lfs_inode.c revision 1.97.2.1
1 /* $NetBSD: lfs_inode.c,v 1.97.2.1 2005/10/20 03:00:30 yamt 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 1.97.2.1 2005/10/20 03:00:30 yamt 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 proc *); 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 + UFS_BALLOC == 2 + 2*NIADDR times */ 202 203 int 204 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, 205 struct ucred *cred, struct proc *p) 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 (length < 0) 227 return (EINVAL); 228 229 /* 230 * Just return and not update modification times. 231 */ 232 if (oip->i_size == length) 233 return (0); 234 235 if (ovp->v_type == VLNK && 236 (oip->i_size < ump->um_maxsymlinklen || 237 (ump->um_maxsymlinklen == 0 && 238 oip->i_ffs1_blocks == 0))) { 239 #ifdef DIAGNOSTIC 240 if (length != 0) 241 panic("lfs_truncate: partial truncate of symlink"); 242 #endif 243 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 244 oip->i_size = oip->i_ffs1_size = 0; 245 oip->i_flag |= IN_CHANGE | IN_UPDATE; 246 return (UFS_UPDATE(ovp, NULL, NULL, 0)); 247 } 248 if (oip->i_size == length) { 249 oip->i_flag |= IN_CHANGE | IN_UPDATE; 250 return (UFS_UPDATE(ovp, NULL, NULL, 0)); 251 } 252 #ifdef QUOTA 253 if ((error = getinoquota(oip)) != 0) 254 return (error); 255 #endif 256 fs = oip->i_lfs; 257 lfs_imtime(fs); 258 osize = oip->i_size; 259 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 260 261 ASSERT_NO_SEGLOCK(fs); 262 /* 263 * Lengthen the size of the file. We must ensure that the 264 * last byte of the file is allocated. Since the smallest 265 * value of osize is 0, length will be at least 1. 266 */ 267 if (osize < length) { 268 if (length > ump->um_maxfilesize) 269 return (EFBIG); 270 aflags = B_CLRBUF; 271 if (ioflag & IO_SYNC) 272 aflags |= B_SYNC; 273 if (usepc) { 274 if (lblkno(fs, osize) < NDADDR && 275 lblkno(fs, osize) != lblkno(fs, length) && 276 blkroundup(fs, osize) != osize) { 277 off_t eob; 278 279 eob = blkroundup(fs, osize); 280 error = ufs_balloc_range(ovp, osize, 281 eob - osize, cred, aflags); 282 if (error) 283 return error; 284 if (ioflag & IO_SYNC) { 285 ovp->v_size = eob; 286 simple_lock(&ovp->v_interlock); 287 VOP_PUTPAGES(ovp, 288 trunc_page(osize & fs->lfs_bmask), 289 round_page(eob), 290 PGO_CLEANIT | PGO_SYNCIO); 291 } 292 } 293 error = ufs_balloc_range(ovp, length - 1, 1, cred, 294 aflags); 295 if (error) { 296 (void) UFS_TRUNCATE(ovp, osize, 297 ioflag & IO_SYNC, cred, p); 298 return error; 299 } 300 uvm_vnp_setsize(ovp, length); 301 oip->i_flag |= IN_CHANGE | IN_UPDATE; 302 KASSERT(ovp->v_size == oip->i_size); 303 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 304 return (UFS_UPDATE(ovp, NULL, NULL, 0)); 305 } else { 306 error = lfs_reserve(fs, ovp, NULL, 307 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 308 if (error) 309 return (error); 310 error = UFS_BALLOC(ovp, length - 1, 1, cred, 311 aflags, &bp); 312 lfs_reserve(fs, ovp, NULL, 313 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 314 if (error) 315 return (error); 316 oip->i_ffs1_size = oip->i_size = length; 317 uvm_vnp_setsize(ovp, length); 318 (void) VOP_BWRITE(bp); 319 oip->i_flag |= IN_CHANGE | IN_UPDATE; 320 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 321 return (UFS_UPDATE(ovp, NULL, NULL, 0)); 322 } 323 } 324 325 if ((error = lfs_reserve(fs, ovp, NULL, 326 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 327 return (error); 328 329 /* 330 * Shorten the size of the file. If the file is not being 331 * truncated to a block boundary, the contents of the 332 * partial block following the end of the file must be 333 * zero'ed in case it ever becomes accessible again because 334 * of subsequent file growth. Directories however are not 335 * zero'ed as they should grow back initialized to empty. 