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/file.h> 73 #include <sys/filedesc.h> 74 #include <sys/proc.h> 75 #include <sys/vnode.h> 76 #include <sys/mount.h> 77 #include <sys/kernel.h> 78 #include <sys/sysctl.h> 79 #include <sys/syslog.h> 80 81 #include <ufs/ufs/extattr.h> 82 #include <ufs/ufs/quota.h> 83 #include <ufs/ufs/inode.h> 84 #include <ufs/ufs/ufs_extern.h> 85 #include <ufs/ufs/ufsmount.h> 86 87 #include <ufs/ffs/fs.h> 88 #include <ufs/ffs/ffs_extern.h> 89 90 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, int cg, ufs2_daddr_t bpref, 91 int size); 92 93 static ufs2_daddr_t ffs_alloccg(struct inode *, int, ufs2_daddr_t, int); 94 static ufs2_daddr_t 95 ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t); 96 #ifdef DIAGNOSTIC 97 static int ffs_checkblk(struct inode *, ufs2_daddr_t, long); 98 #endif 99 static ufs2_daddr_t ffs_clusteralloc(struct inode *, int, ufs2_daddr_t, int); 100 static void ffs_clusteracct(struct ufsmount *, struct fs *, struct cg *, 101 ufs1_daddr_t, int); 102 static ino_t ffs_dirpref(struct inode *); 103 static ufs2_daddr_t ffs_fragextend(struct inode *, int, ufs2_daddr_t, int, int); 104 static void ffs_fserr(struct fs *, ino_t, char *); 105 static ufs2_daddr_t ffs_hashalloc 106 (struct inode *, int, ufs2_daddr_t, int, allocfcn_t *); 107 static ufs2_daddr_t ffs_nodealloccg(struct inode *, int, ufs2_daddr_t, int); 108 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int); 109 static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *); 110 static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *); 111 112 /* 113 * Allocate a block in the filesystem. 114 * 115 * The size of the requested block is given, which must be some 116 * multiple of fs_fsize and <= fs_bsize. 117 * A preference may be optionally specified. If a preference is given 118 * the following hierarchy is used to allocate a block: 119 * 1) allocate the requested block. 120 * 2) allocate a rotationally optimal block in the same cylinder. 121 * 3) allocate a block in the same cylinder group. 122 * 4) quadradically rehash into other cylinder groups, until an 123 * available block is located. 124 * If no block preference is given the following heirarchy is used 125 * to allocate a block: 126 * 1) allocate a block in the cylinder group that contains the 127 * inode for the file. 128 * 2) quadradically rehash into other cylinder groups, until an 129 * available block is located. 130 */ 131 int 132 ffs_alloc(ip, lbn, bpref, size, cred, bnp) 133 struct inode *ip; 134 ufs2_daddr_t lbn, bpref; 135 int size; 136 struct ucred *cred; 137 ufs2_daddr_t *bnp; 138 { 139 struct fs *fs; 140 struct ufsmount *ump; 141 ufs2_daddr_t bno; 142 int cg, reclaimed; 143 #ifdef QUOTA 144 int error; 145 #endif 146 147 *bnp = 0; 148 fs = ip->i_fs; 149 ump = ip->i_ump; 150 mtx_assert(UFS_MTX(ump), MA_OWNED); 151 #ifdef DIAGNOSTIC 152 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 153 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n", 154 devtoname(ip->i_dev), (long)fs->fs_bsize, size, 155 fs->fs_fsmnt); 156 panic("ffs_alloc: bad size"); 157 } 158 if (cred == NOCRED) 159 panic("ffs_alloc: missing credential"); 160 #endif /* DIAGNOSTIC */ 161 reclaimed = 0; 162 retry: 163 #ifdef QUOTA 164 UFS_UNLOCK(ump); 165 error = chkdq(ip, btodb(size), cred, 0); 166 if (error) 167 return (error); 168 UFS_LOCK(ump); 169 #endif 170 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) 171 goto nospace; 172 if (suser_cred(cred, SUSER_ALLOWJAIL) && 173 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) 174 goto nospace; 175 if (bpref >= fs->fs_size) 176 bpref = 0; 177 if (bpref == 0) 178 cg = ino_to_cg(fs, ip->i_number); 179 else 180 cg = dtog(fs, bpref); 181 bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg); 182 if (bno > 0) { 183 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size)); 184 ip->i_flag |= IN_CHANGE | IN_UPDATE; 185 *bnp = bno; 186 return (0); 187 } 188 #ifdef QUOTA 189 UFS_UNLOCK(ump); 190 /* 191 * Restore user's disk quota because allocation failed. 192 */ 193 (void) chkdq(ip, -btodb(size), cred, FORCE); 194 UFS_LOCK(ump); 195 #endif 196 nospace: 197 if (fs->fs_pendingblocks > 0 && reclaimed == 0) { 198 reclaimed = 1; 199 softdep_request_cleanup(fs, ITOV(ip)); 200 goto retry; 201 } 202 UFS_UNLOCK(ump); 203 ffs_fserr(fs, ip->i_number, "filesystem full"); 204 uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); 205 return (ENOSPC); 206 } 207 208 /* 209 * Reallocate a fragment to a bigger size 210 * 211 * The number and size of the old block is given, and a preference 212 * and new size is also specified. The allocator attempts to extend 213 * the original block. Failing that, the regular block allocator is 214 * invoked to get an appropriate block. 215 */ 216 int 217 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, cred, bpp) 218 struct inode *ip; 219 ufs2_daddr_t lbprev; 220 ufs2_daddr_t bprev; 221 ufs2_daddr_t bpref; 222 int osize, nsize; 223 struct ucred *cred; 224 struct buf **bpp; 225 { 226 struct vnode *vp; 227 struct fs *fs; 228 struct buf *bp; 229 struct ufsmount *ump; 230 int cg, request, error, reclaimed; 231 ufs2_daddr_t bno; 232 233 *bpp = 0; 234 vp = ITOV(ip); 235 fs = ip->i_fs; 236 bp = NULL; 237 ump = ip->i_ump; 238 mtx_assert(UFS_MTX(ump), MA_OWNED); 239 #ifdef DIAGNOSTIC 240 if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) 241 panic("ffs_realloccg: allocation on suspended filesystem"); 242 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || 243 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { 244 printf( 245 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", 246 devtoname(ip->i_dev), (long)fs->fs_bsize, osize, 247 nsize, fs->fs_fsmnt); 248 panic("ffs_realloccg: bad size"); 249 } 250 if (cred == NOCRED) 251 panic("ffs_realloccg: missing credential"); 252 #endif /* DIAGNOSTIC */ 253 reclaimed = 0; 254 retry: 255 if (suser_cred(cred, SUSER_ALLOWJAIL) && 256 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) { 257 goto nospace; 258 } 259 if (bprev == 0) { 260 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n", 261 devtoname(ip->i_dev), (long)fs->fs_bsize, (intmax_t)bprev, 262 fs->fs_fsmnt); 263 panic("ffs_realloccg: bad bprev"); 264 } 265 UFS_UNLOCK(ump); 266 /* 267 * Allocate the extra space in the buffer. 268 */ 269 error = bread(vp, lbprev, osize, NOCRED, &bp); 270 if (error) { 271 brelse(bp); 272 return (error); 273 } 274 275 if (bp->b_blkno == bp->b_lblkno) { 276 if (lbprev >= NDADDR) 277 panic("ffs_realloccg: lbprev out of range"); 278 bp->b_blkno = fsbtodb(fs, bprev); 279 } 280 281 #ifdef QUOTA 282 error = chkdq(ip, btodb(nsize - osize), cred, 0); 283 if (error) { 284 brelse(bp); 285 return (error); 286 } 287 #endif 288 /* 289 * Check for extension in the existing location. 290 */ 291 cg = dtog(fs, bprev); 292 UFS_LOCK(ump); 293 bno = ffs_fragextend(ip, cg, bprev, osize, nsize); 294 if (bno) { 295 if (bp->b_blkno != fsbtodb(fs, bno)) 296 panic("ffs_realloccg: bad blockno"); 297 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(nsize - osize)); 298 ip->i_flag |= IN_CHANGE | IN_UPDATE; 299 allocbuf(bp, nsize); 300 bp->b_flags |= B_DONE; 301 if ((bp->b_flags & (B_MALLOC | B_VMIO)) != B_VMIO) 302 bzero((char *)bp->b_data + osize, nsize - osize); 303 else 304 vfs_bio_clrbuf(bp); 305 *bpp = bp; 306 return (0); 307 } 308 /* 309 * Allocate a new disk location. 310 */ 311 if (bpref >= fs->fs_size) 312 bpref = 0; 313 switch ((int)fs->fs_optim) { 314 case FS_OPTSPACE: 315 /* 316 * Allocate an exact sized fragment. Although this makes 317 * best use of space, we will waste time relocating it if 318 * the file continues to grow. If the fragmentation is 319 * less than half of the minimum free reserve, we choose 320 * to begin optimizing for time. 321 */ 322 request = nsize; 323 if (fs->fs_minfree <= 5 || 324 fs->fs_cstotal.cs_nffree > 325 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100)) 326 break; 327 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", 328 fs->fs_fsmnt); 329 fs->fs_optim = FS_OPTTIME; 330 break; 331 case FS_OPTTIME: 332 /* 333 * At this point we have discovered a file that is trying to 334 * grow a small fragment to a larger fragment. To save time, 335 * we allocate a full sized block, then free the unused portion. 336 * If the file continues to grow, the `ffs_fragextend' call 337 * above will be able to grow it in place without further 338 * copying. If aberrant programs cause disk fragmentation to 339 * grow within 2% of the free reserve, we choose to begin 340 * optimizing for space. 341 */ 342 request = fs->fs_bsize; 343 if (fs->fs_cstotal.cs_nffree < 344 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100) 345 break; 346 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", 347 fs->fs_fsmnt); 348 fs->fs_optim = FS_OPTSPACE; 349 break; 350 default: 351 printf("dev = %s, optim = %ld, fs = %s\n", 352 devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt); 353 panic("ffs_realloccg: bad optim"); 354 /* NOTREACHED */ 355 } 356 bno = ffs_hashalloc(ip, cg, bpref, request, ffs_alloccg); 357 if (bno > 0) { 358 bp->b_blkno = fsbtodb(fs, bno); 359 if (!DOINGSOFTDEP(vp)) 360 ffs_blkfree(ump, fs, ip->i_devvp, bprev, (long)osize, 361 ip->i_number); 362 if (nsize < request) 363 ffs_blkfree(ump, fs, ip->i_devvp, 364 bno + numfrags(fs, nsize), 365 (long)(request - nsize), ip->i_number); 366 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(nsize - osize)); 367 ip->i_flag |= IN_CHANGE | IN_UPDATE; 368 allocbuf(bp, nsize); 369 bp->b_flags |= B_DONE; 370 if ((bp->b_flags & (B_MALLOC | B_VMIO)) != B_VMIO) 371 bzero((char *)bp->b_data + osize, nsize - osize); 372 else 373 vfs_bio_clrbuf(bp); 374 *bpp = bp; 375 return (0); 376 } 377 #ifdef QUOTA 378 UFS_UNLOCK(ump); 379 /* 380 * Restore user's disk quota because allocation failed. 