1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95 34 * $FreeBSD: src/sys/ufs/ffs/ffs_alloc.c,v 1.64.2.2 2001/09/21 19:15:21 dillon Exp $ 35 */ 36 37 #include "opt_quota.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/buf.h> 42 #include <sys/conf.h> 43 #include <sys/proc.h> 44 #include <sys/vnode.h> 45 #include <sys/mount.h> 46 #include <sys/kernel.h> 47 #include <sys/sysctl.h> 48 #include <sys/syslog.h> 49 50 #include <ufs/ufs/quota.h> 51 #include <ufs/ufs/inode.h> 52 #include <ufs/ufs/ufs_extern.h> 53 #include <ufs/ufs/ufsmount.h> 54 55 #include <ufs/ffs/fs.h> 56 #include <ufs/ffs/ffs_extern.h> 57 58 typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref, 59 int size)); 60 61 static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int)); 62 static ufs_daddr_t 63 ffs_alloccgblk __P((struct inode *, struct buf *, ufs_daddr_t)); 64 #ifdef DIAGNOSTIC 65 static int ffs_checkblk __P((struct inode *, ufs_daddr_t, long)); 66 #endif 67 static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t, 68 int)); 69 static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t, 70 int)); 71 static ino_t ffs_dirpref __P((struct inode *)); 72 static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int)); 73 static void ffs_fserr __P((struct fs *, u_int, char *)); 74 static u_long ffs_hashalloc 75 __P((struct inode *, int, long, int, allocfcn_t *)); 76 static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int)); 77 static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t, 78 int)); 79 80 /* 81 * Allocate a block in the file system. 82 * 83 * The size of the requested block is given, which must be some 84 * multiple of fs_fsize and <= fs_bsize. 85 * A preference may be optionally specified. If a preference is given 86 * the following hierarchy is used to allocate a block: 87 * 1) allocate the requested block. 88 * 2) allocate a rotationally optimal block in the same cylinder. 89 * 3) allocate a block in the same cylinder group. 90 * 4) quadradically rehash into other cylinder groups, until an 91 * available block is located. 92 * If no block preference is given the following heirarchy is used 93 * to allocate a block: 94 * 1) allocate a block in the cylinder group that contains the 95 * inode for the file. 96 * 2) quadradically rehash into other cylinder groups, until an 97 * available block is located. 98 */ 99 int 100 ffs_alloc(ip, lbn, bpref, size, cred, bnp) 101 register struct inode *ip; 102 ufs_daddr_t lbn, bpref; 103 int size; 104 struct ucred *cred; 105 ufs_daddr_t *bnp; 106 { 107 register struct fs *fs; 108 ufs_daddr_t bno; 109 int cg; 110 #ifdef QUOTA 111 int error; 112 #endif 113 114 *bnp = 0; 115 fs = ip->i_fs; 116 #ifdef DIAGNOSTIC 117 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 118 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n", 119 devtoname(ip->i_dev), (long)fs->fs_bsize, size, 120 fs->fs_fsmnt); 121 panic("ffs_alloc: bad size"); 122 } 123 if (cred == NOCRED) 124 panic("ffs_alloc: missing credential"); 125 #endif /* DIAGNOSTIC */ 126 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) 127 goto nospace; 128 if (cred->cr_uid != 0 && 129 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) 130 goto nospace; 131 #ifdef QUOTA 132 error = chkdq(ip, (long)btodb(size), cred, 0); 133 if (error) 134 return (error); 135 #endif 136 if (bpref >= fs->fs_size) 137 bpref = 0; 138 if (bpref == 0) 139 cg = ino_to_cg(fs, ip->i_number); 140 else 141 cg = dtog(fs, bpref); 142 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size, 143 ffs_alloccg); 144 if (bno > 0) { 145 ip->i_blocks += btodb(size); 146 ip->i_flag |= IN_CHANGE | IN_UPDATE; 147 *bnp = bno; 148 return (0); 149 } 150 #ifdef QUOTA 151 /* 152 * Restore user's disk quota because allocation failed. 153 */ 154 (void) chkdq(ip, (long)-btodb(size), cred, FORCE); 155 #endif 156 nospace: 157 ffs_fserr(fs, cred->cr_uid, "file system full"); 158 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 159 return (ENOSPC); 160 } 161 162 /* 163 * Reallocate a fragment to a bigger size 164 * 165 * The number and size of the old block is given, and a preference 166 * and new size is also specified. The allocator attempts to extend 167 * the original block. Failing that, the regular block allocator is 168 * invoked to get an appropriate block. 169 */ 170 int 171 ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp) 172 register struct inode *ip; 173 ufs_daddr_t lbprev; 174 ufs_daddr_t bpref; 175 int osize, nsize; 176 struct ucred *cred; 177 struct buf **bpp; 178 { 179 register struct fs *fs; 180 struct buf *bp; 181 int cg, request, error; 182 ufs_daddr_t bprev, bno; 183 184 *bpp = 0; 185 fs = ip->i_fs; 186 #ifdef DIAGNOSTIC 187 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || 188 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { 189 printf( 190 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", 191 devtoname(ip->i_dev), (long)fs->fs_bsize, osize, 192 nsize, fs->fs_fsmnt); 193 panic("ffs_realloccg: bad size"); 194 } 195 if (cred == NOCRED) 196 panic("ffs_realloccg: missing credential"); 197 #endif /* DIAGNOSTIC */ 198 if (cred->cr_uid != 0 && 199 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) 200 goto nospace; 201 if ((bprev = ip->i_db[lbprev]) == 0) { 202 printf("dev = %s, bsize = %ld, bprev = %ld, fs = %s\n", 203 devtoname(ip->i_dev), (long)fs->fs_bsize, (long)bprev, 204 fs->fs_fsmnt); 205 panic("ffs_realloccg: bad bprev"); 206 } 207 /* 208 * Allocate the extra space in the buffer. 209 */ 210 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp); 211 if (error) { 212 brelse(bp); 213 return (error); 214 } 215 216 if( bp->b_blkno == bp->b_lblkno) { 217 if( lbprev >= NDADDR) 218 panic("ffs_realloccg: lbprev out of range"); 219 bp->b_blkno = fsbtodb(fs, bprev); 220 } 221 222 #ifdef QUOTA 223 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0); 224 if (error) { 225 brelse(bp); 226 return (error); 227 } 228 #endif 229 /* 230 * Check for extension in the existing location. 231 */ 232 cg = dtog(fs, bprev); 233 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize); 234 if (bno) { 235 if (bp->b_blkno != fsbtodb(fs, bno)) 236 panic("ffs_realloccg: bad blockno"); 237 ip->i_blocks += btodb(nsize - osize); 238 ip->i_flag |= IN_CHANGE | IN_UPDATE; 239 allocbuf(bp, nsize); 240 bp->b_flags |= B_DONE; 241 bzero((char *)bp->b_data + osize, (u_int)nsize - osize); 242 *bpp = bp; 243 return (0); 244 } 245 /* 246 * Allocate a new disk location. 