lfs_inode.c revision 1.63
1 /* $NetBSD: lfs_inode.c,v 1.63 2003/01/25 16:40:29 fvdl 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 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.63 2003/01/25 16:40:29 fvdl Exp $"); 75 76 #if defined(_KERNEL_OPT) 77 #include "opt_quota.h" 78 #endif 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/mount.h> 83 #include <sys/proc.h> 84 #include <sys/file.h> 85 #include <sys/buf.h> 86 #include <sys/vnode.h> 87 #include <sys/kernel.h> 88 #include <sys/malloc.h> 89 #include <sys/trace.h> 90 #include <sys/resourcevar.h> 91 92 #include <ufs/ufs/quota.h> 93 #include <ufs/ufs/inode.h> 94 #include <ufs/ufs/ufsmount.h> 95 #include <ufs/ufs/ufs_extern.h> 96 97 #include <ufs/lfs/lfs.h> 98 #include <ufs/lfs/lfs_extern.h> 99 100 extern int locked_queue_count; 101 extern long locked_queue_bytes; 102 103 static int lfs_update_seguse(struct lfs *, long, size_t); 104 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t, 105 daddr_t, int, long *, long *, long *, size_t *, 106 struct proc *); 107 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *); 108 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int); 109 110 /* Search a block for a specific dinode. */ 111 struct dinode * 112 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 113 { 114 struct dinode *dip = (struct dinode *)bp->b_data; 115 struct dinode *ldip, *fin; 116 117 #ifdef LFS_IFILE_FRAG_ADDRESSING 118 if (fs->lfs_version == 1) 119 fin = dip + INOPB(fs); 120 else 121 fin = dip + INOPF(fs); 122 #else 123 fin = dip + INOPB(fs); 124 #endif 125 126 /* 127 * Read the inode block backwards, since later versions of the 128 * inode will supercede earlier ones. Though it is unlikely, it is 129 * possible that the same inode will appear in the same inode block. 130 */ 131 for (ldip = fin - 1; ldip >= dip; --ldip) 132 if (ldip->di_inumber == ino) 133 return (ldip); 134 135 printf("searched %d entries\n", (int)(fin - dip)); 136 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 137 dtosn(fs, fs->lfs_offset)); 138 printf("block is 0x%llx (seg %lld)\n", 139 (unsigned long long)dbtofsb(fs, bp->b_blkno), 140 (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno))); 141 142 return NULL; 143 } 144 145 int 146 lfs_update(void *v) 147 { 148 struct vop_update_args /* { 149 struct vnode *a_vp; 150 struct timespec *a_access; 151 struct timespec *a_modify; 152 int a_flags; 153 } */ *ap = v; 154 struct inode *ip; 155 struct vnode *vp = ap->a_vp; 156 struct timespec ts; 157 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 158 int s; 159 160 if (vp->v_mount->mnt_flag & MNT_RDONLY) 161 return (0); 162 ip = VTOI(vp); 163 164 /* 165 * If we are called from vinvalbuf, and the file's blocks have 166 * already been scheduled for writing, but the writes have not 167 * yet completed, lfs_vflush will not be called, and vinvalbuf 168 * will cause a panic. So, we must wait until any pending write 169 * for our inode completes, if we are called with UPDATE_WAIT set. 170 */ 171 s = splbio(); 172 while ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 173 WRITEINPROG(vp)) { 174 #ifdef DEBUG_LFS 175 printf("lfs_update: sleeping on inode %d (in-progress)\n", 176 ip->i_number); 177 #endif 178 tsleep(vp, (PRIBIO+1), "lfs_update", 0); 179 } 180 splx(s); 181 TIMEVAL_TO_TIMESPEC(&time, &ts); 182 LFS_ITIMES(ip, 183 ap->a_access ? ap->a_access : &ts, 184 ap->a_modify ? ap->a_modify : &ts, &ts); 185 if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) { 186 return (0); 187 } 188 189 /* If sync, push back the vnode and any dirty blocks it may have. */ 190 if ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 191 /* Avoid flushing VDIROP. */ 192 ++fs->lfs_diropwait; 193 while (vp->v_flag & VDIROP) { 194 #ifdef DEBUG_LFS 195 printf("lfs_update: sleeping on inode %d (dirops)\n", 196 ip->i_number); 197 printf("lfs_update: vflags 0x%x, iflags 0x%x\n", 198 vp->v_flag, ip->i_flag); 199 #endif 200 if (fs->lfs_dirops == 0) 201 lfs_flush_fs(fs, SEGM_SYNC); 202 else 203 tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 204 0); 205 /* XXX KS - by falling out here, are we writing the vn 206 twice? */ 207 } 208 --fs->lfs_diropwait; 209 return lfs_vflush(vp); 210 } 211 return 0; 212 } 213 214 #define SINGLE 0 /* index of single indirect block */ 215 #define DOUBLE 1 /* index of double indirect block */ 216 #define TRIPLE 2 /* index of triple indirect block */ 217 /* 218 * Truncate the inode oip to at most length size, freeing the 219 * disk blocks. 