lfs_inode.c revision 1.101
1 /* $NetBSD: lfs_inode.c,v 1.101 2006/04/08 00:26:34 perseant 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 1.101 2006/04/08 00:26:34 perseant 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/trace.h> 85 #include <sys/resourcevar.h> 86 87 #include <ufs/ufs/quota.h> 88 #include <ufs/ufs/inode.h> 89 #include <ufs/ufs/ufsmount.h> 90 #include <ufs/ufs/ufs_extern.h> 91 92 #include <ufs/lfs/lfs.h> 93 #include <ufs/lfs/lfs_extern.h> 94 95 static int lfs_update_seguse(struct lfs *, long, size_t); 96 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t, 97 daddr_t, int, long *, long *, long *, size_t *, 98 struct lwp *); 99 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *); 100 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int); 101 102 /* Search a block for a specific dinode. */ 103 struct ufs1_dinode * 104 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 105 { 106 struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data; 107 struct ufs1_dinode *ldip, *fin; 108 109 ASSERT_NO_SEGLOCK(fs); 110 /* 111 * Read the inode block backwards, since later versions of the 112 * inode will supercede earlier ones. Though it is unlikely, it is 113 * possible that the same inode will appear in the same inode block. 114 */ 115 fin = dip + INOPB(fs); 116 for (ldip = fin - 1; ldip >= dip; --ldip) 117 if (ldip->di_inumber == ino) 118 return (ldip); 119 120 printf("searched %d entries\n", (int)(fin - dip)); 121 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 122 dtosn(fs, fs->lfs_offset)); 123 printf("block is 0x%llx (seg %lld)\n", 124 (unsigned long long)dbtofsb(fs, bp->b_blkno), 125 (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno))); 126 127 return NULL; 128 } 129 130 int 131 lfs_update(struct vnode *vp, const struct timespec *acc, 132 const struct timespec *mod, int updflags) 133 { 134 struct inode *ip; 135 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 136 int s; 137 int flags; 138 139 ASSERT_NO_SEGLOCK(fs); 140 if (vp->v_mount->mnt_flag & MNT_RDONLY) 141 return (0); 142 ip = VTOI(vp); 143 144 /* 145 * If we are called from vinvalbuf, and the file's blocks have 146 * already been scheduled for writing, but the writes have not 147 * yet completed, lfs_vflush will not be called, and vinvalbuf 148 * will cause a panic. So, we must wait until any pending write 149 * for our inode completes, if we are called with UPDATE_WAIT set. 150 */ 151 s = splbio(); 152 simple_lock(&vp->v_interlock); 153 while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 154 WRITEINPROG(vp)) { 155 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 156 " (in progress)\n", ip->i_number)); 157 ltsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock); 158 } 159 simple_unlock(&vp->v_interlock); 160 splx(s); 161 LFS_ITIMES(ip, acc, mod, NULL); 162 if (updflags & UPDATE_CLOSE) 163 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 164 else 165 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 166 if (flags == 0) 167 return (0); 168 169 /* If sync, push back the vnode and any dirty blocks it may have. */ 170 if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 171 /* Avoid flushing VDIROP. */ 172 simple_lock(&fs->lfs_interlock); 173 ++fs->lfs_diropwait; 174 while (vp->v_flag & VDIROP) { 175 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 176 " (dirops)\n", ip->i_number)); 177 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 178 " 0x%x\n", vp->v_flag, ip->i_flag)); 179 if (fs->lfs_dirops == 0) 180 lfs_flush_fs(fs, SEGM_SYNC); 181 else 182 ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 183 0, &fs->lfs_interlock); 184 /* XXX KS - by falling out here, are we writing the vn 185 twice? */ 186 } 187 --fs->lfs_diropwait; 188 simple_unlock(&fs->lfs_interlock); 189 return lfs_vflush(vp); 190 } 191 return 0; 192 } 193 194 #define SINGLE 0 /* index of single indirect block */ 195 #define DOUBLE 1 /* index of double indirect block */ 196 #define TRIPLE 2 /* index of triple indirect block */ 197 /* 198 * Truncate the inode oip to at most length size, freeing the 199 * disk blocks. 