247 */ 248 if (bpref >= fs->fs_size) 249 bpref = 0; 250 switch ((int)fs->fs_optim) { 251 case FS_OPTSPACE: 252 /* 253 * Allocate an exact sized fragment. Although this makes 254 * best use of space, we will waste time relocating it if 255 * the file continues to grow. If the fragmentation is 256 * less than half of the minimum free reserve, we choose 257 * to begin optimizing for time. 258 */ 259 request = nsize; 260 if (fs->fs_minfree <= 5 || 261 fs->fs_cstotal.cs_nffree > 262 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100)) 263 break; 264 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", 265 fs->fs_fsmnt); 266 fs->fs_optim = FS_OPTTIME; 267 break; 268 case FS_OPTTIME: 269 /* 270 * At this point we have discovered a file that is trying to 271 * grow a small fragment to a larger fragment. To save time, 272 * we allocate a full sized block, then free the unused portion. 273 * If the file continues to grow, the `ffs_fragextend' call 274 * above will be able to grow it in place without further 275 * copying. If aberrant programs cause disk fragmentation to 276 * grow within 2% of the free reserve, we choose to begin 277 * optimizing for space. 278 */ 279 request = fs->fs_bsize; 280 if (fs->fs_cstotal.cs_nffree < 281 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100) 282 break; 283 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", 284 fs->fs_fsmnt); 285 fs->fs_optim = FS_OPTSPACE; 286 break; 287 default: 288 printf("dev = %s, optim = %ld, fs = %s\n", 289 devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt); 290 panic("ffs_realloccg: bad optim"); 291 /* NOTREACHED */ 292 } 293 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request, 294 ffs_alloccg); 295 if (bno > 0) { 296 bp->b_blkno = fsbtodb(fs, bno); 297 if (!DOINGSOFTDEP(ITOV(ip))) 298 ffs_blkfree(ip, bprev, (long)osize); 299 if (nsize < request) 300 ffs_blkfree(ip, bno + numfrags(fs, nsize), 301 (long)(request - nsize)); 302 ip->i_blocks += btodb(nsize - osize); 303 ip->i_flag |= IN_CHANGE | IN_UPDATE; 304 allocbuf(bp, nsize); 305 bp->b_flags |= B_DONE; 306 bzero((char *)bp->b_data + osize, (u_int)nsize - osize); 307 *bpp = bp; 308 return (0); 309 } 310 #ifdef QUOTA 311 /* 312 * Restore user's disk quota because allocation failed. 313 */ 314 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE); 315 #endif 316 brelse(bp); 317 nospace: 318 /* 319 * no space available 320 */ 321 ffs_fserr(fs, cred->cr_uid, "file system full"); 322 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 323 return (ENOSPC); 324 } 325 326 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem"); 327 328 /* 329 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 330 * 331 * The vnode and an array of buffer pointers for a range of sequential 332 * logical blocks to be made contiguous is given. The allocator attempts 333 * to find a range of sequential blocks starting as close as possible to 334 * an fs_rotdelay offset from the end of the allocation for the logical 335 * block immediately preceeding the current range. If successful, the 336 * physical block numbers in the buffer pointers and in the inode are 337 * changed to reflect the new allocation. If unsuccessful, the allocation 338 * is left unchanged. The success in doing the reallocation is returned. 339 * Note that the error return is not reflected back to the user. Rather 340 * the previous block allocation will be used. 341 */ 342 static int doasyncfree = 1; 343 SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, ""); 344 345 static int doreallocblks = 1; 346 SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); 347 348 #ifdef DEBUG 349 static volatile int prtrealloc = 0; 350 #endif 351 352 int 353 ffs_reallocblks(ap) 354 struct vop_reallocblks_args /* { 355 struct vnode *a_vp; 356 struct cluster_save *a_buflist; 357 } */ *ap; 358 { 359 struct fs *fs; 360 struct inode *ip; 361 struct vnode *vp; 362 struct buf *sbp, *ebp; 363 ufs_daddr_t *bap, *sbap, *ebap = 0; 364 struct cluster_save *buflist; 365 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno; 366 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; 367 int i, len, start_lvl, end_lvl, pref, ssize; 368 369 if (doreallocblks == 0) 370 return (ENOSPC); 371 vp = ap->a_vp; 372 ip = VTOI(vp); 373 fs = ip->i_fs; 374 if (fs->fs_contigsumsize <= 0) 375 return (ENOSPC); 376 buflist = ap->a_buflist; 377 len = buflist->bs_nchildren; 378 start_lbn = buflist->bs_children[0]->b_lblkno; 379 end_lbn = start_lbn + len - 1; 380 #ifdef DIAGNOSTIC 381 for (i = 0; i < len; i++) 382 if (!ffs_checkblk(ip, 383 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 384 panic("ffs_reallocblks: unallocated block 1"); 385 for (i = 1; i < len; i++) 386 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 387 panic("ffs_reallocblks: non-logical cluster"); 388 blkno = buflist->bs_children[0]->b_blkno; 389 ssize = fsbtodb(fs, fs->fs_frag); 390 for (i = 1; i < len - 1; i++) 391 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) 392 panic("ffs_reallocblks: non-physical cluster %d", i); 393 #endif 394 /* 395 * If the latest allocation is in a new cylinder group, assume that 396 * the filesystem has decided to move and do not force it back to 397 * the previous cylinder group. 398 */ 399 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 400 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 401 return (ENOSPC); 402 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 403 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 404 return (ENOSPC); 405 /* 406 * Get the starting offset and block map for the first block. 407 */ 408 if (start_lvl == 0) { 409 sbap = &ip->i_db[0]; 410 soff = start_lbn; 411 } else { 412 idp = &start_ap[start_lvl - 1]; 413 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { 414 brelse(sbp); 415 return (ENOSPC); 416 } 417 sbap = (ufs_daddr_t *)sbp->b_data; 418 soff = idp->in_off; 419 } 420 /* 421 * Find the preferred location for the cluster. 422 */ 423 pref = ffs_blkpref(ip, start_lbn, soff, sbap); 424 /* 425 * If the block range spans two block maps, get the second map. 426 */ 427 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 428 ssize = len; 429 } else { 430 #ifdef DIAGNOSTIC 431 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn) 432 panic("ffs_reallocblk: start == end"); 433 #endif 434 ssize = len - (idp->in_off + 1); 435 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) 436 goto fail; 437 ebap = (ufs_daddr_t *)ebp->b_data; 438 } 439 /* 440 * Search the block map looking for an allocation of the desired size. 441 */ 442 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref, 443 len, ffs_clusteralloc)) == 0) 444 goto fail; 445 /* 446 * We have found a new contiguous block. 