1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95 30 * $FreeBSD: src/sys/ufs/ffs/ffs_inode.c,v 1.56.2.5 2002/02/05 18:35:03 dillon Exp $ 31 */ 32 33 #include "opt_quota.h" 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/mount.h> 38 #include <sys/proc.h> 39 #include <sys/buf.h> 40 #include <sys/vnode.h> 41 #include <sys/kernel.h> 42 #include <sys/malloc.h> 43 #include <sys/resourcevar.h> 44 #include <sys/vmmeter.h> 45 46 #include <vm/vm.h> 47 #include <vm/vm_extern.h> 48 49 #include "quota.h" 50 #include "ufsmount.h" 51 #include "inode.h" 52 #include "ufs_extern.h" 53 54 #include "fs.h" 55 #include "ffs_extern.h" 56 57 #include <vm/vm_page2.h> 58 #include <sys/buf2.h> 59 60 static int ffs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t, 61 ufs_daddr_t, int, long *); 62 63 /* 64 * Update the access, modified, and inode change times as specified by the 65 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode 66 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by 67 * the timestamp update). The IN_LAZYMOD flag is set to force a write 68 * later if not now. If we write now, then clear both IN_MODIFIED and 69 * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is 70 * set, then wait for the write to complete. 71 */ 72 int 73 ffs_update(struct vnode *vp, int waitfor) 74 { 75 struct fs *fs; 76 struct buf *bp; 77 struct inode *ip; 78 int error; 79 80 ufs_itimes(vp); 81 ip = VTOI(vp); 82 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0) 83 return (0); 84 ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED); 85 fs = ip->i_fs; 86 if (fs->fs_ronly) 87 return (0); 88 89 /* 90 * The vnode type is usually set to VBAD if an unrecoverable I/O 91 * error has occured (such as when reading the inode). Clear the 92 * modified bits but do not write anything out in this case. 93 */ 94 if (vp->v_type == VBAD) 95 return (0); 96 /* 97 * Ensure that uid and gid are correct. This is a temporary 98 * fix until fsck has been changed to do the update. 99 */ 100 if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */ 101 ip->i_din.di_ouid = ip->i_uid; /* XXX */ 102 ip->i_din.di_ogid = ip->i_gid; /* XXX */ 103 } /* XXX */ 104 error = bread(ip->i_devvp, 105 fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)), 106 (int)fs->fs_bsize, &bp); 107 if (error) { 108 brelse(bp); 109 return (error); 110 } 111 if (DOINGSOFTDEP(vp)) 112 softdep_update_inodeblock(ip, bp, waitfor); 113 else if (ip->i_effnlink != ip->i_nlink) 114 panic("ffs_update: bad link cnt"); 115 *((struct ufs1_dinode *)bp->b_data + 116 ino_to_fsbo(fs, ip->i_number)) = ip->i_din; 117 if (waitfor && !DOINGASYNC(vp)) { 118 return (bwrite(bp)); 119 } else if (vm_paging_severe() || buf_dirty_count_severe()) { 120 return (bwrite(bp)); 121 } else { 122 if (bp->b_bufsize == fs->fs_bsize) 123 bp->b_flags |= B_CLUSTEROK; 124 bdwrite(bp); 125 return (0); 126 } 127 } 128 129 #define SINGLE 0 /* index of single indirect block */ 130 #define DOUBLE 1 /* index of double indirect block */ 131 #define TRIPLE 2 /* index of triple indirect block */ 132 /* 133 * Truncate the inode oip to at most length size, freeing the 134 * disk blocks. 