lfs_inode.c revision 1.54
1 /* $NetBSD: lfs_inode.c,v 1.54 2001/11/06 07:11:29 simonb 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 #ifdef LFS_IFILE_FRAG_ADDRESSING 115 if (fs->lfs_version == 1) 116 fin = dip + INOPB(fs); 117 else 118 fin = dip + INOPF(fs); 119 #else 120 fin = dip + INOPB(fs); 121 #endif 122 123 /* 124 * XXX we used to go from the top down here, presumably with the 125 * idea that the same inode could be written twice in the same 126 * block (which is not supposed to be true). 127 */ 128 for (ldip = dip; ldip < fin; ++ldip) 129 if (ldip->di_inumber == ino) 130 return (ldip); 131 132 printf("searched %d entries\n", (int)(fin - dip)); 133 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 134 dtosn(fs, fs->lfs_offset)); 135 printf("block is 0x%x (seg %d)\n", dbtofsb(fs, bp->b_blkno), 136 dtosn(fs, dbtofsb(fs, bp->b_blkno))); 137 panic("lfs_ifind: dinode %u not found", ino); 138 /* NOTREACHED */ 139 } 140 141 int 142 lfs_update(void *v) 143 { 144 struct vop_update_args /* { 145 struct vnode *a_vp; 146 struct timespec *a_access; 147 struct timespec *a_modify; 148 int a_flags; 149 } */ *ap = v; 150 struct inode *ip; 151 struct vnode *vp = ap->a_vp; 152 struct timespec ts; 153 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 154 155 if (vp->v_mount->mnt_flag & MNT_RDONLY) 156 return (0); 157 ip = VTOI(vp); 158 159 /* 160 * If we are called from vinvalbuf, and the file's blocks have 161 * already been scheduled for writing, but the writes have not 162 * yet completed, lfs_vflush will not be called, and vinvalbuf 163 * will cause a panic. So, we must wait until any pending write 164 * for our inode completes, if we are called with UPDATE_WAIT set. 165 */ 166 while((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 167 WRITEINPROG(vp)) { 168 #ifdef DEBUG_LFS 169 printf("lfs_update: sleeping on inode %d (in-progress)\n", 170 ip->i_number); 171 #endif 172 tsleep(vp, (PRIBIO+1), "lfs_update", 0); 173 } 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; 276 277 /* 278 * Lengthen the size of the file. We must ensure that the 279 * last byte of the file is allocated. Since the smallest 280 * value of osize is 0, length will be at least 1. 281 */ 282 if (osize < length) { 283 if (length > fs->lfs_maxfilesize) 284 return (EFBIG); 285 aflags = B_CLRBUF; 286 if (ap->a_flags & IO_SYNC) 287 aflags |= B_SYNC; 288 error = lfs_reserve(fs, ovp, btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 289 if (error) 290 return (error); 291 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 292 lfs_reserve(fs, ovp, -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 293 if (error) 294 return (error); 295 oip->i_ffs_size = length; 296 uvm_vnp_setsize(ovp, length); 297 (void) VOP_BWRITE(bp); 298 oip->i_flag |= IN_CHANGE | IN_UPDATE; 299 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 300 } 301 302 if ((error = lfs_reserve(fs, ovp, btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 303 return (error); 304 /* 305 * Make sure no writes to this inode can happen while we're 306 * truncating. Otherwise, blocks which are accounted for on the 307 * inode *and* which have been created for cleaning can coexist, 308 * and cause an overcounting. 309 * 310 * (We don't need to *hold* the seglock, though, because we already 311 * hold the inode lock; draining the seglock is sufficient.) 312 */ 313 if (ovp != fs->lfs_unlockvp) { 314 while(fs->lfs_seglock) { 315 tsleep(&fs->lfs_seglock, PRIBIO+1, "lfs_truncate", 0); 316 } 317 } 318 319 /* 320 * Shorten the size of the file. If the file is not being 321 * truncated to a block boundary, the contents of the 322 * partial block following the end of the file must be 323 * zero'ed in case it ever becomes accessible again because 324 * of subsequent file growth. Directories however are not 325 * zero'ed as they should grow back initialized to empty. 