ffs_alloc.c revision 72645
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: head/sys/ufs/ffs/ffs_alloc.c 72645 2001-02-18 10:43:53Z asmodai $ 35 */ 36 37#include "opt_quota.h" 38 39#include <sys/param.h> 40#include <sys/systm.h> 41#include <sys/bio.h> 42#include <sys/buf.h> 43#include <sys/conf.h> 44#include <sys/proc.h> 45#include <sys/vnode.h> 46#include <sys/mount.h> 47#include <sys/kernel.h> 48#include <sys/sysctl.h> 49#include <sys/syslog.h> 50 51#include <ufs/ufs/extattr.h> 52#include <ufs/ufs/quota.h> 53#include <ufs/ufs/inode.h> 54#include <ufs/ufs/ufs_extern.h> 55#include <ufs/ufs/ufsmount.h> 56 57#include <ufs/ffs/fs.h> 58#include <ufs/ffs/ffs_extern.h> 59 60typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref, 61 int size)); 62 63static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int)); 64static ufs_daddr_t 65 ffs_alloccgblk __P((struct inode *, struct buf *, ufs_daddr_t)); 66#ifdef DIAGNOSTIC 67static int ffs_checkblk __P((struct inode *, ufs_daddr_t, long)); 68#endif 69static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t, 70 int)); 71static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t, 72 int)); 73static ino_t ffs_dirpref __P((struct fs *)); 74static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int)); 75static void ffs_fserr __P((struct fs *, u_int, char *)); 76static u_long ffs_hashalloc 77 __P((struct inode *, int, long, int, allocfcn_t *)); 78static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int)); 79static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t, 80 int)); 81 82/* 83 * Allocate a block in the file system. 84 * 85 * The size of the requested block is given, which must be some 86 * multiple of fs_fsize and <= fs_bsize. 87 * A preference may be optionally specified. If a preference is given 88 * the following hierarchy is used to allocate a block: 89 * 1) allocate the requested block. 90 * 2) allocate a rotationally optimal block in the same cylinder. 91 * 3) allocate a block in the same cylinder group. 92 * 4) quadradically rehash into other cylinder groups, until an 93 * available block is located. 94 * If no block preference is given the following heirarchy is used 95 * to allocate a block: 96 * 1) allocate a block in the cylinder group that contains the 97 * inode for the file. 98 * 2) quadradically rehash into other cylinder groups, until an 99 * available block is located. 100 */ 101int 102ffs_alloc(ip, lbn, bpref, size, cred, bnp) 103 register struct inode *ip; 104 ufs_daddr_t lbn, bpref; 105 int size; 106 struct ucred *cred; 107 ufs_daddr_t *bnp; 108{ 109 register struct fs *fs; 110 ufs_daddr_t bno; 111 int cg; 112#ifdef QUOTA 113 int error; 114#endif 115 116 *bnp = 0; 117 fs = ip->i_fs; 118#ifdef DIAGNOSTIC 119 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 120 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n", 121 devtoname(ip->i_dev), (long)fs->fs_bsize, size, 122 fs->fs_fsmnt); 123 panic("ffs_alloc: bad size"); 124 } 125 if (cred == NOCRED) 126 panic("ffs_alloc: missing credential"); 127#endif /* DIAGNOSTIC */ 128 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) 129 goto nospace; 130 if (cred->cr_uid != 0 && 131 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) 132 goto nospace; 133#ifdef QUOTA 134 error = chkdq(ip, (long)btodb(size), cred, 0); 135 if (error) 136 return (error); 137#endif 138 if (bpref >= fs->fs_size) 139 bpref = 0; 140 if (bpref == 0) 141 cg = ino_to_cg(fs, ip->i_number); 142 else 143 cg = dtog(fs, bpref); 144 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size, 145 ffs_alloccg); 146 if (bno > 0) { 147 ip->i_blocks += btodb(size); 148 ip->i_flag |= IN_CHANGE | IN_UPDATE; 149 *bnp = bno; 150 return (0); 151 } 152#ifdef QUOTA 153 /* 154 * Restore user's disk quota because allocation failed. 155 */ 156 (void) chkdq(ip, (long)-btodb(size), cred, FORCE); 157#endif 158nospace: 159 ffs_fserr(fs, cred->cr_uid, "file system full"); 160 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 161 return (ENOSPC); 162} 163 164/* 165 * Reallocate a fragment to a bigger size 166 * 167 * The number and size of the old block is given, and a preference 168 * and new size is also specified. The allocator attempts to extend 169 * the original block. Failing that, the regular block allocator is 170 * invoked to get an appropriate block. 171 */ 172int 173ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp) 174 register struct inode *ip; 175 ufs_daddr_t lbprev; 176 ufs_daddr_t bpref; 177 int osize, nsize; 178 struct ucred *cred; 179 struct buf **bpp; 180{ 181 register struct fs *fs; 182 struct buf *bp; 183 int cg, request, error; 184 ufs_daddr_t bprev, bno; 185 186 *bpp = 0; 187 fs = ip->i_fs; 188#ifdef DIAGNOSTIC 189 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) 190 panic("ffs_realloccg: allocation on suspended filesystem"); 191 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || 192 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { 193 printf( 194 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", 195 devtoname(ip->i_dev), (long)fs->fs_bsize, osize, 196 nsize, fs->fs_fsmnt); 197 panic("ffs_realloccg: bad size"); 198 } 199 if (cred == NOCRED) 200 panic("ffs_realloccg: missing credential"); 201#endif /* DIAGNOSTIC */ 202 if (cred->cr_uid != 0 && 203 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) 204 goto nospace; 205 if ((bprev = ip->i_db[lbprev]) == 0) { 206 printf("dev = %s, bsize = %ld, bprev = %ld, fs = %s\n", 207 devtoname(ip->i_dev), (long)fs->fs_bsize, (long)bprev, 208 fs->fs_fsmnt); 209 panic("ffs_realloccg: bad bprev"); 210 } 211 /* 212 * Allocate the extra space in the buffer. 213 */ 214 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp); 215 if (error) { 216 brelse(bp); 217 return (error); 218 } 219 220 if( bp->b_blkno == bp->b_lblkno) { 221 if( lbprev >= NDADDR) 222 panic("ffs_realloccg: lbprev out of range"); 223 bp->b_blkno = fsbtodb(fs, bprev); 224 } 225 226#ifdef QUOTA 227 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0); 228 if (error) { 229 brelse(bp); 230 return (error); 231 } 232#endif 233 /* 234 * Check for extension in the existing location. 