447 * 448 * First we have to replace the old block pointers with the new 449 * block pointers in the inode and indirect blocks associated 450 * with the file. 451 */ 452 #ifdef DEBUG 453 if (prtrealloc) 454 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number, 455 start_lbn, end_lbn); 456 #endif 457 blkno = newblk; 458 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { 459 if (i == ssize) { 460 bap = ebap; 461 soff = -i; 462 } 463 #ifdef DIAGNOSTIC 464 if (!ffs_checkblk(ip, 465 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 466 panic("ffs_reallocblks: unallocated block 2"); 467 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) 468 panic("ffs_reallocblks: alloc mismatch"); 469 #endif 470 #ifdef DEBUG 471 if (prtrealloc) 472 printf(" %d,", *bap); 473 #endif 474 if (DOINGSOFTDEP(vp)) { 475 if (sbap == &ip->i_db[0] && i < ssize) 476 softdep_setup_allocdirect(ip, start_lbn + i, 477 blkno, *bap, fs->fs_bsize, fs->fs_bsize, 478 buflist->bs_children[i]); 479 else 480 softdep_setup_allocindir_page(ip, start_lbn + i, 481 i < ssize ? sbp : ebp, soff + i, blkno, 482 *bap, buflist->bs_children[i]); 483 } 484 *bap++ = blkno; 485 } 486 /* 487 * Next we must write out the modified inode and indirect blocks. 488 * For strict correctness, the writes should be synchronous since 489 * the old block values may have been written to disk. In practise 490 * they are almost never written, but if we are concerned about 491 * strict correctness, the `doasyncfree' flag should be set to zero. 492 * 493 * The test on `doasyncfree' should be changed to test a flag 494 * that shows whether the associated buffers and inodes have 495 * been written. The flag should be set when the cluster is 496 * started and cleared whenever the buffer or inode is flushed. 497 * We can then check below to see if it is set, and do the 498 * synchronous write only when it has been cleared. 499 */ 500 if (sbap != &ip->i_db[0]) { 501 if (doasyncfree) 502 bdwrite(sbp); 503 else 504 bwrite(sbp); 505 } else { 506 ip->i_flag |= IN_CHANGE | IN_UPDATE; 507 if (!doasyncfree) 508 UFS_UPDATE(vp, 1); 509 } 510 if (ssize < len) { 511 if (doasyncfree) 512 bdwrite(ebp); 513 else 514 bwrite(ebp); 515 } 516 /* 517 * Last, free the old blocks and assign the new blocks to the buffers. 518 */ 519 #ifdef DEBUG 520 if (prtrealloc) 521 printf("\n\tnew:"); 522 #endif 523 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { 524 if (!DOINGSOFTDEP(vp)) 525 ffs_blkfree(ip, 526 dbtofsb(fs, buflist->bs_children[i]->b_blkno), 527 fs->fs_bsize); 528 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 529 #ifdef DIAGNOSTIC 530 if (!ffs_checkblk(ip, 531 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 532 panic("ffs_reallocblks: unallocated block 3"); 533 #endif 534 #ifdef DEBUG 535 if (prtrealloc) 536 printf(" %d,", blkno); 537 #endif 538 } 539 #ifdef DEBUG 540 if (prtrealloc) { 541 prtrealloc--; 542 printf("\n"); 543 } 544 #endif 545 return (0); 546 547 fail: 548 if (ssize < len) 549 brelse(ebp); 550 if (sbap != &ip->i_db[0]) 551 brelse(sbp); 552 return (ENOSPC); 553 } 554 555 /* 556 * Allocate an inode in the file system. 557 * 558 * If allocating a directory, use ffs_dirpref to select the inode. 559 * If allocating in a directory, the following hierarchy is followed: 560 * 1) allocate the preferred inode. 561 * 2) allocate an inode in the same cylinder group. 562 * 3) quadradically rehash into other cylinder groups, until an 563 * available inode is located. 564 * If no inode preference is given the following heirarchy is used 565 * to allocate an inode: 566 * 1) allocate an inode in cylinder group 0. 567 * 2) quadradically rehash into other cylinder groups, until an 568 * available inode is located. 569 */ 570 int 571 ffs_valloc(pvp, mode, cred, vpp) 572 struct vnode *pvp; 573 int mode; 574 struct ucred *cred; 575 struct vnode **vpp; 576 { 577 register struct inode *pip; 578 register struct fs *fs; 579 register struct inode *ip; 580 ino_t ino, ipref; 581 int cg, error; 582 583 *vpp = NULL; 584 pip = VTOI(pvp); 585 fs = pip->i_fs; 586 if (fs->fs_cstotal.cs_nifree == 0) 587 goto noinodes; 588 589 if ((mode & IFMT) == IFDIR) 590 ipref = ffs_dirpref(pip); 591 else 592 ipref = pip->i_number; 593 if (ipref >= fs->fs_ncg * fs->fs_ipg) 594 ipref = 0; 595 cg = ino_to_cg(fs, ipref); 596 /* 597 * Track number of dirs created one after another 598 * in a same cg without intervening by files. 599 */ 600 if ((mode & IFMT) == IFDIR) { 601 if (fs->fs_contigdirs[cg] < 255) 602 fs->fs_contigdirs[cg]++; 603 } else { 604 if (fs->fs_contigdirs[cg] > 0) 605 fs->fs_contigdirs[cg]--; 606 } 607 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode, 608 (allocfcn_t *)ffs_nodealloccg); 609 if (ino == 0) 610 goto noinodes; 611 error = VFS_VGET(pvp->v_mount, ino, vpp); 612 if (error) { 613 UFS_VFREE(pvp, ino, mode); 614 return (error); 615 } 616 ip = VTOI(*vpp); 617 if (ip->i_mode) { 618 printf("mode = 0%o, inum = %lu, fs = %s\n", 619 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt); 620 panic("ffs_valloc: dup alloc"); 621 } 622 if (ip->i_blocks) { /* XXX */ 623 printf("free inode %s/%lu had %ld blocks\n", 624 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks); 625 ip->i_blocks = 0; 626 } 627 ip->i_flags = 0; 628 /* 629 * Set up a new generation number for this inode. 630 */ 631 if (ip->i_gen == 0 || ++ip->i_gen == 0) 632 ip->i_gen = random() / 2 + 1; 633 return (0); 634 noinodes: 635 ffs_fserr(fs, cred->cr_uid, "out of inodes"); 636 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); 637 return (ENOSPC); 638 } 639 640 /* 641 * Find a cylinder group to place a directory. 642 * 643 * The policy implemented by this algorithm is to allocate a 644 * directory inode in the same cylinder group as its parent 645 * directory, but also to reserve space for its files inodes 646 * and data. Restrict the number of directories which may be 647 * allocated one after another in the same cylinder group 648 * without intervening allocation of files. 649 * 650 * If we allocate a first level directory then force allocation 651 * in another cylinder group. 652 */ 653 static ino_t 654 ffs_dirpref(pip) 655 struct inode *pip; 656 { 657 register struct fs *fs; 658 int cg, prefcg, dirsize, cgsize; 659 int avgifree, avgbfree, avgndir, curdirsize; 660 int minifree, minbfree, maxndir; 661 int mincg, minndir; 662 int maxcontigdirs; 663 664 fs = pip->i_fs; 665 666 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 667 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 668 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; 669 670 /* 671 * Force allocation in another cg if creating a first level dir. 672 */ 673 if (ITOV(pip)->v_flag & VROOT) { 674 prefcg = arc4random() % fs->fs_ncg; 675 mincg = prefcg; 676 minndir = fs->fs_ipg; 677 for (cg = prefcg; cg < fs->fs_ncg; cg++) 678 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 679 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 680 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 681 mincg = cg; 682 minndir = fs->fs_cs(fs, cg).cs_ndir; 683 } 684 for (cg = 0; cg < prefcg; cg++) 685 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 686 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 687 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 688 mincg = cg; 689 minndir = fs->fs_cs(fs, cg).cs_ndir; 690 } 691 return ((ino_t)(fs->fs_ipg * mincg)); 692 } 693 694 /* 695 * Count various limits which used for 696 * optimal allocation of a directory inode. 