326 */ 327 offset = blkoff(fs, length); 328 lastseg = -1; 329 bc = 0; 330 if (offset == 0) { 331 oip->i_ffs_size = length; 332 } else { 333 lbn = lblkno(fs, length); 334 aflags = B_CLRBUF; 335 if (ap->a_flags & IO_SYNC) 336 aflags |= B_SYNC; 337 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 338 if (error) { 339 lfs_reserve(fs, ovp, -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 340 return (error); 341 } 342 obufsize = bp->b_bufsize; 343 odb = btofsb(fs, bp->b_bcount); 344 oip->i_ffs_size = length; 345 size = blksize(fs, oip, lbn); 346 if (ovp->v_type != VDIR) 347 memset((char *)bp->b_data + offset, 0, 348 (u_int)(size - offset)); 349 allocbuf(bp, size); 350 if (bp->b_flags & B_DELWRI) { 351 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 352 locked_queue_bytes -= obufsize - bp->b_bufsize; 353 fs->lfs_avail += odb - btofsb(fs, size); 354 } 355 (void) VOP_BWRITE(bp); 356 } 357 uvm_vnp_setsize(ovp, length); 358 /* 359 * Calculate index into inode's block list of 360 * last direct and indirect blocks (if any) 361 * which we want to keep. Lastblock is -1 when 362 * the file is truncated to 0. 363 */ 364 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 365 lastiblock[SINGLE] = lastblock - NDADDR; 366 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 367 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 368 nblocks = btofsb(fs, fs->lfs_bsize); 369 /* 370 * Record changed file and block pointers before we start 371 * freeing blocks. lastiblock values are also normalized to -1 372 * for calls to lfs_indirtrunc below. 373 */ 374 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks); 375 for (level = TRIPLE; level >= SINGLE; level--) 376 if (lastiblock[level] < 0) { 377 newblks[NDADDR+level] = 0; 378 lastiblock[level] = -1; 379 } 380 for (i = NDADDR - 1; i > lastblock; i--) 381 newblks[i] = 0; 382 383 oip->i_ffs_size = osize; 384 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 385 if (error && !allerror) 386 allerror = error; 387 388 /* 389 * Indirect blocks first. 390 */ 391 indir_lbn[SINGLE] = -NDADDR; 392 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 393 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 394 for (level = TRIPLE; level >= SINGLE; level--) { 395 bn = oip->i_ffs_ib[level]; 396 if (bn != 0) { 397 error = lfs_indirtrunc(oip, indir_lbn[level], 398 bn, lastiblock[level], 399 level, &count, &rcount, 400 &lastseg, &bc, ap->a_p); 401 if (error) 402 allerror = error; 403 real_released += rcount; 404 blocksreleased += count; 405 if (lastiblock[level] < 0) { 406 if (oip->i_ffs_ib[level] > 0) 407 real_released += nblocks; 408 blocksreleased += nblocks; 409 oip->i_ffs_ib[level] = 0; 410 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 411 } 412 } 413 if (lastiblock[level] >= 0) 414 goto done; 415 } 416 417 /* 418 * All whole direct blocks or frags. 419 */ 420 for (i = NDADDR - 1; i > lastblock; i--) { 421 long bsize; 422 423 bn = oip->i_ffs_db[i]; 424 if (bn == 0) 425 continue; 426 bsize = blksize(fs, oip, i); 427 if (oip->i_ffs_db[i] > 0) 428 real_released += btofsb(fs, bsize); 429 blocksreleased += btofsb(fs, bsize); 430 oip->i_ffs_db[i] = 0; 431 lfs_blkfree(fs, bn, bsize, &lastseg, &bc); 432 } 433 if (lastblock < 0) 434 goto done; 435 436 /* 437 * Finally, look for a change in size of the 438 * last direct block; release any frags. 439 */ 440 bn = oip->i_ffs_db[lastblock]; 441 if (bn != 0) { 442 long oldspace, newspace; 443 444 /* 445 * Calculate amount of space we're giving 446 * back as old block size minus new block size. 447 */ 448 oldspace = blksize(fs, oip, lastblock); 449 oip->i_ffs_size = length; 450 newspace = blksize(fs, oip, lastblock); 451 if (newspace == 0) 452 panic("itrunc: newspace"); 453 if (oldspace - newspace > 0) { 454 lfs_blkfree(fs, bn, oldspace - newspace, &lastseg, &bc); 455 if (bn > 0) 456 real_released += btofsb(fs, oldspace - newspace); 457 blocksreleased += btofsb(fs, oldspace - newspace); 458 } 459 } 460 461 