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