1 /*- 2 * modified for Lites 1.1 3 * 4 * Aug 1995, Godmar Back (gback@cs.utah.edu) 5 * University of Utah, Department of Computer Science 6 */ 7 /*- 8 * Copyright (c) 1982, 1986, 1989, 1993 9 * The Regents of the University of California. All rights reserved. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)ffs_alloc.c 8.8 (Berkeley) 2/21/94 36 * $FreeBSD$ 37 */ 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/conf.h> 42 #include <sys/vnode.h> 43 #include <sys/stat.h> 44 #include <sys/mount.h> 45 #include <sys/sysctl.h> 46 #include <sys/syslog.h> 47 #include <sys/buf.h> 48 49 #include <fs/ext2fs/fs.h> 50 #include <fs/ext2fs/inode.h> 51 #include <fs/ext2fs/ext2_mount.h> 52 #include <fs/ext2fs/ext2fs.h> 53 #include <fs/ext2fs/ext2_extern.h> 54 55 static daddr_t ext2_alloccg(struct inode *, int, daddr_t, int); 56 static daddr_t ext2_clusteralloc(struct inode *, int, daddr_t, int); 57 static u_long ext2_dirpref(struct inode *); 58 static void ext2_fserr(struct m_ext2fs *, uid_t, char *); 59 static u_long ext2_hashalloc(struct inode *, int, long, int, 60 daddr_t (*)(struct inode *, int, daddr_t, 61 int)); 62 static daddr_t ext2_nodealloccg(struct inode *, int, daddr_t, int); 63 static daddr_t ext2_mapsearch(struct m_ext2fs *, char *, daddr_t); 64 65 /* 66 * Allocate a block in the filesystem. 67 * 68 * A preference may be optionally specified. If a preference is given 69 * the following hierarchy is used to allocate a block: 70 * 1) allocate the requested block. 71 * 2) allocate a rotationally optimal block in the same cylinder. 72 * 3) allocate a block in the same cylinder group. 73 * 4) quadradically rehash into other cylinder groups, until an 74 * available block is located. 75 * If no block preference is given the following hierarchy is used 76 * to allocate a block: 77 * 1) allocate a block in the cylinder group that contains the 78 * inode for the file. 79 * 2) quadradically rehash into other cylinder groups, until an 80 * available block is located. 81 */ 82 int 83 ext2_alloc(struct inode *ip, daddr_t lbn, e4fs_daddr_t bpref, int size, 84 struct ucred *cred, e4fs_daddr_t *bnp) 85 { 86 struct m_ext2fs *fs; 87 struct ext2mount *ump; 88 int32_t bno; 89 int cg; 90 *bnp = 0; 91 fs = ip->i_e2fs; 92 ump = ip->i_ump; 93 mtx_assert(EXT2_MTX(ump), MA_OWNED); 94 #ifdef INVARIANTS 95 if ((u_int)size > fs->e2fs_bsize || blkoff(fs, size) != 0) { 96 vn_printf(ip->i_devvp, "bsize = %lu, size = %d, fs = %s\n", 97 (long unsigned int)fs->e2fs_bsize, size, fs->e2fs_fsmnt); 98 panic("ext2_alloc: bad size"); 99 } 100 if (cred == NOCRED) 101 panic("ext2_alloc: missing credential"); 102 #endif /* INVARIANTS */ 103 if (size == fs->e2fs_bsize && fs->e2fs->e2fs_fbcount == 0) 104 goto nospace; 105 if (cred->cr_uid != 0 && 106 fs->e2fs->e2fs_fbcount < fs->e2fs->e2fs_rbcount) 107 goto nospace; 108 if (bpref >= fs->e2fs->e2fs_bcount) 109 bpref = 0; 110 if (bpref == 0) 111 cg = ino_to_cg(fs, ip->i_number); 112 else 113 cg = dtog(fs, bpref); 114 bno = (daddr_t)ext2_hashalloc(ip, cg, bpref, fs->e2fs_bsize, 115 ext2_alloccg); 116 if (bno > 0) { 117 /* set next_alloc fields as done in block_getblk */ 118 ip->i_next_alloc_block = lbn; 119 ip->i_next_alloc_goal = bno; 120 121 ip->i_blocks += btodb(fs->e2fs_bsize); 122 ip->i_flag |= IN_CHANGE | IN_UPDATE; 123 *bnp = bno; 124 return (0); 125 } 126 nospace: 127 EXT2_UNLOCK(ump); 128 ext2_fserr(fs, cred->cr_uid, "filesystem full"); 129 uprintf("\n%s: write failed, filesystem is full\n", fs->e2fs_fsmnt); 130 return (ENOSPC); 131 } 132 133 /* 134 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 135 * 136 * The vnode and an array of buffer pointers for a range of sequential 137 * logical blocks to be made contiguous is given. The allocator attempts 138 * to find a range of sequential blocks starting as close as possible to 139 * an fs_rotdelay offset from the end of the allocation for the logical 140 * block immediately preceding the current range. If successful, the 141 * physical block numbers in the buffer pointers and in the inode are 142 * changed to reflect the new allocation. If unsuccessful, the allocation 143 * is left unchanged. The success in doing the reallocation is returned. 