lfs_inode.c revision 1.152
1 /* $NetBSD: lfs_inode.c,v 1.152 2017/03/19 22:48:00 riastradh 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 /* 32 * Copyright (c) 1986, 1989, 1991, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * @(#)lfs_inode.c 8.9 (Berkeley) 5/8/95 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.152 2017/03/19 22:48:00 riastradh Exp $"); 64 65 #if defined(_KERNEL_OPT) 66 #include "opt_quota.h" 67 #endif 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/mount.h> 72 #include <sys/malloc.h> 73 #include <sys/proc.h> 74 #include <sys/file.h> 75 #include <sys/buf.h> 76 #include <sys/vnode.h> 77 #include <sys/kernel.h> 78 #include <sys/trace.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kauth.h> 81 82 #include <ufs/lfs/ulfs_quotacommon.h> 83 #include <ufs/lfs/ulfs_inode.h> 84 #include <ufs/lfs/ulfsmount.h> 85 #include <ufs/lfs/ulfs_extern.h> 86 87 #include <ufs/lfs/lfs.h> 88 #include <ufs/lfs/lfs_accessors.h> 89 #include <ufs/lfs/lfs_extern.h> 90 #include <ufs/lfs/lfs_kernel.h> 91 92 static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t); 93 static int lfs_indirtrunc(struct inode *, daddr_t, daddr_t, 94 daddr_t, int, daddr_t *, daddr_t *, 95 long *, size_t *); 96 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *); 97 static int lfs_vtruncbuf(struct vnode *, daddr_t, bool, int); 98 99 /* Search a block for a specific dinode. */ 100 union lfs_dinode * 101 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 102 { 103 union lfs_dinode *ldip; 104 unsigned num, i; 105 106 ASSERT_NO_SEGLOCK(fs); 107 /* 108 * Read the inode block backwards, since later versions of the 109 * inode will supercede earlier ones. Though it is unlikely, it is 110 * possible that the same inode will appear in the same inode block. 111 */ 112 num = LFS_INOPB(fs); 113 for (i = num; i-- > 0; ) { 114 ldip = DINO_IN_BLOCK(fs, bp->b_data, i); 115 if (lfs_dino_getinumber(fs, ldip) == ino) 116 return (ldip); 117 } 118 119 printf("searched %u entries for %ju\n", num, (uintmax_t)ino); 120 printf("offset is 0x%jx (seg %d)\n", (uintmax_t)lfs_sb_getoffset(fs), 121 lfs_dtosn(fs, lfs_sb_getoffset(fs))); 122 printf("block is 0x%jx (seg %d)\n", 123 (uintmax_t)LFS_DBTOFSB(fs, bp->b_blkno), 124 lfs_dtosn(fs, LFS_DBTOFSB(fs, bp->b_blkno))); 125 126 return NULL; 127 } 128 129 int 130 lfs_update(struct vnode *vp, const struct timespec *acc, 131 const struct timespec *mod, int updflags) 132 { 133 struct inode *ip; 134 struct lfs *fs = VFSTOULFS(vp->v_mount)->um_lfs; 135 int flags; 136 137 ASSERT_NO_SEGLOCK(fs); 138 if (vp->v_mount->mnt_flag & MNT_RDONLY) 139 return (0); 140 ip = VTOI(vp); 141 142 /* 143 * If we are called from vinvalbuf, and the file's blocks have 144 * already been scheduled for writing, but the writes have not 145 * yet completed, lfs_vflush will not be called, and vinvalbuf 146 * will cause a panic. So, we must wait until any pending write 147 * for our inode completes, if we are called with UPDATE_WAIT set. 148 */ 149 mutex_enter(vp->v_interlock); 150 while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 151 WRITEINPROG(vp)) { 152 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %llu" 153 " (in progress)\n", (unsigned long long) ip->i_number)); 154 cv_wait(&vp->v_cv, vp->v_interlock); 155 } 156 mutex_exit(vp->v_interlock); 157 LFS_ITIMES(ip, acc, mod, NULL); 158 if (updflags & UPDATE_CLOSE) 159 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 160 else 161 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 162 if (flags == 0) 163 return (0); 164 165 /* If sync, push back the vnode and any dirty blocks it may have. */ 166 if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 167 /* Avoid flushing VU_DIROP. */ 168 mutex_enter(&lfs_lock); 