lfs_inode.c revision 1.85
1 /* $NetBSD: lfs_inode.c,v 1.85 2004/08/15 07:19:56 mycroft Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 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. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_inode.c 8.9 (Berkeley) 5/8/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.85 2004/08/15 07:19:56 mycroft Exp $"); 71 72 #if defined(_KERNEL_OPT) 73 #include "opt_quota.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/mount.h> 79 #include <sys/proc.h> 80 #include <sys/file.h> 81 #include <sys/buf.h> 82 #include <sys/vnode.h> 83 #include <sys/kernel.h> 84 #include <sys/malloc.h> 85 #include <sys/trace.h> 86 #include <sys/resourcevar.h> 87 88 #include <ufs/ufs/quota.h> 89 #include <ufs/ufs/inode.h> 90 #include <ufs/ufs/ufsmount.h> 91 #include <ufs/ufs/ufs_extern.h> 92 93 #include <ufs/lfs/lfs.h> 94 #include <ufs/lfs/lfs_extern.h> 95 96 static int lfs_update_seguse(struct lfs *, long, size_t); 97 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t, 98 daddr_t, int, long *, long *, long *, size_t *, 99 struct proc *); 100 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *); 101 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int); 102 103 /* Search a block for a specific dinode. */ 104 struct ufs1_dinode * 105 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 106 { 107 struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data; 108 struct ufs1_dinode *ldip, *fin; 109 110 #ifdef LFS_IFILE_FRAG_ADDRESSING 111 if (fs->lfs_version == 1) 112 fin = dip + INOPB(fs); 113 else 114 fin = dip + INOPF(fs); 115 #else 116 fin = dip + INOPB(fs); 117 #endif 118 119 /* 120 * Read the inode block backwards, since later versions of the 121 * inode will supercede earlier ones. Though it is unlikely, it is 122 * possible that the same inode will appear in the same inode block. 123 */ 124 for (ldip = fin - 1; ldip >= dip; --ldip) 125 if (ldip->di_inumber == ino) 126 return (ldip); 127 128 printf("searched %d entries\n", (int)(fin - dip)); 129 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 130 dtosn(fs, fs->lfs_offset)); 131 printf("block is 0x%llx (seg %lld)\n", 132 (unsigned long long)dbtofsb(fs, bp->b_blkno), 133 (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno))); 134 135 return NULL; 136 } 137 138 int 139 lfs_update(void *v) 140 { 141 struct vop_update_args /* { 142 struct vnode *a_vp; 143 struct timespec *a_access; 144 struct timespec *a_modify; 145 int a_flags; 146 } */ *ap = v; 147 struct inode *ip; 148 struct vnode *vp = ap->a_vp; 149 struct timespec ts; 150 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 151 int s; 152 int flags; 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 s = splbio(); 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 splx(s); 175 TIMEVAL_TO_TIMESPEC(&time, &ts); 176 LFS_ITIMES(ip, 177 ap->a_access ? ap->a_access : &ts, 178 ap->a_modify ? ap->a_modify : &ts, &ts); 179 if (ap->a_flags & UPDATE_CLOSE) 180 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 181 else 182 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 183 if (flags == 0) 184 return (0); 185 186 /* If sync, push back the vnode and any dirty blocks it may have. */ 187 if ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 188 /* Avoid flushing VDIROP. */ 189 ++fs->lfs_diropwait; 190 while (vp->v_flag & VDIROP) { 191 #ifdef DEBUG_LFS 192 printf("lfs_update: sleeping on inode %d (dirops)\n", 193 ip->i_number); 194 printf("lfs_update: vflags 0x%x, iflags 0x%x\n", 195 vp->v_flag, ip->i_flag); 196 #endif 197 if (fs->lfs_dirops == 0) 198 lfs_flush_fs(fs, SEGM_SYNC); 199 else 200 tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 201 0); 202 /* XXX KS - by falling out here, are we writing the vn 203 twice? */ 204 } 205 --fs->lfs_diropwait; 206 return lfs_vflush(vp); 207 } 208 return 0; 209 } 210 211 #define SINGLE 0 /* index of single indirect block */ 212 #define DOUBLE 1 /* index of double indirect block */ 213 #define TRIPLE 2 /* index of triple indirect block */ 214 /* 215 * Truncate the inode oip to at most length size, freeing the 216 * disk blocks. 