135 */ 136 int 137 ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred) 138 { 139 struct vnode *ovp = vp; 140 ufs_daddr_t lastblock; 141 struct inode *oip; 142 ufs_daddr_t bn, lbn, lastiblock[UFS_NIADDR], indir_lbn[UFS_NIADDR]; 143 ufs_daddr_t oldblks[UFS_NDADDR + UFS_NIADDR]; 144 ufs_daddr_t newblks[UFS_NDADDR + UFS_NIADDR]; 145 struct fs *fs; 146 struct buf *bp; 147 int offset, size, level; 148 long count, nblocks, blocksreleased = 0; 149 int i; 150 int aflags, error, allerror; 151 off_t osize; 152 153 oip = VTOI(ovp); 154 fs = oip->i_fs; 155 if (length < 0) 156 return (EINVAL); 157 if (length > fs->fs_maxfilesize) 158 return (EFBIG); 159 if (ovp->v_type == VLNK && 160 (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) { 161 #ifdef DIAGNOSTIC 162 if (length != 0) 163 panic("ffs_truncate: partial truncate of symlink"); 164 #endif /* DIAGNOSTIC */ 165 bzero((char *)&oip->i_shortlink, (uint)oip->i_size); 166 oip->i_size = 0; 167 oip->i_flag |= IN_CHANGE | IN_UPDATE; 168 return (ffs_update(ovp, 1)); 169 } 170 if (oip->i_size == length) { 171 oip->i_flag |= IN_CHANGE | IN_UPDATE; 172 return (ffs_update(ovp, 0)); 173 } 174 if (fs->fs_ronly) 175 panic("ffs_truncate: read-only filesystem"); 176 #ifdef QUOTA 177 error = ufs_getinoquota(oip); 178 if (error) 179 return (error); 180 #endif 181 if (DOINGSOFTDEP(ovp)) { 182 if (length > 0 || softdep_slowdown(ovp)) { 183 /* 184 * If a file is only partially truncated, then 185 * we have to clean up the data structures 186 * describing the allocation past the truncation 187 * point. Finding and deallocating those structures 188 * is a lot of work. Since partial truncation occurs 189 * rarely, we solve the problem by syncing the file 190 * so that it will have no data structures left. 191 */ 192 if ((error = VOP_FSYNC(ovp, MNT_WAIT, 0)) != 0) 193 return (error); 194 } else { 195 #ifdef QUOTA 196 (void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0); 197 #endif 198 softdep_setup_freeblocks(oip, length); 199 vinvalbuf(ovp, 0, 0, 0); 200 nvnode_pager_setsize(ovp, 0, fs->fs_bsize, 0); 201 oip->i_flag |= IN_CHANGE | IN_UPDATE; 202 return (ffs_update(ovp, 0)); 203 } 204 } 205 osize = oip->i_size; 206 207 /* 208 * Lengthen the size of the file. We must ensure that the 209 * last byte of the file is allocated. Since the smallest 210 * value of osize is 0, length will be at least 1. 211 * 212 * nvextendbuf() only breads the old buffer. The blocksize 213 * of the new buffer must be specified so it knows how large 214 * to make the VM object. 215 */ 216 if (osize < length) { 217 nvextendbuf(vp, osize, length, 218 blkoffsize(fs, oip, osize), /* oblksize */ 219 blkoffresize(fs, length), /* nblksize */ 220 blkoff(fs, osize), 221 blkoff(fs, length), 222 0); 223 224 aflags = B_CLRBUF; 225 if (flags & IO_SYNC) 226 aflags |= B_SYNC; 227 /* BALLOC will reallocate the fragment at the old EOF */ 228 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp); 229 if (error) 230 return (error); 231 oip->i_size = length; 232 if (bp->b_bufsize == fs->fs_bsize) 233 bp->b_flags |= B_CLUSTEROK; 234 if (aflags & B_SYNC) 235 bwrite(bp); 236 else 237 bawrite(bp); 238 oip->i_flag |= IN_CHANGE | IN_UPDATE; 239 return (ffs_update(ovp, 1)); 240 } 241 242 /* 243 * Shorten the size of the file. 244 * 245 * NOTE: The block size specified in nvtruncbuf() is the blocksize 246 * of the buffer containing length prior to any reallocation 247 * of the block. 