169 ++fs->lfs_diropwait; 170 while (vp->v_uflag & VU_DIROP) { 171 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %llu " 172 "(dirops)\n", (unsigned long long) ip->i_number)); 173 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 174 " 0x%x\n", 175 vp->v_iflag | vp->v_vflag | vp->v_uflag, 176 ip->i_flag)); 177 if (fs->lfs_dirops == 0) 178 lfs_flush_fs(fs, SEGM_SYNC); 179 else 180 mtsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 181 0, &lfs_lock); 182 /* XXX KS - by falling out here, are we writing the vn 183 twice? */ 184 } 185 --fs->lfs_diropwait; 186 mutex_exit(&lfs_lock); 187 return lfs_vflush(vp); 188 } 189 return 0; 190 } 191 192 #define SINGLE 0 /* index of single indirect block */ 193 #define DOUBLE 1 /* index of double indirect block */ 194 #define TRIPLE 2 /* index of triple indirect block */ 195 /* 196 * Truncate the inode oip to at most length size, freeing the 197 * disk blocks. 198 */ 199 /* VOP_BWRITE 1 + ULFS_NIADDR + lfs_balloc == 2 + 2*ULFS_NIADDR times */ 200 201 int 202 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred) 203 { 204 daddr_t lastblock; 205 struct inode *oip = VTOI(ovp); 206 daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR]; 207 /* note: newblks is set but only actually used if DIAGNOSTIC */ 208 daddr_t newblks[ULFS_NDADDR + ULFS_NIADDR] __diagused; 209 struct lfs *fs; 210 struct buf *bp; 211 int offset, size, level; 212 daddr_t count, rcount; 213 daddr_t blocksreleased = 0, real_released = 0; 214 int i, nblocks; 215 int aflags, error, allerror = 0; 216 off_t osize; 217 long lastseg; 218 size_t bc; 219 int obufsize, odb; 220 int usepc; 221 222 if (ovp->v_type == VCHR || ovp->v_type == VBLK || 223 ovp->v_type == VFIFO || ovp->v_type == VSOCK) { 224 KASSERT(oip->i_size == 0); 225 return 0; 226 } 227 228 if (length < 0) 229 return (EINVAL); 230 231 /* 232 * Just return and not update modification times. 233 */ 234 if (oip->i_size == length) { 235 /* still do a uvm_vnp_setsize() as writesize may be larger */ 236 uvm_vnp_setsize(ovp, length); 237 return (0); 238 } 239 240 fs = oip->i_lfs; 241 242 if (ovp->v_type == VLNK && 243 (oip->i_size < fs->um_maxsymlinklen || 244 (fs->um_maxsymlinklen == 0 && 245 lfs_dino_getblocks(fs, oip->i_din) == 0))) { 246 KASSERTMSG((length == 0), 247 "partial truncate of symlink: %jd", (intmax_t)length); 248 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 249 oip->i_size = 0; 250 lfs_dino_setsize(fs, oip->i_din, 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 lfs_imtime(fs); 259 osize = oip->i_size; 260 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 261 262 ASSERT_NO_SEGLOCK(fs); 263 /* 264 * Lengthen the size of the file. We must ensure that the 265 * last byte of the file is allocated. Since the smallest 266 * value of osize is 0, length will be at least 1. 267 */ 268 if (osize < length) { 269 if (length > fs->um_maxfilesize) 270 return (EFBIG); 271 aflags = B_CLRBUF; 272 if (ioflag & IO_SYNC) 273 aflags |= B_SYNC; 274 if (usepc) { 275 if (lfs_lblkno(fs, osize) < ULFS_NDADDR && 276 lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) && 277 lfs_blkroundup(fs, osize) != osize) { 278 off_t eob; 279 280 eob = lfs_blkroundup(fs, osize); 281 uvm_vnp_setwritesize(ovp, eob); 282 error = ulfs_balloc_range(ovp, osize, 283 eob - osize, cred, aflags); 284 if (error) { 285 (void) lfs_truncate(ovp, osize, 286 ioflag & IO_SYNC, cred); 287 return error; 288 } 289 if (ioflag & IO_SYNC) { 290 mutex_enter(ovp->v_interlock); 291 VOP_PUTPAGES(ovp, 292 trunc_page(osize & lfs_sb_getbmask(fs)), 293 round_page(eob), 294 PGO_CLEANIT | PGO_SYNCIO); 295 } 296 } 297 uvm_vnp_setwritesize(ovp, length); 298 error = ulfs_balloc_range(ovp, length - 1, 1, cred, 299 aflags); 300 if (error) { 