ffs_alloc.c revision 37555
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 * $Id: ffs_alloc.c,v 1.49 1998/03/30 09:55:50 phk 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/proc.h> 43#include <sys/vnode.h> 44#include <sys/mount.h> 45#include <sys/kernel.h> 46#include <sys/sysctl.h> 47#include <sys/syslog.h> 48 49#include <ufs/ufs/quota.h> 50#include <ufs/ufs/inode.h> 51#include <ufs/ufs/ufsmount.h> 52 53#include <ufs/ffs/fs.h> 54#include <ufs/ffs/ffs_extern.h> 55 56typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref, 57 int size)); 58 59static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int)); 60static ufs_daddr_t 61 ffs_alloccgblk __P((struct inode *, struct buf *, ufs_daddr_t)); 62#ifdef DIAGNOSTIC 63static int ffs_checkblk __P((struct inode *, ufs_daddr_t, long)); 64#endif 65static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t, 66 int)); 67#ifdef notyet 68static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t, 69 int)); 70#endif 71static ino_t ffs_dirpref __P((struct fs *)); 72static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int)); 73static void ffs_fserr __P((struct fs *, u_int, char *)); 74static u_long ffs_hashalloc 75 __P((struct inode *, int, long, int, allocfcn_t *)); 76static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int)); 77static 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 */ 99int 100ffs_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 = 0x%lx, bsize = %ld, size = %d, fs = %s\n", 119 (u_long)ip->i_dev, (long)fs->fs_bsize, size, fs->fs_fsmnt); 120 panic("ffs_alloc: bad size"); 121 } 122 if (cred == NOCRED) 123 panic("ffs_alloc: missing credential"); 124#endif /* DIAGNOSTIC */ 125 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) 126 goto nospace; 127 if (cred->cr_uid != 0 && 128 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) 129 goto nospace; 130#ifdef QUOTA 131 error = chkdq(ip, (long)btodb(size), cred, 0); 132 if (error) 133 return (error); 134#endif 135 if (bpref >= fs->fs_size) 136 bpref = 0; 137 if (bpref == 0) 138 cg = ino_to_cg(fs, ip->i_number); 139 else 140 cg = dtog(fs, bpref); 141 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size, 142 ffs_alloccg); 143 if (bno > 0) { 144 ip->i_blocks += btodb(size); 145 ip->i_flag |= IN_CHANGE | IN_UPDATE; 146 *bnp = bno; 147 return (0); 148 } 149#ifdef QUOTA 150 /* 151 * Restore user's disk quota because allocation failed. 152 */ 153 (void) chkdq(ip, (long)-btodb(size), cred, FORCE); 154#endif 155nospace: 156 ffs_fserr(fs, cred->cr_uid, "file system full"); 157 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 158 return (ENOSPC); 159} 160 161/* 162 * Reallocate a fragment to a bigger size 163 * 164 * The number and size of the old block is given, and a preference 165 * and new size is also specified. The allocator attempts to extend 166 * the original block. Failing that, the regular block allocator is 167 * invoked to get an appropriate block. 168 */ 169int 170ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp) 171 register struct inode *ip; 172 ufs_daddr_t lbprev; 173 ufs_daddr_t bpref; 174 int osize, nsize; 175 struct ucred *cred; 176 struct buf **bpp; 177{ 178 register struct fs *fs; 179 struct buf *bp; 180 int cg, request, error; 181 ufs_daddr_t bprev, bno; 182 183 *bpp = 0; 184 fs = ip->i_fs; 185#ifdef DIAGNOSTIC 186 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || 187 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { 188 printf( 189 "dev = 0x%lx, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", 190 (u_long)ip->i_dev, (long)fs->fs_bsize, osize, 191 nsize, fs->fs_fsmnt); 192 panic("ffs_realloccg: bad size"); 193 } 194 if (cred == NOCRED) 195 panic("ffs_realloccg: missing credential"); 196#endif /* DIAGNOSTIC */ 197 if (cred->cr_uid != 0 && 198 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) 199 goto nospace; 200 if ((bprev = ip->i_db[lbprev]) == 0) { 201 printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n", 202 (u_long)ip->i_dev, (long)fs->fs_bsize, (long)bprev, 203 fs->fs_fsmnt); 204 panic("ffs_realloccg: bad bprev"); 205 } 206 /* 207 * Allocate the extra space in the buffer. 208 */ 209 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp); 210 if (error) { 211 brelse(bp); 212 return (error); 213 } 214 215 if( bp->b_blkno == bp->b_lblkno) { 216 if( lbprev >= NDADDR) 217 panic("ffs_realloccg: lbprev out of range"); 218 bp->b_blkno = fsbtodb(fs, bprev); 219 } 220 221#ifdef QUOTA 222 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0); 223 if (error) { 224 brelse(bp); 225 return (error); 226 } 227#endif 228 /* 229 * Check for extension in the existing location. 230 */ 231 cg = dtog(fs, bprev); 232 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize); 233 if (bno) { 234 if (bp->b_blkno != fsbtodb(fs, bno)) 235 panic("ffs_realloccg: bad blockno"); 236 ip->i_blocks += btodb(nsize - osize); 237 ip->i_flag |= IN_CHANGE | IN_UPDATE; 238 allocbuf(bp, nsize); 239 bp->b_flags |= B_DONE; 240 bzero((char *)bp->b_data + osize, (u_int)nsize - osize); 241 *bpp = bp; 242 return (0); 243 } 244 /* 245 * Allocate a new disk location. 246 */ 247 if (bpref >= fs->fs_size) 248 bpref = 0; 249 switch ((int)fs->fs_optim) { 250 case FS_OPTSPACE: 251 /* 252 * Allocate an exact sized fragment. Although this makes 253 * best use of space, we will waste time relocating it if 254 * the file continues to grow. If the fragmentation is 255 * less than half of the minimum free reserve, we choose 256 * to begin optimizing for time. 257 */ 258 request = nsize; 259 if (fs->fs_minfree <= 5 || 260 fs->fs_cstotal.cs_nffree > 261 fs->fs_dsize * fs->fs_minfree / (2 * 100)) 262 break; 263 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", 264 fs->fs_fsmnt); 265 fs->fs_optim = FS_OPTTIME; 266 break; 267 case FS_OPTTIME: 268 /* 269 * At this point we have discovered a file that is trying to 270 * grow a small fragment to a larger fragment. To save time, 271 * we allocate a full sized block, then free the unused portion. 