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 * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_ufs.h" 38 #include "opt_quota.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/bio.h> 43 #include <sys/buf.h> 44 #include <sys/malloc.h> 45 #include <sys/mount.h> 46 #include <sys/proc.h> 47 #include <sys/racct.h> 48 #include <sys/random.h> 49 #include <sys/resourcevar.h> 50 #include <sys/rwlock.h> 51 #include <sys/stat.h> 52 #include <sys/vmmeter.h> 53 #include <sys/vnode.h> 54 55 #include <vm/vm.h> 56 #include <vm/vm_extern.h> 57 #include <vm/vm_object.h> 58 59 #include <ufs/ufs/extattr.h> 60 #include <ufs/ufs/quota.h> 61 #include <ufs/ufs/ufsmount.h> 62 #include <ufs/ufs/inode.h> 63 #include <ufs/ufs/dir.h> 64 #ifdef UFS_DIRHASH 65 #include <ufs/ufs/dirhash.h> 66 #endif 67 #include <ufs/ufs/ufs_extern.h> 68 69 #include <ufs/ffs/fs.h> 70 #include <ufs/ffs/ffs_extern.h> 71 72 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t, 73 ufs2_daddr_t, int, ufs2_daddr_t *); 74 75 static void 76 ffs_inode_bwrite(struct vnode *vp, struct buf *bp, int flags) 77 { 78 if ((flags & IO_SYNC) != 0) 79 bwrite(bp); 80 else if (DOINGASYNC(vp)) 81 bdwrite(bp); 82 else 83 bawrite(bp); 84 } 85 86 /* 87 * Update the access, modified, and inode change times as specified by the 88 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode 89 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by 90 * the timestamp update). The IN_LAZYMOD flag is set to force a write 91 * later if not now. The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs 92 * is currently being suspended (or is suspended) and vnode has been accessed. 93 * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to 94 * reflect the presumably successful write, and if waitfor is set, then wait 95 * for the write to complete. 96 */ 97 int 98 ffs_update(vp, waitfor) 99 struct vnode *vp; 100 int waitfor; 101 { 102 struct fs *fs; 103 struct buf *bp; 104 struct inode *ip; 105 daddr_t bn; 106 int flags, error; 107 108 ASSERT_VOP_ELOCKED(vp, "ffs_update"); 109 ufs_itimes(vp); 110 ip = VTOI(vp); 111 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0) 112 return (0); 113 ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED); 114 /* 115 * The IN_SIZEMOD and IN_IBLKDATA flags indicate changes to the 116 * file size and block pointer fields in the inode. When these 117 * fields have been changed, the fsync() and fsyncdata() system 118 * calls must write the inode to ensure their semantics that the 119 * file is on stable store. 120 * 121 * The IN_SIZEMOD and IN_IBLKDATA flags cannot be cleared until 122 * a synchronous write of the inode is done. If they are cleared 123 * on an asynchronous write, then the inode may not yet have been 124 * written to the disk when an fsync() or fsyncdata() call is done. 125 * Absent these flags, these calls would not know that they needed 126 * to write the inode. Thus, these flags only can be cleared on 127 * synchronous writes of the inode. Since the inode will be locked 128 * for the duration of the I/O that writes it to disk, no fsync() 129 * or fsyncdata() will be able to run before the on-disk inode 130 * is complete. 131 */ 132 if (waitfor) 133 ip->i_flag &= ~(IN_SIZEMOD | IN_IBLKDATA); 134 fs = ITOFS(ip); 135 if (fs->fs_ronly && ITOUMP(ip)->um_fsckpid == 0) 136 return (0); 137 /* 138 * If we are updating a snapshot and another process is currently 139 * writing the buffer containing the inode for this snapshot then 140 * a deadlock can occur when it tries to check the snapshot to see 141 * if that block needs to be copied. Thus when updating a snapshot 142 * we check to see if the buffer is already locked, and if it is 143 * we drop the snapshot lock until the buffer has been written 144 * and is available to us. We have to grab a reference to the 145 * snapshot vnode to prevent it from being removed while we are 146 * waiting for the buffer. 