381 */ 382 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE); 383 UFS_LOCK(ump); 384 #endif 385 nospace: 386 /* 387 * no space available 388 */ 389 if (fs->fs_pendingblocks > 0 && reclaimed == 0) { 390 reclaimed = 1; 391 softdep_request_cleanup(fs, vp); 392 UFS_UNLOCK(ump); 393 if (bp) 394 brelse(bp); 395 UFS_LOCK(ump); 396 goto retry; 397 } 398 UFS_UNLOCK(ump); 399 if (bp) 400 brelse(bp); 401 ffs_fserr(fs, ip->i_number, "filesystem full"); 402 uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); 403 return (ENOSPC); 404 } 405 406 /* 407 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 408 * 409 * The vnode and an array of buffer pointers for a range of sequential 410 * logical blocks to be made contiguous is given. The allocator attempts 411 * to find a range of sequential blocks starting as close as possible 412 * from the end of the allocation for the logical block immediately 413 * preceding the current range. If successful, the physical block numbers 414 * in the buffer pointers and in the inode are changed to reflect the new 415 * allocation. If unsuccessful, the allocation is left unchanged. The 416 * success in doing the reallocation is returned. Note that the error 417 * return is not reflected back to the user. Rather the previous block 418 * allocation will be used. 419 */ 420 421 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem"); 422 423 static int doasyncfree = 1; 424 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, ""); 425 426 static int doreallocblks = 1; 427 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); 428 429 #ifdef DEBUG 430 static volatile int prtrealloc = 0; 431 #endif 432 433 int 434 ffs_reallocblks(ap) 435 struct vop_reallocblks_args /* { 436 struct vnode *a_vp; 437 struct cluster_save *a_buflist; 438 } */ *ap; 439 { 440 441 if (doreallocblks == 0) 442 return (ENOSPC); 443 if (VTOI(ap->a_vp)->i_ump->um_fstype == UFS1) 444 return (ffs_reallocblks_ufs1(ap)); 445 return (ffs_reallocblks_ufs2(ap)); 446 } 447 448 static int 449 ffs_reallocblks_ufs1(ap) 450 struct vop_reallocblks_args /* { 451 struct vnode *a_vp; 452 struct cluster_save *a_buflist; 453 } */ *ap; 454 { 455 struct fs *fs; 456 struct inode *ip; 457 struct vnode *vp; 458 struct buf *sbp, *ebp; 459 ufs1_daddr_t *bap, *sbap, *ebap = 0; 460 struct cluster_save *buflist; 461 struct ufsmount *ump; 462 ufs_lbn_t start_lbn, end_lbn; 463 ufs1_daddr_t soff, newblk, blkno; 464 ufs2_daddr_t pref; 465 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; 466 int i, len, start_lvl, end_lvl, ssize; 467 468 vp = ap->a_vp; 469 ip = VTOI(vp); 470 fs = ip->i_fs; 471 ump = ip->i_ump; 472 if (fs->fs_contigsumsize <= 0) 473 return (ENOSPC); 474 buflist = ap->a_buflist; 475 len = buflist->bs_nchildren; 476 start_lbn = buflist->bs_children[0]->b_lblkno; 477 end_lbn = start_lbn + len - 1; 478 #ifdef DIAGNOSTIC 479 for (i = 0; i < len; i++) 480 if (!ffs_checkblk(ip, 481 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 482 panic("ffs_reallocblks: unallocated block 1"); 483 for (i = 1; i < len; i++) 484 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 485 panic("ffs_reallocblks: non-logical cluster"); 486 blkno = buflist->bs_children[0]->b_blkno; 487 ssize = fsbtodb(fs, fs->fs_frag); 488 for (i = 1; i < len - 1; i++) 489 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) 490 panic("ffs_reallocblks: non-physical cluster %d", i); 491 #endif 492 /* 493 * If the latest allocation is in a new cylinder group, assume that 494 * the filesystem has decided to move and do not force it back to 495 * the previous cylinder group. 496 */ 497 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 498 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 499 return (ENOSPC); 500 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 501 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 502 return (ENOSPC); 503 /* 504 * Get the starting offset and block map for the first block. 505 */ 506 if (start_lvl == 0) { 507 sbap = &ip->i_din1->di_db[0]; 508 soff = start_lbn; 509 } else { 510 idp = &start_ap[start_lvl - 1]; 511 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { 512 brelse(sbp); 513 return (ENOSPC); 514 } 515 sbap = (ufs1_daddr_t *)sbp->b_data; 516 soff = idp->in_off; 517 } 518 /* 519 * If the block range spans two block maps, get the second map. 520 */ 521 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 522 ssize = len; 523 } else { 524 #ifdef DIAGNOSTIC 525 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn) 526 panic("ffs_reallocblk: start == end"); 527 #endif 528 ssize = len - (idp->in_off + 1); 529 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) 530 goto fail; 531 ebap = (ufs1_daddr_t *)ebp->b_data; 532 } 533 /* 534 * Find the preferred location for the cluster. 535 */ 536 UFS_LOCK(ump); 537 pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap); 538 /* 539 * Search the block map looking for an allocation of the desired size. 540 */ 541 if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref, 542 len, ffs_clusteralloc)) == 0) { 543 UFS_UNLOCK(ump); 544 goto fail; 545 } 546 /* 547 * We have found a new contiguous block. 548 * 549 * First we have to replace the old block pointers with the new 550 * block pointers in the inode and indirect blocks associated 551 * with the file. 552 */ 553 #ifdef DEBUG 554 if (prtrealloc) 555 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number, 556 (intmax_t)start_lbn, (intmax_t)end_lbn); 557 #endif 558 blkno = newblk; 559 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { 560 if (i == ssize) { 561 bap = ebap; 562 soff = -i; 563 } 564 #ifdef DIAGNOSTIC 565 if (!ffs_checkblk(ip, 566 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 567 panic("ffs_reallocblks: unallocated block 2"); 568 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) 569 panic("ffs_reallocblks: alloc mismatch"); 570 #endif 571 #ifdef DEBUG 572 if (prtrealloc) 573 printf(" %d,", *bap); 574 #endif 575 if (DOINGSOFTDEP(vp)) { 576 if (sbap == &ip->i_din1->di_db[0] && i < ssize) 577 softdep_setup_allocdirect(ip, start_lbn + i, 578 blkno, *bap, fs->fs_bsize, fs->fs_bsize, 579 buflist->bs_children[i]); 580 else 581 softdep_setup_allocindir_page(ip, start_lbn + i, 582 i < ssize ? sbp : ebp, soff + i, blkno, 583 *bap, buflist->bs_children[i]); 584 } 585 *bap++ = blkno; 586 } 587 /* 588 * Next we must write out the modified inode and indirect blocks. 589 * For strict correctness, the writes should be synchronous since 590 * the old block values may have been written to disk. In practise 591 * they are almost never written, but if we are concerned about 592 * strict correctness, the `doasyncfree' flag should be set to zero. 593 * 594 * The test on `doasyncfree' should be changed to test a flag 595 * that shows whether the associated buffers and inodes have 596 * been written. The flag should be set when the cluster is 597 * started and cleared whenever the buffer or inode is flushed. 598 * We can then check below to see if it is set, and do the 599 * synchronous write only when it has been cleared. 600 */ 601 if (sbap != &ip->i_din1->di_db[0]) { 602 if (doasyncfree) 603 bdwrite(sbp); 604 else 605 bwrite(sbp); 606 } else { 607 ip->i_flag |= IN_CHANGE | IN_UPDATE; 608 if (!doasyncfree) 609 ffs_update(vp, 1); 610 } 611 if (ssize < len) { 612 if (doasyncfree) 613 bdwrite(ebp); 614 else 615 bwrite(ebp); 616 } 617 /* 618 * Last, free the old blocks and assign the new blocks to the buffers. 