200 */ 201 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */ 202 203 int 204 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, 205 struct ucred *cred, struct lwp *l) 206 { 207 struct genfs_node *gp = VTOG(ovp); 208 daddr_t lastblock; 209 struct inode *oip = VTOI(ovp); 210 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 211 /* XXX ondisk32 */ 212 int32_t newblks[NDADDR + NIADDR]; 213 struct lfs *fs; 214 struct buf *bp; 215 int offset, size, level; 216 long count, rcount, blocksreleased = 0, real_released = 0; 217 int i, nblocks; 218 int aflags, error, allerror = 0; 219 off_t osize; 220 long lastseg; 221 size_t bc; 222 int obufsize, odb; 223 int usepc; 224 struct ufsmount *ump = oip->i_ump; 225 226 if (ovp->v_type == VCHR || ovp->v_type == VBLK || 227 ovp->v_type == VFIFO || ovp->v_type == VSOCK) { 228 KASSERT(oip->i_size == 0); 229 return 0; 230 } 231 232 if (length < 0) 233 return (EINVAL); 234 235 /* 236 * Just return and not update modification times. 237 */ 238 if (oip->i_size == length) 239 return (0); 240 241 if (ovp->v_type == VLNK && 242 (oip->i_size < ump->um_maxsymlinklen || 243 (ump->um_maxsymlinklen == 0 && 244 oip->i_ffs1_blocks == 0))) { 245 #ifdef DIAGNOSTIC 246 if (length != 0) 247 panic("lfs_truncate: partial truncate of symlink"); 248 #endif 249 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 250 oip->i_size = oip->i_ffs1_size = 0; 251 oip->i_flag |= IN_CHANGE | IN_UPDATE; 252 return (lfs_update(ovp, NULL, NULL, 0)); 253 } 254 if (oip->i_size == length) { 255 oip->i_flag |= IN_CHANGE | IN_UPDATE; 256 return (lfs_update(ovp, NULL, NULL, 0)); 257 } 258 #ifdef QUOTA 259 if ((error = getinoquota(oip)) != 0) 260 return (error); 261 #endif 262 fs = oip->i_lfs; 263 lfs_imtime(fs); 264 osize = oip->i_size; 265 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 266 267 ASSERT_NO_SEGLOCK(fs); 268 /* 269 * Lengthen the size of the file. We must ensure that the 270 * last byte of the file is allocated. Since the smallest 271 * value of osize is 0, length will be at least 1. 272 */ 273 if (osize < length) { 274 if (length > ump->um_maxfilesize) 275 return (EFBIG); 276 aflags = B_CLRBUF; 277 if (ioflag & IO_SYNC) 278 aflags |= B_SYNC; 279 if (usepc) { 280 if (lblkno(fs, osize) < NDADDR && 281 lblkno(fs, osize) != lblkno(fs, length) && 282 blkroundup(fs, osize) != osize) { 283 off_t eob; 284 285 eob = blkroundup(fs, osize); 286 error = ufs_balloc_range(ovp, osize, 287 eob - osize, cred, aflags); 288 if (error) 289 return error; 290 if (ioflag & IO_SYNC) { 291 ovp->v_size = eob; 292 simple_lock(&ovp->v_interlock); 293 VOP_PUTPAGES(ovp, 294 trunc_page(osize & fs->lfs_bmask), 295 round_page(eob), 296 PGO_CLEANIT | PGO_SYNCIO); 297 } 298 } 299 error = ufs_balloc_range(ovp, length - 1, 1, cred, 300 aflags); 301 if (error) { 302 (void) lfs_truncate(ovp, osize, 303 ioflag & IO_SYNC, cred, l); 304 return error; 305 } 306 uvm_vnp_setsize(ovp, length); 307 oip->i_flag |= IN_CHANGE | IN_UPDATE; 308 KASSERT(ovp->v_size == oip->i_size); 309 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 310 return (lfs_update(ovp, NULL, NULL, 0)); 311 } else { 312 error = lfs_reserve(fs, ovp, NULL, 313 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 314 if (error) 315 return (error); 316 error = lfs_balloc(ovp, length - 1, 1, cred, 317 aflags, &bp); 318 lfs_reserve(fs, ovp, NULL, 319 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 320 if (error) 321 return (error); 322 oip->i_ffs1_size = oip->i_size = length; 323 uvm_vnp_setsize(ovp, length); 324 (void) VOP_BWRITE(bp); 325 oip->i_flag |= IN_CHANGE | IN_UPDATE; 326 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 327 return (lfs_update(ovp, NULL, NULL, 0)); 328 } 329 } 330 331 if ((error = lfs_reserve(fs, ovp, NULL, 332 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 333 return (error); 334 335 /* 336 * Shorten the size of the file. If the file is not being 337 * truncated to a block boundary, the contents of the 338 * partial block following the end of the file must be 339 * zero'ed in case it ever becomes accessible again because 340 * of subsequent file growth. Directories however are not 341 * zero'ed as they should grow back initialized to empty. 