147 */ 148 flags = 0; 149 if (IS_SNAPSHOT(ip)) 150 flags = GB_LOCK_NOWAIT; 151 loop: 152 bn = fsbtodb(fs, ino_to_fsba(fs, ip->i_number)); 153 error = ffs_breadz(VFSTOUFS(vp->v_mount), ITODEVVP(ip), bn, bn, 154 (int) fs->fs_bsize, NULL, NULL, 0, NOCRED, flags, NULL, &bp); 155 if (error != 0) { 156 if (error != EBUSY) 157 return (error); 158 KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot")); 159 /* 160 * Wait for our inode block to become available. 161 * 162 * Hold a reference to the vnode to protect against 163 * ffs_snapgone(). Since we hold a reference, it can only 164 * get reclaimed (VIRF_DOOMED flag) in a forcible downgrade 165 * or unmount. For an unmount, the entire filesystem will be 166 * gone, so we cannot attempt to touch anything associated 167 * with it while the vnode is unlocked; all we can do is 168 * pause briefly and try again. If when we relock the vnode 169 * we discover that it has been reclaimed, updating it is no 170 * longer necessary and we can just return an error. 171 */ 172 vref(vp); 173 VOP_UNLOCK(vp); 174 pause("ffsupd", 1); 175 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 176 vrele(vp); 177 if (VN_IS_DOOMED(vp)) 178 return (ENOENT); 179 goto loop; 180 } 181 if (DOINGSOFTDEP(vp)) 182 softdep_update_inodeblock(ip, bp, waitfor); 183 else if (ip->i_effnlink != ip->i_nlink) 184 panic("ffs_update: bad link cnt"); 185 if (I_IS_UFS1(ip)) { 186 *((struct ufs1_dinode *)bp->b_data + 187 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1; 188 /* 189 * XXX: FIX? The entropy here is desirable, 190 * but the harvesting may be expensive 191 */ 192 random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), RANDOM_FS_ATIME); 193 } else { 194 ffs_update_dinode_ckhash(fs, ip->i_din2); 195 *((struct ufs2_dinode *)bp->b_data + 196 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2; 197 /* 198 * XXX: FIX? The entropy here is desirable, 199 * but the harvesting may be expensive 200 */ 201 random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), RANDOM_FS_ATIME); 202 } 203 if (waitfor) { 204 error = bwrite(bp); 205 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error)) 206 error = 0; 207 } else if (vm_page_count_severe() || buf_dirty_count_severe()) { 208 bawrite(bp); 209 error = 0; 210 } else { 211 if (bp->b_bufsize == fs->fs_bsize) 212 bp->b_flags |= B_CLUSTEROK; 213 bdwrite(bp); 214 error = 0; 215 } 216 return (error); 217 } 218 219 #define SINGLE 0 /* index of single indirect block */ 220 #define DOUBLE 1 /* index of double indirect block */ 221 #define TRIPLE 2 /* index of triple indirect block */ 222 /* 223 * Truncate the inode ip to at most length size, freeing the 224 * disk blocks. 225 */ 226 int 227 ffs_truncate(vp, length, flags, cred) 228 struct vnode *vp; 229 off_t length; 230 int flags; 231 struct ucred *cred; 232 { 233 struct inode *ip; 234 ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR]; 235 ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR]; 236 ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR]; 237 ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno; 238 struct bufobj *bo; 239 struct fs *fs; 240 struct buf *bp; 241 struct ufsmount *ump; 242 int softdeptrunc, journaltrunc; 243 int needextclean, extblocks; 244 int offset, size, level, nblocks; 245 int i, error, allerror, indiroff, waitforupdate; 246 u_long key; 247 off_t osize; 