lfs_inode.c revision 1.140
1 /* $NetBSD: lfs_inode.c,v 1.140 2015/07/28 05:09:34 dholland 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 1.140 2015/07/28 05:09:34 dholland 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, long *, long *, long *, size_t *); 95 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *); 96 static int lfs_vtruncbuf(struct vnode *, daddr_t, bool, int); 97 98 /* Search a block for a specific dinode. */ 99 struct ulfs1_dinode * 100 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 101 { 102 struct ulfs1_dinode *dip = (struct ulfs1_dinode *)bp->b_data; 103 struct ulfs1_dinode *ldip, *fin; 104 105 ASSERT_NO_SEGLOCK(fs); 106 /* 107 * Read the inode block backwards, since later versions of the 108 * inode will supercede earlier ones. Though it is unlikely, it is 109 * possible that the same inode will appear in the same inode block. 110 */ 111 fin = dip + LFS_INOPB(fs); 112 for (ldip = fin - 1; ldip >= dip; --ldip) 113 if (ldip->di_inumber == ino) 114 return (ldip); 115 116 printf("searched %d entries\n", (int)(fin - dip)); 117 printf("offset is 0x%jx (seg %d)\n", (uintmax_t)lfs_sb_getoffset(fs), 118 lfs_dtosn(fs, lfs_sb_getoffset(fs))); 119 printf("block is 0x%jx (seg %d)\n", 120 (uintmax_t)LFS_DBTOFSB(fs, bp->b_blkno), 121 lfs_dtosn(fs, LFS_DBTOFSB(fs, bp->b_blkno))); 122 123 return NULL; 124 } 125 126 int 127 lfs_update(struct vnode *vp, const struct timespec *acc, 128 const struct timespec *mod, int updflags) 129 { 130 struct inode *ip; 131 struct lfs *fs = VFSTOULFS(vp->v_mount)->um_lfs; 132 int flags; 133 134 ASSERT_NO_SEGLOCK(fs); 135 if (vp->v_mount->mnt_flag & MNT_RDONLY) 136 return (0); 137 ip = VTOI(vp); 138 139 /* 140 * If we are called from vinvalbuf, and the file's blocks have 141 * already been scheduled for writing, but the writes have not 142 * yet completed, lfs_vflush will not be called, and vinvalbuf 143 * will cause a panic. So, we must wait until any pending write 144 * for our inode completes, if we are called with UPDATE_WAIT set. 145 */ 146 mutex_enter(vp->v_interlock); 147 while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 148 WRITEINPROG(vp)) { 149 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 150 " (in progress)\n", ip->i_number)); 151 cv_wait(&vp->v_cv, vp->v_interlock); 152 } 153 mutex_exit(vp->v_interlock); 154 LFS_ITIMES(ip, acc, mod, NULL); 155 if (updflags & UPDATE_CLOSE) 156 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 157 else 158 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 159 if (flags == 0) 160 return (0); 161 162 /* If sync, push back the vnode and any dirty blocks it may have. */ 163 if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 164 /* Avoid flushing VU_DIROP. */ 165 mutex_enter(&lfs_lock); 166 ++fs->lfs_diropwait; 167 while (vp->v_uflag & VU_DIROP) { 168 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 169 " (dirops)\n", ip->i_number)); 170 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 171 " 0x%x\n", 172 vp->v_iflag | vp->v_vflag | vp->v_uflag, 173 ip->i_flag)); 174 if (fs->lfs_dirops == 0) 175 lfs_flush_fs(fs, SEGM_SYNC); 176 else 177 mtsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 178 0, &lfs_lock); 179 /* XXX KS - by falling out here, are we writing the vn 180 twice? */ 181 } 182 --fs->lfs_diropwait; 183 mutex_exit(&lfs_lock); 184 return lfs_vflush(vp); 185 } 186 return 0; 187 } 188 189 #define SINGLE 0 /* index of single indirect block */ 190 #define DOUBLE 1 /* index of double indirect block */ 191 #define TRIPLE 2 /* index of triple indirect block */ 192 /* 193 * Truncate the inode oip to at most length size, freeing the 194 * disk blocks. 