144 * Note that the error return is not reflected back to the user. Rather 145 * the previous block allocation will be used. 146 */ 147 148 static SYSCTL_NODE(_vfs, OID_AUTO, ext2fs, CTLFLAG_RW, 0, "EXT2FS filesystem"); 149 150 static int doasyncfree = 1; 151 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, 152 "Use asychronous writes to update block pointers when freeing blocks"); 153 154 static int doreallocblks = 1; 155 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); 156 157 int 158 ext2_reallocblks(struct vop_reallocblks_args *ap) 159 { 160 struct m_ext2fs *fs; 161 struct inode *ip; 162 struct vnode *vp; 163 struct buf *sbp, *ebp; 164 uint32_t *bap, *sbap, *ebap = 0; 165 struct ext2mount *ump; 166 struct cluster_save *buflist; 167 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; 168 e2fs_lbn_t start_lbn, end_lbn; 169 int soff; 170 e2fs_daddr_t newblk, blkno; 171 int i, len, start_lvl, end_lvl, pref, ssize; 172 173 if (doreallocblks == 0) 174 return (ENOSPC); 175 176 vp = ap->a_vp; 177 ip = VTOI(vp); 178 fs = ip->i_e2fs; 179 ump = ip->i_ump; 180 181 if (fs->e2fs_contigsumsize <= 0) 182 return (ENOSPC); 183 184 buflist = ap->a_buflist; 185 len = buflist->bs_nchildren; 186 start_lbn = buflist->bs_children[0]->b_lblkno; 187 end_lbn = start_lbn + len - 1; 188 #ifdef INVARIANTS 189 for (i = 1; i < len; i++) 190 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 191 panic("ext2_reallocblks: non-cluster"); 192 #endif 193 /* 194 * If the cluster crosses the boundary for the first indirect 195 * block, leave space for the indirect block. Indirect blocks 196 * are initially laid out in a position after the last direct 197 * block. Block reallocation would usually destroy locality by 198 * moving the indirect block out of the way to make room for 199 * data blocks if we didn't compensate here. We should also do 200 * this for other indirect block boundaries, but it is only 201 * important for the first one. 202 */ 203 if (start_lbn < NDADDR && end_lbn >= NDADDR) 204 return (ENOSPC); 205 /* 206 * If the latest allocation is in a new cylinder group, assume that 207 * the filesystem has decided to move and do not force it back to 208 * the previous cylinder group. 209 */ 210 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 211 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 212 return (ENOSPC); 213 if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) || 214 ext2_getlbns(vp, end_lbn, end_ap, &end_lvl)) 215 return (ENOSPC); 216 /* 217 * Get the starting offset and block map for the first block. 218 */ 219 if (start_lvl == 0) { 220 sbap = &ip->i_db[0]; 221 soff = start_lbn; 222 } else { 223 idp = &start_ap[start_lvl - 1]; 224 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) { 225 brelse(sbp); 226 return (ENOSPC); 227 } 228 sbap = (u_int *)sbp->b_data; 229 soff = idp->in_off; 230 } 231 /* 232 * If the block range spans two block maps, get the second map. 233 */ 234 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 235 ssize = len; 236 } else { 237 #ifdef INVARIANTS 238 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn) 239 panic("ext2_reallocblks: start == end"); 240 #endif 241 ssize = len - (idp->in_off + 1); 242 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp)) 243 goto fail; 244 ebap = (u_int *)ebp->b_data; 245 } 246 /* 247 * Find the preferred location for the cluster. 248 */ 249 EXT2_LOCK(ump); 250 pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0); 251 /* 252 * Search the block map looking for an allocation of the desired size. 253 */ 254 if ((newblk = (e2fs_daddr_t)ext2_hashalloc(ip, dtog(fs, pref), pref, 255 len, ext2_clusteralloc)) == 0){ 256 EXT2_UNLOCK(ump); 257 goto fail; 258 } 259 /* 260 * We have found a new contiguous block. 