248 249 ip = VTOI(vp); 250 ump = VFSTOUFS(vp->v_mount); 251 fs = ump->um_fs; 252 bo = &vp->v_bufobj; 253 254 ASSERT_VOP_LOCKED(vp, "ffs_truncate"); 255 256 if (length < 0) 257 return (EINVAL); 258 if (length > fs->fs_maxfilesize) 259 return (EFBIG); 260 #ifdef QUOTA 261 error = getinoquota(ip); 262 if (error) 263 return (error); 264 #endif 265 /* 266 * Historically clients did not have to specify which data 267 * they were truncating. So, if not specified, we assume 268 * traditional behavior, e.g., just the normal data. 269 */ 270 if ((flags & (IO_EXT | IO_NORMAL)) == 0) 271 flags |= IO_NORMAL; 272 if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp)) 273 flags |= IO_SYNC; 274 waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp); 275 /* 276 * If we are truncating the extended-attributes, and cannot 277 * do it with soft updates, then do it slowly here. If we are 278 * truncating both the extended attributes and the file contents 279 * (e.g., the file is being unlinked), then pick it off with 280 * soft updates below. 281 */ 282 allerror = 0; 283 needextclean = 0; 284 softdeptrunc = 0; 285 journaltrunc = DOINGSUJ(vp); 286 journaltrunc = 0; /* XXX temp patch until bug found */ 287 if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0) 288 softdeptrunc = !softdep_slowdown(vp); 289 extblocks = 0; 290 datablocks = DIP(ip, i_blocks); 291 if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) { 292 extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize)); 293 datablocks -= extblocks; 294 } 295 if ((flags & IO_EXT) && extblocks > 0) { 296 if (length != 0) 297 panic("ffs_truncate: partial trunc of extdata"); 298 if (softdeptrunc || journaltrunc) { 299 if ((flags & IO_NORMAL) == 0) 300 goto extclean; 301 needextclean = 1; 302 } else { 303 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 304 return (error); 305 #ifdef QUOTA 306 (void) chkdq(ip, -extblocks, NOCRED, FORCE); 307 #endif 308 vinvalbuf(vp, V_ALT, 0, 0); 309 vn_pages_remove(vp, 310 OFF_TO_IDX(lblktosize(fs, -extblocks)), 0); 311 osize = ip->i_din2->di_extsize; 312 ip->i_din2->di_blocks -= extblocks; 313 ip->i_din2->di_extsize = 0; 314 for (i = 0; i < UFS_NXADDR; i++) { 315 oldblks[i] = ip->i_din2->di_extb[i]; 316 ip->i_din2->di_extb[i] = 0; 317 } 318 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 319 if ((error = ffs_update(vp, waitforupdate))) 320 return (error); 321 for (i = 0; i < UFS_NXADDR; i++) { 322 if (oldblks[i] == 0) 323 continue; 324 ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i], 325 sblksize(fs, osize, i), ip->i_number, 326 vp->v_type, NULL, SINGLETON_KEY); 327 } 328 } 329 } 330 if ((flags & IO_NORMAL) == 0) 331 return (0); 332 if (vp->v_type == VLNK && 333 (ip->i_size < vp->v_mount->mnt_maxsymlinklen || 334 datablocks == 0)) { 335 #ifdef INVARIANTS 336 if (length != 0) 337 panic("ffs_truncate: partial truncate of symlink"); 338 #endif 339 bzero(SHORTLINK(ip), (u_int)ip->i_size); 340 ip->i_size = 0; 341 DIP_SET(ip, i_size, 0); 342 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 343 if (needextclean) 344 goto extclean; 345 return (ffs_update(vp, waitforupdate)); 346 } 347 if (ip->i_size == length) { 348 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 349 if (needextclean) 350 goto extclean; 351 return (ffs_update(vp, 0)); 352 } 353 if (fs->fs_ronly) 354 panic("ffs_truncate: read-only filesystem"); 355 if (IS_SNAPSHOT(ip)) 356 ffs_snapremove(vp); 357 cluster_init_vn(&ip->i_clusterw); 358 osize = ip->i_size; 359 /* 360 * Lengthen the size of the file. We must ensure that the 361 * last byte of the file is allocated. Since the smallest 362 * value of osize is 0, length will be at least 1. 