235 */ 236 cg = dtog(fs, bprev); 237 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize); 238 if (bno) { 239 if (bp->b_blkno != fsbtodb(fs, bno)) 240 panic("ffs_realloccg: bad blockno"); 241 ip->i_blocks += btodb(nsize - osize); 242 ip->i_flag |= IN_CHANGE | IN_UPDATE; 243 allocbuf(bp, nsize); 244 bp->b_flags |= B_DONE; 245 bzero((char *)bp->b_data + osize, (u_int)nsize - osize); 246 *bpp = bp; 247 return (0); 248 } 249 /* 250 * Allocate a new disk location. 251 */ 252 if (bpref >= fs->fs_size) 253 bpref = 0; 254 switch ((int)fs->fs_optim) { 255 case FS_OPTSPACE: 256 /* 257 * Allocate an exact sized fragment. Although this makes 258 * best use of space, we will waste time relocating it if 259 * the file continues to grow. If the fragmentation is 260 * less than half of the minimum free reserve, we choose 261 * to begin optimizing for time. 262 */ 263 request = nsize; 264 if (fs->fs_minfree <= 5 || 265 fs->fs_cstotal.cs_nffree > 266 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100)) 267 break; 268 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", 269 fs->fs_fsmnt); 270 fs->fs_optim = FS_OPTTIME; 271 break; 272 case FS_OPTTIME: 273 /* 274 * At this point we have discovered a file that is trying to 275 * grow a small fragment to a larger fragment. To save time, 276 * we allocate a full sized block, then free the unused portion. 277 * If the file continues to grow, the `ffs_fragextend' call 278 * above will be able to grow it in place without further 279 * copying. If aberrant programs cause disk fragmentation to 280 * grow within 2% of the free reserve, we choose to begin 281 * optimizing for space. 282 */ 283 request = fs->fs_bsize; 284 if (fs->fs_cstotal.cs_nffree < 285 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100) 286 break; 287 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", 288 fs->fs_fsmnt); 289 fs->fs_optim = FS_OPTSPACE; 290 break; 291 default: 292 printf("dev = %s, optim = %ld, fs = %s\n", 293 devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt); 294 panic("ffs_realloccg: bad optim"); 295 /* NOTREACHED */ 296 } 297 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request, 298 ffs_alloccg); 299 if (bno > 0) { 300 bp->b_blkno = fsbtodb(fs, bno); 301 if (!DOINGSOFTDEP(ITOV(ip))) 302 ffs_blkfree(ip, bprev, (long)osize); 303 if (nsize < request) 304 ffs_blkfree(ip, bno + numfrags(fs, nsize), 305 (long)(request - nsize)); 306 ip->i_blocks += btodb(nsize - osize); 307 ip->i_flag |= IN_CHANGE | IN_UPDATE; 308 allocbuf(bp, nsize); 309 bp->b_flags |= B_DONE; 310 bzero((char *)bp->b_data + osize, (u_int)nsize - osize); 311 *bpp = bp; 312 return (0); 313 } 314#ifdef QUOTA 315 /* 316 * Restore user's disk quota because allocation failed. 317 */ 318 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE); 319#endif 320 brelse(bp); 321nospace: 322 /* 323 * no space available 324 */ 325 ffs_fserr(fs, cred->cr_uid, "file system full"); 326 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 327 return (ENOSPC); 328} 329 330SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem"); 331 332/* 333 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 334 * 335 * The vnode and an array of buffer pointers for a range of sequential 336 * logical blocks to be made contiguous is given. The allocator attempts 337 * to find a range of sequential blocks starting as close as possible to 338 * an fs_rotdelay offset from the end of the allocation for the logical 339 * block immediately preceding the current range. If successful, the 340 * physical block numbers in the buffer pointers and in the inode are 341 * changed to reflect the new allocation. If unsuccessful, the allocation 342 * is left unchanged. The success in doing the reallocation is returned. 343 * Note that the error return is not reflected back to the user. Rather 344 * the previous block allocation will be used. 345 */ 346static int doasyncfree = 1; 347SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, ""); 348 349static int doreallocblks = 1; 350SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); 351 352#ifdef DEBUG 353static volatile int prtrealloc = 0; 354#endif 355 356int 357ffs_reallocblks(ap) 358 struct vop_reallocblks_args /* { 359 struct vnode *a_vp; 360 struct cluster_save *a_buflist; 361 } */ *ap; 362{ 363 struct fs *fs; 364 struct inode *ip; 365 struct vnode *vp; 366 struct buf *sbp, *ebp; 367 ufs_daddr_t *bap, *sbap, *ebap = 0; 368 struct cluster_save *buflist; 369 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno; 370 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; 371 int i, len, start_lvl, end_lvl, pref, ssize; 372 373 if (doreallocblks == 0) 374 return (ENOSPC); 375 vp = ap->a_vp; 376 ip = VTOI(vp); 377 fs = ip->i_fs; 378 if (fs->fs_contigsumsize <= 0) 379 return (ENOSPC); 380 buflist = ap->a_buflist; 381 len = buflist->bs_nchildren; 382 start_lbn = buflist->bs_children[0]->b_lblkno; 383 end_lbn = start_lbn + len - 1; 384#ifdef DIAGNOSTIC 385 for (i = 0; i < len; i++) 386 if (!ffs_checkblk(ip, 387 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 388 panic("ffs_reallocblks: unallocated block 1"); 389 for (i = 1; i < len; i++) 390 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 391 panic("ffs_reallocblks: non-logical cluster"); 392 blkno = buflist->bs_children[0]->b_blkno; 393 ssize = fsbtodb(fs, fs->fs_frag); 394 for (i = 1; i < len - 1; i++) 395 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) 396 panic("ffs_reallocblks: non-physical cluster %d", i); 397#endif 398 /* 399 * If the latest allocation is in a new cylinder group, assume that 400 * the filesystem has decided to move and do not force it back to 401 * the previous cylinder group. 402 */ 403 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 404 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 405 return (ENOSPC); 406 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 407 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 408 return (ENOSPC); 409 /* 410 * Get the starting offset and block map for the first block. 411 */ 412 if (start_lvl == 0) { 413 sbap = &ip->i_db[0]; 414 soff = start_lbn; 415 } else { 416 idp = &start_ap[start_lvl - 1]; 417 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { 418 brelse(sbp); 419 return (ENOSPC); 420 } 421 sbap = (ufs_daddr_t *)sbp->b_data; 422 soff = idp->in_off; 423 } 424 /* 425 * Find the preferred location for the cluster. 