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/systm.h> 69 #include <sys/bio.h> 70 #include <sys/buf.h> 71 #include <sys/conf.h> 72 #include <sys/fcntl.h> 73 #include <sys/file.h> 74 #include <sys/filedesc.h> 75 #include <sys/priv.h> 76 #include <sys/proc.h> 77 #include <sys/vnode.h> 78 #include <sys/mount.h> 79 #include <sys/kernel.h> 80 #include <sys/syscallsubr.h> 81 #include <sys/sysctl.h> 82 #include <sys/syslog.h> 83 #include <sys/taskqueue.h> 84 85 #include <security/audit/audit.h> 86 87 #include <geom/geom.h> 88 89 #include <ufs/ufs/dir.h> 90 #include <ufs/ufs/extattr.h> 91 #include <ufs/ufs/quota.h> 92 #include <ufs/ufs/inode.h> 93 #include <ufs/ufs/ufs_extern.h> 94 #include <ufs/ufs/ufsmount.h> 95 96 #include <ufs/ffs/fs.h> 97 #include <ufs/ffs/ffs_extern.h> 98 #include <ufs/ffs/softdep.h> 99 100 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, u_int cg, ufs2_daddr_t bpref, 101 int size, int rsize); 102 103 static ufs2_daddr_t ffs_alloccg(struct inode *, u_int, ufs2_daddr_t, int, int); 104 static ufs2_daddr_t 105 ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t, int); 106 static void ffs_blkfree_cg(struct ufsmount *, struct fs *, 107 struct vnode *, ufs2_daddr_t, long, ino_t, 108 struct workhead *); 109 static void ffs_blkfree_trim_completed(struct bio *); 110 static void ffs_blkfree_trim_task(void *ctx, int pending __unused); 111 #ifdef INVARIANTS 112 static int ffs_checkblk(struct inode *, ufs2_daddr_t, long); 113 #endif 114 static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int, 115 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 void ffs_fserr(struct fs *, ino_t, char *); 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 (fs->fs_pendingblocks > 0 && reclaimed == 0) { 221 reclaimed = 1; 222 softdep_request_cleanup(fs, ITOV(ip)); 223 goto retry; 224 } 225 UFS_UNLOCK(ump); 226 if (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, 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 (fs->fs_pendingblocks > 0 && reclaimed == 0) { 422 reclaimed = 1; 423 softdep_request_cleanup(fs, vp); 424 UFS_UNLOCK(ump); 425 if (bp) { 426 brelse(bp); 427 bp = NULL; 428 } 429 UFS_LOCK(ump); 430 goto retry; 431 } 432 UFS_UNLOCK(ump); 433 if (bp) 434 brelse(bp); 435 if (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, 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, 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; 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 if (fs->fs_cstotal.cs_nifree == 0) 950 goto noinodes; 951 952 if ((mode & IFMT) == IFDIR) 953 ipref = ffs_dirpref(pip); 954 else 955 ipref = pip->i_number; 956 if (ipref >= fs->fs_ncg * fs->fs_ipg) 957 ipref = 0; 958 cg = ino_to_cg(fs, ipref); 959 /* 960 * Track number of dirs created one after another 961 * in a same cg without intervening by files. 962 */ 963 if ((mode & IFMT) == IFDIR) { 964 if (fs->fs_contigdirs[cg] < 255) 965 fs->fs_contigdirs[cg]++; 966 } else { 967 if (fs->fs_contigdirs[cg] > 0) 968 fs->fs_contigdirs[cg]--; 969 } 970 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, 971 (allocfcn_t *)ffs_nodealloccg); 972 if (ino == 0) 973 goto noinodes; 974 error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp); 975 if (error) { 976 error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp, 977 FFSV_FORCEINSMQ); 978 ffs_vfree(pvp, ino, mode); 979 if (error1 == 0) { 980 ip = VTOI(*vpp); 981 if (ip->i_mode) 982 goto dup_alloc; 983 ip->i_flag |= IN_MODIFIED; 984 vput(*vpp); 985 } 986 return (error); 987 } 988 ip = VTOI(*vpp); 989 if (ip->i_mode) { 990 dup_alloc: 991 printf("mode = 0%o, inum = %lu, fs = %s\n", 992 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt); 993 panic("ffs_valloc: dup alloc"); 994 } 995 if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */ 996 printf("free inode %s/%lu had %ld blocks\n", 997 fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks)); 998 DIP_SET(ip, i_blocks, 0); 999 } 1000 ip->i_flags = 0; 1001 DIP_SET(ip, i_flags, 0); 1002 /* 1003 * Set up a new generation number for this inode. 1004 */ 1005 if (ip->i_gen == 0 || ++ip->i_gen == 0) 1006 ip->i_gen = arc4random() / 2 + 1; 1007 DIP_SET(ip, i_gen, ip->i_gen); 1008 if (fs->fs_magic == FS_UFS2_MAGIC) { 1009 vfs_timestamp(&ts); 1010 ip->i_din2->di_birthtime = ts.tv_sec; 1011 ip->i_din2->di_birthnsec = ts.tv_nsec; 1012 } 1013 ip->i_flag = 0; 1014 vnode_destroy_vobject(*vpp); 1015 (*vpp)->v_type = VNON; 1016 if (fs->fs_magic == FS_UFS2_MAGIC) 1017 (*vpp)->v_op = &ffs_vnodeops2; 1018 else 1019 (*vpp)->v_op = &ffs_vnodeops1; 1020 return (0); 1021 noinodes: 1022 UFS_UNLOCK(ump); 1023 if (ppsratecheck(&lastfail, &curfail, 1)) { 1024 ffs_fserr(fs, pip->i_number, "out of inodes"); 1025 uprintf("\n%s: create/symlink failed, no inodes free\n", 1026 fs->fs_fsmnt); 1027 } 1028 return (ENOSPC); 1029 } 1030 1031 /* 1032 * Find a cylinder group to place a directory. 1033 * 1034 * The policy implemented by this algorithm is to allocate a 1035 * directory inode in the same cylinder group as its parent 1036 * directory, but also to reserve space for its files inodes 1037 * and data. Restrict the number of directories which may be 1038 * allocated one after another in the same cylinder group 1039 * without intervening allocation of files. 1040 * 1041 * If we allocate a first level directory then force allocation 1042 * in another cylinder group. 1043 */ 1044 static ino_t 1045 ffs_dirpref(pip) 1046 struct inode *pip; 1047 { 1048 struct fs *fs; 1049 u_int cg, prefcg, dirsize, cgsize; 1050 u_int avgifree, avgbfree, avgndir, curdirsize; 1051 u_int minifree, minbfree, maxndir; 1052 u_int mincg, minndir; 1053 u_int maxcontigdirs; 1054 1055 mtx_assert(UFS_MTX(pip->i_ump), MA_OWNED); 1056 fs = pip->i_fs; 1057 1058 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 1059 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1060 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; 1061 1062 /* 1063 * Force allocation in another cg if creating a first level dir. 1064 */ 1065 ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref"); 1066 if (ITOV(pip)->v_vflag & VV_ROOT) { 1067 prefcg = arc4random() % fs->fs_ncg; 1068 mincg = prefcg; 1069 minndir = fs->fs_ipg; 1070 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1071 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 1072 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 1073 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1074 mincg = cg; 1075 minndir = fs->fs_cs(fs, cg).cs_ndir; 1076 } 1077 for (cg = 0; cg < prefcg; cg++) 1078 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 1079 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 1080 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1081 mincg = cg; 1082 minndir = fs->fs_cs(fs, cg).cs_ndir; 1083 } 1084 return ((ino_t)(fs->fs_ipg * mincg)); 1085 } 1086 1087 /* 1088 * Count various limits which used for 1089 * optimal allocation of a directory inode. 