261 * 262 * First we have to replace the old block pointers with the new 263 * block pointers in the inode and indirect blocks associated 264 * with the file. 265 */ 266 #ifdef DEBUG 267 printf("realloc: ino %ju, lbns %jd-%jd\n\told:", 268 (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); 269 #endif /* DEBUG */ 270 blkno = newblk; 271 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) { 272 if (i == ssize) { 273 bap = ebap; 274 soff = -i; 275 } 276 #ifdef INVARIANTS 277 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap)) 278 panic("ext2_reallocblks: alloc mismatch"); 279 #endif 280 #ifdef DEBUG 281 printf(" %d,", *bap); 282 #endif /* DEBUG */ 283 *bap++ = blkno; 284 } 285 /* 286 * Next we must write out the modified inode and indirect blocks. 287 * For strict correctness, the writes should be synchronous since 288 * the old block values may have been written to disk. In practise 289 * they are almost never written, but if we are concerned about 290 * strict correctness, the `doasyncfree' flag should be set to zero. 291 * 292 * The test on `doasyncfree' should be changed to test a flag 293 * that shows whether the associated buffers and inodes have 294 * been written. The flag should be set when the cluster is 295 * started and cleared whenever the buffer or inode is flushed. 296 * We can then check below to see if it is set, and do the 297 * synchronous write only when it has been cleared. 298 */ 299 if (sbap != &ip->i_db[0]) { 300 if (doasyncfree) 301 bdwrite(sbp); 302 else 303 bwrite(sbp); 304 } else { 305 ip->i_flag |= IN_CHANGE | IN_UPDATE; 306 if (!doasyncfree) 307 ext2_update(vp, 1); 308 } 309 if (ssize < len) { 310 if (doasyncfree) 311 bdwrite(ebp); 312 else 313 bwrite(ebp); 314 } 315 /* 316 * Last, free the old blocks and assign the new blocks to the buffers. 317 */ 318 #ifdef DEBUG 319 printf("\n\tnew:"); 320 #endif /* DEBUG */ 321 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) { 322 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), 323 fs->e2fs_bsize); 324 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 325 #ifdef DEBUG 326 printf(" %d,", blkno); 327 #endif /* DEBUG */ 328 } 329 #ifdef DEBUG 330 printf("\n"); 331 #endif /* DEBUG */ 332 return (0); 333 334 fail: 335 if (ssize < len) 336 brelse(ebp); 337 if (sbap != &ip->i_db[0]) 338 brelse(sbp); 339 return (ENOSPC); 340 } 341 342 /* 343 * Allocate an inode in the filesystem. 344 * 345 */ 346 int 347 ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp) 348 { 349 struct timespec ts; 350 struct inode *pip; 351 struct m_ext2fs *fs; 352 struct inode *ip; 353 struct ext2mount *ump; 354 ino_t ino, ipref; 355 int i, error, cg; 356 357 *vpp = NULL; 358 pip = VTOI(pvp); 359 fs = pip->i_e2fs; 360 ump = pip->i_ump; 361 362 EXT2_LOCK(ump); 363 if (fs->e2fs->e2fs_ficount == 0) 364 goto noinodes; 365 /* 366 * If it is a directory then obtain a cylinder group based on 367 * ext2_dirpref else obtain it using ino_to_cg. The preferred inode is 368 * always the next inode. 369 */ 370 if ((mode & IFMT) == IFDIR) { 371 cg = ext2_dirpref(pip); 372 if (fs->e2fs_contigdirs[cg] < 255) 373 fs->e2fs_contigdirs[cg]++; 374 } else { 375 cg = ino_to_cg(fs, pip->i_number); 376 if (fs->e2fs_contigdirs[cg] > 0) 377 fs->e2fs_contigdirs[cg]--; 378 } 379 ipref = cg * fs->e2fs->e2fs_ipg + 1; 380 ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg); 381 382 if (ino == 0) 383 goto noinodes; 384 error = VFS_VGET(pvp->v_mount, ino, LK_EXCLUSIVE, vpp); 385 if (error) { 386 ext2_vfree(pvp, ino, mode); 387 return (error); 388 } 389 ip = VTOI(*vpp); 390 391 /* 392 * The question is whether using VGET was such good idea at all: 393 * Linux doesn't read the old inode in when it is allocating a 394 * new one. I will set at least i_size and i_blocks to zero. 395 */ 396 ip->i_size = 0; 397 ip->i_blocks = 0; 398 ip->i_mode = 0; 399 ip->i_flags = 0; 400 /* now we want to make sure that the block pointers are zeroed out */ 401 for (i = 0; i < NDADDR; i++) 402 ip->i_db[i] = 0; 403 for (i = 0; i < NIADDR; i++) 404 ip->i_ib[i] = 0; 405 406 /* 407 * Set up a new generation number for this inode. 