336 */ 337 offset = blkoff(fs, length); 338 lastseg = -1; 339 bc = 0; 340 341 if (ovp != fs->lfs_ivnode) 342 lfs_seglock(fs, SEGM_PROT); 343 if (offset == 0) { 344 oip->i_size = oip->i_ffs1_size = length; 345 } else if (!usepc) { 346 lbn = lblkno(fs, length); 347 aflags = B_CLRBUF; 348 if (ioflag & IO_SYNC) 349 aflags |= B_SYNC; 350 error = UFS_BALLOC(ovp, length - 1, 1, cred, aflags, &bp); 351 if (error) { 352 lfs_reserve(fs, ovp, NULL, 353 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 354 goto errout; 355 } 356 obufsize = bp->b_bufsize; 357 odb = btofsb(fs, bp->b_bcount); 358 oip->i_size = oip->i_ffs1_size = length; 359 size = blksize(fs, oip, lbn); 360 if (ovp->v_type != VDIR) 361 memset((char *)bp->b_data + offset, 0, 362 (u_int)(size - offset)); 363 allocbuf(bp, size, 1); 364 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) { 365 simple_lock(&lfs_subsys_lock); 366 locked_queue_bytes -= obufsize - bp->b_bufsize; 367 simple_unlock(&lfs_subsys_lock); 368 } 369 if (bp->b_flags & B_DELWRI) 370 fs->lfs_avail += odb - btofsb(fs, size); 371 (void) VOP_BWRITE(bp); 372 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 373 /* 374 * When truncating a regular file down to a non-block-aligned 375 * size, we must zero the part of last block which is past 376 * the new EOF. We must synchronously flush the zeroed pages 377 * to disk since the new pages will be invalidated as soon 378 * as we inform the VM system of the new, smaller size. 379 * We must do this before acquiring the GLOCK, since fetching 380 * the pages will acquire the GLOCK internally. 381 * So there is a window where another thread could see a whole 382 * zeroed page past EOF, but that's life. 383 */ 384 daddr_t xlbn; 385 voff_t eoz; 386 387 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 388 error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags); 389 if (error) { 390 lfs_reserve(fs, ovp, NULL, 391 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 392 goto errout; 393 } 394 xlbn = lblkno(fs, length); 395 size = blksize(fs, oip, xlbn); 396 eoz = MIN(lblktosize(fs, xlbn) + size, osize); 397 uvm_vnp_zerorange(ovp, length, eoz - length); 398 if (round_page(eoz) > round_page(length)) { 399 simple_lock(&ovp->v_interlock); 400 error = VOP_PUTPAGES(ovp, round_page(length), 401 round_page(eoz), 402 PGO_CLEANIT | PGO_DEACTIVATE | 403 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 404 if (error) { 405 lfs_reserve(fs, ovp, NULL, 406 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 407 goto errout; 408 } 409 } 410 } 411 412 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 413 414 oip->i_size = oip->i_ffs1_size = length; 415 uvm_vnp_setsize(ovp, length); 416 /* 417 * Calculate index into inode's block list of 418 * last direct and indirect blocks (if any) 419 * which we want to keep. Lastblock is -1 when 420 * the file is truncated to 0. 421 */ 422 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 423 lastiblock[SINGLE] = lastblock - NDADDR; 424 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 425 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 426 nblocks = btofsb(fs, fs->lfs_bsize); 427 /* 428 * Record changed file and block pointers before we start 429 * freeing blocks. lastiblock values are also normalized to -1 430 * for calls to lfs_indirtrunc below. 431 */ 432 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 433 for (level = TRIPLE; level >= SINGLE; level--) 434 if (lastiblock[level] < 0) { 435 newblks[NDADDR+level] = 0; 436 lastiblock[level] = -1; 437 } 438 for (i = NDADDR - 1; i > lastblock; i--) 439 newblks[i] = 0; 440 441 oip->i_size = oip->i_ffs1_size = osize; 442 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 443 if (error && !allerror) 444 allerror = error; 445 446 /* 447 * Indirect blocks first. 