619 */ 620 #ifdef DEBUG 621 if (prtrealloc) 622 printf("\n\tnew:"); 623 #endif 624 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { 625 if (!DOINGSOFTDEP(vp)) 626 ffs_blkfree(ump, fs, ip->i_devvp, 627 dbtofsb(fs, buflist->bs_children[i]->b_blkno), 628 fs->fs_bsize, ip->i_number); 629 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 630 #ifdef DIAGNOSTIC 631 if (!ffs_checkblk(ip, 632 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 633 panic("ffs_reallocblks: unallocated block 3"); 634 #endif 635 #ifdef DEBUG 636 if (prtrealloc) 637 printf(" %d,", blkno); 638 #endif 639 } 640 #ifdef DEBUG 641 if (prtrealloc) { 642 prtrealloc--; 643 printf("\n"); 644 } 645 #endif 646 return (0); 647 648 fail: 649 if (ssize < len) 650 brelse(ebp); 651 if (sbap != &ip->i_din1->di_db[0]) 652 brelse(sbp); 653 return (ENOSPC); 654 } 655 656 static int 657 ffs_reallocblks_ufs2(ap) 658 struct vop_reallocblks_args /* { 659 struct vnode *a_vp; 660 struct cluster_save *a_buflist; 661 } */ *ap; 662 { 663 struct fs *fs; 664 struct inode *ip; 665 struct vnode *vp; 666 struct buf *sbp, *ebp; 667 ufs2_daddr_t *bap, *sbap, *ebap = 0; 668 struct cluster_save *buflist; 669 struct ufsmount *ump; 670 ufs_lbn_t start_lbn, end_lbn; 671 ufs2_daddr_t soff, newblk, blkno, pref; 672 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; 673 int i, len, start_lvl, end_lvl, ssize; 674 675 vp = ap->a_vp; 676 ip = VTOI(vp); 677 fs = ip->i_fs; 678 ump = ip->i_ump; 679 if (fs->fs_contigsumsize <= 0) 680 return (ENOSPC); 681 buflist = ap->a_buflist; 682 len = buflist->bs_nchildren; 683 start_lbn = buflist->bs_children[0]->b_lblkno; 684 end_lbn = start_lbn + len - 1; 685 #ifdef DIAGNOSTIC 686 for (i = 0; i < len; i++) 687 if (!ffs_checkblk(ip, 688 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 689 panic("ffs_reallocblks: unallocated block 1"); 690 for (i = 1; i < len; i++) 691 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 692 panic("ffs_reallocblks: non-logical cluster"); 693 blkno = buflist->bs_children[0]->b_blkno; 694 ssize = fsbtodb(fs, fs->fs_frag); 695 for (i = 1; i < len - 1; i++) 696 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) 697 panic("ffs_reallocblks: non-physical cluster %d", i); 698 #endif 699 /* 700 * If the latest allocation is in a new cylinder group, assume that 701 * the filesystem has decided to move and do not force it back to 702 * the previous cylinder group. 703 */ 704 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 705 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 706 return (ENOSPC); 707 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 708 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 709 return (ENOSPC); 710 /* 711 * Get the starting offset and block map for the first block. 712 */ 713 if (start_lvl == 0) { 714 sbap = &ip->i_din2->di_db[0]; 715 soff = start_lbn; 716 } else { 717 idp = &start_ap[start_lvl - 1]; 718 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { 719 brelse(sbp); 720 return (ENOSPC); 721 } 722 sbap = (ufs2_daddr_t *)sbp->b_data; 723 soff = idp->in_off; 724 } 725 /* 726 * If the block range spans two block maps, get the second map. 727 */ 728 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 729 ssize = len; 730 } else { 731 #ifdef DIAGNOSTIC 732 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn) 733 panic("ffs_reallocblk: start == end"); 734 #endif 735 ssize = len - (idp->in_off + 1); 736 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) 737 goto fail; 738 ebap = (ufs2_daddr_t *)ebp->b_data; 739 } 740 /* 741 * Find the preferred location for the cluster. 742 */ 743 UFS_LOCK(ump); 744 pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); 745 /* 746 * Search the block map looking for an allocation of the desired size. 747 */ 748 if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref, 749 len, ffs_clusteralloc)) == 0) { 750 UFS_UNLOCK(ump); 751 goto fail; 752 } 753 /* 754 * We have found a new contiguous block. 755 * 756 * First we have to replace the old block pointers with the new 757 * block pointers in the inode and indirect blocks associated 758 * with the file. 759 */ 760 #ifdef DEBUG 761 if (prtrealloc) 762 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number, 763 (intmax_t)start_lbn, (intmax_t)end_lbn); 764 #endif 765 blkno = newblk; 766 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { 767 if (i == ssize) { 768 bap = ebap; 769 soff = -i; 770 } 771 #ifdef DIAGNOSTIC 772 if (!ffs_checkblk(ip, 773 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 774 panic("ffs_reallocblks: unallocated block 2"); 775 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) 776 panic("ffs_reallocblks: alloc mismatch"); 777 #endif 778 #ifdef DEBUG 779 if (prtrealloc) 780 printf(" %jd,", (intmax_t)*bap); 781 #endif 782 if (DOINGSOFTDEP(vp)) { 783 if (sbap == &ip->i_din2->di_db[0] && i < ssize) 784 softdep_setup_allocdirect(ip, start_lbn + i, 785 blkno, *bap, fs->fs_bsize, fs->fs_bsize, 786 buflist->bs_children[i]); 787 else 788 softdep_setup_allocindir_page(ip, start_lbn + i, 789 i < ssize ? sbp : ebp, soff + i, blkno, 790 *bap, buflist->bs_children[i]); 791 } 792 *bap++ = blkno; 793 } 794 /* 795 * Next we must write out the modified inode and indirect blocks. 796 * For strict correctness, the writes should be synchronous since 797 * the old block values may have been written to disk. In practise 798 * they are almost never written, but if we are concerned about 799 * strict correctness, the `doasyncfree' flag should be set to zero. 800 * 801 * The test on `doasyncfree' should be changed to test a flag 802 * that shows whether the associated buffers and inodes have 803 * been written. The flag should be set when the cluster is 804 * started and cleared whenever the buffer or inode is flushed. 805 * We can then check below to see if it is set, and do the 806 * synchronous write only when it has been cleared. 807 */ 808 if (sbap != &ip->i_din2->di_db[0]) { 809 if (doasyncfree) 810 bdwrite(sbp); 811 else 812 bwrite(sbp); 813 } else { 814 ip->i_flag |= IN_CHANGE | IN_UPDATE; 815 if (!doasyncfree) 816 ffs_update(vp, 1); 817 } 818 if (ssize < len) { 819 if (doasyncfree) 820 bdwrite(ebp); 821 else 822 bwrite(ebp); 823 } 824 /* 825 * Last, free the old blocks and assign the new blocks to the buffers. 826 */ 827 #ifdef DEBUG 828 if (prtrealloc) 829 printf("\n\tnew:"); 830 #endif 831 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { 832 if (!DOINGSOFTDEP(vp)) 833 ffs_blkfree(ump, fs, ip->i_devvp, 834 dbtofsb(fs, buflist->bs_children[i]->b_blkno), 835 fs->fs_bsize, ip->i_number); 836 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 837 #ifdef DIAGNOSTIC 838 if (!ffs_checkblk(ip, 839 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 840 panic("ffs_reallocblks: unallocated block 3"); 841 #endif 842 #ifdef DEBUG 843 if (prtrealloc) 844 printf(" %jd,", (intmax_t)blkno); 845 #endif 846 } 847 #ifdef DEBUG 848 if (prtrealloc) { 849 prtrealloc--; 850 printf("\n"); 851 } 852 #endif 853 return (0); 854 855 fail: 856 if (ssize < len) 857 brelse(ebp); 858 if (sbap != &ip->i_din2->di_db[0]) 859 brelse(sbp); 860 return (ENOSPC); 861 } 862 863 /* 864 * Allocate an inode in the filesystem. 865 * 866 * If allocating a directory, use ffs_dirpref to select the inode. 867 * If allocating in a directory, the following hierarchy is followed: 868 * 1) allocate the preferred inode. 869 * 2) allocate an inode in the same cylinder group. 870 * 3) quadradically rehash into other cylinder groups, until an 871 * available inode is located. 872 * If no inode preference is given the following heirarchy is used 873 * to allocate an inode: 874 * 1) allocate an inode in cylinder group 0. 875 * 2) quadradically rehash into other cylinder groups, until an 876 * available inode is located. 877 */ 878 int 879 ffs_valloc(pvp, mode, cred, vpp) 880 struct vnode *pvp; 881 int mode; 882 struct ucred *cred; 883 struct vnode **vpp; 884 { 885 struct inode *pip; 886 struct fs *fs; 887 struct inode *ip; 888 struct timespec ts; 889 struct ufsmount *ump; 890 ino_t ino, ipref; 891 int cg, error; 892 893 *vpp = NULL; 894 pip = VTOI(pvp); 895 fs = pip->i_fs; 896 ump = pip->i_ump; 897 898 UFS_LOCK(ump); 899 if (fs->fs_cstotal.cs_nifree == 0) 900 goto noinodes; 901 902 if ((mode & IFMT) == IFDIR) 903 ipref = ffs_dirpref(pip); 904 else 905 ipref = pip->i_number; 906 if (ipref >= fs->fs_ncg * fs->fs_ipg) 907 ipref = 0; 908 cg = ino_to_cg(fs, ipref); 909 /* 910 * Track number of dirs created one after another 911 * in a same cg without intervening by files. 912 */ 913 if ((mode & IFMT) == IFDIR) { 914 if (fs->fs_contigdirs[cg] < 255) 915 fs->fs_contigdirs[cg]++; 916 } else { 917 if (fs->fs_contigdirs[cg] > 0) 918 fs->fs_contigdirs[cg]--; 919 } 920 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 921 (allocfcn_t *)ffs_nodealloccg); 922 if (ino == 0) 923 goto noinodes; 924 error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp); 925 if (error) { 926 ffs_vfree(pvp, ino, mode); 927 return (error); 928 } 929 ip = VTOI(*vpp); 930 if (ip->i_mode) { 931 printf("mode = 0%o, inum = %lu, fs = %s\n", 932 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt); 933 panic("ffs_valloc: dup alloc"); 934 } 935 if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */ 936 printf("free inode %s/%lu had %ld blocks\n", 937 fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks)); 938 DIP_SET(ip, i_blocks, 0); 939 } 940 ip->i_flags = 0; 941 DIP_SET(ip, i_flags, 0); 942 /* 943 * Set up a new generation number for this inode. 944 */ 945 if (ip->i_gen == 0 || ++ip->i_gen == 0) 946 ip->i_gen = arc4random() / 2 + 1; 947 DIP_SET(ip, i_gen, ip->i_gen); 948 if (fs->fs_magic == FS_UFS2_MAGIC) { 949 vfs_timestamp(&ts); 950 ip->i_din2->di_birthtime = ts.tv_sec; 951 ip->i_din2->di_birthnsec = ts.tv_nsec; 952 } 953 ip->i_flag = 0; 954 vnode_destroy_vobject(*vpp); 955 (*vpp)->v_type = VNON; 956 if (fs->fs_magic == FS_UFS2_MAGIC) 957 (*vpp)->v_op = &ffs_vnodeops2; 958 else 959 (*vpp)->v_op = &ffs_vnodeops1; 960 return (0); 961 noinodes: 962 UFS_UNLOCK(ump); 963 ffs_fserr(fs, pip->i_number, "out of inodes"); 964 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); 965 return (ENOSPC); 966 } 967 968 /* 969 * Find a cylinder group to place a directory. 970 * 971 * The policy implemented by this algorithm is to allocate a 972 * directory inode in the same cylinder group as its parent 973 * directory, but also to reserve space for its files inodes 974 * and data. Restrict the number of directories which may be 975 * allocated one after another in the same cylinder group 976 * without intervening allocation of files. 