342 */ 343 offset = blkoff(fs, length); 344 lastseg = -1; 345 bc = 0; 346 347 if (ovp != fs->lfs_ivnode) 348 lfs_seglock(fs, SEGM_PROT); 349 if (offset == 0) { 350 oip->i_size = oip->i_ffs1_size = length; 351 } else if (!usepc) { 352 lbn = lblkno(fs, length); 353 aflags = B_CLRBUF; 354 if (ioflag & IO_SYNC) 355 aflags |= B_SYNC; 356 error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp); 357 if (error) { 358 lfs_reserve(fs, ovp, NULL, 359 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 360 goto errout; 361 } 362 obufsize = bp->b_bufsize; 363 odb = btofsb(fs, bp->b_bcount); 364 oip->i_size = oip->i_ffs1_size = length; 365 size = blksize(fs, oip, lbn); 366 if (ovp->v_type != VDIR) 367 memset((char *)bp->b_data + offset, 0, 368 (u_int)(size - offset)); 369 allocbuf(bp, size, 1); 370 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) { 371 simple_lock(&lfs_subsys_lock); 372 locked_queue_bytes -= obufsize - bp->b_bufsize; 373 simple_unlock(&lfs_subsys_lock); 374 } 375 if (bp->b_flags & B_DELWRI) 376 fs->lfs_avail += odb - btofsb(fs, size); 377 (void) VOP_BWRITE(bp); 378 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 379 /* 380 * When truncating a regular file down to a non-block-aligned 381 * size, we must zero the part of last block which is past 382 * the new EOF. We must synchronously flush the zeroed pages 383 * to disk since the new pages will be invalidated as soon 384 * as we inform the VM system of the new, smaller size. 385 * We must do this before acquiring the GLOCK, since fetching 386 * the pages will acquire the GLOCK internally. 387 * So there is a window where another thread could see a whole 388 * zeroed page past EOF, but that's life. 389 */ 390 daddr_t xlbn; 391 voff_t eoz; 392 393 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 394 error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags); 395 if (error) { 396 lfs_reserve(fs, ovp, NULL, 397 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 398 goto errout; 399 } 400 xlbn = lblkno(fs, length); 401 size = blksize(fs, oip, xlbn); 402 eoz = MIN(lblktosize(fs, xlbn) + size, osize); 403 uvm_vnp_zerorange(ovp, length, eoz - length); 404 if (round_page(eoz) > round_page(length)) { 405 simple_lock(&ovp->v_interlock); 406 error = VOP_PUTPAGES(ovp, round_page(length), 407 round_page(eoz), 408 PGO_CLEANIT | PGO_DEACTIVATE | 409 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 410 if (error) { 411 lfs_reserve(fs, ovp, NULL, 412 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 413 goto errout; 414 } 415 } 416 } 417 418 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 419 420 oip->i_size = oip->i_ffs1_size = length; 421 uvm_vnp_setsize(ovp, length); 422 /* 423 * Calculate index into inode's block list of 424 * last direct and indirect blocks (if any) 425 * which we want to keep. Lastblock is -1 when 426 * the file is truncated to 0. 427 */ 428 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 429 lastiblock[SINGLE] = lastblock - NDADDR; 430 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 431 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 432 nblocks = btofsb(fs, fs->lfs_bsize); 433 /* 434 * Record changed file and block pointers before we start 435 * freeing blocks. lastiblock values are also normalized to -1 436 * for calls to lfs_indirtrunc below. 