lfs_inode.c revision 1.51.2.2
1 /* $NetBSD: lfs_inode.c,v 1.51.2.2 2001/06/29 03:56:41 perseant Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1986, 1989, 1991, 1993 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. All advertising materials mentioning features or use of this software 51 * must display the following acknowledgement: 52 * This product includes software developed by the University of 53 * California, Berkeley and its contributors. 54 * 4. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * @(#)lfs_inode.c 8.9 (Berkeley) 5/8/95 71 */ 72 73 #if defined(_KERNEL_OPT) 74 #include "opt_quota.h" 75 #endif 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/mount.h> 80 #include <sys/proc.h> 81 #include <sys/file.h> 82 #include <sys/buf.h> 83 #include <sys/vnode.h> 84 #include <sys/kernel.h> 85 #include <sys/malloc.h> 86 #include <sys/trace.h> 87 #include <sys/resourcevar.h> 88 89 #include <ufs/ufs/quota.h> 90 #include <ufs/ufs/inode.h> 91 #include <ufs/ufs/ufsmount.h> 92 #include <ufs/ufs/ufs_extern.h> 93 94 #include <ufs/lfs/lfs.h> 95 #include <ufs/lfs/lfs_extern.h> 96 97 extern int locked_queue_count; 98 extern long locked_queue_bytes; 99 100 static int lfs_update_seguse(struct lfs *, long, size_t); 101 static int lfs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t, 102 ufs_daddr_t, int, long *, long *, long *, size_t *, 103 struct proc *); 104 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *); 105 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int); 106 107 /* Search a block for a specific dinode. */ 108 struct dinode * 109 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 110 { 111 struct dinode *dip = (struct dinode *)bp->b_data; 112 struct dinode *ldip, *fin; 113 114 if (fs->lfs_version == 1) 115 fin = dip + INOPB(fs); 116 else 117 fin = dip + INOPS(fs); 118 119 /* 120 * XXX we used to go from the top down here, presumably with the 121 * idea that the same inode could be written twice in the same 122 * block (which is not supposed to be true). 123 */ 124 for (ldip = dip; ldip < fin; ++ldip) 125 if (ldip->di_inumber == ino) 126 return (ldip); 127 128 printf("searched %d entries\n", (int)(fs->lfs_version == 1 ? 129 INOPB(fs) : 130 INOPS(fs))); 131 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 132 datosn(fs, fs->lfs_offset)); 133 printf("block is 0x%x (seg %d)\n", bp->b_blkno, 134 datosn(fs, bp->b_blkno)); 135 panic("lfs_ifind: dinode %u not found", ino); 136 /* NOTREACHED */ 137 } 138 139 int 140 lfs_update(void *v) 141 { 142 struct vop_update_args /* { 143 struct vnode *a_vp; 144 struct timespec *a_access; 145 struct timespec *a_modify; 146 int a_flags; 147 } */ *ap = v; 148 struct inode *ip; 149 struct vnode *vp = ap->a_vp; 150 int oflag; 151 struct timespec ts; 152 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 153 154 if (vp->v_mount->mnt_flag & MNT_RDONLY) 155 return (0); 156 ip = VTOI(vp); 157 158 /* 159 * If we are called from vinvalbuf, and the file's blocks have 160 * already been scheduled for writing, but the writes have not 161 * yet completed, lfs_vflush will not be called, and vinvalbuf 162 * will cause a panic. So, we must wait until any pending write 163 * for our inode completes, if we are called with UPDATE_WAIT set. 164 */ 165 while((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 166 WRITEINPROG(vp)) { 167 #ifdef DEBUG_LFS 168 printf("lfs_update: sleeping on inode %d (in-progress)\n", 169 ip->i_number); 170 #endif 171 tsleep(vp, (PRIBIO+1), "lfs_update", 0); 172 } 173 oflag = ip->i_flag; 174 TIMEVAL_TO_TIMESPEC(&time, &ts); 175 LFS_ITIMES(ip, 176 ap->a_access ? ap->a_access : &ts, 177 ap->a_modify ? ap->a_modify : &ts, &ts); 178 if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) { 179 return (0); 180 } 181 182 /* If sync, push back the vnode and any dirty blocks it may have. */ 183 if((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP))==UPDATE_WAIT) { 184 /* Avoid flushing VDIROP. */ 185 ++fs->lfs_diropwait; 186 while(vp->v_flag & VDIROP) { 