1 /*- 2 * Copyright (c) 2002 Networks Associates Technology, Inc. 3 * All rights reserved. 4 * 5 * This software was developed for the FreeBSD Project by Marshall 6 * Kirk McKusick and Network Associates Laboratories, the Security 7 * Research Division of Network Associates, Inc. under DARPA/SPAWAR 8 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS 9 * research program 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * Copyright (c) 1982, 1986, 1989, 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 * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95 60 */ 61 62 #include <sys/cdefs.h> 63 __FBSDID("$FreeBSD$"); 64 65 #include "opt_quota.h" 66 67 #include <sys/param.h> 68 #include <sys/capsicum.h> 69 #include <sys/systm.h> 70 #include <sys/bio.h> 71 #include <sys/buf.h> 72 #include <sys/conf.h> 73 #include <sys/fcntl.h> 74 #include <sys/file.h> 75 #include <sys/filedesc.h> 76 #include <sys/priv.h> 77 #include <sys/proc.h> 78 #include <sys/vnode.h> 79 #include <sys/mount.h> 80 #include <sys/kernel.h> 81 #include <sys/syscallsubr.h> 82 #include <sys/sysctl.h> 83 #include <sys/syslog.h> 84 #include <sys/taskqueue.h> 85 86 #include <security/audit/audit.h> 87 88 #include <geom/geom.h> 89 90 #include <ufs/ufs/dir.h> 91 #include <ufs/ufs/extattr.h> 92 #include <ufs/ufs/quota.h> 93 #include <ufs/ufs/inode.h> 94 #include <ufs/ufs/ufs_extern.h> 95 #include <ufs/ufs/ufsmount.h> 96 97 #include <ufs/ffs/fs.h> 98 #include <ufs/ffs/ffs_extern.h> 99 #include <ufs/ffs/softdep.h> 100 101 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, u_int cg, ufs2_daddr_t bpref, 102 int size, int rsize); 103 104 static ufs2_daddr_t ffs_alloccg(struct inode *, u_int, ufs2_daddr_t, int, int); 105 static ufs2_daddr_t 106 ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t, int); 107 static void ffs_blkfree_cg(struct ufsmount *, struct fs *, 108 struct vnode *, ufs2_daddr_t, long, ino_t, 109 struct workhead *); 110 static void ffs_blkfree_trim_completed(struct bio *); 111 static void ffs_blkfree_trim_task(void *ctx, int pending __unused); 112 #ifdef INVARIANTS 113 static int ffs_checkblk(struct inode *, ufs2_daddr_t, long); 114 #endif 115 static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int); 116 static ino_t ffs_dirpref(struct inode *); 117 static ufs2_daddr_t ffs_fragextend(struct inode *, u_int, ufs2_daddr_t, 118 int, int); 119 static ufs2_daddr_t ffs_hashalloc 120 (struct inode *, u_int, ufs2_daddr_t, int, int, allocfcn_t *); 121 static ufs2_daddr_t ffs_nodealloccg(struct inode *, u_int, ufs2_daddr_t, int, 122 int); 123 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int); 124 static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *); 125 static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *); 126 127 /* 128 * Allocate a block in the filesystem. 129 * 130 * The size of the requested block is given, which must be some 131 * multiple of fs_fsize and <= fs_bsize. 132 * A preference may be optionally specified. If a preference is given 133 * the following hierarchy is used to allocate a block: 134 * 1) allocate the requested block. 135 * 2) allocate a rotationally optimal block in the same cylinder. 136 * 3) allocate a block in the same cylinder group. 137 * 4) quadradically rehash into other cylinder groups, until an 138 * available block is located. 139 * If no block preference is given the following hierarchy is used 140 * to allocate a block: 141 * 1) allocate a block in the cylinder group that contains the 142 * inode for the file. 143 * 2) quadradically rehash into other cylinder groups, until an 144 * available block is located. 145 */ 146 int 147 ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp) 148 struct inode *ip; 149 ufs2_daddr_t lbn, bpref; 150 int size, flags; 151 struct ucred *cred; 152 ufs2_daddr_t *bnp; 153 { 154 struct fs *fs; 155 struct ufsmount *ump; 156 ufs2_daddr_t bno; 157 u_int cg, reclaimed; 158 static struct timeval lastfail; 159 static int curfail; 160 int64_t delta; 161 #ifdef QUOTA 162 int error; 163 #endif 164 165 *bnp = 0; 166 ump = ITOUMP(ip); 167 fs = ump->um_fs; 168 mtx_assert(UFS_MTX(ump), MA_OWNED); 169 #ifdef INVARIANTS 170 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 171 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n", 172 devtoname(ump->um_dev), (long)fs->fs_bsize, size, 173 fs->fs_fsmnt); 174 panic("ffs_alloc: bad size"); 175 } 176 if (cred == NOCRED) 177 panic("ffs_alloc: missing credential"); 178 #endif /* INVARIANTS */ 179 reclaimed = 0; 180 retry: 181 #ifdef QUOTA 182 UFS_UNLOCK(ump); 183 error = chkdq(ip, btodb(size), cred, 0); 184 if (error) 185 return (error); 186 UFS_LOCK(ump); 187 #endif 188 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) 189 goto nospace; 190 if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) && 191 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) 192 goto nospace; 193 if (bpref >= fs->fs_size) 194 bpref = 0; 195 if (bpref == 0) 196 cg = ino_to_cg(fs, ip->i_number); 197 else 198 cg = dtog(fs, bpref); 199 bno = ffs_hashalloc(ip, cg, bpref, size, size, ffs_alloccg); 200 if (bno > 0) { 201 delta = btodb(size); 202 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); 203 if (flags & IO_EXT) 204 ip->i_flag |= IN_CHANGE; 205 else 206 ip->i_flag |= IN_CHANGE | IN_UPDATE; 207 *bnp = bno; 208 return (0); 209 } 210 nospace: 211 #ifdef QUOTA 212 UFS_UNLOCK(ump); 213 /* 214 * Restore user's disk quota because allocation failed. 215 */ 216 (void) chkdq(ip, -btodb(size), cred, FORCE); 217 UFS_LOCK(ump); 218 #endif 219 if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { 220 reclaimed = 1; 221 softdep_request_cleanup(fs, ITOV(ip), cred, FLUSH_BLOCKS_WAIT); 222 goto retry; 223 } 224 UFS_UNLOCK(ump); 225 if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) { 226 ffs_fserr(fs, ip->i_number, "filesystem full"); 227 uprintf("\n%s: write failed, filesystem is full\n", 228 fs->fs_fsmnt); 229 } 230 return (ENOSPC); 231 } 232 233 /* 234 * Reallocate a fragment to a bigger size 235 * 236 * The number and size of the old block is given, and a preference 237 * and new size is also specified. The allocator attempts to extend 238 * the original block. Failing that, the regular block allocator is 239 * invoked to get an appropriate block. 240 */ 241 int 242 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp) 243 struct inode *ip; 244 ufs2_daddr_t lbprev; 245 ufs2_daddr_t bprev; 246 ufs2_daddr_t bpref; 247 int osize, nsize, flags; 248 struct ucred *cred; 249 struct buf **bpp; 250 { 251 struct vnode *vp; 252 struct fs *fs; 253 struct buf *bp; 254 struct ufsmount *ump; 255 u_int cg, request, reclaimed; 256 int error, gbflags; 257 ufs2_daddr_t bno; 258 static struct timeval lastfail; 259 static int curfail; 260 int64_t delta; 261 262 vp = ITOV(ip); 263 ump = ITOUMP(ip); 264 fs = ump->um_fs; 265 bp = NULL; 266 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0; 267 268 mtx_assert(UFS_MTX(ump), MA_OWNED); 269 #ifdef INVARIANTS 270 if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) 271 panic("ffs_realloccg: allocation on suspended filesystem"); 272 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || 273 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { 274 printf( 275 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", 276 devtoname(ump->um_dev), (long)fs->fs_bsize, osize, 277 nsize, fs->fs_fsmnt); 278 panic("ffs_realloccg: bad size"); 279 } 280 if (cred == NOCRED) 281 panic("ffs_realloccg: missing credential"); 282 #endif /* INVARIANTS */ 283 reclaimed = 0; 284 retry: 285 if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) && 286 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) { 287 goto nospace; 288 } 289 if (bprev == 0) { 290 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n", 291 devtoname(ump->um_dev), (long)fs->fs_bsize, (intmax_t)bprev, 292 fs->fs_fsmnt); 293 panic("ffs_realloccg: bad bprev"); 294 } 295 UFS_UNLOCK(ump); 296 /* 297 * Allocate the extra space in the buffer. 298 */ 299 error = bread_gb(vp, lbprev, osize, NOCRED, gbflags, &bp); 300 if (error) { 301 brelse(bp); 302 return (error); 303 } 304 305 if (bp->b_blkno == bp->b_lblkno) { 306 if (lbprev >= UFS_NDADDR) 307 panic("ffs_realloccg: lbprev out of range"); 308 bp->b_blkno = fsbtodb(fs, bprev); 309 } 310 311 #ifdef QUOTA 312 error = chkdq(ip, btodb(nsize - osize), cred, 0); 313 if (error) { 314 brelse(bp); 315 return (error); 316 } 317 #endif 318 /* 319 * Check for extension in the existing location. 320 */ 321 *bpp = NULL; 322 cg = dtog(fs, bprev); 323 UFS_LOCK(ump); 324 bno = ffs_fragextend(ip, cg, bprev, osize, nsize); 325 if (bno) { 326 if (bp->b_blkno != fsbtodb(fs, bno)) 327 panic("ffs_realloccg: bad blockno"); 328 delta = btodb(nsize - osize); 329 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); 330 if (flags & IO_EXT) 331 ip->i_flag |= IN_CHANGE; 332 else 333 ip->i_flag |= IN_CHANGE | IN_UPDATE; 334 allocbuf(bp, nsize); 335 bp->b_flags |= B_DONE; 336 vfs_bio_bzero_buf(bp, osize, nsize - osize); 337 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) 338 vfs_bio_set_valid(bp, osize, nsize - osize); 339 *bpp = bp; 340 return (0); 341 } 342 /* 343 * Allocate a new disk location. 344 */ 345 if (bpref >= fs->fs_size) 346 bpref = 0; 347 switch ((int)fs->fs_optim) { 348 case FS_OPTSPACE: 349 /* 350 * Allocate an exact sized fragment. Although this makes 351 * best use of space, we will waste time relocating it if 352 * the file continues to grow. If the fragmentation is 353 * less than half of the minimum free reserve, we choose 354 * to begin optimizing for time. 355 */ 356 request = nsize; 357 if (fs->fs_minfree <= 5 || 358 fs->fs_cstotal.cs_nffree > 359 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100)) 360 break; 361 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", 362 fs->fs_fsmnt); 363 fs->fs_optim = FS_OPTTIME; 364 break; 365 case FS_OPTTIME: 366 /* 367 * At this point we have discovered a file that is trying to 368 * grow a small fragment to a larger fragment. To save time, 369 * we allocate a full sized block, then free the unused portion. 370 * If the file continues to grow, the `ffs_fragextend' call 371 * above will be able to grow it in place without further 372 * copying. If aberrant programs cause disk fragmentation to 373 * grow within 2% of the free reserve, we choose to begin 374 * optimizing for space. 375 */ 376 request = fs->fs_bsize; 377 if (fs->fs_cstotal.cs_nffree < 378 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100) 379 break; 380 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", 381 fs->fs_fsmnt); 382 fs->fs_optim = FS_OPTSPACE; 383 break; 384 default: 385 printf("dev = %s, optim = %ld, fs = %s\n", 386 devtoname(ump->um_dev), (long)fs->fs_optim, fs->fs_fsmnt); 387 panic("ffs_realloccg: bad optim"); 388 /* NOTREACHED */ 389 } 390 bno = ffs_hashalloc(ip, cg, bpref, request, nsize, ffs_alloccg); 391 if (bno > 0) { 392 bp->b_blkno = fsbtodb(fs, bno); 393 if (!DOINGSOFTDEP(vp)) 394 ffs_blkfree(ump, fs, ump->um_devvp, bprev, (long)osize, 395 ip->i_number, vp->v_type, NULL); 396 delta = btodb(nsize - osize); 397 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); 398 if (flags & IO_EXT) 399 ip->i_flag |= IN_CHANGE; 400 else 401 ip->i_flag |= IN_CHANGE | IN_UPDATE; 402 allocbuf(bp, nsize); 403 bp->b_flags |= B_DONE; 404 vfs_bio_bzero_buf(bp, osize, nsize - osize); 405 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) 406 vfs_bio_set_valid(bp, osize, nsize - osize); 407 *bpp = bp; 408 return (0); 409 } 410 #ifdef QUOTA 411 UFS_UNLOCK(ump); 412 /* 413 * Restore user's disk quota because allocation failed. 