lfs_inode.c revision 1.90.2.4
1 /* $NetBSD: lfs_inode.c,v 1.90.2.4 2006/05/20 22:33:01 riz 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 1.90.2.4 2006/05/20 22:33:01 riz 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(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 ASSERT_NO_SEGLOCK(fs); 147 if (vp->v_mount->mnt_flag & MNT_RDONLY) 148 return (0); 149 ip = VTOI(vp); 150 151 /* 152 * If we are called from vinvalbuf, and the file's blocks have 153 * already been scheduled for writing, but the writes have not 154 * yet completed, lfs_vflush will not be called, and vinvalbuf 155 * will cause a panic. So, we must wait until any pending write 156 * for our inode completes, if we are called with UPDATE_WAIT set. 157 */ 158 s = splbio(); 159 simple_lock(&vp->v_interlock); 160 while ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 161 WRITEINPROG(vp)) { 162 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 163 " (in progress)\n", ip->i_number)); 164 ltsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock); 165 } 166 simple_unlock(&vp->v_interlock); 167 splx(s); 168 TIMEVAL_TO_TIMESPEC(&time, &ts); 169 LFS_ITIMES(ip, 170 ap->a_access ? ap->a_access : &ts, 171 ap->a_modify ? ap->a_modify : &ts, &ts); 172 if (ap->a_flags & UPDATE_CLOSE) 173 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 174 else 175 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 176 if (flags == 0) 177 return (0); 178 179 /* If sync, push back the vnode and any dirty blocks it may have. */ 180 if ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 181 /* Avoid flushing VDIROP. */ 182 simple_lock(&fs->lfs_interlock); 183 ++fs->lfs_diropwait; 184 while (vp->v_flag & VDIROP) { 185 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 186 " (dirops)\n", ip->i_number)); 187 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 188 " 0x%x\n", vp->v_flag, ip->i_flag)); 189 if (fs->lfs_dirops == 0) 190 lfs_flush_fs(fs, SEGM_SYNC); 191 else 192 ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 193 0, &fs->lfs_interlock); 194 /* XXX KS - by falling out here, are we writing the vn 195 twice? */ 196 } 197 --fs->lfs_diropwait; 198 simple_unlock(&fs->lfs_interlock); 199 return lfs_vflush(vp); 200 } 201 return 0; 202 } 203 204 #define SINGLE 0 /* index of single indirect block */ 205 #define DOUBLE 1 /* index of double indirect block */ 206 #define TRIPLE 2 /* index of triple indirect block */ 207 /* 208 * Truncate the inode oip to at most length size, freeing the 209 * disk blocks. 210 */ 211 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */ 212 213 int 214 lfs_truncate(void *v) 215 { 216 struct vop_truncate_args /* { 217 struct vnode *a_vp; 218 off_t a_length; 219 int a_flags; 220 struct ucred *a_cred; 221 struct proc *a_p; 222 } */ *ap = v; 223 struct vnode *ovp = ap->a_vp; 224 struct genfs_node *gp = VTOG(ovp); 225 daddr_t lastblock; 226 struct inode *oip = VTOI(ovp); 227 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 228 /* XXX ondisk32 */ 229 int32_t newblks[NDADDR + NIADDR]; 230 off_t length = ap->a_length; 231 struct lfs *fs; 232 struct buf *bp; 233 int offset, size, level; 234 long count, rcount, blocksreleased = 0, real_released = 0; 235 int i, ioflag, nblocks; 236 int aflags, error, allerror = 0; 237 off_t osize; 238 long lastseg; 239 size_t bc; 240 int obufsize, odb; 241 int usepc; 242 struct ufsmount *ump = oip->i_ump; 243 244 if (length < 0) 245 return (EINVAL); 246 247 /* 248 * Just return and not update modification times. 