408 * XXX check if this makes sense in ext2 409 */ 410 if (ip->i_gen == 0 || ++ip->i_gen == 0) 411 ip->i_gen = random() / 2 + 1; 412 413 vfs_timestamp(&ts); 414 ip->i_birthtime = ts.tv_sec; 415 ip->i_birthnsec = ts.tv_nsec; 416 417 /* 418 printf("ext2_valloc: allocated inode %d\n", ino); 419 */ 420 return (0); 421 noinodes: 422 EXT2_UNLOCK(ump); 423 ext2_fserr(fs, cred->cr_uid, "out of inodes"); 424 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->e2fs_fsmnt); 425 return (ENOSPC); 426 } 427 428 /* 429 * Find a cylinder to place a directory. 430 * 431 * The policy implemented by this algorithm is to allocate a 432 * directory inode in the same cylinder group as its parent 433 * directory, but also to reserve space for its files inodes 434 * and data. Restrict the number of directories which may be 435 * allocated one after another in the same cylinder group 436 * without intervening allocation of files. 437 * 438 * If we allocate a first level directory then force allocation 439 * in another cylinder group. 440 * 441 */ 442 static u_long 443 ext2_dirpref(struct inode *pip) 444 { 445 struct m_ext2fs *fs; 446 int cg, prefcg, cgsize; 447 u_int avgifree, avgbfree, avgndir, curdirsize; 448 u_int minifree, minbfree, maxndir; 449 u_int mincg, minndir; 450 u_int dirsize, maxcontigdirs; 451 452 mtx_assert(EXT2_MTX(pip->i_ump), MA_OWNED); 453 fs = pip->i_e2fs; 454 455 avgifree = fs->e2fs->e2fs_ficount / fs->e2fs_gcount; 456 avgbfree = fs->e2fs->e2fs_fbcount / fs->e2fs_gcount; 457 avgndir = fs->e2fs_total_dir / fs->e2fs_gcount; 458 459 /* 460 * Force allocation in another cg if creating a first level dir. 461 */ 462 ASSERT_VOP_LOCKED(ITOV(pip), "ext2fs_dirpref"); 463 if (ITOV(pip)->v_vflag & VV_ROOT) { 464 prefcg = arc4random() % fs->e2fs_gcount; 465 mincg = prefcg; 466 minndir = fs->e2fs_ipg; 467 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 468 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir && 469 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree && 470 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) { 471 mincg = cg; 472 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs; 473 } 474 for (cg = 0; cg < prefcg; cg++) 475 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir && 476 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree && 477 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) { 478 mincg = cg; 479 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs; 480 } 481 482 return (mincg); 483 } 484 485 /* 486 * Count various limits which used for 487 * optimal allocation of a directory inode. 488 */ 489 maxndir = min(avgndir + fs->e2fs_ipg / 16, fs->e2fs_ipg); 490 minifree = avgifree - avgifree / 4; 491 if (minifree < 1) 492 minifree = 1; 493 minbfree = avgbfree - avgbfree / 4; 494 if (minbfree < 1) 495 minbfree = 1; 496 cgsize = fs->e2fs_fsize * fs->e2fs_fpg; 497 dirsize = AVGDIRSIZE; 498 curdirsize = avgndir ? (cgsize - avgbfree * fs->e2fs_bsize) / avgndir : 0; 499 if (dirsize < curdirsize) 500 dirsize = curdirsize; 501 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255); 502 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR); 503 if (maxcontigdirs == 0) 504 maxcontigdirs = 1; 505 506 /* 507 * Limit number of dirs in one cg and reserve space for 508 * regular files, but only if we have no deficit in 509 * inodes or space. 