437 */ 438 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 439 for (level = TRIPLE; level >= SINGLE; level--) 440 if (lastiblock[level] < 0) { 441 newblks[NDADDR+level] = 0; 442 lastiblock[level] = -1; 443 } 444 for (i = NDADDR - 1; i > lastblock; i--) 445 newblks[i] = 0; 446 447 oip->i_size = oip->i_ffs1_size = osize; 448 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 449 if (error && !allerror) 450 allerror = error; 451 452 /* 453 * Indirect blocks first. 454 */ 455 indir_lbn[SINGLE] = -NDADDR; 456 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 457 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 458 for (level = TRIPLE; level >= SINGLE; level--) { 459 bn = oip->i_ffs1_ib[level]; 460 if (bn != 0) { 461 error = lfs_indirtrunc(oip, indir_lbn[level], 462 bn, lastiblock[level], 463 level, &count, &rcount, 464 &lastseg, &bc, l); 465 if (error) 466 allerror = error; 467 real_released += rcount; 468 blocksreleased += count; 469 if (lastiblock[level] < 0) { 470 if (oip->i_ffs1_ib[level] > 0) 471 real_released += nblocks; 472 blocksreleased += nblocks; 473 oip->i_ffs1_ib[level] = 0; 474 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 475 lfs_deregister_block(ovp, bn); 476 } 477 } 478 if (lastiblock[level] >= 0) 479 goto done; 480 } 481 482 /* 483 * All whole direct blocks or frags. 484 */ 485 for (i = NDADDR - 1; i > lastblock; i--) { 486 long bsize, obsize; 487 488 bn = oip->i_ffs1_db[i]; 489 if (bn == 0) 490 continue; 491 bsize = blksize(fs, oip, i); 492 if (oip->i_ffs1_db[i] > 0) { 493 /* Check for fragment size changes */ 494 obsize = oip->i_lfs_fragsize[i]; 495 real_released += btofsb(fs, obsize); 496 oip->i_lfs_fragsize[i] = 0; 497 } else 498 obsize = 0; 499 blocksreleased += btofsb(fs, bsize); 500 oip->i_ffs1_db[i] = 0; 501 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 502 lfs_deregister_block(ovp, bn); 503 } 504 if (lastblock < 0) 505 goto done; 506 507 /* 508 * Finally, look for a change in size of the 509 * last direct block; release any frags. 510 */ 511 bn = oip->i_ffs1_db[lastblock]; 512 if (bn != 0) { 513 long oldspace, newspace; 514 #if 0 515 long olddspace; 516 #endif 517 518 /* 519 * Calculate amount of space we're giving 520 * back as old block size minus new block size. 521 */ 522 oldspace = blksize(fs, oip, lastblock); 523 #if 0 524 olddspace = oip->i_lfs_fragsize[lastblock]; 525 #endif 526 527 oip->i_size = oip->i_ffs1_size = length; 528 newspace = blksize(fs, oip, lastblock); 529 if (newspace == 0) 530 panic("itrunc: newspace"); 531 if (oldspace - newspace > 0) { 532 blocksreleased += btofsb(fs, oldspace - newspace); 533 } 534 #if 0 535 if (bn > 0 && olddspace - newspace > 0) { 536 /* No segment accounting here, just vnode */ 537 real_released += btofsb(fs, olddspace - newspace); 538 } 539 #endif 540 } 541 542 done: 543 /* Finish segment accounting corrections */ 544 lfs_update_seguse(fs, lastseg, bc); 545 #ifdef DIAGNOSTIC 546 for (level = SINGLE; level <= TRIPLE; level++) 547 if ((newblks[NDADDR + level] == 0) != 548 (oip->i_ffs1_ib[level]) == 0) { 549 panic("lfs itrunc1"); 550 } 551 for (i = 0; i < NDADDR; i++) 552 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 553 panic("lfs itrunc2"); 554 } 555 if (length == 0 && 556 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 557 panic("lfs itrunc3"); 558 #endif /* DIAGNOSTIC */ 559 /* 560 * Put back the real size. 561 */ 562 oip->i_size = oip->i_ffs1_size = length; 563 oip->i_lfs_effnblks -= blocksreleased; 564 oip->i_ffs1_blocks -= real_released; 565 simple_lock(&fs->lfs_interlock); 566 fs->lfs_bfree += blocksreleased; 567 simple_unlock(&fs->lfs_interlock); 568 #ifdef DIAGNOSTIC 569 if (oip->i_size == 0 && 570 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 571 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 572 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 573 panic("lfs_truncate: persistent blocks"); 574 } 575 #endif 576 577 /* 578 * If we truncated to zero, take us off the paging queue. 