426 */ 427 pref = ffs_blkpref(ip, start_lbn, soff, sbap); 428 /* 429 * If the block range spans two block maps, get the second map. 430 */ 431 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 432 ssize = len; 433 } else { 434#ifdef DIAGNOSTIC 435 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn) 436 panic("ffs_reallocblk: start == end"); 437#endif 438 ssize = len - (idp->in_off + 1); 439 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) 440 goto fail; 441 ebap = (ufs_daddr_t *)ebp->b_data; 442 } 443 /* 444 * Search the block map looking for an allocation of the desired size. 445 */ 446 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref, 447 len, ffs_clusteralloc)) == 0) 448 goto fail; 449 /* 450 * We have found a new contiguous block. 451 * 452 * First we have to replace the old block pointers with the new 453 * block pointers in the inode and indirect blocks associated 454 * with the file. 455 */ 456#ifdef DEBUG 457 if (prtrealloc) 458 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number, 459 start_lbn, end_lbn); 460#endif 461 blkno = newblk; 462 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { 463 if (i == ssize) { 464 bap = ebap; 465 soff = -i; 466 } 467#ifdef DIAGNOSTIC 468 if (!ffs_checkblk(ip, 469 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 470 panic("ffs_reallocblks: unallocated block 2"); 471 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) 472 panic("ffs_reallocblks: alloc mismatch"); 473#endif 474#ifdef DEBUG 475 if (prtrealloc) 476 printf(" %d,", *bap); 477#endif 478 if (DOINGSOFTDEP(vp)) { 479 if (sbap == &ip->i_db[0] && i < ssize) 480 softdep_setup_allocdirect(ip, start_lbn + i, 481 blkno, *bap, fs->fs_bsize, fs->fs_bsize, 482 buflist->bs_children[i]); 483 else 484 softdep_setup_allocindir_page(ip, start_lbn + i, 485 i < ssize ? sbp : ebp, soff + i, blkno, 486 *bap, buflist->bs_children[i]); 487 } 488 *bap++ = blkno; 489 } 490 /* 491 * Next we must write out the modified inode and indirect blocks. 492 * For strict correctness, the writes should be synchronous since 493 * the old block values may have been written to disk. In practise 494 * they are almost never written, but if we are concerned about 495 * strict correctness, the `doasyncfree' flag should be set to zero. 496 * 497 * The test on `doasyncfree' should be changed to test a flag 498 * that shows whether the associated buffers and inodes have 499 * been written. The flag should be set when the cluster is 500 * started and cleared whenever the buffer or inode is flushed. 501 * We can then check below to see if it is set, and do the 502 * synchronous write only when it has been cleared. 503 */ 504 if (sbap != &ip->i_db[0]) { 505 if (doasyncfree) 506 bdwrite(sbp); 507 else 508 bwrite(sbp); 509 } else { 510 ip->i_flag |= IN_CHANGE | IN_UPDATE; 511 if (!doasyncfree) 512 UFS_UPDATE(vp, 1); 513 } 514 if (ssize < len) { 515 if (doasyncfree) 516 bdwrite(ebp); 517 else 518 bwrite(ebp); 519 } 520 /* 521 * Last, free the old blocks and assign the new blocks to the buffers. 522 */ 523#ifdef DEBUG 524 if (prtrealloc) 525 printf("\n\tnew:"); 526#endif 527 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { 528 if (!DOINGSOFTDEP(vp)) 529 ffs_blkfree(ip, 530 dbtofsb(fs, buflist->bs_children[i]->b_blkno), 531 fs->fs_bsize); 532 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 533#ifdef DIAGNOSTIC 534 if (!ffs_checkblk(ip, 535 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) 536 panic("ffs_reallocblks: unallocated block 3"); 537#endif 538#ifdef DEBUG 539 if (prtrealloc) 540 printf(" %d,", blkno); 541#endif 542 } 543#ifdef DEBUG 544 if (prtrealloc) { 545 prtrealloc--; 546 printf("\n"); 547 } 548#endif 549 return (0); 550 551fail: 552 if (ssize < len) 553 brelse(ebp); 554 if (sbap != &ip->i_db[0]) 555 brelse(sbp); 556 return (ENOSPC); 557} 558 559/* 560 * Allocate an inode in the file system. 561 * 562 * If allocating a directory, use ffs_dirpref to select the inode. 563 * If allocating in a directory, the following hierarchy is followed: 564 * 1) allocate the preferred inode. 565 * 2) allocate an inode in the same cylinder group. 566 * 3) quadradically rehash into other cylinder groups, until an 567 * available inode is located. 568 * If no inode preference is given the following heirarchy is used 569 * to allocate an inode: 570 * 1) allocate an inode in cylinder group 0. 571 * 2) quadradically rehash into other cylinder groups, until an 572 * available inode is located. 573 */ 574int 575ffs_valloc(pvp, mode, cred, vpp) 576 struct vnode *pvp; 577 int mode; 578 struct ucred *cred; 579 struct vnode **vpp; 580{ 581 register struct inode *pip; 582 register struct fs *fs; 583 register struct inode *ip; 584 ino_t ino, ipref; 585 int cg, error; 586 587 *vpp = NULL; 588 pip = VTOI(pvp); 589 fs = pip->i_fs; 590 if (fs->fs_cstotal.cs_nifree == 0) 591 goto noinodes; 592 593 if ((mode & IFMT) == IFDIR) 594 ipref = ffs_dirpref(fs); 595 else 596 ipref = pip->i_number; 597 if (ipref >= fs->fs_ncg * fs->fs_ipg) 598 ipref = 0; 599 cg = ino_to_cg(fs, ipref); 600 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode, 601 (allocfcn_t *)ffs_nodealloccg); 602 if (ino == 0) 603 goto noinodes; 604 error = VFS_VGET(pvp->v_mount, ino, vpp); 605 if (error) { 606 UFS_VFREE(pvp, ino, mode); 607 return (error); 608 } 609 ip = VTOI(*vpp); 610 if (ip->i_mode) { 611 printf("mode = 0%o, inum = %lu, fs = %s\n", 612 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt); 613 panic("ffs_valloc: dup alloc"); 614 } 615 if (ip->i_blocks) { /* XXX */ 616 printf("free inode %s/%lu had %ld blocks\n", 617 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks); 618 ip->i_blocks = 0; 619 } 620 ip->i_flags = 0; 621 /* 622 * Set up a new generation number for this inode. 623 */ 624 if (ip->i_gen == 0 || ++ip->i_gen == 0) 625 ip->i_gen = random() / 2 + 1; 626 return (0); 627noinodes: 628 ffs_fserr(fs, cred->cr_uid, "out of inodes"); 629 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); 630 return (ENOSPC); 631} 632 633/* 634 * Find a cylinder to place a directory. 635 * 636 * The policy implemented by this algorithm is to select from 637 * among those cylinder groups with above the average number of 638 * free inodes, the one with the smallest number of directories. 