363 */ 364 if (osize < length) { 365 vnode_pager_setsize(vp, length); 366 flags |= BA_CLRBUF; 367 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 368 if (error) { 369 vnode_pager_setsize(vp, osize); 370 return (error); 371 } 372 ip->i_size = length; 373 DIP_SET(ip, i_size, length); 374 if (bp->b_bufsize == fs->fs_bsize) 375 bp->b_flags |= B_CLUSTEROK; 376 ffs_inode_bwrite(vp, bp, flags); 377 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 378 return (ffs_update(vp, waitforupdate)); 379 } 380 /* 381 * Lookup block number for a given offset. Zero length files 382 * have no blocks, so return a blkno of -1. 383 */ 384 lbn = lblkno(fs, length - 1); 385 if (length == 0) { 386 blkno = -1; 387 } else if (lbn < UFS_NDADDR) { 388 blkno = DIP(ip, i_db[lbn]); 389 } else { 390 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, 391 cred, BA_METAONLY, &bp); 392 if (error) 393 return (error); 394 indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); 395 if (I_IS_UFS1(ip)) 396 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; 397 else 398 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; 399 /* 400 * If the block number is non-zero, then the indirect block 401 * must have been previously allocated and need not be written. 402 * If the block number is zero, then we may have allocated 403 * the indirect block and hence need to write it out. 404 */ 405 if (blkno != 0) 406 brelse(bp); 407 else if (flags & IO_SYNC) 408 bwrite(bp); 409 else 410 bdwrite(bp); 411 } 412 /* 413 * If the block number at the new end of the file is zero, 414 * then we must allocate it to ensure that the last block of 415 * the file is allocated. Soft updates does not handle this 416 * case, so here we have to clean up the soft updates data 417 * structures describing the allocation past the truncation 418 * point. Finding and deallocating those structures is a lot of 419 * work. Since partial truncation with a hole at the end occurs 420 * rarely, we solve the problem by syncing the file so that it 421 * will have no soft updates data structures left. 422 */ 423 if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 424 return (error); 425 if (blkno != 0 && DOINGSOFTDEP(vp)) { 426 if (softdeptrunc == 0 && journaltrunc == 0) { 427 /* 428 * If soft updates cannot handle this truncation, 429 * clean up soft dependency data structures and 430 * fall through to the synchronous truncation. 431 */ 432 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 433 return (error); 434 } else { 435 flags = IO_NORMAL | (needextclean ? IO_EXT: 0); 436 if (journaltrunc) 437 softdep_journal_freeblocks(ip, cred, length, 438 flags); 439 else 440 softdep_setup_freeblocks(ip, length, flags); 441 ASSERT_VOP_LOCKED(vp, "ffs_truncate1"); 442 if (journaltrunc == 0) { 443 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 444 error = ffs_update(vp, 0); 445 } 446 return (error); 447 } 448 } 449 /* 450 * Shorten the size of the file. If the last block of the 451 * shortened file is unallocated, we must allocate it. 452 * Additionally, if the file is not being truncated to a 453 * block boundary, the contents of the partial block 454 * following the end of the file must be zero'ed in 455 * case it ever becomes accessible again because of 456 * subsequent file growth. Directories however are not 457 * zero'ed as they should grow back initialized to empty. 