579 */ 580 simple_lock(&fs->lfs_interlock); 581 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 582 oip->i_flags &= ~IN_PAGING; 583 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 584 } 585 simple_unlock(&fs->lfs_interlock); 586 587 oip->i_flag |= IN_CHANGE; 588 #ifdef QUOTA 589 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 590 #endif 591 lfs_reserve(fs, ovp, NULL, 592 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 593 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 594 errout: 595 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 596 if (ovp != fs->lfs_ivnode) 597 lfs_segunlock(fs); 598 return (allerror ? allerror : error); 599 } 600 601 /* Update segment and avail usage information when removing a block. */ 602 static int 603 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 604 size_t *num) 605 { 606 long seg; 607 int error = 0; 608 609 ASSERT_SEGLOCK(fs); 610 bsize = fragroundup(fs, bsize); 611 if (daddr > 0) { 612 if (*lastseg != (seg = dtosn(fs, daddr))) { 613 error = lfs_update_seguse(fs, *lastseg, *num); 614 *num = bsize; 615 *lastseg = seg; 616 } else 617 *num += bsize; 618 } 619 620 return error; 621 } 622 623 /* Finish the accounting updates for a segment. */ 624 static int 625 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 626 { 627 SEGUSE *sup; 628 struct buf *bp; 629 630 ASSERT_SEGLOCK(fs); 631 if (lastseg < 0 || num == 0) 632 return 0; 633 634 LFS_SEGENTRY(sup, fs, lastseg, bp); 635 if (num > sup->su_nbytes) { 636 printf("lfs_truncate: segment %ld short by %ld\n", 637 lastseg, (long)num - sup->su_nbytes); 638 panic("lfs_truncate: negative bytes"); 639 sup->su_nbytes = num; 640 } 641 sup->su_nbytes -= num; 642 LFS_WRITESEGENTRY(sup, fs, lastseg, bp); 643 644 return 0; 645 } 646 647 /* 648 * Release blocks associated with the inode ip and stored in the indirect 649 * block bn. Blocks are free'd in LIFO order up to (but not including) 650 * lastbn. If level is greater than SINGLE, the block is an indirect block 651 * and recursive calls to indirtrunc must be used to cleanse other indirect 652 * blocks. 653 * 654 * NB: triple indirect blocks are untested. 655 */ 656 static int 657 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 658 daddr_t lastbn, int level, long *countp, 659 long *rcountp, long *lastsegp, size_t *bcp, struct lwp *l) 660 { 661 int i; 662 struct buf *bp; 663 struct lfs *fs = ip->i_lfs; 664 int32_t *bap; /* XXX ondisk32 */ 665 struct vnode *vp; 666 daddr_t nb, nlbn, last; 667 int32_t *copy = NULL; /* XXX ondisk32 */ 668 long blkcount, rblkcount, factor; 669 int nblocks, blocksreleased = 0, real_released = 0; 670 int error = 0, allerror = 0; 671 672 ASSERT_SEGLOCK(fs); 673 /* 674 * Calculate index in current block of last 675 * block to be kept. -1 indicates the entire 676 * block so we need not calculate the index. 677 */ 678 factor = 1; 679 for (i = SINGLE; i < level; i++) 680 factor *= NINDIR(fs); 681 last = lastbn; 682 if (lastbn > 0) 683 last /= factor; 684 nblocks = btofsb(fs, fs->lfs_bsize); 685 /* 686 * Get buffer of block pointers, zero those entries corresponding 687 * to blocks to be free'd, and update on disk copy first. Since 688 * double(triple) indirect before single(double) indirect, calls 689 * to bmap on these blocks will fail. However, we already have 690 * the on disk address, so we have to set the b_blkno field 691 * explicitly instead of letting bread do everything for us. 692 */ 693 vp = ITOV(ip); 694 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 695 if (bp->b_flags & (B_DONE | B_DELWRI)) { 696 /* Braces must be here in case trace evaluates to nothing. */ 697 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 698 } else { 699 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 700 l->l_proc->p_stats->p_ru.ru_inblock++; /* pay for read */ 701 bp->b_flags |= B_READ; 702 if (bp->b_bcount > bp->b_bufsize) 703 panic("lfs_indirtrunc: bad buffer size"); 704 bp->b_blkno = fsbtodb(fs, dbn); 705 VOP_STRATEGY(vp, bp); 706 error = biowait(bp); 707 } 708 if (error) { 709 brelse(bp); 710 *countp = *rcountp = 0; 711 return (error); 712 } 713 714 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 715 if (lastbn >= 0) { 716 copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK); 717 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 718 memset((caddr_t)&bap[last + 1], 0, 719 /* XXX ondisk32 */ 720 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 721 error = VOP_BWRITE(bp); 722 if (error) 723 allerror = error; 724 bap = copy; 725 } 726 727 /* 728 * Recursively free totally unused blocks. 