414 */ 415 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE); 416 UFS_LOCK(ump); 417 #endif 418 nospace: 419 /* 420 * no space available 421 */ 422 if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { 423 reclaimed = 1; 424 UFS_UNLOCK(ump); 425 if (bp) { 426 brelse(bp); 427 bp = NULL; 428 } 429 UFS_LOCK(ump); 430 softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); 431 goto retry; 432 } 433 UFS_UNLOCK(ump); 434 if (bp) 435 brelse(bp); 436 if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) { 437 ffs_fserr(fs, ip->i_number, "filesystem full"); 438 uprintf("\n%s: write failed, filesystem is full\n", 439 fs->fs_fsmnt); 440 } 441 return (ENOSPC); 442 } 443 444 /* 445 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 446 * 447 * The vnode and an array of buffer pointers for a range of sequential 448 * logical blocks to be made contiguous is given. The allocator attempts 449 * to find a range of sequential blocks starting as close as possible 450 * from the end of the allocation for the logical block immediately 451 * preceding the current range. If successful, the physical block numbers 452 * in the buffer pointers and in the inode are changed to reflect the new 453 * allocation. If unsuccessful, the allocation is left unchanged. The 454 * success in doing the reallocation is returned. Note that the error 455 * return is not reflected back to the user. Rather the previous block 456 * allocation will be used. 457 */ 458 459 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem"); 460 461 static int doasyncfree = 1; 462 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, 463 "do not force synchronous writes when blocks are reallocated"); 464 465 static int doreallocblks = 1; 466 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, 467 "enable block reallocation"); 468 469 static int maxclustersearch = 10; 470 SYSCTL_INT(_vfs_ffs, OID_AUTO, maxclustersearch, CTLFLAG_RW, &maxclustersearch, 471 0, "max number of cylinder group to search for contigous blocks"); 472 473 #ifdef DEBUG 474 static volatile int prtrealloc = 0; 475 #endif 476 477 int 478 ffs_reallocblks(ap) 479 struct vop_reallocblks_args /* { 480 struct vnode *a_vp; 481 struct cluster_save *a_buflist; 482 } */ *ap; 483 { 484 struct ufsmount *ump; 485 486 /* 487 * If the underlying device can do deletes, then skip reallocating 488 * the blocks of this file into contiguous sequences. Devices that 489 * benefit from BIO_DELETE also benefit from not moving the data. 490 * These devices are flash and therefore work less well with this 491 * optimization. Also skip if reallocblks has been disabled globally. 492 */ 493 ump = ap->a_vp->v_mount->mnt_data; 494 if (ump->um_candelete || doreallocblks == 0) 495 return (ENOSPC); 496 497 /* 498 * We can't wait in softdep prealloc as it may fsync and recurse 499 * here. Instead we simply fail to reallocate blocks if this 500 * rare condition arises. 501 */ 502 if (DOINGSOFTDEP(ap->a_vp)) 503 if (softdep_prealloc(ap->a_vp, MNT_NOWAIT) != 0) 504 return (ENOSPC); 505 if (ump->um_fstype == UFS1) 506 return (ffs_reallocblks_ufs1(ap)); 507 return (ffs_reallocblks_ufs2(ap)); 508 } 509 510 static int 511 ffs_reallocblks_ufs1(ap) 512 struct vop_reallocblks_args /* { 513 struct vnode *a_vp; 514 struct cluster_save *a_buflist; 515 } */ *ap; 516 { 517 struct fs *fs; 518 struct inode *ip; 519 struct vnode *vp; 520 struct buf *sbp, *ebp; 521 ufs1_daddr_t *bap, *sbap, *ebap; 522 struct cluster_save *buflist; 523 struct ufsmount *ump; 524 ufs_lbn_t start_lbn, end_lbn; 525 ufs1_daddr_t soff, newblk, blkno; 526 ufs2_daddr_t pref; 527 struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; 528 int i, cg, len, start_lvl, end_lvl, ssize; 529 530 vp = ap->a_vp; 531 ip = VTOI(vp); 532 ump = ITOUMP(ip); 533 fs = ump->um_fs; 534 /* 535 * If we are not tracking block clusters or if we have less than 4% 536 * free blocks left, then do not attempt to cluster. Running with 537 * less than 5% free block reserve is not recommended and those that 538 * choose to do so do not expect to have good file layout. 539 */ 540 if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) 541 return (ENOSPC); 542 buflist = ap->a_buflist; 543 len = buflist->bs_nchildren; 544 start_lbn = buflist->bs_children[0]->b_lblkno; 545 end_lbn = start_lbn + len - 1; 546 #ifdef INVARIANTS 547 for (i = 0; i < len; i++) 548 if (!ffs_checkblk(ip, 549 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 550 panic("ffs_reallocblks: unallocated block 1"); 551 for (i = 1; i < len; i++) 552 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 553 panic("ffs_reallocblks: non-logical cluster"); 554 blkno = buflist->bs_children[0]->b_blkno; 555 ssize = fsbtodb(fs, fs->fs_frag); 556 for (i = 1; i < len - 1; i++) 557 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) 558 panic("ffs_reallocblks: non-physical cluster %d", i); 559 #endif 560 /* 561 * If the cluster crosses the boundary for the first indirect 562 * block, leave space for the indirect block. Indirect blocks 563 * are initially laid out in a position after the last direct 564 * block. Block reallocation would usually destroy locality by 565 * moving the indirect block out of the way to make room for 566 * data blocks if we didn't compensate here. We should also do 567 * this for other indirect block boundaries, but it is only 568 * important for the first one. 569 */ 570 if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) 571 return (ENOSPC); 572 /* 573 * If the latest allocation is in a new cylinder group, assume that 574 * the filesystem has decided to move and do not force it back to 575 * the previous cylinder group. 576 */ 577 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 578 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 579 return (ENOSPC); 580 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 581 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 582 return (ENOSPC); 583 /* 584 * Get the starting offset and block map for the first block. 585 */ 586 if (start_lvl == 0) { 587 sbap = &ip->i_din1->di_db[0]; 588 soff = start_lbn; 589 } else { 590 idp = &start_ap[start_lvl - 1]; 591 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { 592 brelse(sbp); 593 return (ENOSPC); 594 } 595 sbap = (ufs1_daddr_t *)sbp->b_data; 596 soff = idp->in_off; 597 } 598 /* 599 * If the block range spans two block maps, get the second map. 600 */ 601 ebap = NULL; 602 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 603 ssize = len; 604 } else { 605 #ifdef INVARIANTS 606 if (start_lvl > 0 && 607 start_ap[start_lvl - 1].in_lbn == idp->in_lbn) 608 panic("ffs_reallocblk: start == end"); 609 #endif 610 ssize = len - (idp->in_off + 1); 611 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) 612 goto fail; 613 ebap = (ufs1_daddr_t *)ebp->b_data; 614 } 615 /* 616 * Find the preferred location for the cluster. If we have not 617 * previously failed at this endeavor, then follow our standard 618 * preference calculation. If we have failed at it, then pick up 619 * where we last ended our search. 620 */ 621 UFS_LOCK(ump); 622 if (ip->i_nextclustercg == -1) 623 pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap); 624 else 625 pref = cgdata(fs, ip->i_nextclustercg); 626 /* 627 * Search the block map looking for an allocation of the desired size. 628 * To avoid wasting too much time, we limit the number of cylinder 629 * groups that we will search. 630 */ 631 cg = dtog(fs, pref); 632 for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { 633 if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) 634 break; 635 cg += 1; 636 if (cg >= fs->fs_ncg) 637 cg = 0; 638 } 639 /* 640 * If we have failed in our search, record where we gave up for 641 * next time. Otherwise, fall back to our usual search citerion. 642 */ 643 if (newblk == 0) { 644 ip->i_nextclustercg = cg; 645 UFS_UNLOCK(ump); 646 goto fail; 647 } 648 ip->i_nextclustercg = -1; 649 /* 650 * We have found a new contiguous block. 651 * 652 * First we have to replace the old block pointers with the new 653 * block pointers in the inode and indirect blocks associated 654 * with the file. 655 */ 656 #ifdef DEBUG 657 if (prtrealloc) 658 printf("realloc: ino %ju, lbns %jd-%jd\n\told:", 659 (uintmax_t)ip->i_number, 660 (intmax_t)start_lbn, (intmax_t)end_lbn); 661 #endif 662 blkno = newblk; 663 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { 664 if (i == ssize) { 665 bap = ebap; 666 soff = -i; 667 } 668 #ifdef INVARIANTS 669 if (!ffs_checkblk(ip, 670 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 671 panic("ffs_reallocblks: unallocated block 2"); 672 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) 673 panic("ffs_reallocblks: alloc mismatch"); 674 #endif 675 #ifdef DEBUG 676 if (prtrealloc) 677 printf(" %d,", *bap); 678 #endif 679 if (DOINGSOFTDEP(vp)) { 680 if (sbap == &ip->i_din1->di_db[0] && i < ssize) 681 softdep_setup_allocdirect(ip, start_lbn + i, 682 blkno, *bap, fs->fs_bsize, fs->fs_bsize, 683 buflist->bs_children[i]); 684 else 685 softdep_setup_allocindir_page(ip, start_lbn + i, 686 i < ssize ? sbp : ebp, soff + i, blkno, 687 *bap, buflist->bs_children[i]); 688 } 689 *bap++ = blkno; 690 } 691 /* 692 * Next we must write out the modified inode and indirect blocks. 693 * For strict correctness, the writes should be synchronous since 694 * the old block values may have been written to disk. In practise 695 * they are almost never written, but if we are concerned about 696 * strict correctness, the `doasyncfree' flag should be set to zero. 697 * 698 * The test on `doasyncfree' should be changed to test a flag 699 * that shows whether the associated buffers and inodes have 700 * been written. The flag should be set when the cluster is 701 * started and cleared whenever the buffer or inode is flushed. 702 * We can then check below to see if it is set, and do the 703 * synchronous write only when it has been cleared. 704 */ 705 if (sbap != &ip->i_din1->di_db[0]) { 706 if (doasyncfree) 707 bdwrite(sbp); 708 else 709 bwrite(sbp); 710 } else { 711 ip->i_flag |= IN_CHANGE | IN_UPDATE; 712 if (!doasyncfree) 713 ffs_update(vp, 1); 714 } 715 if (ssize < len) { 716 if (doasyncfree) 717 bdwrite(ebp); 718 else 719 bwrite(ebp); 720 } 721 /* 722 * Last, free the old blocks and assign the new blocks to the buffers. 723 */ 724 #ifdef DEBUG 725 if (prtrealloc) 726 printf("\n\tnew:"); 727 #endif 728 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { 729 if (!DOINGSOFTDEP(vp)) 730 ffs_blkfree(ump, fs, ump->um_devvp, 731 dbtofsb(fs, buflist->bs_children[i]->b_blkno), 732 fs->fs_bsize, ip->i_number, vp->v_type, NULL); 733 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 734 #ifdef INVARIANTS 735 if (!ffs_checkblk(ip, 736 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 737 panic("ffs_reallocblks: unallocated block 3"); 738 #endif 739 #ifdef DEBUG 740 if (prtrealloc) 741 printf(" %d,", blkno); 742 #endif 743 } 744 #ifdef DEBUG 745 if (prtrealloc) { 746 prtrealloc--; 747 printf("\n"); 748 } 749 #endif 750 return (0); 751 752 fail: 753 if (ssize < len) 754 brelse(ebp); 755 if (sbap != &ip->i_din1->di_db[0]) 756 brelse(sbp); 757 return (ENOSPC); 758 } 759 760 static int 761 ffs_reallocblks_ufs2(ap) 762 struct vop_reallocblks_args /* { 763 struct vnode *a_vp; 764 struct cluster_save *a_buflist; 765 } */ *ap; 766 { 767 struct fs *fs; 768 struct inode *ip; 769 struct vnode *vp; 770 struct buf *sbp, *ebp; 771 ufs2_daddr_t *bap, *sbap, *ebap; 772 struct cluster_save *buflist; 773 struct ufsmount *ump; 774 ufs_lbn_t start_lbn, end_lbn; 775 ufs2_daddr_t soff, newblk, blkno, pref; 776 struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; 777 int i, cg, len, start_lvl, end_lvl, ssize; 778 779 vp = ap->a_vp; 780 ip = VTOI(vp); 781 ump = ITOUMP(ip); 782 fs = ump->um_fs; 783 /* 784 * If we are not tracking block clusters or if we have less than 4% 785 * free blocks left, then do not attempt to cluster. Running with 786 * less than 5% free block reserve is not recommended and those that 787 * choose to do so do not expect to have good file layout. 788 */ 789 if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) 790 return (ENOSPC); 791 buflist = ap->a_buflist; 792 len = buflist->bs_nchildren; 793 start_lbn = buflist->bs_children[0]->b_lblkno; 794 end_lbn = start_lbn + len - 1; 795 #ifdef INVARIANTS 796 for (i = 0; i < len; i++) 797 if (!ffs_checkblk(ip, 798 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 799 panic("ffs_reallocblks: unallocated block 1"); 800 for (i = 1; i < len; i++) 801 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 802 panic("ffs_reallocblks: non-logical cluster"); 803 blkno = buflist->bs_children[0]->b_blkno; 804 ssize = fsbtodb(fs, fs->fs_frag); 805 for (i = 1; i < len - 1; i++) 806 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) 807 panic("ffs_reallocblks: non-physical cluster %d", i); 808 #endif 809 /* 810 * If the cluster crosses the boundary for the first indirect 811 * block, do not move anything in it. Indirect blocks are 812 * usually initially laid out in a position between the data 813 * blocks. Block reallocation would usually destroy locality by 814 * moving the indirect block out of the way to make room for 815 * data blocks if we didn't compensate here. We should also do 816 * this for other indirect block boundaries, but it is only 817 * important for the first one. 818 */ 819 if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) 820 return (ENOSPC); 821 /* 822 * If the latest allocation is in a new cylinder group, assume that 823 * the filesystem has decided to move and do not force it back to 824 * the previous cylinder group. 