977 * 978 * If we allocate a first level directory then force allocation 979 * in another cylinder group. 980 */ 981 static ino_t 982 ffs_dirpref(pip) 983 struct inode *pip; 984 { 985 struct fs *fs; 986 int cg, prefcg, dirsize, cgsize; 987 int avgifree, avgbfree, avgndir, curdirsize; 988 int minifree, minbfree, maxndir; 989 int mincg, minndir; 990 int maxcontigdirs; 991 992 mtx_assert(UFS_MTX(pip->i_ump), MA_OWNED); 993 fs = pip->i_fs; 994 995 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 996 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 997 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; 998 999 /* 1000 * Force allocation in another cg if creating a first level dir. 1001 */ 1002 ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref"); 1003 if (ITOV(pip)->v_vflag & VV_ROOT) { 1004 prefcg = arc4random() % fs->fs_ncg; 1005 mincg = prefcg; 1006 minndir = fs->fs_ipg; 1007 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1008 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 1009 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 1010 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1011 mincg = cg; 1012 minndir = fs->fs_cs(fs, cg).cs_ndir; 1013 } 1014 for (cg = 0; cg < prefcg; cg++) 1015 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 1016 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 1017 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1018 mincg = cg; 1019 minndir = fs->fs_cs(fs, cg).cs_ndir; 1020 } 1021 return ((ino_t)(fs->fs_ipg * mincg)); 1022 } 1023 1024 /* 1025 * Count various limits which used for 1026 * optimal allocation of a directory inode. 1027 */ 1028 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); 1029 minifree = avgifree - avgifree / 4; 1030 if (minifree < 1) 1031 minifree = 1; 1032 minbfree = avgbfree - avgbfree / 4; 1033 if (minbfree < 1) 1034 minbfree = 1; 1035 cgsize = fs->fs_fsize * fs->fs_fpg; 1036 dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir; 1037 curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0; 1038 if (dirsize < curdirsize) 1039 dirsize = curdirsize; 1040 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255); 1041 if (fs->fs_avgfpdir > 0) 1042 maxcontigdirs = min(maxcontigdirs, 1043 fs->fs_ipg / fs->fs_avgfpdir); 1044 if (maxcontigdirs == 0) 1045 maxcontigdirs = 1; 1046 1047 /* 1048 * Limit number of dirs in one cg and reserve space for 1049 * regular files, but only if we have no deficit in 1050 * inodes or space. 1051 */ 1052 prefcg = ino_to_cg(fs, pip->i_number); 1053 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1054 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 1055 fs->fs_cs(fs, cg).cs_nifree >= minifree && 1056 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 1057 if (fs->fs_contigdirs[cg] < maxcontigdirs) 1058 return ((ino_t)(fs->fs_ipg * cg)); 1059 } 1060 for (cg = 0; cg < prefcg; cg++) 1061 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 1062 fs->fs_cs(fs, cg).cs_nifree >= minifree && 1063 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 1064 if (fs->fs_contigdirs[cg] < maxcontigdirs) 1065 return ((ino_t)(fs->fs_ipg * cg)); 1066 } 1067 /* 1068 * This is a backstop when we have deficit in space. 1069 */ 1070 for (cg = prefcg; cg < fs->fs_ncg; cg++) 1071 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 1072 return ((ino_t)(fs->fs_ipg * cg)); 1073 for (cg = 0; cg < prefcg; cg++) 1074 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 1075 break; 1076 return ((ino_t)(fs->fs_ipg * cg)); 1077 } 1078 1079 /* 1080 * Select the desired position for the next block in a file. The file is 1081 * logically divided into sections. The first section is composed of the 1082 * direct blocks. Each additional section contains fs_maxbpg blocks. 1083 * 1084 * If no blocks have been allocated in the first section, the policy is to 1085 * request a block in the same cylinder group as the inode that describes 1086 * the file. If no blocks have been allocated in any other section, the 1087 * policy is to place the section in a cylinder group with a greater than 1088 * average number of free blocks. An appropriate cylinder group is found 1089 * by using a rotor that sweeps the cylinder groups. When a new group of 1090 * blocks is needed, the sweep begins in the cylinder group following the 1091 * cylinder group from which the previous allocation was made. The sweep 1092 * continues until a cylinder group with greater than the average number 1093 * of free blocks is found. If the allocation is for the first block in an 1094 * indirect block, the information on the previous allocation is unavailable; 1095 * here a best guess is made based upon the logical block number being 1096 * allocated. 1097 * 1098 * If a section is already partially allocated, the policy is to 1099 * contiguously allocate fs_maxcontig blocks. The end of one of these 1100 * contiguous blocks and the beginning of the next is laid out 1101 * contiguously if possible. 1102 */ 1103 ufs2_daddr_t 1104 ffs_blkpref_ufs1(ip, lbn, indx, bap) 1105 struct inode *ip; 1106 ufs_lbn_t lbn; 1107 int indx; 1108 ufs1_daddr_t *bap; 1109 { 1110 struct fs *fs; 1111 int cg; 1112 int avgbfree, startcg; 1113 1114 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); 1115 fs = ip->i_fs; 1116 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 1117 if (lbn < NDADDR + NINDIR(fs)) { 1118 cg = ino_to_cg(fs, ip->i_number); 1119 return (cgbase(fs, cg) + fs->fs_frag); 1120 } 1121 /* 1122 * Find a cylinder with greater than average number of 1123 * unused data blocks. 1124 */ 1125 if (indx == 0 || bap[indx - 1] == 0) 1126 startcg = 1127 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 1128 else 1129 startcg = dtog(fs, bap[indx - 1]) + 1; 1130 startcg %= fs->fs_ncg; 1131 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1132 for (cg = startcg; cg < fs->fs_ncg; cg++) 1133 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1134 fs->fs_cgrotor = cg; 1135 return (cgbase(fs, cg) + fs->fs_frag); 1136 } 1137 for (cg = 0; cg <= startcg; cg++) 1138 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1139 fs->fs_cgrotor = cg; 1140 return (cgbase(fs, cg) + fs->fs_frag); 1141 } 1142 return (0); 1143 } 1144 /* 1145 * We just always try to lay things out contiguously. 1146 */ 1147 return (bap[indx - 1] + fs->fs_frag); 1148 } 1149 1150 /* 1151 * Same as above, but for UFS2 1152 */ 1153 ufs2_daddr_t 1154 ffs_blkpref_ufs2(ip, lbn, indx, bap) 1155 struct inode *ip; 1156 ufs_lbn_t lbn; 1157 int indx; 1158 ufs2_daddr_t *bap; 1159 { 1160 struct fs *fs; 1161 int cg; 1162 int avgbfree, startcg; 1163 1164 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); 1165 fs = ip->i_fs; 1166 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 1167 if (lbn < NDADDR + NINDIR(fs)) { 1168 cg = ino_to_cg(fs, ip->i_number); 1169 return (cgbase(fs, cg) + fs->fs_frag); 1170 } 1171 /* 1172 * Find a cylinder with greater than average number of 1173 * unused data blocks. 1174 */ 1175 if (indx == 0 || bap[indx - 1] == 0) 1176 startcg = 1177 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 1178 else 1179 startcg = dtog(fs, bap[indx - 1]) + 1; 1180 startcg %= fs->fs_ncg; 1181 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 1182 for (cg = startcg; cg < fs->fs_ncg; cg++) 1183 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1184 fs->fs_cgrotor = cg; 1185 return (cgbase(fs, cg) + fs->fs_frag); 1186 } 1187 for (cg = 0; cg <= startcg; cg++) 1188 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 1189 fs->fs_cgrotor = cg; 1190 return (cgbase(fs, cg) + fs->fs_frag); 1191 } 1192 return (0); 1193 } 1194 /* 1195 * We just always try to lay things out contiguously. 1196 */ 1197 return (bap[indx - 1] + fs->fs_frag); 1198 } 1199 1200 /* 1201 * Implement the cylinder overflow algorithm. 1202 * 1203 * The policy implemented by this algorithm is: 1204 * 1) allocate the block in its requested cylinder group. 1205 * 2) quadradically rehash on the cylinder group number. 1206 * 3) brute force search for a free block. 1207 * 1208 * Must be called with the UFS lock held. Will release the lock on success 1209 * and return with it held on failure. 1210 */ 1211 /*VARARGS5*/ 1212 static ufs2_daddr_t 1213 ffs_hashalloc(ip, cg, pref, size, allocator) 1214 struct inode *ip; 1215 int cg; 1216 ufs2_daddr_t pref; 1217 int size; /* size for data blocks, mode for inodes */ 1218 allocfcn_t *allocator; 1219 { 1220 struct fs *fs; 1221 ufs2_daddr_t result; 1222 int i, icg = cg; 1223 1224 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); 1225 #ifdef DIAGNOSTIC 1226 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) 1227 panic("ffs_hashalloc: allocation on suspended filesystem"); 1228 #endif 1229 fs = ip->i_fs; 1230 /* 1231 * 1: preferred cylinder group 1232 */ 1233 result = (*allocator)(ip, cg, pref, size); 1234 if (result) 1235 return (result); 1236 /* 1237 * 2: quadratic rehash 1238 */ 1239 for (i = 1; i < fs->fs_ncg; i *= 2) { 1240 cg += i; 1241 if (cg >= fs->fs_ncg) 1242 cg -= fs->fs_ncg; 1243 result = (*allocator)(ip, cg, 0, size); 1244 if (result) 1245 return (result); 1246 } 1247 /* 1248 * 3: brute force search 1249 * Note that we start at i == 2, since 0 was checked initially, 1250 * and 1 is always checked in the quadratic rehash. 