1090 */ 1091 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); 1092 minifree = avgifree - avgifree / 4; 1093 if (minifree < 1) 1094 minifree = 1; 1095 minbfree = avgbfree - avgbfree / 4; 1096 if (minbfree < 1) 1097 minbfree = 1; 1098 cgsize = fs->fs_fsize * fs->fs_fpg; 1099 dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir; 1100 curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0; 1101 if (dirsize < curdirsize) 1102 dirsize = curdirsize; 1103 if (dirsize <= 0) 1104 maxcontigdirs = 0; /* dirsize overflowed */ 1105 else 1106 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255); 1107 if (fs->fs_avgfpdir > 0) 1108 maxcontigdirs = min(maxcontigdirs, 1109 fs->fs_ipg / fs->fs_avgfpdir); 1110 if (maxcontigdirs == 0) 1111 maxcontigdirs = 1; 1112 1113 /* 1114 * Limit number of dirs in one cg and reserve space for 1115 * regular files, but only if we have no deficit in 1116 * inodes or space. 1117 */ 1118 prefcg = ino_to_cg(fs, pip->i_number); 1119 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1120 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 1121 fs->fs_cs(fs, cg).cs_nifree >= minifree && 1122 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 1123 if (fs->fs_contigdirs[cg] < maxcontigdirs) 1124 return ((ino_t)(fs->fs_ipg * cg)); 1125 } 1126 for (cg = 0; cg < prefcg; cg++) 1127 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 1128 fs->fs_cs(fs, cg).cs_nifree >= minifree && 1129 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 1130 if (fs->fs_contigdirs[cg] < maxcontigdirs) 1131 return ((ino_t)(fs->fs_ipg * cg)); 1132 } 1133 /* 1134 * This is a backstop when we have deficit in space. 1135 */ 1136 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1137 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 1138 return ((ino_t)(fs->fs_ipg * cg)); 1139 for (cg = 0; cg < prefcg; cg++) 1140 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 1141 break; 1142 return ((ino_t)(fs->fs_ipg * cg)); 1143 } 1144 1145 /* 1146 * Select the desired position for the next block in a file. The file is 1147 * logically divided into sections. The first section is composed of the 1148 * direct blocks. Each additional section contains fs_maxbpg blocks. 1149 * 1150 * If no blocks have been allocated in the first section, the policy is to 1151 * request a block in the same cylinder group as the inode that describes 1152 * the file. If no blocks have been allocated in any other section, the 1153 * policy is to place the section in a cylinder group with a greater than 1154 * average number of free blocks. An appropriate cylinder group is found 1155 * by using a rotor that sweeps the cylinder groups. When a new group of 1156 * blocks is needed, the sweep begins in the cylinder group following the 1157 * cylinder group from which the previous allocation was made. The sweep 1158 * continues until a cylinder group with greater than the average number 1159 * of free blocks is found. If the allocation is for the first block in an 1160 * indirect block, the information on the previous allocation is unavailable; 1161 * here a best guess is made based upon the logical block number being 1162 * allocated. 1163 * 1164 * If a section is already partially allocated, the policy is to 1165 * contiguously allocate fs_maxcontig blocks. The end of one of these 1166 * contiguous blocks and the beginning of the next is laid out 1167 * contiguously if possible. 1168 */ 1169 ufs2_daddr_t 1170 ffs_blkpref_ufs1(ip, lbn, indx, bap) 1171 struct inode *ip; 1172 ufs_lbn_t lbn; 1173 int indx; 1174 ufs1_daddr_t *bap; 1175 { 1176 struct fs *fs; 1177 u_int cg; 1178 u_int avgbfree, startcg; 1179 1180 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); 1181 fs = ip->i_fs; 1182 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 1183 if (lbn < NDADDR + NINDIR(fs)) { 1184 cg = ino_to_cg(fs, ip->i_number); 1185 return (cgbase(fs, cg) + fs->fs_frag); 1186 } 1187 /* 1188 * Find a cylinder with greater than average number of 1189 * unused data blocks. 1190 */ 1191 if (indx == 0 || bap[indx - 1] == 0) 1192 startcg = 1193 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 1194 else 1195 startcg = dtog(fs, bap[indx - 1]) + 1; 1196 startcg %= fs->fs_ncg; 1197 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1198 for (cg = startcg; cg < fs->fs_ncg; cg++) 1199 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1200 fs->fs_cgrotor = cg; 1201 return (cgbase(fs, cg) + fs->fs_frag); 1202 } 1203 for (cg = 0; cg <= startcg; cg++) 1204 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1205 fs->fs_cgrotor = cg; 1206 return (cgbase(fs, cg) + fs->fs_frag); 1207 } 1208 return (0); 1209 } 1210 /* 1211 * We just always try to lay things out contiguously. 1212 */ 1213 return (bap[indx - 1] + fs->fs_frag); 1214 } 1215 1216 /* 1217 * Same as above, but for UFS2 1218 */ 1219 ufs2_daddr_t 1220 ffs_blkpref_ufs2(ip, lbn, indx, bap) 1221 struct inode *ip; 1222 ufs_lbn_t lbn; 1223 int indx; 1224 ufs2_daddr_t *bap; 1225 { 1226 struct fs *fs; 1227 u_int cg; 1228 u_int avgbfree, startcg; 1229 1230 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); 1231 fs = ip->i_fs; 1232 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 1233 if (lbn < NDADDR + NINDIR(fs)) { 1234 cg = ino_to_cg(fs, ip->i_number); 1235 return (cgbase(fs, cg) + fs->fs_frag); 1236 } 1237 /* 1238 * Find a cylinder with greater than average number of 1239 * unused data blocks. 1240 */ 1241 if (indx == 0 || bap[indx - 1] == 0) 1242 startcg = 1243 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 1244 else 1245 startcg = dtog(fs, bap[indx - 1]) + 1; 1246 startcg %= fs->fs_ncg; 1247 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1248 for (cg = startcg; cg < fs->fs_ncg; cg++) 1249 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1250 fs->fs_cgrotor = cg; 1251 return (cgbase(fs, cg) + fs->fs_frag); 1252 } 1253 for (cg = 0; cg <= startcg; cg++) 1254 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1255 fs->fs_cgrotor = cg; 1256 return (cgbase(fs, cg) + fs->fs_frag); 1257 } 1258 return (0); 1259 } 1260 /* 1261 * We just always try to lay things out contiguously. 1262 */ 1263 return (bap[indx - 1] + fs->fs_frag); 1264 } 1265 1266 /* 1267 * Implement the cylinder overflow algorithm. 1268 * 1269 * The policy implemented by this algorithm is: 1270 * 1) allocate the block in its requested cylinder group. 1271 * 2) quadradically rehash on the cylinder group number. 1272 * 3) brute force search for a free block. 1273 * 1274 * Must be called with the UFS lock held. Will release the lock on success 1275 * and return with it held on failure. 