825 */ 826 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 827 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 828 return (ENOSPC); 829 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 830 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 831 return (ENOSPC); 832 /* 833 * Get the starting offset and block map for the first block. 834 */ 835 if (start_lvl == 0) { 836 sbap = &ip->i_din2->di_db[0]; 837 soff = start_lbn; 838 } else { 839 idp = &start_ap[start_lvl - 1]; 840 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { 841 brelse(sbp); 842 return (ENOSPC); 843 } 844 sbap = (ufs2_daddr_t *)sbp->b_data; 845 soff = idp->in_off; 846 } 847 /* 848 * If the block range spans two block maps, get the second map. 849 */ 850 ebap = NULL; 851 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 852 ssize = len; 853 } else { 854 #ifdef INVARIANTS 855 if (start_lvl > 0 && 856 start_ap[start_lvl - 1].in_lbn == idp->in_lbn) 857 panic("ffs_reallocblk: start == end"); 858 #endif 859 ssize = len - (idp->in_off + 1); 860 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) 861 goto fail; 862 ebap = (ufs2_daddr_t *)ebp->b_data; 863 } 864 /* 865 * Find the preferred location for the cluster. If we have not 866 * previously failed at this endeavor, then follow our standard 867 * preference calculation. If we have failed at it, then pick up 868 * where we last ended our search. 869 */ 870 UFS_LOCK(ump); 871 if (ip->i_nextclustercg == -1) 872 pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); 873 else 874 pref = cgdata(fs, ip->i_nextclustercg); 875 /* 876 * Search the block map looking for an allocation of the desired size. 877 * To avoid wasting too much time, we limit the number of cylinder 878 * groups that we will search. 879 */ 880 cg = dtog(fs, pref); 881 for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { 882 if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) 883 break; 884 cg += 1; 885 if (cg >= fs->fs_ncg) 886 cg = 0; 887 } 888 /* 889 * If we have failed in our search, record where we gave up for 890 * next time. Otherwise, fall back to our usual search citerion. 891 */ 892 if (newblk == 0) { 893 ip->i_nextclustercg = cg; 894 UFS_UNLOCK(ump); 895 goto fail; 896 } 897 ip->i_nextclustercg = -1; 898 /* 899 * We have found a new contiguous block. 900 * 901 * First we have to replace the old block pointers with the new 902 * block pointers in the inode and indirect blocks associated 903 * with the file. 904 */ 905 #ifdef DEBUG 906 if (prtrealloc) 907 printf("realloc: ino %ju, lbns %jd-%jd\n\told:", (uintmax_t)ip->i_number, 908 (intmax_t)start_lbn, (intmax_t)end_lbn); 909 #endif 910 blkno = newblk; 911 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { 912 if (i == ssize) { 913 bap = ebap; 914 soff = -i; 915 } 916 #ifdef INVARIANTS 917 if (!ffs_checkblk(ip, 918 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 919 panic("ffs_reallocblks: unallocated block 2"); 920 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) 921 panic("ffs_reallocblks: alloc mismatch"); 922 #endif 923 #ifdef DEBUG 924 if (prtrealloc) 925 printf(" %jd,", (intmax_t)*bap); 926 #endif 927 if (DOINGSOFTDEP(vp)) { 928 if (sbap == &ip->i_din2->di_db[0] && i < ssize) 929 softdep_setup_allocdirect(ip, start_lbn + i, 930 blkno, *bap, fs->fs_bsize, fs->fs_bsize, 931 buflist->bs_children[i]); 932 else 933 softdep_setup_allocindir_page(ip, start_lbn + i, 934 i < ssize ? sbp : ebp, soff + i, blkno, 935 *bap, buflist->bs_children[i]); 936 } 937 *bap++ = blkno; 938 } 939 /* 940 * Next we must write out the modified inode and indirect blocks. 941 * For strict correctness, the writes should be synchronous since 942 * the old block values may have been written to disk. In practise 943 * they are almost never written, but if we are concerned about 944 * strict correctness, the `doasyncfree' flag should be set to zero. 945 * 946 * The test on `doasyncfree' should be changed to test a flag 947 * that shows whether the associated buffers and inodes have 948 * been written. The flag should be set when the cluster is 949 * started and cleared whenever the buffer or inode is flushed. 950 * We can then check below to see if it is set, and do the 951 * synchronous write only when it has been cleared. 952 */ 953 if (sbap != &ip->i_din2->di_db[0]) { 954 if (doasyncfree) 955 bdwrite(sbp); 956 else 957 bwrite(sbp); 958 } else { 959 ip->i_flag |= IN_CHANGE | IN_UPDATE; 960 if (!doasyncfree) 961 ffs_update(vp, 1); 962 } 963 if (ssize < len) { 964 if (doasyncfree) 965 bdwrite(ebp); 966 else 967 bwrite(ebp); 968 } 969 /* 970 * Last, free the old blocks and assign the new blocks to the buffers. 971 */ 972 #ifdef DEBUG 973 if (prtrealloc) 974 printf("\n\tnew:"); 975 #endif 976 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { 977 if (!DOINGSOFTDEP(vp)) 978 ffs_blkfree(ump, fs, ump->um_devvp, 979 dbtofsb(fs, buflist->bs_children[i]->b_blkno), 980 fs->fs_bsize, ip->i_number, vp->v_type, NULL); 981 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 982 #ifdef INVARIANTS 983 if (!ffs_checkblk(ip, 984 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 985 panic("ffs_reallocblks: unallocated block 3"); 986 #endif 987 #ifdef DEBUG 988 if (prtrealloc) 989 printf(" %jd,", (intmax_t)blkno); 990 #endif 991 } 992 #ifdef DEBUG 993 if (prtrealloc) { 994 prtrealloc--; 995 printf("\n"); 996 } 997 #endif 998 return (0); 999 1000 fail: 1001 if (ssize < len) 1002 brelse(ebp); 1003 if (sbap != &ip->i_din2->di_db[0]) 1004 brelse(sbp); 1005 return (ENOSPC); 1006 } 1007 1008 /* 1009 * Allocate an inode in the filesystem. 1010 * 1011 * If allocating a directory, use ffs_dirpref to select the inode. 1012 * If allocating in a directory, the following hierarchy is followed: 1013 * 1) allocate the preferred inode. 1014 * 2) allocate an inode in the same cylinder group. 1015 * 3) quadradically rehash into other cylinder groups, until an 1016 * available inode is located. 1017 * If no inode preference is given the following hierarchy is used 1018 * to allocate an inode: 1019 * 1) allocate an inode in cylinder group 0. 1020 * 2) quadradically rehash into other cylinder groups, until an 1021 * available inode is located. 1022 */ 1023 int 1024 ffs_valloc(pvp, mode, cred, vpp) 1025 struct vnode *pvp; 1026 int mode; 1027 struct ucred *cred; 1028 struct vnode **vpp; 1029 { 1030 struct inode *pip; 1031 struct fs *fs; 1032 struct inode *ip; 1033 struct timespec ts; 1034 struct ufsmount *ump; 1035 ino_t ino, ipref; 1036 u_int cg; 1037 int error, error1, reclaimed; 1038 static struct timeval lastfail; 1039 static int curfail; 1040 1041 *vpp = NULL; 1042 pip = VTOI(pvp); 1043 ump = ITOUMP(pip); 1044 fs = ump->um_fs; 1045 1046 UFS_LOCK(ump); 1047 reclaimed = 0; 1048 retry: 1049 if (fs->fs_cstotal.cs_nifree == 0) 1050 goto noinodes; 1051 1052 if ((mode & IFMT) == IFDIR) 1053 ipref = ffs_dirpref(pip); 1054 else 1055 ipref = pip->i_number; 1056 if (ipref >= fs->fs_ncg * fs->fs_ipg) 1057 ipref = 0; 1058 cg = ino_to_cg(fs, ipref); 1059 /* 1060 * Track number of dirs created one after another 1061 * in a same cg without intervening by files. 1062 */ 1063 if ((mode & IFMT) == IFDIR) { 1064 if (fs->fs_contigdirs[cg] < 255) 1065 fs->fs_contigdirs[cg]++; 1066 } else { 1067 if (fs->fs_contigdirs[cg] > 0) 1068 fs->fs_contigdirs[cg]--; 1069 } 1070 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, 1071 (allocfcn_t *)ffs_nodealloccg); 1072 if (ino == 0) 1073 goto noinodes; 1074 error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp); 1075 if (error) { 1076 error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp, 1077 FFSV_FORCEINSMQ); 1078 ffs_vfree(pvp, ino, mode); 1079 if (error1 == 0) { 1080 ip = VTOI(*vpp); 1081 if (ip->i_mode) 1082 goto dup_alloc; 1083 ip->i_flag |= IN_MODIFIED; 1084 vput(*vpp); 1085 } 1086 return (error); 1087 } 1088 ip = VTOI(*vpp); 1089 if (ip->i_mode) { 1090 dup_alloc: 1091 printf("mode = 0%o, inum = %ju, fs = %s\n", 1092 ip->i_mode, (uintmax_t)ip->i_number, fs->fs_fsmnt); 1093 panic("ffs_valloc: dup alloc"); 1094 } 1095 if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */ 1096 printf("free inode %s/%lu had %ld blocks\n", 1097 fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks)); 1098 DIP_SET(ip, i_blocks, 0); 1099 } 1100 ip->i_flags = 0; 1101 DIP_SET(ip, i_flags, 0); 1102 /* 1103 * Set up a new generation number for this inode. 1104 */ 1105 while (ip->i_gen == 0 || ++ip->i_gen == 0) 1106 ip->i_gen = arc4random(); 1107 DIP_SET(ip, i_gen, ip->i_gen); 1108 if (fs->fs_magic == FS_UFS2_MAGIC) { 1109 vfs_timestamp(&ts); 1110 ip->i_din2->di_birthtime = ts.tv_sec; 1111 ip->i_din2->di_birthnsec = ts.tv_nsec; 1112 } 1113 ufs_prepare_reclaim(*vpp); 1114 ip->i_flag = 0; 1115 (*vpp)->v_vflag = 0; 1116 (*vpp)->v_type = VNON; 1117 if (fs->fs_magic == FS_UFS2_MAGIC) { 1118 (*vpp)->v_op = &ffs_vnodeops2; 1119 ip->i_flag |= IN_UFS2; 1120 } else { 1121 (*vpp)->v_op = &ffs_vnodeops1; 1122 } 1123 return (0); 1124 noinodes: 1125 if (reclaimed == 0) { 1126 reclaimed = 1; 1127 softdep_request_cleanup(fs, pvp, cred, FLUSH_INODES_WAIT); 1128 goto retry; 1129 } 1130 UFS_UNLOCK(ump); 1131 if (ppsratecheck(&lastfail, &curfail, 1)) { 1132 ffs_fserr(fs, pip->i_number, "out of inodes"); 1133 uprintf("\n%s: create/symlink failed, no inodes free\n", 1134 fs->fs_fsmnt); 1135 } 1136 return (ENOSPC); 1137 } 1138 1139 /* 1140 * Find a cylinder group to place a directory. 1141 * 1142 * The policy implemented by this algorithm is to allocate a 1143 * directory inode in the same cylinder group as its parent 1144 * directory, but also to reserve space for its files inodes 1145 * and data. Restrict the number of directories which may be 1146 * allocated one after another in the same cylinder group 1147 * without intervening allocation of files. 1148 * 1149 * If we allocate a first level directory then force allocation 1150 * in another cylinder group. 1151 */ 1152 static ino_t 1153 ffs_dirpref(pip) 1154 struct inode *pip; 1155 { 1156 struct fs *fs; 1157 int cg, prefcg, dirsize, cgsize; 1158 u_int avgifree, avgbfree, avgndir, curdirsize; 1159 u_int minifree, minbfree, maxndir; 1160 u_int mincg, minndir; 1161 u_int maxcontigdirs; 1162 1163 mtx_assert(UFS_MTX(ITOUMP(pip)), MA_OWNED); 1164 fs = ITOFS(pip); 1165 1166 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 1167 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1168 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; 1169 1170 /* 1171 * Force allocation in another cg if creating a first level dir. 1172 */ 1173 ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref"); 1174 if (ITOV(pip)->v_vflag & VV_ROOT) { 1175 prefcg = arc4random() % fs->fs_ncg; 1176 mincg = prefcg; 1177 minndir = fs->fs_ipg; 1178 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1179 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 1180 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 1181 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1182 mincg = cg; 1183 minndir = fs->fs_cs(fs, cg).cs_ndir; 1184 } 1185 for (cg = 0; cg < prefcg; cg++) 1186 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 1187 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 1188 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1189 mincg = cg; 1190 minndir = fs->fs_cs(fs, cg).cs_ndir; 1191 } 1192 return ((ino_t)(fs->fs_ipg * mincg)); 1193 } 1194 1195 /* 1196 * Count various limits which used for 1197 * optimal allocation of a directory inode. 