195 */ 196 /* VOP_BWRITE 1 + ULFS_NIADDR + lfs_balloc == 2 + 2*ULFS_NIADDR times */ 197 198 int 199 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred) 200 { 201 daddr_t lastblock; 202 struct inode *oip = VTOI(ovp); 203 daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR]; 204 /* XXX ondisk32 */ 205 int32_t newblks[ULFS_NDADDR + ULFS_NIADDR]; 206 struct lfs *fs; 207 struct buf *bp; 208 int offset, size, level; 209 long count, rcount, blocksreleased = 0, real_released = 0; 210 int i, nblocks; 211 int aflags, error, allerror = 0; 212 off_t osize; 213 long lastseg; 214 size_t bc; 215 int obufsize, odb; 216 int usepc; 217 218 if (ovp->v_type == VCHR || ovp->v_type == VBLK || 219 ovp->v_type == VFIFO || ovp->v_type == VSOCK) { 220 KASSERT(oip->i_size == 0); 221 return 0; 222 } 223 224 if (length < 0) 225 return (EINVAL); 226 227 /* 228 * Just return and not update modification times. 229 */ 230 if (oip->i_size == length) { 231 /* still do a uvm_vnp_setsize() as writesize may be larger */ 232 uvm_vnp_setsize(ovp, length); 233 return (0); 234 } 235 236 fs = oip->i_lfs; 237 238 if (ovp->v_type == VLNK && 239 (oip->i_size < fs->um_maxsymlinklen || 240 (fs->um_maxsymlinklen == 0 && 241 oip->i_ffs1_blocks == 0))) { 242 #ifdef DIAGNOSTIC 243 if (length != 0) 244 panic("lfs_truncate: partial truncate of symlink"); 245 #endif 246 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 247 oip->i_size = oip->i_ffs1_size = 0; 248 oip->i_flag |= IN_CHANGE | IN_UPDATE; 249 return (lfs_update(ovp, NULL, NULL, 0)); 250 } 251 if (oip->i_size == length) { 252 oip->i_flag |= IN_CHANGE | IN_UPDATE; 253 return (lfs_update(ovp, NULL, NULL, 0)); 254 } 255 lfs_imtime(fs); 256 osize = oip->i_size; 257 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 258 259 ASSERT_NO_SEGLOCK(fs); 260 /* 261 * Lengthen the size of the file. We must ensure that the 262 * last byte of the file is allocated. Since the smallest 263 * value of osize is 0, length will be at least 1. 264 */ 265 if (osize < length) { 266 if (length > fs->um_maxfilesize) 267 return (EFBIG); 268 aflags = B_CLRBUF; 269 if (ioflag & IO_SYNC) 270 aflags |= B_SYNC; 271 if (usepc) { 272 if (lfs_lblkno(fs, osize) < ULFS_NDADDR && 273 lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) && 274 lfs_blkroundup(fs, osize) != osize) { 275 off_t eob; 276 277 eob = lfs_blkroundup(fs, osize); 278 uvm_vnp_setwritesize(ovp, eob); 279 error = ulfs_balloc_range(ovp, osize, 280 eob - osize, cred, aflags); 281 if (error) { 282 (void) lfs_truncate(ovp, osize, 283 ioflag & IO_SYNC, cred); 284 return error; 285 } 286 if (ioflag & IO_SYNC) { 287 mutex_enter(ovp->v_interlock); 288 VOP_PUTPAGES(ovp, 289 trunc_page(osize & lfs_sb_getbmask(fs)), 290 round_page(eob), 291 PGO_CLEANIT | PGO_SYNCIO); 292 } 293 } 294 uvm_vnp_setwritesize(ovp, length); 295 error = ulfs_balloc_range(ovp, length - 1, 1, cred, 296 aflags); 297 if (error) { 298 (void) lfs_truncate(ovp, osize, 299 ioflag & IO_SYNC, cred); 300 return error; 301 } 302 uvm_vnp_setsize(ovp, length); 303 oip->i_flag |= IN_CHANGE | IN_UPDATE; 304 KASSERT(ovp->v_size == oip->i_size); 305 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 306 return (lfs_update(ovp, NULL, NULL, 0)); 307 } else { 308 error = lfs_reserve(fs, ovp, NULL, 309 lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs))); 310 if (error) 311 return (error); 312 error = lfs_balloc(ovp, length - 1, 1, cred, 313 aflags, &bp); 314 lfs_reserve(fs, ovp, NULL, 