697 */ 698 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); 699 minifree = avgifree - fs->fs_ipg / 4; 700 if (minifree < 0) 701 minifree = 0; 702 minbfree = avgbfree - fs->fs_fpg / fs->fs_frag / 4; 703 if (minbfree < 0) 704 minbfree = 0; 705 cgsize = fs->fs_fsize * fs->fs_fpg; 706 dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir; 707 curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0; 708 if (dirsize < curdirsize) 709 dirsize = curdirsize; 710 maxcontigdirs = min(cgsize / dirsize, 255); 711 if (fs->fs_avgfpdir > 0) 712 maxcontigdirs = min(maxcontigdirs, 713 fs->fs_ipg / fs->fs_avgfpdir); 714 if (maxcontigdirs == 0) 715 maxcontigdirs = 1; 716 717 /* 718 * Limit number of dirs in one cg and reserve space for 719 * regular files, but only if we have no deficit in 720 * inodes or space. 721 */ 722 prefcg = ino_to_cg(fs, pip->i_number); 723 for (cg = prefcg; cg < fs->fs_ncg; cg++) 724 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 725 fs->fs_cs(fs, cg).cs_nifree >= minifree && 726 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 727 if (fs->fs_contigdirs[cg] < maxcontigdirs) 728 return ((ino_t)(fs->fs_ipg * cg)); 729 } 730 for (cg = 0; cg < prefcg; cg++) 731 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 732 fs->fs_cs(fs, cg).cs_nifree >= minifree && 733 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 734 if (fs->fs_contigdirs[cg] < maxcontigdirs) 735 return ((ino_t)(fs->fs_ipg * cg)); 736 } 737 /* 738 * This is a backstop when we have deficit in space. 739 */ 740 for (cg = prefcg; cg < fs->fs_ncg; cg++) 741 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 742 return ((ino_t)(fs->fs_ipg * cg)); 743 for (cg = 0; cg < prefcg; cg++) 744 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 745 break; 746 return ((ino_t)(fs->fs_ipg * cg)); 747 } 748 749 /* 750 * Select the desired position for the next block in a file. The file is 751 * logically divided into sections. The first section is composed of the 752 * direct blocks. Each additional section contains fs_maxbpg blocks. 753 * 754 * If no blocks have been allocated in the first section, the policy is to 755 * request a block in the same cylinder group as the inode that describes 756 * the file. If no blocks have been allocated in any other section, the 757 * policy is to place the section in a cylinder group with a greater than 758 * average number of free blocks. An appropriate cylinder group is found 759 * by using a rotor that sweeps the cylinder groups. When a new group of 760 * blocks is needed, the sweep begins in the cylinder group following the 761 * cylinder group from which the previous allocation was made. The sweep 762 * continues until a cylinder group with greater than the average number 763 * of free blocks is found. If the allocation is for the first block in an 764 * indirect block, the information on the previous allocation is unavailable; 765 * here a best guess is made based upon the logical block number being 766 * allocated. 767 * 768 * If a section is already partially allocated, the policy is to 769 * contiguously allocate fs_maxcontig blocks. The end of one of these 770 * contiguous blocks and the beginning of the next is physically separated 771 * so that the disk head will be in transit between them for at least 772 * fs_rotdelay milliseconds. This is to allow time for the processor to 773 * schedule another I/O transfer. 774 */ 775 ufs_daddr_t 776 ffs_blkpref(ip, lbn, indx, bap) 777 struct inode *ip; 778 ufs_daddr_t lbn; 779 int indx; 780 ufs_daddr_t *bap; 781 { 782 register struct fs *fs; 783 register int cg; 784 int avgbfree, startcg; 785 ufs_daddr_t nextblk; 786 787 fs = ip->i_fs; 788 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 789 if (lbn < NDADDR + NINDIR(fs)) { 790 cg = ino_to_cg(fs, ip->i_number); 791 return (fs->fs_fpg * cg + fs->fs_frag); 792 } 793 /* 794 * Find a cylinder with greater than average number of 795 * unused data blocks. 796 */ 797 if (indx == 0 || bap[indx - 1] == 0) 798 startcg = 799 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 800 else 801 startcg = dtog(fs, bap[indx - 1]) + 1; 802 startcg %= fs->fs_ncg; 803 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 804 for (cg = startcg; cg < fs->fs_ncg; cg++) 805 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 806 fs->fs_cgrotor = cg; 807 return (fs->fs_fpg * cg + fs->fs_frag); 808 } 809 for (cg = 0; cg <= startcg; cg++) 810 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 811 fs->fs_cgrotor = cg; 812 return (fs->fs_fpg * cg + fs->fs_frag); 813 } 814 return (0); 815 } 816 /* 817 * One or more previous blocks have been laid out. If less 818 * than fs_maxcontig previous blocks are contiguous, the 819 * next block is requested contiguously, otherwise it is 820 * requested rotationally delayed by fs_rotdelay milliseconds. 821 */ 822 nextblk = bap[indx - 1] + fs->fs_frag; 823 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig || 824 bap[indx - fs->fs_maxcontig] + 825 blkstofrags(fs, fs->fs_maxcontig) != nextblk) 826 return (nextblk); 827 /* 828 * Here we convert ms of delay to frags as: 829 * (frags) = (ms) * (rev/sec) * (sect/rev) / 830 * ((sect/frag) * (ms/sec)) 831 * then round up to the next block. 832 */ 833 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect / 834 (NSPF(fs) * 1000), fs->fs_frag); 835 return (nextblk); 836 } 837 838 /* 839 * Implement the cylinder overflow algorithm. 840 * 841 * The policy implemented by this algorithm is: 842 * 1) allocate the block in its requested cylinder group. 843 * 2) quadradically rehash on the cylinder group number. 844 * 3) brute force search for a free block. 