272 * If the file continues to grow, the `ffs_fragextend' call 273 * above will be able to grow it in place without further 274 * copying. If aberrant programs cause disk fragmentation to 275 * grow within 2% of the free reserve, we choose to begin 276 * optimizing for space. 277 */ 278 request = fs->fs_bsize; 279 if (fs->fs_cstotal.cs_nffree < 280 fs->fs_dsize * (fs->fs_minfree - 2) / 100) 281 break; 282 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", 283 fs->fs_fsmnt); 284 fs->fs_optim = FS_OPTSPACE; 285 break; 286 default: 287 printf("dev = 0x%lx, optim = %ld, fs = %s\n", 288 (u_long)ip->i_dev, (long)fs->fs_optim, fs->fs_fsmnt); 289 panic("ffs_realloccg: bad optim"); 290 /* NOTREACHED */ 291 } 292 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request, 293 ffs_alloccg); 294 if (bno > 0) { 295 bp->b_blkno = fsbtodb(fs, bno); 296 if (!DOINGSOFTDEP(ITOV(ip))) 297 ffs_blkfree(ip, bprev, (long)osize); 298 if (nsize < request) 299 ffs_blkfree(ip, bno + numfrags(fs, nsize), 300 (long)(request - nsize)); 301 ip->i_blocks += btodb(nsize - osize); 302 ip->i_flag |= IN_CHANGE | IN_UPDATE; 303 allocbuf(bp, nsize); 304 bp->b_flags |= B_DONE; 305 bzero((char *)bp->b_data + osize, (u_int)nsize - osize); 306 *bpp = bp; 307 return (0); 308 } 309#ifdef QUOTA 310 /* 311 * Restore user's disk quota because allocation failed. 312 */ 313 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE); 314#endif 315 brelse(bp); 316nospace: 317 /* 318 * no space available 319 */ 320 ffs_fserr(fs, cred->cr_uid, "file system full"); 321 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 322 return (ENOSPC); 323} 324 325#ifdef notyet 326/* 327 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 328 * 329 * The vnode and an array of buffer pointers for a range of sequential 330 * logical blocks to be made contiguous is given. The allocator attempts 331 * to find a range of sequential blocks starting as close as possible to 332 * an fs_rotdelay offset from the end of the allocation for the logical 333 * block immediately preceeding the current range. If successful, the 334 * physical block numbers in the buffer pointers and in the inode are 335 * changed to reflect the new allocation. If unsuccessful, the allocation 336 * is left unchanged. The success in doing the reallocation is returned. 337 * Note that the error return is not reflected back to the user. Rather 338 * the previous block allocation will be used. 339 */ 340static int doasyncfree = 1; 341SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, ""); 342 343static int doreallocblks = 1; 344SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); 345 346static int prtrealloc = 0; 347#endif 348 349int 350ffs_reallocblks(ap) 351 struct vop_reallocblks_args /* { 352 struct vnode *a_vp; 353 struct cluster_save *a_buflist; 354 } */ *ap; 355{ 356#if !defined (not_yes) 357 return (ENOSPC); 358#else 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 struct timeval tv; 369 370 if (doreallocblks == 0) 371 return (ENOSPC); 372 vp = ap->a_vp; 373 ip = VTOI(vp); 374 fs = ip->i_fs; 375 if (fs->fs_contigsumsize <= 0) 376 return (ENOSPC); 377 buflist = ap->a_buflist; 378 len = buflist->bs_nchildren; 379 start_lbn = buflist->bs_children[0]->b_lblkno; 380 end_lbn = start_lbn + len - 1; 381#ifdef DIAGNOSTIC 382 for (i = 0; i < len; i++) 383 if (!ffs_checkblk(ip, 384 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 385 panic("ffs_reallocblks: unallocated block 1"); 386 for (i = 1; i < len; i++) 387 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 388 panic("ffs_reallocblks: non-logical cluster"); 389 blkno = buflist->bs_children[0]->b_blkno; 390 ssize = fsbtodb(fs, fs->fs_frag); 391 for (i = 1; i < len - 1; i++) 392 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) 393 panic("ffs_reallocblks: non-physical cluster %d", i); 394#endif 395 /* 396 * If the latest allocation is in a new cylinder group, assume that 397 * the filesystem has decided to move and do not force it back to 398 * the previous cylinder group. 399 */ 400 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 401 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 402 return (ENOSPC); 403 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 404 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 405 return (ENOSPC); 406 /* 407 * Get the starting offset and block map for the first block. 408 */ 409 if (start_lvl == 0) { 410 sbap = &ip->i_db[0]; 411 soff = start_lbn; 412 } else { 413 idp = &start_ap[start_lvl - 1]; 414 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { 415 brelse(sbp); 416 return (ENOSPC); 417 } 418 sbap = (ufs_daddr_t *)sbp->b_data; 419 soff = idp->in_off; 420 } 421 /* 422 * Find the preferred location for the cluster. 423 */ 424 pref = ffs_blkpref(ip, start_lbn, soff, sbap); 425 /* 426 * If the block range spans two block maps, get the second map. 427 */ 428 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 429 ssize = len; 430 } else { 431#ifdef DIAGNOSTIC 432 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn) 433 panic("ffs_reallocblk: start == end"); 434#endif 435 ssize = len - (idp->in_off + 1); 436 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) 437 goto fail; 438 ebap = (ufs_daddr_t *)ebp->b_data; 439 } 440 /* 441 * Search the block map looking for an allocation of the desired size. 442 */ 443 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref, 444 len, ffs_clusteralloc)) == 0) 445 goto fail; 446 /* 447 * We have found a new contiguous block. 