639 */ 640static ino_t 641ffs_dirpref(fs) 642 register struct fs *fs; 643{ 644 int cg, minndir, mincg, avgifree; 645 646 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 647 minndir = fs->fs_ipg; 648 mincg = 0; 649 for (cg = 0; cg < fs->fs_ncg; cg++) 650 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 651 fs->fs_cs(fs, cg).cs_nifree >= avgifree) { 652 mincg = cg; 653 minndir = fs->fs_cs(fs, cg).cs_ndir; 654 } 655 return ((ino_t)(fs->fs_ipg * mincg)); 656} 657 658/* 659 * Select the desired position for the next block in a file. The file is 660 * logically divided into sections. The first section is composed of the 661 * direct blocks. Each additional section contains fs_maxbpg blocks. 662 * 663 * If no blocks have been allocated in the first section, the policy is to 664 * request a block in the same cylinder group as the inode that describes 665 * the file. If no blocks have been allocated in any other section, the 666 * policy is to place the section in a cylinder group with a greater than 667 * average number of free blocks. An appropriate cylinder group is found 668 * by using a rotor that sweeps the cylinder groups. When a new group of 669 * blocks is needed, the sweep begins in the cylinder group following the 670 * cylinder group from which the previous allocation was made. The sweep 671 * continues until a cylinder group with greater than the average number 672 * of free blocks is found. If the allocation is for the first block in an 673 * indirect block, the information on the previous allocation is unavailable; 674 * here a best guess is made based upon the logical block number being 675 * allocated. 676 * 677 * If a section is already partially allocated, the policy is to 678 * contiguously allocate fs_maxcontig blocks. The end of one of these 679 * contiguous blocks and the beginning of the next is physically separated 680 * so that the disk head will be in transit between them for at least 681 * fs_rotdelay milliseconds. This is to allow time for the processor to 682 * schedule another I/O transfer. 683 */ 684ufs_daddr_t 685ffs_blkpref(ip, lbn, indx, bap) 686 struct inode *ip; 687 ufs_daddr_t lbn; 688 int indx; 689 ufs_daddr_t *bap; 690{ 691 register struct fs *fs; 692 register int cg; 693 int avgbfree, startcg; 694 ufs_daddr_t nextblk; 695 696 fs = ip->i_fs; 697 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 698 if (lbn < NDADDR + NINDIR(fs)) { 699 cg = ino_to_cg(fs, ip->i_number); 700 return (fs->fs_fpg * cg + fs->fs_frag); 701 } 702 /* 703 * Find a cylinder with greater than average number of 704 * unused data blocks. 705 */ 706 if (indx == 0 || bap[indx - 1] == 0) 707 startcg = 708 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 709 else 710 startcg = dtog(fs, bap[indx - 1]) + 1; 711 startcg %= fs->fs_ncg; 712 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 713 for (cg = startcg; cg < fs->fs_ncg; cg++) 714 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 715 fs->fs_cgrotor = cg; 716 return (fs->fs_fpg * cg + fs->fs_frag); 717 } 718 for (cg = 0; cg <= startcg; cg++) 719 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 720 fs->fs_cgrotor = cg; 721 return (fs->fs_fpg * cg + fs->fs_frag); 722 } 723 return (0); 724 } 725 /* 726 * One or more previous blocks have been laid out. If less 727 * than fs_maxcontig previous blocks are contiguous, the 728 * next block is requested contiguously, otherwise it is 729 * requested rotationally delayed by fs_rotdelay milliseconds. 730 */ 731 nextblk = bap[indx - 1] + fs->fs_frag; 732 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig || 733 bap[indx - fs->fs_maxcontig] + 734 blkstofrags(fs, fs->fs_maxcontig) != nextblk) 735 return (nextblk); 736 /* 737 * Here we convert ms of delay to frags as: 738 * (frags) = (ms) * (rev/sec) * (sect/rev) / 739 * ((sect/frag) * (ms/sec)) 740 * then round up to the next block. 741 */ 742 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect / 743 (NSPF(fs) * 1000), fs->fs_frag); 744 return (nextblk); 745} 746 747/* 748 * Implement the cylinder overflow algorithm. 749 * 750 * The policy implemented by this algorithm is: 751 * 1) allocate the block in its requested cylinder group. 752 * 2) quadradically rehash on the cylinder group number. 753 * 3) brute force search for a free block. 754 */ 755/*VARARGS5*/ 756static u_long 757ffs_hashalloc(ip, cg, pref, size, allocator) 758 struct inode *ip; 759 int cg; 760 long pref; 761 int size; /* size for data blocks, mode for inodes */ 762 allocfcn_t *allocator; 763{ 764 register struct fs *fs; 765 long result; /* XXX why not same type as we return? */ 766 int i, icg = cg; 767 768#ifdef DIAGNOSTIC 769 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) 770 panic("ffs_hashalloc: allocation on suspended filesystem"); 771#endif 772 fs = ip->i_fs; 773 /* 774 * 1: preferred cylinder group 775 */ 776 result = (*allocator)(ip, cg, pref, size); 777 if (result) 778 return (result); 779 /* 780 * 2: quadratic rehash 781 */ 782 for (i = 1; i < fs->fs_ncg; i *= 2) { 783 cg += i; 784 if (cg >= fs->fs_ncg) 785 cg -= fs->fs_ncg; 786 result = (*allocator)(ip, cg, 0, size); 787 if (result) 788 return (result); 789 } 790 /* 791 * 3: brute force search 792 * Note that we start at i == 2, since 0 was checked initially, 793 * and 1 is always checked in the quadratic rehash. 794 */ 795 cg = (icg + 2) % fs->fs_ncg; 796 for (i = 2; i < fs->fs_ncg; i++) { 797 result = (*allocator)(ip, cg, 0, size); 798 if (result) 799 return (result); 800 cg++; 801 if (cg == fs->fs_ncg) 802 cg = 0; 803 } 804 return (0); 805} 806 807/* 808 * Determine whether a fragment can be extended. 