458 */ 459 offset = blkoff(fs, length); 460 if (blkno != 0 && offset == 0) { 461 ip->i_size = length; 462 DIP_SET(ip, i_size, length); 463 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 464 #ifdef UFS_DIRHASH 465 if (vp->v_type == VDIR && ip->i_dirhash != NULL) 466 ufsdirhash_dirtrunc(ip, length); 467 #endif 468 } else { 469 lbn = lblkno(fs, length); 470 flags |= BA_CLRBUF; 471 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 472 if (error) 473 return (error); 474 ffs_inode_bwrite(vp, bp, flags); 475 476 /* 477 * When we are doing soft updates and the UFS_BALLOC 478 * above fills in a direct block hole with a full sized 479 * block that will be truncated down to a fragment below, 480 * we must flush out the block dependency with an FSYNC 481 * so that we do not get a soft updates inconsistency 482 * when we create the fragment below. 483 */ 484 if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR && 485 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize && 486 (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 487 return (error); 488 489 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 490 if (error) 491 return (error); 492 ip->i_size = length; 493 DIP_SET(ip, i_size, length); 494 #ifdef UFS_DIRHASH 495 if (vp->v_type == VDIR && ip->i_dirhash != NULL) 496 ufsdirhash_dirtrunc(ip, length); 497 #endif 498 size = blksize(fs, ip, lbn); 499 if (vp->v_type != VDIR && offset != 0) 500 bzero((char *)bp->b_data + offset, 501 (u_int)(size - offset)); 502 /* Kirk's code has reallocbuf(bp, size, 1) here */ 503 allocbuf(bp, size); 504 if (bp->b_bufsize == fs->fs_bsize) 505 bp->b_flags |= B_CLUSTEROK; 506 ffs_inode_bwrite(vp, bp, flags); 507 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 508 } 509 /* 510 * Calculate index into inode's block list of 511 * last direct and indirect blocks (if any) 512 * which we want to keep. Lastblock is -1 when 513 * the file is truncated to 0. 514 */ 515 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 516 lastiblock[SINGLE] = lastblock - UFS_NDADDR; 517 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 518 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 519 nblocks = btodb(fs->fs_bsize); 520 /* 521 * Update file and block pointers on disk before we start freeing 522 * blocks. If we crash before free'ing blocks below, the blocks 523 * will be returned to the free list. lastiblock values are also 524 * normalized to -1 for calls to ffs_indirtrunc below. 525 */ 526 for (level = TRIPLE; level >= SINGLE; level--) { 527 oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]); 528 if (lastiblock[level] < 0) { 529 DIP_SET(ip, i_ib[level], 0); 530 lastiblock[level] = -1; 531 } 532 } 533 for (i = 0; i < UFS_NDADDR; i++) { 534 oldblks[i] = DIP(ip, i_db[i]); 535 if (i > lastblock) 536 DIP_SET(ip, i_db[i], 0); 537 } 538 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); 539 allerror = ffs_update(vp, waitforupdate); 540 541 /* 542 * Having written the new inode to disk, save its new configuration 543 * and put back the old block pointers long enough to process them. 544 * Note that we save the new block configuration so we can check it 545 * when we are done. 546 */ 547 for (i = 0; i < UFS_NDADDR; i++) { 548 newblks[i] = DIP(ip, i_db[i]); 549 DIP_SET(ip, i_db[i], oldblks[i]); 550 } 551 for (i = 0; i < UFS_NIADDR; i++) { 552 newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]); 553 DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]); 554 } 555 ip->i_size = osize; 556 DIP_SET(ip, i_size, osize); 557 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 558 559 error = vtruncbuf(vp, length, fs->fs_bsize); 560 if (error && (allerror == 0)) 561 allerror = error; 562 563 /* 564 * Indirect blocks first. 