448 * 449 * First we have to replace the old block pointers with the new 450 * block pointers in the inode and indirect blocks associated 451 * with the file. 452 */ 453#ifdef DEBUG 454 if (prtrealloc) 455 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number, 456 start_lbn, end_lbn); 457#endif 458 blkno = newblk; 459 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { 460 if (i == ssize) { 461 bap = ebap; 462 soff = -i; 463 } 464#ifdef DIAGNOSTIC 465 if (!ffs_checkblk(ip, 466 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 467 panic("ffs_reallocblks: unallocated block 2"); 468 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) 469 panic("ffs_reallocblks: alloc mismatch"); 470#endif 471#ifdef DEBUG 472 if (prtrealloc) 473 printf(" %d,", *bap); 474#endif 475 if (DOINGSOFTDEP(vp)) { 476 if (sbap == &ip->i_db[0] && i < ssize) 477 softdep_setup_allocdirect(ip, start_lbn + i, 478 blkno, *bap, fs->fs_bsize, fs->fs_bsize, 479 buflist->bs_children[i]); 480 else 481 softdep_setup_allocindir_page(ip, start_lbn + i, 482 i < ssize ? sbp : ebp, soff + i, blkno, 483 *bap, buflist->bs_children[i]); 484 } 485 *bap++ = blkno; 486 } 487 /* 488 * Next we must write out the modified inode and indirect blocks. 489 * For strict correctness, the writes should be synchronous since 490 * the old block values may have been written to disk. In practise 491 * they are almost never written, but if we are concerned about 492 * strict correctness, the `doasyncfree' flag should be set to zero. 493 * 494 * The test on `doasyncfree' should be changed to test a flag 495 * that shows whether the associated buffers and inodes have 496 * been written. The flag should be set when the cluster is 497 * started and cleared whenever the buffer or inode is flushed. 498 * We can then check below to see if it is set, and do the 499 * synchronous write only when it has been cleared. 500 */ 501 if (sbap != &ip->i_db[0]) { 502 if (doasyncfree) 503 bdwrite(sbp); 504 else 505 bwrite(sbp); 506 } else { 507 ip->i_flag |= IN_CHANGE | IN_UPDATE; 508 if (!doasyncfree) { 509 gettime(&tv); 510 UFS_UPDATE(vp, &tv, &tv, 1); 511 } 512 } 513 if (ssize < len) 514 if (doasyncfree) 515 bdwrite(ebp); 516 else 517 bwrite(ebp); 518 /* 519 * Last, free the old blocks and assign the new blocks to the buffers. 520 */ 521#ifdef DEBUG 522 if (prtrealloc) 523 printf("\n\tnew:"); 524#endif 525 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { 526 if (!DOINGSOFTDEP(vp)) 527 ffs_blkfree(ip, 528 dbtofsb(fs, buflist->bs_children[i]->b_blkno), 529 fs->fs_bsize); 530 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 531#ifdef DEBUG 532 if (!ffs_checkblk(ip, 533 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 534 panic("ffs_reallocblks: unallocated block 3"); 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 547fail: 548 if (ssize < len) 549 brelse(ebp); 550 if (sbap != &ip->i_db[0]) 551 brelse(sbp); 552 return (ENOSPC); 553#endif 554} 555 556/* 557 * Allocate an inode in the file system. 558 * 559 * If allocating a directory, use ffs_dirpref to select the inode. 560 * If allocating in a directory, the following hierarchy is followed: 561 * 1) allocate the preferred inode. 562 * 2) allocate an inode in the same cylinder group. 563 * 3) quadradically rehash into other cylinder groups, until an 564 * available inode is located. 565 * If no inode preference is given the following heirarchy is used 566 * to allocate an inode: 567 * 1) allocate an inode in cylinder group 0. 568 * 2) quadradically rehash into other cylinder groups, until an 569 * available inode is located. 570 */ 571int 572ffs_valloc(pvp, mode, cred, vpp) 573 struct vnode *pvp; 574 int mode; 575 struct ucred *cred; 576 struct vnode **vpp; 577{ 578 register struct inode *pip; 579 register struct fs *fs; 580 register struct inode *ip; 581 ino_t ino, ipref; 582 int cg, error; 583 584 *vpp = NULL; 585 pip = VTOI(pvp); 586 fs = pip->i_fs; 587 if (fs->fs_cstotal.cs_nifree == 0) 588 goto noinodes; 589 590 if ((mode & IFMT) == IFDIR) 591 ipref = ffs_dirpref(fs); 592 else 593 ipref = pip->i_number; 594 if (ipref >= fs->fs_ncg * fs->fs_ipg) 595 ipref = 0; 596 cg = ino_to_cg(fs, ipref); 597 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode, 598 (allocfcn_t *)ffs_nodealloccg); 599 if (ino == 0) 600 goto noinodes; 601 error = VFS_VGET(pvp->v_mount, ino, vpp); 602 if (error) { 603 UFS_VFREE(pvp, ino, mode); 604 return (error); 605 } 606 ip = VTOI(*vpp); 607 if (ip->i_mode) { 608 printf("mode = 0%o, inum = %lu, fs = %s\n", 609 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt); 610 panic("ffs_valloc: dup alloc"); 611 } 612 if (ip->i_blocks) { /* XXX */ 613 printf("free inode %s/%lu had %ld blocks\n", 614 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks); 615 ip->i_blocks = 0; 616 } 617 ip->i_flags = 0; 618 /* 619 * Set up a new generation number for this inode. 620 */ 621 if (ip->i_gen == 0 || ++ip->i_gen == 0) 622 ip->i_gen = random() / 2 + 1; 623 return (0); 624noinodes: 625 ffs_fserr(fs, cred->cr_uid, "out of inodes"); 626 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); 627 return (ENOSPC); 628} 629 630/* 631 * Find a cylinder to place a directory. 632 * 633 * The policy implemented by this algorithm is to select from 634 * among those cylinder groups with above the average number of 635 * free inodes, the one with the smallest number of directories. 636 */ 637static ino_t 638ffs_dirpref(fs) 639 register struct fs *fs; 640{ 641 int cg, minndir, mincg, avgifree; 642 643 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 644 minndir = fs->fs_ipg; 645 mincg = 0; 646 for (cg = 0; cg < fs->fs_ncg; cg++) 647 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 648 fs->fs_cs(fs, cg).cs_nifree >= avgifree) { 649 mincg = cg; 650 minndir = fs->fs_cs(fs, cg).cs_ndir; 651 } 652 return ((ino_t)(fs->fs_ipg * mincg)); 653} 654 655/* 656 * Select the desired position for the next block in a file. The file is 657 * logically divided into sections. The first section is composed of the 658 * direct blocks. Each additional section contains fs_maxbpg blocks. 