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 * 3. 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 91 *bnp = 0; 92 fs = ip->i_e2fs; 93 ump = ip->i_ump; 94 mtx_assert(EXT2_MTX(ump), MA_OWNED); 95 #ifdef INVARIANTS 96 if ((u_int)size > fs->e2fs_bsize || blkoff(fs, size) != 0) { 97 vn_printf(ip->i_devvp, "bsize = %lu, size = %d, fs = %s\n", 98 (long unsigned int)fs->e2fs_bsize, size, fs->e2fs_fsmnt); 99 panic("ext2_alloc: bad size"); 100 } 101 if (cred == NOCRED) 102 panic("ext2_alloc: missing credential"); 103 #endif /* INVARIANTS */ 104 if (size == fs->e2fs_bsize && fs->e2fs->e2fs_fbcount == 0) 105 goto nospace; 106 if (cred->cr_uid != 0 && 107 fs->e2fs->e2fs_fbcount < fs->e2fs->e2fs_rbcount) 108 goto nospace; 109 if (bpref >= fs->e2fs->e2fs_bcount) 110 bpref = 0; 111 if (bpref == 0) 112 cg = ino_to_cg(fs, ip->i_number); 113 else 114 cg = dtog(fs, bpref); 115 bno = (daddr_t)ext2_hashalloc(ip, cg, bpref, fs->e2fs_bsize, 116 ext2_alloccg); 117 if (bno > 0) { 118 /* set next_alloc fields as done in block_getblk */ 119 ip->i_next_alloc_block = lbn; 120 ip->i_next_alloc_goal = bno; 121 122 ip->i_blocks += btodb(fs->e2fs_bsize); 123 ip->i_flag |= IN_CHANGE | IN_UPDATE; 124 *bnp = bno; 125 return (0); 126 } 127 nospace: 128 EXT2_UNLOCK(ump); 129 ext2_fserr(fs, cred->cr_uid, "filesystem full"); 130 uprintf("\n%s: write failed, filesystem is full\n", fs->e2fs_fsmnt); 131 return (ENOSPC); 132 } 133 134 /* 135 * Allocate EA's block for inode. 136 */ 137 daddr_t 138 ext2_allocfacl(struct inode *ip) 139 { 140 struct m_ext2fs *fs; 141 daddr_t facl; 142 143 fs = ip->i_e2fs; 144 145 EXT2_LOCK(ip->i_ump); 146 facl = ext2_alloccg(ip, ino_to_cg(fs, ip->i_number), 0, fs->e2fs_bsize); 147 if (0 == facl) 148 EXT2_UNLOCK(ip->i_ump); 149 150 return (facl); 151 } 152 153 /* 154 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 155 * 156 * The vnode and an array of buffer pointers for a range of sequential 157 * logical blocks to be made contiguous is given. The allocator attempts 158 * to find a range of sequential blocks starting as close as possible to 159 * an fs_rotdelay offset from the end of the allocation for the logical 160 * block immediately preceding the current range. If successful, the 161 * physical block numbers in the buffer pointers and in the inode are 162 * changed to reflect the new allocation. If unsuccessful, the allocation 163 * is left unchanged. The success in doing the reallocation is returned. 164 * Note that the error return is not reflected back to the user. Rather 165 * the previous block allocation will be used. 166 */ 167 168 static SYSCTL_NODE(_vfs, OID_AUTO, ext2fs, CTLFLAG_RW, 0, "EXT2FS filesystem"); 169 170 static int doasyncfree = 1; 171 172 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, 173 "Use asychronous writes to update block pointers when freeing blocks"); 174 175 static int doreallocblks = 1; 176 177 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); 178 179 int 180 ext2_reallocblks(struct vop_reallocblks_args *ap) 181 { 182 struct m_ext2fs *fs; 183 struct inode *ip; 184 struct vnode *vp; 185 struct buf *sbp, *ebp; 186 uint32_t *bap, *sbap, *ebap; 187 struct ext2mount *ump; 188 struct cluster_save *buflist; 189 struct indir start_ap[EXT2_NIADDR + 1], end_ap[EXT2_NIADDR + 1], *idp; 190 e2fs_lbn_t start_lbn, end_lbn; 191 int soff; 192 