lfs_inode.c revision 1.90
1 /* $NetBSD: lfs_inode.c,v 1.90 2005/03/08 00:18:19 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 1.90 2005/03/08 00:18:19 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/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 /* 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(void *v) 132 { 133 struct vop_update_args /* { 134 struct vnode *a_vp; 135 struct timespec *a_access; 136 struct timespec *a_modify; 137 int a_flags; 138 } */ *ap = v; 139 struct inode *ip; 140 struct vnode *vp = ap->a_vp; 141 struct timespec ts; 142 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 143 int s; 144 int flags; 145 146 if (vp->v_mount->mnt_flag & MNT_RDONLY) 147 return (0); 148 ip = VTOI(vp); 149 150 /* 151 * If we are called from vinvalbuf, and the file's blocks have 152 * already been scheduled for writing, but the writes have not 153 * yet completed, lfs_vflush will not be called, and vinvalbuf 154 * will cause a panic. So, we must wait until any pending write 155 * for our inode completes, if we are called with UPDATE_WAIT set. 156 */ 157 s = splbio(); 158 while ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 159 WRITEINPROG(vp)) { 160 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 161 " (in progress)\n", ip->i_number)); 162 tsleep(vp, (PRIBIO+1), "lfs_update", 0); 163 } 164 splx(s); 165 TIMEVAL_TO_TIMESPEC(&time, &ts); 166 LFS_ITIMES(ip, 167 ap->a_access ? ap->a_access : &ts, 168 ap->a_modify ? ap->a_modify : &ts, &ts); 169 if (ap->a_flags & UPDATE_CLOSE) 170 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 171 else 172 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 173 if (flags == 0) 174 return (0); 175 176 /* If sync, push back the vnode and any dirty blocks it may have. */ 177 if ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 178 /* Avoid flushing VDIROP. */ 179 ++fs->lfs_diropwait; 180 while (vp->v_flag & VDIROP) { 181 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 182 " (dirops)\n", ip->i_number)); 183 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 184 " 0x%x\n", vp->v_flag, ip->i_flag)); 185 if (fs->lfs_dirops == 0) 186 lfs_flush_fs(fs, SEGM_SYNC); 187 else 188 tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 189 0); 190 /* XXX KS - by falling out here, are we writing the vn 191 twice? */ 192 } 193 --fs->lfs_diropwait; 194 return lfs_vflush(vp); 195 } 196 return 0; 197 } 198 199 #define SINGLE 0 /* index of single indirect block */ 200 #define DOUBLE 1 /* index of double indirect block */ 201 #define TRIPLE 2 /* index of triple indirect block */ 202 /* 203 * Truncate the inode oip to at most length size, freeing the 204 * disk blocks. 205 */ 206 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */ 207 208 int 209 lfs_truncate(void *v) 210 { 211 struct vop_truncate_args /* { 212 struct vnode *a_vp; 213 off_t a_length; 214 int a_flags; 215 struct ucred *a_cred; 216 struct proc *a_p; 217 } */ *ap = v; 218 struct vnode *ovp = ap->a_vp; 219 struct genfs_node *gp = VTOG(ovp); 220 daddr_t lastblock; 221 struct inode *oip = VTOI(ovp); 222 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 223 /* XXX ondisk32 */ 224 int32_t newblks[NDADDR + NIADDR]; 225 off_t length = ap->a_length; 226 struct lfs *fs; 227 struct buf *bp; 228 int offset, size, level; 229 long count, rcount, blocksreleased = 0, real_released = 0; 230 int i, ioflag, nblocks; 231 int aflags, error, allerror = 0; 232 off_t osize; 233 long lastseg; 234 size_t bc; 235 int obufsize, odb; 236 int usepc; 237 struct ufsmount *ump = oip->i_ump; 238 239 if (length < 0) 240 return (EINVAL); 241 242 /* 243 * Just return and not update modification times. 