659 * 660 * If no blocks have been allocated in the first section, the policy is to 661 * request a block in the same cylinder group as the inode that describes 662 * the file. If no blocks have been allocated in any other section, the 663 * policy is to place the section in a cylinder group with a greater than 664 * average number of free blocks. An appropriate cylinder group is found 665 * by using a rotor that sweeps the cylinder groups. When a new group of 666 * blocks is needed, the sweep begins in the cylinder group following the 667 * cylinder group from which the previous allocation was made. The sweep 668 * continues until a cylinder group with greater than the average number 669 * of free blocks is found. If the allocation is for the first block in an 670 * indirect block, the information on the previous allocation is unavailable; 671 * here a best guess is made based upon the logical block number being 672 * allocated. 673 * 674 * If a section is already partially allocated, the policy is to 675 * contiguously allocate fs_maxcontig blocks. The end of one of these 676 * contiguous blocks and the beginning of the next is physically separated 677 * so that the disk head will be in transit between them for at least 678 * fs_rotdelay milliseconds. This is to allow time for the processor to 679 * schedule another I/O transfer. 680 */ 681ufs_daddr_t 682ffs_blkpref(ip, lbn, indx, bap) 683 struct inode *ip; 684 ufs_daddr_t lbn; 685 int indx; 686 ufs_daddr_t *bap; 687{ 688 register struct fs *fs; 689 register int cg; 690 int avgbfree, startcg; 691 ufs_daddr_t nextblk; 692 693 fs = ip->i_fs; 694 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 695 if (lbn < NDADDR) { 696 cg = ino_to_cg(fs, ip->i_number); 697 return (fs->fs_fpg * cg + fs->fs_frag); 698 } 699 /* 700 * Find a cylinder with greater than average number of 701 * unused data blocks. 702 */ 703 if (indx == 0 || bap[indx - 1] == 0) 704 startcg = 705 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 706 else 707 startcg = dtog(fs, bap[indx - 1]) + 1; 708 startcg %= fs->fs_ncg; 709 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 710 for (cg = startcg; cg < fs->fs_ncg; cg++) 711 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 712 fs->fs_cgrotor = cg; 713 return (fs->fs_fpg * cg + fs->fs_frag); 714 } 715 for (cg = 0; cg <= startcg; cg++) 716 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 717 fs->fs_cgrotor = cg; 718 return (fs->fs_fpg * cg + fs->fs_frag); 719 } 720 return (0); 721 } 722 /* 723 * One or more previous blocks have been laid out. If less 724 * than fs_maxcontig previous blocks are contiguous, the 725 * next block is requested contiguously, otherwise it is 726 * requested rotationally delayed by fs_rotdelay milliseconds. 727 */ 728 nextblk = bap[indx - 1] + fs->fs_frag; 729 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig || 730 bap[indx - fs->fs_maxcontig] + 731 blkstofrags(fs, fs->fs_maxcontig) != nextblk) 732 return (nextblk); 733 /* 734 * Here we convert ms of delay to frags as: 735 * (frags) = (ms) * (rev/sec) * (sect/rev) / 736 * ((sect/frag) * (ms/sec)) 737 * then round up to the next block. 738 */ 739 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect / 740 (NSPF(fs) * 1000), fs->fs_frag); 741 return (nextblk); 742} 743 744/* 745 * Implement the cylinder overflow algorithm. 746 * 747 * The policy implemented by this algorithm is: 748 * 1) allocate the block in its requested cylinder group. 749 * 2) quadradically rehash on the cylinder group number. 750 * 3) brute force search for a free block. 751 */ 752/*VARARGS5*/ 753static u_long 754ffs_hashalloc(ip, cg, pref, size, allocator) 755 struct inode *ip; 756 int cg; 757 long pref; 758 int size; /* size for data blocks, mode for inodes */ 759 allocfcn_t *allocator; 760{ 761 register struct fs *fs; 762 long result; /* XXX why not same type as we return? */ 763 int i, icg = cg; 764 765 fs = ip->i_fs; 766 /* 767 * 1: preferred cylinder group 768 */ 769 result = (*allocator)(ip, cg, pref, size); 770 if (result) 771 return (result); 772 /* 773 * 2: quadratic rehash 774 */ 775 for (i = 1; i < fs->fs_ncg; i *= 2) { 776 cg += i; 777 if (cg >= fs->fs_ncg) 778 cg -= fs->fs_ncg; 779 result = (*allocator)(ip, cg, 0, size); 780 if (result) 781 return (result); 782 } 783 /* 784 * 3: brute force search 785 * Note that we start at i == 2, since 0 was checked initially, 786 * and 1 is always checked in the quadratic rehash. 787 */ 788 cg = (icg + 2) % fs->fs_ncg; 789 for (i = 2; i < fs->fs_ncg; i++) { 790 result = (*allocator)(ip, cg, 0, size); 791 if (result) 792 return (result); 793 cg++; 794 if (cg == fs->fs_ncg) 795 cg = 0; 796 } 797 return (0); 798} 799 800/* 801 * Determine whether a fragment can be extended. 802 * 803 * Check to see if the necessary fragments are available, and 804 * if they are, allocate them. 805 */ 806static ufs_daddr_t 807ffs_fragextend(ip, cg, bprev, osize, nsize) 808 struct inode *ip; 809 int cg; 810 long bprev; 811 int osize, nsize; 812{ 813 register struct fs *fs; 814 register struct cg *cgp; 815 struct buf *bp; 816 long bno; 817 int frags, bbase; 818 int i, error; 819 820 fs = ip->i_fs; 821 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 822 return (0); 823 frags = numfrags(fs, nsize); 824 bbase = fragnum(fs, bprev); 825 if (bbase > fragnum(fs, (bprev + frags - 1))) { 826 /* cannot extend across a block boundary */ 827 return (0); 828 } 829 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 830 (int)fs->fs_cgsize, NOCRED, &bp); 831 if (error) { 832 brelse(bp); 833 return (0); 834 } 835 cgp = (struct cg *)bp->b_data; 836 if (!cg_chkmagic(cgp)) { 837 brelse(bp); 838 return (0); 839 } 840 cgp->cg_time = time_second; 841 bno = dtogd(fs, bprev); 842 for (i = numfrags(fs, osize); i < frags; i++) 843 if (isclr(cg_blksfree(cgp), bno + i)) { 844 brelse(bp); 845 return (0); 846 } 847 /* 848 * the current fragment can be extended 849 * deduct the count on fragment being extended into 850 * increase the count on the remaining fragment (if any) 851 * allocate the extended piece 852 */ 853 for (i = frags; i < fs->fs_frag - bbase; i++) 854 if (isclr(cg_blksfree(cgp), bno + i)) 855 break; 856 cgp->cg_frsum[i - numfrags(fs, osize)]--; 857 if (i != frags) 858 cgp->cg_frsum[i - frags]++; 859 for (i = numfrags(fs, osize); i < frags; i++) { 860 clrbit(cg_blksfree(cgp), bno + i); 861 cgp->cg_cs.cs_nffree--; 862 fs->fs_cstotal.cs_nffree--; 863 fs->fs_cs(fs, cg).cs_nffree--; 864 } 865 fs->fs_fmod = 1; 866 if (DOINGSOFTDEP(ITOV(ip))) 867 softdep_setup_blkmapdep(bp, fs, bprev); 868 bdwrite(bp); 869 return (bprev); 870} 871 872/* 873 * Determine whether a block can be allocated. 