187 #ifdef DEBUG_LFS 188 printf("lfs_update: sleeping on inode %d (dirops)\n", 189 ip->i_number); 190 printf("lfs_update: vflags 0x%lx, iflags 0x%x\n", 191 vp->v_flag, ip->i_flag); 192 #endif 193 if(fs->lfs_dirops == 0) 194 lfs_flush_fs(fs, SEGM_SYNC); 195 else 196 tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 197 0); 198 /* XXX KS - by falling out here, are we writing the vn 199 twice? */ 200 } 201 --fs->lfs_diropwait; 202 return lfs_vflush(vp); 203 } 204 return 0; 205 } 206 207 #define SINGLE 0 /* index of single indirect block */ 208 #define DOUBLE 1 /* index of double indirect block */ 209 #define TRIPLE 2 /* index of triple indirect block */ 210 /* 211 * Truncate the inode oip to at most length size, freeing the 212 * disk blocks. 213 */ 214 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */ 215 int 216 lfs_truncate(void *v) 217 { 218 struct vop_truncate_args /* { 219 struct vnode *a_vp; 220 off_t a_length; 221 int a_flags; 222 struct ucred *a_cred; 223 struct proc *a_p; 224 } */ *ap = v; 225 struct vnode *ovp = ap->a_vp; 226 ufs_daddr_t lastblock; 227 struct inode *oip; 228 ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 229 ufs_daddr_t newblks[NDADDR + NIADDR]; 230 off_t length = ap->a_length; 231 struct lfs *fs; 232 struct buf *bp; 233 int offset, size, level; 234 long count, rcount, nblocks, blocksreleased = 0, real_released = 0; 235 int i; 236 int aflags, error, allerror = 0; 237 off_t osize; 238 long lastseg; 239 size_t bc; 240 int obufsize, odb; 241 242 if (length < 0) 243 return (EINVAL); 244 oip = VTOI(ovp); 245 246 /* 247 * Just return and not update modification times. 248 */ 249 if (oip->i_ffs_size == length) 250 return (0); 251 252 if (ovp->v_type == VLNK && 253 (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen || 254 (ovp->v_mount->mnt_maxsymlinklen == 0 && 255 oip->i_din.ffs_din.di_blocks == 0))) { 256 #ifdef DIAGNOSTIC 257 if (length != 0) 258 panic("lfs_truncate: partial truncate of symlink"); 259 #endif 260 memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size); 261 oip->i_ffs_size = 0; 262 oip->i_flag |= IN_CHANGE | IN_UPDATE; 263 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 264 } 265 if (oip->i_ffs_size == length) { 266 oip->i_flag |= IN_CHANGE | IN_UPDATE; 267 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 268 } 269 #ifdef QUOTA 270 if ((error = getinoquota(oip)) != 0) 271 return (error); 272 #endif 273 fs = oip->i_lfs; 274 lfs_imtime(fs); 275 osize = oip->i_ffs_size; 276 ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0; 277 278 /* 279 * Lengthen the size of the file. We must ensure that the 280 * last byte of the file is allocated. Since the smallest 281 * value of osize is 0, length will be at least 1. 282 */ 283 if (osize < length) { 284 if (length > fs->lfs_maxfilesize) 285 return (EFBIG); 286 aflags = B_CLRBUF; 287 if (ap->a_flags & IO_SYNC) 288 aflags |= B_SYNC; 289 error = lfs_reserve(fs, ovp, fsbtodb(fs, NIADDR + 2)); 290 if (error) 291 return (error); 292 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 293 lfs_reserve(fs, ovp, -fsbtodb(fs, NIADDR + 2)); 294 if (error) 295 return (error); 296 oip->i_ffs_size = length; 297 uvm_vnp_setsize(ovp, length); 298 (void) VOP_BWRITE(bp); 299 oip->i_flag |= IN_CHANGE | IN_UPDATE; 300 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 301 } 302 303 if ((error = lfs_reserve(fs, ovp, fsbtodb(fs, 2 * NIADDR + 3))) != 0) 304 return (error); 305 /* 306 * Make sure no writes to this inode can happen while we're 307 * truncating. Otherwise, blocks which are accounted for on the 308 * inode *and* which have been created for cleaning can coexist, 309 * and cause an overcounting. 310 * 311 * (We don't need to *hold* the seglock, though, because we already 312 * hold the inode lock; draining the seglock is sufficient.) 