244 */ 245 if (oip->i_size == length) 246 return (0); 247 248 if (ovp->v_type == VLNK && 249 (oip->i_size < ump->um_maxsymlinklen || 250 (ump->um_maxsymlinklen == 0 && 251 oip->i_ffs1_blocks == 0))) { 252 #ifdef DIAGNOSTIC 253 if (length != 0) 254 panic("lfs_truncate: partial truncate of symlink"); 255 #endif 256 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 257 oip->i_size = oip->i_ffs1_size = 0; 258 oip->i_flag |= IN_CHANGE | IN_UPDATE; 259 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 260 } 261 if (oip->i_size == length) { 262 oip->i_flag |= IN_CHANGE | IN_UPDATE; 263 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 264 } 265 #ifdef QUOTA 266 if ((error = getinoquota(oip)) != 0) 267 return (error); 268 #endif 269 fs = oip->i_lfs; 270 lfs_imtime(fs); 271 osize = oip->i_size; 272 ioflag = ap->a_flags; 273 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 274 275 /* 276 * Lengthen the size of the file. We must ensure that the 277 * last byte of the file is allocated. Since the smallest 278 * value of osize is 0, length will be at least 1. 279 */ 280 if (osize < length) { 281 if (length > ump->um_maxfilesize) 282 return (EFBIG); 283 aflags = B_CLRBUF; 284 if (ioflag & IO_SYNC) 285 aflags |= B_SYNC; 286 if (usepc) { 287 if (lblkno(fs, osize) < NDADDR && 288 lblkno(fs, osize) != lblkno(fs, length) && 289 blkroundup(fs, osize) != osize) { 290 off_t eob; 291 292 eob = blkroundup(fs, osize); 293 error = ufs_balloc_range(ovp, osize, 294 eob - osize, ap->a_cred, aflags); 295 if (error) 296 return error; 297 if (ioflag & IO_SYNC) { 298 ovp->v_size = eob; 299 simple_lock(&ovp->v_interlock); 300 VOP_PUTPAGES(ovp, 301 trunc_page(osize & fs->lfs_bmask), 302 round_page(eob), 303 PGO_CLEANIT | PGO_SYNCIO); 304 } 305 } 306 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 307 aflags); 308 if (error) { 309 (void) VOP_TRUNCATE(ovp, osize, 310 ioflag & IO_SYNC, 311 ap->a_cred, ap->a_p); 312 return error; 313 } 314 uvm_vnp_setsize(ovp, length); 315 oip->i_flag |= IN_CHANGE | IN_UPDATE; 316 KASSERT(ovp->v_size == oip->i_size); 317 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 318 } else { 319 error = lfs_reserve(fs, ovp, NULL, 320 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 321 if (error) 322 return (error); 323 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, 324 aflags, &bp); 325 lfs_reserve(fs, ovp, NULL, 326 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 327 if (error) 328 return (error); 329 oip->i_ffs1_size = oip->i_size = length; 330 uvm_vnp_setsize(ovp, length); 331 (void) VOP_BWRITE(bp); 332 oip->i_flag |= IN_CHANGE | IN_UPDATE; 333 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 334 } 335 } 336 337 if ((error = lfs_reserve(fs, ovp, NULL, 338 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 339 return (error); 340 341 /* 342 * Shorten the size of the file. If the file is not being 343 * truncated to a block boundary, the contents of the 344 * partial block following the end of the file must be 345 * zero'ed in case it ever becomes accessible again because 346 * of subsequent file growth. Directories however are not 347 * zero'ed as they should grow back initialized to empty. 348 */ 349 offset = blkoff(fs, length); 350 lastseg = -1; 351 bc = 0; 352 353 lfs_seglock(fs, SEGM_PROT); 354 if (offset == 0) { 355 oip->i_size = oip->i_ffs1_size = length; 356 } else if (!usepc) { 357 lbn = lblkno(fs, length); 358 aflags = B_CLRBUF; 359 if (ioflag & IO_SYNC) 360 aflags |= B_SYNC; 361 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 362 if (error) { 363 lfs_reserve(fs, ovp, NULL, 364 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 365 goto errout; 366 } 367 obufsize = bp->b_bufsize; 368 odb = btofsb(fs, bp->b_bcount); 369 oip->i_size = oip->i_ffs1_size = length; 370 size = blksize(fs, oip, lbn); 371 if (ovp->v_type != VDIR) 372 memset((char *)bp->b_data + offset, 0, 373 (u_int)(size - offset)); 374 allocbuf(bp, size, 1); 375 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 376 locked_queue_bytes -= obufsize - bp->b_bufsize; 377 if (bp->b_flags & B_DELWRI) 378 fs->lfs_avail += odb - btofsb(fs, size); 379 (void) VOP_BWRITE(bp); 380 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 381 /* 382 * When truncating a regular file down to a non-block-aligned 383 * size, we must zero the part of last block which is past 384 * the new EOF. We must synchronously flush the zeroed pages 385 * to disk since the new pages will be invalidated as soon 386 * as we inform the VM system of the new, smaller size. 387 * We must do this before acquiring the GLOCK, since fetching 388 * the pages will acquire the GLOCK internally. 