e2fs_daddr_t newblk, blkno; 193 int i, len, start_lvl, end_lvl, pref, ssize; 194 195 if (doreallocblks == 0) 196 return (ENOSPC); 197 198 vp = ap->a_vp; 199 ip = VTOI(vp); 200 fs = ip->i_e2fs; 201 ump = ip->i_ump; 202 203 if (fs->e2fs_contigsumsize <= 0) 204 return (ENOSPC); 205 206 buflist = ap->a_buflist; 207 len = buflist->bs_nchildren; 208 start_lbn = buflist->bs_children[0]->b_lblkno; 209 end_lbn = start_lbn + len - 1; 210 #ifdef INVARIANTS 211 for (i = 1; i < len; i++) 212 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 213 panic("ext2_reallocblks: non-cluster"); 214 #endif 215 /* 216 * If the cluster crosses the boundary for the first indirect 217 * block, leave space for the indirect block. Indirect blocks 218 * are initially laid out in a position after the last direct 219 * block. Block reallocation would usually destroy locality by 220 * moving the indirect block out of the way to make room for 221 * data blocks if we didn't compensate here. We should also do 222 * this for other indirect block boundaries, but it is only 223 * important for the first one. 224 */ 225 if (start_lbn < EXT2_NDADDR && end_lbn >= EXT2_NDADDR) 226 return (ENOSPC); 227 /* 228 * If the latest allocation is in a new cylinder group, assume that 229 * the filesystem has decided to move and do not force it back to 230 * the previous cylinder group. 231 */ 232 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 233 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 234 return (ENOSPC); 235 if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) || 236 ext2_getlbns(vp, end_lbn, end_ap, &end_lvl)) 237 return (ENOSPC); 238 /* 239 * Get the starting offset and block map for the first block. 240 */ 241 if (start_lvl == 0) { 242 sbap = &ip->i_db[0]; 243 soff = start_lbn; 244 } else { 245 idp = &start_ap[start_lvl - 1]; 246 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) { 247 brelse(sbp); 248 return (ENOSPC); 249 } 250 sbap = (u_int *)sbp->b_data; 251 soff = idp->in_off; 252 } 253 /* 254 * If the block range spans two block maps, get the second map. 255 */ 256 ebap = NULL; 257 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 258 ssize = len; 259 } else { 260 #ifdef INVARIANTS 261 if (start_ap[start_lvl - 1].in_lbn == idp->in_lbn) 262 panic("ext2_reallocblks: start == end"); 263 #endif 264 ssize = len - (idp->in_off + 1); 265 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp)) 266 goto fail; 267 ebap = (u_int *)ebp->b_data; 268 } 269 /* 270 * Find the preferred location for the cluster. 271 */ 272 EXT2_LOCK(ump); 273 pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0); 274 /* 275 * Search the block map looking for an allocation of the desired size. 276 */ 277 if ((newblk = (e2fs_daddr_t)ext2_hashalloc(ip, dtog(fs, pref), pref, 278 len, ext2_clusteralloc)) == 0) { 279 EXT2_UNLOCK(ump); 280 goto fail; 281 } 282 /* 283 * We have found a new contiguous block. 284 * 285 * First we have to replace the old block pointers with the new 286 * block pointers in the inode and indirect blocks associated 287 * with the file. 288 */ 289 #ifdef DEBUG 290 printf("realloc: ino %ju, lbns %jd-%jd\n\told:", 291 (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); 292 #endif /* DEBUG */ 293 blkno = newblk; 294 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) { 295 if (i == ssize) { 296 bap = ebap; 297 soff = -i; 298 } 299 #ifdef INVARIANTS 300 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap)) 301 panic("ext2_reallocblks: alloc mismatch"); 302 #endif 303 #ifdef DEBUG 304 printf(" %d,", *bap); 305 #endif /* DEBUG */ 306 *bap++ = blkno; 307 } 308 /* 309 * Next we must write out the modified inode and indirect blocks. 