1276 */ 1277 /*VARARGS5*/ 1278 static ufs2_daddr_t 1279 ffs_hashalloc(ip, cg, pref, size, rsize, allocator) 1280 struct inode *ip; 1281 u_int cg; 1282 ufs2_daddr_t pref; 1283 int size; /* Search size for data blocks, mode for inodes */ 1284 int rsize; /* Real allocated size. */ 1285 allocfcn_t *allocator; 1286 { 1287 struct fs *fs; 1288 ufs2_daddr_t result; 1289 u_int i, icg = cg; 1290 1291 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); 1292 #ifdef INVARIANTS 1293 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) 1294 panic("ffs_hashalloc: allocation on suspended filesystem"); 1295 #endif 1296 fs = ip->i_fs; 1297 /* 1298 * 1: preferred cylinder group 1299 */ 1300 result = (*allocator)(ip, cg, pref, size, rsize); 1301 if (result) 1302 return (result); 1303 /* 1304 * 2: quadratic rehash 1305 */ 1306 for (i = 1; i < fs->fs_ncg; i *= 2) { 1307 cg += i; 1308 if (cg >= fs->fs_ncg) 1309 cg -= fs->fs_ncg; 1310 result = (*allocator)(ip, cg, 0, size, rsize); 1311 if (result) 1312 return (result); 1313 } 1314 /* 1315 * 3: brute force search 1316 * Note that we start at i == 2, since 0 was checked initially, 1317 * and 1 is always checked in the quadratic rehash. 1318 */ 1319 cg = (icg + 2) % fs->fs_ncg; 1320 for (i = 2; i < fs->fs_ncg; i++) { 1321 result = (*allocator)(ip, cg, 0, size, rsize); 1322 if (result) 1323 return (result); 1324 cg++; 1325 if (cg == fs->fs_ncg) 1326 cg = 0; 1327 } 1328 return (0); 1329 } 1330 1331 /* 1332 * Determine whether a fragment can be extended. 1333 * 1334 * Check to see if the necessary fragments are available, and 1335 * if they are, allocate them. 1336 */ 1337 static ufs2_daddr_t 1338 ffs_fragextend(ip, cg, bprev, osize, nsize) 1339 struct inode *ip; 1340 u_int cg; 1341 ufs2_daddr_t bprev; 1342 int osize, nsize; 1343 { 1344 struct fs *fs; 1345 struct cg *cgp; 1346 struct buf *bp; 1347 struct ufsmount *ump; 1348 int nffree; 1349 long bno; 1350 int frags, bbase; 1351 int i, error; 1352 u_int8_t *blksfree; 1353 1354 ump = ip->i_ump; 1355 fs = ip->i_fs; 1356 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 1357 return (0); 1358 frags = numfrags(fs, nsize); 1359 bbase = fragnum(fs, bprev); 1360 if (bbase > fragnum(fs, (bprev + frags - 1))) { 1361 /* cannot extend across a block boundary */ 1362 return (0); 1363 } 1364 UFS_UNLOCK(ump); 1365 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1366 (int)fs->fs_cgsize, NOCRED, &bp); 1367 if (error) 1368 goto fail; 1369 cgp = (struct cg *)bp->b_data; 1370 if (!cg_chkmagic(cgp)) 1371 goto fail; 1372 bp->b_xflags |= BX_BKGRDWRITE; 1373 cgp->cg_old_time = cgp->cg_time = time_second; 1374 bno = dtogd(fs, bprev); 1375 blksfree = cg_blksfree(cgp); 1376 for (i = numfrags(fs, osize); i < frags; i++) 1377 if (isclr(blksfree, bno + i)) 1378 goto fail; 1379 /* 1380 * the current fragment can be extended 1381 * deduct the count on fragment being extended into 1382 * increase the count on the remaining fragment (if any) 1383 * allocate the extended piece 1384 */ 1385 for (i = frags; i < fs->fs_frag - bbase; i++) 1386 if (isclr(blksfree, bno + i)) 1387 break; 1388 cgp->cg_frsum[i - numfrags(fs, osize)]--; 1389 if (i != frags) 1390 cgp->cg_frsum[i - frags]++; 1391 for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) { 1392 clrbit(blksfree, bno + i); 1393 cgp->cg_cs.cs_nffree--; 1394 nffree++; 1395 } 1396 UFS_LOCK(ump); 1397 fs->fs_cstotal.cs_nffree -= nffree; 1398 fs->fs_cs(fs, cg).cs_nffree -= nffree; 1399 fs->fs_fmod = 1; 1400 ACTIVECLEAR(fs, cg); 1401 UFS_UNLOCK(ump); 1402 if (DOINGSOFTDEP(ITOV(ip))) 1403 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev, 1404 frags, numfrags(fs, osize)); 1405 bdwrite(bp); 1406 return (bprev); 1407 1408 fail: 1409 brelse(bp); 1410 UFS_LOCK(ump); 1411 return (0); 1412 1413 } 1414 1415 /* 1416 * Determine whether a block can be allocated. 1417 * 1418 * Check to see if a block of the appropriate size is available, 1419 * and if it is, allocate it. 1420 */ 1421 static ufs2_daddr_t 1422 ffs_alloccg(ip, cg, bpref, size, rsize) 1423 struct inode *ip; 1424 u_int cg; 1425 ufs2_daddr_t bpref; 1426 int size; 1427 int rsize; 1428 { 1429 struct fs *fs; 1430 struct cg *cgp; 1431 struct buf *bp; 1432 struct ufsmount *ump; 1433 ufs1_daddr_t bno; 1434 ufs2_daddr_t blkno; 1435 int i, allocsiz, error, frags; 1436 u_int8_t *blksfree; 1437 1438 ump = ip->i_ump; 1439 fs = ip->i_fs; 1440 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 1441 return (0); 1442 UFS_UNLOCK(ump); 1443 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1444 (int)fs->fs_cgsize, NOCRED, &bp); 1445 if (error) 1446 goto fail; 1447 cgp = (struct cg *)bp->b_data; 1448 if (!cg_chkmagic(cgp) || 1449 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) 1450 goto fail; 1451 bp->b_xflags |= BX_BKGRDWRITE; 1452 cgp->cg_old_time = cgp->cg_time = time_second; 1453 if (size == fs->fs_bsize) { 1454 UFS_LOCK(ump); 1455 blkno = ffs_alloccgblk(ip, bp, bpref, rsize); 1456 ACTIVECLEAR(fs, cg); 1457 UFS_UNLOCK(ump); 1458 bdwrite(bp); 1459 return (blkno); 1460 } 1461 /* 1462 * check to see if any fragments are already available 1463 * allocsiz is the size which will be allocated, hacking 1464 * it down to a smaller size if necessary 1465 */ 1466 blksfree = cg_blksfree(cgp); 1467 frags = numfrags(fs, size); 1468 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 1469 if (cgp->cg_frsum[allocsiz] != 0) 1470 break; 1471 if (allocsiz == fs->fs_frag) { 1472 /* 1473 * no fragments were available, so a block will be 1474 * allocated, and hacked up 1475 */ 1476 if (cgp->cg_cs.cs_nbfree == 0) 1477 goto fail; 1478 UFS_LOCK(ump); 1479 blkno = ffs_alloccgblk(ip, bp, bpref, rsize); 1480 ACTIVECLEAR(fs, cg); 1481 UFS_UNLOCK(ump); 1482 bdwrite(bp); 1483 return (blkno); 1484 } 1485 KASSERT(size == rsize, 1486 ("ffs_alloccg: size(%d) != rsize(%d)", size, rsize)); 1487 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 1488 if (bno < 0) 1489 goto fail; 1490 for (i = 0; i < frags; i++) 1491 clrbit(blksfree, bno + i); 1492 cgp->cg_cs.cs_nffree -= frags; 1493 cgp->cg_frsum[allocsiz]--; 1494 if (frags != allocsiz) 1495 cgp->cg_frsum[allocsiz - frags]++; 1496 UFS_LOCK(ump); 1497 fs->fs_cstotal.cs_nffree -= frags; 1498 fs->fs_cs(fs, cg).cs_nffree -= frags; 1499 fs->fs_fmod = 1; 1500 blkno = cgbase(fs, cg) + bno; 1501 ACTIVECLEAR(fs, cg); 1502 UFS_UNLOCK(ump); 1503 if (DOINGSOFTDEP(ITOV(ip))) 1504 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, frags, 0); 1505 bdwrite(bp); 1506 return (blkno); 1507 1508 fail: 1509 brelse(bp); 1510 UFS_LOCK(ump); 1511 return (0); 1512 } 1513 1514 /* 1515 * Allocate a block in a cylinder group. 1516 * 1517 * This algorithm implements the following policy: 1518 * 1) allocate the requested block. 1519 * 2) allocate a rotationally optimal block in the same cylinder. 1520 * 3) allocate the next available block on the block rotor for the 1521 * specified cylinder group. 