389 * So there is a window where another thread could see a whole 390 * zeroed page past EOF, but that's life. 391 */ 392 daddr_t lbn; 393 voff_t eoz; 394 395 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 396 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 397 aflags); 398 if (error) { 399 lfs_reserve(fs, ovp, NULL, 400 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 401 goto errout; 402 } 403 lbn = lblkno(fs, length); 404 size = blksize(fs, oip, lbn); 405 eoz = MIN(lblktosize(fs, lbn) + size, osize); 406 uvm_vnp_zerorange(ovp, length, eoz - length); 407 if (round_page(eoz) > round_page(length)) { 408 simple_lock(&ovp->v_interlock); 409 error = VOP_PUTPAGES(ovp, round_page(length), 410 round_page(eoz), 411 PGO_CLEANIT | PGO_DEACTIVATE | 412 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 413 if (error) { 414 lfs_reserve(fs, ovp, NULL, 415 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 416 goto errout; 417 } 418 } 419 } 420 421 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 422 423 oip->i_size = oip->i_ffs1_size = length; 424 uvm_vnp_setsize(ovp, length); 425 /* 426 * Calculate index into inode's block list of 427 * last direct and indirect blocks (if any) 428 * which we want to keep. Lastblock is -1 when 429 * the file is truncated to 0. 430 */ 431 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 432 lastiblock[SINGLE] = lastblock - NDADDR; 433 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 434 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 435 nblocks = btofsb(fs, fs->lfs_bsize); 436 /* 437 * Record changed file and block pointers before we start 438 * freeing blocks. lastiblock values are also normalized to -1 439 * for calls to lfs_indirtrunc below. 440 */ 441 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 442 for (level = TRIPLE; level >= SINGLE; level--) 443 if (lastiblock[level] < 0) { 444 newblks[NDADDR+level] = 0; 445 lastiblock[level] = -1; 446 } 447 for (i = NDADDR - 1; i > lastblock; i--) 448 newblks[i] = 0; 449 450 oip->i_size = oip->i_ffs1_size = osize; 451 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 452 if (error && !allerror) 453 allerror = error; 454 455 /* 456 * Indirect blocks first. 457 */ 458 indir_lbn[SINGLE] = -NDADDR; 459 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 460 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 461 for (level = TRIPLE; level >= SINGLE; level--) { 462 bn = oip->i_ffs1_ib[level]; 463 if (bn != 0) { 464 error = lfs_indirtrunc(oip, indir_lbn[level], 465 bn, lastiblock[level], 466 level, &count, &rcount, 467 &lastseg, &bc, ap->a_p); 468 if (error) 469 allerror = error; 470 real_released += rcount; 471 blocksreleased += count; 472 if (lastiblock[level] < 0) { 473 if (oip->i_ffs1_ib[level] > 0) 474 real_released += nblocks; 475 blocksreleased += nblocks; 476 oip->i_ffs1_ib[level] = 0; 477 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 478 } 479 } 480 if (lastiblock[level] >= 0) 481 goto done; 482 } 483 484 /* 485 * All whole direct blocks or frags. 486 */ 487 for (i = NDADDR - 1; i > lastblock; i--) { 488 long bsize, obsize; 489 490 bn = oip->i_ffs1_db[i]; 491 if (bn == 0) 492 continue; 493 bsize = blksize(fs, oip, i); 494 if (oip->i_ffs1_db[i] > 0) { 495 /* Check for fragment size changes */ 496 obsize = oip->i_lfs_fragsize[i]; 497 real_released += btofsb(fs, obsize); 498 oip->i_lfs_fragsize[i] = 0; 499 } else 500 obsize = 0; 501 blocksreleased += btofsb(fs, bsize); 502 oip->i_ffs1_db[i] = 0; 503 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 504 } 505 if (lastblock < 0) 506 goto done; 507 508 /* 509 * Finally, look for a change in size of the 510 * last direct block; release any frags. 