249 */ 250 if (oip->i_size == length) 251 return (0); 252 253 if (ovp->v_type == VLNK && 254 (oip->i_size < ump->um_maxsymlinklen || 255 (ump->um_maxsymlinklen == 0 && 256 oip->i_ffs1_blocks == 0))) { 257 #ifdef DIAGNOSTIC 258 if (length != 0) 259 panic("lfs_truncate: partial truncate of symlink"); 260 #endif 261 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 262 oip->i_size = oip->i_ffs1_size = 0; 263 oip->i_flag |= IN_CHANGE | IN_UPDATE; 264 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 265 } 266 if (oip->i_size == length) { 267 oip->i_flag |= IN_CHANGE | IN_UPDATE; 268 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 269 } 270 #ifdef QUOTA 271 if ((error = getinoquota(oip)) != 0) 272 return (error); 273 #endif 274 fs = oip->i_lfs; 275 lfs_imtime(fs); 276 osize = oip->i_size; 277 ioflag = ap->a_flags; 278 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 279 280 ASSERT_NO_SEGLOCK(fs); 281 /* 282 * Lengthen the size of the file. We must ensure that the 283 * last byte of the file is allocated. Since the smallest 284 * value of osize is 0, length will be at least 1. 285 */ 286 if (osize < length) { 287 if (length > ump->um_maxfilesize) 288 return (EFBIG); 289 aflags = B_CLRBUF; 290 if (ioflag & IO_SYNC) 291 aflags |= B_SYNC; 292 if (usepc) { 293 if (lblkno(fs, osize) < NDADDR && 294 lblkno(fs, osize) != lblkno(fs, length) && 295 blkroundup(fs, osize) != osize) { 296 off_t eob; 297 298 eob = blkroundup(fs, osize); 299 error = ufs_balloc_range(ovp, osize, 300 eob - osize, ap->a_cred, aflags); 301 if (error) 302 return error; 303 if (ioflag & IO_SYNC) { 304 ovp->v_size = eob; 305 simple_lock(&ovp->v_interlock); 306 VOP_PUTPAGES(ovp, 307 trunc_page(osize & fs->lfs_bmask), 308 round_page(eob), 309 PGO_CLEANIT | PGO_SYNCIO); 310 } 311 } 312 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 313 aflags); 314 if (error) { 315 (void) VOP_TRUNCATE(ovp, osize, 316 ioflag & IO_SYNC, 317 ap->a_cred, ap->a_p); 318 return error; 319 } 320 uvm_vnp_setsize(ovp, length); 321 oip->i_flag |= IN_CHANGE | IN_UPDATE; 322 KASSERT(ovp->v_size == oip->i_size); 323 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 324 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 325 } else { 326 error = lfs_reserve(fs, ovp, NULL, 327 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 328 if (error) 329 return (error); 330 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, 331 aflags, &bp); 332 lfs_reserve(fs, ovp, NULL, 333 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 334 if (error) 335 return (error); 336 oip->i_ffs1_size = oip->i_size = length; 337 uvm_vnp_setsize(ovp, length); 338 (void) VOP_BWRITE(bp); 339 oip->i_flag |= IN_CHANGE | IN_UPDATE; 340 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 341 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 342 } 343 } 344 345 if ((error = lfs_reserve(fs, ovp, NULL, 346 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 347 return (error); 348 349 /* 350 * Shorten the size of the file. If the file is not being 351 * truncated to a block boundary, the contents of the 352 * partial block following the end of the file must be 353 * zero'ed in case it ever becomes accessible again because 354 * of subsequent file growth. Directories however are not 355 * zero'ed as they should grow back initialized to empty. 