1522 * Note that this routine only allocates fs_bsize blocks; these 1523 * blocks may be fragmented by the routine that allocates them. 1524 */ 1525 static ufs2_daddr_t 1526 ffs_alloccgblk(ip, bp, bpref, size) 1527 struct inode *ip; 1528 struct buf *bp; 1529 ufs2_daddr_t bpref; 1530 int size; 1531 { 1532 struct fs *fs; 1533 struct cg *cgp; 1534 struct ufsmount *ump; 1535 ufs1_daddr_t bno; 1536 ufs2_daddr_t blkno; 1537 u_int8_t *blksfree; 1538 int i; 1539 1540 fs = ip->i_fs; 1541 ump = ip->i_ump; 1542 mtx_assert(UFS_MTX(ump), MA_OWNED); 1543 cgp = (struct cg *)bp->b_data; 1544 blksfree = cg_blksfree(cgp); 1545 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) { 1546 bpref = cgp->cg_rotor; 1547 } else { 1548 bpref = blknum(fs, bpref); 1549 bno = dtogd(fs, bpref); 1550 /* 1551 * if the requested block is available, use it 1552 */ 1553 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) 1554 goto gotit; 1555 } 1556 /* 1557 * Take the next available block in this cylinder group. 1558 */ 1559 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1560 if (bno < 0) 1561 return (0); 1562 cgp->cg_rotor = bno; 1563 gotit: 1564 blkno = fragstoblks(fs, bno); 1565 ffs_clrblock(fs, blksfree, (long)blkno); 1566 ffs_clusteracct(fs, cgp, blkno, -1); 1567 cgp->cg_cs.cs_nbfree--; 1568 fs->fs_cstotal.cs_nbfree--; 1569 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; 1570 fs->fs_fmod = 1; 1571 blkno = cgbase(fs, cgp->cg_cgx) + bno; 1572 /* 1573 * If the caller didn't want the whole block free the frags here. 1574 */ 1575 size = numfrags(fs, size); 1576 if (size != fs->fs_frag) { 1577 bno = dtogd(fs, blkno); 1578 for (i = size; i < fs->fs_frag; i++) 1579 setbit(blksfree, bno + i); 1580 i = fs->fs_frag - size; 1581 cgp->cg_cs.cs_nffree += i; 1582 fs->fs_cstotal.cs_nffree += i; 1583 fs->fs_cs(fs, cgp->cg_cgx).cs_nffree += i; 1584 fs->fs_fmod = 1; 1585 cgp->cg_frsum[i]++; 1586 } 1587 /* XXX Fixme. */ 1588 UFS_UNLOCK(ump); 1589 if (DOINGSOFTDEP(ITOV(ip))) 1590 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, 1591 size, 0); 1592 UFS_LOCK(ump); 1593 return (blkno); 1594 } 1595 1596 /* 1597 * Determine whether a cluster can be allocated. 1598 * 1599 * We do not currently check for optimal rotational layout if there 1600 * are multiple choices in the same cylinder group. Instead we just 1601 * take the first one that we find following bpref. 1602 */ 1603 static ufs2_daddr_t 1604 ffs_clusteralloc(ip, cg, bpref, len, unused) 1605 struct inode *ip; 1606 u_int cg; 1607 ufs2_daddr_t bpref; 1608 int len; 1609 int unused; 1610 { 1611 struct fs *fs; 1612 struct cg *cgp; 1613 struct buf *bp; 1614 struct ufsmount *ump; 1615 int i, run, bit, map, got; 1616 ufs2_daddr_t bno; 1617 u_char *mapp; 1618 int32_t *lp; 1619 u_int8_t *blksfree; 1620 1621 fs = ip->i_fs; 1622 ump = ip->i_ump; 1623 if (fs->fs_maxcluster[cg] < len) 1624 return (0); 1625 UFS_UNLOCK(ump); 1626 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize, 1627 NOCRED, &bp)) 1628 goto fail_lock; 1629 cgp = (struct cg *)bp->b_data; 1630 if (!cg_chkmagic(cgp)) 1631 goto fail_lock; 1632 bp->b_xflags |= BX_BKGRDWRITE; 1633 /* 1634 * Check to see if a cluster of the needed size (or bigger) is 1635 * available in this cylinder group. 1636 */ 1637 lp = &cg_clustersum(cgp)[len]; 1638 for (i = len; i <= fs->fs_contigsumsize; i++) 1639 if (*lp++ > 0) 1640 break; 1641 if (i > fs->fs_contigsumsize) { 1642 /* 1643 * This is the first time looking for a cluster in this 1644 * cylinder group. Update the cluster summary information 1645 * to reflect the true maximum sized cluster so that 1646 * future cluster allocation requests can avoid reading 1647 * the cylinder group map only to find no clusters. 1648 */ 1649 lp = &cg_clustersum(cgp)[len - 1]; 1650 for (i = len - 1; i > 0; i--) 1651 if (*lp-- > 0) 1652 break; 1653 UFS_LOCK(ump); 1654 fs->fs_maxcluster[cg] = i; 1655 goto fail; 1656 } 1657 /* 1658 * Search the cluster map to find a big enough cluster. 1659 * We take the first one that we find, even if it is larger 1660 * than we need as we prefer to get one close to the previous 1661 * block allocation. We do not search before the current 1662 * preference point as we do not want to allocate a block 1663 * that is allocated before the previous one (as we will 1664 * then have to wait for another pass of the elevator 1665 * algorithm before it will be read). We prefer to fail and 1666 * be recalled to try an allocation in the next cylinder group. 1667 */ 1668 if (dtog(fs, bpref) != cg) 1669 bpref = 0; 1670 else 1671 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref))); 1672 mapp = &cg_clustersfree(cgp)[bpref / NBBY]; 1673 map = *mapp++; 1674 bit = 1 << (bpref % NBBY); 1675 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { 1676 if ((map & bit) == 0) { 1677 run = 0; 1678 } else { 1679 run++; 1680 if (run == len) 1681 break; 1682 } 1683 if ((got & (NBBY - 1)) != (NBBY - 1)) { 1684 bit <<= 1; 1685 } else { 1686 map = *mapp++; 1687 bit = 1; 1688 } 1689 } 1690 if (got >= cgp->cg_nclusterblks) 1691 goto fail_lock; 1692 /* 1693 * Allocate the cluster that we have found. 1694 */ 1695 blksfree = cg_blksfree(cgp); 1696 for (i = 1; i <= len; i++) 1697 if (!ffs_isblock(fs, blksfree, got - run + i)) 1698 panic("ffs_clusteralloc: map mismatch"); 1699 bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1); 1700 if (dtog(fs, bno) != cg) 1701 panic("ffs_clusteralloc: allocated out of group"); 1702 len = blkstofrags(fs, len); 1703 UFS_LOCK(ump); 1704 for (i = 0; i < len; i += fs->fs_frag) 1705 if (ffs_alloccgblk(ip, bp, bno + i, fs->fs_bsize) != bno + i) 1706 panic("ffs_clusteralloc: lost block"); 1707 ACTIVECLEAR(fs, cg); 1708 UFS_UNLOCK(ump); 1709 bdwrite(bp); 1710 return (bno); 1711 1712 fail_lock: 1713 UFS_LOCK(ump); 1714 fail: 1715 brelse(bp); 1716 return (0); 1717 } 1718 1719 /* 1720 * Determine whether an inode can be allocated. 1721 * 1722 * Check to see if an inode is available, and if it is, 1723 * allocate it using the following policy: 1724 * 1) allocate the requested inode. 1725 * 2) allocate the next available inode after the requested 1726 * inode in the specified cylinder group. 1727 */ 1728 static ufs2_daddr_t 1729 ffs_nodealloccg(ip, cg, ipref, mode, unused) 1730 struct inode *ip; 1731 u_int cg; 1732 ufs2_daddr_t ipref; 1733 int mode; 1734 int unused; 1735 { 1736 struct fs *fs; 1737 struct cg *cgp; 1738 struct buf *bp, *ibp; 1739 struct ufsmount *ump; 1740 u_int8_t *inosused; 1741 struct ufs2_dinode *dp2; 1742 int error, start, len, loc, map, i; 1743 1744 fs = ip->i_fs; 1745 ump = ip->i_ump; 1746 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1747 return (0); 1748 UFS_UNLOCK(ump); 1749 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1750 (int)fs->fs_cgsize, NOCRED, &bp); 1751 if (error) { 1752 brelse(bp); 1753 UFS_LOCK(ump); 1754 return (0); 1755 } 1756 cgp = (struct cg *)bp->b_data; 1757 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) { 1758 brelse(bp); 1759 UFS_LOCK(ump); 1760 return (0); 1761 } 1762 bp->b_xflags |= BX_BKGRDWRITE; 1763 cgp->cg_old_time = cgp->cg_time = time_second; 1764 inosused = cg_inosused(cgp); 1765 if (ipref) { 1766 ipref %= fs->fs_ipg; 1767 if (isclr(inosused, ipref)) 1768 goto gotit; 1769 } 1770 start = cgp->cg_irotor / NBBY; 1771 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); 1772 loc = skpc(0xff, len, &inosused[start]); 1773 if (loc == 0) { 1774 len = start + 1; 1775 start = 0; 1776 loc = skpc(0xff, len, &inosused[0]); 1777 if (loc == 0) { 1778 printf("cg = %d, irotor = %ld, fs = %s\n", 1779 cg, (long)cgp->cg_irotor, fs->fs_fsmnt); 1780 panic("ffs_nodealloccg: map corrupted"); 1781 /* NOTREACHED */ 1782 } 1783 } 1784 i = start + len - loc; 1785 map = inosused[i] ^ 0xff; 1786 if (map == 0) { 1787 printf("fs = %s\n", fs->fs_fsmnt); 1788 panic("ffs_nodealloccg: block not in map"); 1789 } 1790 ipref = i * NBBY + ffs(map) - 1; 1791 cgp->cg_irotor = ipref; 1792 gotit: 1793 /* 1794 * Check to see if we need to initialize more inodes. 