248 */ 249 allerror = nvtruncbuf(ovp, length, blkoffsize(fs, oip, length), 250 blkoff(fs, length), 0); 251 offset = blkoff(fs, length); 252 if (offset == 0) { 253 oip->i_size = length; 254 } else { 255 lbn = lblkno(fs, length); 256 aflags = B_CLRBUF; 257 if (flags & IO_SYNC) 258 aflags |= B_SYNC; 259 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp); 260 if (error) 261 return (error); 262 263 /* 264 * When we are doing soft updates and the UFS_BALLOC 265 * above fills in a direct block hole with a full sized 266 * block that will be truncated down to a fragment below, 267 * we must flush out the block dependency with an FSYNC 268 * so that we do not get a soft updates inconsistency 269 * when we create the fragment below. 270 * 271 * nvtruncbuf() may have re-dirtied the underlying block 272 * as part of its truncation zeroing code. To avoid a 273 * 'locking against myself' panic in the second fsync we 274 * can simply undirty the bp since the redirtying was 275 * related to areas of the buffer that we are going to 276 * throw away anyway, and we will b*write() the remainder 277 * anyway down below. 278 */ 279 if (DOINGSOFTDEP(ovp) && lbn < UFS_NDADDR && 280 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize) { 281 bundirty(bp); 282 error = VOP_FSYNC(ovp, MNT_WAIT, 0); 283 if (error) { 284 bdwrite(bp); 285 return (error); 286 } 287 } 288 oip->i_size = length; 289 size = blksize(fs, oip, lbn); 290 #if 0 291 /* remove - nvtruncbuf deals with this */ 292 if (ovp->v_type != VDIR) 293 bzero((char *)bp->b_data + offset, 294 (uint)(size - offset)); 295 #endif 296 /* Kirk's code has reallocbuf(bp, size, 1) here */ 297 allocbuf(bp, size); 298 if (bp->b_bufsize == fs->fs_bsize) 299 bp->b_flags |= B_CLUSTEROK; 300 if (aflags & B_SYNC) 301 bwrite(bp); 302 else 303 bawrite(bp); 304 } 305 /* 306 * Calculate index into inode's block list of 307 * last direct and indirect blocks (if any) 308 * which we want to keep. Lastblock is -1 when 309 * the file is truncated to 0. 310 */ 311 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 312 lastiblock[SINGLE] = lastblock - UFS_NDADDR; 313 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 314 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 315 nblocks = btodb(fs->fs_bsize); 316 317 /* 318 * Update file and block pointers on disk before we start freeing 319 * blocks. If we crash before free'ing blocks below, the blocks 320 * will be returned to the free list. lastiblock values are also 321 * normalized to -1 for calls to ffs_indirtrunc below. 322 */ 323 for (level = TRIPLE; level >= SINGLE; level--) { 324 oldblks[UFS_NDADDR + level] = oip->i_ib[level]; 325 if (lastiblock[level] < 0) { 326 oip->i_ib[level] = 0; 327 lastiblock[level] = -1; 328 } 329 } 330 for (i = 0; i < UFS_NDADDR; i++) { 331 oldblks[i] = oip->i_db[i]; 332 if (i > lastblock) 333 oip->i_db[i] = 0; 334 } 335 oip->i_flag |= IN_CHANGE | IN_UPDATE; 336 error = ffs_update(ovp, 1); 337 if (error && allerror == 0) 338 allerror = error; 339 340 /* 341 * Having written the new inode to disk, save its new configuration 342 * and put back the old block pointers long enough to process them. 343 * Note that we save the new block configuration so we can check it 344 * when we are done. 345 */ 346 for (i = 0; i < UFS_NDADDR; i++) { 347 newblks[i] = oip->i_db[i]; 348 oip->i_db[i] = oldblks[i]; 349 } 350 for (i = 0; i < UFS_NIADDR; i++) { 351 newblks[UFS_NDADDR + i] = oip->i_ib[i]; 352 oip->i_ib[i] = oldblks[UFS_NDADDR + i]; 353 } 354 oip->i_size = osize; 355 356 if (error && allerror == 0) 357 allerror = error; 358 359 /* 360 * Indirect blocks first. 