310 * For strict correctness, the writes should be synchronous since 311 * the old block values may have been written to disk. In practise 312 * they are almost never written, but if we are concerned about 313 * strict correctness, the `doasyncfree' flag should be set to zero. 314 * 315 * The test on `doasyncfree' should be changed to test a flag 316 * that shows whether the associated buffers and inodes have 317 * been written. The flag should be set when the cluster is 318 * started and cleared whenever the buffer or inode is flushed. 319 * We can then check below to see if it is set, and do the 320 * synchronous write only when it has been cleared. 321 */ 322 if (sbap != &ip->i_db[0]) { 323 if (doasyncfree) 324 bdwrite(sbp); 325 else 326 bwrite(sbp); 327 } else { 328 ip->i_flag |= IN_CHANGE | IN_UPDATE; 329 if (!doasyncfree) 330 ext2_update(vp, 1); 331 } 332 if (ssize < len) { 333 if (doasyncfree) 334 bdwrite(ebp); 335 else 336 bwrite(ebp); 337 } 338 /* 339 * Last, free the old blocks and assign the new blocks to the buffers. 340 */ 341 #ifdef DEBUG 342 printf("\n\tnew:"); 343 #endif /* DEBUG */ 344 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) { 345 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), 346 fs->e2fs_bsize); 347 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 348 #ifdef DEBUG 349 printf(" %d,", blkno); 350 #endif /* DEBUG */ 351 } 352 #ifdef DEBUG 353 printf("\n"); 354 #endif /* DEBUG */ 355 return (0); 356 357 fail: 358 if (ssize < len) 359 brelse(ebp); 360 if (sbap != &ip->i_db[0]) 361 brelse(sbp); 362 return (ENOSPC); 363 } 364 365 /* 366 * Allocate an inode in the filesystem. 367 * 368 */ 369 int 370 ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp) 371 { 372 struct timespec ts; 373 struct inode *pip; 374 struct m_ext2fs *fs; 375 struct inode *ip; 376 struct ext2mount *ump; 377 ino_t ino, ipref; 378 int i, error, cg; 379 380 *vpp = NULL; 381 pip = VTOI(pvp); 382 fs = pip->i_e2fs; 383 ump = pip->i_ump; 384 385 EXT2_LOCK(ump); 386 if (fs->e2fs->e2fs_ficount == 0) 387 goto noinodes; 388 /* 389 * If it is a directory then obtain a cylinder group based on 390 * ext2_dirpref else obtain it using ino_to_cg. The preferred inode is 391 * always the next inode. 392 */ 393 if ((mode & IFMT) == IFDIR) { 394 cg = ext2_dirpref(pip); 395 if (fs->e2fs_contigdirs[cg] < 255) 396 fs->e2fs_contigdirs[cg]++; 397 } else { 398 cg = ino_to_cg(fs, pip->i_number); 399 if (fs->e2fs_contigdirs[cg] > 0) 400 fs->e2fs_contigdirs[cg]--; 401 } 402 ipref = cg * fs->e2fs->e2fs_ipg + 1; 403 ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg); 404 405 if (ino == 0) 406 goto noinodes; 407 error = VFS_VGET(pvp->v_mount, ino, LK_EXCLUSIVE, vpp); 408 if (error) { 409 ext2_vfree(pvp, ino, mode); 410 return (error); 411 } 412 ip = VTOI(*vpp); 413 414 /* 415 * The question is whether using VGET was such good idea at all: 416 * Linux doesn't read the old inode in when it is allocating a 417 * new one. I will set at least i_size and i_blocks to zero. 418 */ 419 ip->i_flag = 0; 420 ip->i_size = 0; 421 ip->i_blocks = 0; 422 ip->i_mode = 0; 423 ip->i_flags = 0; 424 /* now we want to make sure that the block pointers are zeroed out */ 425 for (i = 0; i < EXT2_NDADDR; i++) 426 ip->i_db[i] = 0; 427 for (i = 0; i < EXT2_NIADDR; i++) 428 ip->i_ib[i] = 0; 429 430 /* 431 * Set up a new generation number for this inode. 