874 * 875 * Check to see if a block of the appropriate size is available, 876 * and if it is, allocate it. 877 */ 878static ufs_daddr_t 879ffs_alloccg(ip, cg, bpref, size) 880 struct inode *ip; 881 int cg; 882 ufs_daddr_t bpref; 883 int size; 884{ 885 register struct fs *fs; 886 register struct cg *cgp; 887 struct buf *bp; 888 register int i; 889 ufs_daddr_t bno, blkno; 890 int allocsiz, error, frags; 891 892 fs = ip->i_fs; 893 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 894 return (0); 895 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 896 (int)fs->fs_cgsize, NOCRED, &bp); 897 if (error) { 898 brelse(bp); 899 return (0); 900 } 901 cgp = (struct cg *)bp->b_data; 902 if (!cg_chkmagic(cgp) || 903 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) { 904 brelse(bp); 905 return (0); 906 } 907 cgp->cg_time = time_second; 908 if (size == fs->fs_bsize) { 909 bno = ffs_alloccgblk(ip, bp, bpref); 910 bdwrite(bp); 911 return (bno); 912 } 913 /* 914 * check to see if any fragments are already available 915 * allocsiz is the size which will be allocated, hacking 916 * it down to a smaller size if necessary 917 */ 918 frags = numfrags(fs, size); 919 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 920 if (cgp->cg_frsum[allocsiz] != 0) 921 break; 922 if (allocsiz == fs->fs_frag) { 923 /* 924 * no fragments were available, so a block will be 925 * allocated, and hacked up 926 */ 927 if (cgp->cg_cs.cs_nbfree == 0) { 928 brelse(bp); 929 return (0); 930 } 931 bno = ffs_alloccgblk(ip, bp, bpref); 932 bpref = dtogd(fs, bno); 933 for (i = frags; i < fs->fs_frag; i++) 934 setbit(cg_blksfree(cgp), bpref + i); 935 i = fs->fs_frag - frags; 936 cgp->cg_cs.cs_nffree += i; 937 fs->fs_cstotal.cs_nffree += i; 938 fs->fs_cs(fs, cg).cs_nffree += i; 939 fs->fs_fmod = 1; 940 cgp->cg_frsum[i]++; 941 bdwrite(bp); 942 return (bno); 943 } 944 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 945 if (bno < 0) { 946 brelse(bp); 947 return (0); 948 } 949 for (i = 0; i < frags; i++) 950 clrbit(cg_blksfree(cgp), bno + i); 951 cgp->cg_cs.cs_nffree -= frags; 952 fs->fs_cstotal.cs_nffree -= frags; 953 fs->fs_cs(fs, cg).cs_nffree -= frags; 954 fs->fs_fmod = 1; 955 cgp->cg_frsum[allocsiz]--; 956 if (frags != allocsiz) 957 cgp->cg_frsum[allocsiz - frags]++; 958 blkno = cg * fs->fs_fpg + bno; 959 if (DOINGSOFTDEP(ITOV(ip))) 960 softdep_setup_blkmapdep(bp, fs, blkno); 961 bdwrite(bp); 962 return ((u_long)blkno); 963} 964 965/* 966 * Allocate a block in a cylinder group. 967 * 968 * This algorithm implements the following policy: 969 * 1) allocate the requested block. 970 * 2) allocate a rotationally optimal block in the same cylinder. 971 * 3) allocate the next available block on the block rotor for the 972 * specified cylinder group. 973 * Note that this routine only allocates fs_bsize blocks; these 974 * blocks may be fragmented by the routine that allocates them. 975 */ 976static ufs_daddr_t 977ffs_alloccgblk(ip, bp, bpref) 978 struct inode *ip; 979 struct buf *bp; 980 ufs_daddr_t bpref; 981{ 982 struct fs *fs; 983 struct cg *cgp; 984 ufs_daddr_t bno, blkno; 985 int cylno, pos, delta; 986 short *cylbp; 987 register int i; 988 989 fs = ip->i_fs; 990 cgp = (struct cg *)bp->b_data; 991 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) { 992 bpref = cgp->cg_rotor; 993 goto norot; 994 } 995 bpref = blknum(fs, bpref); 996 bpref = dtogd(fs, bpref); 997 /* 998 * if the requested block is available, use it 999 */ 1000 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) { 1001 bno = bpref; 1002 goto gotit; 1003 } 1004 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) { 1005 /* 1006 * Block layout information is not available. 1007 * Leaving bpref unchanged means we take the 1008 * next available free block following the one 1009 * we just allocated. Hopefully this will at 1010 * least hit a track cache on drives of unknown 1011 * geometry (e.g. SCSI). 1012 */ 1013 goto norot; 1014 } 1015 /* 1016 * check for a block available on the same cylinder 1017 */ 1018 cylno = cbtocylno(fs, bpref); 1019 if (cg_blktot(cgp)[cylno] == 0) 1020 goto norot; 1021 /* 1022 * check the summary information to see if a block is 1023 * available in the requested cylinder starting at the 1024 * requested rotational position and proceeding around. 1025 */ 1026 cylbp = cg_blks(fs, cgp, cylno); 1027 pos = cbtorpos(fs, bpref); 1028 for (i = pos; i < fs->fs_nrpos; i++) 1029 if (cylbp[i] > 0) 1030 break; 1031 if (i == fs->fs_nrpos) 1032 for (i = 0; i < pos; i++) 1033 if (cylbp[i] > 0) 1034 break; 1035 if (cylbp[i] > 0) { 1036 /* 1037 * found a rotational position, now find the actual 1038 * block. A panic if none is actually there. 1039 */ 1040 pos = cylno % fs->fs_cpc; 1041 bno = (cylno - pos) * fs->fs_spc / NSPB(fs); 1042 if (fs_postbl(fs, pos)[i] == -1) { 1043 printf("pos = %d, i = %d, fs = %s\n", 1044 pos, i, fs->fs_fsmnt); 1045 panic("ffs_alloccgblk: cyl groups corrupted"); 1046 } 1047 for (i = fs_postbl(fs, pos)[i];; ) { 1048 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) { 1049 bno = blkstofrags(fs, (bno + i)); 1050 goto gotit; 1051 } 1052 delta = fs_rotbl(fs)[i]; 1053 if (delta <= 0 || 1054 delta + i > fragstoblks(fs, fs->fs_fpg)) 1055 break; 1056 i += delta; 1057 } 1058 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt); 1059 panic("ffs_alloccgblk: can't find blk in cyl"); 1060 } 1061norot: 1062 /* 1063 * no blocks in the requested cylinder, so take next 1064 * available one in this cylinder group. 