845 */ 846 /*VARARGS5*/ 847 static u_long 848 ffs_hashalloc(ip, cg, pref, size, allocator) 849 struct inode *ip; 850 int cg; 851 long pref; 852 int size; /* size for data blocks, mode for inodes */ 853 allocfcn_t *allocator; 854 { 855 register struct fs *fs; 856 long result; /* XXX why not same type as we return? */ 857 int i, icg = cg; 858 859 fs = ip->i_fs; 860 /* 861 * 1: preferred cylinder group 862 */ 863 result = (*allocator)(ip, cg, pref, size); 864 if (result) 865 return (result); 866 /* 867 * 2: quadratic rehash 868 */ 869 for (i = 1; i < fs->fs_ncg; i *= 2) { 870 cg += i; 871 if (cg >= fs->fs_ncg) 872 cg -= fs->fs_ncg; 873 result = (*allocator)(ip, cg, 0, size); 874 if (result) 875 return (result); 876 } 877 /* 878 * 3: brute force search 879 * Note that we start at i == 2, since 0 was checked initially, 880 * and 1 is always checked in the quadratic rehash. 881 */ 882 cg = (icg + 2) % fs->fs_ncg; 883 for (i = 2; i < fs->fs_ncg; i++) { 884 result = (*allocator)(ip, cg, 0, size); 885 if (result) 886 return (result); 887 cg++; 888 if (cg == fs->fs_ncg) 889 cg = 0; 890 } 891 return (0); 892 } 893 894 /* 895 * Determine whether a fragment can be extended. 896 * 897 * Check to see if the necessary fragments are available, and 898 * if they are, allocate them. 899 */ 900 static ufs_daddr_t 901 ffs_fragextend(ip, cg, bprev, osize, nsize) 902 struct inode *ip; 903 int cg; 904 long bprev; 905 int osize, nsize; 906 { 907 register struct fs *fs; 908 register struct cg *cgp; 909 struct buf *bp; 910 long bno; 911 int frags, bbase; 912 int i, error; 913 u_int8_t *blksfree; 914 915 fs = ip->i_fs; 916 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 917 return (0); 918 frags = numfrags(fs, nsize); 919 bbase = fragnum(fs, bprev); 920 if (bbase > fragnum(fs, (bprev + frags - 1))) { 921 /* cannot extend across a block boundary */ 922 return (0); 923 } 924 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 925 (int)fs->fs_cgsize, NOCRED, &bp); 926 if (error) { 927 brelse(bp); 928 return (0); 929 } 930 cgp = (struct cg *)bp->b_data; 931 if (!cg_chkmagic(cgp)) { 932 brelse(bp); 933 return (0); 934 } 935 bp->b_xflags |= BX_BKGRDWRITE; 936 cgp->cg_time = time_second; 937 bno = dtogd(fs, bprev); 938 blksfree = cg_blksfree(cgp); 939 for (i = numfrags(fs, osize); i < frags; i++) 940 if (isclr(blksfree, bno + i)) { 941 brelse(bp); 942 return (0); 943 } 944 /* 945 * the current fragment can be extended 946 * deduct the count on fragment being extended into 947 * increase the count on the remaining fragment (if any) 948 * allocate the extended piece 949 */ 950 for (i = frags; i < fs->fs_frag - bbase; i++) 951 if (isclr(blksfree, bno + i)) 952 break; 953 cgp->cg_frsum[i - numfrags(fs, osize)]--; 954 if (i != frags) 955 cgp->cg_frsum[i - frags]++; 956 for (i = numfrags(fs, osize); i < frags; i++) { 957 clrbit(blksfree, bno + i); 958 cgp->cg_cs.cs_nffree--; 959 fs->fs_cstotal.cs_nffree--; 960 fs->fs_cs(fs, cg).cs_nffree--; 961 } 962 fs->fs_fmod = 1; 963 if (DOINGSOFTDEP(ITOV(ip))) 964 softdep_setup_blkmapdep(bp, fs, bprev); 965 bdwrite(bp); 966 return (bprev); 967 } 968 969 /* 970 * Determine whether a block can be allocated. 971 * 972 * Check to see if a block of the appropriate size is available, 973 * and if it is, allocate it. 974 */ 975 static ufs_daddr_t 976 ffs_alloccg(ip, cg, bpref, size) 977 struct inode *ip; 978 int cg; 979 ufs_daddr_t bpref; 980 int size; 981 { 982 register struct fs *fs; 983 register struct cg *cgp; 984 struct buf *bp; 985 register int i; 986 ufs_daddr_t bno, blkno; 987 int allocsiz, error, frags; 988 u_int8_t *blksfree; 989 990 fs = ip->i_fs; 991 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 992 return (0); 993 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 994 (int)fs->fs_cgsize, NOCRED, &bp); 995 if (error) { 996 brelse(bp); 997 return (0); 998 } 999 cgp = (struct cg *)bp->b_data; 1000 if (!cg_chkmagic(cgp) || 1001 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) { 1002 brelse(bp); 1003 return (0); 1004 } 1005 bp->b_xflags |= BX_BKGRDWRITE; 1006 cgp->cg_time = time_second; 1007 if (size == fs->fs_bsize) { 1008 bno = ffs_alloccgblk(ip, bp, bpref); 1009 bdwrite(bp); 1010 return (bno); 1011 } 1012 /* 1013 * check to see if any fragments are already available 1014 * allocsiz is the size which will be allocated, hacking 1015 * it down to a smaller size if necessary 1016 */ 1017 blksfree = cg_blksfree(cgp); 1018 frags = numfrags(fs, size); 1019 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 1020 if (cgp->cg_frsum[allocsiz] != 0) 1021 break; 1022 if (allocsiz == fs->fs_frag) { 1023 /* 1024 * no fragments were available, so a block will be 1025 * allocated, and hacked up 1026 */ 1027 if (cgp->cg_cs.cs_nbfree == 0) { 1028 brelse(bp); 1029 return (0); 1030 } 1031 bno = ffs_alloccgblk(ip, bp, bpref); 1032 bpref = dtogd(fs, bno); 1033 for (i = frags; i < fs->fs_frag; i++) 1034 setbit(blksfree, bpref + i); 1035 i = fs->fs_frag - frags; 1036 cgp->cg_cs.cs_nffree += i; 1037 fs->fs_cstotal.cs_nffree += i; 1038 fs->fs_cs(fs, cg).cs_nffree += i; 1039 fs->fs_fmod = 1; 1040 cgp->cg_frsum[i]++; 1041 bdwrite(bp); 1042 return (bno); 1043 } 1044 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 1045 if (bno < 0) { 1046 brelse(bp); 1047 return (0); 1048 } 1049 for (i = 0; i < frags; i++) 1050 clrbit(blksfree, bno + i); 1051 cgp->cg_cs.cs_nffree -= frags; 1052 fs->fs_cstotal.cs_nffree -= frags; 1053 fs->fs_cs(fs, cg).cs_nffree -= frags; 1054 fs->fs_fmod = 1; 1055 cgp->cg_frsum[allocsiz]--; 1056 if (frags != allocsiz) 1057 cgp->cg_frsum[allocsiz - frags]++; 1058 blkno = cg * fs->fs_fpg + bno; 1059 if (DOINGSOFTDEP(ITOV(ip))) 1060 softdep_setup_blkmapdep(bp, fs, blkno); 1061 bdwrite(bp); 1062 return ((u_long)blkno); 1063 } 1064 1065 /* 1066 * Allocate a block in a cylinder group. 1067 * 1068 * This algorithm implements the following policy: 1069 * 1) allocate the requested block. 1070 * 2) allocate a rotationally optimal block in the same cylinder. 1071 * 3) allocate the next available block on the block rotor for the 1072 * specified cylinder group. 1073 * Note that this routine only allocates fs_bsize blocks; these 1074 * blocks may be fragmented by the routine that allocates them. 1075 */ 1076 static ufs_daddr_t 1077 ffs_alloccgblk(ip, bp, bpref) 1078 struct inode *ip; 1079 struct buf *bp; 1080 ufs_daddr_t bpref; 1081 { 1082 struct fs *fs; 1083 struct cg *cgp; 1084 ufs_daddr_t bno, blkno; 1085 int cylno, pos, delta; 1086 short *cylbp; 1087 register int i; 1088 u_int8_t *blksfree; 1089 1090 fs = ip->i_fs; 1091 cgp = (struct cg *)bp->b_data; 1092 blksfree = cg_blksfree(cgp); 1093 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) { 1094 bpref = cgp->cg_rotor; 1095 goto norot; 1096 } 1097 bpref = blknum(fs, bpref); 1098 bpref = dtogd(fs, bpref); 1099 /* 1100 * if the requested block is available, use it 1101 */ 1102 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bpref))) { 1103 bno = bpref; 1104 goto gotit; 1105 } 1106 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) { 1107 /* 1108 * Block layout information is not available. 