448 */ 449 indir_lbn[SINGLE] = -NDADDR; 450 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 451 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 452 for (level = TRIPLE; level >= SINGLE; level--) { 453 bn = oip->i_ffs1_ib[level]; 454 if (bn != 0) { 455 error = lfs_indirtrunc(oip, indir_lbn[level], 456 bn, lastiblock[level], 457 level, &count, &rcount, 458 &lastseg, &bc, p); 459 if (error) 460 allerror = error; 461 real_released += rcount; 462 blocksreleased += count; 463 if (lastiblock[level] < 0) { 464 if (oip->i_ffs1_ib[level] > 0) 465 real_released += nblocks; 466 blocksreleased += nblocks; 467 oip->i_ffs1_ib[level] = 0; 468 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 469 lfs_deregister_block(ovp, bn); 470 } 471 } 472 if (lastiblock[level] >= 0) 473 goto done; 474 } 475 476 /* 477 * All whole direct blocks or frags. 478 */ 479 for (i = NDADDR - 1; i > lastblock; i--) { 480 long bsize, obsize; 481 482 bn = oip->i_ffs1_db[i]; 483 if (bn == 0) 484 continue; 485 bsize = blksize(fs, oip, i); 486 if (oip->i_ffs1_db[i] > 0) { 487 /* Check for fragment size changes */ 488 obsize = oip->i_lfs_fragsize[i]; 489 real_released += btofsb(fs, obsize); 490 oip->i_lfs_fragsize[i] = 0; 491 } else 492 obsize = 0; 493 blocksreleased += btofsb(fs, bsize); 494 oip->i_ffs1_db[i] = 0; 495 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 496 lfs_deregister_block(ovp, bn); 497 } 498 if (lastblock < 0) 499 goto done; 500 501 /* 502 * Finally, look for a change in size of the 503 * last direct block; release any frags. 504 */ 505 bn = oip->i_ffs1_db[lastblock]; 506 if (bn != 0) { 507 long oldspace, newspace; 508 #if 0 509 long olddspace; 510 #endif 511 512 /* 513 * Calculate amount of space we're giving 514 * back as old block size minus new block size. 515 */ 516 oldspace = blksize(fs, oip, lastblock); 517 #if 0 518 olddspace = oip->i_lfs_fragsize[lastblock]; 519 #endif 520 521 oip->i_size = oip->i_ffs1_size = length; 522 newspace = blksize(fs, oip, lastblock); 523 if (newspace == 0) 524 panic("itrunc: newspace"); 525 if (oldspace - newspace > 0) { 526 blocksreleased += btofsb(fs, oldspace - newspace); 527 } 528 #if 0 529 if (bn > 0 && olddspace - newspace > 0) { 530 /* No segment accounting here, just vnode */ 531 real_released += btofsb(fs, olddspace - newspace); 532 } 533 #endif 534 } 535 536 done: 537 /* Finish segment accounting corrections */ 538 lfs_update_seguse(fs, lastseg, bc); 539 #ifdef DIAGNOSTIC 540 for (level = SINGLE; level <= TRIPLE; level++) 541 if ((newblks[NDADDR + level] == 0) != 542 (oip->i_ffs1_ib[level]) == 0) { 543 panic("lfs itrunc1"); 544 } 545 for (i = 0; i < NDADDR; i++) 546 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 547 panic("lfs itrunc2"); 548 } 549 if (length == 0 && 550 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 551 panic("lfs itrunc3"); 552 #endif /* DIAGNOSTIC */ 553 /* 554 * Put back the real size. 555 */ 556 oip->i_size = oip->i_ffs1_size = length; 557 oip->i_lfs_effnblks -= blocksreleased; 558 oip->i_ffs1_blocks -= real_released; 559 simple_lock(&fs->lfs_interlock); 560 fs->lfs_bfree += blocksreleased; 561 simple_unlock(&fs->lfs_interlock); 562 #ifdef DIAGNOSTIC 563 if (oip->i_size == 0 && 564 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 565 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 566 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 567 panic("lfs_truncate: persistent blocks"); 568 } 569 #endif 570 oip->i_flag |= IN_CHANGE; 571 #ifdef QUOTA 572 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 573 #endif 574 lfs_reserve(fs, ovp, NULL, 575 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 576 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 577 