1198 */ 1199 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); 1200 minifree = avgifree - avgifree / 4; 1201 if (minifree < 1) 1202 minifree = 1; 1203 minbfree = avgbfree - avgbfree / 4; 1204 if (minbfree < 1) 1205 minbfree = 1; 1206 cgsize = fs->fs_fsize * fs->fs_fpg; 1207 dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir; 1208 curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0; 1209 if (dirsize < curdirsize) 1210 dirsize = curdirsize; 1211 if (dirsize <= 0) 1212 maxcontigdirs = 0; /* dirsize overflowed */ 1213 else 1214 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255); 1215 if (fs->fs_avgfpdir > 0) 1216 maxcontigdirs = min(maxcontigdirs, 1217 fs->fs_ipg / fs->fs_avgfpdir); 1218 if (maxcontigdirs == 0) 1219 maxcontigdirs = 1; 1220 1221 /* 1222 * Limit number of dirs in one cg and reserve space for 1223 * regular files, but only if we have no deficit in 1224 * inodes or space. 1225 * 1226 * We are trying to find a suitable cylinder group nearby 1227 * our preferred cylinder group to place a new directory. 1228 * We scan from our preferred cylinder group forward looking 1229 * for a cylinder group that meets our criterion. If we get 1230 * to the final cylinder group and do not find anything, 1231 * we start scanning forwards from the beginning of the 1232 * filesystem. While it might seem sensible to start scanning 1233 * backwards or even to alternate looking forward and backward, 1234 * this approach fails badly when the filesystem is nearly full. 1235 * Specifically, we first search all the areas that have no space 1236 * and finally try the one preceding that. We repeat this on 1237 * every request and in the case of the final block end up 1238 * searching the entire filesystem. By jumping to the front 1239 * of the filesystem, our future forward searches always look 1240 * in new cylinder groups so finds every possible block after 1241 * one pass over the filesystem. 1242 */ 1243 prefcg = ino_to_cg(fs, pip->i_number); 1244 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1245 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 1246 fs->fs_cs(fs, cg).cs_nifree >= minifree && 1247 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 1248 if (fs->fs_contigdirs[cg] < maxcontigdirs) 1249 return ((ino_t)(fs->fs_ipg * cg)); 1250 } 1251 for (cg = 0; cg < prefcg; cg++) 1252 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 1253 fs->fs_cs(fs, cg).cs_nifree >= minifree && 1254 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 1255 if (fs->fs_contigdirs[cg] < maxcontigdirs) 1256 return ((ino_t)(fs->fs_ipg * cg)); 1257 } 1258 /* 1259 * This is a backstop when we have deficit in space. 1260 */ 1261 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1262 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 1263 return ((ino_t)(fs->fs_ipg * cg)); 1264 for (cg = 0; cg < prefcg; cg++) 1265 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 1266 break; 1267 return ((ino_t)(fs->fs_ipg * cg)); 1268 } 1269 1270 /* 1271 * Select the desired position for the next block in a file. The file is 1272 * logically divided into sections. The first section is composed of the 1273 * direct blocks and the next fs_maxbpg blocks. Each additional section 1274 * contains fs_maxbpg blocks. 1275 * 1276 * If no blocks have been allocated in the first section, the policy is to 1277 * request a block in the same cylinder group as the inode that describes 1278 * the file. The first indirect is allocated immediately following the last 1279 * direct block and the data blocks for the first indirect immediately 1280 * follow it. 1281 * 1282 * If no blocks have been allocated in any other section, the indirect 1283 * block(s) are allocated in the same cylinder group as its inode in an 1284 * area reserved immediately following the inode blocks. The policy for 1285 * the data blocks is to place them in a cylinder group with a greater than 1286 * average number of free blocks. An appropriate cylinder group is found 1287 * by using a rotor that sweeps the cylinder groups. When a new group of 1288 * blocks is needed, the sweep begins in the cylinder group following the 1289 * cylinder group from which the previous allocation was made. The sweep 1290 * continues until a cylinder group with greater than the average number 1291 * of free blocks is found. If the allocation is for the first block in an 1292 * indirect block or the previous block is a hole, then the information on 1293 * the previous allocation is unavailable; here a best guess is made based 1294 * on the logical block number being allocated. 1295 * 1296 * If a section is already partially allocated, the policy is to 1297 * allocate blocks contiguously within the section if possible. 1298 */ 1299 ufs2_daddr_t 1300 ffs_blkpref_ufs1(ip, lbn, indx, bap) 1301 struct inode *ip; 1302 ufs_lbn_t lbn; 1303 int indx; 1304 ufs1_daddr_t *bap; 1305 { 1306 struct fs *fs; 1307 u_int cg, inocg; 1308 u_int avgbfree, startcg; 1309 ufs2_daddr_t pref; 1310 1311 KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); 1312 mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); 1313 fs = ITOFS(ip); 1314 /* 1315 * Allocation of indirect blocks is indicated by passing negative 1316 * values in indx: -1 for single indirect, -2 for double indirect, 1317 * -3 for triple indirect. As noted below, we attempt to allocate 1318 * the first indirect inline with the file data. For all later 1319 * indirect blocks, the data is often allocated in other cylinder 1320 * groups. However to speed random file access and to speed up 1321 * fsck, the filesystem reserves the first fs_metaspace blocks 1322 * (typically half of fs_minfree) of the data area of each cylinder 1323 * group to hold these later indirect blocks. 1324 */ 1325 inocg = ino_to_cg(fs, ip->i_number); 1326 if (indx < 0) { 1327 /* 1328 * Our preference for indirect blocks is the zone at the 1329 * beginning of the inode's cylinder group data area that 1330 * we try to reserve for indirect blocks. 1331 */ 1332 pref = cgmeta(fs, inocg); 1333 /* 1334 * If we are allocating the first indirect block, try to 1335 * place it immediately following the last direct block. 1336 */ 1337 if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && 1338 ip->i_din1->di_db[UFS_NDADDR - 1] != 0) 1339 pref = ip->i_din1->di_db[UFS_NDADDR - 1] + fs->fs_frag; 1340 return (pref); 1341 } 1342 /* 1343 * If we are allocating the first data block in the first indirect 1344 * block and the indirect has been allocated in the data block area, 1345 * try to place it immediately following the indirect block. 1346 */ 1347 if (lbn == UFS_NDADDR) { 1348 pref = ip->i_din1->di_ib[0]; 1349 if (pref != 0 && pref >= cgdata(fs, inocg) && 1350 pref < cgbase(fs, inocg + 1)) 1351 return (pref + fs->fs_frag); 1352 } 1353 /* 1354 * If we are at the beginning of a file, or we have already allocated 1355 * the maximum number of blocks per cylinder group, or we do not 1356 * have a block allocated immediately preceding us, then we need 1357 * to decide where to start allocating new blocks. 1358 */ 1359 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 1360 /* 1361 * If we are allocating a directory data block, we want 1362 * to place it in the metadata area. 1363 */ 1364 if ((ip->i_mode & IFMT) == IFDIR) 1365 return (cgmeta(fs, inocg)); 1366 /* 1367 * Until we fill all the direct and all the first indirect's 1368 * blocks, we try to allocate in the data area of the inode's 1369 * cylinder group. 1370 */ 1371 if (lbn < UFS_NDADDR + NINDIR(fs)) 1372 return (cgdata(fs, inocg)); 1373 /* 1374 * Find a cylinder with greater than average number of 1375 * unused data blocks. 1376 */ 1377 if (indx == 0 || bap[indx - 1] == 0) 1378 startcg = inocg + lbn / fs->fs_maxbpg; 1379 else 1380 startcg = dtog(fs, bap[indx - 1]) + 1; 1381 startcg %= fs->fs_ncg; 1382 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1383 for (cg = startcg; cg < fs->fs_ncg; cg++) 1384 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1385 fs->fs_cgrotor = cg; 1386 return (cgdata(fs, cg)); 1387 } 1388 for (cg = 0; cg <= startcg; cg++) 1389 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1390 fs->fs_cgrotor = cg; 1391 return (cgdata(fs, cg)); 1392 } 1393 return (0); 1394 } 1395 /* 1396 * Otherwise, we just always try to lay things out contiguously. 1397 */ 1398 return (bap[indx - 1] + fs->fs_frag); 1399 } 1400 1401 /* 1402 * Same as above, but for UFS2 1403 */ 1404 ufs2_daddr_t 1405 ffs_blkpref_ufs2(ip, lbn, indx, bap) 1406 struct inode *ip; 1407 ufs_lbn_t lbn; 1408 int indx; 1409 ufs2_daddr_t *bap; 1410 { 1411 struct fs *fs; 1412 u_int cg, inocg; 1413 u_int avgbfree, startcg; 1414 ufs2_daddr_t pref; 1415 1416 KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); 1417 mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); 1418 fs = ITOFS(ip); 1419 /* 1420 * Allocation of indirect blocks is indicated by passing negative 1421 * values in indx: -1 for single indirect, -2 for double indirect, 1422 * -3 for triple indirect. As noted below, we attempt to allocate 1423 * the first indirect inline with the file data. For all later 1424 * indirect blocks, the data is often allocated in other cylinder 1425 * groups. However to speed random file access and to speed up 1426 * fsck, the filesystem reserves the first fs_metaspace blocks 1427 * (typically half of fs_minfree) of the data area of each cylinder 1428 * group to hold these later indirect blocks. 1429 */ 1430 inocg = ino_to_cg(fs, ip->i_number); 1431 if (indx < 0) { 1432 /* 1433 * Our preference for indirect blocks is the zone at the 1434 * beginning of the inode's cylinder group data area that 1435 * we try to reserve for indirect blocks. 1436 */ 1437 pref = cgmeta(fs, inocg); 1438 /* 1439 * If we are allocating the first indirect block, try to 1440 * place it immediately following the last direct block. 1441 */ 1442 if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && 1443 ip->i_din2->di_db[UFS_NDADDR - 1] != 0) 1444 pref = ip->i_din2->di_db[UFS_NDADDR - 1] + fs->fs_frag; 1445 return (pref); 1446 } 1447 /* 1448 * If we are allocating the first data block in the first indirect 1449 * block and the indirect has been allocated in the data block area, 1450 * try to place it immediately following the indirect block. 1451 */ 1452 if (lbn == UFS_NDADDR) { 1453 pref = ip->i_din2->di_ib[0]; 1454 if (pref != 0 && pref >= cgdata(fs, inocg) && 1455 pref < cgbase(fs, inocg + 1)) 1456 return (pref + fs->fs_frag); 1457 } 1458 /* 1459 * If we are at the beginning of a file, or we have already allocated 1460 * the maximum number of blocks per cylinder group, or we do not 1461 * have a block allocated immediately preceding us, then we need 1462 * to decide where to start allocating new blocks. 1463 */ 1464 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 1465 /* 1466 * If we are allocating a directory data block, we want 1467 * to place it in the metadata area. 1468 */ 1469 if ((ip->i_mode & IFMT) == IFDIR) 1470 return (cgmeta(fs, inocg)); 1471 /* 1472 * Until we fill all the direct and all the first indirect's 1473 * blocks, we try to allocate in the data area of the inode's 1474 * cylinder group. 1475 */ 1476 if (lbn < UFS_NDADDR + NINDIR(fs)) 1477 return (cgdata(fs, inocg)); 1478 /* 1479 * Find a cylinder with greater than average number of 1480 * unused data blocks. 1481 */ 1482 if (indx == 0 || bap[indx - 1] == 0) 1483 startcg = inocg + lbn / fs->fs_maxbpg; 1484 else 1485 startcg = dtog(fs, bap[indx - 1]) + 1; 1486 startcg %= fs->fs_ncg; 1487 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1488 for (cg = startcg; cg < fs->fs_ncg; cg++) 1489 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1490 fs->fs_cgrotor = cg; 1491 return (cgdata(fs, cg)); 1492 } 1493 for (cg = 0; cg <= startcg; cg++) 1494 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1495 fs->fs_cgrotor = cg; 1496 return (cgdata(fs, cg)); 1497 } 1498 return (0); 1499 } 1500 /* 1501 * Otherwise, we just always try to lay things out contiguously. 1502 */ 1503 return (bap[indx - 1] + fs->fs_frag); 1504 } 1505 1506 /* 1507 * Implement the cylinder overflow algorithm. 1508 * 1509 * The policy implemented by this algorithm is: 1510 * 1) allocate the block in its requested cylinder group. 1511 * 2) quadradically rehash on the cylinder group number. 1512 * 3) brute force search for a free block. 1513 * 1514 * Must be called with the UFS lock held. Will release the lock on success 1515 * and return with it held on failure. 