511 */ 512 bn = oip->i_ffs1_db[lastblock]; 513 if (bn != 0) { 514 long oldspace, newspace; 515 #if 0 516 long olddspace; 517 #endif 518 519 /* 520 * Calculate amount of space we're giving 521 * back as old block size minus new block size. 522 */ 523 oldspace = blksize(fs, oip, lastblock); 524 #if 0 525 olddspace = oip->i_lfs_fragsize[lastblock]; 526 #endif 527 528 oip->i_size = oip->i_ffs1_size = length; 529 newspace = blksize(fs, oip, lastblock); 530 if (newspace == 0) 531 panic("itrunc: newspace"); 532 if (oldspace - newspace > 0) { 533 blocksreleased += btofsb(fs, oldspace - newspace); 534 } 535 #if 0 536 if (bn > 0 && olddspace - newspace > 0) { 537 /* No segment accounting here, just vnode */ 538 real_released += btofsb(fs, olddspace - newspace); 539 } 540 #endif 541 } 542 543 done: 544 /* Finish segment accounting corrections */ 545 lfs_update_seguse(fs, lastseg, bc); 546 #ifdef DIAGNOSTIC 547 for (level = SINGLE; level <= TRIPLE; level++) 548 if ((newblks[NDADDR + level] == 0) != 549 (oip->i_ffs1_ib[level]) == 0) { 550 panic("lfs itrunc1"); 551 } 552 for (i = 0; i < NDADDR; i++) 553 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 554 panic("lfs itrunc2"); 555 } 556 if (length == 0 && 557 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 558 panic("lfs itrunc3"); 559 #endif /* DIAGNOSTIC */ 560 /* 561 * Put back the real size. 562 */ 563 oip->i_size = oip->i_ffs1_size = length; 564 oip->i_lfs_effnblks -= blocksreleased; 565 oip->i_ffs1_blocks -= real_released; 566 fs->lfs_bfree += blocksreleased; 567 #ifdef DIAGNOSTIC 568 if (oip->i_size == 0 && 569 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 570 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 571 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 572 panic("lfs_truncate: persistent blocks"); 573 } 574 #endif 575 oip->i_flag |= IN_CHANGE; 576 #ifdef QUOTA 577 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 578 #endif 579 lfs_reserve(fs, ovp, NULL, 580 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 581 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 582 errout: 583 lfs_segunlock(fs); 584 return (allerror ? allerror : error); 585 } 586 587 /* Update segment usage information when removing a block. */ 588 static int 589 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 590 size_t *num) 591 { 592 long seg; 593 int error = 0; 594 595 bsize = fragroundup(fs, bsize); 596 if (daddr > 0) { 597 if (*lastseg != (seg = dtosn(fs, daddr))) { 598 error = lfs_update_seguse(fs, *lastseg, *num); 599 *num = bsize; 600 *lastseg = seg; 601 } else 602 *num += bsize; 603 } 604 return error; 605 } 606 607 /* Finish the accounting updates for a segment. */ 608 static int 609 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 610 { 611 SEGUSE *sup; 612 struct buf *bp; 613 614 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 /* 656 * Calculate index in current block of last 657 * block to be kept. -1 indicates the entire 658 * block so we need not calculate the index. 659 */ 660 factor = 1; 661 for (i = SINGLE; i < level; i++) 662 factor *= NINDIR(fs); 663 last = lastbn; 664 if (lastbn > 0) 665 last /= factor; 666 nblocks = btofsb(fs, fs->lfs_bsize); 667 /* 668 * Get buffer of block pointers, zero those entries corresponding 669 * to blocks to be free'd, and update on disk copy first. Since 670 * double(triple) indirect before single(double) indirect, calls 671 * to bmap on these blocks will fail. However, we already have 672 * the on disk address, so we have to set the b_blkno field 673 * explicitly instead of letting bread do everything for us. 674 */ 675 vp = ITOV(ip); 676 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 677 if (bp->b_flags & (B_DONE | B_DELWRI)) { 678 /* Braces must be here in case trace evaluates to nothing. */ 679 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 680 } else { 681 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 682 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 683 bp->b_flags |= B_READ; 684 if (bp->b_bcount > bp->b_bufsize) 685 panic("lfs_indirtrunc: bad buffer size"); 686 bp->b_blkno = fsbtodb(fs, dbn); 687 VOP_STRATEGY(vp, bp); 688 error = biowait(bp); 689 } 690 if (error) { 691 brelse(bp); 692 *countp = *rcountp = 0; 693 return (error); 694 } 695 696 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 697 if (lastbn >= 0) { 698 MALLOC(copy, int32_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 699 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 700 memset((caddr_t)&bap[last + 1], 0, 701 /* XXX ondisk32 */ 702 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 703 error = VOP_BWRITE(bp); 704 if (error) 705 allerror = error; 706 bap = copy; 707 } 708 709 /* 710 * Recursively free totally unused blocks. 