313 */ 314 if (ovp != fs->lfs_unlockvp) { 315 while(fs->lfs_seglock) { 316 tsleep(&fs->lfs_seglock, PRIBIO+1, "lfs_truncate", 0); 317 } 318 } 319 320 /* 321 * Shorten the size of the file. If the file is not being 322 * truncated to a block boundary, the contents of the 323 * partial block following the end of the file must be 324 * zero'ed in case it ever becomes accessible again because 325 * of subsequent file growth. Directories however are not 326 * zero'ed as they should grow back initialized to empty. 327 */ 328 offset = blkoff(fs, length); 329 lastseg = -1; 330 bc = 0; 331 if (offset == 0) { 332 oip->i_ffs_size = length; 333 } else { 334 lbn = lblkno(fs, length); 335 aflags = B_CLRBUF; 336 if (ap->a_flags & IO_SYNC) 337 aflags |= B_SYNC; 338 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 339 if (error) { 340 lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3)); 341 return (error); 342 } 343 obufsize = bp->b_bufsize; 344 odb = btodb(bp->b_bcount); 345 oip->i_ffs_size = length; 346 size = blksize(fs, oip, lbn); 347 if (ovp->v_type != VDIR) 348 memset((char *)bp->b_data + offset, 0, 349 (u_int)(size - offset)); 350 allocbuf(bp, size); 351 if (bp->b_flags & B_DELWRI) { 352 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 353 locked_queue_bytes -= obufsize - bp->b_bufsize; 354 fs->lfs_avail += odb - btodb(size); 355 } 356 (void) VOP_BWRITE(bp); 357 } 358 uvm_vnp_setsize(ovp, length); 359 /* 360 * Calculate index into inode's block list of 361 * last direct and indirect blocks (if any) 362 * which we want to keep. Lastblock is -1 when 363 * the file is truncated to 0. 364 */ 365 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 366 lastiblock[SINGLE] = lastblock - NDADDR; 367 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 368 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 369 nblocks = btodb(fs->lfs_bsize); 370 /* 371 * Record changed file and block pointers before we start 372 * freeing blocks. lastiblock values are also normalized to -1 373 * for calls to lfs_indirtrunc below. 374 */ 375 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks); 376 for (level = TRIPLE; level >= SINGLE; level--) 377 if (lastiblock[level] < 0) { 378 newblks[NDADDR+level] = 0; 379 lastiblock[level] = -1; 380 } 381 for (i = NDADDR - 1; i > lastblock; i--) 382 newblks[i] = 0; 383 384 oip->i_ffs_size = osize; 385 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 386 if (error && !allerror) 387 allerror = error; 388 389 /* 390 * Indirect blocks first. 391 */ 392 indir_lbn[SINGLE] = -NDADDR; 393 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 394 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 395 for (level = TRIPLE; level >= SINGLE; level--) { 396 bn = oip->i_ffs_ib[level]; 397 if (bn != 0) { 398 error = lfs_indirtrunc(oip, indir_lbn[level], 399 bn, lastiblock[level], 400 level, &count, &rcount, 401 &lastseg, &bc, ap->a_p); 402 if (error) 403 allerror = error; 404 real_released += rcount; 405 blocksreleased += count; 406 if (lastiblock[level] < 0) { 407 if (oip->i_ffs_ib[level] > 0) 408 real_released += nblocks; 409 blocksreleased += nblocks; 410 oip->i_ffs_ib[level] = 0; 411 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 412 } 413 } 414 if (lastiblock[level] >= 0) 415 goto done; 416 } 417 418 /* 419 * All whole direct blocks or frags. 420 */ 421 for (i = NDADDR - 1; i > lastblock; i--) { 422 long bsize; 423 424 bn = oip->i_ffs_db[i]; 425 if (bn == 0) 426 continue; 427 bsize = blksize(fs, oip, i); 428 if (oip->i_ffs_db[i] > 0) 429 real_released += btodb(bsize); 430 blocksreleased += btodb(bsize); 431 oip->i_ffs_db[i] = 0; 432 lfs_blkfree(fs, bn, bsize, &lastseg, &bc); 433 } 434 if (lastblock < 0) 435 goto done; 436 437 /* 438 * Finally, look for a change in size of the 439 * last direct block; release any frags. 