1516 */ 1517 /*VARARGS5*/ 1518 static ufs2_daddr_t 1519 ffs_hashalloc(ip, cg, pref, size, rsize, allocator) 1520 struct inode *ip; 1521 u_int cg; 1522 ufs2_daddr_t pref; 1523 int size; /* Search size for data blocks, mode for inodes */ 1524 int rsize; /* Real allocated size. */ 1525 allocfcn_t *allocator; 1526 { 1527 struct fs *fs; 1528 ufs2_daddr_t result; 1529 u_int i, icg = cg; 1530 1531 mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); 1532 #ifdef INVARIANTS 1533 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) 1534 panic("ffs_hashalloc: allocation on suspended filesystem"); 1535 #endif 1536 fs = ITOFS(ip); 1537 /* 1538 * 1: preferred cylinder group 1539 */ 1540 result = (*allocator)(ip, cg, pref, size, rsize); 1541 if (result) 1542 return (result); 1543 /* 1544 * 2: quadratic rehash 1545 */ 1546 for (i = 1; i < fs->fs_ncg; i *= 2) { 1547 cg += i; 1548 if (cg >= fs->fs_ncg) 1549 cg -= fs->fs_ncg; 1550 result = (*allocator)(ip, cg, 0, size, rsize); 1551 if (result) 1552 return (result); 1553 } 1554 /* 1555 * 3: brute force search 1556 * Note that we start at i == 2, since 0 was checked initially, 1557 * and 1 is always checked in the quadratic rehash. 1558 */ 1559 cg = (icg + 2) % fs->fs_ncg; 1560 for (i = 2; i < fs->fs_ncg; i++) { 1561 result = (*allocator)(ip, cg, 0, size, rsize); 1562 if (result) 1563 return (result); 1564 cg++; 1565 if (cg == fs->fs_ncg) 1566 cg = 0; 1567 } 1568 return (0); 1569 } 1570 1571 /* 1572 * Determine whether a fragment can be extended. 1573 * 1574 * Check to see if the necessary fragments are available, and 1575 * if they are, allocate them. 1576 */ 1577 static ufs2_daddr_t 1578 ffs_fragextend(ip, cg, bprev, osize, nsize) 1579 struct inode *ip; 1580 u_int cg; 1581 ufs2_daddr_t bprev; 1582 int osize, nsize; 1583 { 1584 struct fs *fs; 1585 struct cg *cgp; 1586 struct buf *bp; 1587 struct ufsmount *ump; 1588 int nffree; 1589 long bno; 1590 int frags, bbase; 1591 int i, error; 1592 u_int8_t *blksfree; 1593 1594 ump = ITOUMP(ip); 1595 fs = ump->um_fs; 1596 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 1597 return (0); 1598 frags = numfrags(fs, nsize); 1599 bbase = fragnum(fs, bprev); 1600 if (bbase > fragnum(fs, (bprev + frags - 1))) { 1601 /* cannot extend across a block boundary */ 1602 return (0); 1603 } 1604 UFS_UNLOCK(ump); 1605 if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0) 1606 goto fail; 1607 bno = dtogd(fs, bprev); 1608 blksfree = cg_blksfree(cgp); 1609 for (i = numfrags(fs, osize); i < frags; i++) 1610 if (isclr(blksfree, bno + i)) 1611 goto fail; 1612 /* 1613 * the current fragment can be extended 1614 * deduct the count on fragment being extended into 1615 * increase the count on the remaining fragment (if any) 1616 * allocate the extended piece 1617 */ 1618 for (i = frags; i < fs->fs_frag - bbase; i++) 1619 if (isclr(blksfree, bno + i)) 1620 break; 1621 cgp->cg_frsum[i - numfrags(fs, osize)]--; 1622 if (i != frags) 1623 cgp->cg_frsum[i - frags]++; 1624 for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) { 1625 clrbit(blksfree, bno + i); 1626 cgp->cg_cs.cs_nffree--; 1627 nffree++; 1628 } 1629 UFS_LOCK(ump); 1630 fs->fs_cstotal.cs_nffree -= nffree; 1631 fs->fs_cs(fs, cg).cs_nffree -= nffree; 1632 fs->fs_fmod = 1; 1633 ACTIVECLEAR(fs, cg); 1634 UFS_UNLOCK(ump); 1635 if (DOINGSOFTDEP(ITOV(ip))) 1636 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev, 1637 frags, numfrags(fs, osize)); 1638 bdwrite(bp); 1639 return (bprev); 1640 1641 fail: 1642 brelse(bp); 1643 UFS_LOCK(ump); 1644 return (0); 1645 1646 } 1647 1648 /* 1649 * Determine whether a block can be allocated. 1650 * 1651 * Check to see if a block of the appropriate size is available, 1652 * and if it is, allocate it. 1653 */ 1654 static ufs2_daddr_t 1655 ffs_alloccg(ip, cg, bpref, size, rsize) 1656 struct inode *ip; 1657 u_int cg; 1658 ufs2_daddr_t bpref; 1659 int size; 1660 int rsize; 1661 { 1662 struct fs *fs; 1663 struct cg *cgp; 1664 struct buf *bp; 1665 struct ufsmount *ump; 1666 ufs1_daddr_t bno; 1667 ufs2_daddr_t blkno; 1668 int i, allocsiz, error, frags; 1669 u_int8_t *blksfree; 1670 1671 ump = ITOUMP(ip); 1672 fs = ump->um_fs; 1673 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 1674 return (0); 1675 UFS_UNLOCK(ump); 1676 if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0 || 1677 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) 1678 goto fail; 1679 if (size == fs->fs_bsize) { 1680 UFS_LOCK(ump); 1681 blkno = ffs_alloccgblk(ip, bp, bpref, rsize); 1682 ACTIVECLEAR(fs, cg); 1683 UFS_UNLOCK(ump); 1684 bdwrite(bp); 1685 return (blkno); 1686 } 1687 /* 1688 * check to see if any fragments are already available 1689 * allocsiz is the size which will be allocated, hacking 1690 * it down to a smaller size if necessary 1691 */ 1692 blksfree = cg_blksfree(cgp); 1693 frags = numfrags(fs, size); 1694 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 1695 if (cgp->cg_frsum[allocsiz] != 0) 1696 break; 1697 if (allocsiz == fs->fs_frag) { 1698 /* 1699 * no fragments were available, so a block will be 1700 * allocated, and hacked up 1701 */ 1702 if (cgp->cg_cs.cs_nbfree == 0) 1703 goto fail; 1704 UFS_LOCK(ump); 1705 blkno = ffs_alloccgblk(ip, bp, bpref, rsize); 1706 ACTIVECLEAR(fs, cg); 1707 UFS_UNLOCK(ump); 1708 bdwrite(bp); 1709 return (blkno); 1710 } 1711 KASSERT(size == rsize, 1712 ("ffs_alloccg: size(%d) != rsize(%d)", size, rsize)); 1713 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 1714 if (bno < 0) 1715 goto fail; 1716 for (i = 0; i < frags; i++) 1717 clrbit(blksfree, bno + i); 1718 cgp->cg_cs.cs_nffree -= frags; 1719 cgp->cg_frsum[allocsiz]--; 1720 if (frags != allocsiz) 1721 cgp->cg_frsum[allocsiz - frags]++; 1722 UFS_LOCK(ump); 1723 fs->fs_cstotal.cs_nffree -= frags; 1724 fs->fs_cs(fs, cg).cs_nffree -= frags; 1725 fs->fs_fmod = 1; 1726 blkno = cgbase(fs, cg) + bno; 1727 ACTIVECLEAR(fs, cg); 1728 UFS_UNLOCK(ump); 1729 if (DOINGSOFTDEP(ITOV(ip))) 1730 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, frags, 0); 1731 bdwrite(bp); 1732 return (blkno); 1733 1734 fail: 1735 brelse(bp); 1736 UFS_LOCK(ump); 1737 return (0); 1738 } 1739 1740 /* 1741 * Allocate a block in a cylinder group. 1742 * 1743 * This algorithm implements the following policy: 1744 * 1) allocate the requested block. 1745 * 2) allocate a rotationally optimal block in the same cylinder. 1746 * 3) allocate the next available block on the block rotor for the 1747 * specified cylinder group. 1748 * Note that this routine only allocates fs_bsize blocks; these 1749 * blocks may be fragmented by the routine that allocates them. 1750 */ 1751 static ufs2_daddr_t 1752 ffs_alloccgblk(ip, bp, bpref, size) 1753 struct inode *ip; 1754 struct buf *bp; 1755 ufs2_daddr_t bpref; 1756 int size; 1757 { 1758 struct fs *fs; 1759 struct cg *cgp; 1760 struct ufsmount *ump; 1761 ufs1_daddr_t bno; 1762 ufs2_daddr_t blkno; 1763 u_int8_t *blksfree; 1764 int i, cgbpref; 1765 1766 ump = ITOUMP(ip); 1767 fs = ump->um_fs; 1768 mtx_assert(UFS_MTX(ump), MA_OWNED); 1769 cgp = (struct cg *)bp->b_data; 1770 blksfree = cg_blksfree(cgp); 1771 if (bpref == 0) { 1772 bpref = cgbase(fs, cgp->cg_cgx) + cgp->cg_rotor + fs->fs_frag; 1773 } else if ((cgbpref = dtog(fs, bpref)) != cgp->cg_cgx) { 1774 /* map bpref to correct zone in this cg */ 1775 if (bpref < cgdata(fs, cgbpref)) 1776 bpref = cgmeta(fs, cgp->cg_cgx); 1777 else 1778 bpref = cgdata(fs, cgp->cg_cgx); 1779 } 1780 /* 1781 * if the requested block is available, use it 1782 */ 1783 bno = dtogd(fs, blknum(fs, bpref)); 1784 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) 1785 goto gotit; 1786 /* 1787 * Take the next available block in this cylinder group. 1788 */ 1789 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1790 if (bno < 0) 1791 return (0); 1792 /* Update cg_rotor only if allocated from the data zone */ 1793 if (bno >= dtogd(fs, cgdata(fs, cgp->cg_cgx))) 1794 cgp->cg_rotor = bno; 1795 gotit: 1796 blkno = fragstoblks(fs, bno); 1797 ffs_clrblock(fs, blksfree, (long)blkno); 1798 ffs_clusteracct(fs, cgp, blkno, -1); 1799 cgp->cg_cs.cs_nbfree--; 1800 fs->fs_cstotal.cs_nbfree--; 1801 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; 1802 fs->fs_fmod = 1; 1803 blkno = cgbase(fs, cgp->cg_cgx) + bno; 1804 /* 1805 * If the caller didn't want the whole block free the frags here. 1806 */ 1807 size = numfrags(fs, size); 1808 if (size != fs->fs_frag) { 1809 bno = dtogd(fs, blkno); 1810 for (i = size; i < fs->fs_frag; i++) 1811 setbit(blksfree, bno + i); 1812 i = fs->fs_frag - size; 1813 cgp->cg_cs.cs_nffree += i; 1814 fs->fs_cstotal.cs_nffree += i; 1815 fs->fs_cs(fs, cgp->cg_cgx).cs_nffree += i; 1816 fs->fs_fmod = 1; 1817 cgp->cg_frsum[i]++; 1818 } 1819 /* XXX Fixme. */ 1820 UFS_UNLOCK(ump); 1821 if (DOINGSOFTDEP(ITOV(ip))) 1822 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, 1823 size, 0); 1824 UFS_LOCK(ump); 1825 return (blkno); 1826 } 1827 1828 /* 1829 * Determine whether a cluster can be allocated. 1830 * 1831 * We do not currently check for optimal rotational layout if there 1832 * are multiple choices in the same cylinder group. Instead we just 1833 * take the first one that we find following bpref. 1834 */ 1835 static ufs2_daddr_t 1836 ffs_clusteralloc(ip, cg, bpref, len) 1837 struct inode *ip; 1838 u_int cg; 1839 ufs2_daddr_t bpref; 1840 int len; 1841 { 1842 struct fs *fs; 1843 struct cg *cgp; 1844 struct buf *bp; 1845 struct ufsmount *ump; 1846 int i, run, bit, map, got, error; 1847 ufs2_daddr_t bno; 1848 u_char *mapp; 1849 int32_t *lp; 1850 u_int8_t *blksfree; 1851 1852 ump = ITOUMP(ip); 1853 fs = ump->um_fs; 1854 if (fs->fs_maxcluster[cg] < len) 1855 return (0); 1856 UFS_UNLOCK(ump); 1857 if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0) { 1858 UFS_LOCK(ump); 1859 return (0); 1860 } 1861 /* 1862 * Check to see if a cluster of the needed size (or bigger) is 1863 * available in this cylinder group. 1864 */ 1865 lp = &cg_clustersum(cgp)[len]; 1866 for (i = len; i <= fs->fs_contigsumsize; i++) 1867 if (*lp++ > 0) 1868 break; 1869 if (i > fs->fs_contigsumsize) { 1870 /* 1871 * This is the first time looking for a cluster in this 1872 * cylinder group. Update the cluster summary information 1873 * to reflect the true maximum sized cluster so that 1874 * future cluster allocation requests can avoid reading 1875 * the cylinder group map only to find no clusters. 1876 */ 1877 lp = &cg_clustersum(cgp)[len - 1]; 1878 for (i = len - 1; i > 0; i--) 1879 if (*lp-- > 0) 1880 break; 1881 UFS_LOCK(ump); 1882 fs->fs_maxcluster[cg] = i; 1883 brelse(bp); 1884 return (0); 1885 } 1886 /* 1887 * Search the cluster map to find a big enough cluster. 1888 * We take the first one that we find, even if it is larger 1889 * than we need as we prefer to get one close to the previous 1890 * block allocation. We do not search before the current 1891 * preference point as we do not want to allocate a block 1892 * that is allocated before the previous one (as we will 1893 * then have to wait for another pass of the elevator 1894 * algorithm before it will be read). We prefer to fail and 1895 * be recalled to try an allocation in the next cylinder group. 1896 */ 1897 if (dtog(fs, bpref) != cg) 1898 bpref = cgdata(fs, cg); 1899 else 1900 bpref = blknum(fs, bpref); 1901 bpref = fragstoblks(fs, dtogd(fs, bpref)); 1902 mapp = &cg_clustersfree(cgp)[bpref / NBBY]; 1903 map = *mapp++; 1904 bit = 1 << (bpref % NBBY); 1905 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { 1906 if ((map & bit) == 0) { 1907 run = 0; 1908 } else { 1909 run++; 1910 if (run == len) 1911 break; 1912 } 1913 if ((got & (NBBY - 1)) != (NBBY - 1)) { 1914 bit <<= 1; 1915 } else { 1916 map = *mapp++; 1917 bit = 1; 1918 } 1919 } 1920 if (got >= cgp->cg_nclusterblks) { 1921 UFS_LOCK(ump); 1922 brelse(bp); 1923 return (0); 1924 } 1925 /* 1926 * Allocate the cluster that we have found. 1927 */ 1928 blksfree = cg_blksfree(cgp); 1929 for (i = 1; i <= len; i++) 1930 if (!ffs_isblock(fs, blksfree, got - run + i)) 1931 panic("ffs_clusteralloc: map mismatch"); 1932 bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1); 1933 if (dtog(fs, bno) != cg) 1934 panic("ffs_clusteralloc: allocated out of group"); 1935 len = blkstofrags(fs, len); 1936 UFS_LOCK(ump); 1937 for (i = 0; i < len; i += fs->fs_frag) 1938 if (ffs_alloccgblk(ip, bp, bno + i, fs->fs_bsize) != bno + i) 1939 panic("ffs_clusteralloc: lost block"); 1940 ACTIVECLEAR(fs, cg); 1941 UFS_UNLOCK(ump); 1942 bdwrite(bp); 1943 return (bno); 1944 } 1945 1946 static inline struct buf * 1947 getinobuf(struct inode *ip, u_int cg, u_int32_t cginoblk, int gbflags) 1948 { 1949 struct fs *fs; 1950 1951 fs = ITOFS(ip); 1952 return (getblk(ITODEVVP(ip), fsbtodb(fs, ino_to_fsba(fs, 1953 cg * fs->fs_ipg + cginoblk)), (int)fs->fs_bsize, 0, 0, 1954 gbflags)); 1955 } 1956 1957 /* 1958 * Determine whether an inode can be allocated. 