220 */ 221 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */ 222 223 int 224 lfs_truncate(void *v) 225 { 226 struct vop_truncate_args /* { 227 struct vnode *a_vp; 228 off_t a_length; 229 int a_flags; 230 struct ucred *a_cred; 231 struct proc *a_p; 232 } */ *ap = v; 233 struct vnode *ovp = ap->a_vp; 234 daddr_t lastblock; 235 struct inode *oip; 236 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 237 /* XXX ondisk32 */ 238 int32_t newblks[NDADDR + NIADDR]; 239 off_t length = ap->a_length; 240 struct lfs *fs; 241 struct buf *bp; 242 int offset, size, level; 243 long count, rcount, nblocks, blocksreleased = 0, real_released = 0; 244 int i; 245 int aflags, error, allerror = 0; 246 off_t osize; 247 long lastseg; 248 size_t bc; 249 int obufsize, odb; 250 251 if (length < 0) 252 return (EINVAL); 253 oip = VTOI(ovp); 254 255 /* 256 * Just return and not update modification times. 257 */ 258 if (oip->i_ffs_size == length) 259 return (0); 260 261 if (ovp->v_type == VLNK && 262 (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen || 263 (ovp->v_mount->mnt_maxsymlinklen == 0 && 264 oip->i_din.ffs_din.di_blocks == 0))) { 265 #ifdef DIAGNOSTIC 266 if (length != 0) 267 panic("lfs_truncate: partial truncate of symlink"); 268 #endif 269 memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size); 270 oip->i_ffs_size = 0; 271 oip->i_flag |= IN_CHANGE | IN_UPDATE; 272 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 273 } 274 if (oip->i_ffs_size == length) { 275 oip->i_flag |= IN_CHANGE | IN_UPDATE; 276 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 277 } 278 #ifdef QUOTA 279 if ((error = getinoquota(oip)) != 0) 280 return (error); 281 #endif 282 fs = oip->i_lfs; 283 lfs_imtime(fs); 284 osize = oip->i_ffs_size; 285 286 /* 287 * Lengthen the size of the file. We must ensure that the 288 * last byte of the file is allocated. Since the smallest 289 * value of osize is 0, length will be at least 1. 290 */ 291 if (osize < length) { 292 if (length > fs->lfs_maxfilesize) 293 return (EFBIG); 294 aflags = B_CLRBUF; 295 if (ap->a_flags & IO_SYNC) 296 aflags |= B_SYNC; 297 error = lfs_reserve(fs, ovp, NULL, 298 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 299 if (error) 300 return (error); 301 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 302 lfs_reserve(fs, ovp, NULL, 303 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 304 if (error) 305 return (error); 306 oip->i_ffs_size = length; 307 uvm_vnp_setsize(ovp, length); 308 (void) VOP_BWRITE(bp); 309 oip->i_flag |= IN_CHANGE | IN_UPDATE; 310 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 311 } 312 313 if ((error = lfs_reserve(fs, ovp, NULL, 314 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 315 return (error); 316 /* 317 * Make sure no writes to this inode can happen while we're 318 * truncating. Otherwise, blocks which are accounted for on the 319 * inode *and* which have been created for cleaning can coexist, 320 * and cause an overcounting. 321 */ 322 #ifdef LFS_FRAGSIZE_SEGLOCK 323 lfs_seglock(fs, SEGM_PROT); 324 #else 325 lockmgr(&fs->lfs_fraglock, LK_SHARED, 0); 326 #endif 327 328 /* 329 * Shorten the size of the file. If the file is not being 330 * truncated to a block boundary, the contents of the 331 * partial block following the end of the file must be 332 * zero'ed in case it ever becomes accessible again because 333 * of subsequent file growth. Directories however are not 334 * zero'ed as they should grow back initialized to empty. 