356 */ 357 offset = blkoff(fs, length); 358 lastseg = -1; 359 bc = 0; 360 361 if (ovp != fs->lfs_ivnode) 362 lfs_seglock(fs, SEGM_PROT); 363 if (offset == 0) { 364 oip->i_size = oip->i_ffs1_size = length; 365 } else if (!usepc) { 366 lbn = lblkno(fs, length); 367 aflags = B_CLRBUF; 368 if (ioflag & IO_SYNC) 369 aflags |= B_SYNC; 370 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 371 if (error) { 372 lfs_reserve(fs, ovp, NULL, 373 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 374 goto errout; 375 } 376 obufsize = bp->b_bufsize; 377 odb = btofsb(fs, bp->b_bcount); 378 oip->i_size = oip->i_ffs1_size = length; 379 size = blksize(fs, oip, lbn); 380 if (ovp->v_type != VDIR) 381 memset((char *)bp->b_data + offset, 0, 382 (u_int)(size - offset)); 383 allocbuf(bp, size, 1); 384 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) { 385 simple_lock(&lfs_subsys_lock); 386 locked_queue_bytes -= obufsize - bp->b_bufsize; 387 simple_unlock(&lfs_subsys_lock); 388 } 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 daddr_t xlbn; 405 voff_t eoz; 406 407 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 408 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 409 aflags); 410 if (error) { 411 lfs_reserve(fs, ovp, NULL, 412 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 413 goto errout; 414 } 415 xlbn = lblkno(fs, length); 416 size = blksize(fs, oip, xlbn); 417 eoz = MIN(lblktosize(fs, xlbn) + size, osize); 418 uvm_vnp_zerorange(ovp, length, eoz - length); 419 if (round_page(eoz) > round_page(length)) { 420 simple_lock(&ovp->v_interlock); 421 error = VOP_PUTPAGES(ovp, round_page(length), 422 round_page(eoz), 423 PGO_CLEANIT | PGO_DEACTIVATE | 424 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 425 if (error) { 426 lfs_reserve(fs, ovp, NULL, 427 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 428 goto errout; 429 } 430 } 431 } 432 433 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 434 435 oip->i_size = oip->i_ffs1_size = length; 436 uvm_vnp_setsize(ovp, length); 437 /* 438 * Calculate index into inode's block list of 439 * last direct and indirect blocks (if any) 440 * which we want to keep. Lastblock is -1 when 441 * the file is truncated to 0. 442 */ 443 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */ 444 if (length > QUAD_MAX - fs->lfs_bsize) 445 lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize); 446 else 447 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 448 lastiblock[SINGLE] = lastblock - NDADDR; 449 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 450 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 451 nblocks = btofsb(fs, fs->lfs_bsize); 452 /* 453 * Record changed file and block pointers before we start 454 * freeing blocks. lastiblock values are also normalized to -1 455 * for calls to lfs_indirtrunc below. 456 */ 457 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 458 for (level = TRIPLE; level >= SINGLE; level--) 459 if (lastiblock[level] < 0) { 460 newblks[NDADDR+level] = 0; 461 lastiblock[level] = -1; 462 } 463 for (i = NDADDR - 1; i > lastblock; i--) 464 newblks[i] = 0; 465 466 oip->i_size = oip->i_ffs1_size = osize; 467 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 468 if (error && !allerror) 469 allerror = error; 470 471 /* 472 * Indirect blocks first. 473 */ 474 indir_lbn[SINGLE] = -NDADDR; 475 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 476 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 477 for (level = TRIPLE; level >= SINGLE; level--) { 478 bn = oip->i_ffs1_ib[level]; 479 if (bn != 0) { 480 error = lfs_indirtrunc(oip, indir_lbn[level], 481 bn, lastiblock[level], 482 level, &count, &rcount, 483 &lastseg, &bc, ap->a_p); 484 if (error) 485 allerror = error; 486 real_released += rcount; 487 blocksreleased += count; 488 if (lastiblock[level] < 0) { 489 if (oip->i_ffs1_ib[level] > 0) 490 real_released += nblocks; 491 blocksreleased += nblocks; 492 oip->i_ffs1_ib[level] = 0; 493 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 494 lfs_deregister_block(ovp, bn); 495 } 496 } 497 if (lastiblock[level] >= 0) 498 goto done; 499 } 500 501 /* 502 * All whole direct blocks or frags. 