809 * 810 * Check to see if the necessary fragments are available, and 811 * if they are, allocate them. 812 */ 813static ufs_daddr_t 814ffs_fragextend(ip, cg, bprev, osize, nsize) 815 struct inode *ip; 816 int cg; 817 long bprev; 818 int osize, nsize; 819{ 820 register struct fs *fs; 821 register struct cg *cgp; 822 struct buf *bp; 823 long bno; 824 int frags, bbase; 825 int i, error; 826 u_int8_t *blksfree; 827 828 fs = ip->i_fs; 829 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 830 return (0); 831 frags = numfrags(fs, nsize); 832 bbase = fragnum(fs, bprev); 833 if (bbase > fragnum(fs, (bprev + frags - 1))) { 834 /* cannot extend across a block boundary */ 835 return (0); 836 } 837 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 838 (int)fs->fs_cgsize, NOCRED, &bp); 839 if (error) { 840 brelse(bp); 841 return (0); 842 } 843 cgp = (struct cg *)bp->b_data; 844 if (!cg_chkmagic(cgp)) { 845 brelse(bp); 846 return (0); 847 } 848 bp->b_xflags |= BX_BKGRDWRITE; 849 cgp->cg_time = time_second; 850 bno = dtogd(fs, bprev); 851 blksfree = cg_blksfree(cgp); 852 for (i = numfrags(fs, osize); i < frags; i++) 853 if (isclr(blksfree, bno + i)) { 854 brelse(bp); 855 return (0); 856 } 857 /* 858 * the current fragment can be extended 859 * deduct the count on fragment being extended into 860 * increase the count on the remaining fragment (if any) 861 * allocate the extended piece 862 */ 863 for (i = frags; i < fs->fs_frag - bbase; i++) 864 if (isclr(blksfree, bno + i)) 865 break; 866 cgp->cg_frsum[i - numfrags(fs, osize)]--; 867 if (i != frags) 868 cgp->cg_frsum[i - frags]++; 869 for (i = numfrags(fs, osize); i < frags; i++) { 870 clrbit(blksfree, bno + i); 871 cgp->cg_cs.cs_nffree--; 872 fs->fs_cstotal.cs_nffree--; 873 fs->fs_cs(fs, cg).cs_nffree--; 874 } 875 fs->fs_fmod = 1; 876 if (DOINGSOFTDEP(ITOV(ip))) 877 softdep_setup_blkmapdep(bp, fs, bprev); 878 bdwrite(bp); 879 return (bprev); 880} 881 882/* 883 * Determine whether a block can be allocated. 884 * 885 * Check to see if a block of the appropriate size is available, 886 * and if it is, allocate it. 887 */ 888static ufs_daddr_t 889ffs_alloccg(ip, cg, bpref, size) 890 struct inode *ip; 891 int cg; 892 ufs_daddr_t bpref; 893 int size; 894{ 895 register struct fs *fs; 896 register struct cg *cgp; 897 struct buf *bp; 898 register int i; 899 ufs_daddr_t bno, blkno; 900 int allocsiz, error, frags; 901 u_int8_t *blksfree; 902 903 fs = ip->i_fs; 904 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 905 return (0); 906 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 907 (int)fs->fs_cgsize, NOCRED, &bp); 908 if (error) { 909 brelse(bp); 910 return (0); 911 } 912 cgp = (struct cg *)bp->b_data; 913 if (!cg_chkmagic(cgp) || 914 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) { 915 brelse(bp); 916 return (0); 917 } 918 bp->b_xflags |= BX_BKGRDWRITE; 919 cgp->cg_time = time_second; 920 if (size == fs->fs_bsize) { 921 bno = ffs_alloccgblk(ip, bp, bpref); 922 bdwrite(bp); 923 return (bno); 924 } 925 /* 926 * check to see if any fragments are already available 927 * allocsiz is the size which will be allocated, hacking 928 * it down to a smaller size if necessary 929 */ 930 blksfree = cg_blksfree(cgp); 931 frags = numfrags(fs, size); 932 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 933 if (cgp->cg_frsum[allocsiz] != 0) 934 break; 935 if (allocsiz == fs->fs_frag) { 936 /* 937 * no fragments were available, so a block will be 938 * allocated, and hacked up 939 */ 940 if (cgp->cg_cs.cs_nbfree == 0) { 941 brelse(bp); 942 return (0); 943 } 944 bno = ffs_alloccgblk(ip, bp, bpref); 945 bpref = dtogd(fs, bno); 946 for (i = frags; i < fs->fs_frag; i++) 947 setbit(blksfree, bpref + i); 948 i = fs->fs_frag - frags; 949 cgp->cg_cs.cs_nffree += i; 950 fs->fs_cstotal.cs_nffree += i; 951 fs->fs_cs(fs, cg).cs_nffree += i; 952 fs->fs_fmod = 1; 953 cgp->cg_frsum[i]++; 954 bdwrite(bp); 955 return (bno); 956 } 957 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 958 if (bno < 0) { 959 brelse(bp); 960 return (0); 961 } 962 for (i = 0; i < frags; i++) 963 clrbit(blksfree, bno + i); 964 cgp->cg_cs.cs_nffree -= frags; 965 fs->fs_cstotal.cs_nffree -= frags; 966 fs->fs_cs(fs, cg).cs_nffree -= frags; 967 fs->fs_fmod = 1; 968 cgp->cg_frsum[allocsiz]--; 969 if (frags != allocsiz) 970 cgp->cg_frsum[allocsiz - frags]++; 971 blkno = cg * fs->fs_fpg + bno; 972 if (DOINGSOFTDEP(ITOV(ip))) 973 softdep_setup_blkmapdep(bp, fs, blkno); 974 bdwrite(bp); 975 return ((u_long)blkno); 976} 977 978/* 979 * Allocate a block in a cylinder group. 980 * 981 * This algorithm implements the following policy: 982 * 1) allocate the requested block. 983 * 2) allocate a rotationally optimal block in the same cylinder. 984 * 3) allocate the next available block on the block rotor for the 985 * specified cylinder group. 986 * Note that this routine only allocates fs_bsize blocks; these 987 * blocks may be fragmented by the routine that allocates them. 988 */ 989static ufs_daddr_t 990ffs_alloccgblk(ip, bp, bpref) 991 struct inode *ip; 992 struct buf *bp; 993 ufs_daddr_t bpref; 994{ 995 struct fs *fs; 996 struct cg *cgp; 997 ufs_daddr_t bno, blkno; 998 int cylno, pos, delta; 999 short *cylbp; 1000 register int i; 1001 u_int8_t *blksfree; 1002 1003 fs = ip->i_fs; 1004 cgp = (struct cg *)bp->b_data; 1005 blksfree = cg_blksfree(cgp); 1006 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) { 1007 bpref = cgp->cg_rotor; 1008 goto norot; 1009 } 1010 bpref = blknum(fs, bpref); 1011 bpref = dtogd(fs, bpref); 1012 /* 1013 * if the requested block is available, use it 1014 */ 1015 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bpref))) { 1016 bno = bpref; 1017 goto gotit; 1018 } 1019 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) { 1020 /* 1021 * Block layout information is not available. 1022 * Leaving bpref unchanged means we take the 1023 * next available free block following the one 1024 * we just allocated. Hopefully this will at 1025 * least hit a track cache on drives of unknown 1026 * geometry (e.g. SCSI). 1027 */ 1028 goto norot; 1029 } 1030 /* 1031 * check for a block available on the same cylinder 1032 */ 1033 cylno = cbtocylno(fs, bpref); 1034 if (cg_blktot(cgp)[cylno] == 0) 1035 goto norot; 1036 /* 1037 * check the summary information to see if a block is 1038 * available in the requested cylinder starting at the 1039 * requested rotational position and proceeding around. 1040 */ 1041 cylbp = cg_blks(fs, cgp, cylno); 1042 pos = cbtorpos(fs, bpref); 1043 for (i = pos; i < fs->fs_nrpos; i++) 1044 if (cylbp[i] > 0) 1045 break; 1046 if (i == fs->fs_nrpos) 1047 for (i = 0; i < pos; i++) 1048 if (cylbp[i] > 0) 1049 break; 1050 if (cylbp[i] > 0) { 1051 /* 1052 * found a rotational position, now find the actual 1053 * block. A panic if none is actually there. 