315 -lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs))); 316 if (error) 317 return (error); 318 oip->i_ffs1_size = oip->i_size = length; 319 uvm_vnp_setsize(ovp, length); 320 (void) VOP_BWRITE(bp->b_vp, bp); 321 oip->i_flag |= IN_CHANGE | IN_UPDATE; 322 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 323 return (lfs_update(ovp, NULL, NULL, 0)); 324 } 325 } 326 327 if ((error = lfs_reserve(fs, ovp, NULL, 328 lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 0) 329 return (error); 330 331 /* 332 * Shorten the size of the file. If the file is not being 333 * truncated to a block boundary, the contents of the 334 * partial block following the end of the file must be 335 * zero'ed in case it ever becomes accessible again because 336 * of subsequent file growth. Directories however are not 337 * zero'ed as they should grow back initialized to empty. 338 */ 339 offset = lfs_blkoff(fs, length); 340 lastseg = -1; 341 bc = 0; 342 343 if (ovp != fs->lfs_ivnode) 344 lfs_seglock(fs, SEGM_PROT); 345 if (offset == 0) { 346 oip->i_size = oip->i_ffs1_size = length; 347 } else if (!usepc) { 348 lbn = lfs_lblkno(fs, length); 349 aflags = B_CLRBUF; 350 if (ioflag & IO_SYNC) 351 aflags |= B_SYNC; 352 error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp); 353 if (error) { 354 lfs_reserve(fs, ovp, NULL, 355 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 356 goto errout; 357 } 358 obufsize = bp->b_bufsize; 359 odb = lfs_btofsb(fs, bp->b_bcount); 360 oip->i_size = oip->i_ffs1_size = length; 361 size = lfs_blksize(fs, oip, lbn); 362 if (ovp->v_type != VDIR) 363 memset((char *)bp->b_data + offset, 0, 364 (u_int)(size - offset)); 365 allocbuf(bp, size, 1); 366 if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) { 367 mutex_enter(&lfs_lock); 368 locked_queue_bytes -= obufsize - bp->b_bufsize; 369 mutex_exit(&lfs_lock); 370 } 371 if (bp->b_oflags & BO_DELWRI) { 372 lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size)); 373 /* XXX shouldn't this wake up on lfs_availsleep? */ 374 } 375 (void) VOP_BWRITE(bp->b_vp, bp); 376 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 377 /* 378 * When truncating a regular file down to a non-block-aligned 379 * size, we must zero the part of last block which is past 380 * the new EOF. We must synchronously flush the zeroed pages 381 * to disk since the new pages will be invalidated as soon 382 * as we inform the VM system of the new, smaller size. 383 * We must do this before acquiring the GLOCK, since fetching 384 * the pages will acquire the GLOCK internally. 385 * So there is a window where another thread could see a whole 386 * zeroed page past EOF, but that's life. 387 */ 388 daddr_t xlbn; 389 voff_t eoz; 390 391 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 392 error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags); 393 if (error) { 394 lfs_reserve(fs, ovp, NULL, 395 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 396 goto errout; 397 } 398 xlbn = lfs_lblkno(fs, length); 399 size = lfs_blksize(fs, oip, xlbn); 400 eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize); 401 ubc_zerorange(&ovp->v_uobj, length, eoz - length, 402 UBC_UNMAP_FLAG(ovp)); 403 if (round_page(eoz) > round_page(length)) { 404 mutex_enter(ovp->v_interlock); 405 error = VOP_PUTPAGES(ovp, round_page(length), 406 round_page(eoz), 407 PGO_CLEANIT | PGO_DEACTIVATE | 408 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 409 if (error) { 410 lfs_reserve(fs, ovp, NULL, 411 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 412 goto errout; 413 } 414 } 415 } 416 417 genfs_node_wrlock(ovp); 418 419 oip->i_size = oip->i_ffs1_size = length; 420 uvm_vnp_setsize(ovp, length); 421 422 /* 423 * Calculate index into inode's block list of 424 * last direct and indirect blocks (if any) 425 * which we want to keep. Lastblock is -1 when 426 * the file is truncated to 0. 