217 */ 218 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */ 219 220 int 221 lfs_truncate(void *v) 222 { 223 struct vop_truncate_args /* { 224 struct vnode *a_vp; 225 off_t a_length; 226 int a_flags; 227 struct ucred *a_cred; 228 struct proc *a_p; 229 } */ *ap = v; 230 struct vnode *ovp = ap->a_vp; 231 struct genfs_node *gp = VTOG(ovp); 232 daddr_t lastblock; 233 struct inode *oip = VTOI(ovp); 234 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 235 /* XXX ondisk32 */ 236 int32_t newblks[NDADDR + NIADDR]; 237 off_t length = ap->a_length; 238 struct lfs *fs; 239 struct buf *bp; 240 int offset, size, level; 241 long count, rcount, nblocks, blocksreleased = 0, real_released = 0; 242 int i; 243 int aflags, error, allerror = 0; 244 off_t osize; 245 voff_t eoz; 246 long lastseg; 247 size_t bc; 248 int obufsize, odb; 249 int usepc; 250 struct ufsmount *ump = oip->i_ump; 251 252 if (length < 0) 253 return (EINVAL); 254 255 /* 256 * Just return and not update modification times. 257 */ 258 if (oip->i_size == length) 259 return (0); 260 261 if (ovp->v_type == VLNK && 262 (oip->i_size < ump->um_maxsymlinklen || 263 (ump->um_maxsymlinklen == 0 && 264 oip->i_ffs1_blocks == 0))) { 265 #ifdef DIAGNOSTIC 266 if (length != 0) 267 panic("lfs_truncate: partial truncate of symlink"); 268 #endif 269 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 270 oip->i_size = oip->i_ffs1_size = 0; 271 oip->i_flag |= IN_CHANGE | IN_UPDATE; 272 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 273 } 274 if (oip->i_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_size; 285 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 286 287 /* 288 * Lengthen the size of the file. We must ensure that the 289 * last byte of the file is allocated. Since the smallest 290 * value of osize is 0, length will be at least 1. 291 */ 292 if (osize < length) { 293 if (length > ump->um_maxfilesize) 294 return (EFBIG); 295 aflags = B_CLRBUF; 296 if (ap->a_flags & IO_SYNC) 297 aflags |= B_SYNC; 298 if (usepc) { 299 if (lblkno(fs, osize) < NDADDR && 300 lblkno(fs, osize) != lblkno(fs, length) && 301 blkroundup(fs, osize) != osize) { 302 error = ufs_balloc_range(ovp, osize, 303 blkroundup(fs, osize) - 304 osize, ap->a_cred, 305 aflags); 306 if (error) { 307 return error; 308 } 309 if (ap->a_flags & IO_SYNC) { 310 ovp->v_size = blkroundup(fs, osize); 311 simple_lock(&ovp->v_interlock); 312 VOP_PUTPAGES(ovp, 313 trunc_page(osize & fs->lfs_bmask), 314 round_page(ovp->v_size), 315 PGO_CLEANIT | PGO_SYNCIO); 316 } 317 } 318 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 319 aflags); 320 if (error) { 321 (void) VOP_TRUNCATE(ovp, osize, 322 ap->a_flags & IO_SYNC, 323 ap->a_cred, ap->a_p); 324 return error; 325 } 326 uvm_vnp_setsize(ovp, length); 327 oip->i_flag |= IN_CHANGE | IN_UPDATE; 328 KASSERT(ovp->v_size == oip->i_size); 329 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 330 } else { 331 error = lfs_reserve(fs, ovp, NULL, 332 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 333 if (error) 334 return (error); 335 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, 336 aflags, &bp); 337 lfs_reserve(fs, ovp, NULL, 338 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 339 if (error) 340 return (error); 341 oip->i_ffs1_size = oip->i_size = length; 342 uvm_vnp_setsize(ovp, length); 343 (void) VOP_BWRITE(bp); 344 oip->i_flag |= IN_CHANGE | IN_UPDATE; 345 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 346 } 347 } 348 349 if ((error = lfs_reserve(fs, ovp, NULL, 350 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 351 return (error); 352 353 /* 354 * Shorten the size of the file. If the file is not being 355 * truncated to a block boundary, the contents of the 356 * partial block following the end of the file must be 357 * zero'ed in case it ever becomes accessible again because 358 * of subsequent file growth. Directories however are not 359 * zero'ed as they should grow back initialized to empty. 360 */ 361 offset = blkoff(fs, length); 362 lastseg = -1; 363 bc = 0; 364 365 lfs_seglock(fs, SEGM_PROT); 366 if (offset == 0) { 367 oip->i_size = oip->i_ffs1_size = length; 368 } else if (!usepc) { 369 lbn = lblkno(fs, length); 370 aflags = B_CLRBUF; 371 if (ap->a_flags & IO_SYNC) 372 aflags |= B_SYNC; 373 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 374 if (error) { 375 lfs_reserve(fs, ovp, NULL, 376 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 377 goto errout; 378 } 379 obufsize = bp->b_bufsize; 380 odb = btofsb(fs, bp->b_bcount); 381 oip->i_size = oip->i_ffs1_size = length; 382 size = blksize(fs, oip, lbn); 383 if (ovp->v_type != VDIR) 384 memset((char *)bp->b_data + offset, 0, 385 (u_int)(size - offset)); 386 allocbuf(bp, size, 1); 387 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 388 locked_queue_bytes -= obufsize - bp->b_bufsize; 389 if (bp->b_flags & B_DELWRI) 390 fs->lfs_avail += odb - btofsb(fs, size); 391 (void) VOP_BWRITE(bp); 392 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 393 /* 394 * When truncating a regular file down to a non-block-aligned 395 * size, we must zero the part of last block which is past 396 * the new EOF. We must synchronously flush the zeroed pages 397 * to disk since the new pages will be invalidated as soon 398 * as we inform the VM system of the new, smaller size. 399 * We must do this before acquiring the GLOCK, since fetching 400 * the pages will acquire the GLOCK internally. 401 * So there is a window where another thread could see a whole 402 * zeroed page past EOF, but that's life. 403 */ 404 aflags = ap->a_flags & IO_SYNC ? B_SYNC : 0; 405 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 406 aflags); 407 if (error) { 408 lfs_reserve(fs, ovp, NULL, 409 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 410 goto errout; 411 } 412 eoz = blkroundup(fs, length); 413 uvm_vnp_zerorange(ovp, length, eoz - length); 414 simple_lock(&ovp->v_interlock); 415 error = VOP_PUTPAGES(ovp, trunc_page(length), round_page(eoz), 416 PGO_CLEANIT | PGO_DEACTIVATE | (aflags ? PGO_SYNCIO : 0)); 417 if (error) { 418 lfs_reserve(fs, ovp, NULL, 419 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 420 goto errout; 421 } 422 } 423 424 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 425 426 oip->i_size = oip->i_ffs1_size = length; 427 uvm_vnp_setsize(ovp, length); 428 /* 429 * Calculate index into inode's block list of 430 * last direct and indirect blocks (if any) 431 * which we want to keep. Lastblock is -1 when 432 * the file is truncated to 0. 433 */ 434 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 435 lastiblock[SINGLE] = lastblock - NDADDR; 436 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 437 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 438 nblocks = btofsb(fs, fs->lfs_bsize); 439 /* 440 * Record changed file and block pointers before we start 441 * freeing blocks. lastiblock values are also normalized to -1 442 * for calls to lfs_indirtrunc below. 