errout: 578 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 579 if (ovp != fs->lfs_ivnode) 580 lfs_segunlock(fs); 581 return (allerror ? allerror : error); 582 } 583 584 /* Update segment and avail usage information when removing a block. */ 585 static int 586 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 587 size_t *num) 588 { 589 long seg; 590 int error = 0; 591 592 ASSERT_SEGLOCK(fs); 593 bsize = fragroundup(fs, bsize); 594 if (daddr > 0) { 595 if (*lastseg != (seg = dtosn(fs, daddr))) { 596 error = lfs_update_seguse(fs, *lastseg, *num); 597 *num = bsize; 598 *lastseg = seg; 599 } else 600 *num += bsize; 601 } 602 603 return error; 604 } 605 606 /* Finish the accounting updates for a segment. */ 607 static int 608 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 609 { 610 SEGUSE *sup; 611 struct buf *bp; 612 613 ASSERT_SEGLOCK(fs); 614 if (lastseg < 0 || num == 0) 615 return 0; 616 617 LFS_SEGENTRY(sup, fs, lastseg, bp); 618 if (num > sup->su_nbytes) { 619 printf("lfs_truncate: segment %ld short by %ld\n", 620 lastseg, (long)num - sup->su_nbytes); 621 panic("lfs_truncate: negative bytes"); 622 sup->su_nbytes = num; 623 } 624 sup->su_nbytes -= num; 625 LFS_WRITESEGENTRY(sup, fs, lastseg, bp); 626 627 return 0; 628 } 629 630 /* 631 * Release blocks associated with the inode ip and stored in the indirect 632 * block bn. Blocks are free'd in LIFO order up to (but not including) 633 * lastbn. If level is greater than SINGLE, the block is an indirect block 634 * and recursive calls to indirtrunc must be used to cleanse other indirect 635 * blocks. 636 * 637 * NB: triple indirect blocks are untested. 638 */ 639 static int 640 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 641 daddr_t lastbn, int level, long *countp, 642 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 643 { 644 int i; 645 struct buf *bp; 646 struct lfs *fs = ip->i_lfs; 647 int32_t *bap; /* XXX ondisk32 */ 648 struct vnode *vp; 649 daddr_t nb, nlbn, last; 650 int32_t *copy = NULL; /* XXX ondisk32 */ 651 long blkcount, rblkcount, factor; 652 int nblocks, blocksreleased = 0, real_released = 0; 653 int error = 0, allerror = 0; 654 655 ASSERT_SEGLOCK(fs); 656 /* 657 * Calculate index in current block of last 658 * block to be kept. -1 indicates the entire 659 * block so we need not calculate the index. 660 */ 661 factor = 1; 662 for (i = SINGLE; i < level; i++) 663 factor *= NINDIR(fs); 664 last = lastbn; 665 if (lastbn > 0) 666 last /= factor; 667 nblocks = btofsb(fs, fs->lfs_bsize); 668 /* 669 * Get buffer of block pointers, zero those entries corresponding 670 * to blocks to be free'd, and update on disk copy first. Since 671 * double(triple) indirect before single(double) indirect, calls 672 * to bmap on these blocks will fail. However, we already have 673 * the on disk address, so we have to set the b_blkno field 674 * explicitly instead of letting bread do everything for us. 675 */ 676 vp = ITOV(ip); 677 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 678 if (bp->b_flags & (B_DONE | B_DELWRI)) { 679 /* Braces must be here in case trace evaluates to nothing. */ 680 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 681 } else { 682 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 683 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 684 bp->b_flags |= B_READ; 685 if (bp->b_bcount > bp->b_bufsize) 686 panic("lfs_indirtrunc: bad buffer size"); 687 bp->b_blkno = fsbtodb(fs, dbn); 688 VOP_STRATEGY(vp, bp); 689 error = biowait(bp); 690 } 691 if (error) { 692 brelse(bp); 693 *countp = *rcountp = 0; 694 return (error); 695 } 696 697 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 698 if (lastbn >= 0) { 699 copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK); 700 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 701 memset((caddr_t)&bap[last + 1], 0, 702 /* XXX ondisk32 */ 703 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 704 error = VOP_BWRITE(bp); 705 if (error) 706 allerror = error; 707 bap = copy; 708 } 709 710 /* 711 * Recursively free totally unused blocks. 