335 */ 336 offset = blkoff(fs, length); 337 lastseg = -1; 338 bc = 0; 339 if (offset == 0) { 340 oip->i_ffs_size = length; 341 } else { 342 lbn = lblkno(fs, length); 343 aflags = B_CLRBUF; 344 if (ap->a_flags & IO_SYNC) 345 aflags |= B_SYNC; 346 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 347 if (error) { 348 lfs_reserve(fs, ovp, NULL, 349 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 350 #ifdef LFS_FRAGSIZE_SEGLOCK 351 lfs_segunlock(fs); 352 #else 353 lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0); 354 #endif 355 return (error); 356 } 357 obufsize = bp->b_bufsize; 358 odb = btofsb(fs, bp->b_bcount); 359 oip->i_ffs_size = length; 360 size = blksize(fs, oip, lbn); 361 if (ovp->v_type != VDIR) 362 memset((char *)bp->b_data + offset, 0, 363 (u_int)(size - offset)); 364 allocbuf(bp, size); 365 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 366 locked_queue_bytes -= obufsize - bp->b_bufsize; 367 if (bp->b_flags & B_DELWRI) 368 fs->lfs_avail += odb - btofsb(fs, size); 369 (void) VOP_BWRITE(bp); 370 } 371 uvm_vnp_setsize(ovp, length); 372 /* 373 * Calculate index into inode's block list of 374 * last direct and indirect blocks (if any) 375 * which we want to keep. Lastblock is -1 when 376 * the file is truncated to 0. 377 */ 378 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 379 lastiblock[SINGLE] = lastblock - NDADDR; 380 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 381 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 382 nblocks = btofsb(fs, fs->lfs_bsize); 383 /* 384 * Record changed file and block pointers before we start 385 * freeing blocks. lastiblock values are also normalized to -1 386 * for calls to lfs_indirtrunc below. 387 */ 388 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks); 389 for (level = TRIPLE; level >= SINGLE; level--) 390 if (lastiblock[level] < 0) { 391 newblks[NDADDR+level] = 0; 392 lastiblock[level] = -1; 393 } 394 for (i = NDADDR - 1; i > lastblock; i--) 395 newblks[i] = 0; 396 397 oip->i_ffs_size = osize; 398 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 399 if (error && !allerror) 400 allerror = error; 401 402 /* 403 * Indirect blocks first. 404 */ 405 indir_lbn[SINGLE] = -NDADDR; 406 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 407 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 408 for (level = TRIPLE; level >= SINGLE; level--) { 409 bn = oip->i_ffs_ib[level]; 410 if (bn != 0) { 411 error = lfs_indirtrunc(oip, indir_lbn[level], 412 bn, lastiblock[level], 413 level, &count, &rcount, 414 &lastseg, &bc, ap->a_p); 415 if (error) 416 allerror = error; 417 real_released += rcount; 418 blocksreleased += count; 419 if (lastiblock[level] < 0) { 420 if (oip->i_ffs_ib[level] > 0) 421 real_released += nblocks; 422 blocksreleased += nblocks; 423 oip->i_ffs_ib[level] = 0; 424 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 425 } 426 } 427 if (lastiblock[level] >= 0) 428 goto done; 429 } 430 431 /* 432 * All whole direct blocks or frags. 433 */ 434 for (i = NDADDR - 1; i > lastblock; i--) { 435 long bsize, obsize; 436 437 bn = oip->i_ffs_db[i]; 438 if (bn == 0) 439 continue; 440 bsize = blksize(fs, oip, i); 441 if (oip->i_ffs_db[i] > 0) { 442 /* Check for fragment size changes */ 443 obsize = oip->i_lfs_fragsize[i]; 444 real_released += btofsb(fs, obsize); 445 oip->i_lfs_fragsize[i] = 0; 446 } else 447 obsize = 0; 448 blocksreleased += btofsb(fs, bsize); 449 oip->i_ffs_db[i] = 0; 450 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 451 } 452 if (lastblock < 0) 453 goto done; 454 455 /* 456 * Finally, look for a change in size of the 457 * last direct block; release any frags. 458 */ 459 bn = oip->i_ffs_db[lastblock]; 460 if (bn != 0) { 461 long oldspace, newspace, olddspace; 462 463 /* 464 * Calculate amount of space we're giving 465 * back as old block size minus new block size. 466 */ 467 oldspace = blksize(fs, oip, lastblock); 468 olddspace = oip->i_lfs_fragsize[lastblock]; 469 470 oip->i_ffs_size = length; 471 newspace = blksize(fs, oip, lastblock); 472 if (newspace == 0) 473 panic("itrunc: newspace"); 474 if (oldspace - newspace > 0) { 475 blocksreleased += btofsb(fs, oldspace - newspace); 476 } 477 #if 0 478 if (bn > 0 && olddspace - newspace > 0) { 479 /* No segment accounting