711 */ 712 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 713 i--, nlbn += factor) { 714 nb = bap[i]; 715 if (nb == 0) 716 continue; 717 if (level > SINGLE) { 718 error = lfs_indirtrunc(ip, nlbn, nb, 719 (daddr_t)-1, level - 1, 720 &blkcount, &rblkcount, 721 lastsegp, bcp, p); 722 if (error) 723 allerror = error; 724 blocksreleased += blkcount; 725 real_released += rblkcount; 726 } 727 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 728 if (bap[i] > 0) 729 real_released += nblocks; 730 blocksreleased += nblocks; 731 } 732 733 /* 734 * Recursively free last partial block. 735 */ 736 if (level > SINGLE && lastbn >= 0) { 737 last = lastbn % factor; 738 nb = bap[i]; 739 if (nb != 0) { 740 error = lfs_indirtrunc(ip, nlbn, nb, 741 last, level - 1, &blkcount, 742 &rblkcount, lastsegp, bcp, p); 743 if (error) 744 allerror = error; 745 real_released += rblkcount; 746 blocksreleased += blkcount; 747 } 748 } 749 750 if (copy != NULL) { 751 FREE(copy, M_TEMP); 752 } else { 753 if (bp->b_flags & B_DELWRI) { 754 LFS_UNLOCK_BUF(bp); 755 fs->lfs_avail += btofsb(fs, bp->b_bcount); 756 wakeup(&fs->lfs_avail); 757 } 758 bp->b_flags |= B_INVAL; 759 brelse(bp); 760 } 761 762 *countp = blocksreleased; 763 *rcountp = real_released; 764 return (allerror); 765 } 766 767 /* 768 * Destroy any in core blocks past the truncation length. 769 * Inlined from vtruncbuf, so that lfs_avail could be updated. 770 * We take the seglock to prevent cleaning from occurring while we are 771 * invalidating blocks. 772 */ 773 static int 774 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 775 { 776 struct buf *bp, *nbp; 777 int s, error; 778 struct lfs *fs; 779 voff_t off; 780 781 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 782 simple_lock(&vp->v_interlock); 783 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 784 if (error) 785 return error; 786 787 fs = VTOI(vp)->i_lfs; 788 s = splbio(); 789 790 restart: 791 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 792 nbp = LIST_NEXT(bp, b_vnbufs); 793 if (bp->b_lblkno < lbn) 794 continue; 795 simple_lock(&bp->b_interlock); 796 if (bp->b_flags & B_BUSY) { 797 bp->b_flags |= B_WANTED; 798 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 799 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 800 if (error) { 801 splx(s); 802 return (error); 803 } 804 goto restart; 805 } 806 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 807 if (bp->b_flags & B_DELWRI) { 808 bp->b_flags &= ~B_DELWRI; 809 fs->lfs_avail += btofsb(fs, bp->b_bcount); 810 wakeup(&fs->lfs_avail); 811 } 812 LFS_UNLOCK_BUF(bp); 813 simple_unlock(&bp->b_interlock); 814 brelse(bp); 815 } 816 817 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 818 nbp = LIST_NEXT(bp, b_vnbufs); 819 if (bp->b_lblkno < lbn) 820 continue; 821 simple_lock(&bp->b_interlock); 822 if (bp->b_flags & B_BUSY) { 823 bp->b_flags |= B_WANTED; 824 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 825 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 826 if (error) { 827 splx(s); 828 return (error); 829 } 830 goto restart; 831 } 832 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 833 if (bp->b_flags & B_DELWRI) { 834 bp->b_flags &= ~B_DELWRI; 835 fs->lfs_avail += btofsb(fs, bp->b_bcount); 836 wakeup(&fs->lfs_avail); 837 } 838 LFS_UNLOCK_BUF(bp); 839 simple_unlock(&bp->b_interlock); 840 brelse(bp); 841 } 842 843 splx(s); 844 845 return (0); 846 } 847 848
Indexes created Fri Sep 26 20:09:58 GMT 2025