301 (void) lfs_truncate(ovp, osize, 302 ioflag & IO_SYNC, cred); 303 return error; 304 } 305 uvm_vnp_setsize(ovp, length); 306 oip->i_flag |= IN_CHANGE | IN_UPDATE; 307 KASSERT(ovp->v_size == oip->i_size); 308 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 309 return (lfs_update(ovp, NULL, NULL, 0)); 310 } else { 311 error = lfs_reserve(fs, ovp, NULL, 312 lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs))); 313 if (error) 314 return (error); 315 error = lfs_balloc(ovp, length - 1, 1, cred, 316 aflags, &bp); 317 lfs_reserve(fs, ovp, NULL, 318 -lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs))); 319 if (error) 320 return (error); 321 oip->i_size = length; 322 lfs_dino_setsize(fs, oip->i_din, oip->i_size); 323 uvm_vnp_setsize(ovp, length); 324 (void) VOP_BWRITE(bp->b_vp, bp); 325 oip->i_flag |= IN_CHANGE | IN_UPDATE; 326 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 327 return (lfs_update(ovp, NULL, NULL, 0)); 328 } 329 } 330 331 if ((error = lfs_reserve(fs, ovp, NULL, 332 lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 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 = lfs_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 = length; 351 lfs_dino_setsize(fs, oip->i_din, oip->i_size); 352 } else if (!usepc) { 353 lbn = lfs_lblkno(fs, length); 354 aflags = B_CLRBUF; 355 if (ioflag & IO_SYNC) 356 aflags |= B_SYNC; 357 error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp); 358 if (error) { 359 lfs_reserve(fs, ovp, NULL, 360 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 361 goto errout; 362 } 363 obufsize = bp->b_bufsize; 364 odb = lfs_btofsb(fs, bp->b_bcount); 365 oip->i_size = length; 366 lfs_dino_setsize(fs, oip->i_din, oip->i_size); 367 size = lfs_blksize(fs, oip, lbn); 368 if (ovp->v_type != VDIR) 369 memset((char *)bp->b_data + offset, 0, 370 (u_int)(size - offset)); 371 allocbuf(bp, size, 1); 372 if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) { 373 mutex_enter(&lfs_lock); 374 locked_queue_bytes -= obufsize - bp->b_bufsize; 375 mutex_exit(&lfs_lock); 376 } 377 if (bp->b_oflags & BO_DELWRI) { 378 lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size)); 379 /* XXX shouldn't this wake up on lfs_availsleep? */ 380 } 381 (void) VOP_BWRITE(bp->b_vp, bp); 382 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 383 /* 384 * When truncating a regular file down to a non-block-aligned 385 * size, we must zero the part of last block which is past 386 * the new EOF. We must synchronously flush the zeroed pages 387 * to disk since the new pages will be invalidated as soon 388 * as we inform the VM system of the new, smaller size. 389 * We must do this before acquiring the GLOCK, since fetching 390 * the pages will acquire the GLOCK internally. 391 * So there is a window where another thread could see a whole 392 * zeroed page past EOF, but that's life. 393 */ 394 daddr_t xlbn; 395 voff_t eoz; 396 397 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 398 error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags); 399 if (error) { 400 lfs_reserve(fs, ovp, NULL, 401 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 402 goto errout; 403 } 404 xlbn = lfs_lblkno(fs, length); 405 size = lfs_blksize(fs, oip, xlbn); 406 eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize); 407 ubc_zerorange(&ovp->v_uobj, length, eoz - length, 408 UBC_UNMAP_FLAG(ovp)); 409 if (round_page(eoz) > round_page(length)) { 410 mutex_enter(ovp->v_interlock); 411 error = VOP_PUTPAGES(ovp, round_page(length), 412 round_page(eoz), 413 PGO_CLEANIT | PGO_DEACTIVATE | 414 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 415 if (error) { 416 lfs_reserve(fs, ovp, NULL, 417 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 418 goto errout; 419 } 420 } 421 } 422 423 genfs_node_wrlock(ovp); 424 425 oip->i_size = length; 426 lfs_dino_setsize(fs, oip->i_din, oip->i_size); 427 uvm_vnp_setsize(ovp, length); 428 429 /* 430 * Calculate index into inode's block list of 431 * last direct and indirect blocks (if any) 432 * which we want to keep. Lastblock is -1 when 433 * the file is truncated to 0. 