729 */ 730 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 731 i--, nlbn += factor) { 732 nb = bap[i]; 733 if (nb == 0) 734 continue; 735 if (level > SINGLE) { 736 error = lfs_indirtrunc(ip, nlbn, nb, 737 (daddr_t)-1, level - 1, 738 &blkcount, &rblkcount, 739 lastsegp, bcp, l); 740 if (error) 741 allerror = error; 742 blocksreleased += blkcount; 743 real_released += rblkcount; 744 } 745 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 746 if (bap[i] > 0) 747 real_released += nblocks; 748 blocksreleased += nblocks; 749 } 750 751 /* 752 * Recursively free last partial block. 753 */ 754 if (level > SINGLE && lastbn >= 0) { 755 last = lastbn % factor; 756 nb = bap[i]; 757 if (nb != 0) { 758 error = lfs_indirtrunc(ip, nlbn, nb, 759 last, level - 1, &blkcount, 760 &rblkcount, lastsegp, bcp, l); 761 if (error) 762 allerror = error; 763 real_released += rblkcount; 764 blocksreleased += blkcount; 765 } 766 } 767 768 if (copy != NULL) { 769 lfs_free(fs, copy, LFS_NB_IBLOCK); 770 } else { 771 if (bp->b_flags & B_DELWRI) { 772 LFS_UNLOCK_BUF(bp); 773 fs->lfs_avail += btofsb(fs, bp->b_bcount); 774 wakeup(&fs->lfs_avail); 775 } 776 bp->b_flags |= B_INVAL; 777 brelse(bp); 778 } 779 780 *countp = blocksreleased; 781 *rcountp = real_released; 782 return (allerror); 783 } 784 785 /* 786 * Destroy any in core blocks past the truncation length. 787 * Inlined from vtruncbuf, so that lfs_avail could be updated. 788 * We take the seglock to prevent cleaning from occurring while we are 789 * invalidating blocks. 790 */ 791 static int 792 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 793 { 794 struct buf *bp, *nbp; 795 int s, error; 796 struct lfs *fs; 797 voff_t off; 798 799 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 800 simple_lock(&vp->v_interlock); 801 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 802 if (error) 803 return error; 804 805 fs = VTOI(vp)->i_lfs; 806 s = splbio(); 807 808 ASSERT_SEGLOCK(fs); 809 restart: 810 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 811 nbp = LIST_NEXT(bp, b_vnbufs); 812 if (bp->b_lblkno < lbn) 813 continue; 814 simple_lock(&bp->b_interlock); 815 if (bp->b_flags & B_BUSY) { 816 bp->b_flags |= B_WANTED; 817 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 818 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 819 if (error) { 820 splx(s); 821 return (error); 822 } 823 goto restart; 824 } 825 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 826 if (bp->b_flags & B_DELWRI) { 827 bp->b_flags &= ~B_DELWRI; 828 fs->lfs_avail += btofsb(fs, bp->b_bcount); 829 wakeup(&fs->lfs_avail); 830 } 831 LFS_UNLOCK_BUF(bp); 832 simple_unlock(&bp->b_interlock); 833 brelse(bp); 834 } 835 836 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 837 nbp = LIST_NEXT(bp, b_vnbufs); 838 if (bp->b_lblkno < lbn) 839 continue; 840 simple_lock(&bp->b_interlock); 841 if (bp->b_flags & B_BUSY) { 842 bp->b_flags |= B_WANTED; 843 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 844 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 845 if (error) { 846 splx(s); 847 return (error); 848 } 849 goto restart; 850 } 851 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 852 if (bp->b_flags & B_DELWRI) { 853 bp->b_flags &= ~B_DELWRI; 854 fs->lfs_avail += btofsb(fs, bp->b_bcount); 855 wakeup(&fs->lfs_avail); 856 } 857 LFS_UNLOCK_BUF(bp); 858 simple_unlock(&bp->b_interlock); 859 brelse(bp); 860 } 861 862 splx(s); 863 864 return (0); 865 } 866 867
Indexes created Sat Sep 27 14:09:57 GMT 2025