1795 */ 1796 ibp = NULL; 1797 if (fs->fs_magic == FS_UFS2_MAGIC && 1798 ipref + INOPB(fs) > cgp->cg_initediblk && 1799 cgp->cg_initediblk < cgp->cg_niblk) { 1800 ibp = getblk(ip->i_devvp, fsbtodb(fs, 1801 ino_to_fsba(fs, cg * fs->fs_ipg + cgp->cg_initediblk)), 1802 (int)fs->fs_bsize, 0, 0, 0); 1803 bzero(ibp->b_data, (int)fs->fs_bsize); 1804 dp2 = (struct ufs2_dinode *)(ibp->b_data); 1805 for (i = 0; i < INOPB(fs); i++) { 1806 dp2->di_gen = arc4random() / 2 + 1; 1807 dp2++; 1808 } 1809 cgp->cg_initediblk += INOPB(fs); 1810 } 1811 UFS_LOCK(ump); 1812 ACTIVECLEAR(fs, cg); 1813 setbit(inosused, ipref); 1814 cgp->cg_cs.cs_nifree--; 1815 fs->fs_cstotal.cs_nifree--; 1816 fs->fs_cs(fs, cg).cs_nifree--; 1817 fs->fs_fmod = 1; 1818 if ((mode & IFMT) == IFDIR) { 1819 cgp->cg_cs.cs_ndir++; 1820 fs->fs_cstotal.cs_ndir++; 1821 fs->fs_cs(fs, cg).cs_ndir++; 1822 } 1823 UFS_UNLOCK(ump); 1824 if (DOINGSOFTDEP(ITOV(ip))) 1825 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref); 1826 bdwrite(bp); 1827 if (ibp != NULL) 1828 bawrite(ibp); 1829 return ((ino_t)(cg * fs->fs_ipg + ipref)); 1830 } 1831 1832 /* 1833 * Free a block or fragment. 1834 * 1835 * The specified block or fragment is placed back in the 1836 * free map. If a fragment is deallocated, a possible 1837 * block reassembly is checked. 1838 */ 1839 static void 1840 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd) 1841 struct ufsmount *ump; 1842 struct fs *fs; 1843 struct vnode *devvp; 1844 ufs2_daddr_t bno; 1845 long size; 1846 ino_t inum; 1847 struct workhead *dephd; 1848 { 1849 struct mount *mp; 1850 struct cg *cgp; 1851 struct buf *bp; 1852 ufs1_daddr_t fragno, cgbno; 1853 ufs2_daddr_t cgblkno; 1854 int i, blk, frags, bbase; 1855 u_int cg; 1856 u_int8_t *blksfree; 1857 struct cdev *dev; 1858 1859 cg = dtog(fs, bno); 1860 if (devvp->v_type == VREG) { 1861 /* devvp is a snapshot */ 1862 dev = VTOI(devvp)->i_devvp->v_rdev; 1863 cgblkno = fragstoblks(fs, cgtod(fs, cg)); 1864 } else { 1865 /* devvp is a normal disk device */ 1866 dev = devvp->v_rdev; 1867 cgblkno = fsbtodb(fs, cgtod(fs, cg)); 1868 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree"); 1869 if ((devvp->v_vflag & VV_COPYONWRITE) && 1870 ffs_snapblkfree(fs, devvp, bno, size, inum)) 1871 return; 1872 } 1873 #ifdef INVARIANTS 1874 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 1875 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 1876 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n", 1877 devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize, 1878 size, fs->fs_fsmnt); 1879 panic("ffs_blkfree: bad size"); 1880 } 1881 #endif 1882 if ((u_int)bno >= fs->fs_size) { 1883 printf("bad block %jd, ino %lu\n", (intmax_t)bno, 1884 (u_long)inum); 1885 ffs_fserr(fs, inum, "bad block"); 1886 return; 1887 } 1888 if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) { 1889 brelse(bp); 1890 return; 1891 } 1892 cgp = (struct cg *)bp->b_data; 1893 if (!cg_chkmagic(cgp)) { 1894 brelse(bp); 1895 return; 1896 } 1897 bp->b_xflags |= BX_BKGRDWRITE; 1898 cgp->cg_old_time = cgp->cg_time = time_second; 1899 cgbno = dtogd(fs, bno); 1900 blksfree = cg_blksfree(cgp); 1901 UFS_LOCK(ump); 1902 if (size == fs->fs_bsize) { 1903 fragno = fragstoblks(fs, cgbno); 1904 if (!ffs_isfreeblock(fs, blksfree, fragno)) { 1905 if (devvp->v_type == VREG) { 1906 UFS_UNLOCK(ump); 1907 /* devvp is a snapshot */ 1908 brelse(bp); 1909 return; 1910 } 1911 printf("dev = %s, block = %jd, fs = %s\n", 1912 devtoname(dev), (intmax_t)bno, fs->fs_fsmnt); 1913 panic("ffs_blkfree: freeing free block"); 1914 } 1915 ffs_setblock(fs, blksfree, fragno); 1916 ffs_clusteracct(fs, cgp, fragno, 1); 1917 cgp->cg_cs.cs_nbfree++; 1918 fs->fs_cstotal.cs_nbfree++; 1919 fs->fs_cs(fs, cg).cs_nbfree++; 1920 } else { 1921 bbase = cgbno - fragnum(fs, cgbno); 1922 /* 1923 * decrement the counts associated with the old frags 1924 */ 1925 blk = blkmap(fs, blksfree, bbase); 1926 ffs_fragacct(fs, blk, cgp->cg_frsum, -1); 1927 /* 1928 * deallocate the fragment 1929 */ 1930 frags = numfrags(fs, size); 1931 for (i = 0; i < frags; i++) { 1932 if (isset(blksfree, cgbno + i)) { 1933 printf("dev = %s, block = %jd, fs = %s\n", 1934 devtoname(dev), (intmax_t)(bno + i), 1935 fs->fs_fsmnt); 1936 panic("ffs_blkfree: freeing free frag"); 1937 } 1938 setbit(blksfree, cgbno + i); 1939 } 1940 cgp->cg_cs.cs_nffree += i; 1941 fs->fs_cstotal.cs_nffree += i; 1942 fs->fs_cs(fs, cg).cs_nffree += i; 1943 /* 1944 * add back in counts associated with the new frags 1945 */ 1946 blk = blkmap(fs, blksfree, bbase); 1947 ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 1948 /* 1949 * if a complete block has been reassembled, account for it 1950 */ 1951 fragno = fragstoblks(fs, bbase); 1952 if (ffs_isblock(fs, blksfree, fragno)) { 1953 cgp->cg_cs.cs_nffree -= fs->fs_frag; 1954 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 1955 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 1956 ffs_clusteracct(fs, cgp, fragno, 1); 1957 cgp->cg_cs.cs_nbfree++; 1958 fs->fs_cstotal.cs_nbfree++; 1959 fs->fs_cs(fs, cg).cs_nbfree++; 1960 } 1961 } 1962 fs->fs_fmod = 1; 1963 ACTIVECLEAR(fs, cg); 1964 UFS_UNLOCK(ump); 1965 mp = UFSTOVFS(ump); 1966 if (mp->mnt_flag & MNT_SOFTDEP && devvp->v_type != VREG) 1967 softdep_setup_blkfree(UFSTOVFS(ump), bp, bno, 1968 numfrags(fs, size), dephd); 1969 bdwrite(bp); 1970 } 1971 1972 TASKQUEUE_DEFINE_THREAD(ffs_trim); 1973 1974 struct ffs_blkfree_trim_params { 1975 struct task task; 1976 struct ufsmount *ump; 1977 struct vnode *devvp; 1978 ufs2_daddr_t bno; 1979 long size; 1980 ino_t inum; 1981 struct workhead *pdephd; 1982 struct workhead dephd; 1983 }; 1984 1985 static void 1986 ffs_blkfree_trim_task(ctx, pending) 1987 void *ctx; 1988 int pending; 1989 { 1990 struct ffs_blkfree_trim_params *tp; 1991 1992 tp = ctx; 1993 ffs_blkfree_cg(tp->ump, tp->ump->um_fs, tp->devvp, tp->bno, tp->size, 1994 tp->inum, tp->pdephd); 1995 vn_finished_secondary_write(UFSTOVFS(tp->ump)); 1996 free(tp, M_TEMP); 1997 } 1998 1999 static void 2000 ffs_blkfree_trim_completed(bip) 2001 struct bio *bip; 2002 { 2003 struct ffs_blkfree_trim_params *tp; 2004 2005 tp = bip->bio_caller2; 2006 g_destroy_bio(bip); 2007 TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp); 2008 taskqueue_enqueue(taskqueue_ffs_trim, &tp->task); 