503 */ 504 for (i = NDADDR - 1; i > lastblock; i--) { 505 long bsize, obsize; 506 507 bn = oip->i_ffs1_db[i]; 508 if (bn == 0) 509 continue; 510 bsize = blksize(fs, oip, i); 511 if (oip->i_ffs1_db[i] > 0) { 512 /* Check for fragment size changes */ 513 obsize = oip->i_lfs_fragsize[i]; 514 real_released += btofsb(fs, obsize); 515 oip->i_lfs_fragsize[i] = 0; 516 } else 517 obsize = 0; 518 blocksreleased += btofsb(fs, bsize); 519 oip->i_ffs1_db[i] = 0; 520 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 521 lfs_deregister_block(ovp, bn); 522 } 523 if (lastblock < 0) 524 goto done; 525 526 /* 527 * Finally, look for a change in size of the 528 * last direct block; release any frags. 529 */ 530 bn = oip->i_ffs1_db[lastblock]; 531 if (bn != 0) { 532 long oldspace, newspace; 533 #if 0 534 long olddspace; 535 #endif 536 537 /* 538 * Calculate amount of space we're giving 539 * back as old block size minus new block size. 540 */ 541 oldspace = blksize(fs, oip, lastblock); 542 #if 0 543 olddspace = oip->i_lfs_fragsize[lastblock]; 544 #endif 545 546 oip->i_size = oip->i_ffs1_size = length; 547 newspace = blksize(fs, oip, lastblock); 548 if (newspace == 0) 549 panic("itrunc: newspace"); 550 if (oldspace - newspace > 0) { 551 blocksreleased += btofsb(fs, oldspace - newspace); 552 } 553 #if 0 554 if (bn > 0 && olddspace - newspace > 0) { 555 /* No segment accounting here, just vnode */ 556 real_released += btofsb(fs, olddspace - newspace); 557 } 558 #endif 559 } 560 561 done: 562 /* Finish segment accounting corrections */ 563 lfs_update_seguse(fs, lastseg, bc); 564 #ifdef DIAGNOSTIC 565 for (level = SINGLE; level <= TRIPLE; level++) 566 if ((newblks[NDADDR + level] == 0) != 567 (oip->i_ffs1_ib[level]) == 0) { 568 panic("lfs itrunc1"); 569 } 570 for (i = 0; i < NDADDR; i++) 571 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 572 panic("lfs itrunc2"); 573 } 574 if (length == 0 && 575 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 576 panic("lfs itrunc3"); 577 #endif /* DIAGNOSTIC */ 578 /* 579 * Put back the real size. 580 */ 581 oip->i_size = oip->i_ffs1_size = length; 582 oip->i_lfs_effnblks -= blocksreleased; 583 oip->i_ffs1_blocks -= real_released; 584 simple_lock(&fs->lfs_interlock); 585 fs->lfs_bfree += blocksreleased; 586 simple_unlock(&fs->lfs_interlock); 587 #ifdef DIAGNOSTIC 588 if (oip->i_size == 0 && 589 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 590 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 591 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 592 panic("lfs_truncate: persistent blocks"); 593 } 594 #endif 595 596 /* 597 * If we truncated to zero, take us off the paging queue. 598 */ 599 simple_lock(&fs->lfs_interlock); 600 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 601 oip->i_flags &= ~IN_PAGING; 602 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 603 } 604 simple_unlock(&fs->lfs_interlock); 605 606 oip->i_flag |= IN_CHANGE; 607 #ifdef QUOTA 608 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 609 #endif 610 lfs_reserve(fs, ovp, NULL, 611 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 612 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 613 errout: 614 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 615 if (ovp != fs->lfs_ivnode) 616 lfs_segunlock(fs); 617 return (allerror ? allerror : error); 618 } 619 620 /* Update segment and avail usage information when removing a block. */ 621 static int 622 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 