510 */ 511 prefcg = ino_to_cg(fs, pip->i_number); 512 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 513 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir && 514 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree && 515 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) { 516 if (fs->e2fs_contigdirs[cg] < maxcontigdirs) 517 return (cg); 518 } 519 for (cg = 0; cg < prefcg; cg++) 520 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir && 521 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree && 522 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) { 523 if (fs->e2fs_contigdirs[cg] < maxcontigdirs) 524 return (cg); 525 } 526 /* 527 * This is a backstop when we have deficit in space. 528 */ 529 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 530 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree) 531 return (cg); 532 for (cg = 0; cg < prefcg; cg++) 533 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree) 534 break; 535 return (cg); 536 } 537 538 /* 539 * Select the desired position for the next block in a file. 540 * 541 * we try to mimic what Remy does in inode_getblk/block_getblk 542 * 543 * we note: blocknr == 0 means that we're about to allocate either 544 * a direct block or a pointer block at the first level of indirection 545 * (In other words, stuff that will go in i_db[] or i_ib[]) 546 * 547 * blocknr != 0 means that we're allocating a block that is none 548 * of the above. Then, blocknr tells us the number of the block 549 * that will hold the pointer 550 */ 551 e4fs_daddr_t 552 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, e2fs_daddr_t *bap, 553 e2fs_daddr_t blocknr) 554 { 555 int tmp; 556 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED); 557 558 /* if the next block is actually what we thought it is, 559 then set the goal to what we thought it should be 560 */ 561 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0) 562 return ip->i_next_alloc_goal; 563 564 /* now check whether we were provided with an array that basically 565 tells us previous blocks to which we want to stay closeby 566 */ 567 if (bap) 568 for (tmp = indx - 1; tmp >= 0; tmp--) 569 if (bap[tmp]) 570 return bap[tmp]; 571 572 /* else let's fall back to the blocknr, or, if there is none, 573 follow the rule that a block should be allocated near its inode 574 */ 575 return blocknr ? blocknr : 576 (e2fs_daddr_t)(ip->i_block_group * 577 EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) + 578 ip->i_e2fs->e2fs->e2fs_first_dblock; 579 } 580 581 /* 582 * Implement the cylinder overflow algorithm. 583 * 584 * The policy implemented by this algorithm is: 585 * 1) allocate the block in its requested cylinder group. 586 * 2) quadradically rehash on the cylinder group number. 587 * 3) brute force search for a free block. 588 */ 589 static u_long 590 ext2_hashalloc(struct inode *ip, int cg, long pref, int size, 591 daddr_t (*allocator)(struct inode *, int, daddr_t, int)) 592 { 593 struct m_ext2fs *fs; 594 ino_t result; 595 int i, icg = cg; 596 597 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED); 598 fs = ip->i_e2fs; 599 /* 600 * 1: preferred cylinder group 601 */ 602 result = (*allocator)(ip, cg, pref, size); 603 if (result) 604 return (result); 605 /* 606 * 2: quadratic rehash 607 */ 608 for (i = 1; i < fs->e2fs_gcount; i *= 2) { 609 cg += i; 610 if (cg >= fs->e2fs_gcount) 611 cg -= fs->e2fs_gcount; 612 result = (*allocator)(ip, cg, 0, size); 613 if (result) 614 return (result); 615 } 616 /* 617 * 3: brute force search 618 * Note that we start at i == 2, since 0 was checked initially, 619 * and 1 is always checked in the quadratic rehash. 620 */ 621 cg = (icg + 2) % fs->e2fs_gcount; 622 for (i = 2; i < fs->e2fs_gcount; i++) { 623 result = (*allocator)(ip, cg, 0, size); 624 if (result) 625 return (result); 626 cg++; 627 if (cg == fs->e2fs_gcount) 628 cg = 0; 629 } 630 return (0); 631 } 632 633 /* 634 * Determine whether a block can be allocated. 