done: 462 /* Finish segment accounting corrections */ 463 lfs_update_seguse(fs, lastseg, bc); 464 #ifdef DIAGNOSTIC 465 for (level = SINGLE; level <= TRIPLE; level++) 466 if (newblks[NDADDR + level] != oip->i_ffs_ib[level]) 467 panic("lfs itrunc1"); 468 for (i = 0; i < NDADDR; i++) 469 if (newblks[i] != oip->i_ffs_db[i]) 470 panic("lfs itrunc2"); 471 if (length == 0 && 472 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 473 panic("lfs itrunc3"); 474 #endif /* DIAGNOSTIC */ 475 /* 476 * Put back the real size. 477 */ 478 oip->i_ffs_size = length; 479 oip->i_lfs_effnblks -= blocksreleased; 480 oip->i_ffs_blocks -= real_released; 481 fs->lfs_bfree += blocksreleased; 482 #ifdef DIAGNOSTIC 483 if (oip->i_ffs_size == 0 && oip->i_ffs_blocks != 0) { 484 printf("lfs_truncate: truncate to 0 but %d blocks on inode\n", 485 oip->i_ffs_blocks); 486 panic("lfs_truncate: persistent blocks\n"); 487 } 488 #endif 489 oip->i_flag |= IN_CHANGE; 490 #ifdef QUOTA 491 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 492 #endif 493 lfs_reserve(fs, ovp, -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 494 return (allerror); 495 } 496 497 /* Update segment usage information when removing a block. */ 498 static int 499 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 500 size_t *num) 501 { 502 long seg; 503 int error = 0; 504 505 bsize = fragroundup(fs, bsize); 506 if (daddr > 0) { 507 if (*lastseg != (seg = dtosn(fs, daddr))) { 508 error = lfs_update_seguse(fs, *lastseg, *num); 509 *num = bsize; 510 *lastseg = seg; 511 } else 512 *num += bsize; 513 } 514 return error; 515 } 516 517 /* Finish the accounting updates for a segment. */ 518 static int 519 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 520 { 521 SEGUSE *sup; 522 struct buf *bp; 523 524 if (lastseg < 0 || num == 0) 525 return 0; 526 527 528 LFS_SEGENTRY(sup, fs, lastseg, bp); 529 if (num > sup->su_nbytes) { 530 printf("lfs_truncate: segment %ld short by %ld\n", 531 lastseg, (long)num - sup->su_nbytes); 532 panic("lfs_truncate: negative bytes"); 533 sup->su_nbytes = num; 534 } 535 sup->su_nbytes -= num; 536 return (VOP_BWRITE(bp)); /* Ifile */ 537 } 538 539 /* 540 * Release blocks associated with the inode ip and stored in the indirect 541 * block bn. Blocks are free'd in LIFO order up to (but not including) 542 * lastbn. If level is greater than SINGLE, the block is an indirect block 543 * and recursive calls to indirtrunc must be used to cleanse other indirect 544 * blocks. 545 * 546 * NB: triple indirect blocks are untested. 547 */ 548 static int 549 lfs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, daddr_t dbn, 550 ufs_daddr_t lastbn, int level, long *countp, 551 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 552 { 553 int i; 554 struct buf *bp; 555 struct lfs *fs = ip->i_lfs; 556 ufs_daddr_t *bap; 557 struct vnode *vp; 558 ufs_daddr_t *copy = NULL, nb, nlbn, last; 559 long blkcount, rblkcount, factor; 560 int nblocks, blocksreleased = 0, real_released = 0; 561 int error = 0, allerror = 0; 562 563 /* 564 * Calculate index in current block of last 565 * block to be kept. -1 indicates the entire 566 * block so we need not calculate the index. 567 */ 568 factor = 1; 569 for (i = SINGLE; i < level; i++) 570 factor *= NINDIR(fs); 571 last = lastbn; 572 if (lastbn > 0) 573 last /= factor; 574 nblocks = btofsb(fs, fs->lfs_bsize); 575 /* 576 * Get buffer of block pointers, zero those entries corresponding 577 * to blocks to be free'd, and update on disk copy first. Since 578 * double(triple) indirect before single(double) indirect, calls 579 * to bmap on these blocks will fail. However, we already have 580 * the on disk address, so we have to set the b_blkno field 581 * explicitly instead of letting bread do everything for us. 582 */ 583 vp = ITOV(ip); 584 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 585 if (bp->b_flags & (B_DONE | B_DELWRI)) { 586 /* Braces must be