2009 } 2010 2011 void 2012 ffs_blkfree(ump, fs, devvp, bno, size, inum, dephd) 2013 struct ufsmount *ump; 2014 struct fs *fs; 2015 struct vnode *devvp; 2016 ufs2_daddr_t bno; 2017 long size; 2018 ino_t inum; 2019 struct workhead *dephd; 2020 { 2021 struct mount *mp; 2022 struct bio *bip; 2023 struct ffs_blkfree_trim_params *tp; 2024 2025 if (!ump->um_candelete) { 2026 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd); 2027 return; 2028 } 2029 2030 /* 2031 * Postpone the set of the free bit in the cg bitmap until the 2032 * BIO_DELETE is completed. Otherwise, due to disk queue 2033 * reordering, TRIM might be issued after we reuse the block 2034 * and write some new data into it. 2035 */ 2036 tp = malloc(sizeof(struct ffs_blkfree_trim_params), M_TEMP, M_WAITOK); 2037 tp->ump = ump; 2038 tp->devvp = devvp; 2039 tp->bno = bno; 2040 tp->size = size; 2041 tp->inum = inum; 2042 if (dephd != NULL) { 2043 LIST_INIT(&tp->dephd); 2044 LIST_SWAP(dephd, &tp->dephd, worklist, wk_list); 2045 tp->pdephd = &tp->dephd; 2046 } else 2047 tp->pdephd = NULL; 2048 2049 bip = g_alloc_bio(); 2050 bip->bio_cmd = BIO_DELETE; 2051 bip->bio_offset = dbtob(fsbtodb(fs, bno)); 2052 bip->bio_done = ffs_blkfree_trim_completed; 2053 bip->bio_length = size; 2054 bip->bio_caller2 = tp; 2055 2056 mp = UFSTOVFS(ump); 2057 vn_start_secondary_write(NULL, &mp, 0); 2058 g_io_request(bip, (struct g_consumer *)devvp->v_bufobj.bo_private); 2059 } 2060 2061 #ifdef INVARIANTS 2062 /* 2063 * Verify allocation of a block or fragment. Returns true if block or 2064 * fragment is allocated, false if it is free. 2065 */ 2066 static int 2067 ffs_checkblk(ip, bno, size) 2068 struct inode *ip; 2069 ufs2_daddr_t bno; 2070 long size; 2071 { 2072 struct fs *fs; 2073 struct cg *cgp; 2074 struct buf *bp; 2075 ufs1_daddr_t cgbno; 2076 int i, error, frags, free; 2077 u_int8_t *blksfree; 2078 2079 fs = ip->i_fs; 2080 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 2081 printf("bsize = %ld, size = %ld, fs = %s\n", 2082 (long)fs->fs_bsize, size, fs->fs_fsmnt); 2083 panic("ffs_checkblk: bad size"); 2084 } 2085 if ((u_int)bno >= fs->fs_size) 2086 panic("ffs_checkblk: bad block %jd", (intmax_t)bno); 2087 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))), 2088 (int)fs->fs_cgsize, NOCRED, &bp); 2089 if (error) 2090 panic("ffs_checkblk: cg bread failed"); 2091 cgp = (struct cg *)bp->b_data; 2092 if (!cg_chkmagic(cgp)) 2093 panic("ffs_checkblk: cg magic mismatch"); 2094 bp->b_xflags |= BX_BKGRDWRITE; 2095 blksfree = cg_blksfree(cgp); 2096 cgbno = dtogd(fs, bno); 2097 if (size == fs->fs_bsize) { 2098 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno)); 2099 } else { 2100 frags = numfrags(fs, size); 2101 for (free = 0, i = 0; i < frags; i++) 2102 if (isset(blksfree, cgbno + i)) 2103 free++; 2104 if (free != 0 && free != frags) 2105 panic("ffs_checkblk: partially free fragment"); 2106 } 2107 brelse(bp); 2108 return (!free); 2109 } 2110 #endif /* INVARIANTS */ 2111 2112 /* 2113 * Free an inode. 2114 */ 2115 int 2116 ffs_vfree(pvp, ino, mode) 2117 struct vnode *pvp; 2118 ino_t ino; 2119 int mode; 2120 { 2121 struct inode *ip; 2122 2123 if (DOINGSOFTDEP(pvp)) { 2124 softdep_freefile(pvp, ino, mode); 2125 return (0); 2126 } 2127 ip = VTOI(pvp); 2128 return (ffs_freefile(ip->i_ump, ip->i_fs, ip->i_devvp, ino, mode, 2129 NULL)); 2130 } 2131 2132 /* 2133 * Do the actual free operation. 2134 * The specified inode is placed back in the free map. 2135 */ 2136 int 2137 ffs_freefile(ump, fs, devvp, ino, mode, wkhd) 2138 struct ufsmount *ump; 2139 struct fs *fs; 2140 struct vnode *devvp; 2141 ino_t ino; 2142 int mode; 2143 struct workhead *wkhd; 2144 { 2145 struct cg *cgp; 2146 struct buf *bp; 2147 ufs2_daddr_t cgbno; 2148 int error; 2149 u_int cg; 2150 u_int8_t *inosused; 2151 struct cdev *dev; 2152 2153 cg = ino_to_cg(fs, ino); 2154 if (devvp->v_type == VREG) { 2155 /* devvp is a snapshot */ 2156 dev = VTOI(devvp)->i_devvp->v_rdev; 2157 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2158 } else { 2159 /* devvp is a normal disk device */ 2160 dev = devvp->v_rdev; 2161 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2162 } 2163 if (ino >= fs->fs_ipg * fs->fs_ncg) 2164 panic("ffs_freefile: range: dev = %s, ino = %lu, fs = %s", 2165 devtoname(dev), (u_long)ino, fs->fs_fsmnt); 2166 if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) { 2167 brelse(bp); 2168 return (error); 2169 } 2170 cgp = (struct cg *)bp->b_data; 2171 if (!cg_chkmagic(cgp)) { 2172 brelse(bp); 2173 return (0); 2174 } 2175 bp->b_xflags |= BX_BKGRDWRITE; 2176 cgp->cg_old_time = cgp->cg_time = time_second; 2177 inosused = cg_inosused(cgp); 2178 ino %= fs->fs_ipg; 2179 if (isclr(inosused, ino)) { 2180 printf("dev = %s, ino = %u, fs = %s\n", devtoname(dev), 2181 ino + cg * fs->fs_ipg, fs->fs_fsmnt); 2182 if (fs->fs_ronly == 0) 2183 panic("ffs_freefile: freeing free inode"); 2184 } 2185 clrbit(inosused, ino); 2186 if (ino < cgp->cg_irotor) 2187 cgp->cg_irotor = ino; 2188 cgp->cg_cs.cs_nifree++; 2189 UFS_LOCK(ump); 2190 fs->fs_cstotal.cs_nifree++; 2191 fs->fs_cs(fs, cg).cs_nifree++; 2192 if ((mode & IFMT) == IFDIR) { 2193 cgp->cg_cs.cs_ndir--; 2194 fs->fs_cstotal.cs_ndir--; 2195 fs->fs_cs(fs, cg).cs_ndir--; 2196 } 2197 fs->fs_fmod = 1; 2198 ACTIVECLEAR(fs, cg); 2199 UFS_UNLOCK(ump); 2200 if (UFSTOVFS(ump)->mnt_flag & MNT_SOFTDEP && devvp->v_type != VREG) 2201 softdep_setup_inofree(UFSTOVFS(ump), bp, 2202 ino + cg * fs->fs_ipg, wkhd); 2203 bdwrite(bp); 2204 return (0); 2205 } 2206 2207 /* 2208 * Check to see if a file is free. 2209 */ 2210 int 2211 ffs_checkfreefile(fs, devvp, ino) 2212 struct fs *fs; 2213 struct vnode *devvp; 2214 ino_t ino; 2215 { 2216 struct cg *cgp; 2217 struct buf *bp; 2218 ufs2_daddr_t cgbno; 2219 int ret; 2220 u_int cg; 2221 u_int8_t *inosused; 2222 2223 cg = ino_to_cg(fs, ino); 2224 if (devvp->v_type == VREG) { 2225 /* devvp is a snapshot */ 2226 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2227 } else { 2228 /* devvp is a normal disk device */ 2229 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2230 } 2231 if (ino >= fs->fs_ipg * fs->fs_ncg) 2232 return (1); 2233 if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) { 2234 brelse(bp); 2235 return (1); 2236 } 2237 cgp = (struct cg *)bp->b_data; 2238 if (!cg_chkmagic(cgp)) { 2239 brelse(bp); 2240 return (1); 2241 } 2242 inosused = cg_inosused(cgp); 2243 ino %= fs->fs_ipg; 2244 ret = isclr(inosused, ino); 2245 brelse(bp); 2246 return (ret); 2247 } 2248 2249 /* 2250 * Find a block of the specified size in the specified cylinder group. 2251 * 2252 * It is a panic if a request is made to find a block if none are 2253 * available. 