1054 */ 1055 pos = cylno % fs->fs_cpc; 1056 bno = (cylno - pos) * fs->fs_spc / NSPB(fs); 1057 if (fs_postbl(fs, pos)[i] == -1) { 1058 printf("pos = %d, i = %d, fs = %s\n", 1059 pos, i, fs->fs_fsmnt); 1060 panic("ffs_alloccgblk: cyl groups corrupted"); 1061 } 1062 for (i = fs_postbl(fs, pos)[i];; ) { 1063 if (ffs_isblock(fs, blksfree, bno + i)) { 1064 bno = blkstofrags(fs, (bno + i)); 1065 goto gotit; 1066 } 1067 delta = fs_rotbl(fs)[i]; 1068 if (delta <= 0 || 1069 delta + i > fragstoblks(fs, fs->fs_fpg)) 1070 break; 1071 i += delta; 1072 } 1073 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt); 1074 panic("ffs_alloccgblk: can't find blk in cyl"); 1075 } 1076norot: 1077 /* 1078 * no blocks in the requested cylinder, so take next 1079 * available one in this cylinder group. 1080 */ 1081 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1082 if (bno < 0) 1083 return (0); 1084 cgp->cg_rotor = bno; 1085gotit: 1086 blkno = fragstoblks(fs, bno); 1087 ffs_clrblock(fs, blksfree, (long)blkno); 1088 ffs_clusteracct(fs, cgp, blkno, -1); 1089 cgp->cg_cs.cs_nbfree--; 1090 fs->fs_cstotal.cs_nbfree--; 1091 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; 1092 cylno = cbtocylno(fs, bno); 1093 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--; 1094 cg_blktot(cgp)[cylno]--; 1095 fs->fs_fmod = 1; 1096 blkno = cgp->cg_cgx * fs->fs_fpg + bno; 1097 if (DOINGSOFTDEP(ITOV(ip))) 1098 softdep_setup_blkmapdep(bp, fs, blkno); 1099 return (blkno); 1100} 1101 1102/* 1103 * Determine whether a cluster can be allocated. 1104 * 1105 * We do not currently check for optimal rotational layout if there 1106 * are multiple choices in the same cylinder group. Instead we just 1107 * take the first one that we find following bpref. 1108 */ 1109static ufs_daddr_t 1110ffs_clusteralloc(ip, cg, bpref, len) 1111 struct inode *ip; 1112 int cg; 1113 ufs_daddr_t bpref; 1114 int len; 1115{ 1116 register struct fs *fs; 1117 register struct cg *cgp; 1118 struct buf *bp; 1119 int i, got, run, bno, bit, map; 1120 u_char *mapp; 1121 int32_t *lp; 1122 u_int8_t *blksfree; 1123 1124 fs = ip->i_fs; 1125 if (fs->fs_maxcluster[cg] < len) 1126 return (0); 1127 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize, 1128 NOCRED, &bp)) 1129 goto fail; 1130 cgp = (struct cg *)bp->b_data; 1131 if (!cg_chkmagic(cgp)) 1132 goto fail; 1133 bp->b_xflags |= BX_BKGRDWRITE; 1134 /* 1135 * Check to see if a cluster of the needed size (or bigger) is 1136 * available in this cylinder group. 1137 */ 1138 lp = &cg_clustersum(cgp)[len]; 1139 for (i = len; i <= fs->fs_contigsumsize; i++) 1140 if (*lp++ > 0) 1141 break; 1142 if (i > fs->fs_contigsumsize) { 1143 /* 1144 * This is the first time looking for a cluster in this 1145 * cylinder group. Update the cluster summary information 1146 * to reflect the true maximum sized cluster so that 1147 * future cluster allocation requests can avoid reading 1148 * the cylinder group map only to find no clusters. 1149 */ 1150 lp = &cg_clustersum(cgp)[len - 1]; 1151 for (i = len - 1; i > 0; i--) 1152 if (*lp-- > 0) 1153 break; 1154 fs->fs_maxcluster[cg] = i; 1155 goto fail; 1156 } 1157 /* 1158 * Search the cluster map to find a big enough cluster. 1159 * We take the first one that we find, even if it is larger 1160 * than we need as we prefer to get one close to the previous 1161 * block allocation. We do not search before the current 1162 * preference point as we do not want to allocate a block 1163 * that is allocated before the previous one (as we will 1164 * then have to wait for another pass of the elevator 1165 * algorithm before it will be read). We prefer to fail and 1166 * be recalled to try an allocation in the next cylinder group. 1167 */ 1168 if (dtog(fs, bpref) != cg) 1169 bpref = 0; 1170 else 1171 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref))); 1172 mapp = &cg_clustersfree(cgp)[bpref / NBBY]; 1173 map = *mapp++; 1174 bit = 1 << (bpref % NBBY); 1175 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { 1176 if ((map & bit) == 0) { 1177 run = 0; 1178 } else { 1179 run++; 1180 if (run == len) 1181 break; 1182 } 1183 if ((got & (NBBY - 1)) != (NBBY - 1)) { 1184 bit <<= 1; 1185 } else { 1186 map = *mapp++; 1187 bit = 1; 1188 } 1189 } 1190 if (got >= cgp->cg_nclusterblks) 1191 goto fail; 1192 /* 1193 * Allocate the cluster that we have found. 1194 */ 1195 blksfree = cg_blksfree(cgp); 1196 for (i = 1; i <= len; i++) 1197 if (!ffs_isblock(fs, blksfree, got - run + i)) 1198 panic("ffs_clusteralloc: map mismatch"); 1199 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1); 1200 if (dtog(fs, bno) != cg) 1201 panic("ffs_clusteralloc: allocated out of group"); 1202 len = blkstofrags(fs, len); 1203 for (i = 0; i < len; i += fs->fs_frag) 1204 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i) 1205 panic("ffs_clusteralloc: lost block"); 1206 bdwrite(bp); 1207 return (bno); 1208 1209fail: 1210 brelse(bp); 1211 return (0); 1212} 1213 1214/* 1215 * Determine whether an inode can be allocated. 1216 * 1217 * Check to see if an inode is available, and if it is, 1218 * allocate it using the following policy: 1219 * 1) allocate the requested inode. 1220 * 2) allocate the next available inode after the requested 1221 * inode in the specified cylinder group. 1222 */ 1223static ino_t 1224ffs_nodealloccg(ip, cg, ipref, mode) 1225 struct inode *ip; 1226 int cg; 1227 ufs_daddr_t ipref; 1228 int mode; 1229{ 1230 register struct fs *fs; 1231 register struct cg *cgp; 1232 struct buf *bp; 1233 u_int8_t *inosused; 1234 int error, start, len, loc, map, i; 1235 1236 fs = ip->i_fs; 1237 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1238 return (0); 1239 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1240 (int)fs->fs_cgsize, NOCRED, &bp); 1241 if (error) { 1242 brelse(bp); 1243 return (0); 1244 } 1245 cgp = (struct cg *)bp->b_data; 1246 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) { 1247 brelse(bp); 1248 return (0); 1249 } 1250 bp->b_xflags |= BX_BKGRDWRITE; 1251 cgp->cg_time = time_second; 1252 inosused = cg_inosused(cgp); 1253 if (ipref) { 1254 ipref %= fs->fs_ipg; 1255 if (isclr(inosused, ipref)) 1256 goto gotit; 1257 } 1258 start = cgp->cg_irotor / NBBY; 1259 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); 1260 loc = skpc(0xff, len, &inosused[start]); 1261 if (loc == 0) { 1262 len = start + 1; 1263 start = 0; 1264 loc = skpc(0xff, len, &inosused[0]); 1265 if (loc == 0) { 1266 printf("cg = %d, irotor = %ld, fs = %s\n", 1267 cg, (long)cgp->cg_irotor, fs->fs_fsmnt); 1268 panic("ffs_nodealloccg: map corrupted"); 1269 /* NOTREACHED */ 1270 } 1271 } 1272 i = start + len - loc; 1273 map = inosused[i]; 1274 ipref = i * NBBY; 1275 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) { 1276 if ((map & i) == 0) { 1277 cgp->cg_irotor = ipref; 1278 goto gotit; 1279 } 1280 } 1281 printf("fs = %s\n", fs->fs_fsmnt); 1282 panic("ffs_nodealloccg: block not in map"); 1283 /* NOTREACHED */ 1284gotit: 1285 if (DOINGSOFTDEP(ITOV(ip))) 1286 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref); 1287 setbit(inosused, ipref); 1288 cgp->cg_cs.cs_nifree--; 1289 fs->fs_cstotal.cs_nifree--; 1290 fs->fs_cs(fs, cg).cs_nifree--; 1291 fs->fs_fmod = 1; 1292 if ((mode & IFMT) == IFDIR) { 1293 cgp->cg_cs.cs_ndir++; 1294 fs->fs_cstotal.cs_ndir++; 1295 fs->fs_cs(fs, cg).cs_ndir++; 1296 } 1297 bdwrite(bp); 1298 return (cg * fs->fs_ipg + ipref); 1299} 1300 1301/* 1302 * Free a block or fragment. 