440 */ 441 bn = oip->i_ffs_db[lastblock]; 442 if (bn != 0) { 443 long oldspace, newspace; 444 445 /* 446 * Calculate amount of space we're giving 447 * back as old block size minus new block size. 448 */ 449 oldspace = blksize(fs, oip, lastblock); 450 oip->i_ffs_size = length; 451 newspace = blksize(fs, oip, lastblock); 452 if (newspace == 0) 453 panic("itrunc: newspace"); 454 if (oldspace - newspace > 0) { 455 lfs_blkfree(fs, bn, oldspace - newspace, &lastseg, &bc); 456 if (bn > 0) 457 real_released += btodb(oldspace - newspace); 458 blocksreleased += btodb(oldspace - newspace); 459 } 460 } 461 462 done: 463 /* Finish segment accounting corrections */ 464 lfs_update_seguse(fs, lastseg, bc); 465 #ifdef DIAGNOSTIC 466 for (level = SINGLE; level <= TRIPLE; level++) 467 if (newblks[NDADDR + level] != oip->i_ffs_ib[level]) 468 panic("lfs itrunc1"); 469 for (i = 0; i < NDADDR; i++) 470 if (newblks[i] != oip->i_ffs_db[i]) 471 panic("lfs itrunc2"); 472 if (length == 0 && 473 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 474 panic("lfs itrunc3"); 475 #endif /* DIAGNOSTIC */ 476 /* 477 * Put back the real size. 478 */ 479 oip->i_ffs_size = length; 480 oip->i_lfs_effnblks -= blocksreleased; 481 oip->i_ffs_blocks -= real_released; 482 fs->lfs_bfree += blocksreleased; 483 #ifdef DIAGNOSTIC 484 if (oip->i_ffs_size == 0 && oip->i_ffs_blocks != 0) { 485 printf("lfs_truncate: truncate to 0 but %d blocks on inode\n", 486 oip->i_ffs_blocks); 487 panic("lfs_truncate: persistent blocks\n"); 488 } 489 #endif 490 oip->i_flag |= IN_CHANGE; 491 #ifdef QUOTA 492 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 493 #endif 494 lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3)); 495 return (allerror); 496 } 497 498 /* Update segment usage information when removing a block. */ 499 static int 500 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 501 size_t *num) 502 { 503 long seg; 504 int error = 0; 505 506 bsize = fragroundup(fs, bsize); 507 if (daddr > 0) { 508 if (*lastseg != (seg = datosn(fs, daddr))) { 509 error = lfs_update_seguse(fs, *lastseg, *num); 510 *num = bsize; 511 *lastseg = seg; 512 } else 513 *num += bsize; 514 } 515 return error; 516 } 517 518 /* Finish the accounting updates for a segment. */ 519 static int 520 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 521 { 522 SEGUSE *sup; 523 struct buf *bp; 524 525 if (lastseg < 0 || num == 0) 526 return 0; 527 528 529 LFS_SEGENTRY(sup, fs, lastseg, bp); 530 if (num > sup->su_nbytes) { 531 printf("lfs_truncate: segment %ld short by %ld\n", 532 lastseg, (long)num - sup->su_nbytes); 533 panic("lfs_truncate: negative bytes"); 534 sup->su_nbytes = num; 535 } 536 sup->su_nbytes -= num; 537 return (VOP_BWRITE(bp)); /* Ifile */ 538 } 539 540 /* 541 * Release blocks associated with the inode ip and stored in the indirect 542 * block bn. Blocks are free'd in LIFO order up to (but not including) 543 * lastbn. If level is greater than SINGLE, the block is an indirect block 544 * and recursive calls to indirtrunc must be used to cleanse other indirect 545 * blocks. 546 * 547 * NB: triple indirect blocks are untested. 548 */ 549 static int 550 lfs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, daddr_t dbn, 551 ufs_daddr_t lastbn, int level, long *countp, 552 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 553 { 554 int i; 555 struct buf *bp; 556 struct lfs *fs = ip->i_lfs; 557 ufs_daddr_t *bap; 558 struct vnode *vp; 559 ufs_daddr_t *copy = NULL, nb, nlbn, last; 560 long blkcount, rblkcount, factor; 561 int nblocks, blocksreleased = 0, real_released = 0; 562 int error = 0, allerror = 0; 563 564 /* 565 * Calculate index in current block of last 566 * block to be kept. -1 indicates the entire 567 * block so we need not calculate the index. 568 */ 569 factor = 1; 570 for (i = SINGLE; i < level; i++) 571 factor *= NINDIR(fs); 572 last = lastbn; 573 if (lastbn > 0) 574 last /= factor; 575 nblocks = btodb(fs->lfs_bsize); 576 /* 577 * Get buffer of block pointers, zero those entries corresponding 578 * to blocks to be free'd, and update on disk copy first. Since 579 * double(triple) indirect before single(double) indirect, calls 580 * to bmap on these blocks will fail. However, we already have 581 * the on disk address, so we have to set the b_blkno field 582 * explicitly instead of letting bread do everything for us. 583 */ 584 vp = ITOV(ip); 585 