2254 */ 2255 static ufs1_daddr_t 2256 ffs_mapsearch(fs, cgp, bpref, allocsiz) 2257 struct fs *fs; 2258 struct cg *cgp; 2259 ufs2_daddr_t bpref; 2260 int allocsiz; 2261 { 2262 ufs1_daddr_t bno; 2263 int start, len, loc, i; 2264 int blk, field, subfield, pos; 2265 u_int8_t *blksfree; 2266 2267 /* 2268 * find the fragment by searching through the free block 2269 * map for an appropriate bit pattern 2270 */ 2271 if (bpref) 2272 start = dtogd(fs, bpref) / NBBY; 2273 else 2274 start = cgp->cg_frotor / NBBY; 2275 blksfree = cg_blksfree(cgp); 2276 len = howmany(fs->fs_fpg, NBBY) - start; 2277 loc = scanc((u_int)len, (u_char *)&blksfree[start], 2278 fragtbl[fs->fs_frag], 2279 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 2280 if (loc == 0) { 2281 len = start + 1; 2282 start = 0; 2283 loc = scanc((u_int)len, (u_char *)&blksfree[0], 2284 fragtbl[fs->fs_frag], 2285 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 2286 if (loc == 0) { 2287 printf("start = %d, len = %d, fs = %s\n", 2288 start, len, fs->fs_fsmnt); 2289 panic("ffs_alloccg: map corrupted"); 2290 /* NOTREACHED */ 2291 } 2292 } 2293 bno = (start + len - loc) * NBBY; 2294 cgp->cg_frotor = bno; 2295 /* 2296 * found the byte in the map 2297 * sift through the bits to find the selected frag 2298 */ 2299 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 2300 blk = blkmap(fs, blksfree, bno); 2301 blk <<= 1; 2302 field = around[allocsiz]; 2303 subfield = inside[allocsiz]; 2304 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 2305 if ((blk & field) == subfield) 2306 return (bno + pos); 2307 field <<= 1; 2308 subfield <<= 1; 2309 } 2310 } 2311 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); 2312 panic("ffs_alloccg: block not in map"); 2313 return (-1); 2314 } 2315 2316 /* 2317 * Fserr prints the name of a filesystem with an error diagnostic. 2318 * 2319 * The form of the error message is: 2320 * fs: error message 2321 */ 2322 static void 2323 ffs_fserr(fs, inum, cp) 2324 struct fs *fs; 2325 ino_t inum; 2326 char *cp; 2327 { 2328 struct thread *td = curthread; /* XXX */ 2329 struct proc *p = td->td_proc; 2330 2331 log(LOG_ERR, "pid %d (%s), uid %d inumber %d on %s: %s\n", 2332 p->p_pid, p->p_comm, td->td_ucred->cr_uid, inum, fs->fs_fsmnt, cp); 2333 } 2334 2335 /* 2336 * This function provides the capability for the fsck program to 2337 * update an active filesystem. Fourteen operations are provided: 2338 * 2339 * adjrefcnt(inode, amt) - adjusts the reference count on the 2340 * specified inode by the specified amount. Under normal 2341 * operation the count should always go down. Decrementing 2342 * the count to zero will cause the inode to be freed. 2343 * adjblkcnt(inode, amt) - adjust the number of blocks used to 2344 * by the specifed amount. 2345 * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) - 2346 * adjust the superblock summary. 2347 * freedirs(inode, count) - directory inodes [inode..inode + count - 1] 2348 * are marked as free. Inodes should never have to be marked 2349 * as in use. 2350 * freefiles(inode, count) - file inodes [inode..inode + count - 1] 2351 * are marked as free. Inodes should never have to be marked 2352 * as in use. 2353 * freeblks(blockno, size) - blocks [blockno..blockno + size - 1] 2354 * are marked as free. Blocks should never have to be marked 2355 * as in use. 2356 * setflags(flags, set/clear) - the fs_flags field has the specified 2357 * flags set (second parameter +1) or cleared (second parameter -1). 2358 * setcwd(dirinode) - set the current directory to dirinode in the 2359 * filesystem associated with the snapshot. 2360 * setdotdot(oldvalue, newvalue) - Verify that the inode number for ".." 2361 * in the current directory is oldvalue then change it to newvalue. 2362 * unlink(nameptr, oldvalue) - Verify that the inode number associated 2363 * with nameptr in the current directory is oldvalue then unlink it. 2364 */ 2365 2366 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS); 2367 2368 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT, 2369 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count"); 2370 2371 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR, 2372 sysctl_ffs_fsck, "Adjust Inode Used Blocks Count"); 2373 2374 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR, 2375 sysctl_ffs_fsck, "Adjust number of directories"); 2376 2377 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR, 2378 sysctl_ffs_fsck, "Adjust number of free blocks"); 2379 2380 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR, 2381 sysctl_ffs_fsck, "Adjust number of free inodes"); 2382 2383 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR, 2384 sysctl_ffs_fsck, "Adjust number of free frags"); 2385 2386 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR, 2387 sysctl_ffs_fsck, "Adjust number of free clusters"); 2388 2389 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR, 2390 sysctl_ffs_fsck, "Free Range of Directory Inodes"); 2391 2392 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR, 2393 sysctl_ffs_fsck, "Free Range of File Inodes"); 2394 2395 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR, 2396 sysctl_ffs_fsck, "Free Range of Blocks"); 2397 2398 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR, 2399 sysctl_ffs_fsck, "Change Filesystem Flags"); 2400 2401 static SYSCTL_NODE(_vfs_ffs, FFS_SET_CWD, setcwd, CTLFLAG_WR, 2402 sysctl_ffs_fsck, "Set Current Working Directory"); 2403 2404 static SYSCTL_NODE(_vfs_ffs, FFS_SET_DOTDOT, setdotdot, CTLFLAG_WR, 2405 sysctl_ffs_fsck, "Change Value of .. Entry"); 2406 2407 static SYSCTL_NODE(_vfs_ffs, FFS_UNLINK, unlink, CTLFLAG_WR, 2408 sysctl_ffs_fsck, "Unlink a Duplicate Name"); 2409 2410 #ifdef DEBUG 2411 static int fsckcmds = 0; 2412 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, ""); 2413 #endif /* DEBUG */ 2414 2415 static int 2416 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS) 2417 { 2418 struct thread *td = curthread; 2419 struct fsck_cmd cmd; 2420 struct ufsmount *ump; 2421 struct vnode *vp, *vpold, *dvp, *fdvp; 2422 struct inode *ip, *dp; 2423 struct mount *mp; 2424 struct fs *fs; 2425 ufs2_daddr_t blkno; 2426 long blkcnt, blksize; 2427 struct filedesc *fdp; 2428 struct file *fp; 2429 int vfslocked, filetype, error; 2430 2431 if (req->newlen > sizeof cmd) 2432 return (EBADRPC); 2433 if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0) 2434 return (error); 2435 if (cmd.version != FFS_CMD_VERSION) 2436 return (ERPCMISMATCH); 2437 if ((error = getvnode(curproc->p_fd, cmd.handle, &fp)) != 0) 2438 return (error); 2439 vp = fp->f_data; 2440 if (vp->v_type != VREG && vp->v_type != VDIR) { 2441 fdrop(fp, td); 2442 return (EINVAL); 2443 } 2444 vn_start_write(vp, &mp, V_WAIT); 2445 if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) { 2446 vn_finished_write(mp); 2447 fdrop(fp, td); 2448 return (EINVAL); 2449 } 2450 if (mp->mnt_flag & MNT_RDONLY) { 2451 vn_finished_write(mp); 2452 fdrop(fp, td); 2453 return (EROFS); 2454 } 2455 ump = VFSTOUFS(mp); 2456 fs = ump->um_fs; 2457 filetype = IFREG; 2458 2459 switch (oidp->oid_number) { 2460 2461 case FFS_SET_FLAGS: 2462 #ifdef DEBUG 2463 if (fsckcmds) 2464 printf("%s: %s