443 */ 444 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 445 for (level = TRIPLE; level >= SINGLE; level--) 446 if (lastiblock[level] < 0) { 447 newblks[NDADDR+level] = 0; 448 lastiblock[level] = -1; 449 } 450 for (i = NDADDR - 1; i > lastblock; i--) 451 newblks[i] = 0; 452 453 oip->i_size = oip->i_ffs1_size = osize; 454 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 455 if (error && !allerror) 456 allerror = error; 457 458 /* 459 * Indirect blocks first. 460 */ 461 indir_lbn[SINGLE] = -NDADDR; 462 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 463 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 464 for (level = TRIPLE; level >= SINGLE; level--) { 465 bn = oip->i_ffs1_ib[level]; 466 if (bn != 0) { 467 error = lfs_indirtrunc(oip, indir_lbn[level], 468 bn, lastiblock[level], 469 level, &count, &rcount, 470 &lastseg, &bc, ap->a_p); 471 if (error) 472 allerror = error; 473 real_released += rcount; 474 blocksreleased += count; 475 if (lastiblock[level] < 0) { 476 if (oip->i_ffs1_ib[level] > 0) 477 real_released += nblocks; 478 blocksreleased += nblocks; 479 oip->i_ffs1_ib[level] = 0; 480 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 481 } 482 } 483 if (lastiblock[level] >= 0) 484 goto done; 485 } 486 487 /* 488 * All whole direct blocks or frags. 489 */ 490 for (i = NDADDR - 1; i > lastblock; i--) { 491 long bsize, obsize; 492 493 bn = oip->i_ffs1_db[i]; 494 if (bn == 0) 495 continue; 496 bsize = blksize(fs, oip, i); 497 if (oip->i_ffs1_db[i] > 0) { 498 /* Check for fragment size changes */ 499 obsize = oip->i_lfs_fragsize[i]; 500 real_released += btofsb(fs, obsize); 501 oip->i_lfs_fragsize[i] = 0; 502 } else 503 obsize = 0; 504 blocksreleased += btofsb(fs, bsize); 505 oip->i_ffs1_db[i] = 0; 506 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 507 } 508 if (lastblock < 0) 509 goto done; 510 511 /* 512 * Finally, look for a change in size of the 513 * last direct block; release any frags. 514 */ 515 bn = oip->i_ffs1_db[lastblock]; 516 if (bn != 0) { 517 long oldspace, newspace; 518 #if 0 519 long olddspace; 520 #endif 521 522 /* 523 * Calculate amount of space we're giving 524 * back as old block size minus new block size. 525 */ 526 oldspace = blksize(fs, oip, lastblock); 527 #if 0 528 olddspace = oip->i_lfs_fragsize[lastblock]; 529 #endif 530 531 oip->i_size = oip->i_ffs1_size = length; 532 newspace = blksize(fs, oip, lastblock); 533 if (newspace == 0) 534 panic("itrunc: newspace"); 535 if (oldspace - newspace > 0) { 536 blocksreleased += btofsb(fs, oldspace - newspace); 537 } 538 #if 0 539 if (bn > 0 && olddspace - newspace > 0) { 540 /* No segment accounting here, just vnode */ 541 real_released += btofsb(fs, olddspace - newspace); 542 } 543 #endif 544 } 545 546 done: 547 /* Finish segment accounting corrections */ 548 lfs_update_seguse(fs, lastseg, bc); 549 #ifdef DIAGNOSTIC 550 for (level = SINGLE; level <= TRIPLE; level++) 551 if ((newblks[NDADDR + level] == 0) != 552 (oip->i_ffs1_ib[level]) == 0) { 553 panic("lfs itrunc1"); 554 } 555 for (i = 0; i < NDADDR; i++) 556 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 557 panic("lfs itrunc2"); 558 } 559 if (length == 0 && 560 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 561 panic("lfs itrunc3"); 562 #endif /* DIAGNOSTIC */ 563 /* 564 * Put back the real size. 565 */ 566 oip->i_size = oip->i_ffs1_size = length; 567 oip->i_lfs_effnblks -= blocksreleased; 568 oip->i_ffs1_blocks -= real_released; 569 fs->lfs_bfree += blocksreleased; 570 #ifdef DIAGNOSTIC 571 if (oip->i_size == 0 && 572 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 573 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 574 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 575 panic("lfs_truncate: persistent blocks"); 576 } 577 #endif 578 oip->i_flag |= IN_CHANGE; 579 #ifdef QUOTA 580 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 581 #endif 582 lfs_reserve(fs, ovp, NULL, 