635 * 636 * Check to see if a block of the appropriate size is available, 637 * and if it is, allocate it. 638 */ 639 static daddr_t 640 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size) 641 { 642 struct m_ext2fs *fs; 643 struct buf *bp; 644 struct ext2mount *ump; 645 daddr_t bno, runstart, runlen; 646 int bit, loc, end, error, start; 647 char *bbp; 648 /* XXX ondisk32 */ 649 fs = ip->i_e2fs; 650 ump = ip->i_ump; 651 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0) 652 return (0); 653 EXT2_UNLOCK(ump); 654 error = bread(ip->i_devvp, fsbtodb(fs, 655 fs->e2fs_gd[cg].ext2bgd_b_bitmap), 656 (int)fs->e2fs_bsize, NOCRED, &bp); 657 if (error) { 658 brelse(bp); 659 EXT2_LOCK(ump); 660 return (0); 661 } 662 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0) { 663 /* 664 * Another thread allocated the last block in this 665 * group while we were waiting for the buffer. 666 */ 667 brelse(bp); 668 EXT2_LOCK(ump); 669 return (0); 670 } 671 bbp = (char *)bp->b_data; 672 673 if (dtog(fs, bpref) != cg) 674 bpref = 0; 675 if (bpref != 0) { 676 bpref = dtogd(fs, bpref); 677 /* 678 * if the requested block is available, use it 679 */ 680 if (isclr(bbp, bpref)) { 681 bno = bpref; 682 goto gotit; 683 } 684 } 685 /* 686 * no blocks in the requested cylinder, so take next 687 * available one in this cylinder group. 688 * first try to get 8 contigous blocks, then fall back to a single 689 * block. 690 */ 691 if (bpref) 692 start = dtogd(fs, bpref) / NBBY; 693 else 694 start = 0; 695 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start; 696 retry: 697 runlen = 0; 698 runstart = 0; 699 for (loc = start; loc < end; loc++) { 700 if (bbp[loc] == (char)0xff) { 701 runlen = 0; 702 continue; 703 } 704 705 /* Start of a run, find the number of high clear bits. */ 706 if (runlen == 0) { 707 bit = fls(bbp[loc]); 708 runlen = NBBY - bit; 709 runstart = loc * NBBY + bit; 710 } else if (bbp[loc] == 0) { 711 /* Continue a run. */ 712 runlen += NBBY; 713 } else { 714 /* 715 * Finish the current run. If it isn't long 716 * enough, start a new one. 717 */ 718 bit = ffs(bbp[loc]) - 1; 719 runlen += bit; 720 if (runlen >= 8) { 721 bno = runstart; 722 goto gotit; 723 } 724 725 /* Run was too short, start a new one. */ 726 bit = fls(bbp[loc]); 727 runlen = NBBY - bit; 728 runstart = loc * NBBY + bit; 729 } 730 731 /* If the current run is long enough, use it. */ 732 if (runlen >= 8) { 733 bno = runstart; 734 goto gotit; 735 } 736 } 737 if (start != 0) { 738 end = start; 739 start = 0; 740 goto retry; 741 } 742 743 bno = ext2_mapsearch(fs, bbp, bpref); 744 if (bno < 0){ 745 brelse(bp); 746 EXT2_LOCK(ump); 747 return (0); 748 } 749 gotit: 750 #ifdef INVARIANTS 751 if (isset(bbp, bno)) { 752 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n", 753 cg, (intmax_t)bno, fs->e2fs_fsmnt); 754 panic("ext2fs_alloccg: dup alloc"); 755 } 756 #endif 757 setbit(bbp, bno); 758 EXT2_LOCK(ump); 759 ext2_clusteracct(fs, bbp, cg, bno, -1); 760 fs->e2fs->e2fs_fbcount--; 761 fs->e2fs_gd[cg].ext2bgd_nbfree--; 762 fs->e2fs_fmod = 1; 763 EXT2_UNLOCK(ump); 764 bdwrite(bp); 765 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno); 766 } 767 768 /* 769 * Determine whether a cluster can be allocated. 770 */ 771 static daddr_t 772 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len) 773 { 774 struct m_ext2fs *fs; 775 struct ext2mount *ump; 776 struct buf *bp; 777 char *bbp; 778 int bit, error, got, i, loc, run; 779 int32_t *lp; 780 daddr_t bno; 781 782 fs = ip->i_e2fs; 783 ump = ip->i_ump; 784 785 if (fs->e2fs_maxcluster[cg] < len) 786 return (0); 787 788 EXT2_UNLOCK(ump); 789 error = bread(ip->i_devvp, 790 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap), 791 (int)fs->e2fs_bsize, NOCRED, &bp); 792 if (error) 793 goto fail_lock; 794 795 bbp = (char *)bp->b_data; 796 EXT2_LOCK(ump); 797 /* 798 * Check to see if a cluster of the needed size (or bigger) is 799 * available in this cylinder group. 