1303 * 1304 * The specified block or fragment is placed back in the 1305 * free map. If a fragment is deallocated, a possible 1306 * block reassembly is checked. 1307 */ 1308void 1309ffs_blkfree(ip, bno, size) 1310 register struct inode *ip; 1311 ufs_daddr_t bno; 1312 long size; 1313{ 1314 register struct fs *fs; 1315 register struct cg *cgp; 1316 struct buf *bp; 1317 ufs_daddr_t blkno; 1318 int i, error, cg, blk, frags, bbase; 1319 u_int8_t *blksfree; 1320#ifdef DIAGNOSTIC 1321 struct vnode *vp; 1322#endif 1323 1324 fs = ip->i_fs; 1325#ifdef DIAGNOSTIC 1326 if ((vp = ITOV(ip)) != NULL && vp->v_mount != NULL && 1327 (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED)) 1328 panic("ffs_blkfree: deallocation on suspended filesystem"); 1329 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 1330 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 1331 printf("dev=%s, bno = %ld, bsize = %ld, size = %ld, fs = %s\n", 1332 devtoname(ip->i_dev), (long)bno, (long)fs->fs_bsize, size, 1333 fs->fs_fsmnt); 1334 panic("ffs_blkfree: bad size"); 1335 } 1336#endif 1337 if ((ip->i_devvp->v_flag & VCOPYONWRITE) && 1338 ffs_snapblkfree(ip, bno, size)) 1339 return; 1340 VOP_FREEBLKS(ip->i_devvp, fsbtodb(fs, bno), size); 1341 cg = dtog(fs, bno); 1342 if ((u_int)bno >= fs->fs_size) { 1343 printf("bad block %ld, ino %lu\n", 1344 (long)bno, (u_long)ip->i_number); 1345 ffs_fserr(fs, ip->i_uid, "bad block"); 1346 return; 1347 } 1348 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1349 (int)fs->fs_cgsize, NOCRED, &bp); 1350 if (error) { 1351 brelse(bp); 1352 return; 1353 } 1354 cgp = (struct cg *)bp->b_data; 1355 if (!cg_chkmagic(cgp)) { 1356 brelse(bp); 1357 return; 1358 } 1359 bp->b_xflags |= BX_BKGRDWRITE; 1360 cgp->cg_time = time_second; 1361 bno = dtogd(fs, bno); 1362 blksfree = cg_blksfree(cgp); 1363 if (size == fs->fs_bsize) { 1364 blkno = fragstoblks(fs, bno); 1365 if (!ffs_isfreeblock(fs, blksfree, blkno)) { 1366 printf("dev = %s, block = %ld, fs = %s\n", 1367 devtoname(ip->i_dev), (long)bno, fs->fs_fsmnt); 1368 panic("ffs_blkfree: freeing free block"); 1369 } 1370 ffs_setblock(fs, blksfree, blkno); 1371 ffs_clusteracct(fs, cgp, blkno, 1); 1372 cgp->cg_cs.cs_nbfree++; 1373 fs->fs_cstotal.cs_nbfree++; 1374 fs->fs_cs(fs, cg).cs_nbfree++; 1375 i = cbtocylno(fs, bno); 1376 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++; 1377 cg_blktot(cgp)[i]++; 1378 } else { 1379 bbase = bno - fragnum(fs, bno); 1380 /* 1381 * decrement the counts associated with the old frags 1382 */ 1383 blk = blkmap(fs, blksfree, bbase); 1384 ffs_fragacct(fs, blk, cgp->cg_frsum, -1); 1385 /* 1386 * deallocate the fragment 1387 */ 1388 frags = numfrags(fs, size); 1389 for (i = 0; i < frags; i++) { 1390 if (isset(blksfree, bno + i)) { 1391 printf("dev = %s, block = %ld, fs = %s\n", 1392 devtoname(ip->i_dev), (long)(bno + i), 1393 fs->fs_fsmnt); 1394 panic("ffs_blkfree: freeing free frag"); 1395 } 1396 setbit(blksfree, bno + i); 1397 } 1398 cgp->cg_cs.cs_nffree += i; 1399 fs->fs_cstotal.cs_nffree += i; 1400 fs->fs_cs(fs, cg).cs_nffree += i; 1401 /* 1402 * add back in counts associated with the new frags 1403 */ 1404 blk = blkmap(fs, blksfree, bbase); 1405 ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 1406 /* 1407 * if a complete block has been reassembled, account for it 1408 */ 1409 blkno = fragstoblks(fs, bbase); 1410 if (ffs_isblock(fs, blksfree, blkno)) { 1411 cgp->cg_cs.cs_nffree -= fs->fs_frag; 1412 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 1413 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 1414 ffs_clusteracct(fs, cgp, blkno, 1); 1415 cgp->cg_cs.cs_nbfree++; 1416 fs->fs_cstotal.cs_nbfree++; 1417 fs->fs_cs(fs, cg).cs_nbfree++; 1418 i = cbtocylno(fs, bbase); 1419 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++; 1420 cg_blktot(cgp)[i]++; 1421 } 1422 } 1423 fs->fs_fmod = 1; 1424 bdwrite(bp); 1425} 1426 1427#ifdef DIAGNOSTIC 1428/* 1429 * Verify allocation of a block or fragment. Returns true if block or 1430 * fragment is allocated, false if it is free. 1431 */ 1432static int 1433ffs_checkblk(ip, bno, size) 1434 struct inode *ip; 1435 ufs_daddr_t bno; 1436 long size; 1437{ 1438 struct fs *fs; 1439 struct cg *cgp; 1440 struct buf *bp; 1441 int i, error, frags, free; 1442 u_int8_t *blksfree; 1443 1444 fs = ip->i_fs; 1445 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 1446 printf("bsize = %ld, size = %ld, fs = %s\n", 1447 (long)fs->fs_bsize, size, fs->fs_fsmnt); 1448 panic("ffs_checkblk: bad size"); 1449 } 1450 if ((u_int)bno >= fs->fs_size) 1451 panic("ffs_checkblk: bad block %d", bno); 1452 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))), 1453 (int)fs->fs_cgsize, NOCRED, &bp); 1454 if (error) 1455 panic("ffs_checkblk: cg bread failed"); 1456 cgp = (struct cg *)bp->b_data; 1457 if (!cg_chkmagic(cgp)) 1458 panic("ffs_checkblk: cg magic mismatch"); 1459 bp->b_xflags |= BX_BKGRDWRITE; 1460 blksfree = cg_blksfree(cgp); 1461 bno = dtogd(fs, bno); 1462 if (size == fs->fs_bsize) { 1463 free = ffs_isblock(fs, blksfree, fragstoblks(fs, bno)); 1464 } else { 1465 frags = numfrags(fs, size); 1466 for (free = 0, i = 0; i < frags; i++) 1467 if (isset(blksfree, bno + i)) 1468 free++; 1469 if (free != 0 && free != frags) 1470 panic("ffs_checkblk: partially free fragment"); 1471 } 1472 brelse(bp); 1473 return (!free); 1474} 1475#endif /* DIAGNOSTIC */ 1476 1477/* 1478 * Free an inode. 1479 */ 1480int 1481ffs_vfree( pvp, ino, mode) 1482 struct vnode *pvp; 1483 ino_t ino; 1484 int mode; 1485{ 1486 if (DOINGSOFTDEP(pvp)) { 1487 softdep_freefile(pvp, ino, mode); 1488 return (0); 1489 } 1490 return (ffs_freefile(pvp, ino, mode)); 1491} 1492 1493/* 1494 * Do the actual free operation. 