565 */ 566 indir_lbn[SINGLE] = -UFS_NDADDR; 567 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 568 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 569 for (level = TRIPLE; level >= SINGLE; level--) { 570 bn = DIP(ip, i_ib[level]); 571 if (bn != 0) { 572 error = ffs_indirtrunc(ip, indir_lbn[level], 573 fsbtodb(fs, bn), lastiblock[level], level, &count); 574 if (error) 575 allerror = error; 576 blocksreleased += count; 577 if (lastiblock[level] < 0) { 578 DIP_SET(ip, i_ib[level], 0); 579 ffs_blkfree(ump, fs, ump->um_devvp, bn, 580 fs->fs_bsize, ip->i_number, 581 vp->v_type, NULL, SINGLETON_KEY); 582 blocksreleased += nblocks; 583 } 584 } 585 if (lastiblock[level] >= 0) 586 goto done; 587 } 588 589 /* 590 * All whole direct blocks or frags. 591 */ 592 key = ffs_blkrelease_start(ump, ump->um_devvp, ip->i_number); 593 for (i = UFS_NDADDR - 1; i > lastblock; i--) { 594 long bsize; 595 596 bn = DIP(ip, i_db[i]); 597 if (bn == 0) 598 continue; 599 DIP_SET(ip, i_db[i], 0); 600 bsize = blksize(fs, ip, i); 601 ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number, 602 vp->v_type, NULL, key); 603 blocksreleased += btodb(bsize); 604 } 605 ffs_blkrelease_finish(ump, key); 606 if (lastblock < 0) 607 goto done; 608 609 /* 610 * Finally, look for a change in size of the 611 * last direct block; release any frags. 612 */ 613 bn = DIP(ip, i_db[lastblock]); 614 if (bn != 0) { 615 long oldspace, newspace; 616 617 /* 618 * Calculate amount of space we're giving 619 * back as old block size minus new block size. 620 */ 621 oldspace = blksize(fs, ip, lastblock); 622 ip->i_size = length; 623 DIP_SET(ip, i_size, length); 624 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); 625 newspace = blksize(fs, ip, lastblock); 626 if (newspace == 0) 627 panic("ffs_truncate: newspace"); 628 if (oldspace - newspace > 0) { 629 /* 630 * Block number of space to be free'd is 631 * the old block # plus the number of frags 632 * required for the storage we're keeping. 633 */ 634 bn += numfrags(fs, newspace); 635 ffs_blkfree(ump, fs, ump->um_devvp, bn, 636 oldspace - newspace, ip->i_number, vp->v_type, 637 NULL, SINGLETON_KEY); 638 blocksreleased += btodb(oldspace - newspace); 639 } 640 } 641 done: 642 #ifdef INVARIANTS 643 for (level = SINGLE; level <= TRIPLE; level++) 644 if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level])) 645 panic("ffs_truncate1: level %d newblks %jd != i_ib %jd", 646 level, (intmax_t)newblks[UFS_NDADDR + level], 647 (intmax_t)DIP(ip, i_ib[level])); 648 for (i = 0; i < UFS_NDADDR; i++) 649 if (newblks[i] != DIP(ip, i_db[i])) 650 panic("ffs_truncate2: blkno %d newblks %jd != i_db %jd", 651 i, (intmax_t)newblks[UFS_NDADDR + level], 652 (intmax_t)DIP(ip, i_ib[level])); 653 BO_LOCK(bo); 654 if (length == 0 && 655 (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) && 656 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 657 panic("ffs_truncate3: vp = %p, buffers: dirty = %d, clean = %d", 658 vp, bo->bo_dirty.bv_cnt, bo->bo_clean.bv_cnt); 659 BO_UNLOCK(bo); 660 #endif /* INVARIANTS */ 661 /* 662 * Put back the real size. 663 */ 664 ip->i_size = length; 665 DIP_SET(ip, i_size, length); 666 if (DIP(ip, i_blocks) >= blocksreleased) 667 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased); 668 else /* sanity */ 669 DIP_SET(ip, i_blocks, 0); 670 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE); 671 #ifdef QUOTA 672 (void) chkdq(ip, -blocksreleased, NOCRED, FORCE); 673 #endif 674 return (allerror); 675 676 extclean: 677 if (journaltrunc) 678 softdep_journal_freeblocks(ip, cred, length, IO_EXT); 679 else 680 softdep_setup_freeblocks(ip, length, IO_EXT); 681 return (ffs_update(vp, waitforupdate)); 682 } 683 684 /* 685 * Release blocks associated with the inode ip and stored in the indirect 686 * block bn. Blocks are free'd in LIFO order up to (but not including) 687 * lastbn. If level is greater than SINGLE, the block is an indirect block 688 * and recursive calls to indirtrunc must be used to cleanse other indirect 689 * blocks. 