427 */ 428 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */ 429 if (length > QUAD_MAX - lfs_sb_getbsize(fs)) 430 lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs)); 431 else 432 lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1; 433 lastiblock[SINGLE] = lastblock - ULFS_NDADDR; 434 lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs); 435 lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs); 436 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs)); 437 /* 438 * Record changed file and block pointers before we start 439 * freeing blocks. lastiblock values are also normalized to -1 440 * for calls to lfs_indirtrunc below. 441 */ 442 memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks); 443 for (level = TRIPLE; level >= SINGLE; level--) 444 if (lastiblock[level] < 0) { 445 newblks[ULFS_NDADDR+level] = 0; 446 lastiblock[level] = -1; 447 } 448 for (i = ULFS_NDADDR - 1; i > lastblock; i--) 449 newblks[i] = 0; 450 451 oip->i_size = oip->i_ffs1_size = osize; 452 error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0); 453 if (error && !allerror) 454 allerror = error; 455 456 /* 457 * Indirect blocks first. 458 */ 459 indir_lbn[SINGLE] = -ULFS_NDADDR; 460 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1; 461 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1; 462 for (level = TRIPLE; level >= SINGLE; level--) { 463 bn = oip->i_ffs1_ib[level]; 464 if (bn != 0) { 465 error = lfs_indirtrunc(oip, indir_lbn[level], 466 bn, lastiblock[level], 467 level, &count, &rcount, 468 &lastseg, &bc); 469 if (error) 470 allerror = error; 471 real_released += rcount; 472 blocksreleased += count; 473 if (lastiblock[level] < 0) { 474 if (oip->i_ffs1_ib[level] > 0) 475 real_released += nblocks; 476 blocksreleased += nblocks; 477 oip->i_ffs1_ib[level] = 0; 478 lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs), 479 &lastseg, &bc); 480 lfs_deregister_block(ovp, bn); 481 } 482 } 483 if (lastiblock[level] >= 0) 484 goto done; 485 } 486 487 /* 488 * All whole direct blocks or frags. 489 */ 490 for (i = ULFS_NDADDR - 1; i > lastblock; i--) { 491 long bsize, obsize; 492 493 bn = oip->i_ffs1_db[i]; 494 if (bn == 0) 495 continue; 496 bsize = lfs_blksize(fs, oip, i); 497 if (oip->i_ffs1_db[i] > 0) { 498 /* Check for fragment size changes */ 499 obsize = oip->i_lfs_fragsize[i]; 500 real_released += lfs_btofsb(fs, obsize); 501 oip->i_lfs_fragsize[i] = 0; 502 } else 503 obsize = 0; 504 blocksreleased += lfs_btofsb(fs, bsize); 505 oip->i_ffs1_db[i] = 0; 506 lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc); 507 lfs_deregister_block(ovp, bn); 508 } 509 if (lastblock < 0) 510 goto done; 511 512 /* 513 * Finally, look for a change in size of the 514 * last direct block; release any frags. 