1251 */ 1252 cg = (icg + 2) % fs->fs_ncg; 1253 for (i = 2; i < fs->fs_ncg; i++) { 1254 result = (*allocator)(ip, cg, 0, size); 1255 if (result) 1256 return (result); 1257 cg++; 1258 if (cg == fs->fs_ncg) 1259 cg = 0; 1260 } 1261 return (0); 1262 } 1263 1264 /* 1265 * Determine whether a fragment can be extended. 1266 * 1267 * Check to see if the necessary fragments are available, and 1268 * if they are, allocate them. 1269 */ 1270 static ufs2_daddr_t 1271 ffs_fragextend(ip, cg, bprev, osize, nsize) 1272 struct inode *ip; 1273 int cg; 1274 ufs2_daddr_t bprev; 1275 int osize, nsize; 1276 { 1277 struct fs *fs; 1278 struct cg *cgp; 1279 struct buf *bp; 1280 struct ufsmount *ump; 1281 int nffree; 1282 long bno; 1283 int frags, bbase; 1284 int i, error; 1285 u_int8_t *blksfree; 1286 1287 ump = ip->i_ump; 1288 fs = ip->i_fs; 1289 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 1290 return (0); 1291 frags = numfrags(fs, nsize); 1292 bbase = fragnum(fs, bprev); 1293 if (bbase > fragnum(fs, (bprev + frags - 1))) { 1294 /* cannot extend across a block boundary */ 1295 return (0); 1296 } 1297 UFS_UNLOCK(ump); 1298 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1299 (int)fs->fs_cgsize, NOCRED, &bp); 1300 if (error) 1301 goto fail; 1302 cgp = (struct cg *)bp->b_data; 1303 if (!cg_chkmagic(cgp)) 1304 goto fail; 1305 bp->b_xflags |= BX_BKGRDWRITE; 1306 cgp->cg_old_time = cgp->cg_time = time_second; 1307 bno = dtogd(fs, bprev); 1308 blksfree = cg_blksfree(cgp); 1309 for (i = numfrags(fs, osize); i < frags; i++) 1310 if (isclr(blksfree, bno + i)) 1311 goto fail; 1312 /* 1313 * the current fragment can be extended 1314 * deduct the count on fragment being extended into 1315 * increase the count on the remaining fragment (if any) 1316 * allocate the extended piece 1317 */ 1318 for (i = frags; i < fs->fs_frag - bbase; i++) 1319 if (isclr(blksfree, bno + i)) 1320 break; 1321 cgp->cg_frsum[i - numfrags(fs, osize)]--; 1322 if (i != frags) 1323 cgp->cg_frsum[i - frags]++; 1324 for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) { 1325 clrbit(blksfree, bno + i); 1326 cgp->cg_cs.cs_nffree--; 1327 nffree++; 1328 } 1329 UFS_LOCK(ump); 1330 fs->fs_cstotal.cs_nffree -= nffree; 1331 fs->fs_cs(fs, cg).cs_nffree -= nffree; 1332 fs->fs_fmod = 1; 1333 ACTIVECLEAR(fs, cg); 1334 UFS_UNLOCK(ump); 1335 if (DOINGSOFTDEP(ITOV(ip))) 1336 softdep_setup_blkmapdep(bp, fs, bprev); 1337 bdwrite(bp); 1338 return (bprev); 1339 1340 fail: 1341 brelse(bp); 1342 UFS_LOCK(ump); 1343 return (0); 1344 1345 } 1346 1347 /* 1348 * Determine whether a block can be allocated. 1349 * 1350 * Check to see if a block of the appropriate size is available, 1351 * and if it is, allocate it. 1352 */ 1353 static ufs2_daddr_t 1354 ffs_alloccg(ip, cg, bpref, size) 1355 struct inode *ip; 1356 int cg; 1357 ufs2_daddr_t bpref; 1358 int size; 1359 { 1360 struct fs *fs; 1361 struct cg *cgp; 1362 struct buf *bp; 1363 struct ufsmount *ump; 1364 ufs1_daddr_t bno; 1365 ufs2_daddr_t blkno; 1366 int i, allocsiz, error, frags; 1367 u_int8_t *blksfree; 1368 1369 ump = ip->i_ump; 1370 fs = ip->i_fs; 1371 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 1372 return (0); 1373 UFS_UNLOCK(ump); 1374 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1375 (int)fs->fs_cgsize, NOCRED, &bp); 1376 if (error) 1377 goto fail; 1378 cgp = (struct cg *)bp->b_data; 1379 if (!cg_chkmagic(cgp) || 1380 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) 1381 goto fail; 1382 bp->b_xflags |= BX_BKGRDWRITE; 1383 cgp->cg_old_time = cgp->cg_time = time_second; 1384 if (size == fs->fs_bsize) { 1385 UFS_LOCK(ump); 1386 blkno = ffs_alloccgblk(ip, bp, bpref); 1387 ACTIVECLEAR(fs, cg); 1388 UFS_UNLOCK(ump); 1389 bdwrite(bp); 1390 return (blkno); 1391 } 1392 /* 1393 * check to see if any fragments are already available 1394 * allocsiz is the size which will be allocated, hacking 1395 * it down to a smaller size if necessary 1396 */ 1397 blksfree = cg_blksfree(cgp); 1398 frags = numfrags(fs, size); 1399 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 1400 if (cgp->cg_frsum[allocsiz] != 0) 1401 break; 1402 if (allocsiz == fs->fs_frag) { 1403 /* 1404 * no fragments were available, so a block will be 1405 * allocated, and hacked up 1406 */ 1407 if (cgp->cg_cs.cs_nbfree == 0) 1408 goto fail; 1409 UFS_LOCK(ump); 1410 blkno = ffs_alloccgblk(ip, bp, bpref); 1411 bno = dtogd(fs, blkno); 1412 for (i = frags; i < fs->fs_frag; i++) 1413 setbit(blksfree, bno + i); 1414 i = fs->fs_frag - frags; 1415 cgp->cg_cs.cs_nffree += i; 1416 fs->fs_cstotal.cs_nffree += i; 1417 fs->fs_cs(fs, cg).cs_nffree += i; 1418 fs->fs_fmod = 1; 1419 cgp->cg_frsum[i]++; 1420 ACTIVECLEAR(fs, cg); 1421 UFS_UNLOCK(ump); 1422 bdwrite(bp); 1423 return (blkno); 1424 } 1425 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 1426 if (bno < 0) 1427 goto fail; 1428 for (i = 0; i < frags; i++) 1429 clrbit(blksfree, bno + i); 1430 cgp->cg_cs.cs_nffree -= frags; 1431 cgp->cg_frsum[allocsiz]--; 1432 if (frags != allocsiz) 1433 cgp->cg_frsum[allocsiz - frags]++; 1434 UFS_LOCK(ump); 1435 fs->fs_cstotal.cs_nffree -= frags; 1436 fs->fs_cs(fs, cg).cs_nffree -= frags; 1437 fs->fs_fmod = 1; 1438 blkno = cgbase(fs, cg) + bno; 1439 ACTIVECLEAR(fs, cg); 1440 UFS_UNLOCK(ump); 1441 if (DOINGSOFTDEP(ITOV(ip))) 1442 softdep_setup_blkmapdep(bp, fs, blkno); 1443 bdwrite(bp); 1444 return (blkno); 1445 1446 fail: 1447 brelse(bp); 1448 UFS_LOCK(ump); 1449 return (0); 1450 } 1451 1452 /* 1453 * Allocate a block in a cylinder group. 1454 * 1455 * This algorithm implements the following policy: 1456 * 1) allocate the requested block. 1457 * 2) allocate a rotationally optimal block in the same cylinder. 1458 * 3) allocate the next available block on the block rotor for the 1459 * specified cylinder group. 1460 * Note that this routine only allocates fs_bsize blocks; these 1461 * blocks may be fragmented by the routine that allocates them. 1462 */ 1463 static ufs2_daddr_t 1464 ffs_alloccgblk(ip, bp, bpref) 1465 struct inode *ip; 1466 struct buf *bp; 1467 ufs2_daddr_t bpref; 1468 { 1469 struct fs *fs; 1470 struct cg *cgp; 1471 struct ufsmount *ump; 1472 ufs1_daddr_t bno; 1473 ufs2_daddr_t blkno; 1474 u_int8_t *blksfree; 1475 1476 fs = ip->i_fs; 1477 ump = ip->i_ump; 1478 mtx_assert(UFS_MTX(ump), MA_OWNED); 1479 cgp = (struct cg *)bp->b_data; 1480 blksfree = cg_blksfree(cgp); 1481 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) { 1482 bpref = cgp->cg_rotor; 1483 } else { 1484 bpref = blknum(fs, bpref); 1485 bno = dtogd(fs, bpref); 1486 /* 1487 * if the requested block is available, use it 1488 */ 1489 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) 1490 goto gotit; 1491 } 1492 /* 1493 * Take the next available block in this cylinder group. 1494 */ 1495 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1496 if (bno < 0) 1497 return (0); 1498 cgp->cg_rotor = bno; 1499 gotit: 1500 blkno = fragstoblks(fs, bno); 1501 ffs_clrblock(fs, blksfree, (long)blkno); 1502 ffs_clusteracct(ump, fs, cgp, blkno, -1); 1503 cgp->cg_cs.cs_nbfree--; 1504 fs->fs_cstotal.cs_nbfree--; 1505 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; 1506 fs->fs_fmod = 1; 1507 blkno = cgbase(fs, cgp->cg_cgx) + bno; 1508 /* XXX Fixme. */ 1509 UFS_UNLOCK(ump); 1510 if (DOINGSOFTDEP(ITOV(ip))) 1511 softdep_setup_blkmapdep(bp, fs, blkno); 1512 UFS_LOCK(ump); 1513 return (blkno); 1514 } 1515 1516 /* 1517 * Determine whether a cluster can be allocated. 1518 * 1519 * We do not currently check for optimal rotational layout if there 1520 * are multiple choices in the same cylinder group. Instead we just 1521 * take the first one that we find following bpref. 1522 */ 1523 static ufs2_daddr_t 1524 ffs_clusteralloc(ip, cg, bpref, len) 1525 struct inode *ip; 1526 int cg; 1527 ufs2_daddr_t bpref; 1528 int len; 1529 { 1530 struct fs *fs; 1531 struct cg *cgp; 1532 struct buf *bp; 1533 struct ufsmount *ump; 1534 int i, run, bit, map, got; 1535 ufs2_daddr_t bno; 1536 u_char *mapp; 1537 int32_t *lp; 1538 u_int8_t *blksfree; 1539 1540 fs = ip->i_fs; 1541 ump = ip->i_ump; 1542 if (fs->fs_maxcluster[cg] < len) 1543 return (0); 1544 UFS_UNLOCK(ump); 1545 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize, 1546 NOCRED, &bp)) 1547 goto fail_lock; 1548 cgp = (struct cg *)bp->b_data; 1549 if (!cg_chkmagic(cgp)) 1550 goto fail_lock; 1551 bp->b_xflags |= BX_BKGRDWRITE; 1552 /* 1553 * Check to see if a cluster of the needed size (or bigger) is 1554 * available in this cylinder group. 1555 */ 1556 lp = &cg_clustersum(cgp)[len]; 1557 for (i = len; i <= fs->fs_contigsumsize; i++) 1558 if (*lp++ > 0) 1559 break; 1560 if (i > fs->fs_contigsumsize) { 1561 /* 1562 * This is the first time looking for a cluster in this 1563 * cylinder group. Update the cluster summary information 1564 * to reflect the true maximum sized cluster so that 1565 * future cluster allocation requests can avoid reading 1566 * the cylinder group map only to find no clusters. 1567 */ 1568 lp = &cg_clustersum(cgp)[len - 1]; 1569 for (i = len - 1; i > 0; i--) 1570 if (*lp-- > 0) 1571 break; 1572 UFS_LOCK(ump); 1573 fs->fs_maxcluster[cg] = i; 1574 goto fail; 1575 } 1576 /* 1577 * Search the cluster map to find a big enough cluster. 