583 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 584 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 585 errout: 586 lfs_segunlock(fs); 587 return (allerror ? allerror : error); 588 } 589 590 /* Update segment usage information when removing a block. */ 591 static int 592 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 593 size_t *num) 594 { 595 long seg; 596 int error = 0; 597 598 bsize = fragroundup(fs, bsize); 599 if (daddr > 0) { 600 if (*lastseg != (seg = dtosn(fs, daddr))) { 601 error = lfs_update_seguse(fs, *lastseg, *num); 602 *num = bsize; 603 *lastseg = seg; 604 } else 605 *num += bsize; 606 } 607 return error; 608 } 609 610 /* Finish the accounting updates for a segment. */ 611 static int 612 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 613 { 614 SEGUSE *sup; 615 struct buf *bp; 616 617 if (lastseg < 0 || num == 0) 618 return 0; 619 620 LFS_SEGENTRY(sup, fs, lastseg, bp); 621 if (num > sup->su_nbytes) { 622 printf("lfs_truncate: segment %ld short by %ld\n", 623 lastseg, (long)num - sup->su_nbytes); 624 panic("lfs_truncate: negative bytes"); 625 sup->su_nbytes = num; 626 } 627 sup->su_nbytes -= num; 628 LFS_WRITESEGENTRY(sup, fs, lastseg, bp); 629 630 return 0; 631 } 632 633 /* 634 * Release blocks associated with the inode ip and stored in the indirect 635 * block bn. Blocks are free'd in LIFO order up to (but not including) 636 * lastbn. If level is greater than SINGLE, the block is an indirect block 637 * and recursive calls to indirtrunc must be used to cleanse other indirect 638 * blocks. 639 * 640 * NB: triple indirect blocks are untested. 641 */ 642 static int 643 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 644 daddr_t lastbn, int level, long *countp, 645 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 646 { 647 int i; 648 struct buf *bp; 649 struct lfs *fs = ip->i_lfs; 650 int32_t *bap; /* XXX ondisk32 */ 651 struct vnode *vp; 652 daddr_t nb, nlbn, last; 653 int32_t *copy = NULL; /* XXX ondisk32 */ 654 long blkcount, rblkcount, factor; 655 int nblocks, blocksreleased = 0, real_released = 0; 656 int error = 0, allerror = 0; 657 658 /* 659 * Calculate index in current block of last 660 * block to be kept. -1 indicates the entire 661 * block so we need not calculate the index. 662 */ 663 factor = 1; 664 for (i = SINGLE; i < level; i++) 665 factor *= NINDIR(fs); 666 last = lastbn; 667 if (lastbn > 0) 668 last /= factor; 669 nblocks = btofsb(fs, fs->lfs_bsize); 670 /* 671 * Get buffer of block pointers, zero those entries corresponding 672 * to blocks to be free'd, and update on disk copy first. Since 673 * double(triple) indirect before single(double) indirect, calls 674 * to bmap on these blocks will fail. However, we already have 675 * the on disk address, so we have to set the b_blkno field 676 * explicitly instead of letting bread do everything for us. 677 */ 678 vp = ITOV(ip); 679 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 680 if (bp->b_flags & (B_DONE | B_DELWRI)) { 681 /* Braces must be here in case trace evaluates to nothing. */ 682 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 683 } else { 684 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 685 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 686 bp->b_flags |= B_READ; 687 if (bp->b_bcount > bp->b_bufsize) 688 panic("lfs_indirtrunc: bad buffer size"); 689 bp->b_blkno = fsbtodb(fs, dbn); 690 VOP_STRATEGY(vp, bp); 691 error = biowait(bp); 692 } 693 if (error) { 694 brelse(bp); 695 *countp = *rcountp = 0; 696 return (error); 697 } 698 699 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 700 if (lastbn >= 0) { 701 MALLOC(copy, int32_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 702 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 703 memset((caddr_t)&bap[last + 1], 0, 704 /* XXX ondisk32 */ 705 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 706 error = VOP_BWRITE(bp); 707 if (error) 708 allerror = error; 709 bap = copy; 710 } 711 712 /* 713 * Recursively free totally unused blocks. 