flags\n", mp->mnt_stat.f_mntonname, 2465 cmd.size > 0 ? "set" : "clear"); 2466 #endif /* DEBUG */ 2467 if (cmd.size > 0) 2468 fs->fs_flags |= (long)cmd.value; 2469 else 2470 fs->fs_flags &= ~(long)cmd.value; 2471 break; 2472 2473 case FFS_ADJ_REFCNT: 2474 #ifdef DEBUG 2475 if (fsckcmds) { 2476 printf("%s: adjust inode %jd count by %jd\n", 2477 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2478 (intmax_t)cmd.size); 2479 } 2480 #endif /* DEBUG */ 2481 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) 2482 break; 2483 ip = VTOI(vp); 2484 ip->i_nlink += cmd.size; 2485 DIP_SET(ip, i_nlink, ip->i_nlink); 2486 ip->i_effnlink += cmd.size; 2487 ip->i_flag |= IN_CHANGE; 2488 if (DOINGSOFTDEP(vp)) 2489 softdep_change_linkcnt(ip); 2490 vput(vp); 2491 break; 2492 2493 case FFS_ADJ_BLKCNT: 2494 #ifdef DEBUG 2495 if (fsckcmds) { 2496 printf("%s: adjust inode %jd block count by %jd\n", 2497 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2498 (intmax_t)cmd.size); 2499 } 2500 #endif /* DEBUG */ 2501 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) 2502 break; 2503 ip = VTOI(vp); 2504 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size); 2505 ip->i_flag |= IN_CHANGE; 2506 vput(vp); 2507 break; 2508 2509 case FFS_DIR_FREE: 2510 filetype = IFDIR; 2511 /* fall through */ 2512 2513 case FFS_FILE_FREE: 2514 #ifdef DEBUG 2515 if (fsckcmds) { 2516 if (cmd.size == 1) 2517 printf("%s: free %s inode %d\n", 2518 mp->mnt_stat.f_mntonname, 2519 filetype == IFDIR ? "directory" : "file", 2520 (ino_t)cmd.value); 2521 else 2522 printf("%s: free %s inodes %d-%d\n", 2523 mp->mnt_stat.f_mntonname, 2524 filetype == IFDIR ? "directory" : "file", 2525 (ino_t)cmd.value, 2526 (ino_t)(cmd.value + cmd.size - 1)); 2527 } 2528 #endif /* DEBUG */ 2529 while (cmd.size > 0) { 2530 if ((error = ffs_freefile(ump, fs, ump->um_devvp, 2531 cmd.value, filetype, NULL))) 2532 break; 2533 cmd.size -= 1; 2534 cmd.value += 1; 2535 } 2536 break; 2537 2538 case FFS_BLK_FREE: 2539 #ifdef DEBUG 2540 if (fsckcmds) { 2541 if (cmd.size == 1) 2542 printf("%s: free block %jd\n", 2543 mp->mnt_stat.f_mntonname, 2544 (intmax_t)cmd.value); 2545 else 2546 printf("%s: free blocks %jd-%jd\n", 2547 mp->mnt_stat.f_mntonname, 2548 (intmax_t)cmd.value, 2549 (intmax_t)cmd.value + cmd.size - 1); 2550 } 2551 #endif /* DEBUG */ 2552 blkno = cmd.value; 2553 blkcnt = cmd.size; 2554 blksize = fs->fs_frag - (blkno % fs->fs_frag); 2555 while (blkcnt > 0) { 2556 if (blksize > blkcnt) 2557 blksize = blkcnt; 2558 ffs_blkfree(ump, fs, ump->um_devvp, blkno, 2559 blksize * fs->fs_fsize, ROOTINO, NULL); 2560 blkno += blksize; 2561 blkcnt -= blksize; 2562 blksize = fs->fs_frag; 2563 } 2564 break; 2565 2566 /* 2567 * Adjust superblock summaries. fsck(8) is expected to 2568 * submit deltas when necessary. 2569 */ 2570 case FFS_ADJ_NDIR: 2571 #ifdef DEBUG 2572 if (fsckcmds) { 2573 printf("%s: adjust number of directories by %jd\n", 2574 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2575 } 2576 #endif /* DEBUG */ 2577 fs->fs_cstotal.cs_ndir += cmd.value; 2578 break; 2579 2580 case FFS_ADJ_NBFREE: 2581 #ifdef DEBUG 2582 if (fsckcmds) { 2583 printf("%s: adjust number of free blocks by %+jd\n", 2584 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2585 } 2586 #endif /* DEBUG */ 2587 fs->fs_cstotal.cs_nbfree += cmd.value; 2588 break; 2589 2590 case FFS_ADJ_NIFREE: 2591 #ifdef DEBUG 2592 if (fsckcmds) { 2593 printf("%s: adjust number of free inodes by %+jd\n", 2594 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2595 } 2596 #endif /* DEBUG */ 2597 fs->fs_cstotal.cs_nifree += cmd.value; 2598 break; 2599 2600 case FFS_ADJ_NFFREE: 2601 #ifdef DEBUG 2602 if (fsckcmds) { 2603 printf("%s: adjust number of free frags by %+jd\n", 2604 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2605 } 2606 #endif /* DEBUG */ 2607 fs->fs_cstotal.cs_nffree += cmd.value; 2608 break; 2609 2610 case FFS_ADJ_NUMCLUSTERS: 2611 #ifdef DEBUG 2612 if (fsckcmds) { 2613 printf("%s: adjust number of free clusters by %+jd\n", 2614 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2615 } 2616 #endif /* DEBUG */ 2617 fs->fs_cstotal.cs_numclusters += cmd.value; 2618 break; 2619 2620 case FFS_SET_CWD: 2621 #ifdef DEBUG 2622 if (fsckcmds) { 2623 printf("%s: set current directory to inode %jd\n", 2624 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2625 } 2626 #endif /* DEBUG */ 2627 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_SHARED, &vp))) 2628 break; 2629 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 2630 AUDIT_ARG_VNODE1(vp); 2631 if ((error = change_dir(vp, td)) != 0) { 2632 vput(vp); 2633 VFS_UNLOCK_GIANT(vfslocked); 2634 break; 2635 } 2636 VOP_UNLOCK(vp, 0); 2637 VFS_UNLOCK_GIANT(vfslocked); 2638 fdp = td->td_proc->p_fd; 2639 FILEDESC_XLOCK(fdp); 2640 vpold = fdp->fd_cdir; 2641 fdp->fd_cdir = vp; 2642 FILEDESC_XUNLOCK(fdp); 2643 vfslocked = VFS_LOCK_GIANT(vpold->v_mount); 2644 vrele(vpold); 2645 VFS_UNLOCK_GIANT(vfslocked); 2646 break; 2647 2648 case FFS_SET_DOTDOT: 2649 #ifdef DEBUG 2650 if (fsckcmds) { 2651 printf("%s: change .. in cwd from %jd to %jd\n", 2652 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2653 (intmax_t)cmd.size); 2654 } 2655 #endif /* DEBUG */ 2656 /* 2657 * First we have to get and lock the parent directory 2658 * to which ".." points. 2659 */ 2660 error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &fdvp); 2661 if (error) 2662 break; 2663 /* 2664 * Now we get and lock the child directory containing "..". 2665 */ 2666 FILEDESC_SLOCK(td->td_proc->p_fd); 2667 dvp = td->td_proc->p_fd->fd_cdir; 2668 FILEDESC_SUNLOCK(td->td_proc->p_fd); 2669 if ((error = vget(dvp, LK_EXCLUSIVE, td)) != 0) { 2670 vput(fdvp); 2671 break; 2672 } 2673 dp = VTOI(dvp); 2674 dp->i_offset = 12; /* XXX mastertemplate.dot_reclen */ 2675 error = ufs_dirrewrite(dp, VTOI(fdvp), (ino_t)cmd.size, 2676 DT_DIR, 0); 2677 cache_purge(fdvp); 2678 cache_purge(dvp); 2679 vput(dvp); 2680 vput(fdvp); 2681 break; 2682 2683 case FFS_UNLINK: 2684 #ifdef DEBUG 2685 if (fsckcmds) { 2686 char buf[32]; 2687 2688 if (copyinstr((char *)(intptr_t)cmd.value, buf,32,NULL)) 2689 strncpy(buf, "Name_too_long", 32); 2690 printf("%s: unlink %s (inode %jd)\n", 2691 mp->mnt_stat.f_mntonname, buf, (intmax_t)cmd.size); 2692 } 2693 #endif /* DEBUG */ 2694 /* 2695 * kern_unlinkat will do its own start/finish writes and 2696 * they do not nest, so drop ours here. Setting mp == NULL 2697 * indicates that vn_finished_write is not needed down below. 2698 */ 2699 vn_finished_write(mp); 2700 mp = NULL; 2701 error = kern_unlinkat(td, AT_FDCWD, (char *)(intptr_t)cmd.value, 2702 UIO_USERSPACE, (ino_t)cmd.size); 2703 break; 2704 2705 default: 2706 #ifdef DEBUG 2707 if (fsckcmds) { 2708 printf("Invalid request %d from fsck\n", 2709 oidp->oid_number); 2710 } 2711 #endif /* DEBUG */ 2712 error = EINVAL; 2713 break; 2714 2715 } 2716 fdrop(fp, td); 2717 vn_finished_write(mp); 2718 return (error); 2719 } 2720