here, just vnode */ 480 real_released += btofsb(fs, olddspace - newspace); 481 } 482 #endif 483 } 484 485 done: 486 /* Finish segment accounting corrections */ 487 lfs_update_seguse(fs, lastseg, bc); 488 #ifdef DIAGNOSTIC 489 for (level = SINGLE; level <= TRIPLE; level++) 490 if (newblks[NDADDR + level] != oip->i_ffs_ib[level]) 491 panic("lfs itrunc1"); 492 for (i = 0; i < NDADDR; i++) 493 if (newblks[i] != oip->i_ffs_db[i]) 494 panic("lfs itrunc2"); 495 if (length == 0 && 496 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 497 panic("lfs itrunc3"); 498 #endif /* DIAGNOSTIC */ 499 /* 500 * Put back the real size. 501 */ 502 oip->i_ffs_size = length; 503 oip->i_lfs_effnblks -= blocksreleased; 504 oip->i_ffs_blocks -= real_released; 505 fs->lfs_bfree += blocksreleased; 506 #ifdef DIAGNOSTIC 507 if (oip->i_ffs_size == 0 && oip->i_ffs_blocks != 0) { 508 printf("lfs_truncate: truncate to 0 but %d blocks on inode\n", 509 oip->i_ffs_blocks); 510 panic("lfs_truncate: persistent blocks"); 511 } 512 #endif 513 oip->i_flag |= IN_CHANGE; 514 #ifdef QUOTA 515 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 516 #endif 517 lfs_reserve(fs, ovp, NULL, 518 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 519 #ifdef LFS_FRAGSIZE_SEGLOCK 520 lfs_segunlock(fs); 521 #else 522 lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0); 523 #endif 524 return (allerror); 525 } 526 527 /* Update segment usage information when removing a block. */ 528 static int 529 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 530 size_t *num) 531 { 532 long seg; 533 int error = 0; 534 535 bsize = fragroundup(fs, bsize); 536 if (daddr > 0) { 537 if (*lastseg != (seg = dtosn(fs, daddr))) { 538 error = lfs_update_seguse(fs, *lastseg, *num); 539 *num = bsize; 540 *lastseg = seg; 541 } else 542 *num += bsize; 543 } 544 return error; 545 } 546 547 /* Finish the accounting updates for a segment. */ 548 static int 549 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 550 { 551 SEGUSE *sup; 552 struct buf *bp; 553 int error; 554 555 if (lastseg < 0 || num == 0) 556 return 0; 557 558 LFS_SEGENTRY(sup, fs, lastseg, bp); 559 if (num > sup->su_nbytes) { 560 printf("lfs_truncate: segment %ld short by %ld\n", 561 lastseg, (long)num - sup->su_nbytes); 562 panic("lfs_truncate: negative bytes"); 563 sup->su_nbytes = num; 564 } 565 sup->su_nbytes -= num; 566 error = LFS_BWRITE_LOG(bp); /* Ifile */ 567 return error; 568 } 569 570 /* 571 * Release blocks associated with the inode ip and stored in the indirect 572 * block bn. Blocks are free'd in LIFO order up to (but not including) 573 * lastbn. If level is greater than SINGLE, the block is an indirect block 574 * and recursive calls to indirtrunc must be used to cleanse other indirect 575 * blocks. 576 * 577 * NB: triple indirect blocks are untested. 578 */ 579 static int 580 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 581 daddr_t lastbn, int level, long *countp, 582 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 583 { 584 int i; 585 struct buf *bp; 586 struct lfs *fs = ip->i_lfs; 587 int32_t *bap; /* XXX ondisk32 */ 588 struct vnode *vp; 589 daddr_t nb, nlbn, last; 590 int32_t *copy = NULL; /* XXX ondisk32 */ 591 long blkcount, rblkcount, factor; 592 int nblocks, blocksreleased = 0, real_released = 0; 593 int error = 0, allerror = 0; 594 595 /* 596 * Calculate index in current block of last 597 * block to be kept. -1 indicates the entire 598 * block so we need not calculate the index. 