434 */ 435 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */ 436 if (length > QUAD_MAX - lfs_sb_getbsize(fs)) 437 lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs)); 438 else 439 lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1; 440 lastiblock[SINGLE] = lastblock - ULFS_NDADDR; 441 lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs); 442 lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs); 443 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs)); 444 /* 445 * Record changed file and block pointers before we start 446 * freeing blocks. lastiblock values are also normalized to -1 447 * for calls to lfs_indirtrunc below. 448 */ 449 for (i=0; i<ULFS_NDADDR; i++) { 450 newblks[i] = lfs_dino_getdb(fs, oip->i_din, i); 451 } 452 for (i=0; i<ULFS_NIADDR; i++) { 453 newblks[ULFS_NDADDR + i] = lfs_dino_getib(fs, oip->i_din, i); 454 } 455 for (level = TRIPLE; level >= SINGLE; level--) 456 if (lastiblock[level] < 0) { 457 newblks[ULFS_NDADDR+level] = 0; 458 lastiblock[level] = -1; 459 } 460 for (i = ULFS_NDADDR - 1; i > lastblock; i--) 461 newblks[i] = 0; 462 463 oip->i_size = osize; 464 lfs_dino_setsize(fs, oip->i_din, oip->i_size); 465 error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0); 466 if (error && !allerror) 467 allerror = error; 468 469 /* 470 * Indirect blocks first. 471 */ 472 indir_lbn[SINGLE] = -ULFS_NDADDR; 473 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1; 474 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1; 475 for (level = TRIPLE; level >= SINGLE; level--) { 476 bn = lfs_dino_getib(fs, oip->i_din, level); 477 if (bn != 0) { 478 error = lfs_indirtrunc(oip, indir_lbn[level], 479 bn, lastiblock[level], 480 level, &count, &rcount, 481 &lastseg, &bc); 482 if (error) 483 allerror = error; 484 real_released += rcount; 485 blocksreleased += count; 486 if (lastiblock[level] < 0) { 487 if (lfs_dino_getib(fs, oip->i_din, level) > 0) 488 real_released += nblocks; 489 blocksreleased += nblocks; 490 lfs_dino_setib(fs, oip->i_din, level, 0); 491 lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs), 492 &lastseg, &bc); 493 lfs_deregister_block(ovp, bn); 494 } 495 } 496 if (lastiblock[level] >= 0) 497 goto done; 498 } 499 500 /* 501 * All whole direct blocks or frags. 502 */ 503 for (i = ULFS_NDADDR - 1; i > lastblock; i--) { 504 long bsize, obsize; 505 506 bn = lfs_dino_getdb(fs, oip->i_din, i); 507 if (bn == 0) 508 continue; 509 bsize = lfs_blksize(fs, oip, i); 510 if (lfs_dino_getdb(fs, oip->i_din, i) > 0) { 511 /* Check for fragment size changes */ 512 obsize = oip->i_lfs_fragsize[i]; 513 real_released += lfs_btofsb(fs, obsize); 514 oip->i_lfs_fragsize[i] = 0; 515 } else 516 obsize = 0; 517 blocksreleased += lfs_btofsb(fs, bsize); 518 lfs_dino_setdb(fs, oip->i_din, i, 0); 519 lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc); 520 lfs_deregister_block(ovp, bn); 521 } 522 if (lastblock < 0) 523 goto done; 524 525 /* 526 * Finally, look for a change in size of the 527 * last direct block; release any frags. 