1578 * We take the first one that we find, even if it is larger 1579 * than we need as we prefer to get one close to the previous 1580 * block allocation. We do not search before the current 1581 * preference point as we do not want to allocate a block 1582 * that is allocated before the previous one (as we will 1583 * then have to wait for another pass of the elevator 1584 * algorithm before it will be read). We prefer to fail and 1585 * be recalled to try an allocation in the next cylinder group. 1586 */ 1587 if (dtog(fs, bpref) != cg) 1588 bpref = 0; 1589 else 1590 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref))); 1591 mapp = &cg_clustersfree(cgp)[bpref / NBBY]; 1592 map = *mapp++; 1593 bit = 1 << (bpref % NBBY); 1594 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { 1595 if ((map & bit) == 0) { 1596 run = 0; 1597 } else { 1598 run++; 1599 if (run == len) 1600 break; 1601 } 1602 if ((got & (NBBY - 1)) != (NBBY - 1)) { 1603 bit <<= 1; 1604 } else { 1605 map = *mapp++; 1606 bit = 1; 1607 } 1608 } 1609 if (got >= cgp->cg_nclusterblks) 1610 goto fail_lock; 1611 /* 1612 * Allocate the cluster that we have found. 1613 */ 1614 blksfree = cg_blksfree(cgp); 1615 for (i = 1; i <= len; i++) 1616 if (!ffs_isblock(fs, blksfree, got - run + i)) 1617 panic("ffs_clusteralloc: map mismatch"); 1618 bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1); 1619 if (dtog(fs, bno) != cg) 1620 panic("ffs_clusteralloc: allocated out of group"); 1621 len = blkstofrags(fs, len); 1622 UFS_LOCK(ump); 1623 for (i = 0; i < len; i += fs->fs_frag) 1624 if (ffs_alloccgblk(ip, bp, bno + i) != bno + i) 1625 panic("ffs_clusteralloc: lost block"); 1626 ACTIVECLEAR(fs, cg); 1627 UFS_UNLOCK(ump); 1628 bdwrite(bp); 1629 return (bno); 1630 1631 fail_lock: 1632 UFS_LOCK(ump); 1633 fail: 1634 brelse(bp); 1635 return (0); 1636 } 1637 1638 /* 1639 * Determine whether an inode can be allocated. 1640 * 1641 * Check to see if an inode is available, and if it is, 1642 * allocate it using the following policy: 1643 * 1) allocate the requested inode. 1644 * 2) allocate the next available inode after the requested 1645 * inode in the specified cylinder group. 1646 */ 1647 static ufs2_daddr_t 1648 ffs_nodealloccg(ip, cg, ipref, mode) 1649 struct inode *ip; 1650 int cg; 1651 ufs2_daddr_t ipref; 1652 int mode; 1653 { 1654 struct fs *fs; 1655 struct cg *cgp; 1656 struct buf *bp, *ibp; 1657 struct ufsmount *ump; 1658 u_int8_t *inosused; 1659 struct ufs2_dinode *dp2; 1660 int error, start, len, loc, map, i; 1661 1662 fs = ip->i_fs; 1663 ump = ip->i_ump; 1664 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1665 return (0); 1666 UFS_UNLOCK(ump); 1667 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1668 (int)fs->fs_cgsize, NOCRED, &bp); 1669 if (error) { 1670 brelse(bp); 1671 UFS_LOCK(ump); 1672 return (0); 1673 } 1674 cgp = (struct cg *)bp->b_data; 1675 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) { 1676 brelse(bp); 1677 UFS_LOCK(ump); 1678 return (0); 1679 } 1680 bp->b_xflags |= BX_BKGRDWRITE; 1681 cgp->cg_old_time = cgp->cg_time = time_second; 1682 inosused = cg_inosused(cgp); 1683 if (ipref) { 1684 ipref %= fs->fs_ipg; 1685 if (isclr(inosused, ipref)) 1686 goto gotit; 1687 } 1688 start = cgp->cg_irotor / NBBY; 1689 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); 1690 loc = skpc(0xff, len, &inosused[start]); 1691 if (loc == 0) { 1692 len = start + 1; 1693 start = 0; 1694 loc = skpc(0xff, len, &inosused[0]); 1695 if (loc == 0) { 1696 printf("cg = %d, irotor = %ld, fs = %s\n", 1697 cg, (long)cgp->cg_irotor, fs->fs_fsmnt); 1698 panic("ffs_nodealloccg: map corrupted"); 1699 /* NOTREACHED */ 1700 } 1701 } 1702 i = start + len - loc; 1703 map = inosused[i]; 1704 ipref = i * NBBY; 1705 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) { 1706 if ((map & i) == 0) { 1707 cgp->cg_irotor = ipref; 1708 goto gotit; 1709 } 1710 } 1711 printf("fs = %s\n", fs->fs_fsmnt); 1712 panic("ffs_nodealloccg: block not in map"); 1713 /* NOTREACHED */ 1714 gotit: 1715 /* 1716 * Check to see if we need to initialize more inodes. 1717 */ 1718 ibp = NULL; 1719 if (fs->fs_magic == FS_UFS2_MAGIC && 1720 ipref + INOPB(fs) > cgp->cg_initediblk && 1721 cgp->cg_initediblk < cgp->cg_niblk) { 1722 ibp = getblk(ip->i_devvp, fsbtodb(fs, 1723 ino_to_fsba(fs, cg * fs->fs_ipg + cgp->cg_initediblk)), 1724 (int)fs->fs_bsize, 0, 0, 0); 1725 bzero(ibp->b_data, (int)fs->fs_bsize); 1726 dp2 = (struct ufs2_dinode *)(ibp->b_data); 1727 for (i = 0; i < INOPB(fs); i++) { 1728 dp2->di_gen = arc4random() / 2 + 1; 1729 dp2++; 1730 } 1731 cgp->cg_initediblk += INOPB(fs); 1732 } 1733 UFS_LOCK(ump); 1734 ACTIVECLEAR(fs, cg); 1735 setbit(inosused, ipref); 1736 cgp->cg_cs.cs_nifree--; 1737 fs->fs_cstotal.cs_nifree--; 1738 fs->fs_cs(fs, cg).cs_nifree--; 1739 fs->fs_fmod = 1; 1740 if ((mode & IFMT) == IFDIR) { 1741 cgp->cg_cs.cs_ndir++; 1742 fs->fs_cstotal.cs_ndir++; 1743 fs->fs_cs(fs, cg).cs_ndir++; 1744 } 1745 UFS_UNLOCK(ump); 1746 if (DOINGSOFTDEP(ITOV(ip))) 1747 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref); 1748 bdwrite(bp); 1749 if (ibp != NULL) 1750 bawrite(ibp); 1751 return (cg * fs->fs_ipg + ipref); 1752 } 1753 1754 /* 1755 * check if a block is free 1756 */ 1757 static int 1758 ffs_isfreeblock(struct fs *fs, u_char *cp, ufs1_daddr_t h) 1759 { 1760 1761 switch ((int)fs->fs_frag) { 1762 case 8: 1763 return (cp[h] == 0); 1764 case 4: 1765 return ((cp[h >> 1] & (0x0f << ((h & 0x1) << 2))) == 0); 1766 case 2: 1767 return ((cp[h >> 2] & (0x03 << ((h & 0x3) << 1))) == 0); 1768 case 1: 1769 return ((cp[h >> 3] & (0x01 << (h & 0x7))) == 0); 1770 default: 1771 panic("ffs_isfreeblock"); 1772 } 1773 return (0); 1774 } 1775 1776 /* 1777 * Free a block or fragment. 1778 * 1779 * The specified block or fragment is placed back in the 1780 * free map. If a fragment is deallocated, a possible 1781 * block reassembly is checked. 1782 */ 1783 void 1784 ffs_blkfree(ump, fs, devvp, bno, size, inum) 1785 struct ufsmount *ump; 1786 struct fs *fs; 1787 struct vnode *devvp; 1788 ufs2_daddr_t bno; 1789 long size; 1790 ino_t inum; 1791 { 1792 struct cg *cgp; 1793 struct buf *bp; 1794 ufs1_daddr_t fragno, cgbno; 1795 ufs2_daddr_t cgblkno; 1796 int i, cg, blk, frags, bbase; 1797 u_int8_t *blksfree; 1798 struct cdev *dev; 1799 1800 cg = dtog(fs, bno); 1801 if (devvp->v_type != VCHR) { 1802 /* devvp is a snapshot */ 1803 dev = VTOI(devvp)->i_devvp->v_rdev; 1804 cgblkno = fragstoblks(fs, cgtod(fs, cg)); 1805 } else { 1806 /* devvp is a normal disk device */ 1807 dev = devvp->v_rdev; 1808 cgblkno = fsbtodb(fs, cgtod(fs, cg)); 1809 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree"); 1810 if ((devvp->v_vflag & VV_COPYONWRITE) && 1811 ffs_snapblkfree(fs, devvp, bno, size, inum)) 1812 return; 1813 } 1814 #ifdef DIAGNOSTIC 1815 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 1816 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 1817 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n", 1818 devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize, 1819 size, fs->fs_fsmnt); 1820 panic("ffs_blkfree: bad size"); 1821 } 1822 #endif 1823 if ((u_int)bno >= fs->fs_size) { 1824 printf("bad block %jd, ino %lu\n", (intmax_t)bno, 1825 (u_long)inum); 1826 ffs_fserr(fs, inum, "bad block"); 1827 return; 1828 } 1829 if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) { 1830 brelse(bp); 1831 return; 1832 } 1833 cgp = (struct cg *)bp->b_data; 1834 if (!cg_chkmagic(cgp)) { 1835 brelse(bp); 1836 return; 1837 } 1838 bp->b_xflags |= BX_BKGRDWRITE; 1839 cgp->cg_old_time = cgp->cg_time = time_second; 1840 cgbno = dtogd(fs, bno); 1841 blksfree = cg_blksfree(cgp); 1842 UFS_LOCK(ump); 1843 if (size == fs->fs_bsize) { 1844 fragno = fragstoblks(fs, cgbno); 1845 if (!ffs_isfreeblock(fs, blksfree, fragno)) { 1846 if (devvp->v_type != VCHR) { 1847 UFS_UNLOCK(ump); 1848 /* devvp is a snapshot */ 1849 brelse(bp); 1850 return; 1851 } 1852 printf("dev = %s, block = %jd, fs = %s\n", 1853 devtoname(dev), (intmax_t)bno, fs->fs_fsmnt); 1854 panic("ffs_blkfree: freeing free block"); 1855 } 1856 ffs_setblock(fs, blksfree, fragno); 1857 ffs_clusteracct(ump, fs, cgp, fragno, 1); 1858 cgp->cg_cs.cs_nbfree++; 1859 fs->fs_cstotal.cs_nbfree++; 1860 fs->fs_cs(fs, cg).cs_nbfree++; 1861 } else { 1862 bbase = cgbno - fragnum(fs, cgbno); 1863 /* 1864 * decrement the counts associated with the old frags 1865 */ 1866 blk = blkmap(fs, blksfree, bbase); 1867 ffs_fragacct(fs, blk, cgp->cg_frsum, -1); 1868 /* 1869 * deallocate the fragment 1870 */ 1871 frags = numfrags(fs, size); 1872 for (i = 0; i < frags; i++) { 1873 if (isset(blksfree, cgbno + i)) { 1874 printf("dev = %s, block = %jd, fs = %s\n", 1875 devtoname(dev), (intmax_t)(bno + i), 1876 fs->fs_fsmnt); 1877 panic("ffs_blkfree: freeing free frag"); 1878 } 1879 setbit(blksfree, cgbno + i); 1880 } 1881 cgp->cg_cs.cs_nffree += i; 1882 fs->fs_cstotal.cs_nffree += i; 1883 fs->fs_cs(fs, cg).cs_nffree += i; 1884 /* 1885 * add back in counts associated with the new frags 1886 */ 1887 blk = blkmap(fs, blksfree, bbase); 1888 ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 1889 /* 1890 * if a complete