1959 * 1960 * Check to see if an inode is available, and if it is, 1961 * allocate it using the following policy: 1962 * 1) allocate the requested inode. 1963 * 2) allocate the next available inode after the requested 1964 * inode in the specified cylinder group. 1965 */ 1966 static ufs2_daddr_t 1967 ffs_nodealloccg(ip, cg, ipref, mode, unused) 1968 struct inode *ip; 1969 u_int cg; 1970 ufs2_daddr_t ipref; 1971 int mode; 1972 int unused; 1973 { 1974 struct fs *fs; 1975 struct cg *cgp; 1976 struct buf *bp, *ibp; 1977 struct ufsmount *ump; 1978 u_int8_t *inosused, *loc; 1979 struct ufs2_dinode *dp2; 1980 int error, start, len, i; 1981 u_int32_t old_initediblk; 1982 1983 ump = ITOUMP(ip); 1984 fs = ump->um_fs; 1985 check_nifree: 1986 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1987 return (0); 1988 UFS_UNLOCK(ump); 1989 if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0) { 1990 UFS_LOCK(ump); 1991 return (0); 1992 } 1993 restart: 1994 if (cgp->cg_cs.cs_nifree == 0) { 1995 brelse(bp); 1996 UFS_LOCK(ump); 1997 return (0); 1998 } 1999 inosused = cg_inosused(cgp); 2000 if (ipref) { 2001 ipref %= fs->fs_ipg; 2002 if (isclr(inosused, ipref)) 2003 goto gotit; 2004 } 2005 start = cgp->cg_irotor / NBBY; 2006 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); 2007 loc = memcchr(&inosused[start], 0xff, len); 2008 if (loc == NULL) { 2009 len = start + 1; 2010 start = 0; 2011 loc = memcchr(&inosused[start], 0xff, len); 2012 if (loc == NULL) { 2013 printf("cg = %d, irotor = %ld, fs = %s\n", 2014 cg, (long)cgp->cg_irotor, fs->fs_fsmnt); 2015 panic("ffs_nodealloccg: map corrupted"); 2016 /* NOTREACHED */ 2017 } 2018 } 2019 ipref = (loc - inosused) * NBBY + ffs(~*loc) - 1; 2020 gotit: 2021 /* 2022 * Check to see if we need to initialize more inodes. 2023 */ 2024 if (fs->fs_magic == FS_UFS2_MAGIC && 2025 ipref + INOPB(fs) > cgp->cg_initediblk && 2026 cgp->cg_initediblk < cgp->cg_niblk) { 2027 old_initediblk = cgp->cg_initediblk; 2028 2029 /* 2030 * Free the cylinder group lock before writing the 2031 * initialized inode block. Entering the 2032 * babarrierwrite() with the cylinder group lock 2033 * causes lock order violation between the lock and 2034 * snaplk. 2035 * 2036 * Another thread can decide to initialize the same 2037 * inode block, but whichever thread first gets the 2038 * cylinder group lock after writing the newly 2039 * allocated inode block will update it and the other 2040 * will realize that it has lost and leave the 2041 * cylinder group unchanged. 2042 */ 2043 ibp = getinobuf(ip, cg, old_initediblk, GB_LOCK_NOWAIT); 2044 brelse(bp); 2045 if (ibp == NULL) { 2046 /* 2047 * The inode block buffer is already owned by 2048 * another thread, which must initialize it. 2049 * Wait on the buffer to allow another thread 2050 * to finish the updates, with dropped cg 2051 * buffer lock, then retry. 2052 */ 2053 ibp = getinobuf(ip, cg, old_initediblk, 0); 2054 brelse(ibp); 2055 UFS_LOCK(ump); 2056 goto check_nifree; 2057 } 2058 bzero(ibp->b_data, (int)fs->fs_bsize); 2059 dp2 = (struct ufs2_dinode *)(ibp->b_data); 2060 for (i = 0; i < INOPB(fs); i++) { 2061 while (dp2->di_gen == 0) 2062 dp2->di_gen = arc4random(); 2063 dp2++; 2064 } 2065 /* 2066 * Rather than adding a soft updates dependency to ensure 2067 * that the new inode block is written before it is claimed 2068 * by the cylinder group map, we just do a barrier write 2069 * here. The barrier write will ensure that the inode block 2070 * gets written before the updated cylinder group map can be 2071 * written. The barrier write should only slow down bulk 2072 * loading of newly created filesystems. 2073 */ 2074 babarrierwrite(ibp); 2075 2076 /* 2077 * After the inode block is written, try to update the 2078 * cg initediblk pointer. If another thread beat us 2079 * to it, then leave it unchanged as the other thread 2080 * has already set it correctly. 2081 */ 2082 error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp); 2083 UFS_LOCK(ump); 2084 ACTIVECLEAR(fs, cg); 2085 UFS_UNLOCK(ump); 2086 if (error != 0) 2087 return (error); 2088 if (cgp->cg_initediblk == old_initediblk) 2089 cgp->cg_initediblk += INOPB(fs); 2090 goto restart; 2091 } 2092 cgp->cg_irotor = ipref; 2093 UFS_LOCK(ump); 2094 ACTIVECLEAR(fs, cg); 2095 setbit(inosused, ipref); 2096 cgp->cg_cs.cs_nifree--; 2097 fs->fs_cstotal.cs_nifree--; 2098 fs->fs_cs(fs, cg).cs_nifree--; 2099 fs->fs_fmod = 1; 2100 if ((mode & IFMT) == IFDIR) { 2101 cgp->cg_cs.cs_ndir++; 2102 fs->fs_cstotal.cs_ndir++; 2103 fs->fs_cs(fs, cg).cs_ndir++; 2104 } 2105 UFS_UNLOCK(ump); 2106 if (DOINGSOFTDEP(ITOV(ip))) 2107 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref, mode); 2108 bdwrite(bp); 2109 return ((ino_t)(cg * fs->fs_ipg + ipref)); 2110 } 2111 2112 /* 2113 * Free a block or fragment. 2114 * 2115 * The specified block or fragment is placed back in the 2116 * free map. If a fragment is deallocated, a possible 2117 * block reassembly is checked. 2118 */ 2119 static void 2120 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd) 2121 struct ufsmount *ump; 2122 struct fs *fs; 2123 struct vnode *devvp; 2124 ufs2_daddr_t bno; 2125 long size; 2126 ino_t inum; 2127 struct workhead *dephd; 2128 { 2129 struct mount *mp; 2130 struct cg *cgp; 2131 struct buf *bp; 2132 ufs1_daddr_t fragno, cgbno; 2133 int i, blk, frags, bbase, error; 2134 u_int cg; 2135 u_int8_t *blksfree; 2136 struct cdev *dev; 2137 2138 cg = dtog(fs, bno); 2139 if (devvp->v_type == VREG) { 2140 /* devvp is a snapshot */ 2141 MPASS(devvp->v_mount->mnt_data == ump); 2142 dev = ump->um_devvp->v_rdev; 2143 } else if (devvp->v_type == VCHR) { 2144 /* devvp is a normal disk device */ 2145 dev = devvp->v_rdev; 2146 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree_cg"); 2147 } else 2148 return; 2149 #ifdef INVARIANTS 2150 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 2151 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 2152 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n", 2153 devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize, 2154 size, fs->fs_fsmnt); 2155 panic("ffs_blkfree_cg: bad size"); 2156 } 2157 #endif 2158 if ((u_int)bno >= fs->fs_size) { 2159 printf("bad block %jd, ino %lu\n", (intmax_t)bno, 2160 (u_long)inum); 2161 ffs_fserr(fs, inum, "bad block"); 2162 return; 2163 } 2164 if ((error = ffs_getcg(fs, devvp, cg, &bp, &cgp)) != 0) 2165 return; 2166 cgbno = dtogd(fs, bno); 2167 blksfree = cg_blksfree(cgp); 2168 UFS_LOCK(ump); 2169 if (size == fs->fs_bsize) { 2170 fragno = fragstoblks(fs, cgbno); 2171 if (!ffs_isfreeblock(fs, blksfree, fragno)) { 2172 if (devvp->v_type == VREG) { 2173 UFS_UNLOCK(ump); 2174 /* devvp is a snapshot */ 2175 brelse(bp); 2176 return; 2177 } 2178 printf("dev = %s, block = %jd, fs = %s\n", 2179 devtoname(dev), (intmax_t)bno, fs->fs_fsmnt); 2180 panic("ffs_blkfree_cg: freeing free block"); 2181 } 2182 ffs_setblock(fs, blksfree, fragno); 2183 ffs_clusteracct(fs, cgp, fragno, 1); 2184 cgp->cg_cs.cs_nbfree++; 2185 fs->fs_cstotal.cs_nbfree++; 2186 fs->fs_cs(fs, cg).cs_nbfree++; 2187 } else { 2188 bbase = cgbno - fragnum(fs, cgbno); 2189 /* 2190 * decrement the counts associated with the old frags 2191 */ 2192 blk = blkmap(fs, blksfree, bbase); 2193 ffs_fragacct(fs, blk, cgp->cg_frsum, -1); 2194 /* 2195 * deallocate the fragment 2196 */ 2197 frags = numfrags(fs, size); 2198 for (i = 0; i < frags; i++) { 2199 if (isset(blksfree, cgbno + i)) { 2200 printf("dev = %s, block = %jd, fs = %s\n", 2201 devtoname(dev), (intmax_t)(bno + i), 2202 fs->fs_fsmnt); 2203 panic("ffs_blkfree_cg: freeing free frag"); 2204 } 2205 setbit(blksfree, cgbno + i); 2206 } 2207 cgp->cg_cs.cs_nffree += i; 2208 fs->fs_cstotal.cs_nffree += i; 2209 fs->fs_cs(fs, cg).cs_nffree += i; 2210 /* 2211 * add back in counts associated with the new frags 2212 */ 2213 blk = blkmap(fs, blksfree, bbase); 2214 ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 2215 /* 2216 * if a complete block has been reassembled, account for it 2217 */ 2218 fragno = fragstoblks(fs, bbase); 2219 if (ffs_isblock(fs, blksfree, fragno)) { 2220 cgp->cg_cs.cs_nffree -= fs->fs_frag; 2221 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 2222 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 2223 ffs_clusteracct(fs, cgp, fragno, 1); 2224 cgp->cg_cs.cs_nbfree++; 2225 fs->fs_cstotal.cs_nbfree++; 2226 fs->fs_cs(fs, cg).cs_nbfree++; 2227 } 2228 } 2229 fs->fs_fmod = 1; 2230 ACTIVECLEAR(fs, cg); 2231 UFS_UNLOCK(ump); 2232 mp = UFSTOVFS(ump); 2233 if (MOUNTEDSOFTDEP(mp) && devvp->v_type == VCHR) 2234 softdep_setup_blkfree(UFSTOVFS(ump), bp, bno, 2235 numfrags(fs, size), dephd); 2236 bdwrite(bp); 2237 } 2238 2239 struct ffs_blkfree_trim_params { 2240 struct task task; 2241 struct ufsmount *ump; 2242 struct vnode *devvp; 2243 ufs2_daddr_t bno; 2244 long size; 2245 ino_t inum; 2246 struct workhead *pdephd; 2247 struct workhead dephd; 2248 }; 2249 2250 static void 2251 ffs_blkfree_trim_task(ctx, pending) 2252 void *ctx; 2253 int pending; 2254 { 2255 struct ffs_blkfree_trim_params *tp; 2256 2257 tp = ctx; 2258 ffs_blkfree_cg(tp->ump, tp->ump->um_fs, tp->devvp, tp->bno, tp->size, 2259 tp->inum, tp->pdephd); 2260 vn_finished_secondary_write(UFSTOVFS(tp->ump)); 2261 atomic_add_int(&tp->ump->um_trim_inflight, -1); 2262 free(tp, M_TEMP); 2263 } 2264 2265 static void 2266 ffs_blkfree_trim_completed(bip) 2267 struct bio *bip; 2268 { 2269 struct ffs_blkfree_trim_params *tp; 2270 2271 tp = bip->bio_caller2; 2272 g_destroy_bio(bip); 2273 TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp); 2274 taskqueue_enqueue(tp->ump->um_trim_tq, &tp->task); 2275 } 2276 2277 void 2278 ffs_blkfree(ump, fs, devvp, bno, size, inum, vtype, dephd) 2279 struct ufsmount *ump; 2280 struct fs *fs; 2281 struct vnode *devvp; 2282 ufs2_daddr_t bno; 2283 long size; 2284 ino_t inum; 2285 enum vtype vtype; 2286 struct workhead *dephd; 2287 { 2288 struct mount *mp; 2289 struct bio *bip; 2290 struct ffs_blkfree_trim_params *tp; 2291 2292 /* 2293 * Check to see if a snapshot wants to claim the block. 2294 * Check that devvp is a normal disk device, not a snapshot, 2295 * it has a snapshot(s) associated with it, and one of the 2296 * snapshots wants to claim the block. 2297 */ 2298 if (devvp->v_type == VCHR && 2299 (devvp->v_vflag & VV_COPYONWRITE) && 2300 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, dephd)) { 2301 return; 2302 } 2303 /* 2304 * Nothing to delay if TRIM is disabled, or the operation is 2305 * performed on the snapshot. 2306 */ 2307 if (!ump->um_candelete || devvp->v_type == VREG) { 2308 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd); 2309 return; 2310 } 2311 2312 /* 2313 * Postpone the set of the free bit in the cg bitmap until the 2314 * BIO_DELETE is completed. Otherwise, due to disk queue 2315 * reordering, TRIM might be issued after we reuse the block 2316 * and write some new data into it. 2317 */ 2318 atomic_add_int(&ump->um_trim_inflight, 1); 2319 tp = malloc(sizeof(struct ffs_blkfree_trim_params), M_TEMP, M_WAITOK); 2320 tp->ump = ump; 2321 tp->devvp = devvp; 2322 tp->bno = bno; 2323 tp->size = size; 2324 tp->inum = inum; 2325 if (dephd != NULL) { 2326 LIST_INIT(&tp->dephd); 2327 LIST_SWAP(dephd, &tp->dephd, worklist, wk_list); 2328 tp->pdephd = &tp->dephd; 2329 } else 2330 tp->pdephd = NULL; 2331 2332 bip = g_alloc_bio(); 2333 bip->bio_cmd = BIO_DELETE; 2334 bip->bio_offset = dbtob(fsbtodb(fs, bno)); 2335 bip->bio_done = ffs_blkfree_trim_completed; 2336 bip->bio_length = size; 2337 bip->bio_caller2 = tp; 2338 2339 mp = UFSTOVFS(ump); 2340 vn_start_secondary_write(NULL, &mp, 0); 2341 g_io_request(bip, (struct g_consumer *)devvp->v_bufobj.bo_private); 2342 } 2343 2344 #ifdef INVARIANTS 2345 /* 2346 * Verify allocation of a block or fragment. Returns true if block or 2347 * fragment is allocated, false if it is free. 2348 */ 2349 static int 2350 ffs_checkblk(ip, bno, size) 2351 struct inode *ip; 2352 ufs2_daddr_t bno; 2353 long size; 2354 { 2355 struct fs *fs; 2356 struct cg *cgp; 2357 struct buf *bp; 2358 ufs1_daddr_t cgbno; 2359 int i, error, frags, free; 2360 u_int8_t *blksfree; 2361 2362 fs = ITOFS(ip); 2363 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 2364 printf("bsize = %ld, size = %ld, fs = %s\n", 2365 (long)fs->fs_bsize, size, fs->fs_fsmnt); 2366 panic("ffs_checkblk: bad size"); 2367 } 2368 if ((u_int)bno >= fs->fs_size) 2369 panic("ffs_checkblk: bad block %jd", (intmax_t)bno); 2370 error = ffs_getcg(fs, ITODEVVP(ip), dtog(fs, bno), &bp, &cgp); 2371 if (error) 2372 panic("ffs_checkblk: cylinder group read failed"); 2373 blksfree = cg_blksfree(cgp); 2374 cgbno = dtogd(fs, bno); 2375 if (size == fs->fs_bsize) { 2376 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno)); 2377 } else { 2378 frags = numfrags(fs, size); 2379 for (free = 0, i = 0; i < frags; i++) 2380 if (isset(blksfree, cgbno + i)) 2381 free++; 2382 if (free != 0 && free != frags) 2383 panic("ffs_checkblk: partially free fragment"); 2384 } 2385 brelse(bp); 2386 return (!free); 2387 } 2388 #endif /* INVARIANTS */ 2389 2390 /* 2391 * Free an inode. 