528 */ 529 bn = lfs_dino_getdb(fs, oip->i_din, lastblock); 530 if (bn != 0) { 531 long oldspace, newspace; 532 #if 0 533 long olddspace; 534 #endif 535 536 /* 537 * Calculate amount of space we're giving 538 * back as old block size minus new block size. 539 */ 540 oldspace = lfs_blksize(fs, oip, lastblock); 541 #if 0 542 olddspace = oip->i_lfs_fragsize[lastblock]; 543 #endif 544 545 oip->i_size = length; 546 lfs_dino_setsize(fs, oip->i_din, oip->i_size); 547 newspace = lfs_blksize(fs, oip, lastblock); 548 if (newspace == 0) 549 panic("itrunc: newspace"); 550 if (oldspace - newspace > 0) { 551 blocksreleased += lfs_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 += lfs_btofsb(fs, olddspace - newspace); 557 } 558 #endif 559 } 560 561 done: 562 /* Finish segment accounting corrections */ 563 lfs_update_seguse(fs, oip, lastseg, bc); 564 for (level = SINGLE; level <= TRIPLE; level++) 565 KASSERTMSG(((newblks[ULFS_NDADDR + level] == 0) == 566 (lfs_dino_getib(fs, oip->i_din, level) == 0)), 567 "lfs itrunc1"); 568 for (i = 0; i < ULFS_NDADDR; i++) 569 KASSERTMSG(((newblks[i] == 0) == 570 (lfs_dino_getdb(fs, oip->i_din, i) == 0)), 571 "lfs itrunc2"); 572 KASSERTMSG((length != 0 || LIST_EMPTY(&ovp->v_cleanblkhd)), 573 "lfs itrunc3a"); 574 KASSERTMSG((length != 0 || LIST_EMPTY(&ovp->v_dirtyblkhd)), 575 "lfs itrunc3b"); 576 577 /* 578 * Put back the real size. 579 */ 580 oip->i_size = length; 581 lfs_dino_setsize(fs, oip->i_din, oip->i_size); 582 oip->i_lfs_effnblks -= blocksreleased; 583 lfs_dino_setblocks(fs, oip->i_din, 584 lfs_dino_getblocks(fs, oip->i_din) - real_released); 585 mutex_enter(&lfs_lock); 586 lfs_sb_addbfree(fs, blocksreleased); 587 mutex_exit(&lfs_lock); 588 589 KASSERTMSG((oip->i_size != 0 || 590 lfs_dino_getblocks(fs, oip->i_din) == 0), 591 "ino %llu truncate to 0 but %jd blks/%jd effblks", 592 (unsigned long long) oip->i_number, 593 lfs_dino_getblocks(fs, oip->i_din), oip->i_lfs_effnblks); 594 KASSERTMSG((oip->i_size != 0 || oip->i_lfs_effnblks == 0), 595 "ino %llu truncate to 0 but %jd blks/%jd effblks", 596 (unsigned long long) oip->i_number, 597 lfs_dino_getblocks(fs, oip->i_din), oip->i_lfs_effnblks); 598 599 /* 600 * If we truncated to zero, take us off the paging queue. 601 */ 602 mutex_enter(&lfs_lock); 603 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 604 oip->i_flags &= ~IN_PAGING; 605 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 606 } 607 mutex_exit(&lfs_lock); 608 609 oip->i_flag |= IN_CHANGE; 610 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 611 (void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0); 612 #endif 613 lfs_reserve(fs, ovp, NULL, 614 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 615 genfs_node_unlock(ovp); 616 errout: 617 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 618 if (ovp != fs->lfs_ivnode) 619 lfs_segunlock(fs); 620 return (allerror ? allerror : error); 621 } 622 623 /* Update segment and avail usage information when removing a block. */ 624 static int 625 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr, 626 size_t bsize, long *lastseg, size_t *num) 627 { 628 long seg; 629 int error = 0; 630 631 ASSERT_SEGLOCK(fs); 632 bsize = lfs_fragroundup(fs, bsize); 633 if (daddr > 0) { 634 if (*lastseg != (seg = lfs_dtosn(fs, daddr))) { 635 error = lfs_update_seguse(fs, ip, *lastseg, *num); 636 *num = bsize; 637 *lastseg = seg; 638 } else 639 *num += bsize; 640 } 641 642 return error; 643 } 644 645 /* Finish the accounting updates for a segment. */ 646 static int 647 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num) 648 { 649 struct segdelta *sd; 650 651 ASSERT_SEGLOCK(fs); 652 if (lastseg < 0 || num == 0) 653 return 0; 654 655 LIST_FOREACH(sd, &ip->i_lfs_segdhd, list) 656 if (sd->segnum == lastseg) 657 break; 658 if (sd == NULL) { 659 sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK); 660 sd->segnum = lastseg; 661 sd->num = 0; 662 LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list); 663 } 664 sd->num += num; 665 666 return 0; 667 } 668 669 static void 670 lfs_finalize_seguse(struct lfs *fs, void *v) 671 { 672 SEGUSE *sup; 673 struct buf *bp; 674 struct segdelta *sd; 675 LIST_HEAD(, segdelta) *hd = v; 676 677 ASSERT_SEGLOCK(fs); 678 while((sd = LIST_FIRST(hd)) != NULL) { 679 LIST_REMOVE(sd, list); 680 LFS_SEGENTRY(sup, fs, sd->segnum, bp); 681 if (sd->num > sup->su_nbytes) { 682 printf("lfs_finalize_seguse: segment %ld short by %ld\n", 683 sd->segnum, (long)(sd->num - sup->su_nbytes)); 684 panic("lfs_finalize_seguse: negative bytes"); 685 sup->su_nbytes = sd->num; 686 } 687 sup->su_nbytes -= sd->num; 688 LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp); 689 free(sd, M_SEGMENT); 690 } 691 } 692 693 /* Finish the accounting updates for a segment. */ 694 void 695 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip) 696 { 697 ASSERT_SEGLOCK(fs); 698 lfs_finalize_seguse(fs, &ip->i_lfs_segdhd); 699 } 700 701 /* Finish the accounting updates for a segment. */ 702 void 703 lfs_finalize_fs_seguse(struct lfs *fs) 704 { 705 ASSERT_SEGLOCK(fs); 706 lfs_finalize_seguse(fs, &fs->lfs_segdhd); 707 } 708 709 /* 710 * Release blocks associated with the inode ip and stored in the indirect 711 * block bn. Blocks are free'd in LIFO order up to (but not including) 712 * lastbn. If level is greater than SINGLE, the block is an indirect block 713 * and recursive calls to indirtrunc must be used to cleanse other indirect 714 * blocks. 715 * 716 * NB: triple indirect blocks are untested. 717 */ 718 static int 719 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 720 daddr_t lastbn, int level, daddr_t *countp, 721 daddr_t *rcountp, long *lastsegp, size_t *bcp) 722 { 723 int i; 724 struct buf *bp; 725 struct lfs *fs = ip->i_lfs; 726 void *bap; 727 bool bap_needs_free; 728 struct vnode *vp; 729 daddr_t nb, nlbn, last; 730 daddr_t blkcount, rblkcount, factor; 731 int nblocks; 732 daddr_t blocksreleased = 0, real_released = 0; 733 int error = 0, allerror = 0; 734 735 ASSERT_SEGLOCK(fs); 736 /* 737 * Calculate index in current block of last 738 * block to be kept. -1 indicates the entire 739 * block so we need not calculate the index. 740 */ 741 factor = 1; 742 for (i = SINGLE; i < level; i++) 743 factor *= LFS_NINDIR(fs); 744 last = lastbn; 745 if (lastbn > 0) 746 last /= factor; 747 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs)); 748 /* 749 * Get buffer of block pointers, zero those entries corresponding 750 * to blocks to be free'd, and update on disk copy first. Since 751 * double(triple) indirect before single(double) indirect, calls 752 * to bmap on these blocks will fail. However, we already have 753 * the on disk address, so we have to set the b_blkno field 754 * explicitly instead of letting bread do everything for us. 755 */ 756 vp = ITOV(ip); 757 bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0); 758 if (bp->b_oflags & (BO_DONE | BO_DELWRI)) { 759 /* Braces must be here in case trace evaluates to nothing. */ 760 trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn); 761 } else { 762 trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn); 763 curlwp->l_ru.ru_inblock++; /* pay for read */ 764 bp->b_flags |= B_READ; 765 if (bp->b_bcount > bp->b_bufsize) 766 panic("lfs_indirtrunc: bad buffer size"); 767 bp->b_blkno = LFS_FSBTODB(fs, dbn); 768 VOP_STRATEGY(vp, bp); 769 error = biowait(bp); 770 } 771 if (error) { 772 brelse(bp, 0); 773 *countp = *rcountp = 0; 774 return (error); 775 } 776 777 if (lastbn >= 0) { 778 /* 779 * We still need this block, so copy the data for 780 * subsequent processing; then in the original block, 781 * zero out the dying block pointers and send it off. 782 */ 783 bap = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK); 784 memcpy(bap, bp->b_data, lfs_sb_getbsize(fs)); 785 bap_needs_free = true; 786 787 for (i = last + 1; i < LFS_NINDIR(fs); i++) { 788 lfs_iblock_set(fs, bp->b_data, i, 0); 789 } 790 error = VOP_BWRITE(bp->b_vp, bp); 791 if (error) 792 allerror = error; 793 } else { 794 bap = bp->b_data; 795 bap_needs_free = false; 796 } 797 798 /* 799 * Recursively free totally unused blocks. 