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 586 if (bp->b_flags & (B_DONE | B_DELWRI)) { 587 /* Braces must be here in case trace evaluates to nothing. */ 588 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 589 } else { 590 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 591 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 592 bp->b_flags |= B_READ; 593 if (bp->b_bcount > bp->b_bufsize) 594 panic("lfs_indirtrunc: bad buffer size"); 595 bp->b_blkno = dbn; 596 VOP_STRATEGY(bp); 597 error = biowait(bp); 598 } 599 if (error) { 600 brelse(bp); 601 *countp = *rcountp = 0; 602 return (error); 603 } 604 605 bap = (ufs_daddr_t *)bp->b_data; 606 if (lastbn >= 0) { 607 MALLOC(copy, ufs_daddr_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 608 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 609 memset((caddr_t)&bap[last + 1], 0, 610 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t)); 611 error = VOP_BWRITE(bp); 612 if (error) 613 allerror = error; 614 bap = copy; 615 } 616 617 /* 618 * Recursively free totally unused blocks. 619 */ 620 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 621 i--, nlbn += factor) { 622 nb = bap[i]; 623 if (nb == 0) 624 continue; 625 if (level > SINGLE) { 626 error = lfs_indirtrunc(ip, nlbn, nb, 627 (ufs_daddr_t)-1, level - 1, 628 &blkcount, &rblkcount, 629 lastsegp, bcp, p); 630 if (error) 631 allerror = error; 632 blocksreleased += blkcount; 633 real_released += rblkcount; 634 } 635 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 636 if (bap[i] > 0) 637 real_released += nblocks; 638 blocksreleased += nblocks; 639 } 640 641 /* 642 * Recursively free last partial block. 643 */ 644 if (level > SINGLE && lastbn >= 0) { 645 last = lastbn % factor; 646 nb = bap[i]; 647 if (nb != 0) { 648 error = lfs_indirtrunc(ip, nlbn, nb, 649 last, level - 1, &blkcount, 650 &rblkcount, lastsegp, bcp, p); 651 if (error) 652 allerror = error; 653 real_released += rblkcount; 654 blocksreleased += blkcount; 655 } 656 } 657 658 if (copy != NULL) { 659 FREE(copy, M_TEMP); 660 } else { 661 if (bp->b_flags & B_DELWRI) { 662 LFS_UNLOCK_BUF(bp); 663 fs->lfs_avail += btodb(bp->b_bcount); 664 wakeup(&fs->lfs_avail); 665 } 666 bp->b_flags |= B_INVAL; 667 brelse(bp); 668 } 669 670 *countp = blocksreleased; 671 *rcountp = real_released; 672 return (allerror); 673 } 674 675 /* 676 * Destroy any in core blocks past the truncation length. 677 * Inlined from vtruncbuf, so that lfs_avail could be updated. 678 */ 679 static int 680 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 681 { 682 struct buf *bp, *nbp; 683 int s, error; 684 struct lfs *fs; 685 686 fs = VTOI(vp)->i_lfs; 687 s = splbio(); 688 689 restart: 690 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 691 nbp = LIST_NEXT(bp, b_vnbufs); 692 if (bp->b_lblkno < lbn) 693 continue; 694 if (bp->b_flags & B_BUSY) { 695 bp->b_flags |= B_WANTED; 696 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 697 "lfs_vtruncbuf", slptimeo); 698 if (error) { 699 splx(s); 700 return (error); 701 } 702 goto restart; 703 } 704 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 705 if (bp->b_flags & B_DELWRI) { 706 bp->b_flags &= ~B_DELWRI; 707 fs->lfs_avail += btodb(bp->b_bcount); 708 wakeup(&fs->lfs_avail); 709 } 710 LFS_UNLOCK_BUF(bp); 711 brelse(bp); 712 } 713 714 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 715 nbp = LIST_NEXT(bp, b_vnbufs); 716 if (bp->b_lblkno < lbn) 717 continue; 718 if (bp->b_flags & B_BUSY) { 719 bp->b_flags |= B_WANTED; 720 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 721 "lfs_vtruncbuf", slptimeo); 722 if (error) { 723 splx(s); 724 return (error); 725 } 726 goto restart; 727 } 728 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 729 if (bp->b_flags & B_DELWRI) { 730 bp->b_flags &= ~B_DELWRI; 731 fs->lfs_avail += btodb(bp->b_bcount); 732 wakeup(&fs->lfs_avail); 733 } 734 LFS_UNLOCK_BUF(bp); 735 brelse(bp); 736 } 737 738 splx(s); 739 740 return (0); 741 } 742 743
Indexes created Thu Oct 02 14:10:14 GMT 2025