714 */ 715 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 716 i--, nlbn += factor) { 717 nb = bap[i]; 718 if (nb == 0) 719 continue; 720 if (level > SINGLE) { 721 error = lfs_indirtrunc(ip, nlbn, nb, 722 (daddr_t)-1, level - 1, 723 &blkcount, &rblkcount, 724 lastsegp, bcp, p); 725 if (error) 726 allerror = error; 727 blocksreleased += blkcount; 728 real_released += rblkcount; 729 } 730 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 731 if (bap[i] > 0) 732 real_released += nblocks; 733 blocksreleased += nblocks; 734 } 735 736 /* 737 * Recursively free last partial block. 738 */ 739 if (level > SINGLE && lastbn >= 0) { 740 last = lastbn % factor; 741 nb = bap[i]; 742 if (nb != 0) { 743 error = lfs_indirtrunc(ip, nlbn, nb, 744 last, level - 1, &blkcount, 745 &rblkcount, lastsegp, bcp, p); 746 if (error) 747 allerror = error; 748 real_released += rblkcount; 749 blocksreleased += blkcount; 750 } 751 } 752 753 if (copy != NULL) { 754 FREE(copy, M_TEMP); 755 } else { 756 if (bp->b_flags & B_DELWRI) { 757 LFS_UNLOCK_BUF(bp); 758 fs->lfs_avail += btofsb(fs, bp->b_bcount); 759 wakeup(&fs->lfs_avail); 760 } 761 bp->b_flags |= B_INVAL; 762 brelse(bp); 763 } 764 765 *countp = blocksreleased; 766 *rcountp = real_released; 767 return (allerror); 768 } 769 770 /* 771 * Destroy any in core blocks past the truncation length. 772 * Inlined from vtruncbuf, so that lfs_avail could be updated. 773 * We take the seglock to prevent cleaning from occurring while we are 774 * invalidating blocks. 775 */ 776 static int 777 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 778 { 779 struct buf *bp, *nbp; 780 int s, error; 781 struct lfs *fs; 782 voff_t off; 783 784 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 785 simple_lock(&vp->v_interlock); 786 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 787 if (error) { 788 return error; 789 } 790 791 fs = VTOI(vp)->i_lfs; 792 s = splbio(); 793 794 restart: 795 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 796 nbp = LIST_NEXT(bp, b_vnbufs); 797 if (bp->b_lblkno < lbn) 798 continue; 799 simple_lock(&bp->b_interlock); 800 if (bp->b_flags & B_BUSY) { 801 bp->b_flags |= B_WANTED; 802 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 803 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 804 if (error) { 805 splx(s); 806 return (error); 807 } 808 goto restart; 809 } 810 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 811 if (bp->b_flags & B_DELWRI) { 812 bp->b_flags &= ~B_DELWRI; 813 fs->lfs_avail += btofsb(fs, bp->b_bcount); 814 wakeup(&fs->lfs_avail); 815 } 816 LFS_UNLOCK_BUF(bp); 817 simple_unlock(&bp->b_interlock); 818 brelse(bp); 819 } 820 821 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 822 nbp = LIST_NEXT(bp, b_vnbufs); 823 if (bp->b_lblkno < lbn) 824 continue; 825 simple_lock(&bp->b_interlock); 826 if (bp->b_flags & B_BUSY) { 827 bp->b_flags |= B_WANTED; 828 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 829 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 830 if (error) { 831 splx(s); 832 return (error); 833 } 834 goto restart; 835 } 836 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 837 if (bp->b_flags & B_DELWRI) { 838 bp->b_flags &= ~B_DELWRI; 839 fs->lfs_avail += btofsb(fs, bp->b_bcount); 840 wakeup(&fs->lfs_avail); 841 } 842 LFS_UNLOCK_BUF(bp); 843 simple_unlock(&bp->b_interlock); 844 brelse(bp); 845 } 846 847 splx(s); 848 849 return (0); 850 } 851 852
Indexes created Fri Sep 26 20:09:58 GMT 2025