432 * Avoid zero values. 433 */ 434 do { 435 ip->i_gen = arc4random(); 436 } while (ip->i_gen == 0); 437 438 vfs_timestamp(&ts); 439 ip->i_birthtime = ts.tv_sec; 440 ip->i_birthnsec = ts.tv_nsec; 441 442 /* 443 printf("ext2_valloc: allocated inode %d\n", ino); 444 */ 445 return (0); 446 noinodes: 447 EXT2_UNLOCK(ump); 448 ext2_fserr(fs, cred->cr_uid, "out of inodes"); 449 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->e2fs_fsmnt); 450 return (ENOSPC); 451 } 452 453 /* 454 * Find a cylinder to place a directory. 455 * 456 * The policy implemented by this algorithm is to allocate a 457 * directory inode in the same cylinder group as its parent 458 * directory, but also to reserve space for its files inodes 459 * and data. Restrict the number of directories which may be 460 * allocated one after another in the same cylinder group 461 * without intervening allocation of files. 462 * 463 * If we allocate a first level directory then force allocation 464 * in another cylinder group. 465 * 466 */ 467 static u_long 468 ext2_dirpref(struct inode *pip) 469 { 470 struct m_ext2fs *fs; 471 int cg, prefcg, cgsize; 472 u_int avgifree, avgbfree, avgndir, curdirsize; 473 u_int minifree, minbfree, maxndir; 474 u_int mincg, minndir; 475 u_int dirsize, maxcontigdirs; 476 477 mtx_assert(EXT2_MTX(pip->i_ump), MA_OWNED); 478 fs = pip->i_e2fs; 479 480 avgifree = fs->e2fs->e2fs_ficount / fs->e2fs_gcount; 481 avgbfree = fs->e2fs->e2fs_fbcount / fs->e2fs_gcount; 482 avgndir = fs->e2fs_total_dir / fs->e2fs_gcount; 483 484 /* 485 * Force allocation in another cg if creating a first level dir. 486 */ 487 ASSERT_VOP_LOCKED(ITOV(pip), "ext2fs_dirpref"); 488 if (ITOV(pip)->v_vflag & VV_ROOT) { 489 prefcg = arc4random() % fs->e2fs_gcount; 490 mincg = prefcg; 491 minndir = fs->e2fs_ipg; 492 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 493 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir && 494 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree && 495 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) { 496 mincg = cg; 497 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs; 498 } 499 for (cg = 0; cg < prefcg; cg++) 500 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir && 501 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree && 502 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) { 503 mincg = cg; 504 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs; 505 } 506 return (mincg); 507 } 508 /* 509 * Count various limits which used for 510 * optimal allocation of a directory inode. 511 */ 512 maxndir = min(avgndir + fs->e2fs_ipg / 16, fs->e2fs_ipg); 513 minifree = avgifree - avgifree / 4; 514 if (minifree < 1) 515 minifree = 1; 516 minbfree = avgbfree - avgbfree / 4; 517 if (minbfree < 1) 518 minbfree = 1; 519 cgsize = fs->e2fs_fsize * fs->e2fs_fpg; 520 dirsize = AVGDIRSIZE; 521 curdirsize = avgndir ? (cgsize - avgbfree * fs->e2fs_bsize) / avgndir : 0; 522 if (dirsize < curdirsize) 523 dirsize = curdirsize; 524 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255); 525 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR); 526 if (maxcontigdirs == 0) 527 maxcontigdirs = 1; 528 529 /* 530 * Limit number of dirs in one cg and reserve space for 531 * regular files, but only if we have no deficit in 532 * inodes or space. 