block has been reassembled, account for it 1891 */ 1892 fragno = fragstoblks(fs, bbase); 1893 if (ffs_isblock(fs, blksfree, fragno)) { 1894 cgp->cg_cs.cs_nffree -= fs->fs_frag; 1895 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 1896 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 1897 ffs_clusteracct(ump, fs, cgp, fragno, 1); 1898 cgp->cg_cs.cs_nbfree++; 1899 fs->fs_cstotal.cs_nbfree++; 1900 fs->fs_cs(fs, cg).cs_nbfree++; 1901 } 1902 } 1903 fs->fs_fmod = 1; 1904 ACTIVECLEAR(fs, cg); 1905 UFS_UNLOCK(ump); 1906 bdwrite(bp); 1907 } 1908 1909 #ifdef DIAGNOSTIC 1910 /* 1911 * Verify allocation of a block or fragment. Returns true if block or 1912 * fragment is allocated, false if it is free. 1913 */ 1914 static int 1915 ffs_checkblk(ip, bno, size) 1916 struct inode *ip; 1917 ufs2_daddr_t bno; 1918 long size; 1919 { 1920 struct fs *fs; 1921 struct cg *cgp; 1922 struct buf *bp; 1923 ufs1_daddr_t cgbno; 1924 int i, error, frags, free; 1925 u_int8_t *blksfree; 1926 1927 fs = ip->i_fs; 1928 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 1929 printf("bsize = %ld, size = %ld, fs = %s\n", 1930 (long)fs->fs_bsize, size, fs->fs_fsmnt); 1931 panic("ffs_checkblk: bad size"); 1932 } 1933 if ((u_int)bno >= fs->fs_size) 1934 panic("ffs_checkblk: bad block %jd", (intmax_t)bno); 1935 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))), 1936 (int)fs->fs_cgsize, NOCRED, &bp); 1937 if (error) 1938 panic("ffs_checkblk: cg bread failed"); 1939 cgp = (struct cg *)bp->b_data; 1940 if (!cg_chkmagic(cgp)) 1941 panic("ffs_checkblk: cg magic mismatch"); 1942 bp->b_xflags |= BX_BKGRDWRITE; 1943 blksfree = cg_blksfree(cgp); 1944 cgbno = dtogd(fs, bno); 1945 if (size == fs->fs_bsize) { 1946 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno)); 1947 } else { 1948 frags = numfrags(fs, size); 1949 for (free = 0, i = 0; i < frags; i++) 1950 if (isset(blksfree, cgbno + i)) 1951 free++; 1952 if (free != 0 && free != frags) 1953 panic("ffs_checkblk: partially free fragment"); 1954 } 1955 brelse(bp); 1956 return (!free); 1957 } 1958 #endif /* DIAGNOSTIC */ 1959 1960 /* 1961 * Free an inode. 1962 */ 1963 int 1964 ffs_vfree(pvp, ino, mode) 1965 struct vnode *pvp; 1966 ino_t ino; 1967 int mode; 1968 { 1969 struct inode *ip; 1970 1971 if (DOINGSOFTDEP(pvp)) { 1972 softdep_freefile(pvp, ino, mode); 1973 return (0); 1974 } 1975 ip = VTOI(pvp); 1976 return (ffs_freefile(ip->i_ump, ip->i_fs, ip->i_devvp, ino, mode)); 1977 } 1978 1979 /* 1980 * Do the actual free operation. 1981 * The specified inode is placed back in the free map. 1982 */ 1983 int 1984 ffs_freefile(ump, fs, devvp, ino, mode) 1985 struct ufsmount *ump; 1986 struct fs *fs; 1987 struct vnode *devvp; 1988 ino_t ino; 1989 int mode; 1990 { 1991 struct cg *cgp; 1992 struct buf *bp; 1993 ufs2_daddr_t cgbno; 1994 int error, cg; 1995 u_int8_t *inosused; 1996 struct cdev *dev; 1997 1998 cg = ino_to_cg(fs, ino); 1999 if (devvp->v_type != VCHR) { 2000 /* devvp is a snapshot */ 2001 dev = VTOI(devvp)->i_devvp->v_rdev; 2002 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2003 } else { 2004 /* devvp is a normal disk device */ 2005 dev = devvp->v_rdev; 2006 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2007 } 2008 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 2009 panic("ffs_freefile: range: dev = %s, ino = %lu, fs = %s", 2010 devtoname(dev), (u_long)ino, fs->fs_fsmnt); 2011 if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) { 2012 brelse(bp); 2013 return (error); 2014 } 2015 cgp = (struct cg *)bp->b_data; 2016 if (!cg_chkmagic(cgp)) { 2017 brelse(bp); 2018 return (0); 2019 } 2020 bp->b_xflags |= BX_BKGRDWRITE; 2021 cgp->cg_old_time = cgp->cg_time = time_second; 2022 inosused = cg_inosused(cgp); 2023 ino %= fs->fs_ipg; 2024 if (isclr(inosused, ino)) { 2025 printf("dev = %s, ino = %lu, fs = %s\n", devtoname(dev), 2026 (u_long)ino + cg * fs->fs_ipg, fs->fs_fsmnt); 2027 if (fs->fs_ronly == 0) 2028 panic("ffs_freefile: freeing free inode"); 2029 } 2030 clrbit(inosused, ino); 2031 if (ino < cgp->cg_irotor) 2032 cgp->cg_irotor = ino; 2033 cgp->cg_cs.cs_nifree++; 2034 UFS_LOCK(ump); 2035 fs->fs_cstotal.cs_nifree++; 2036 fs->fs_cs(fs, cg).cs_nifree++; 2037 if ((mode & IFMT) == IFDIR) { 2038 cgp->cg_cs.cs_ndir--; 2039 fs->fs_cstotal.cs_ndir--; 2040 fs->fs_cs(fs, cg).cs_ndir--; 2041 } 2042 fs->fs_fmod = 1; 2043 ACTIVECLEAR(fs, cg); 2044 UFS_UNLOCK(ump); 2045 bdwrite(bp); 2046 return (0); 2047 } 2048 2049 /* 2050 * Check to see if a file is free. 2051 */ 2052 int 2053 ffs_checkfreefile(fs, devvp, ino) 2054 struct fs *fs; 2055 struct vnode *devvp; 2056 ino_t ino; 2057 { 2058 struct cg *cgp; 2059 struct buf *bp; 2060 ufs2_daddr_t cgbno; 2061 int ret, cg; 2062 u_int8_t *inosused; 2063 2064 cg = ino_to_cg(fs, ino); 2065 if (devvp->v_type != VCHR) { 2066 /* devvp is a snapshot */ 2067 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2068 } else { 2069 /* devvp is a normal disk device */ 2070 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2071 } 2072 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 2073 return (1); 2074 if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) { 2075 brelse(bp); 2076 return (1); 2077 } 2078 cgp = (struct cg *)bp->b_data; 2079 if (!cg_chkmagic(cgp)) { 2080 brelse(bp); 2081 return (1); 2082 } 2083 inosused = cg_inosused(cgp); 2084 ino %= fs->fs_ipg; 2085 ret = isclr(inosused, ino); 2086 brelse(bp); 2087 return (ret); 2088 } 2089 2090 /* 2091 * Find a block of the specified size in the specified cylinder group. 2092 * 2093 * It is a panic if a request is made to find a block if none are 2094 * available. 2095 */ 2096 static ufs1_daddr_t 2097 ffs_mapsearch(fs, cgp, bpref, allocsiz) 2098 struct fs *fs; 2099 struct cg *cgp; 2100 ufs2_daddr_t bpref; 2101 int allocsiz; 2102 { 2103 ufs1_daddr_t bno; 2104 int start, len, loc, i; 2105 int blk, field, subfield, pos; 2106 u_int8_t *blksfree; 2107 2108 /* 2109 * find the fragment by searching through the free block 2110 * map for an appropriate bit pattern 2111 */ 2112 if (bpref) 2113 start = dtogd(fs, bpref) / NBBY; 2114 else 2115 start = cgp->cg_frotor / NBBY; 2116 blksfree = cg_blksfree(cgp); 2117 len = howmany(fs->fs_fpg, NBBY) - start; 2118 loc = scanc((u_int)len, (u_char *)&blksfree[start], 2119 (u_char *)fragtbl[fs->fs_frag], 2120 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 2121 if (loc == 0) { 2122 len = start + 1; 2123 start = 0; 2124 loc = scanc((u_int)len, (u_char *)&blksfree[0], 2125 (u_char *)fragtbl[fs->fs_frag], 2126 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 2127 if (loc == 0) { 2128 printf("start = %d, len = %d, fs = %s\n", 2129 start, len, fs->fs_fsmnt); 2130 panic("ffs_alloccg: map corrupted"); 2131 /* NOTREACHED */ 2132 } 2133 } 2134 bno = (start + len - loc) * NBBY; 2135 cgp->cg_frotor = bno; 2136 /* 2137 * found the byte in the map 2138 * sift through the bits to find the selected frag 2139 */ 2140 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 2141 blk = blkmap(fs, blksfree, bno); 2142 blk <<= 1; 2143 field = around[allocsiz]; 2144 subfield = inside[allocsiz]; 2145 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 2146 if ((blk & field) == subfield) 2147 return (bno + pos); 2148 field <<= 1; 2149 subfield <<= 1; 2150 } 2151 } 2152 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); 2153 panic("ffs_alloccg: block not in map"); 2154 return (-1); 2155 } 2156 2157 /* 2158 * Update the cluster map because of an allocation or free. 2159 * 2160 * Cnt == 1 means free; cnt == -1 means allocating. 2161 */ 2162 void 2163 ffs_clusteracct(ump, fs, cgp, blkno, cnt) 2164 struct ufsmount *ump; 2165 struct fs *fs; 2166 struct cg *cgp; 2167 ufs1_daddr_t blkno; 2168 int cnt; 2169 { 2170 int32_t *sump; 2171 int32_t *lp; 2172 u_char *freemapp, *mapp; 2173 int i, start, end, forw, back, map, bit; 2174 2175 mtx_assert(UFS_MTX(ump), MA_OWNED); 2176 2177 if (fs->fs_contigsumsize <= 0) 2178 return; 2179 freemapp = cg_clustersfree(cgp); 2180 sump = cg_clustersum(cgp); 2181 /* 2182 * Allocate or clear the actual block. 2183 */ 2184 if (cnt > 0) 2185 setbit(freemapp, blkno); 2186 else 2187 clrbit(freemapp, blkno); 2188 /* 2189 * Find the size of the cluster going forward. 2190 */ 2191 start = blkno + 1; 2192 end = start + fs->fs_contigsumsize; 2193 if (end >= cgp->cg_nclusterblks) 2194 end = cgp->cg_nclusterblks; 2195 mapp = &freemapp[start / NBBY]; 2196 map = *mapp++; 2197 bit = 1 << (start % NBBY); 2198 for (i = start; i < end; i++) { 2199 if ((map & bit) == 0) 2200 break; 2201 if ((i & (NBBY - 1)) != (NBBY - 1)) { 2202 bit <<= 1; 2203 } else { 2204 map = *mapp++; 2205 bit = 1; 2206 } 2207 } 2208 forw = i - start; 2209 /* 2210 * Find the size of the cluster going backward. 2211 */ 2212 start = blkno - 1; 2213 end = start - fs->fs_contigsumsize; 2214 if (end < 0) 2215 end = -1; 2216 mapp = &freemapp[start / NBBY]; 2217 map = *mapp--; 2218 bit = 1 << (start % NBBY); 2219 for (i = start; i > end; i--) { 2220 if ((map & bit) == 0) 2221 break; 2222 if ((i & (NBBY - 1)) != 0) { 2223 bit >>= 1; 2224 } else { 2225 map = *mapp--; 2226 bit = 1 << (NBBY - 1); 2227 } 2228 } 2229 back = start - i; 2230 /* 2231 * Account for old cluster and the possibly new forward and 2232 * back clusters. 