515 */ 516 bn = oip->i_ffs1_db[lastblock]; 517 if (bn != 0) { 518 long oldspace, newspace; 519 #if 0 520 long olddspace; 521 #endif 522 523 /* 524 * Calculate amount of space we're giving 525 * back as old block size minus new block size. 526 */ 527 oldspace = lfs_blksize(fs, oip, lastblock); 528 #if 0 529 olddspace = oip->i_lfs_fragsize[lastblock]; 530 #endif 531 532 oip->i_size = oip->i_ffs1_size = length; 533 newspace = lfs_blksize(fs, oip, lastblock); 534 if (newspace == 0) 535 panic("itrunc: newspace"); 536 if (oldspace - newspace > 0) { 537 blocksreleased += lfs_btofsb(fs, oldspace - newspace); 538 } 539 #if 0 540 if (bn > 0 && olddspace - newspace > 0) { 541 /* No segment accounting here, just vnode */ 542 real_released += lfs_btofsb(fs, olddspace - newspace); 543 } 544 #endif 545 } 546 547 done: 548 /* Finish segment accounting corrections */ 549 lfs_update_seguse(fs, oip, lastseg, bc); 550 #ifdef DIAGNOSTIC 551 for (level = SINGLE; level <= TRIPLE; level++) 552 if ((newblks[ULFS_NDADDR + level] == 0) != 553 ((oip->i_ffs1_ib[level]) == 0)) { 554 panic("lfs itrunc1"); 555 } 556 for (i = 0; i < ULFS_NDADDR; i++) 557 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 558 panic("lfs itrunc2"); 559 } 560 if (length == 0 && 561 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 562 panic("lfs itrunc3"); 563 #endif /* DIAGNOSTIC */ 564 /* 565 * Put back the real size. 566 */ 567 oip->i_size = oip->i_ffs1_size = length; 568 oip->i_lfs_effnblks -= blocksreleased; 569 oip->i_ffs1_blocks -= real_released; 570 mutex_enter(&lfs_lock); 571 lfs_sb_addbfree(fs, blocksreleased); 572 mutex_exit(&lfs_lock); 573 #ifdef DIAGNOSTIC 574 if (oip->i_size == 0 && 575 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 576 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 577 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 578 panic("lfs_truncate: persistent blocks"); 579 } 580 #endif 581 582 /* 583 * If we truncated to zero, take us off the paging queue. 584 */ 585 mutex_enter(&lfs_lock); 586 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 587 oip->i_flags &= ~IN_PAGING; 588 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 589 } 590 mutex_exit(&lfs_lock); 591 592 oip->i_flag |= IN_CHANGE; 593 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 594 (void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0); 595 #endif 596 lfs_reserve(fs, ovp, NULL, 597 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 598 genfs_node_unlock(ovp); 599 errout: 600 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 601 if (ovp != fs->lfs_ivnode) 602 lfs_segunlock(fs); 603 return (allerror ? allerror : error); 604 } 605 606 /* Update segment and avail usage information when removing a block. */ 607 static int 608 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr, 609 size_t bsize, long *lastseg, size_t *num) 610 { 611 long seg; 612 int error = 0; 613 614 ASSERT_SEGLOCK(fs); 615 bsize = lfs_fragroundup(fs, bsize); 616 if (daddr > 0) { 617 if (*lastseg != (seg = lfs_dtosn(fs, daddr))) { 618 error = lfs_update_seguse(fs, ip, *lastseg, *num); 619 *num = bsize; 620 *lastseg = seg; 621 } else 622 *num += bsize; 623 } 624 625 return error; 626 } 627 628 /* Finish the accounting updates for a segment. */ 629 static int 630 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num) 631 { 632 struct segdelta *sd; 633 634 ASSERT_SEGLOCK(fs); 635 if (lastseg < 0 || num == 0) 636 return 0; 637 638 LIST_FOREACH(sd, &ip->i_lfs_segdhd, list) 639 if (sd->segnum == lastseg) 640 break; 641 if (sd == NULL) { 642 sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK); 643 sd->segnum = lastseg; 644 sd->num = 0; 645 LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list); 646 } 647 sd->num += num; 648 649 return 0; 650 } 651 652 static void 653 lfs_finalize_seguse(struct lfs *fs, void *v) 654 { 655 SEGUSE *sup; 656 struct buf *bp; 657 struct segdelta *sd; 658 LIST_HEAD(, segdelta) *hd = v; 659 660 ASSERT_SEGLOCK(fs); 661 while((sd = LIST_FIRST(hd)) != NULL) { 662 LIST_REMOVE(sd, list); 663 LFS_SEGENTRY(sup, fs, sd->segnum, bp); 664 if (sd->num > sup->su_nbytes) { 665 printf("lfs_finalize_seguse: segment %ld short by %ld\n", 666 sd->segnum, (long)(sd->num - sup->su_nbytes)); 667 panic("lfs_finalize_seguse: negative bytes"); 668 sup->su_nbytes = sd->num; 669 } 670 sup->su_nbytes -= sd->num; 671 LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp); 672 free(sd, M_SEGMENT); 673 } 674 } 675 676 /* Finish the accounting updates for a segment. */ 677 void 678 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip) 679 { 680 ASSERT_SEGLOCK(fs); 681 lfs_finalize_seguse(fs, &ip->i_lfs_segdhd); 682 } 683 684 /* Finish the accounting updates for a segment. */ 685 void 686 lfs_finalize_fs_seguse(struct lfs *fs) 687 { 688 ASSERT_SEGLOCK(fs); 689 lfs_finalize_seguse(fs, &fs->lfs_segdhd); 690 } 691 692 /* 693 * Release blocks associated with the inode ip and stored in the indirect 694 * block bn. Blocks are free'd in LIFO order up to (but not including) 695 * lastbn. If level is greater than SINGLE, the block is an indirect block 696 * and recursive calls to indirtrunc must be used to cleanse other indirect 697 * blocks. 698 * 699 * NB: triple indirect blocks are untested. 700 */ 701 static int 702 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 703 daddr_t lastbn, int level, long *countp, 704 long *rcountp, long *lastsegp, size_t *bcp) 705 { 706 int i; 707 struct buf *bp; 708 struct lfs *fs = ip->i_lfs; 709 int32_t *bap; /* XXX ondisk32 */ 710 struct vnode *vp; 711 daddr_t nb, nlbn, last; 712 int32_t *copy = NULL; /* XXX ondisk32 */ 713 long blkcount, rblkcount, factor; 714 int nblocks, blocksreleased = 0, real_released = 0; 715 int error = 0, allerror = 0; 716 717 ASSERT_SEGLOCK(fs); 718 /* 719 * Calculate index in current block of last 720 * block to be kept. -1 indicates the entire 721 * block so we need not calculate the index. 722 */ 723 factor = 1; 724 for (i = SINGLE; i < level; i++) 725 factor *= LFS_NINDIR(fs); 726 last = lastbn; 727 if (lastbn > 0) 728 last /= factor; 729 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs)); 730 /* 731 * Get buffer of block pointers, zero those entries corresponding 732 * to blocks to be free'd, and update on disk copy first. Since 733 * double(triple) indirect before single(double) indirect, calls 734 * to bmap on these blocks will fail. However, we already have 735 * the on disk address, so we have to set the b_blkno field 736 * explicitly instead of letting bread do everything for us. 737 */ 738 vp = ITOV(ip); 739 bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0); 740 if (bp->b_oflags & (BO_DONE | BO_DELWRI)) { 741 /* Braces must be here in case trace evaluates to nothing. */ 742 trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn); 743 } else { 744 trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn); 745 curlwp->l_ru.ru_inblock++; /* pay for read */ 746 bp->b_flags |= B_READ; 747 if (bp->b_bcount > bp->b_bufsize) 748 panic("lfs_indirtrunc: bad buffer size"); 749 bp->b_blkno = LFS_FSBTODB(fs, dbn); 750 VOP_STRATEGY(vp, bp); 751 error = biowait(bp); 752 } 753 if (error) { 754 brelse(bp, 0); 755 *countp = *rcountp = 0; 756 return (error); 757 } 758 759 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 760 if (lastbn >= 0) { 761 copy = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK); 762 memcpy((void *)copy, (void *)bap, lfs_sb_getbsize(fs)); 763 memset((void *)&bap[last + 1], 0, 764 /* XXX ondisk32 */ 765 (u_int)(LFS_NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 766 error = VOP_BWRITE(bp->b_vp, bp); 767 if (error) 768 allerror = error; 769 bap = copy; 770 } 771 772 /* 773 * Recursively free totally unused blocks. 