533 */ 534 prefcg = ino_to_cg(fs, pip->i_number); 535 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 536 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir && 537 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree && 538 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) { 539 if (fs->e2fs_contigdirs[cg] < maxcontigdirs) 540 return (cg); 541 } 542 for (cg = 0; cg < prefcg; cg++) 543 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir && 544 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree && 545 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) { 546 if (fs->e2fs_contigdirs[cg] < maxcontigdirs) 547 return (cg); 548 } 549 /* 550 * This is a backstop when we have deficit in space. 551 */ 552 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 553 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree) 554 return (cg); 555 for (cg = 0; cg < prefcg; cg++) 556 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree) 557 break; 558 return (cg); 559 } 560 561 /* 562 * Select the desired position for the next block in a file. 563 * 564 * we try to mimic what Remy does in inode_getblk/block_getblk 565 * 566 * we note: blocknr == 0 means that we're about to allocate either 567 * a direct block or a pointer block at the first level of indirection 568 * (In other words, stuff that will go in i_db[] or i_ib[]) 569 * 570 * blocknr != 0 means that we're allocating a block that is none 571 * of the above. Then, blocknr tells us the number of the block 572 * that will hold the pointer 573 */ 574 e4fs_daddr_t 575 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, e2fs_daddr_t *bap, 576 e2fs_daddr_t blocknr) 577 { 578 int tmp; 579 580 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED); 581 582 /* 583 * If the next block is actually what we thought it is, then set the 584 * goal to what we thought it should be. 585 */ 586 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0) 587 return ip->i_next_alloc_goal; 588 589 /* 590 * Now check whether we were provided with an array that basically 591 * tells us previous blocks to which we want to stay close. 592 */ 593 if (bap) 594 for (tmp = indx - 1; tmp >= 0; tmp--) 595 if (bap[tmp]) 596 return bap[tmp]; 597 598 /* 599 * Else lets fall back to the blocknr or, if there is none, follow 600 * the rule that a block should be allocated near its inode. 601 */ 602 return blocknr ? blocknr : 603 (e2fs_daddr_t)(ip->i_block_group * 604 EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) + 605 ip->i_e2fs->e2fs->e2fs_first_dblock; 606 } 607 608 /* 609 * Implement the cylinder overflow algorithm. 610 * 611 * The policy implemented by this algorithm is: 612 * 1) allocate the block in its requested cylinder group. 613 * 2) quadradically rehash on the cylinder group number. 614 * 3) brute force search for a free block. 