2392 */ 2393 int 2394 ffs_vfree(pvp, ino, mode) 2395 struct vnode *pvp; 2396 ino_t ino; 2397 int mode; 2398 { 2399 struct ufsmount *ump; 2400 struct inode *ip; 2401 2402 if (DOINGSOFTDEP(pvp)) { 2403 softdep_freefile(pvp, ino, mode); 2404 return (0); 2405 } 2406 ip = VTOI(pvp); 2407 ump = VFSTOUFS(pvp->v_mount); 2408 return (ffs_freefile(ump, ump->um_fs, ump->um_devvp, ino, mode, NULL)); 2409 } 2410 2411 /* 2412 * Do the actual free operation. 2413 * The specified inode is placed back in the free map. 2414 */ 2415 int 2416 ffs_freefile(ump, fs, devvp, ino, mode, wkhd) 2417 struct ufsmount *ump; 2418 struct fs *fs; 2419 struct vnode *devvp; 2420 ino_t ino; 2421 int mode; 2422 struct workhead *wkhd; 2423 { 2424 struct cg *cgp; 2425 struct buf *bp; 2426 ufs2_daddr_t cgbno; 2427 int error; 2428 u_int cg; 2429 u_int8_t *inosused; 2430 struct cdev *dev; 2431 2432 cg = ino_to_cg(fs, ino); 2433 if (devvp->v_type == VREG) { 2434 /* devvp is a snapshot */ 2435 MPASS(devvp->v_mount->mnt_data == ump); 2436 dev = ump->um_devvp->v_rdev; 2437 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2438 } else if (devvp->v_type == VCHR) { 2439 /* devvp is a normal disk device */ 2440 dev = devvp->v_rdev; 2441 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2442 } else { 2443 bp = NULL; 2444 return (0); 2445 } 2446 if (ino >= fs->fs_ipg * fs->fs_ncg) 2447 panic("ffs_freefile: range: dev = %s, ino = %ju, fs = %s", 2448 devtoname(dev), (uintmax_t)ino, fs->fs_fsmnt); 2449 if ((error = ffs_getcg(fs, devvp, cg, &bp, &cgp)) != 0) 2450 return (error); 2451 inosused = cg_inosused(cgp); 2452 ino %= fs->fs_ipg; 2453 if (isclr(inosused, ino)) { 2454 printf("dev = %s, ino = %ju, fs = %s\n", devtoname(dev), 2455 (uintmax_t)(ino + cg * fs->fs_ipg), fs->fs_fsmnt); 2456 if (fs->fs_ronly == 0) 2457 panic("ffs_freefile: freeing free inode"); 2458 } 2459 clrbit(inosused, ino); 2460 if (ino < cgp->cg_irotor) 2461 cgp->cg_irotor = ino; 2462 cgp->cg_cs.cs_nifree++; 2463 UFS_LOCK(ump); 2464 fs->fs_cstotal.cs_nifree++; 2465 fs->fs_cs(fs, cg).cs_nifree++; 2466 if ((mode & IFMT) == IFDIR) { 2467 cgp->cg_cs.cs_ndir--; 2468 fs->fs_cstotal.cs_ndir--; 2469 fs->fs_cs(fs, cg).cs_ndir--; 2470 } 2471 fs->fs_fmod = 1; 2472 ACTIVECLEAR(fs, cg); 2473 UFS_UNLOCK(ump); 2474 if (MOUNTEDSOFTDEP(UFSTOVFS(ump)) && devvp->v_type == VCHR) 2475 softdep_setup_inofree(UFSTOVFS(ump), bp, 2476 ino + cg * fs->fs_ipg, wkhd); 2477 bdwrite(bp); 2478 return (0); 2479 } 2480 2481 /* 2482 * Check to see if a file is free. 2483 * Used to check for allocated files in snapshots. 2484 */ 2485 int 2486 ffs_checkfreefile(fs, devvp, ino) 2487 struct fs *fs; 2488 struct vnode *devvp; 2489 ino_t ino; 2490 { 2491 struct cg *cgp; 2492 struct buf *bp; 2493 ufs2_daddr_t cgbno; 2494 int ret, error; 2495 u_int cg; 2496 u_int8_t *inosused; 2497 2498 cg = ino_to_cg(fs, ino); 2499 if (devvp->v_type == VREG) { 2500 /* devvp is a snapshot */ 2501 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2502 } else if (devvp->v_type == VCHR) { 2503 /* devvp is a normal disk device */ 2504 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2505 } else { 2506 return (1); 2507 } 2508 if (ino >= fs->fs_ipg * fs->fs_ncg) 2509 return (1); 2510 if ((error = ffs_getcg(fs, devvp, cg, &bp, &cgp)) != 0) 2511 return (1); 2512 inosused = cg_inosused(cgp); 2513 ino %= fs->fs_ipg; 2514 ret = isclr(inosused, ino); 2515 brelse(bp); 2516 return (ret); 2517 } 2518 2519 /* 2520 * Find a block of the specified size in the specified cylinder group. 2521 * 2522 * It is a panic if a request is made to find a block if none are 2523 * available. 2524 */ 2525 static ufs1_daddr_t 2526 ffs_mapsearch(fs, cgp, bpref, allocsiz) 2527 struct fs *fs; 2528 struct cg *cgp; 2529 ufs2_daddr_t bpref; 2530 int allocsiz; 2531 { 2532 ufs1_daddr_t bno; 2533 int start, len, loc, i; 2534 int blk, field, subfield, pos; 2535 u_int8_t *blksfree; 2536 2537 /* 2538 * find the fragment by searching through the free block 2539 * map for an appropriate bit pattern 2540 */ 2541 if (bpref) 2542 start = dtogd(fs, bpref) / NBBY; 2543 else 2544 start = cgp->cg_frotor / NBBY; 2545 blksfree = cg_blksfree(cgp); 2546 len = howmany(fs->fs_fpg, NBBY) - start; 2547 loc = scanc((u_int)len, (u_char *)&blksfree[start], 2548 fragtbl[fs->fs_frag], 2549 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 2550 if (loc == 0) { 2551 len = start + 1; 2552 start = 0; 2553 loc = scanc((u_int)len, (u_char *)&blksfree[0], 2554 fragtbl[fs->fs_frag], 2555 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 2556 if (loc == 0) { 2557 printf("start = %d, len = %d, fs = %s\n", 2558 start, len, fs->fs_fsmnt); 2559 panic("ffs_alloccg: map corrupted"); 2560 /* NOTREACHED */ 2561 } 2562 } 2563 bno = (start + len - loc) * NBBY; 2564 cgp->cg_frotor = bno; 2565 /* 2566 * found the byte in the map 2567 * sift through the bits to find the selected frag 2568 */ 2569 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 2570 blk = blkmap(fs, blksfree, bno); 2571 blk <<= 1; 2572 field = around[allocsiz]; 2573 subfield = inside[allocsiz]; 2574 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 2575 if ((blk & field) == subfield) 2576 return (bno + pos); 2577 field <<= 1; 2578 subfield <<= 1; 2579 } 2580 } 2581 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); 2582 panic("ffs_alloccg: block not in map"); 2583 return (-1); 2584 } 2585 2586 /* 2587 * Fetch and verify a cylinder group. 2588 */ 2589 int 2590 ffs_getcg(fs, devvp, cg, bpp, cgpp) 2591 struct fs *fs; 2592 struct vnode *devvp; 2593 u_int cg; 2594 struct buf **bpp; 2595 struct cg **cgpp; 2596 { 2597 struct buf *bp; 2598 struct cg *cgp; 2599 int error; 2600 2601 *bpp = NULL; 2602 *cgpp = NULL; 2603 error = bread(devvp, devvp->v_type == VREG ? 2604 fragstoblks(fs, cgtod(fs, cg)) : fsbtodb(fs, cgtod(fs, cg)), 2605 (int)fs->fs_cgsize, NOCRED, &bp); 2606 if (error != 0) 2607 return (error); 2608 cgp = (struct cg *)bp->b_data; 2609 if (!cg_chkmagic(cgp) || cgp->cg_cgx != cg) { 2610 brelse(bp); 2611 return (EIO); 2612 } 2613 bp->b_xflags |= BX_BKGRDWRITE; 2614 cgp->cg_old_time = cgp->cg_time = time_second; 2615 *bpp = bp; 2616 *cgpp = cgp; 2617 return (0); 2618 } 2619 2620 /* 2621 * Fserr prints the name of a filesystem with an error diagnostic. 2622 * 2623 * The form of the error message is: 2624 * fs: error message 2625 */ 2626 void 2627 ffs_fserr(fs, inum, cp) 2628 struct fs *fs; 2629 ino_t inum; 2630 char *cp; 2631 { 2632 struct thread *td = curthread; /* XXX */ 2633 struct proc *p = td->td_proc; 2634 2635 log(LOG_ERR, "pid %d (%s), uid %d inumber %ju on %s: %s\n", 2636 p->p_pid, p->p_comm, td->td_ucred->cr_uid, (uintmax_t)inum, 2637 fs->fs_fsmnt, cp); 2638 } 2639 2640 /* 2641 * This function provides the capability for the fsck program to 2642 * update an active filesystem. Fourteen operations are provided: 2643 * 2644 * adjrefcnt(inode, amt) - adjusts the reference count on the 2645 * specified inode by the specified amount. Under normal 2646 * operation the count should always go down. Decrementing 2647 * the count to zero will cause the inode to be freed. 2648 * adjblkcnt(inode, amt) - adjust the number of blocks used by the 2649 * inode by the specified amount. 2650 * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) - 2651 * adjust the superblock summary. 2652 * freedirs(inode, count) - directory inodes [inode..inode + count - 1] 2653 * are marked as free. Inodes should never have to be marked 2654 * as in use. 2655 * freefiles(inode, count) - file inodes [inode..inode + count - 1] 2656 * are marked as free. Inodes should never have to be marked 2657 * as in use. 2658 * freeblks(blockno, size) - blocks [blockno..blockno + size - 1] 2659 * are marked as free. Blocks should never have to be marked 2660 * as in use. 2661 * setflags(flags, set/clear) - the fs_flags field has the specified 2662 * flags set (second parameter +1) or cleared (second parameter -1). 2663 * setcwd(dirinode) - set the current directory to dirinode in the 2664 * filesystem associated with the snapshot. 2665 * setdotdot(oldvalue, newvalue) - Verify that the inode number for ".." 2666 * in the current directory is oldvalue then change it to newvalue. 2667 * unlink(nameptr, oldvalue) - Verify that the inode number associated 2668 * with nameptr in the current directory is oldvalue then unlink it. 2669 * 2670 * The following functions may only be used on a quiescent filesystem 2671 * by the soft updates journal. They are not safe to be run on an active 2672 * filesystem. 2673 * 2674 * setinode(inode, dip) - the specified disk inode is replaced with the 2675 * contents pointed to by dip. 2676 * setbufoutput(fd, flags) - output associated with the specified file 2677 * descriptor (which must reference the character device supporting 2678 * the filesystem) switches from using physio to running through the 2679 * buffer cache when flags is set to 1. The descriptor reverts to 2680 * physio for output when flags is set to zero. 2681 */ 2682 2683 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS); 2684 2685 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT, 2686 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count"); 2687 2688 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR, 2689 sysctl_ffs_fsck, "Adjust Inode Used Blocks Count"); 2690 2691 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR, 2692 sysctl_ffs_fsck, "Adjust number of directories"); 2693 2694 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR, 2695 sysctl_ffs_fsck, "Adjust number of free blocks"); 2696 2697 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR, 2698 sysctl_ffs_fsck, "Adjust number of free inodes"); 2699 2700 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR, 2701 sysctl_ffs_fsck, "Adjust number of free frags"); 2702 2703 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR, 2704 sysctl_ffs_fsck, "Adjust number of free clusters"); 2705 2706 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR, 2707 sysctl_ffs_fsck, "Free Range of Directory Inodes"); 2708 2709 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR, 2710 sysctl_ffs_fsck, "Free Range of File Inodes"); 2711 2712 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR, 2713 sysctl_ffs_fsck, "Free Range of Blocks"); 2714 2715 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR, 2716 sysctl_ffs_fsck, "Change Filesystem Flags"); 2717 2718 static SYSCTL_NODE(_vfs_ffs, FFS_SET_CWD, setcwd, CTLFLAG_WR, 2719 sysctl_ffs_fsck, "Set Current Working Directory"); 2720 2721 static SYSCTL_NODE(_vfs_ffs, FFS_SET_DOTDOT, setdotdot, CTLFLAG_WR, 2722 sysctl_ffs_fsck, "Change Value of .. Entry"); 2723 2724 static SYSCTL_NODE(_vfs_ffs, FFS_UNLINK, unlink, CTLFLAG_WR, 2725 sysctl_ffs_fsck, "Unlink a Duplicate Name"); 2726 2727 static SYSCTL_NODE(_vfs_ffs, FFS_SET_INODE, setinode, CTLFLAG_WR, 2728 sysctl_ffs_fsck, "Update an On-Disk Inode"); 2729 2730 static SYSCTL_NODE(_vfs_ffs, FFS_SET_BUFOUTPUT, setbufoutput, CTLFLAG_WR, 2731 sysctl_ffs_fsck, "Set Buffered Writing for Descriptor"); 2732 2733 #define DEBUG 1 2734 #ifdef DEBUG 2735 static int fsckcmds = 0; 2736 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, ""); 2737 #endif /* DEBUG */ 2738 2739 static int buffered_write(struct file *, struct uio *, struct ucred *, 2740 int, struct thread *); 2741 2742 static int 2743 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS) 2744 { 2745 struct thread *td = curthread; 2746 struct fsck_cmd cmd; 2747 struct ufsmount *ump; 2748 struct vnode *vp, *dvp, *fdvp; 2749 struct inode *ip, *dp; 2750 struct mount *mp; 2751 struct fs *fs; 2752 ufs2_daddr_t blkno; 2753 long blkcnt, blksize; 2754 struct file *fp, *vfp; 2755 cap_rights_t rights; 2756 int filetype, error; 2757 static struct fileops *origops, bufferedops; 2758 2759 if (req->newlen > sizeof cmd) 2760 return (EBADRPC); 2761 if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0) 2762 return (error); 2763 if (cmd.version != FFS_CMD_VERSION) 2764 return (ERPCMISMATCH); 2765 if ((error = getvnode(td, cmd.handle, 2766 cap_rights_init(&rights, CAP_FSCK), &fp)) != 0) 2767 return (error); 2768 vp = fp->f_data; 2769 if (vp->v_type != VREG && vp->v_type != VDIR) { 2770 fdrop(fp, td); 2771 return (EINVAL); 2772 } 2773 vn_start_write(vp, &mp, V_WAIT); 2774 if (mp == NULL || 2775 strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) { 2776 vn_finished_write(mp); 2777 fdrop(fp, td); 2778 return (EINVAL); 2779 } 2780 ump = VFSTOUFS(mp); 2781 if ((mp->mnt_flag & MNT_RDONLY) && 2782 ump->um_fsckpid != td->td_proc->p_pid) { 2783 vn_finished_write(mp); 2784 fdrop(fp, td); 2785 return (EROFS); 2786 } 2787 fs = ump->um_fs; 2788 filetype = IFREG; 2789 2790 switch (oidp->oid_number) { 2791 2792 case FFS_SET_FLAGS: 2793 #ifdef DEBUG 2794 if (fsckcmds) 2795 printf("%s: %s flags\n", mp->mnt_stat.f_mntonname, 2796 cmd.size > 0 ? "set" : "clear"); 2797 #endif /* DEBUG */ 2798 if (cmd.size > 0) 2799 fs->fs_flags |= (long)cmd.value; 2800 else 2801 fs->fs_flags &= ~(long)cmd.value; 2802 break; 2803 2804 case FFS_ADJ_REFCNT: 2805 #ifdef DEBUG 2806 if (fsckcmds) { 2807 printf("%s: adjust inode %jd link count by %jd\n", 2808 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2809 (intmax_t)cmd.size); 2810 } 2811 #endif /* DEBUG */ 2812 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) 2813 break; 2814 ip = VTOI(vp); 2815 ip->i_nlink += cmd.size; 2816 DIP_SET(ip, i_nlink, ip->i_nlink); 2817 ip->i_effnlink += cmd.size; 2818 ip->i_flag |= IN_CHANGE | IN_MODIFIED; 2819 error = ffs_update(vp, 1); 2820 if (DOINGSOFTDEP(vp)) 2821 softdep_change_linkcnt(ip); 2822 vput(vp); 2823 break; 2824 2825 case FFS_ADJ_BLKCNT: 2826 #ifdef DEBUG 2827 if (fsckcmds) { 2828 printf("%s: adjust inode %jd block count by %jd\n", 2829 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2830 (intmax_t)cmd.size); 2831 } 2832 #endif /* DEBUG */ 2833 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) 2834 break; 2835 ip = VTOI(vp); 2836 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size); 2837 ip->i_flag |= IN_CHANGE | IN_MODIFIED; 2838 error = ffs_update(vp, 1); 2839 vput(vp); 2840 break; 2841 2842 case FFS_DIR_FREE: 2843 filetype = IFDIR; 2844 /* fall through */ 2845 2846 case FFS_FILE_FREE: 2847 #ifdef DEBUG 2848 if (fsckcmds) { 2849 if (cmd.size == 1) 2850 printf("%s: free %s inode %ju\n", 2851 mp->mnt_stat.f_mntonname, 2852 filetype == IFDIR ? "directory" : "file", 2853 (uintmax_t)cmd.value); 2854 else 2855 printf("%s: free %s inodes %ju-%ju\n", 2856 mp->mnt_stat.f_mntonname, 2857 filetype == IFDIR ? "directory" : "file", 2858 (uintmax_t)cmd.value, 2859 (uintmax_t)(cmd.value + cmd.size - 1)); 2860 } 2861 #endif /* DEBUG */ 2862 while (cmd.size > 0) { 2863 if ((error = ffs_freefile(ump, fs, ump->um_devvp, 2864 cmd.value, filetype, NULL))) 2865 break; 2866 cmd.size -= 1; 2867 cmd.value += 1; 2868 } 2869 break; 2870 2871 case FFS_BLK_FREE: 2872 #ifdef DEBUG 2873 if (fsckcmds) { 2874 if (cmd.size == 1) 2875 printf("%s: free block %jd\n", 2876 mp->mnt_stat.f_mntonname, 2877 (intmax_t)cmd.value); 2878 else 2879 printf("%s: free blocks %jd-%jd\n", 2880 mp->mnt_stat.f_mntonname, 2881 (intmax_t)cmd.value, 2882 (intmax_t)cmd.value + cmd.size - 1); 2883 } 2884 #endif /* DEBUG */ 2885 blkno = cmd.value; 2886 blkcnt = cmd.size; 2887 blksize = fs->fs_frag - (blkno % fs->fs_frag); 2888 while (blkcnt > 0) { 2889 if (blksize > blkcnt) 2890 blksize = blkcnt; 2891 ffs_blkfree(ump, fs, ump->um_devvp, blkno, 2892 blksize * fs->fs_fsize, UFS_ROOTINO, VDIR, NULL); 2893 blkno += blksize; 2894 blkcnt -= blksize; 2895 blksize = fs->fs_frag; 2896 } 2897 break; 2898 2899 /* 2900 * Adjust superblock summaries. fsck(8) is expected to 2901 * submit deltas when necessary. 2902 */ 2903 case FFS_ADJ_NDIR: 2904 #ifdef DEBUG 2905 if (fsckcmds) { 2906 printf("%s: adjust number of directories by %jd\n", 2907 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2908 } 2909 #endif /* DEBUG */ 2910 fs->fs_cstotal.cs_ndir += cmd.value; 2911 break; 2912 2913 case FFS_ADJ_NBFREE: 2914 #ifdef DEBUG 2915 if (fsckcmds) { 2916 printf("%s: adjust number of free blocks by %+jd\n", 2917 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2918 } 2919 #endif /* DEBUG */ 2920 fs->fs_cstotal.cs_nbfree += cmd.value; 2921 break; 2922 2923 case FFS_ADJ_NIFREE: 2924 #ifdef DEBUG 2925 if (fsckcmds) { 2926 printf("%s: adjust number of free inodes by %+jd\n", 2927 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2928 } 2929 #endif /* DEBUG */ 2930 fs->fs_cstotal.cs_nifree += cmd.value; 2931 break; 2932 2933 case FFS_ADJ_NFFREE: 2934 #ifdef DEBUG 2935 if (fsckcmds) { 2936 printf("%s: adjust number of free frags by %+jd\n", 2937 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2938 } 2939 #endif /* DEBUG */ 2940 fs->fs_cstotal.cs_nffree += cmd.value; 2941 break; 2942 2943 case FFS_ADJ_NUMCLUSTERS: 2944 #ifdef DEBUG 2945 if (fsckcmds) { 2946 printf("%s: adjust number of free clusters by %+jd\n", 2947 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2948 } 2949 #endif /* DEBUG */ 2950 fs->fs_cstotal.cs_numclusters += cmd.value; 2951 break; 2952 2953 case FFS_SET_CWD: 2954 #ifdef DEBUG 2955 if (fsckcmds) { 2956 printf("%s: set current directory to inode %jd\n", 2957 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2958 } 2959 #endif /* DEBUG */ 2960 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_SHARED, &vp))) 2961 break; 2962 AUDIT_ARG_VNODE1(vp); 2963 if ((error = change_dir(vp, td)) != 0) { 2964 vput(vp); 2965 break; 2966 } 2967 VOP_UNLOCK(vp, 0); 2968 pwd_chdir(td, vp); 2969 break; 2970 2971 case FFS_SET_DOTDOT: 2972 #ifdef DEBUG 2973 if (fsckcmds) { 2974 printf("%s: change .. in cwd from %jd to %jd\n", 2975 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2976 (intmax_t)cmd.size); 2977 } 2978 #endif /* DEBUG */ 2979 /* 2980 * First we have to get and lock the parent directory 2981 * to which ".." points. 2982 */ 2983 error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &fdvp); 2984 if (error) 2985 break; 2986 /* 2987 * Now we get and lock the child directory containing "..". 2988 */ 2989 FILEDESC_SLOCK(td->td_proc->p_fd); 2990 dvp = td->td_proc->p_fd->fd_cdir; 2991 FILEDESC_SUNLOCK(td->td_proc->p_fd); 2992 if ((error = vget(dvp, LK_EXCLUSIVE, td)) != 0) { 2993 vput(fdvp); 2994 break; 2995 } 2996 dp = VTOI(dvp); 2997 dp->i_offset = 12; /* XXX mastertemplate.dot_reclen */ 2998 error = ufs_dirrewrite(dp, VTOI(fdvp), (ino_t)cmd.size, 2999 DT_DIR, 0); 3000 cache_purge(fdvp); 3001 cache_purge(dvp); 3002 vput(dvp); 3003 vput(fdvp); 3004 break; 3005 3006 case FFS_UNLINK: 3007 #ifdef DEBUG 3008 if (fsckcmds) { 3009 char buf[32]; 3010 3011 if (copyinstr((char *)(intptr_t)cmd.value, buf,32,NULL)) 3012 strncpy(buf, "Name_too_long", 32); 3013 printf("%s: unlink %s (inode %jd)\n", 3014 mp->mnt_stat.f_mntonname, buf, (intmax_t)cmd.size); 3015 } 3016 #endif /* DEBUG */ 3017 /* 3018 * kern_unlinkat will do its own start/finish writes and 3019 * they do not nest, so drop ours here. Setting mp == NULL 3020 * indicates that vn_finished_write is not needed down below. 3021 */ 3022 vn_finished_write(mp); 3023 mp = NULL; 3024 error = kern_unlinkat(td, AT_FDCWD, (char *)(intptr_t)cmd.value, 3025 UIO_USERSPACE, (ino_t)cmd.size); 3026 break; 3027 3028 case FFS_SET_INODE: 3029 if (ump->um_fsckpid != td->td_proc->p_pid) { 3030 error = EPERM; 3031 break; 3032 } 3033 #ifdef DEBUG 3034 if (fsckcmds) { 3035 printf("%s: update inode %jd\n", 3036 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 3037 } 3038 #endif /* DEBUG */ 3039 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) 3040 break; 3041 AUDIT_ARG_VNODE1(vp); 3042 ip = VTOI(vp); 3043 if (I_IS_UFS1(ip)) 3044 error = copyin((void *)(intptr_t)cmd.size, ip->i_din1, 3045 sizeof(struct ufs1_dinode)); 3046 else 3047 error = copyin((void *)(intptr_t)cmd.size, ip->i_din2, 3048 sizeof(struct ufs2_dinode)); 3049 if (error) { 3050 vput(vp); 3051 break; 3052 } 3053 ip->i_flag |= IN_CHANGE | IN_MODIFIED; 3054 error = ffs_update(vp, 1); 3055 vput(vp); 3056 break; 3057 3058 case FFS_SET_BUFOUTPUT: 3059 if (ump->um_fsckpid != td->td_proc->p_pid) { 3060 error = EPERM; 3061 break; 3062 } 3063 if (ITOUMP(VTOI(vp)) != ump) { 3064 error = EINVAL; 3065 break; 3066 } 3067 #ifdef DEBUG 3068 if (fsckcmds) { 3069 printf("%s: %s buffered output for descriptor %jd\n", 3070 mp->mnt_stat.f_mntonname, 3071 cmd.size == 1 ? "enable" : "disable", 3072 (intmax_t)cmd.value); 3073 } 3074 #endif /* DEBUG */ 3075 if ((error = getvnode(td, cmd.value, 3076 cap_rights_init(&rights, CAP_FSCK), &vfp)) != 0) 3077 break; 3078 if (vfp->f_vnode->v_type != VCHR) { 3079 fdrop(vfp, td); 3080 error = EINVAL; 3081 break; 3082 } 3083 if (origops == NULL) { 3084 origops = vfp->f_ops; 3085 bcopy((void *)origops, (void *)&bufferedops, 3086 sizeof(bufferedops)); 3087 bufferedops.fo_write = buffered_write; 3088 } 3089 if (cmd.size == 1) 3090 atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops, 3091 (uintptr_t)&bufferedops); 3092 else 3093 atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops, 3094 (uintptr_t)origops); 3095 fdrop(vfp, td); 3096 break; 3097 3098 default: 3099 #ifdef DEBUG 3100 if (fsckcmds) { 3101 printf("Invalid request %d from fsck\n", 3102 oidp->oid_number); 3103 } 3104 #endif /* DEBUG */ 3105 error = EINVAL; 3106 break; 3107 3108 } 3109 fdrop(fp, td); 3110 vn_finished_write(mp); 3111 return (error); 3112 } 3113 3114 /* 3115 * Function to switch a descriptor to use the buffer cache to stage 3116 * its I/O. This is needed so that writes to the filesystem device 3117 * will give snapshots a chance to copy modified blocks for which it 3118 * needs to retain copies. 3119 */ 3120 static int 3121 buffered_write(fp, uio, active_cred, flags, td) 3122 struct file *fp; 3123 struct uio *uio; 3124 struct ucred *active_cred; 3125 int flags; 3126 struct thread *td; 3127 { 3128 struct vnode *devvp, *vp; 3129 struct inode *ip; 3130 struct buf *bp; 3131 struct fs *fs; 3132 struct filedesc *fdp; 3133 int error; 3134 daddr_t lbn; 3135 3136 /* 3137 * The devvp is associated with the /dev filesystem. To discover 3138 * the filesystem with which the device is associated, we depend 3139 * on the application setting the current directory to a location 3140 * within the filesystem being written. Yes, this is an ugly hack. 3141 */ 3142 devvp = fp->f_vnode; 3143 if (!vn_isdisk(devvp, NULL)) 3144 return (EINVAL); 3145 fdp = td->td_proc->p_fd; 3146 FILEDESC_SLOCK(fdp); 3147 vp = fdp->fd_cdir; 3148 vref(vp); 3149 FILEDESC_SUNLOCK(fdp); 3150 vn_lock(vp, LK_SHARED | LK_RETRY); 3151 /* 3152 * Check that the current directory vnode indeed belongs to 3153 * UFS before trying to dereference UFS-specific v_data fields. 3154 */ 3155 if (vp->v_op != &ffs_vnodeops1 && vp->v_op != &ffs_vnodeops2) { 3156 vput(vp); 3157 return (EINVAL); 3158 } 3159 ip = VTOI(vp); 3160 if (ITODEVVP(ip) != devvp) { 3161 vput(vp); 3162 return (EINVAL); 3163 } 3164 fs = ITOFS(ip); 3165 vput(vp); 3166 foffset_lock_uio(fp, uio, flags); 3167 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 3168 #ifdef DEBUG 3169 if (fsckcmds) { 3170 printf("%s: buffered write for block %jd\n", 3171 fs->fs_fsmnt, (intmax_t)btodb(uio->uio_offset)); 3172 } 3173 #endif /* DEBUG */ 3174 /* 3175 * All I/O must be contained within a filesystem block, start on 3176 * a fragment boundary, and be a multiple of fragments in length. 3177 */ 3178 if (uio->uio_resid > fs->fs_bsize - (uio->uio_offset % fs->fs_bsize) || 3179 fragoff(fs, uio->uio_offset) != 0 || 3180 fragoff(fs, uio->uio_resid) != 0) { 3181 error = EINVAL; 3182 goto out; 3183 } 3184 lbn = numfrags(fs, uio->uio_offset); 3185 bp = getblk(devvp, lbn, uio->uio_resid, 0, 0, 0); 3186 bp->b_flags |= B_RELBUF; 3187 if ((error = uiomove((char *)bp->b_data, uio->uio_resid, uio)) != 0) { 3188 brelse(bp); 3189 goto out; 3190 } 3191 error = bwrite(bp); 3192 out: 3193 VOP_UNLOCK(devvp, 0); 3194 foffset_unlock_uio(fp, uio, flags | FOF_NEXTOFF); 3195 return (error); 3196 } 3197