599 */ 600 factor = 1; 601 for (i = SINGLE; i < level; i++) 602 factor *= NINDIR(fs); 603 last = lastbn; 604 if (lastbn > 0) 605 last /= factor; 606 nblocks = btofsb(fs, fs->lfs_bsize); 607 /* 608 * Get buffer of block pointers, zero those entries corresponding 609 * to blocks to be free'd, and update on disk copy first. Since 610 * double(triple) indirect before single(double) indirect, calls 611 * to bmap on these blocks will fail. However, we already have 612 * the on disk address, so we have to set the b_blkno field 613 * explicitly instead of letting bread do everything for us. 614 */ 615 vp = ITOV(ip); 616 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 617 if (bp->b_flags & (B_DONE | B_DELWRI)) { 618 /* Braces must be here in case trace evaluates to nothing. */ 619 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 620 } else { 621 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 622 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 623 bp->b_flags |= B_READ; 624 if (bp->b_bcount > bp->b_bufsize) 625 panic("lfs_indirtrunc: bad buffer size"); 626 bp->b_blkno = fsbtodb(fs, dbn); 627 VOP_STRATEGY(bp); 628 error = biowait(bp); 629 } 630 if (error) { 631 brelse(bp); 632 *countp = *rcountp = 0; 633 return (error); 634 } 635 636 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 637 if (lastbn >= 0) { 638 MALLOC(copy, int32_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 639 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 640 memset((caddr_t)&bap[last + 1], 0, 641 /* XXX ondisk32 */ 642 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 643 error = VOP_BWRITE(bp); 644 if (error) 645 allerror = error; 646 bap = copy; 647 } 648 649 /* 650 * Recursively free totally unused blocks. 651 */ 652 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 653 i--, nlbn += factor) { 654 nb = bap[i]; 655 if (nb == 0) 656 continue; 657 if (level > SINGLE) { 658 error = lfs_indirtrunc(ip, nlbn, nb, 659 (daddr_t)-1, level - 1, 660 &blkcount, &rblkcount, 661 lastsegp, bcp, p); 662 if (error) 663 allerror = error; 664 blocksreleased += blkcount; 665 real_released += rblkcount; 666 } 667 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 668 if (bap[i] > 0) 669 real_released += nblocks; 670 blocksreleased += nblocks; 671 } 672 673 /* 674 * Recursively free last partial block. 675 */ 676 if (level > SINGLE && lastbn >= 0) { 677 last = lastbn % factor; 678 nb = bap[i]; 679 if (nb != 0) { 680 error = lfs_indirtrunc(ip, nlbn, nb, 681 last, level - 1, &blkcount, 682 &rblkcount, lastsegp, bcp, p); 683 if (error) 684 allerror = error; 685 real_released += rblkcount; 686 blocksreleased += blkcount; 687 } 688 } 689 690 if (copy != NULL) { 691 FREE(copy, M_TEMP); 692 } else { 693 if (bp->b_flags & B_DELWRI) { 694 LFS_UNLOCK_BUF(bp); 695 fs->lfs_avail += btofsb(fs, bp->b_bcount); 696 wakeup(&fs->lfs_avail); 697 } 698 bp->b_flags |= B_INVAL; 699 brelse(bp); 700 } 701 702 *countp = blocksreleased; 703 *rcountp = real_released; 704 return (allerror); 705 } 706 707 /* 708 * Destroy any in core blocks past the truncation length. 709 * Inlined from vtruncbuf, so that lfs_avail could be updated. 710 */ 711 static int 712 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 713 { 714 struct buf *bp, *nbp; 715 int s, error; 716 struct lfs *fs; 717 718 fs = VTOI(vp)->i_lfs; 719 s = splbio(); 720 721 restart: 722 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 723 nbp = LIST_NEXT(bp, b_vnbufs); 724 if (bp->b_lblkno < lbn) 725 continue; 726 if (bp->b_flags & B_BUSY) { 727 bp->b_flags |= B_WANTED; 728 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 729 "lfs_vtruncbuf", slptimeo); 730 if (error) { 731 splx(s); 732 return (error); 733 } 734 goto restart; 735 } 736 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 737 if (bp->b_flags & B_DELWRI) { 738 bp->b_flags &= ~B_DELWRI; 739 fs->lfs_avail += btofsb(fs, bp->b_bcount); 740 wakeup(&fs->lfs_avail); 741 } 742 LFS_UNLOCK_BUF(bp); 743 brelse(bp); 744 } 745 746 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 747 nbp = LIST_NEXT(bp, b_vnbufs); 748 if (bp->b_lblkno < lbn) 749 continue; 750 if (bp->b_flags & B_BUSY) { 751 bp->b_flags |= B_WANTED; 752 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 753 "lfs_vtruncbuf", slptimeo); 754 if (error) { 755 splx(s); 756 return (error); 757 } 758 goto restart; 759 } 760 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 761 if (bp->b_flags & B_DELWRI) { 762 bp->b_flags &= ~B_DELWRI; 763 fs->lfs_avail += btofsb(fs, bp->b_bcount); 764 wakeup(&fs->lfs_avail); 765 } 766 LFS_UNLOCK_BUF(bp); 767 brelse(bp); 768 } 769 770 splx(s); 771 772 return (0); 773 } 774 775
Indexes created Sun Sep 28 12:09:53 GMT 2025