517 lastiblock[SINGLE] = lastblock - UFS_NDADDR; 518 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 519 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 520 nblocks = btodb(fs->fs_bsize); 521 /* 522 * Update file and block pointers on disk before we start freeing 523 * blocks. If we crash before free'ing blocks below, the blocks 524 * will be returned to the free list. lastiblock values are also 525 * normalized to -1 for calls to ffs_indirtrunc below. 526 */ 527 for (level = TRIPLE; level >= SINGLE; level--) { 528 oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]); 529 if (lastiblock[level] < 0) { 530 DIP_SET(ip, i_ib[level], 0); 531 lastiblock[level] = -1; 532 } 533 } 534 for (i = 0; i < UFS_NDADDR; i++) { 535 oldblks[i] = DIP(ip, i_db[i]); 536 if (i > lastblock) 537 DIP_SET(ip, i_db[i], 0); 538 } 539 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 540 allerror = ffs_update(vp, waitforupdate); 541 542 /* 543 * Having written the new inode to disk, save its new configuration 544 * and put back the old block pointers long enough to process them. 545 * Note that we save the new block configuration so we can check it 546 * when we are done. 547 */ 548 for (i = 0; i < UFS_NDADDR; i++) { 549 #ifdef INVARIANTS 550 newblks[i] = DIP(ip, i_db[i]); 551 #endif 552 DIP_SET(ip, i_db[i], oldblks[i]); 553 } 554 for (i = 0; i < UFS_NIADDR; i++) { 555 #ifdef INVARIANTS 556 newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]); 557 #endif 558 DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]); 559 } 560 ip->i_size = osize; 561 DIP_SET(ip, i_size, osize); 562 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 563 564 error = vtruncbuf(vp, length, fs->fs_bsize); 565 if (error && (allerror == 0)) 566 allerror = error; 567 568 /* 569 * Indirect blocks first. 570 */ 571 indir_lbn[SINGLE] = -UFS_NDADDR; 572 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 573 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 574 for (level = TRIPLE; level >= SINGLE; level--) { 575 bn = DIP(ip, i_ib[level]); 576 if (bn != 0) { 577 error = ffs_indirtrunc(ip, indir_lbn[level], 578 fsbtodb(fs, bn), lastiblock[level], level, &count); 579 if (error) 580 allerror = error; 581 blocksreleased += count; 582 if (lastiblock[level] < 0) { 583 DIP_SET(ip, i_ib[level], 0); 584 ffs_blkfree(ump, fs, ump->um_devvp, bn, 585 fs->fs_bsize, ip->i_number, 586 vp->v_type, NULL, SINGLETON_KEY); 587 blocksreleased += nblocks; 588 } 589 } 590 if (lastiblock[level] >= 0) 591 goto done; 592 } 593 594 /* 595 * All whole direct blocks or frags. 596 */ 597 key = ffs_blkrelease_start(ump, ump->um_devvp, ip->i_number); 598 for (i = UFS_NDADDR - 1; i > lastblock; i--) { 599 long bsize; 600 601 bn = DIP(ip, i_db[i]); 602 if (bn == 0) 603 continue; 604 DIP_SET(ip, i_db[i], 0); 605 bsize = blksize(fs, ip, i); 606 ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number, 607 vp->v_type, NULL, key); 608 blocksreleased += btodb(bsize); 609 } 610 ffs_blkrelease_finish(ump, key); 611 if (lastblock < 0) 612 goto done; 613 614 /* 615 * Finally, look for a change in size of the 616 * last direct block; release any frags. 617 */ 618 bn = DIP(ip, i_db[lastblock]); 619 if (bn != 0) { 620 long oldspace, newspace; 621 622 /* 623 * Calculate amount of space we're giving 624 * back as old block size minus new block size. 625 */ 626 oldspace = blksize(fs, ip, lastblock); 627 ip->i_size = length; 628 DIP_SET(ip, i_size, length); 629 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 630 newspace = blksize(fs, ip, lastblock); 631 if (newspace == 0) 632 panic("ffs_truncate: newspace"); 633 if (oldspace - newspace > 0) { 634 /* 635 * Block number of space to be free'd is 636 * the old block # plus the number of frags 637 * required for the storage we're keeping. 638 */ 639 bn += numfrags(fs, newspace); 640 ffs_blkfree(ump, fs, ump->um_devvp, bn, 641 oldspace - newspace, ip->i_number, vp->v_type, 642 NULL, SINGLETON_KEY); 643 blocksreleased += btodb(oldspace - newspace); 644 } 645 } 646 done: 647 #ifdef INVARIANTS 648 for (level = SINGLE; level <= TRIPLE; level++) 649 if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level])) 650 panic("ffs_truncate1: level %d newblks %jd != i_ib %jd", 651 level, (intmax_t)newblks[UFS_NDADDR + level], 652 (intmax_t)DIP(ip, i_ib[level])); 653 for (i = 0; i < UFS_NDADDR; i++) 654 if (newblks[i] != DIP(ip, i_db[i])) 655 panic("ffs_truncate2: blkno %d newblks %jd != i_db %jd", 656 i, (intmax_t)newblks[UFS_NDADDR + level], 657 (intmax_t)DIP(ip, i_ib[level])); 658 BO_LOCK(bo); 659 if (length == 0 && 660 (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) && 661 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 662 panic("ffs_truncate3: vp = %p, buffers: dirty = %d, clean = %d", 663 vp, bo->bo_dirty.bv_cnt, bo->bo_clean.bv_cnt); 664 BO_UNLOCK(bo); 665 #endif /* INVARIANTS */ 666 /* 667 * Put back the real size. 668 */ 669 ip->i_size = length; 670 DIP_SET(ip, i_size, length); 671 if (DIP(ip, i_blocks) >= blocksreleased) 672 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased); 673 else /* sanity */ 674 DIP_SET(ip, i_blocks, 0); 675 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 676 #ifdef QUOTA 677 (void) chkdq(ip, -blocksreleased, NOCRED, FORCE); 678 #endif 679 return (allerror); 680 681 extclean: 682 if (journaltrunc) 683 softdep_journal_freeblocks(ip, cred, length, IO_EXT); 684 else 685 softdep_setup_freeblocks(ip, length, IO_EXT); 686 return (ffs_update(vp, waitforupdate)); 687 } 688 689 /* 690 * Release blocks associated with the inode ip and stored in the indirect 691 * block bn. Blocks are free'd in LIFO order up to (but not including) 692 * lastbn. If level is greater than SINGLE, the block is an indirect block 693 * and recursive calls to indirtrunc must be used to cleanse other indirect 694 * blocks. 695 */ 696 static int 697 ffs_indirtrunc(struct inode *ip, 698 ufs2_daddr_t lbn, 699 ufs2_daddr_t dbn, 700 ufs2_daddr_t lastbn, 701 int level, 702 ufs2_daddr_t *countp) 703 { 704 struct buf *bp; 705 struct fs *fs; 706 struct ufsmount *ump; 707 struct vnode *vp; 708 caddr_t copy = NULL; 709 uint64_t key; 710 int i, nblocks, error = 0, allerror = 0; 711 ufs2_daddr_t nb, nlbn, last; 712 ufs2_daddr_t blkcount, factor, blocksreleased = 0; 713 ufs1_daddr_t *bap1 = NULL; 714 ufs2_daddr_t *bap2 = NULL; 715 #define BAP(ip, i) (I_IS_UFS1(ip) ? bap1[i] : bap2[i]) 716 717 fs = ITOFS(ip); 718 ump = ITOUMP(ip); 719 720 /* 721 * Calculate index in current block of last 722 * block to be kept. -1 indicates the entire 723 * block so we need not calculate the index. 724 */ 725 factor = lbn_offset(fs, level); 726 last = lastbn; 727 if (lastbn > 0) 728 last /= factor; 729 nblocks = btodb(fs->fs_bsize); 730 /* 731 * Get buffer of block pointers, zero those entries corresponding 732 * to blocks to be free'd, and update on disk copy first. Since 733 * double(triple) indirect before single(double) indirect, calls 734 * to VOP_BMAP() on these blocks will fail. However, we already 735 * have the on-disk address, so we just pass it to bread() instead 736 * of having bread() attempt to calculate it using VOP_BMAP(). 737 */ 738 vp = ITOV(ip); 739 error = ffs_breadz(ump, vp, lbn, dbn, (int)fs->fs_bsize, NULL, NULL, 0, 740 NOCRED, 0, NULL, &bp); 741 if (error) { 742 *countp = 0; 743 return (error); 744 } 745 746 if (I_IS_UFS1(ip)) 747 bap1 = (ufs1_daddr_t *)bp->b_data; 748 else 749 bap2 = (ufs2_daddr_t *)bp->b_data; 750 if (lastbn != -1) { 751 copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK); 752 bcopy((caddr_t)bp->b_data, copy, (uint64_t)fs->fs_bsize); 753 for (i = last + 1; i < NINDIR(fs); i++) 754 if (I_IS_UFS1(ip)) 755 bap1[i] = 0; 756 else 757 bap2[i] = 0; 758 if (DOINGASYNC(vp)) { 759 bdwrite(bp); 760 } else { 761 error = bwrite(bp); 762 if (error) 763 allerror = error; 764 } 765 if (I_IS_UFS1(ip)) 766 bap1 = (ufs1_daddr_t *)copy; 767 else 768 bap2 = (ufs2_daddr_t *)copy; 769 } 770 771 /* 772 * Recursively free totally unused blocks. 773 */ 774 key = ffs_blkrelease_start(ump, ITODEVVP(ip), ip->i_number); 775 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 776 i--, nlbn += factor) { 777 nb = BAP(ip, i); 778 if (nb == 0) 779 continue; 780 if (level > SINGLE) { 781 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 782 (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0) 783 allerror = error; 784 blocksreleased += blkcount; 785 } 786 ffs_blkfree(ump, fs, ITODEVVP(ip), nb, fs->fs_bsize, 787 ip->i_number, vp->v_type, NULL, key); 788 blocksreleased += nblocks; 789 } 790 ffs_blkrelease_finish(ump, key); 791 792 /* 793 * Recursively free last partial block. 794 */ 795 if (level > SINGLE && lastbn >= 0) { 796 last = lastbn % factor; 797 nb = BAP(ip, i); 798 if (nb != 0) { 799 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 800 last, level - 1, &blkcount); 801 if (error) 802 allerror = error; 803 blocksreleased += blkcount; 804 } 805 } 806 if (copy != NULL) { 807 free(copy, M_TEMP); 808 } else { 809 bp->b_flags |= B_INVAL | B_NOCACHE; 810 brelse(bp); 811 } 812 813 *countp = blocksreleased; 814 return (allerror); 815 } 816 817 int 818 ffs_rdonly(struct inode *ip) 819 { 820 821 return (ITOFS(ip)->fs_ronly != 0); 822 } 823