615 */ 616 static u_long 617 ext2_hashalloc(struct inode *ip, int cg, long pref, int size, 618 daddr_t (*allocator) (struct inode *, int, daddr_t, int)) 619 { 620 struct m_ext2fs *fs; 621 ino_t result; 622 int i, icg = cg; 623 624 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED); 625 fs = ip->i_e2fs; 626 /* 627 * 1: preferred cylinder group 628 */ 629 result = (*allocator)(ip, cg, pref, size); 630 if (result) 631 return (result); 632 /* 633 * 2: quadratic rehash 634 */ 635 for (i = 1; i < fs->e2fs_gcount; i *= 2) { 636 cg += i; 637 if (cg >= fs->e2fs_gcount) 638 cg -= fs->e2fs_gcount; 639 result = (*allocator)(ip, cg, 0, size); 640 if (result) 641 return (result); 642 } 643 /* 644 * 3: brute force search 645 * Note that we start at i == 2, since 0 was checked initially, 646 * and 1 is always checked in the quadratic rehash. 647 */ 648 cg = (icg + 2) % fs->e2fs_gcount; 649 for (i = 2; i < fs->e2fs_gcount; i++) { 650 result = (*allocator)(ip, cg, 0, size); 651 if (result) 652 return (result); 653 cg++; 654 if (cg == fs->e2fs_gcount) 655 cg = 0; 656 } 657 return (0); 658 } 659 660 /* 661 * Determine whether a block can be allocated. 662 * 663 * Check to see if a block of the appropriate size is available, 664 * and if it is, allocate it. 665 */ 666 static daddr_t 667 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size) 668 { 669 struct m_ext2fs *fs; 670 struct buf *bp; 671 struct ext2mount *ump; 672 daddr_t bno, runstart, runlen; 673 int bit, loc, end, error, start; 674 char *bbp; 675 /* XXX ondisk32 */ 676 fs = ip->i_e2fs; 677 ump = ip->i_ump; 678 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0) 679 return (0); 680 EXT2_UNLOCK(ump); 681 error = bread(ip->i_devvp, fsbtodb(fs, 682 fs->e2fs_gd[cg].ext2bgd_b_bitmap), 683 (int)fs->e2fs_bsize, NOCRED, &bp); 684 if (error) { 685 brelse(bp); 686 EXT2_LOCK(ump); 687 return (0); 688 } 689 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0) { 690 /* 691 * Another thread allocated the last block in this 692 * group while we were waiting for the buffer. 693 */ 694 brelse(bp); 695 EXT2_LOCK(ump); 696 return (0); 697 } 698 bbp = (char *)bp->b_data; 699 700 if (dtog(fs, bpref) != cg) 701 bpref = 0; 702 if (bpref != 0) { 703 bpref = dtogd(fs, bpref); 704 /* 705 * if the requested block is available, use it 706 */ 707 if (isclr(bbp, bpref)) { 708 bno = bpref; 709 goto gotit; 710 } 711 } 712 /* 713 * no blocks in the requested cylinder, so take next 714 * available one in this cylinder group. 715 * first try to get 8 contigous blocks, then fall back to a single 716 * block. 717 */ 718 if (bpref) 719 start = dtogd(fs, bpref) / NBBY; 720 else 721 start = 0; 722 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start; 723 retry: 724 runlen = 0; 725 runstart = 0; 726 for (loc = start; loc < end; loc++) { 727 if (bbp[loc] == (char)0xff) { 728 runlen = 0; 729 continue; 730 } 731 732 /* Start of a run, find the number of high clear bits. */ 733 if (runlen == 0) { 734 bit = fls(bbp[loc]); 735 runlen = NBBY - bit; 736 runstart = loc * NBBY + bit; 737 } else if (bbp[loc] == 0) { 738 /* Continue a run. */ 739 runlen += NBBY; 740 } else { 741 /* 742 * Finish the current run. If it isn't long 743 * enough, start a new one. 744 */ 745 bit = ffs(bbp[loc]) - 1; 746 runlen += bit; 747 if (runlen >= 8) { 748 bno = runstart; 749 goto gotit; 750 } 751 752 /* Run was too short, start a new one. */ 753 bit = fls(bbp[loc]); 754 runlen = NBBY - bit; 755 runstart = loc * NBBY + bit; 756 } 757 758 /* If the current run