800 */ 801 for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 802 i--, nlbn += factor) { 803 nb = lfs_iblock_get(fs, bap, i); 804 if (nb == 0) 805 continue; 806 if (level > SINGLE) { 807 error = lfs_indirtrunc(ip, nlbn, nb, 808 (daddr_t)-1, level - 1, 809 &blkcount, &rblkcount, 810 lastsegp, bcp); 811 if (error) 812 allerror = error; 813 blocksreleased += blkcount; 814 real_released += rblkcount; 815 } 816 lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp); 817 if (lfs_iblock_get(fs, bap, i) > 0) 818 real_released += nblocks; 819 blocksreleased += nblocks; 820 } 821 822 /* 823 * Recursively free last partial block. 824 */ 825 if (level > SINGLE && lastbn >= 0) { 826 last = lastbn % factor; 827 nb = lfs_iblock_get(fs, bap, i); 828 if (nb != 0) { 829 error = lfs_indirtrunc(ip, nlbn, nb, 830 last, level - 1, &blkcount, 831 &rblkcount, lastsegp, bcp); 832 if (error) 833 allerror = error; 834 real_released += rblkcount; 835 blocksreleased += blkcount; 836 } 837 } 838 839 if (bap_needs_free) { 840 lfs_free(fs, bap, LFS_NB_IBLOCK); 841 } else { 842 mutex_enter(&bufcache_lock); 843 if (bp->b_oflags & BO_DELWRI) { 844 LFS_UNLOCK_BUF(bp); 845 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 846 wakeup(&fs->lfs_availsleep); 847 } 848 brelsel(bp, BC_INVAL); 849 mutex_exit(&bufcache_lock); 850 } 851 852 *countp = blocksreleased; 853 *rcountp = real_released; 854 return (allerror); 855 } 856 857 /* 858 * Destroy any in core blocks past the truncation length. 859 * Inlined from vtruncbuf, so that lfs_avail could be updated. 860 * We take the seglock to prevent cleaning from occurring while we are 861 * invalidating blocks. 862 */ 863 static int 864 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo) 865 { 866 struct buf *bp, *nbp; 867 int error; 868 struct lfs *fs; 869 voff_t off; 870 871 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 872 mutex_enter(vp->v_interlock); 873 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 874 if (error) 875 return error; 876 877 fs = VTOI(vp)->i_lfs; 878 879 ASSERT_SEGLOCK(fs); 880 881 mutex_enter(&bufcache_lock); 882 restart: 883 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 884 nbp = LIST_NEXT(bp, b_vnbufs); 885 if (bp->b_lblkno < lbn) 886 continue; 887 error = bbusy(bp, catch, slptimeo, NULL); 888 if (error == EPASSTHROUGH) 889 goto restart; 890 if (error != 0) { 891 mutex_exit(&bufcache_lock); 892 return (error); 893 } 894 mutex_enter(bp->b_objlock); 895 if (bp->b_oflags & BO_DELWRI) { 896 bp->b_oflags &= ~BO_DELWRI; 897 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 898 wakeup(&fs->lfs_availsleep); 899 } 900 mutex_exit(bp->b_objlock); 901 LFS_UNLOCK_BUF(bp); 902 brelsel(bp, BC_INVAL | BC_VFLUSH); 903 } 904 905 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 906 nbp = LIST_NEXT(bp, b_vnbufs); 907 if (bp->b_lblkno < lbn) 908 continue; 909 error = bbusy(bp, catch, slptimeo, NULL); 910 if (error == EPASSTHROUGH) 911 goto restart; 912 if (error != 0) { 913 mutex_exit(&bufcache_lock); 914 return (error); 915 } 916 mutex_enter(bp->b_objlock); 917 if (bp->b_oflags & BO_DELWRI) { 918 bp->b_oflags &= ~BO_DELWRI; 919 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 920 wakeup(&fs->lfs_availsleep); 921 } 922 mutex_exit(bp->b_objlock); 923 LFS_UNLOCK_BUF(bp); 924 brelsel(bp, BC_INVAL | BC_VFLUSH); 925 } 926 mutex_exit(&bufcache_lock); 927 928 return (0); 929 } 930 931
Indexes created Mon Sep 29 03:10:08 GMT 2025