623 size_t *num) 624 { 625 long seg; 626 int error = 0; 627 628 ASSERT_SEGLOCK(fs); 629 bsize = fragroundup(fs, bsize); 630 if (daddr > 0) { 631 if (*lastseg != (seg = dtosn(fs, daddr))) { 632 error = lfs_update_seguse(fs, *lastseg, *num); 633 *num = bsize; 634 *lastseg = seg; 635 } else 636 *num += bsize; 637 } 638 639 return error; 640 } 641 642 /* Finish the accounting updates for a segment. */ 643 static int 644 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 645 { 646 SEGUSE *sup; 647 struct buf *bp; 648 649 ASSERT_SEGLOCK(fs); 650 if (lastseg < 0 || num == 0) 651 return 0; 652 653 LFS_SEGENTRY(sup, fs, lastseg, bp); 654 if (num > sup->su_nbytes) { 655 printf("lfs_truncate: segment %ld short by %ld\n", 656 lastseg, (long)num - sup->su_nbytes); 657 panic("lfs_truncate: negative bytes"); 658 sup->su_nbytes = num; 659 } 660 sup->su_nbytes -= num; 661 LFS_WRITESEGENTRY(sup, fs, lastseg, bp); 662 663 return 0; 664 } 665 666 /* 667 * Release blocks associated with the inode ip and stored in the indirect 668 * block bn. Blocks are free'd in LIFO order up to (but not including) 669 * lastbn. If level is greater than SINGLE, the block is an indirect block 670 * and recursive calls to indirtrunc must be used to cleanse other indirect 671 * blocks. 672 * 673 * NB: triple indirect blocks are untested. 674 */ 675 static int 676 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 677 daddr_t lastbn, int level, long *countp, 678 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 679 { 680 int i; 681 struct buf *bp; 682 struct lfs *fs = ip->i_lfs; 683 int32_t *bap; /* XXX ondisk32 */ 684 struct vnode *vp; 685 daddr_t nb, nlbn, last; 686 int32_t *copy = NULL; /* XXX ondisk32 */ 687 long blkcount, rblkcount, factor; 688 int nblocks, blocksreleased = 0, real_released = 0; 689 int error = 0, allerror = 0; 690 691 ASSERT_SEGLOCK(fs); 692 /* 693 * Calculate index in current block of last 694 * block to be kept. -1 indicates the entire 695 * block so we need not calculate the index. 696 */ 697 factor = 1; 698 for (i = SINGLE; i < level; i++) 699 factor *= NINDIR(fs); 700 last = lastbn; 701 if (lastbn > 0) 702 last /= factor; 703 nblocks = btofsb(fs, fs->lfs_bsize); 704 /* 705 * Get buffer of block pointers, zero those entries corresponding 706 * to blocks to be free'd, and update on disk copy first. Since 707 * double(triple) indirect before single(double) indirect, calls 708 * to bmap on these blocks will fail. However, we already have 709 * the on disk address, so we have to set the b_blkno field 710 * explicitly instead of letting bread do everything for us. 711 */ 712 vp = ITOV(ip); 713 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 714 if (bp->b_flags & (B_DONE | B_DELWRI)) { 715 /* Braces must be here in case trace evaluates to nothing. */ 716 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 717 } else { 718 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 719 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 720 bp->b_flags |= B_READ; 721 if (bp->b_bcount > bp->b_bufsize) 722 panic("lfs_indirtrunc: bad buffer size"); 723 bp->b_blkno = fsbtodb(fs, dbn); 724 VOP_STRATEGY(vp, bp); 725 error = biowait(bp); 726 } 727 if (error) { 728 brelse(bp); 729 *countp = *rcountp = 0; 730 return (error); 731 } 732 733 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 734 if (lastbn >= 0) { 735 copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK); 736 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 737 memset((caddr_t)&bap[last + 1], 0, 738 /* XXX ondisk32 */ 739 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 740 error = VOP_BWRITE(bp); 741 if (error) 742 allerror = error; 743 bap = copy; 744 } 745 746 /* 747 * Recursively free totally unused blocks. 