361 */ 362 indir_lbn[SINGLE] = -UFS_NDADDR; 363 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 364 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 365 for (level = TRIPLE; level >= SINGLE; level--) { 366 bn = oip->i_ib[level]; 367 if (bn != 0) { 368 error = ffs_indirtrunc(oip, indir_lbn[level], 369 fsbtodb(fs, bn), lastiblock[level], level, &count); 370 if (error) 371 allerror = error; 372 blocksreleased += count; 373 if (lastiblock[level] < 0) { 374 oip->i_ib[level] = 0; 375 ffs_blkfree(oip, bn, fs->fs_bsize); 376 blocksreleased += nblocks; 377 } 378 } 379 if (lastiblock[level] >= 0) 380 goto done; 381 } 382 383 /* 384 * All whole direct blocks or frags. 385 */ 386 for (i = UFS_NDADDR - 1; i > lastblock; i--) { 387 long bsize; 388 389 bn = oip->i_db[i]; 390 if (bn == 0) 391 continue; 392 oip->i_db[i] = 0; 393 bsize = blksize(fs, oip, i); 394 ffs_blkfree(oip, bn, bsize); 395 blocksreleased += btodb(bsize); 396 } 397 if (lastblock < 0) 398 goto done; 399 400 /* 401 * Finally, look for a change in size of the 402 * last direct block; release any frags. 403 */ 404 bn = oip->i_db[lastblock]; 405 if (bn != 0) { 406 long oldspace, newspace; 407 408 /* 409 * Calculate amount of space we're giving 410 * back as old block size minus new block size. 411 */ 412 oldspace = blksize(fs, oip, lastblock); 413 oip->i_size = length; 414 newspace = blksize(fs, oip, lastblock); 415 if (newspace == 0) 416 panic("ffs_truncate: newspace"); 417 if (oldspace - newspace > 0) { 418 /* 419 * Block number of space to be free'd is 420 * the old block # plus the number of frags 421 * required for the storage we're keeping. 422 */ 423 bn += numfrags(fs, newspace); 424 ffs_blkfree(oip, bn, oldspace - newspace); 425 blocksreleased += btodb(oldspace - newspace); 426 } 427 } 428 done: 429 #ifdef DIAGNOSTIC 430 for (level = SINGLE; level <= TRIPLE; level++) 431 if (newblks[UFS_NDADDR + level] != oip->i_ib[level]) 432 panic("ffs_truncate1"); 433 for (i = 0; i < UFS_NDADDR; i++) 434 if (newblks[i] != oip->i_db[i]) 435 panic("ffs_truncate2"); 436 if (length == 0 && !RB_EMPTY(&ovp->v_rbdirty_tree)) 437 panic("ffs_truncate3"); 438 #endif /* DIAGNOSTIC */ 439 /* 440 * Put back the real size. 441 */ 442 oip->i_size = length; 443 oip->i_blocks -= blocksreleased; 444 445 if (oip->i_blocks < 0) /* sanity */ 446 oip->i_blocks = 0; 447 oip->i_flag |= IN_CHANGE; 448 #ifdef QUOTA 449 (void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0); 450 #endif 451 return (allerror); 452 } 453 454 /* 455 * Release blocks associated with the inode ip and stored in the indirect 456 * block bn. Blocks are free'd in LIFO order up to (but not including) 457 * lastbn. If level is greater than SINGLE, the block is an indirect block 458 * and recursive calls to indirtrunc must be used to cleanse other indirect 459 * blocks. 460 * 461 * NB: triple indirect blocks are untested. 462 */ 463 static int 464 ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn, 465 ufs_daddr_t lastbn, int level, long *countp) 466 { 467 int i; 468 struct buf *bp; 469 struct fs *fs = ip->i_fs; 470 ufs_daddr_t *bap; 471 struct vnode *vp; 472 ufs_daddr_t *copy = NULL, nb, nlbn, last; 473 long blkcount, factor; 474 int nblocks, blocksreleased = 0; 475 int error = 0, allerror = 0; 476 477 /* 478 * Calculate index in current block of last 479 * block to be kept. -1 indicates the entire 480 * block so we need not calculate the index. 