is long enough, use it. */ 759 if (runlen >= 8) { 760 bno = runstart; 761 goto gotit; 762 } 763 } 764 if (start != 0) { 765 end = start; 766 start = 0; 767 goto retry; 768 } 769 bno = ext2_mapsearch(fs, bbp, bpref); 770 if (bno < 0) { 771 brelse(bp); 772 EXT2_LOCK(ump); 773 return (0); 774 } 775 gotit: 776 #ifdef INVARIANTS 777 if (isset(bbp, bno)) { 778 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n", 779 cg, (intmax_t)bno, fs->e2fs_fsmnt); 780 panic("ext2fs_alloccg: dup alloc"); 781 } 782 #endif 783 setbit(bbp, bno); 784 EXT2_LOCK(ump); 785 ext2_clusteracct(fs, bbp, cg, bno, -1); 786 fs->e2fs->e2fs_fbcount--; 787 fs->e2fs_gd[cg].ext2bgd_nbfree--; 788 fs->e2fs_fmod = 1; 789 EXT2_UNLOCK(ump); 790 bdwrite(bp); 791 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno); 792 } 793 794 /* 795 * Determine whether a cluster can be allocated. 796 */ 797 static daddr_t 798 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len) 799 { 800 struct m_ext2fs *fs; 801 struct ext2mount *ump; 802 struct buf *bp; 803 char *bbp; 804 int bit, error, got, i, loc, run; 805 int32_t *lp; 806 daddr_t bno; 807 808 fs = ip->i_e2fs; 809 ump = ip->i_ump; 810 811 if (fs->e2fs_maxcluster[cg] < len) 812 return (0); 813 814 EXT2_UNLOCK(ump); 815 error = bread(ip->i_devvp, 816 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap), 817 (int)fs->e2fs_bsize, NOCRED, &bp); 818 if (error) 819 goto fail_lock; 820 821 bbp = (char *)bp->b_data; 822 EXT2_LOCK(ump); 823 /* 824 * Check to see if a cluster of the needed size (or bigger) is 825 * available in this cylinder group. 826 */ 827 lp = &fs->e2fs_clustersum[cg].cs_sum[len]; 828 for (i = len; i <= fs->e2fs_contigsumsize; i++) 829 if (*lp++ > 0) 830 break; 831 if (i > fs->e2fs_contigsumsize) { 832 /* 833 * Update the cluster summary information to reflect 834 * the true maximum-sized cluster so that future cluster 835 * allocation requests can avoid reading the bitmap only 836 * to find no cluster. 837 */ 838 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1]; 839 for (i = len - 1; i > 0; i--) 840 if (*lp-- > 0) 841 break; 842 fs->e2fs_maxcluster[cg] = i; 843 goto fail; 844 } 845 EXT2_UNLOCK(ump); 846 847 /* Search the bitmap to find a big enough cluster like in FFS. */ 848 if (dtog(fs, bpref) != cg) 849 bpref = 0; 850 if (bpref != 0) 851 bpref = dtogd(fs, bpref); 852 loc = bpref / NBBY; 853 bit = 1 << (bpref % NBBY); 854 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) { 855 if ((bbp[loc] & bit) != 0) 856 run = 0; 857 else { 858 run++; 859 if (run == len) 860 break; 861 } 862 if ((got & (NBBY - 1)) != (NBBY - 1)) 863 bit <<= 1; 864 else { 865 loc++; 866 bit = 1; 867 } 868 } 869 870 if (got >= fs->e2fs->e2fs_fpg) 871 goto fail_lock; 872 873 /* Allocate the cluster that we found. */ 874 for (i = 1; i < len; i++) 875 if (!isclr(bbp, got - run + i)) 876 panic("ext2_clusteralloc: map mismatch"); 877 878 bno = got - run + 1; 879 if (bno >= fs->e2fs->e2fs_fpg) 880 panic("ext2_clusteralloc: allocated out of group"); 881 882 EXT2_LOCK(ump); 883 for (i = 0; i < len; i += fs->e2fs_fpb) { 884 setbit(bbp, bno + i); 885 ext2_clusteracct(fs, bbp, cg, bno + i, -1); 886 fs->e2fs->e2fs_fbcount--; 887 fs->e2fs_gd[cg].ext2bgd_nbfree--; 888 } 889 fs->e2fs_fmod = 1; 890 EXT2_UNLOCK(ump); 891 892 bdwrite(bp); 893 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno); 894 895 fail_lock: 896 EXT2_LOCK(ump); 897 fail: 898 brelse(bp); 899 return (0); 900 } 901 902 /* 903 * Determine whether an inode can be allocated. 