1109 * Leaving bpref unchanged means we take the 1110 * next available free block following the one 1111 * we just allocated. Hopefully this will at 1112 * least hit a track cache on drives of unknown 1113 * geometry (e.g. SCSI). 1114 */ 1115 goto norot; 1116 } 1117 /* 1118 * check for a block available on the same cylinder 1119 */ 1120 cylno = cbtocylno(fs, bpref); 1121 if (cg_blktot(cgp)[cylno] == 0) 1122 goto norot; 1123 /* 1124 * check the summary information to see if a block is 1125 * available in the requested cylinder starting at the 1126 * requested rotational position and proceeding around. 1127 */ 1128 cylbp = cg_blks(fs, cgp, cylno); 1129 pos = cbtorpos(fs, bpref); 1130 for (i = pos; i < fs->fs_nrpos; i++) 1131 if (cylbp[i] > 0) 1132 break; 1133 if (i == fs->fs_nrpos) 1134 for (i = 0; i < pos; i++) 1135 if (cylbp[i] > 0) 1136 break; 1137 if (cylbp[i] > 0) { 1138 /* 1139 * found a rotational position, now find the actual 1140 * block. A panic if none is actually there. 1141 */ 1142 pos = cylno % fs->fs_cpc; 1143 bno = (cylno - pos) * fs->fs_spc / NSPB(fs); 1144 if (fs_postbl(fs, pos)[i] == -1) { 1145 printf("pos = %d, i = %d, fs = %s\n", 1146 pos, i, fs->fs_fsmnt); 1147 panic("ffs_alloccgblk: cyl groups corrupted"); 1148 } 1149 for (i = fs_postbl(fs, pos)[i];; ) { 1150 if (ffs_isblock(fs, blksfree, bno + i)) { 1151 bno = blkstofrags(fs, (bno + i)); 1152 goto gotit; 1153 } 1154 delta = fs_rotbl(fs)[i]; 1155 if (delta <= 0 || 1156 delta + i > fragstoblks(fs, fs->fs_fpg)) 1157 break; 1158 i += delta; 1159 } 1160 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt); 1161 panic("ffs_alloccgblk: can't find blk in cyl"); 1162 } 1163 norot: 1164 /* 1165 * no blocks in the requested cylinder, so take next 1166 * available one in this cylinder group. 1167 */ 1168 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1169 if (bno < 0) 1170 return (0); 1171 cgp->cg_rotor = bno; 1172 gotit: 1173 blkno = fragstoblks(fs, bno); 1174 ffs_clrblock(fs, blksfree, (long)blkno); 1175 ffs_clusteracct(fs, cgp, blkno, -1); 1176 cgp->cg_cs.cs_nbfree--; 1177 fs->fs_cstotal.cs_nbfree--; 1178 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; 1179 cylno = cbtocylno(fs, bno); 1180 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--; 1181 cg_blktot(cgp)[cylno]--; 1182 fs->fs_fmod = 1; 1183 blkno = cgp->cg_cgx * fs->fs_fpg + bno; 1184 if (DOINGSOFTDEP(ITOV(ip))) 1185 softdep_setup_blkmapdep(bp, fs, blkno); 1186 return (blkno); 1187 } 1188 1189 /* 1190 * Determine whether a cluster can be allocated. 1191 * 1192 * We do not currently check for optimal rotational layout if there 1193 * are multiple choices in the same cylinder group. Instead we just 1194 * take the first one that we find following bpref. 1195 */ 1196 static ufs_daddr_t 1197 ffs_clusteralloc(ip, cg, bpref, len) 1198 struct inode *ip; 1199 int cg; 1200 ufs_daddr_t bpref; 1201 int len; 1202 { 1203 register struct fs *fs; 1204 register struct cg *cgp; 1205 struct buf *bp; 1206 int i, got, run, bno, bit, map; 1207 u_char *mapp; 1208 int32_t *lp; 1209 u_int8_t *blksfree; 1210 1211 fs = ip->i_fs; 1212 if (fs->fs_maxcluster[cg] < len) 1213 return (0); 1214 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize, 1215 NOCRED, &bp)) 1216 goto fail; 1217 cgp = (struct cg *)bp->b_data; 1218 if (!cg_chkmagic(cgp)) 1219 goto fail; 1220 bp->b_xflags |= BX_BKGRDWRITE; 1221 /* 1222 * Check to see if a cluster of the needed size (or bigger) is 1223 * available in this cylinder group. 1224 */ 1225 lp = &cg_clustersum(cgp)[len]; 1226 for (i = len; i <= fs->fs_contigsumsize; i++) 1227 if (*lp++ > 0) 1228 break; 1229 if (i > fs->fs_contigsumsize) { 1230 /* 1231 * This is the first time looking for a cluster in this 1232 * cylinder group. Update the cluster summary information 1233 * to reflect the true maximum sized cluster so that 1234 * future cluster allocation requests can avoid reading 1235 * the cylinder group map only to find no clusters. 1236 */ 1237 lp = &cg_clustersum(cgp)[len - 1]; 1238 for (i = len - 1; i > 0; i--) 1239 if (*lp-- > 0) 1240 break; 1241 fs->fs_maxcluster[cg] = i; 1242 goto fail; 1243 } 1244 /* 1245 * Search the cluster map to find a big enough cluster. 1246 * We take the first one that we find, even if it is larger 1247 * than we need as we prefer to get one close to the previous 1248 * block allocation. We do not search before the current 1249 * preference point as we do not want to allocate a block 1250 * that is allocated before the previous one (as we will 1251 * then have to wait for another pass of the elevator 1252 * algorithm before it will be read). We prefer to fail and 1253 * be recalled to try an allocation in the next cylinder group. 1254 */ 1255 if (dtog(fs, bpref) != cg) 1256 bpref = 0; 1257 else 1258 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref))); 1259 mapp = &cg_clustersfree(cgp)[bpref / NBBY]; 1260 map = *mapp++; 1261 bit = 1 << (bpref % NBBY); 1262 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { 1263 if ((map & bit) == 0) { 1264 run = 0; 1265 } else { 1266 run++; 1267 if (run == len) 1268 break; 1269 } 1270 if ((got & (NBBY - 1)) != (NBBY - 1)) { 1271 bit <<= 1; 1272 } else { 1273 map = *mapp++; 1274 bit = 1; 1275 } 1276 } 1277 if (got >= cgp->cg_nclusterblks) 1278 goto fail; 1279 /* 1280 * Allocate the cluster that we have found. 1281 */ 1282 blksfree = cg_blksfree(cgp); 1283 for (i = 1; i <= len; i++) 1284 if (!ffs_isblock(fs, blksfree, got - run + i)) 1285 panic("ffs_clusteralloc: map mismatch"); 1286 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1); 1287 if (dtog(fs, bno) != cg) 1288 panic("ffs_clusteralloc: allocated out of group"); 1289 len = blkstofrags(fs, len); 1290 for (i = 0; i < len; i += fs->fs_frag) 1291 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i) 1292 panic("ffs_clusteralloc: lost block"); 1293 bdwrite(bp); 1294 return (bno); 1295 1296 fail: 1297 brelse(bp); 1298 return (0); 1299 } 1300 1301 /* 1302 * Determine whether an inode can be allocated. 1303 * 1304 * Check to see if an inode is available, and if it is, 1305 * allocate it using the following policy: 1306 * 1) allocate the requested inode. 1307 * 2) allocate the next available inode after the requested 1308 * inode in the specified cylinder group. 