712 */ 713 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 714 i--, nlbn += factor) { 715 nb = bap[i]; 716 if (nb == 0) 717 continue; 718 if (level > SINGLE) { 719 error = lfs_indirtrunc(ip, nlbn, nb, 720 (daddr_t)-1, level - 1, 721 &blkcount, &rblkcount, 722 lastsegp, bcp, p); 723 if (error) 724 allerror = error; 725 blocksreleased += blkcount; 726 real_released += rblkcount; 727 } 728 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 729 if (bap[i] > 0) 730 real_released += nblocks; 731 blocksreleased += nblocks; 732 } 733 734 /* 735 * Recursively free last partial block. 736 */ 737 if (level > SINGLE && lastbn >= 0) { 738 last = lastbn % factor; 739 nb = bap[i]; 740 if (nb != 0) { 741 error = lfs_indirtrunc(ip, nlbn, nb, 742 last, level - 1, &blkcount, 743 &rblkcount, lastsegp, bcp, p); 744 if (error) 745 allerror = error; 746 real_released += rblkcount; 747 blocksreleased += blkcount; 748 } 749 } 750 751 if (copy != NULL) { 752 lfs_free(fs, copy, LFS_NB_IBLOCK); 753 } else { 754 if (bp->b_flags & B_DELWRI) { 755 LFS_UNLOCK_BUF(bp); 756 fs->lfs_avail += btofsb(fs, bp->b_bcount); 757 wakeup(&fs->lfs_avail); 758 } 759 bp->b_flags |= B_INVAL; 760 brelse(bp); 761 } 762 763 *countp = blocksreleased; 764 *rcountp = real_released; 765 return (allerror); 766 } 767 768 /* 769 * Destroy any in core blocks past the truncation length. 770 * Inlined from vtruncbuf, so that lfs_avail could be updated. 771 * We take the seglock to prevent cleaning from occurring while we are 772 * invalidating blocks. 773 */ 774 static int 775 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 776 { 777 struct buf *bp, *nbp; 778 int s, error; 779 struct lfs *fs; 780 voff_t off; 781 782 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 783 simple_lock(&vp->v_interlock); 784 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 785 if (error) 786 return error; 787 788 fs = VTOI(vp)->i_lfs; 789 s = splbio(); 790 791 ASSERT_SEGLOCK(fs); 792 restart: 793 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 794 nbp = LIST_NEXT(bp, b_vnbufs); 795 if (bp->b_lblkno < lbn) 796 continue; 797 simple_lock(&bp->b_interlock); 798 if (bp->b_flags & B_BUSY) { 799 bp->b_flags |= B_WANTED; 800 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 801 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 802 if (error) { 803 splx(s); 804 return (error); 805 } 806 goto restart; 807 } 808 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 809 if (bp->b_flags & B_DELWRI) { 810 bp->b_flags &= ~B_DELWRI; 811 fs->lfs_avail += btofsb(fs, bp->b_bcount); 812 wakeup(&fs->lfs_avail); 813 } 814 LFS_UNLOCK_BUF(bp); 815 simple_unlock(&bp->b_interlock); 816 brelse(bp); 817 } 818 819 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 820 nbp = LIST_NEXT(bp, b_vnbufs); 821 if (bp->b_lblkno < lbn) 822 continue; 823 simple_lock(&bp->b_interlock); 824 if (bp->b_flags & B_BUSY) { 825 bp->b_flags |= B_WANTED; 826 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 827 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 828 if (error) { 829 splx(s); 830 return (error); 831 } 832 goto restart; 833 } 834 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 835 if (bp->b_flags & B_DELWRI) { 836 bp->b_flags &= ~B_DELWRI; 837 fs->lfs_avail += btofsb(fs, bp->b_bcount); 838 wakeup(&fs->lfs_avail); 839 } 840 LFS_UNLOCK_BUF(bp); 841 simple_unlock(&bp->b_interlock); 842 brelse(bp); 843 } 844 845 splx(s); 846 847 return (0); 848 } 849 850
Indexes created Thu Oct 02 14:10:14 GMT 2025