748 */ 749 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 750 i--, nlbn += factor) { 751 nb = bap[i]; 752 if (nb == 0) 753 continue; 754 if (level > SINGLE) { 755 error = lfs_indirtrunc(ip, nlbn, nb, 756 (daddr_t)-1, level - 1, 757 &blkcount, &rblkcount, 758 lastsegp, bcp, p); 759 if (error) 760 allerror = error; 761 blocksreleased += blkcount; 762 real_released += rblkcount; 763 } 764 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 765 if (bap[i] > 0) 766 real_released += nblocks; 767 blocksreleased += nblocks; 768 } 769 770 /* 771 * Recursively free last partial block. 772 */ 773 if (level > SINGLE && lastbn >= 0) { 774 last = lastbn % factor; 775 nb = bap[i]; 776 if (nb != 0) { 777 error = lfs_indirtrunc(ip, nlbn, nb, 778 last, level - 1, &blkcount, 779 &rblkcount, lastsegp, bcp, p); 780 if (error) 781 allerror = error; 782 real_released += rblkcount; 783 blocksreleased += blkcount; 784 } 785 } 786 787 if (copy != NULL) { 788 lfs_free(fs, copy, LFS_NB_IBLOCK); 789 } else { 790 if (bp->b_flags & B_DELWRI) { 791 LFS_UNLOCK_BUF(bp); 792 fs->lfs_avail += btofsb(fs, bp->b_bcount); 793 wakeup(&fs->lfs_avail); 794 } 795 bp->b_flags |= B_INVAL; 796 brelse(bp); 797 } 798 799 *countp = blocksreleased; 800 *rcountp = real_released; 801 return (allerror); 802 } 803 804 /* 805 * Destroy any in core blocks past the truncation length. 806 * Inlined from vtruncbuf, so that lfs_avail could be updated. 807 * We take the seglock to prevent cleaning from occurring while we are 808 * invalidating blocks. 809 */ 810 static int 811 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 812 { 813 struct buf *bp, *nbp; 814 int s, error; 815 struct lfs *fs; 816 voff_t off; 817 818 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 819 simple_lock(&vp->v_interlock); 820 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 821 if (error) 822 return error; 823 824 fs = VTOI(vp)->i_lfs; 825 s = splbio(); 826 827 ASSERT_SEGLOCK(fs); 828 restart: 829 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 830 nbp = LIST_NEXT(bp, b_vnbufs); 831 if (bp->b_lblkno < lbn) 832 continue; 833 simple_lock(&bp->b_interlock); 834 if (bp->b_flags & B_BUSY) { 835 bp->b_flags |= B_WANTED; 836 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 837 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 838 if (error) { 839 splx(s); 840 return (error); 841 } 842 goto restart; 843 } 844 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 845 if (bp->b_flags & B_DELWRI) { 846 bp->b_flags &= ~B_DELWRI; 847 fs->lfs_avail += btofsb(fs, bp->b_bcount); 848 wakeup(&fs->lfs_avail); 849 } 850 LFS_UNLOCK_BUF(bp); 851 simple_unlock(&bp->b_interlock); 852 brelse(bp); 853 } 854 855 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 856 nbp = LIST_NEXT(bp, b_vnbufs); 857 if (bp->b_lblkno < lbn) 858 continue; 859 simple_lock(&bp->b_interlock); 860 if (bp->b_flags & B_BUSY) { 861 bp->b_flags |= B_WANTED; 862 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 863 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 864 if (error) { 865 splx(s); 866 return (error); 867 } 868 goto restart; 869 } 870 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 871 if (bp->b_flags & B_DELWRI) { 872 bp->b_flags &= ~B_DELWRI; 873 fs->lfs_avail += btofsb(fs, bp->b_bcount); 874 wakeup(&fs->lfs_avail); 875 } 876 LFS_UNLOCK_BUF(bp); 877 simple_unlock(&bp->b_interlock); 878 brelse(bp); 879 } 880 881 splx(s); 882 883 return (0); 884 } 885 886
Indexes created Thu Oct 02 14:10:14 GMT 2025