800 */ 801 lp = &fs->e2fs_clustersum[cg].cs_sum[len]; 802 for (i = len; i <= fs->e2fs_contigsumsize; i++) 803 if (*lp++ > 0) 804 break; 805 if (i > fs->e2fs_contigsumsize) { 806 /* 807 * Update the cluster summary information to reflect 808 * the true maximum-sized cluster so that future cluster 809 * allocation requests can avoid reading the bitmap only 810 * to find no cluster. 811 */ 812 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1]; 813 for (i = len - 1; i > 0; i--) 814 if (*lp-- > 0) 815 break; 816 fs->e2fs_maxcluster[cg] = i; 817 goto fail; 818 } 819 EXT2_UNLOCK(ump); 820 821 /* Search the bitmap to find a big enough cluster like in FFS. */ 822 if (dtog(fs, bpref) != cg) 823 bpref = 0; 824 if (bpref != 0) 825 bpref = dtogd(fs, bpref); 826 loc = bpref / NBBY; 827 bit = 1 << (bpref % NBBY); 828 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) { 829 if ((bbp[loc] & bit) != 0) 830 run = 0; 831 else { 832 run++; 833 if (run == len) 834 break; 835 } 836 if ((got & (NBBY - 1)) != (NBBY - 1)) 837 bit <<= 1; 838 else { 839 loc++; 840 bit = 1; 841 } 842 } 843 844 if (got >= fs->e2fs->e2fs_fpg) 845 goto fail_lock; 846 847 /* Allocate the cluster that we found. */ 848 for (i = 1; i < len; i++) 849 if (!isclr(bbp, got - run + i)) 850 panic("ext2_clusteralloc: map mismatch"); 851 852 bno = got - run + 1; 853 if (bno >= fs->e2fs->e2fs_fpg) 854 panic("ext2_clusteralloc: allocated out of group"); 855 856 EXT2_LOCK(ump); 857 for (i = 0; i < len; i += fs->e2fs_fpb) { 858 setbit(bbp, bno + i); 859 ext2_clusteracct(fs, bbp, cg, bno + i, -1); 860 fs->e2fs->e2fs_fbcount--; 861 fs->e2fs_gd[cg].ext2bgd_nbfree--; 862 } 863 fs->e2fs_fmod = 1; 864 EXT2_UNLOCK(ump); 865 866 bdwrite(bp); 867 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno); 868 869 fail_lock: 870 EXT2_LOCK(ump); 871 fail: 872 brelse(bp); 873 return (0); 874 } 875 876 /* 877 * Determine whether an inode can be allocated. 878 * 879 * Check to see if an inode is available, and if it is, 880 * allocate it using tode in the specified cylinder group. 881 */ 882 static daddr_t 883 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode) 884 { 885 struct m_ext2fs *fs; 886 struct buf *bp; 887 struct ext2mount *ump; 888 int error, start, len; 889 char *ibp, *loc; 890 ipref--; /* to avoid a lot of (ipref -1) */ 891 if (ipref == -1) 892 ipref = 0; 893 fs = ip->i_e2fs; 894 ump = ip->i_ump; 895 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0) 896 return (0); 897 EXT2_UNLOCK(ump); 898 error = bread(ip->i_devvp, fsbtodb(fs, 899 fs->e2fs_gd[cg].ext2bgd_i_bitmap), 900 (int)fs->e2fs_bsize, NOCRED, &bp); 901 if (error) { 902 brelse(bp); 903 EXT2_LOCK(ump); 904 return (0); 905 } 906 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0) { 907 /* 908 * Another thread allocated the last i-node in this 909 * group while we were waiting for the buffer. 910 */ 911 brelse(bp); 912 EXT2_LOCK(ump); 913 return (0); 914 } 915 ibp = (char *)bp->b_data; 916 if (ipref) { 917 ipref %= fs->e2fs->e2fs_ipg; 918 if (isclr(ibp, ipref)) 919 goto gotit; 920 } 921 start = ipref / NBBY; 922 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY); 923 loc = memcchr(&ibp[start], 0xff, len); 924 if (loc == NULL) { 925 len = start + 1; 926 start = 0; 927 loc = memcchr(&ibp[start], 0xff, len); 928 if (loc == NULL) { 929 printf("cg = %d, ipref = %lld, fs = %s\n", 930 cg, (long long)ipref, fs->e2fs_fsmnt); 931 panic("ext2fs_nodealloccg: map corrupted"); 932 /* NOTREACHED */ 933 } 934 } 935 ipref = (loc - ibp) * NBBY + ffs(~*loc) - 1; 936 gotit: 937 setbit(ibp, ipref); 938 EXT2_LOCK(ump); 939 fs->e2fs_gd[cg].ext2bgd_nifree--; 940 fs->e2fs->e2fs_ficount--; 941 fs->e2fs_fmod = 1; 942 if ((mode & IFMT) == IFDIR) { 943 fs->e2fs_gd[cg].ext2bgd_ndirs++; 944 fs->e2fs_total_dir++; 945 } 946 EXT2_UNLOCK(ump); 947 bdwrite(bp); 948 return (cg * fs->e2fs->e2fs_ipg + ipref +1); 949 } 950 951 /* 952 * Free a block or fragment. 