690 */ 691 static int 692 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp) 693 struct inode *ip; 694 ufs2_daddr_t lbn, lastbn; 695 ufs2_daddr_t dbn; 696 int level; 697 ufs2_daddr_t *countp; 698 { 699 struct buf *bp; 700 struct fs *fs; 701 struct ufsmount *ump; 702 struct vnode *vp; 703 caddr_t copy = NULL; 704 u_long key; 705 int i, nblocks, error = 0, allerror = 0; 706 ufs2_daddr_t nb, nlbn, last; 707 ufs2_daddr_t blkcount, factor, blocksreleased = 0; 708 ufs1_daddr_t *bap1 = NULL; 709 ufs2_daddr_t *bap2 = NULL; 710 #define BAP(ip, i) (I_IS_UFS1(ip) ? bap1[i] : bap2[i]) 711 712 fs = ITOFS(ip); 713 ump = ITOUMP(ip); 714 715 /* 716 * Calculate index in current block of last 717 * block to be kept. -1 indicates the entire 718 * block so we need not calculate the index. 719 */ 720 factor = lbn_offset(fs, level); 721 last = lastbn; 722 if (lastbn > 0) 723 last /= factor; 724 nblocks = btodb(fs->fs_bsize); 725 /* 726 * Get buffer of block pointers, zero those entries corresponding 727 * to blocks to be free'd, and update on disk copy first. Since 728 * double(triple) indirect before single(double) indirect, calls 729 * to VOP_BMAP() on these blocks will fail. However, we already 730 * have the on-disk address, so we just pass it to bread() instead 731 * of having bread() attempt to calculate it using VOP_BMAP(). 732 */ 733 vp = ITOV(ip); 734 error = ffs_breadz(ump, vp, lbn, dbn, (int)fs->fs_bsize, NULL, NULL, 0, 735 NOCRED, 0, NULL, &bp); 736 if (error) { 737 *countp = 0; 738 return (error); 739 } 740 741 if (I_IS_UFS1(ip)) 742 bap1 = (ufs1_daddr_t *)bp->b_data; 743 else 744 bap2 = (ufs2_daddr_t *)bp->b_data; 745 if (lastbn != -1) { 746 copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK); 747 bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize); 748 for (i = last + 1; i < NINDIR(fs); i++) 749 if (I_IS_UFS1(ip)) 750 bap1[i] = 0; 751 else 752 bap2[i] = 0; 753 if (DOINGASYNC(vp)) { 754 bdwrite(bp); 755 } else { 756 error = bwrite(bp); 757 if (error) 758 allerror = error; 759 } 760 if (I_IS_UFS1(ip)) 761 bap1 = (ufs1_daddr_t *)copy; 762 else 763 bap2 = (ufs2_daddr_t *)copy; 764 } 765 766 /* 767 * Recursively free totally unused blocks. 768 */ 769 key = ffs_blkrelease_start(ump, ITODEVVP(ip), ip->i_number); 770 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 771 i--, nlbn += factor) { 772 nb = BAP(ip, i); 773 if (nb == 0) 774 continue; 775 if (level > SINGLE) { 776 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 777 (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0) 778 allerror = error; 779 blocksreleased += blkcount; 780 } 781 ffs_blkfree(ump, fs, ITODEVVP(ip), nb, fs->fs_bsize, 782 ip->i_number, vp->v_type, NULL, key); 783 blocksreleased += nblocks; 784 } 785 ffs_blkrelease_finish(ump, key); 786 787 /* 788 * Recursively free last partial block. 789 */ 790 if (level > SINGLE && lastbn >= 0) { 791 last = lastbn % factor; 792 nb = BAP(ip, i); 793 if (nb != 0) { 794 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 795 last, level - 1, &blkcount); 796 if (error) 797 allerror = error; 798 blocksreleased += blkcount; 799 } 800 } 801 if (copy != NULL) { 802 free(copy, M_TEMP); 803 } else { 804 bp->b_flags |= B_INVAL | B_NOCACHE; 805 brelse(bp); 806 } 807 808 *countp = blocksreleased; 809 return (allerror); 810 } 811 812 int 813 ffs_rdonly(struct inode *ip) 814 { 815 816 return (ITOFS(ip)->fs_ronly != 0); 817 } 818