1309 */ 1310 static ino_t 1311 ffs_nodealloccg(ip, cg, ipref, mode) 1312 struct inode *ip; 1313 int cg; 1314 ufs_daddr_t ipref; 1315 int mode; 1316 { 1317 register struct fs *fs; 1318 register struct cg *cgp; 1319 struct buf *bp; 1320 u_int8_t *inosused; 1321 int error, start, len, loc, map, i; 1322 1323 fs = ip->i_fs; 1324 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1325 return (0); 1326 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1327 (int)fs->fs_cgsize, NOCRED, &bp); 1328 if (error) { 1329 brelse(bp); 1330 return (0); 1331 } 1332 cgp = (struct cg *)bp->b_data; 1333 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) { 1334 brelse(bp); 1335 return (0); 1336 } 1337 bp->b_xflags |= BX_BKGRDWRITE; 1338 cgp->cg_time = time_second; 1339 inosused = cg_inosused(cgp); 1340 if (ipref) { 1341 ipref %= fs->fs_ipg; 1342 if (isclr(inosused, ipref)) 1343 goto gotit; 1344 } 1345 start = cgp->cg_irotor / NBBY; 1346 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); 1347 loc = skpc(0xff, len, &inosused[start]); 1348 if (loc == 0) { 1349 len = start + 1; 1350 start = 0; 1351 loc = skpc(0xff, len, &inosused[0]); 1352 if (loc == 0) { 1353 printf("cg = %d, irotor = %ld, fs = %s\n", 1354 cg, (long)cgp->cg_irotor, fs->fs_fsmnt); 1355 panic("ffs_nodealloccg: map corrupted"); 1356 /* NOTREACHED */ 1357 } 1358 } 1359 i = start + len - loc; 1360 map = inosused[i]; 1361 ipref = i * NBBY; 1362 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) { 1363 if ((map & i) == 0) { 1364 cgp->cg_irotor = ipref; 1365 goto gotit; 1366 } 1367 } 1368 printf("fs = %s\n", fs->fs_fsmnt); 1369 panic("ffs_nodealloccg: block not in map"); 1370 /* NOTREACHED */ 1371 gotit: 1372 if (DOINGSOFTDEP(ITOV(ip))) 1373 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref); 1374 setbit(inosused, ipref); 1375 cgp->cg_cs.cs_nifree--; 1376 fs->fs_cstotal.cs_nifree--; 1377 fs->fs_cs(fs, cg).cs_nifree--; 1378 fs->fs_fmod = 1; 1379 if ((mode & IFMT) == IFDIR) { 1380 cgp->cg_cs.cs_ndir++; 1381 fs->fs_cstotal.cs_ndir++; 1382 fs->fs_cs(fs, cg).cs_ndir++; 1383 } 1384 bdwrite(bp); 1385 return (cg * fs->fs_ipg + ipref); 1386 } 1387 1388 /* 1389 * Free a block or fragment. 1390 * 1391 * The specified block or fragment is placed back in the 1392 * free map. If a fragment is deallocated, a possible 1393 * block reassembly is checked. 1394 */ 1395 void 1396 ffs_blkfree(ip, bno, size) 1397 register struct inode *ip; 1398 ufs_daddr_t bno; 1399 long size; 1400 { 1401 register struct fs *fs; 1402 register struct cg *cgp; 1403 struct buf *bp; 1404 ufs_daddr_t blkno; 1405 int i, error, cg, blk, frags, bbase; 1406 u_int8_t *blksfree; 1407 1408 fs = ip->i_fs; 1409 VOP_FREEBLKS(ip->i_devvp, fsbtodb(fs, bno), size); 1410 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 1411 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 1412 printf("dev=%s, bno = %ld, bsize = %ld, size = %ld, fs = %s\n", 1413 devtoname(ip->i_dev), (long)bno, (long)fs->fs_bsize, size, 1414 fs->fs_fsmnt); 1415 panic("ffs_blkfree: bad size"); 1416 } 1417 cg = dtog(fs, bno); 1418 if ((u_int)bno >= fs->fs_size) { 1419 printf("bad block %ld, ino %lu\n", 1420 (long)bno, (u_long)ip->i_number); 1421 ffs_fserr(fs, ip->i_uid, "bad block"); 1422 return; 1423 } 1424 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1425 (int)fs->fs_cgsize, NOCRED, &bp); 1426 if (error) { 1427 brelse(bp); 1428 return; 1429 } 1430 cgp = (struct cg *)bp->b_data; 1431 if (!cg_chkmagic(cgp)) { 1432 brelse(bp); 1433 return; 1434 } 1435 bp->b_xflags |= BX_BKGRDWRITE; 1436 cgp->cg_time = time_second; 1437 bno = dtogd(fs, bno); 1438 blksfree = cg_blksfree(cgp); 1439 if (size == fs->fs_bsize) { 1440 blkno = fragstoblks(fs, bno); 1441 if (!ffs_isfreeblock(fs, blksfree, blkno)) { 1442 printf("dev = %s, block = %ld, fs = %s\n", 1443 devtoname(ip->i_dev), (long)bno, fs->fs_fsmnt); 1444 panic("ffs_blkfree: freeing free block"); 1445 } 1446 ffs_setblock(fs, blksfree, blkno); 1447 ffs_clusteracct(fs, cgp, blkno, 1); 1448 cgp->cg_cs.cs_nbfree++; 1449 fs->fs_cstotal.cs_nbfree++; 1450 fs->fs_cs(fs, cg).cs_nbfree++; 1451 i = cbtocylno(fs, bno); 1452 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++; 1453 cg_blktot(cgp)[i]++; 1454 } else { 1455 bbase = bno - fragnum(fs, bno); 1456 /* 1457 * decrement the counts associated with the old frags 1458 */ 1459 blk = blkmap(fs, blksfree, bbase); 1460 ffs_fragacct(fs, blk, cgp->cg_frsum, -1); 1461 /* 1462 * deallocate the fragment 1463 */ 1464 frags = numfrags(fs, size); 1465 for (i = 0; i < frags; i++) { 1466 if (isset(blksfree, bno + i)) { 1467 printf("dev = %s, block = %ld, fs = %s\n", 1468 devtoname(ip->i_dev), (long)(bno + i), 1469 fs->fs_fsmnt); 1470 panic("ffs_blkfree: freeing free frag"); 1471 } 1472 setbit(blksfree, bno + i); 1473 } 1474 cgp->cg_cs.cs_nffree += i; 1475 fs->fs_cstotal.cs_nffree += i; 1476 fs->fs_cs(fs, cg).cs_nffree += i; 1477 /* 1478 * add back in counts associated with the new frags 1479 */ 1480 blk = blkmap(fs, blksfree, bbase); 1481 ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 1482 /* 1483 * if a complete block has been reassembled, account for it 1484 */ 1485 blkno = fragstoblks(fs, bbase); 1486 if (ffs_isblock(fs, blksfree, blkno)) { 1487 cgp->cg_cs.cs_nffree -= fs->fs_frag; 1488 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 1489 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 1490 ffs_clusteracct(fs, cgp, blkno, 1); 1491 cgp->cg_cs.cs_nbfree++; 1492 fs->fs_cstotal.cs_nbfree++; 1493 fs->fs_cs(fs, cg).cs_nbfree++; 1494 i = cbtocylno(fs, bbase); 1495 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++; 1496 cg_blktot(cgp)[i]++; 1497 } 1498 } 1499 fs->fs_fmod = 1; 1500 bdwrite(bp); 1501 } 1502 1503 #ifdef DIAGNOSTIC 1504 /* 1505 * Verify allocation of a block or fragment. Returns true if block or 1506 * fragment is allocated, false if it is free. 1507 */ 1508 static int 1509 ffs_checkblk(ip, bno, size) 1510 struct inode *ip; 1511 ufs_daddr_t bno; 1512 long size; 1513 { 1514 struct fs *fs; 1515 struct cg *cgp; 1516 struct buf *bp; 1517 int i, error, frags, free; 1518 u_int8_t *blksfree; 1519 1520 fs = ip->i_fs; 1521 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 1522 printf("bsize = %ld, size = %ld, fs = %s\n", 1523 (long)fs->fs_bsize, size, fs->fs_fsmnt); 1524 panic("ffs_checkblk: bad size"); 1525 } 1526 if ((u_int)bno >= fs->fs_size) 1527 panic("ffs_checkblk: bad block %d", bno); 1528 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))), 1529 (int)fs->fs_cgsize, NOCRED, &bp); 1530 if (error) 1531 panic("ffs_checkblk: cg bread failed"); 1532 cgp = (struct cg *)bp->b_data; 1533 if (!cg_chkmagic(cgp)) 1534 panic("ffs_checkblk: cg magic mismatch"); 1535 bp->b_xflags |= BX_BKGRDWRITE; 1536 blksfree = cg_blksfree(cgp); 1537 bno = dtogd(fs, bno); 1538 if (size == fs->fs_bsize) { 1539 free = ffs_isblock(fs, blksfree, fragstoblks(fs, bno)); 1540 } else { 1541 frags = numfrags(fs, size); 1542 for (free = 0, i = 0; i < frags; i++) 1543 if (isset(blksfree, bno + i)) 1544 free++; 1545 if (free != 0 && free != frags) 1546 panic("ffs_checkblk: partially free fragment"); 1547 } 1548 brelse(bp); 1549 return (!free); 1550 } 1551 #endif /* DIAGNOSTIC */ 1552 1553 /* 1554 * Free an inode. 