774 */ 775 for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 776 i--, nlbn += factor) { 777 nb = bap[i]; 778 if (nb == 0) 779 continue; 780 if (level > SINGLE) { 781 error = lfs_indirtrunc(ip, nlbn, nb, 782 (daddr_t)-1, level - 1, 783 &blkcount, &rblkcount, 784 lastsegp, bcp); 785 if (error) 786 allerror = error; 787 blocksreleased += blkcount; 788 real_released += rblkcount; 789 } 790 lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp); 791 if (bap[i] > 0) 792 real_released += nblocks; 793 blocksreleased += nblocks; 794 } 795 796 /* 797 * Recursively free last partial block. 798 */ 799 if (level > SINGLE && lastbn >= 0) { 800 last = lastbn % factor; 801 nb = bap[i]; 802 if (nb != 0) { 803 error = lfs_indirtrunc(ip, nlbn, nb, 804 last, level - 1, &blkcount, 805 &rblkcount, lastsegp, bcp); 806 if (error) 807 allerror = error; 808 real_released += rblkcount; 809 blocksreleased += blkcount; 810 } 811 } 812 813 if (copy != NULL) { 814 lfs_free(fs, copy, LFS_NB_IBLOCK); 815 } else { 816 mutex_enter(&bufcache_lock); 817 if (bp->b_oflags & BO_DELWRI) { 818 LFS_UNLOCK_BUF(bp); 819 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 820 wakeup(&fs->lfs_availsleep); 821 } 822 brelsel(bp, BC_INVAL); 823 mutex_exit(&bufcache_lock); 824 } 825 826 *countp = blocksreleased; 827 *rcountp = real_released; 828 return (allerror); 829 } 830 831 /* 832 * Destroy any in core blocks past the truncation length. 833 * Inlined from vtruncbuf, so that lfs_avail could be updated. 834 * We take the seglock to prevent cleaning from occurring while we are 835 * invalidating blocks. 836 */ 837 static int 838 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo) 839 { 840 struct buf *bp, *nbp; 841 int error; 842 struct lfs *fs; 843 voff_t off; 844 845 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 846 mutex_enter(vp->v_interlock); 847 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 848 if (error) 849 return error; 850 851 fs = VTOI(vp)->i_lfs; 852 853 ASSERT_SEGLOCK(fs); 854 855 mutex_enter(&bufcache_lock); 856 restart: 857 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 858 nbp = LIST_NEXT(bp, b_vnbufs); 859 if (bp->b_lblkno < lbn) 860 continue; 861 error = bbusy(bp, catch, slptimeo, NULL); 862 if (error == EPASSTHROUGH) 863 goto restart; 864 if (error != 0) { 865 mutex_exit(&bufcache_lock); 866 return (error); 867 } 868 mutex_enter(bp->b_objlock); 869 if (bp->b_oflags & BO_DELWRI) { 870 bp->b_oflags &= ~BO_DELWRI; 871 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 872 wakeup(&fs->lfs_availsleep); 873 } 874 mutex_exit(bp->b_objlock); 875 LFS_UNLOCK_BUF(bp); 876 brelsel(bp, BC_INVAL | BC_VFLUSH); 877 } 878 879 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 880 nbp = LIST_NEXT(bp, b_vnbufs); 881 if (bp->b_lblkno < lbn) 882 continue; 883 error = bbusy(bp, catch, slptimeo, NULL); 884 if (error == EPASSTHROUGH) 885 goto restart; 886 if (error != 0) { 887 mutex_exit(&bufcache_lock); 888 return (error); 889 } 890 mutex_enter(bp->b_objlock); 891 if (bp->b_oflags & BO_DELWRI) { 892 bp->b_oflags &= ~BO_DELWRI; 893 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 894 wakeup(&fs->lfs_availsleep); 895 } 896 mutex_exit(bp->b_objlock); 897 LFS_UNLOCK_BUF(bp); 898 brelsel(bp, BC_INVAL | BC_VFLUSH); 899 } 900 mutex_exit(&bufcache_lock); 901 902 return (0); 903 } 904 905
Indexes created Mon Sep 29 03:10:08 GMT 2025