953 * 954 */ 955 void 956 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size) 957 { 958 struct m_ext2fs *fs; 959 struct buf *bp; 960 struct ext2mount *ump; 961 int cg, error; 962 char *bbp; 963 964 fs = ip->i_e2fs; 965 ump = ip->i_ump; 966 cg = dtog(fs, bno); 967 if ((u_int)bno >= fs->e2fs->e2fs_bcount) { 968 printf("bad block %lld, ino %ju\n", (long long)bno, 969 (uintmax_t)ip->i_number); 970 ext2_fserr(fs, ip->i_uid, "bad block"); 971 return; 972 } 973 error = bread(ip->i_devvp, 974 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap), 975 (int)fs->e2fs_bsize, NOCRED, &bp); 976 if (error) { 977 brelse(bp); 978 return; 979 } 980 bbp = (char *)bp->b_data; 981 bno = dtogd(fs, bno); 982 if (isclr(bbp, bno)) { 983 printf("block = %lld, fs = %s\n", 984 (long long)bno, fs->e2fs_fsmnt); 985 panic("ext2_blkfree: freeing free block"); 986 } 987 clrbit(bbp, bno); 988 EXT2_LOCK(ump); 989 ext2_clusteracct(fs, bbp, cg, bno, 1); 990 fs->e2fs->e2fs_fbcount++; 991 fs->e2fs_gd[cg].ext2bgd_nbfree++; 992 fs->e2fs_fmod = 1; 993 EXT2_UNLOCK(ump); 994 bdwrite(bp); 995 } 996 997 /* 998 * Free an inode. 999 * 1000 */ 1001 int 1002 ext2_vfree(struct vnode *pvp, ino_t ino, int mode) 1003 { 1004 struct m_ext2fs *fs; 1005 struct inode *pip; 1006 struct buf *bp; 1007 struct ext2mount *ump; 1008 int error, cg; 1009 char * ibp; 1010 1011 pip = VTOI(pvp); 1012 fs = pip->i_e2fs; 1013 ump = pip->i_ump; 1014 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount) 1015 panic("ext2_vfree: range: devvp = %p, ino = %ju, fs = %s", 1016 pip->i_devvp, (uintmax_t)ino, fs->e2fs_fsmnt); 1017 1018 cg = ino_to_cg(fs, ino); 1019 error = bread(pip->i_devvp, 1020 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_i_bitmap), 1021 (int)fs->e2fs_bsize, NOCRED, &bp); 1022 if (error) { 1023 brelse(bp); 1024 return (0); 1025 } 1026 ibp = (char *)bp->b_data; 1027 ino = (ino - 1) % fs->e2fs->e2fs_ipg; 1028 if (isclr(ibp, ino)) { 1029 printf("ino = %llu, fs = %s\n", 1030 (unsigned long long)ino, fs->e2fs_fsmnt); 1031 if (fs->e2fs_ronly == 0) 1032 panic("ext2_vfree: freeing free inode"); 1033 } 1034 clrbit(ibp, ino); 1035 EXT2_LOCK(ump); 1036 fs->e2fs->e2fs_ficount++; 1037 fs->e2fs_gd[cg].ext2bgd_nifree++; 1038 if ((mode & IFMT) == IFDIR) { 1039 fs->e2fs_gd[cg].ext2bgd_ndirs--; 1040 fs->e2fs_total_dir--; 1041 } 1042 fs->e2fs_fmod = 1; 1043 EXT2_UNLOCK(ump); 1044 bdwrite(bp); 1045 return (0); 1046 } 1047 1048 /* 1049 * Find a block in the specified cylinder group. 1050 * 1051 * It is a panic if a request is made to find a block if none are 1052 * available. 1053 */ 1054 static daddr_t 1055 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref) 1056 { 1057 char *loc; 1058 int start, len; 1059 1060 /* 1061 * find the fragment by searching through the free block 1062 * map for an appropriate bit pattern 1063 */ 1064 if (bpref) 1065 start = dtogd(fs, bpref) / NBBY; 1066 else 1067 start = 0; 1068 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start; 1069 loc = memcchr(&bbp[start], 0xff, len); 1070 if (loc == NULL) { 1071 len = start + 1; 1072 start = 0; 1073 loc = memcchr(&bbp[start], 0xff, len); 1074 if (loc == NULL) { 1075 printf("start = %d, len = %d, fs = %s\n", 1076 start, len, fs->e2fs_fsmnt); 1077 panic("ext2_mapsearch: map corrupted"); 1078 /* NOTREACHED */ 1079 } 1080 } 1081 return ((loc - bbp) * NBBY + ffs(~*loc) - 1); 1082 } 1083 1084 /* 1085 * Fserr prints the name of a filesystem with an error diagnostic. 1086 * 1087 * The form of the error message is: 1088 * fs: error message 1089 */ 1090 static void 1091 ext2_fserr(struct m_ext2fs *fs, uid_t uid, char *cp) 1092 { 1093 1094 log(LOG_ERR, "uid %u on %s: %s\n", uid, fs->e2fs_fsmnt, cp); 1095 } 1096 1097 int 1098 cg_has_sb(int i) 1099 { 1100 int a3, a5, a7; 1101 1102 if (i == 0 || i == 1) 1103 return 1; 1104 for (a3 = 3, a5 = 5, a7 = 7; 1105 a3 <= i || a5 <= i || a7 <= i; 1106 a3 *= 3, a5 *= 5, a7 *= 7) 1107 if (i == a3 || i == a5 || i == a7) 1108 return 1; 1109 return 0; 1110 } 1111