481 */ 482 factor = 1; 483 for (i = SINGLE; i < level; i++) 484 factor *= NINDIR(fs); 485 last = lastbn; 486 if (lastbn > 0) 487 last /= factor; 488 nblocks = btodb(fs->fs_bsize); 489 /* 490 * Get buffer of block pointers, zero those entries corresponding 491 * to blocks to be free'd, and update on disk copy first. Since 492 * double(triple) indirect before single(double) indirect, calls 493 * to bmap on these blocks will fail. However, we already have 494 * the on disk address, so we have to set the bio_offset field 495 * explicitly instead of letting bread do everything for us. 496 */ 497 vp = ITOV(ip); 498 bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, 0, 0); 499 if ((bp->b_flags & B_CACHE) == 0) { 500 bp->b_flags &= ~(B_ERROR|B_INVAL); 501 bp->b_cmd = BUF_CMD_READ; 502 if (bp->b_bcount > bp->b_bufsize) 503 panic("ffs_indirtrunc: bad buffer size"); 504 /* 505 * BIO is bio2 which chains back to bio1. We wait 506 * on bio1. 507 */ 508 bp->b_bio2.bio_offset = dbtodoff(fs, dbn); 509 bp->b_bio1.bio_done = biodone_sync; 510 bp->b_bio1.bio_flags |= BIO_SYNC; 511 vfs_busy_pages(vp, bp); 512 /* 513 * Access the block device layer using the device vnode 514 * and the translated block number (bio2) instead of the 515 * file vnode (vp) and logical block number (bio1). 516 * 517 * Even though we are bypassing the vnode layer, we still 518 * want the vnode state to indicate that an I/O on its behalf 519 * is in progress. 520 */ 521 bio_start_transaction(&bp->b_bio1, &vp->v_track_read); 522 vn_strategy(ip->i_devvp, &bp->b_bio2); 523 error = biowait(&bp->b_bio1, "biord"); 524 } 525 if (error) { 526 brelse(bp); 527 *countp = 0; 528 return (error); 529 } 530 531 bap = (ufs_daddr_t *)bp->b_data; 532 if (lastbn != -1) { 533 copy = kmalloc(fs->fs_bsize, M_TEMP, M_WAITOK); 534 bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize); 535 bzero((caddr_t)&bap[last + 1], 536 (uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t)); 537 if (DOINGASYNC(vp)) { 538 bawrite(bp); 539 } else { 540 error = bwrite(bp); 541 if (error) 542 allerror = error; 543 } 544 bap = copy; 545 } 546 547 /* 548 * Recursively free totally unused blocks. 549 */ 550 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 551 i--, nlbn += factor) { 552 nb = bap[i]; 553 if (nb == 0) 554 continue; 555 if (level > SINGLE) { 556 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 557 (ufs_daddr_t)-1, level - 1, &blkcount)) != 0) 558 allerror = error; 559 blocksreleased += blkcount; 560 } 561 ffs_blkfree(ip, nb, fs->fs_bsize); 562 blocksreleased += nblocks; 563 } 564 565 /* 566 * Recursively free last partial block. 567 */ 568 if (level > SINGLE && lastbn >= 0) { 569 last = lastbn % factor; 570 nb = bap[i]; 571 if (nb != 0) { 572 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 573 last, level - 1, &blkcount); 574 if (error) 575 allerror = error; 576 blocksreleased += blkcount; 577 } 578 } 579 if (copy != NULL) { 580 kfree(copy, M_TEMP); 581 } else { 582 bp->b_flags |= B_INVAL | B_NOCACHE; 583 brelse(bp); 584 } 585 586 *countp = blocksreleased; 587 return (allerror); 588 } 589