here in case trace evaluates to nothing. */ 587 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 588 } else { 589 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 590 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 591 bp->b_flags |= B_READ; 592 if (bp->b_bcount > bp->b_bufsize) 593 panic("lfs_indirtrunc: bad buffer size"); 594 bp->b_blkno = fsbtodb(fs, dbn); 595 VOP_STRATEGY(bp); 596 error = biowait(bp); 597 } 598 if (error) { 599 brelse(bp); 600 *countp = *rcountp = 0; 601 return (error); 602 } 603 604 bap = (ufs_daddr_t *)bp->b_data; 605 if (lastbn >= 0) { 606 MALLOC(copy, ufs_daddr_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 607 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 608 memset((caddr_t)&bap[last + 1], 0, 609 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t)); 610 error = VOP_BWRITE(bp); 611 if (error) 612 allerror = error; 613 bap = copy; 614 } 615 616 /* 617 * Recursively free totally unused blocks. 618 */ 619 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 620 i--, nlbn += factor) { 621 nb = bap[i]; 622 if (nb == 0) 623 continue; 624 if (level > SINGLE) { 625 error = lfs_indirtrunc(ip, nlbn, nb, 626 (ufs_daddr_t)-1, level - 1, 627 &blkcount, &rblkcount, 628 lastsegp, bcp, p); 629 if (error) 630 allerror = error; 631 blocksreleased += blkcount; 632 real_released += rblkcount; 633 } 634 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 635 if (bap[i] > 0) 636 real_released += nblocks; 637 blocksreleased += nblocks; 638 } 639 640 /* 641 * Recursively free last partial block. 642 */ 643 if (level > SINGLE && lastbn >= 0) { 644 last = lastbn % factor; 645 nb = bap[i]; 646 if (nb != 0) { 647 error = lfs_indirtrunc(ip, nlbn, nb, 648 last, level - 1, &blkcount, 649 &rblkcount, lastsegp, bcp, p); 650 if (error) 651 allerror = error; 652 real_released += rblkcount; 653 blocksreleased += blkcount; 654 } 655 } 656 657 if (copy != NULL) { 658 FREE(copy, M_TEMP); 659 } else { 660 if (bp->b_flags & B_DELWRI) { 661 LFS_UNLOCK_BUF(bp); 662 fs->lfs_avail += btofsb(fs, bp->b_bcount); 663 wakeup(&fs->lfs_avail); 664 } 665 bp->b_flags |= B_INVAL; 666 brelse(bp); 667 } 668 669 *countp = blocksreleased; 670 *rcountp = real_released; 671 return (allerror); 672 } 673 674 /* 675 * Destroy any in core blocks past the truncation length. 676 * Inlined from vtruncbuf, so that lfs_avail could be updated. 677 */ 678 static int 679 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 680 { 681 struct buf *bp, *nbp; 682 int s, error; 683 struct lfs *fs; 684 685 fs = VTOI(vp)->i_lfs; 686 s = splbio(); 687 688 restart: 689 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 690 nbp = LIST_NEXT(bp, b_vnbufs); 691 if (bp->b_lblkno < lbn) 692 continue; 693 if (bp->b_flags & B_BUSY) { 694 bp->b_flags |= B_WANTED; 695 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 696 "lfs_vtruncbuf", slptimeo); 697 if (error) { 698 splx(s); 699 return (error); 700 } 701 goto restart; 702 } 703 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 704 if (bp->b_flags & B_DELWRI) { 705 bp->b_flags &= ~B_DELWRI; 706 fs->lfs_avail += btofsb(fs, bp->b_bcount); 707 wakeup(&fs->lfs_avail); 708 } 709 LFS_UNLOCK_BUF(bp); 710 brelse(bp); 711 } 712 713 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 714 nbp = LIST_NEXT(bp, b_vnbufs); 715 if (bp->b_lblkno < lbn) 716 continue; 717 if (bp->b_flags & B_BUSY) { 718 bp->b_flags |= B_WANTED; 719 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 720 "lfs_vtruncbuf", slptimeo); 721 if (error) { 722 splx(s); 723 return (error); 724 } 725 goto restart; 726 } 727 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 728 if (bp->b_flags & B_DELWRI) { 729 bp->b_flags &= ~B_DELWRI; 730 fs->lfs_avail += btofsb(fs, bp->b_bcount); 731 wakeup(&fs->lfs_avail); 732 } 733 LFS_UNLOCK_BUF(bp); 734 brelse(bp); 735 } 736 737 splx(s); 738 739 return (0); 740 } 741 742
Indexes created Thu Oct 02 14:10:14 GMT 2025