1065 */ 1066 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1067 if (bno < 0) 1068 return (0); 1069 cgp->cg_rotor = bno; 1070gotit: 1071 blkno = fragstoblks(fs, bno); 1072 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno); 1073 ffs_clusteracct(fs, cgp, blkno, -1); 1074 cgp->cg_cs.cs_nbfree--; 1075 fs->fs_cstotal.cs_nbfree--; 1076 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; 1077 cylno = cbtocylno(fs, bno); 1078 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--; 1079 cg_blktot(cgp)[cylno]--; 1080 fs->fs_fmod = 1; 1081 blkno = cgp->cg_cgx * fs->fs_fpg + bno; 1082 if (DOINGSOFTDEP(ITOV(ip))) 1083 softdep_setup_blkmapdep(bp, fs, blkno); 1084 return (blkno); 1085} 1086 1087#ifdef notyet 1088/* 1089 * Determine whether a cluster can be allocated. 1090 * 1091 * We do not currently check for optimal rotational layout if there 1092 * are multiple choices in the same cylinder group. Instead we just 1093 * take the first one that we find following bpref. 1094 */ 1095static ufs_daddr_t 1096ffs_clusteralloc(ip, cg, bpref, len) 1097 struct inode *ip; 1098 int cg; 1099 ufs_daddr_t bpref; 1100 int len; 1101{ 1102 register struct fs *fs; 1103 register struct cg *cgp; 1104 struct buf *bp; 1105 int i, got, run, bno, bit, map; 1106 u_char *mapp; 1107 int32_t *lp; 1108 1109 fs = ip->i_fs; 1110 if (fs->fs_maxcluster[cg] < len) 1111 return (NULL); 1112 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize, 1113 NOCRED, &bp)) 1114 goto fail; 1115 cgp = (struct cg *)bp->b_data; 1116 if (!cg_chkmagic(cgp)) 1117 goto fail; 1118 /* 1119 * Check to see if a cluster of the needed size (or bigger) is 1120 * available in this cylinder group. 1121 */ 1122 lp = &cg_clustersum(cgp)[len]; 1123 for (i = len; i <= fs->fs_contigsumsize; i++) 1124 if (*lp++ > 0) 1125 break; 1126 if (i > fs->fs_contigsumsize) { 1127 /* 1128 * This is the first time looking for a cluster in this 1129 * cylinder group. Update the cluster summary information 1130 * to reflect the true maximum sized cluster so that 1131 * future cluster allocation requests can avoid reading 1132 * the cylinder group map only to find no clusters. 1133 */ 1134 lp = &cg_clustersum(cgp)[len - 1]; 1135 for (i = len - 1; i > 0; i--) 1136 if (*lp-- > 0) 1137 break; 1138 fs->fs_maxcluster[cg] = i; 1139 goto fail; 1140 } 1141 /* 1142 * Search the cluster map to find a big enough cluster. 1143 * We take the first one that we find, even if it is larger 1144 * than we need as we prefer to get one close to the previous 1145 * block allocation. We do not search before the current 1146 * preference point as we do not want to allocate a block 1147 * that is allocated before the previous one (as we will 1148 * then have to wait for another pass of the elevator 1149 * algorithm before it will be read). We prefer to fail and 1150 * be recalled to try an allocation in the next cylinder group. 1151 */ 1152 if (dtog(fs, bpref) != cg) 1153 bpref = 0; 1154 else 1155 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref))); 1156 mapp = &cg_clustersfree(cgp)[bpref / NBBY]; 1157 map = *mapp++; 1158 bit = 1 << (bpref % NBBY); 1159 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { 1160 if ((map & bit) == 0) { 1161 run = 0; 1162 } else { 1163 run++; 1164 if (run == len) 1165 break; 1166 } 1167 if ((got & (NBBY - 1)) != (NBBY - 1)) { 1168 bit <<= 1; 1169 } else { 1170 map = *mapp++; 1171 bit = 1; 1172 } 1173 } 1174 if (got >= cgp->cg_nclusterblks) 1175 goto fail; 1176 /* 1177 * Allocate the cluster that we have found. 1178 */ 1179 for (i = 1; i <= len; i++) 1180 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i)) 1181 panic("ffs_clusteralloc: map mismatch"); 1182 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1); 1183 if (dtog(fs, bno) != cg) 1184 panic("ffs_clusteralloc: allocated out of group"); 1185 len = blkstofrags(fs, len); 1186 for (i = 0; i < len; i += fs->fs_frag) 1187 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i) 1188 panic("ffs_clusteralloc: lost block"); 1189 bdwrite(bp); 1190 return (bno); 1191 1192fail: 1193 brelse(bp); 1194 return (0); 1195} 1196#endif 1197 1198/* 1199 * Determine whether an inode can be allocated. 1200 * 1201 * Check to see if an inode is available, and if it is, 1202 * allocate it using the following policy: 1203 * 1) allocate the requested inode. 1204 * 2) allocate the next available inode after the requested 1205 * inode in the specified cylinder group. 1206 */ 1207static ino_t 1208ffs_nodealloccg(ip, cg, ipref, mode) 1209 struct inode *ip; 1210 int cg; 1211 ufs_daddr_t ipref; 1212 int mode; 1213{ 1214 register struct fs *fs; 1215 register struct cg *cgp; 1216 struct buf *bp; 1217 int error, start, len, loc, map, i; 1218 1219 fs = ip->i_fs; 1220 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1221 return (0); 1222 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1223 (int)fs->fs_cgsize, NOCRED, &bp); 1224 if (error) { 1225 brelse(bp); 1226 return (0); 1227 } 1228 cgp = (struct cg *)bp->b_data; 1229 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) { 1230 brelse(bp); 1231 return (0); 1232 } 1233 cgp->cg_time = time_second; 1234 if (ipref) { 1235 ipref %= fs->fs_ipg; 1236 if (isclr(cg_inosused(cgp), ipref)) 1237 goto gotit; 1238 } 1239 start = cgp->cg_irotor / NBBY; 1240 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); 1241 loc = skpc(0xff, len, &cg_inosused(cgp)[start]); 1242 if (loc == 0) { 1243 len = start + 1; 1244 start = 0; 1245 loc = skpc(0xff, len, &cg_inosused(cgp)[0]); 1246 if (loc == 0) { 1247 printf("cg = %d, irotor = %ld, fs = %s\n", 1248 cg, (long)cgp->cg_irotor, fs->fs_fsmnt); 1249 panic("ffs_nodealloccg: map corrupted"); 1250 /* NOTREACHED */ 1251 } 1252 } 1253 i = start + len - loc; 1254 map = cg_inosused(cgp)[i]; 1255 ipref = i * NBBY; 1256 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) { 1257 if ((map & i) == 0) { 1258 cgp->cg_irotor = ipref; 1259 goto gotit; 1260 } 1261 } 1262 printf("fs = %s\n", fs->fs_fsmnt); 1263 panic("ffs_nodealloccg: block not in map"); 1264 /* NOTREACHED */ 1265gotit: 1266 if (DOINGSOFTDEP(ITOV(ip))) 1267 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref); 1268 setbit(cg_inosused(cgp), ipref); 1269 cgp->cg_cs.cs_nifree--; 1270 fs->fs_cstotal.cs_nifree--; 1271 fs->fs_cs(fs, cg).cs_nifree--; 1272 fs->fs_fmod = 1; 1273 if ((mode & IFMT) == IFDIR) { 1274 cgp->cg_cs.cs_ndir++; 1275 fs->fs_cstotal.cs_ndir++; 1276 fs->fs_cs(fs, cg).cs_ndir++; 1277 } 1278 bdwrite(bp); 1279 return (cg * fs->fs_ipg + ipref); 1280} 1281 1282/* 1283 * Free a block or fragment. 