2233 */ 2234 i = back + forw + 1; 2235 if (i > fs->fs_contigsumsize) 2236 i = fs->fs_contigsumsize; 2237 sump[i] += cnt; 2238 if (back > 0) 2239 sump[back] -= cnt; 2240 if (forw > 0) 2241 sump[forw] -= cnt; 2242 /* 2243 * Update cluster summary information. 2244 */ 2245 lp = &sump[fs->fs_contigsumsize]; 2246 for (i = fs->fs_contigsumsize; i > 0; i--) 2247 if (*lp-- > 0) 2248 break; 2249 fs->fs_maxcluster[cgp->cg_cgx] = i; 2250 } 2251 2252 /* 2253 * Fserr prints the name of a filesystem with an error diagnostic. 2254 * 2255 * The form of the error message is: 2256 * fs: error message 2257 */ 2258 static void 2259 ffs_fserr(fs, inum, cp) 2260 struct fs *fs; 2261 ino_t inum; 2262 char *cp; 2263 { 2264 struct thread *td = curthread; /* XXX */ 2265 struct proc *p = td->td_proc; 2266 2267 log(LOG_ERR, "pid %d (%s), uid %d inumber %d on %s: %s\n", 2268 p->p_pid, p->p_comm, td->td_ucred->cr_uid, inum, fs->fs_fsmnt, cp); 2269 } 2270 2271 /* 2272 * This function provides the capability for the fsck program to 2273 * update an active filesystem. Eleven operations are provided: 2274 * 2275 * adjrefcnt(inode, amt) - adjusts the reference count on the 2276 * specified inode by the specified amount. Under normal 2277 * operation the count should always go down. Decrementing 2278 * the count to zero will cause the inode to be freed. 2279 * adjblkcnt(inode, amt) - adjust the number of blocks used to 2280 * by the specifed amount. 2281 * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) - 2282 * adjust the superblock summary. 2283 * freedirs(inode, count) - directory inodes [inode..inode + count - 1] 2284 * are marked as free. Inodes should never have to be marked 2285 * as in use. 2286 * freefiles(inode, count) - file inodes [inode..inode + count - 1] 2287 * are marked as free. Inodes should never have to be marked 2288 * as in use. 2289 * freeblks(blockno, size) - blocks [blockno..blockno + size - 1] 2290 * are marked as free. Blocks should never have to be marked 2291 * as in use. 2292 * setflags(flags, set/clear) - the fs_flags field has the specified 2293 * flags set (second parameter +1) or cleared (second parameter -1). 2294 */ 2295 2296 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS); 2297 2298 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT, 2299 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count"); 2300 2301 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR, 2302 sysctl_ffs_fsck, "Adjust Inode Used Blocks Count"); 2303 2304 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR, 2305 sysctl_ffs_fsck, "Adjust number of directories"); 2306 2307 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR, 2308 sysctl_ffs_fsck, "Adjust number of free blocks"); 2309 2310 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR, 2311 sysctl_ffs_fsck, "Adjust number of free inodes"); 2312 2313 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR, 2314 sysctl_ffs_fsck, "Adjust number of free frags"); 2315 2316 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR, 2317 sysctl_ffs_fsck, "Adjust number of free clusters"); 2318 2319 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR, 2320 sysctl_ffs_fsck, "Free Range of Directory Inodes"); 2321 2322 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR, 2323 sysctl_ffs_fsck, "Free Range of File Inodes"); 2324 2325 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR, 2326 sysctl_ffs_fsck, "Free Range of Blocks"); 2327 2328 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR, 2329 sysctl_ffs_fsck, "Change Filesystem Flags"); 2330 2331 #ifdef DEBUG 2332 static int fsckcmds = 0; 2333 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, ""); 2334 #endif /* DEBUG */ 2335 2336 static int 2337 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS) 2338 { 2339 struct fsck_cmd cmd; 2340 struct ufsmount *ump; 2341 struct vnode *vp; 2342 struct inode *ip; 2343 struct mount *mp; 2344 struct fs *fs; 2345 ufs2_daddr_t blkno; 2346 long blkcnt, blksize; 2347 struct file *fp; 2348 int filetype, error; 2349 2350 if (req->newlen > sizeof cmd) 2351 return (EBADRPC); 2352 if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0) 2353 return (error); 2354 if (cmd.version != FFS_CMD_VERSION) 2355 return (ERPCMISMATCH); 2356 if ((error = getvnode(curproc->p_fd, cmd.handle, &fp)) != 0) 2357 return (error); 2358 vn_start_write(fp->f_data, &mp, V_WAIT); 2359 if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) { 2360 vn_finished_write(mp); 2361 fdrop(fp, curthread); 2362 return (EINVAL); 2363 } 2364 if (mp->mnt_flag & MNT_RDONLY) { 2365 vn_finished_write(mp); 2366 fdrop(fp, curthread); 2367 return (EROFS); 2368 } 2369 ump = VFSTOUFS(mp); 2370 fs = ump->um_fs; 2371 filetype = IFREG; 2372 2373 switch (oidp->oid_number) { 2374 2375 case FFS_SET_FLAGS: 2376 #ifdef DEBUG 2377 if (fsckcmds) 2378 printf("%s: %s flags\n", mp->mnt_stat.f_mntonname, 2379 cmd.size > 0 ? "set" : "clear"); 2380 #endif /* DEBUG */ 2381 if (cmd.size > 0) 2382 fs->fs_flags |= (long)cmd.value; 2383 else 2384 fs->fs_flags &= ~(long)cmd.value; 2385 break; 2386 2387 case FFS_ADJ_REFCNT: 2388 #ifdef DEBUG 2389 if (fsckcmds) { 2390 printf("%s: adjust inode %jd count by %jd\n", 2391 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2392 (intmax_t)cmd.size); 2393 } 2394 #endif /* DEBUG */ 2395 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) 2396 break; 2397 ip = VTOI(vp); 2398 ip->i_nlink += cmd.size; 2399 DIP_SET(ip, i_nlink, ip->i_nlink); 2400 ip->i_effnlink += cmd.size; 2401 ip->i_flag |= IN_CHANGE; 2402 if (DOINGSOFTDEP(vp)) 2403 softdep_change_linkcnt(ip); 2404 vput(vp); 2405 break; 2406 2407 case FFS_ADJ_BLKCNT: 2408 #ifdef DEBUG 2409 if (fsckcmds) { 2410 printf("%s: adjust inode %jd block count by %jd\n", 2411 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, 2412 (intmax_t)cmd.size); 2413 } 2414 #endif /* DEBUG */ 2415 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) 2416 break; 2417 ip = VTOI(vp); 2418 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size); 2419 ip->i_flag |= IN_CHANGE; 2420 vput(vp); 2421 break; 2422 2423 case FFS_DIR_FREE: 2424 filetype = IFDIR; 2425 /* fall through */ 2426 2427 case FFS_FILE_FREE: 2428 #ifdef DEBUG 2429 if (fsckcmds) { 2430 if (cmd.size == 1) 2431 printf("%s: free %s inode %d\n", 2432 mp->mnt_stat.f_mntonname, 2433 filetype == IFDIR ? "directory" : "file", 2434 (ino_t)cmd.value); 2435 else 2436 printf("%s: free %s inodes %d-%d\n", 2437 mp->mnt_stat.f_mntonname, 2438 filetype == IFDIR ? "directory" : "file", 2439 (ino_t)cmd.value, 2440 (ino_t)(cmd.value + cmd.size - 1)); 2441 } 2442 #endif /* DEBUG */ 2443 while (cmd.size > 0) { 2444 if ((error = ffs_freefile(ump, fs, ump->um_devvp, 2445 cmd.value, filetype))) 2446 break; 2447 cmd.size -= 1; 2448 cmd.value += 1; 2449 } 2450 break; 2451 2452 case FFS_BLK_FREE: 2453 #ifdef DEBUG 2454 if (fsckcmds) { 2455 if (cmd.size == 1) 2456 printf("%s: free block %jd\n", 2457 mp->mnt_stat.f_mntonname, 2458 (intmax_t)cmd.value); 2459 else 2460 printf("%s: free blocks %jd-%jd\n", 2461 mp->mnt_stat.f_mntonname, 2462 (intmax_t)cmd.value, 2463 (intmax_t)cmd.value + cmd.size - 1); 2464 } 2465 #endif /* DEBUG */ 2466 blkno = cmd.value; 2467 blkcnt = cmd.size; 2468 blksize = fs->fs_frag - (blkno % fs->fs_frag); 2469 while (blkcnt > 0) { 2470 if (blksize > blkcnt) 2471 blksize = blkcnt; 2472 ffs_blkfree(ump, fs, ump->um_devvp, blkno, 2473 blksize * fs->fs_fsize, ROOTINO); 2474 blkno += blksize; 2475 blkcnt -= blksize; 2476 blksize = fs->fs_frag; 2477 } 2478 break; 2479 2480 /* 2481 * Adjust superblock summaries. fsck(8) is expected to 2482 * submit deltas when necessary. 2483 */ 2484 case FFS_ADJ_NDIR: 2485 #ifdef DEBUG 2486 if (fsckcmds) { 2487 printf("%s: adjust number of directories by %jd\n", 2488 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2489 } 2490 #endif /* DEBUG */ 2491 fs->fs_cstotal.cs_ndir += cmd.value; 2492 break; 2493 case FFS_ADJ_NBFREE: 2494 #ifdef DEBUG 2495 if (fsckcmds) { 2496 printf("%s: adjust number of free blocks by %+jd\n", 2497 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2498 } 2499 #endif /* DEBUG */ 2500 fs->fs_cstotal.cs_nbfree += cmd.value; 2501 break; 2502 case FFS_ADJ_NIFREE: 2503 #ifdef DEBUG 2504 if (fsckcmds) { 2505 printf("%s: adjust number of free inodes by %+jd\n", 2506 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2507 } 2508 #endif /* DEBUG */ 2509 fs->fs_cstotal.cs_nifree += cmd.value; 2510 break; 2511 case FFS_ADJ_NFFREE: 2512 #ifdef DEBUG 2513 if (fsckcmds) { 2514 printf("%s: adjust number of free frags by %+jd\n", 2515 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2516 } 2517 #endif /* DEBUG */ 2518 fs->fs_cstotal.cs_nffree += cmd.value; 2519 break; 2520 case FFS_ADJ_NUMCLUSTERS: 2521 #ifdef DEBUG 2522 if (fsckcmds) { 2523 printf("%s: adjust number of free clusters by %+jd\n", 2524 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); 2525 } 2526 #endif /* DEBUG */ 2527 fs->fs_cstotal.cs_numclusters += cmd.value; 2528 break; 2529 2530 default: 2531 #ifdef DEBUG 2532 if (fsckcmds) { 2533 printf("Invalid request %d from fsck\n", 2534 oidp->oid_number); 2535 } 2536 #endif /* DEBUG */ 2537 error = EINVAL; 2538 break; 2539 2540 } 2541 fdrop(fp, curthread); 2542 vn_finished_write(mp); 2543 return (error); 2544 } 2545