1495 * The specified inode is placed back in the free map. 1496 */ 1497 int 1498 ffs_freefile( pvp, ino, mode) 1499 struct vnode *pvp; 1500 ino_t ino; 1501 int mode; 1502{ 1503 register struct fs *fs; 1504 register struct cg *cgp; 1505 register struct inode *pip; 1506 struct buf *bp; 1507 int error, cg; 1508 u_int8_t *inosused; 1509 1510 pip = VTOI(pvp); 1511 fs = pip->i_fs; 1512 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 1513 panic("ffs_vfree: range: dev = (%d,%d), ino = %d, fs = %s", 1514 major(pip->i_dev), minor(pip->i_dev), ino, fs->fs_fsmnt); 1515 cg = ino_to_cg(fs, ino); 1516 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1517 (int)fs->fs_cgsize, NOCRED, &bp); 1518 if (error) { 1519 brelse(bp); 1520 return (error); 1521 } 1522 cgp = (struct cg *)bp->b_data; 1523 if (!cg_chkmagic(cgp)) { 1524 brelse(bp); 1525 return (0); 1526 } 1527 bp->b_xflags |= BX_BKGRDWRITE; 1528 cgp->cg_time = time_second; 1529 inosused = cg_inosused(cgp); 1530 ino %= fs->fs_ipg; 1531 if (isclr(inosused, ino)) { 1532 printf("dev = %s, ino = %lu, fs = %s\n", 1533 devtoname(pip->i_dev), (u_long)ino, fs->fs_fsmnt); 1534 if (fs->fs_ronly == 0) 1535 panic("ffs_vfree: freeing free inode"); 1536 } 1537 clrbit(inosused, ino); 1538 if (ino < cgp->cg_irotor) 1539 cgp->cg_irotor = ino; 1540 cgp->cg_cs.cs_nifree++; 1541 fs->fs_cstotal.cs_nifree++; 1542 fs->fs_cs(fs, cg).cs_nifree++; 1543 if ((mode & IFMT) == IFDIR) { 1544 cgp->cg_cs.cs_ndir--; 1545 fs->fs_cstotal.cs_ndir--; 1546 fs->fs_cs(fs, cg).cs_ndir--; 1547 } 1548 fs->fs_fmod = 1; 1549 bdwrite(bp); 1550 return (0); 1551} 1552 1553/* 1554 * Find a block of the specified size in the specified cylinder group. 1555 * 1556 * It is a panic if a request is made to find a block if none are 1557 * available. 1558 */ 1559static ufs_daddr_t 1560ffs_mapsearch(fs, cgp, bpref, allocsiz) 1561 register struct fs *fs; 1562 register struct cg *cgp; 1563 ufs_daddr_t bpref; 1564 int allocsiz; 1565{ 1566 ufs_daddr_t bno; 1567 int start, len, loc, i; 1568 int blk, field, subfield, pos; 1569 u_int8_t *blksfree; 1570 1571 /* 1572 * find the fragment by searching through the free block 1573 * map for an appropriate bit pattern 1574 */ 1575 if (bpref) 1576 start = dtogd(fs, bpref) / NBBY; 1577 else 1578 start = cgp->cg_frotor / NBBY; 1579 blksfree = cg_blksfree(cgp); 1580 len = howmany(fs->fs_fpg, NBBY) - start; 1581 loc = scanc((u_int)len, (u_char *)&blksfree[start], 1582 (u_char *)fragtbl[fs->fs_frag], 1583 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 1584 if (loc == 0) { 1585 len = start + 1; 1586 start = 0; 1587 loc = scanc((u_int)len, (u_char *)&blksfree[0], 1588 (u_char *)fragtbl[fs->fs_frag], 1589 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); 1590 if (loc == 0) { 1591 printf("start = %d, len = %d, fs = %s\n", 1592 start, len, fs->fs_fsmnt); 1593 panic("ffs_alloccg: map corrupted"); 1594 /* NOTREACHED */ 1595 } 1596 } 1597 bno = (start + len - loc) * NBBY; 1598 cgp->cg_frotor = bno; 1599 /* 1600 * found the byte in the map 1601 * sift through the bits to find the selected frag 1602 */ 1603 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 1604 blk = blkmap(fs, blksfree, bno); 1605 blk <<= 1; 1606 field = around[allocsiz]; 1607 subfield = inside[allocsiz]; 1608 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 1609 if ((blk & field) == subfield) 1610 return (bno + pos); 1611 field <<= 1; 1612 subfield <<= 1; 1613 } 1614 } 1615 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); 1616 panic("ffs_alloccg: block not in map"); 1617 return (-1); 1618} 1619 1620/* 1621 * Update the cluster map because of an allocation or free. 1622 * 1623 * Cnt == 1 means free; cnt == -1 means allocating. 1624 */ 1625static void 1626ffs_clusteracct(fs, cgp, blkno, cnt) 1627 struct fs *fs; 1628 struct cg *cgp; 1629 ufs_daddr_t blkno; 1630 int cnt; 1631{ 1632 int32_t *sump; 1633 int32_t *lp; 1634 u_char *freemapp, *mapp; 1635 int i, start, end, forw, back, map, bit; 1636 1637 if (fs->fs_contigsumsize <= 0) 1638 return; 1639 freemapp = cg_clustersfree(cgp); 1640 sump = cg_clustersum(cgp); 1641 /* 1642 * Allocate or clear the actual block. 1643 */ 1644 if (cnt > 0) 1645 setbit(freemapp, blkno); 1646 else 1647 clrbit(freemapp, blkno); 1648 /* 1649 * Find the size of the cluster going forward. 1650 */ 1651 start = blkno + 1; 1652 end = start + fs->fs_contigsumsize; 1653 if (end >= cgp->cg_nclusterblks) 1654 end = cgp->cg_nclusterblks; 1655 mapp = &freemapp[start / NBBY]; 1656 map = *mapp++; 1657 bit = 1 << (start % NBBY); 1658 for (i = start; i < end; i++) { 1659 if ((map & bit) == 0) 1660 break; 1661 if ((i & (NBBY - 1)) != (NBBY - 1)) { 1662 bit <<= 1; 1663 } else { 1664 map = *mapp++; 1665 bit = 1; 1666 } 1667 } 1668 forw = i - start; 1669 /* 1670 * Find the size of the cluster going backward. 1671 */ 1672 start = blkno - 1; 1673 end = start - fs->fs_contigsumsize; 1674 if (end < 0) 1675 end = -1; 1676 mapp = &freemapp[start / NBBY]; 1677 map = *mapp--; 1678 bit = 1 << (start % NBBY); 1679 for (i = start; i > end; i--) { 1680 if ((map & bit) == 0) 1681 break; 1682 if ((i & (NBBY - 1)) != 0) { 1683 bit >>= 1; 1684 } else { 1685 map = *mapp--; 1686 bit = 1 << (NBBY - 1); 1687 } 1688 } 1689 back = start - i; 1690 /* 1691 * Account for old cluster and the possibly new forward and 1692 * back clusters. 1693 */ 1694 i = back + forw + 1; 1695 if (i > fs->fs_contigsumsize) 1696 i = fs->fs_contigsumsize; 1697 sump[i] += cnt; 1698 if (back > 0) 1699 sump[back] -= cnt; 1700 if (forw > 0) 1701 sump[forw] -= cnt; 1702 /* 1703 * Update cluster summary information. 1704 */ 1705 lp = &sump[fs->fs_contigsumsize]; 1706 for (i = fs->fs_contigsumsize; i > 0; i--) 1707 if (*lp-- > 0) 1708 break; 1709 fs->fs_maxcluster[cgp->cg_cgx] = i; 1710} 1711 1712/* 1713 * Fserr prints the name of a file system with an error diagnostic. 1714 * 1715 * The form of the error message is: 1716 * fs: error message 1717 */ 1718static void 1719ffs_fserr(fs, uid, cp) 1720 struct fs *fs; 1721 u_int uid; 1722 char *cp; 1723{ 1724 struct proc *p = curproc; /* XXX */ 1725 1726 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1, 1727 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp); 1728} 1729