1284 * 1285 * The specified block or fragment is placed back in the 1286 * free map. If a fragment is deallocated, a possible 1287 * block reassembly is checked. 1288 */ 1289void 1290ffs_blkfree(ip, bno, size) 1291 register struct inode *ip; 1292 ufs_daddr_t bno; 1293 long size; 1294{ 1295 register struct fs *fs; 1296 register struct cg *cgp; 1297 struct buf *bp; 1298 ufs_daddr_t blkno; 1299 int i, error, cg, blk, frags, bbase; 1300 1301 fs = ip->i_fs; 1302 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 1303 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 1304 printf("dev=0x%lx, bno = %d, bsize = %d, size = %ld, fs = %s\n", 1305 (u_long)ip->i_dev, bno, fs->fs_bsize, size, fs->fs_fsmnt); 1306 panic("ffs_blkfree: bad size"); 1307 } 1308 cg = dtog(fs, bno); 1309 if ((u_int)bno >= fs->fs_size) { 1310 printf("bad block %ld, ino %lu\n", 1311 (long)bno, (u_long)ip->i_number); 1312 ffs_fserr(fs, ip->i_uid, "bad block"); 1313 return; 1314 } 1315 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1316 (int)fs->fs_cgsize, NOCRED, &bp); 1317 if (error) { 1318 brelse(bp); 1319 return; 1320 } 1321 cgp = (struct cg *)bp->b_data; 1322 if (!cg_chkmagic(cgp)) { 1323 brelse(bp); 1324 return; 1325 } 1326 cgp->cg_time = time_second; 1327 bno = dtogd(fs, bno); 1328 if (size == fs->fs_bsize) { 1329 blkno = fragstoblks(fs, bno); 1330 if (!ffs_isfreeblock(fs, cg_blksfree(cgp), blkno)) { 1331 printf("dev = 0x%lx, block = %ld, fs = %s\n", 1332 (u_long)ip->i_dev, (long)bno, fs->fs_fsmnt); 1333 panic("ffs_blkfree: freeing free block"); 1334 } 1335 ffs_setblock(fs, cg_blksfree(cgp), blkno); 1336 ffs_clusteracct(fs, cgp, blkno, 1); 1337 cgp->cg_cs.cs_nbfree++; 1338 fs->fs_cstotal.cs_nbfree++; 1339 fs->fs_cs(fs, cg).cs_nbfree++; 1340 i = cbtocylno(fs, bno); 1341 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++; 1342 cg_blktot(cgp)[i]++; 1343 } else { 1344 bbase = bno - fragnum(fs, bno); 1345 /* 1346 * decrement the counts associated with the old frags 1347 */ 1348 blk = blkmap(fs, cg_blksfree(cgp), bbase); 1349 ffs_fragacct(fs, blk, cgp->cg_frsum, -1); 1350 /* 1351 * deallocate the fragment 1352 */ 1353 frags = numfrags(fs, size); 1354 for (i = 0; i < frags; i++) { 1355 if (isset(cg_blksfree(cgp), bno + i)) { 1356 printf("dev = 0x%lx, block = %ld, fs = %s\n", 1357 (u_long)ip->i_dev, (long)(bno + i), 1358 fs->fs_fsmnt); 1359 panic("ffs_blkfree: freeing free frag"); 1360 } 1361 setbit(cg_blksfree(cgp), bno + i); 1362 } 1363 cgp->cg_cs.cs_nffree += i; 1364 fs->fs_cstotal.cs_nffree += i; 1365 fs->fs_cs(fs, cg).cs_nffree += i; 1366 /* 1367 * add back in counts associated with the new frags 1368 */ 1369 blk = blkmap(fs, cg_blksfree(cgp), bbase); 1370 ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 1371 /* 1372 * if a complete block has been reassembled, account for it 1373 */ 1374 blkno = fragstoblks(fs, bbase); 1375 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) { 1376 cgp->cg_cs.cs_nffree -= fs->fs_frag; 1377 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 1378 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 1379 ffs_clusteracct(fs, cgp, blkno, 1); 1380 cgp->cg_cs.cs_nbfree++; 1381 fs->fs_cstotal.cs_nbfree++; 1382 fs->fs_cs(fs, cg).cs_nbfree++; 1383 i = cbtocylno(fs, bbase); 1384 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++; 1385 cg_blktot(cgp)[i]++; 1386 } 1387 } 1388 fs->fs_fmod = 1; 1389 bdwrite(bp); 1390} 1391 1392#ifdef DIAGNOSTIC 1393/* 1394 * Verify allocation of a block or fragment. Returns true if block or 1395 * fragment is allocated, false if it is free. 1396 */ 1397static int 1398ffs_checkblk(ip, bno, size) 1399 struct inode *ip; 1400 ufs_daddr_t bno; 1401 long size; 1402{ 1403 struct fs *fs; 1404 struct cg *cgp; 1405 struct buf *bp; 1406 int i, error, frags, free; 1407 1408 fs = ip->i_fs; 1409 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 1410 printf("bsize = %ld, size = %ld, fs = %s\n", 1411 (long)fs->fs_bsize, size, fs->fs_fsmnt); 1412 panic("ffs_checkblk: bad size"); 1413 } 1414 if ((u_int)bno >= fs->fs_size) 1415 panic("ffs_checkblk: bad block %d", bno); 1416 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))), 1417 (int)fs->fs_cgsize, NOCRED, &bp); 1418 if (error) 1419 panic("ffs_checkblk: cg bread failed"); 1420 cgp = (struct cg *)bp->b_data; 1421 if (!cg_chkmagic(cgp)) 1422 panic("ffs_checkblk: cg magic mismatch"); 1423 bno = dtogd(fs, bno); 1424 if (size == fs->fs_bsize) { 1425 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno)); 1426 } else { 1427 frags = numfrags(fs, size); 1428 for (free = 0, i = 0; i < frags; i++) 1429 if (isset(cg_blksfree(cgp), bno + i)) 1430 free++; 1431 if (free != 0 && free != frags) 1432 panic("ffs_checkblk: partially free fragment"); 1433 } 1434 brelse(bp); 1435 return (!free); 1436} 1437#endif /* DIAGNOSTIC */ 1438 1439/* 1440 * Free an inode. 1441 */ 1442int 1443ffs_vfree( pvp, ino, mode) 1444 struct vnode *pvp; 1445 ino_t ino; 1446 int mode; 1447{ 1448 if (DOINGSOFTDEP(pvp)) { 1449 softdep_freefile(pvp, ino, mode); 1450 return (0); 1451 } 1452 return (ffs_freefile(pvp, ino, mode)); 1453} 1454 1455/* 1456 * Do the actual free operation. 