1555 */ 1556 int 1557 ffs_vfree( pvp, ino, mode) 1558 struct vnode *pvp; 1559 ino_t ino; 1560 int mode; 1561 { 1562 if (DOINGSOFTDEP(pvp)) { 1563 softdep_freefile(pvp, ino, mode); 1564 return (0); 1565 } 1566 return (ffs_freefile(pvp, ino, mode)); 1567 } 1568 1569 /* 1570 * Do the actual free operation. 1571 * The specified inode is placed back in the free map. 1572 */ 1573 int 1574 ffs_freefile( pvp, ino, mode) 1575 struct vnode *pvp; 1576 ino_t ino; 1577 int mode; 1578 { 1579 register struct fs *fs; 1580 register struct cg *cgp; 1581 register struct inode *pip; 1582 struct buf *bp; 1583 int error, cg; 1584 u_int8_t *inosused; 1585 1586 pip = VTOI(pvp); 1587 fs = pip->i_fs; 1588 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 1589 panic("ffs_vfree: range: dev = (%d,%d), ino = %d, fs = %s", 1590 major(pip->i_dev), minor(pip->i_dev), ino, fs->fs_fsmnt); 1591 cg = ino_to_cg(fs, ino); 1592 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1593 (int)fs->fs_cgsize, NOCRED, &bp); 1594 if (error) { 1595 brelse(bp); 1596 return (error); 1597 } 1598 cgp = (struct cg *)bp->b_data; 1599 if (!cg_chkmagic(cgp)) { 1600 brelse(bp); 1601 return (0); 1602 } 1603 bp->b_xflags |= BX_BKGRDWRITE; 1604 cgp->cg_time = time_second; 1605 inosused = cg_inosused(cgp); 1606 ino %= fs->fs_ipg; 1607 if (isclr(inosused, ino)) { 1608 printf("dev = %s, ino = %lu, fs = %s\n", 1609 devtoname(pip->i_dev), (u_long)ino, fs->fs_fsmnt); 1610 if (fs->fs_ronly == 0) 1611 panic("ffs_vfree: freeing free inode"); 1612 } 1613 clrbit(inosused, ino); 1614 if (ino < cgp->cg_irotor) 1615 cgp->cg_irotor = ino; 1616 cgp->cg_cs.cs_nifree++; 1617 fs->fs_cstotal.cs_nifree++; 1618 fs->fs_cs(fs, cg).cs_nifree++; 1619 if ((mode & IFMT) == IFDIR) { 1620 cgp->cg_cs.cs_ndir--; 1621 fs->fs_cstotal.cs_ndir--; 1622 fs->fs_cs(fs, cg).cs_ndir--; 1623 } 1624 fs->fs_fmod = 1; 1625 bdwrite(bp); 1626 return (0); 1627 } 1628 1629 /* 1630 * Find a block of the specified size in the specified cylinder group. 1631 * 1632 * It is a panic if a request is made to find a block if none are 1633 * available. 1634 */ 1635 static ufs_daddr_t 1636 ffs_mapsearch(fs, cgp, bpref, allocsiz) 1637 register struct fs *fs; 1638 register struct cg *cgp; 1639 ufs_daddr_t bpref; 1640 int allocsiz; 1641 { 1642 ufs_daddr_t bno; 1643 int start, len, loc, i; 1644 int blk, field, subfield, pos; 1645 u_int8_t *blksfree; 1646 1647 /* 1648 * find the fragment by searching through the free block 1649 * map for an appropriate bit pattern 1650 */ 1651 if (bpref) 1652 start = dtogd(fs, bpref) / NBBY; 1653 else 1654 start = cgp->cg_frotor / NBBY; 1655 blksfree = cg_blksfree(cgp); 1656 len = howmany(fs->fs_fpg, NBBY) - start; 1657 loc = scanc((u_int)len, (u_char *)&blksfree[start], 1658 (u_char *)fragtbl[fs->fs_frag], 1659 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 1660 if (loc == 0) { 1661 len = start + 1; 1662 start = 0; 1663 loc = scanc((u_int)len, (u_char *)&blksfree[0], 1664 (u_char *)fragtbl[fs->fs_frag], 1665 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 1666 if (loc == 0) { 1667 printf("start = %d, len = %d, fs = %s\n", 1668 start, len, fs->fs_fsmnt); 1669 panic("ffs_alloccg: map corrupted"); 1670 /* NOTREACHED */ 1671 } 1672 } 1673 bno = (start + len - loc) * NBBY; 1674 cgp->cg_frotor = bno; 1675 /* 1676 * found the byte in the map 1677 * sift through the bits to find the selected frag 1678 */ 1679 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 1680 blk = blkmap(fs, blksfree, bno); 1681 blk <<= 1; 1682 field = around[allocsiz]; 1683 subfield = inside[allocsiz]; 1684 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 1685 if ((blk & field) == subfield) 1686 return (bno + pos); 1687 field <<= 1; 1688 subfield <<= 1; 1689 } 1690 } 1691 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); 1692 panic("ffs_alloccg: block not in map"); 1693 return (-1); 1694 } 1695 1696 /* 1697 * Update the cluster map because of an allocation or free. 1698 * 1699 * Cnt == 1 means free; cnt == -1 means allocating. 1700 */ 1701 static void 1702 ffs_clusteracct(fs, cgp, blkno, cnt) 1703 struct fs *fs; 1704 struct cg *cgp; 1705 ufs_daddr_t blkno; 1706 int cnt; 1707 { 1708 int32_t *sump; 1709 int32_t *lp; 1710 u_char *freemapp, *mapp; 1711 int i, start, end, forw, back, map, bit; 1712 1713 if (fs->fs_contigsumsize <= 0) 1714 return; 1715 freemapp = cg_clustersfree(cgp); 1716 sump = cg_clustersum(cgp); 1717 /* 1718 * Allocate or clear the actual block. 1719 */ 1720 if (cnt > 0) 1721 setbit(freemapp, blkno); 1722 else 1723 clrbit(freemapp, blkno); 1724 /* 1725 * Find the size of the cluster going forward. 1726 */ 1727 start = blkno + 1; 1728 end = start + fs->fs_contigsumsize; 1729 if (end >= cgp->cg_nclusterblks) 1730 end = cgp->cg_nclusterblks; 1731 mapp = &freemapp[start / NBBY]; 1732 map = *mapp++; 1733 bit = 1 << (start % NBBY); 1734 for (i = start; i < end; i++) { 1735 if ((map & bit) == 0) 1736 break; 1737 if ((i & (NBBY - 1)) != (NBBY - 1)) { 1738 bit <<= 1; 1739 } else { 1740 map = *mapp++; 1741 bit = 1; 1742 } 1743 } 1744 forw = i - start; 1745 /* 1746 * Find the size of the cluster going backward. 1747 */ 1748 start = blkno - 1; 1749 end = start - fs->fs_contigsumsize; 1750 if (end < 0) 1751 end = -1; 1752 mapp = &freemapp[start / NBBY]; 1753 map = *mapp--; 1754 bit = 1 << (start % NBBY); 1755 for (i = start; i > end; i--) { 1756 if ((map & bit) == 0) 1757 break; 1758 if ((i & (NBBY - 1)) != 0) { 1759 bit >>= 1; 1760 } else { 1761 map = *mapp--; 1762 bit = 1 << (NBBY - 1); 1763 } 1764 } 1765 back = start - i; 1766 /* 1767 * Account for old cluster and the possibly new forward and 1768 * back clusters. 1769 */ 1770 i = back + forw + 1; 1771 if (i > fs->fs_contigsumsize) 1772 i = fs->fs_contigsumsize; 1773 sump[i] += cnt; 1774 if (back > 0) 1775 sump[back] -= cnt; 1776 if (forw > 0) 1777 sump[forw] -= cnt; 1778 /* 1779 * Update cluster summary information. 1780 */ 1781 lp = &sump[fs->fs_contigsumsize]; 1782 for (i = fs->fs_contigsumsize; i > 0; i--) 1783 if (*lp-- > 0) 1784 break; 1785 fs->fs_maxcluster[cgp->cg_cgx] = i; 1786 } 1787 1788 /* 1789 * Fserr prints the name of a file system with an error diagnostic. 1790 * 1791 * The form of the error message is: 1792 * fs: error message 1793 */ 1794 static void 1795 ffs_fserr(fs, uid, cp) 1796 struct fs *fs; 1797 u_int uid; 1798 char *cp; 1799 { 1800 struct proc *p = curproc; /* XXX */ 1801 1802 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1, 1803 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp); 1804 } 1805