1457 * The specified inode is placed back in the free map. 1458 */ 1459 int 1460 ffs_freefile( pvp, ino, mode) 1461 struct vnode *pvp; 1462 ino_t ino; 1463 int mode; 1464{ 1465 register struct fs *fs; 1466 register struct cg *cgp; 1467 register struct inode *pip; 1468 struct buf *bp; 1469 int error, cg; 1470 1471 pip = VTOI(pvp); 1472 fs = pip->i_fs; 1473 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 1474 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s", 1475 pip->i_dev, ino, fs->fs_fsmnt); 1476 cg = ino_to_cg(fs, ino); 1477 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1478 (int)fs->fs_cgsize, NOCRED, &bp); 1479 if (error) { 1480 brelse(bp); 1481 return (error); 1482 } 1483 cgp = (struct cg *)bp->b_data; 1484 if (!cg_chkmagic(cgp)) { 1485 brelse(bp); 1486 return (0); 1487 } 1488 cgp->cg_time = time_second; 1489 ino %= fs->fs_ipg; 1490 if (isclr(cg_inosused(cgp), ino)) { 1491 printf("dev = 0x%lx, ino = %lu, fs = %s\n", 1492 (u_long)pip->i_dev, (u_long)ino, fs->fs_fsmnt); 1493 if (fs->fs_ronly == 0) 1494 panic("ffs_vfree: freeing free inode"); 1495 } 1496 clrbit(cg_inosused(cgp), ino); 1497 if (ino < cgp->cg_irotor) 1498 cgp->cg_irotor = ino; 1499 cgp->cg_cs.cs_nifree++; 1500 fs->fs_cstotal.cs_nifree++; 1501 fs->fs_cs(fs, cg).cs_nifree++; 1502 if ((mode & IFMT) == IFDIR) { 1503 cgp->cg_cs.cs_ndir--; 1504 fs->fs_cstotal.cs_ndir--; 1505 fs->fs_cs(fs, cg).cs_ndir--; 1506 } 1507 fs->fs_fmod = 1; 1508 bdwrite(bp); 1509 return (0); 1510} 1511 1512/* 1513 * Find a block of the specified size in the specified cylinder group. 1514 * 1515 * It is a panic if a request is made to find a block if none are 1516 * available. 1517 */ 1518static ufs_daddr_t 1519ffs_mapsearch(fs, cgp, bpref, allocsiz) 1520 register struct fs *fs; 1521 register struct cg *cgp; 1522 ufs_daddr_t bpref; 1523 int allocsiz; 1524{ 1525 ufs_daddr_t bno; 1526 int start, len, loc, i; 1527 int blk, field, subfield, pos; 1528 1529 /* 1530 * find the fragment by searching through the free block 1531 * map for an appropriate bit pattern 1532 */ 1533 if (bpref) 1534 start = dtogd(fs, bpref) / NBBY; 1535 else 1536 start = cgp->cg_frotor / NBBY; 1537 len = howmany(fs->fs_fpg, NBBY) - start; 1538 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start], 1539 (u_char *)fragtbl[fs->fs_frag], 1540 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 1541 if (loc == 0) { 1542 len = start + 1; 1543 start = 0; 1544 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0], 1545 (u_char *)fragtbl[fs->fs_frag], 1546 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 1547 if (loc == 0) { 1548 printf("start = %d, len = %d, fs = %s\n", 1549 start, len, fs->fs_fsmnt); 1550 panic("ffs_alloccg: map corrupted"); 1551 /* NOTREACHED */ 1552 } 1553 } 1554 bno = (start + len - loc) * NBBY; 1555 cgp->cg_frotor = bno; 1556 /* 1557 * found the byte in the map 1558 * sift through the bits to find the selected frag 1559 */ 1560 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 1561 blk = blkmap(fs, cg_blksfree(cgp), bno); 1562 blk <<= 1; 1563 field = around[allocsiz]; 1564 subfield = inside[allocsiz]; 1565 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 1566 if ((blk & field) == subfield) 1567 return (bno + pos); 1568 field <<= 1; 1569 subfield <<= 1; 1570 } 1571 } 1572 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); 1573 panic("ffs_alloccg: block not in map"); 1574 return (-1); 1575} 1576 1577/* 1578 * Update the cluster map because of an allocation or free. 1579 * 1580 * Cnt == 1 means free; cnt == -1 means allocating. 1581 */ 1582static void 1583ffs_clusteracct(fs, cgp, blkno, cnt) 1584 struct fs *fs; 1585 struct cg *cgp; 1586 ufs_daddr_t blkno; 1587 int cnt; 1588{ 1589 int32_t *sump; 1590 int32_t *lp; 1591 u_char *freemapp, *mapp; 1592 int i, start, end, forw, back, map, bit; 1593 1594 if (fs->fs_contigsumsize <= 0) 1595 return; 1596 freemapp = cg_clustersfree(cgp); 1597 sump = cg_clustersum(cgp); 1598 /* 1599 * Allocate or clear the actual block. 1600 */ 1601 if (cnt > 0) 1602 setbit(freemapp, blkno); 1603 else 1604 clrbit(freemapp, blkno); 1605 /* 1606 * Find the size of the cluster going forward. 1607 */ 1608 start = blkno + 1; 1609 end = start + fs->fs_contigsumsize; 1610 if (end >= cgp->cg_nclusterblks) 1611 end = cgp->cg_nclusterblks; 1612 mapp = &freemapp[start / NBBY]; 1613 map = *mapp++; 1614 bit = 1 << (start % NBBY); 1615 for (i = start; i < end; i++) { 1616 if ((map & bit) == 0) 1617 break; 1618 if ((i & (NBBY - 1)) != (NBBY - 1)) { 1619 bit <<= 1; 1620 } else { 1621 map = *mapp++; 1622 bit = 1; 1623 } 1624 } 1625 forw = i - start; 1626 /* 1627 * Find the size of the cluster going backward. 1628 */ 1629 start = blkno - 1; 1630 end = start - fs->fs_contigsumsize; 1631 if (end < 0) 1632 end = -1; 1633 mapp = &freemapp[start / NBBY]; 1634 map = *mapp--; 1635 bit = 1 << (start % NBBY); 1636 for (i = start; i > end; i--) { 1637 if ((map & bit) == 0) 1638 break; 1639 if ((i & (NBBY - 1)) != 0) { 1640 bit >>= 1; 1641 } else { 1642 map = *mapp--; 1643 bit = 1 << (NBBY - 1); 1644 } 1645 } 1646 back = start - i; 1647 /* 1648 * Account for old cluster and the possibly new forward and 1649 * back clusters. 1650 */ 1651 i = back + forw + 1; 1652 if (i > fs->fs_contigsumsize) 1653 i = fs->fs_contigsumsize; 1654 sump[i] += cnt; 1655 if (back > 0) 1656 sump[back] -= cnt; 1657 if (forw > 0) 1658 sump[forw] -= cnt; 1659 /* 1660 * Update cluster summary information. 1661 */ 1662 lp = &sump[fs->fs_contigsumsize]; 1663 for (i = fs->fs_contigsumsize; i > 0; i--) 1664 if (*lp-- > 0) 1665 break; 1666 fs->fs_maxcluster[cgp->cg_cgx] = i; 1667} 1668 1669/* 1670 * Fserr prints the name of a file system with an error diagnostic. 1671 * 1672 * The form of the error message is: 1673 * fs: error message 1674 */ 1675static void 1676ffs_fserr(fs, uid, cp) 1677 struct fs *fs; 1678 u_int uid; 1679 char *cp; 1680{ 1681 struct proc *p = curproc; /* XXX */ 1682 1683 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1, 1684 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp); 1685} 1686