904 * 905 * Check to see if an inode is available, and if it is, 906 * allocate it using tode in the specified cylinder group. 907 */ 908 static daddr_t 909 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode) 910 { 911 struct m_ext2fs *fs; 912 struct buf *bp; 913 struct ext2mount *ump; 914 int error, start, len; 915 char *ibp, *loc; 916 917 ipref--; /* to avoid a lot of (ipref -1) */ 918 if (ipref == -1) 919 ipref = 0; 920 fs = ip->i_e2fs; 921 ump = ip->i_ump; 922 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0) 923 return (0); 924 EXT2_UNLOCK(ump); 925 error = bread(ip->i_devvp, fsbtodb(fs, 926 fs->e2fs_gd[cg].ext2bgd_i_bitmap), 927 (int)fs->e2fs_bsize, NOCRED, &bp); 928 if (error) { 929 brelse(bp); 930 EXT2_LOCK(ump); 931 return (0); 932 } 933 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0) { 934 /* 935 * Another thread allocated the last i-node in this 936 * group while we were waiting for the buffer. 937 */ 938 brelse(bp); 939 EXT2_LOCK(ump); 940 return (0); 941 } 942 ibp = (char *)bp->b_data; 943 if (ipref) { 944 ipref %= fs->e2fs->e2fs_ipg; 945 if (isclr(ibp, ipref)) 946 goto gotit; 947 } 948 start = ipref / NBBY; 949 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY); 950 loc = memcchr(&ibp[start], 0xff, len); 951 if (loc == NULL) { 952 len = start + 1; 953 start = 0; 954 loc = memcchr(&ibp[start], 0xff, len); 955 if (loc == NULL) { 956 printf("cg = %d, ipref = %lld, fs = %s\n", 957 cg, (long long)ipref, fs->e2fs_fsmnt); 958 panic("ext2fs_nodealloccg: map corrupted"); 959 /* NOTREACHED */ 960 } 961 } 962 ipref = (loc - ibp) * NBBY + ffs(~*loc) - 1; 963 gotit: 964 setbit(ibp, ipref); 965 EXT2_LOCK(ump); 966 fs->e2fs_gd[cg].ext2bgd_nifree--; 967 fs->e2fs->e2fs_ficount--; 968 fs->e2fs_fmod = 1; 969 if ((mode & IFMT) == IFDIR) { 970 fs->e2fs_gd[cg].ext2bgd_ndirs++; 971 fs->e2fs_total_dir++; 972 } 973 EXT2_UNLOCK(ump); 974 bdwrite(bp); 975 return (cg * fs->e2fs->e2fs_ipg + ipref + 1); 976 } 977 978 /* 979 * Free a block or fragment. 980 * 981 */ 982 void 983 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size) 984 { 985 struct m_ext2fs *fs; 986 struct buf *bp; 987 struct ext2mount *ump; 988 int cg, error; 989 char *bbp; 990 991 fs = ip->i_e2fs; 992 ump = ip->i_ump; 993 cg = dtog(fs, bno); 994 if ((u_int)bno >= fs->e2fs->e2fs_bcount) { 995 printf("bad block %lld, ino %ju\n", (long long)bno, 996 (uintmax_t)ip->i_number); 997 ext2_fserr(fs, ip->i_uid, "bad block"); 998 return; 999 } 1000 error = bread(ip->i_devvp, 1001 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap), 1002 (int)fs->e2fs_bsize, NOCRED, &bp); 1003 if (error) { 1004 brelse(bp); 1005 return; 1006 } 1007 bbp = (char *)bp->b_data; 1008 bno = dtogd(fs, bno); 1009 if (isclr(bbp, bno)) { 1010 printf("block = %lld, fs = %s\n", 1011 (long long)bno, fs->e2fs_fsmnt); 1012 panic("ext2_blkfree: freeing free block"); 1013 } 1014 clrbit(bbp, bno); 1015 EXT2_LOCK(ump); 1016 ext2_clusteracct(fs, bbp, cg, bno, 1); 1017 fs->e2fs->e2fs_fbcount++; 1018 fs->e2fs_gd[cg].ext2bgd_nbfree++; 1019 fs->e2fs_fmod = 1; 1020 EXT2_UNLOCK(ump); 1021 bdwrite(bp); 1022 } 1023 1024 /* 1025 * Free an inode. 1026 * 1027 */ 1028 int 1029 ext2_vfree(struct vnode *pvp, ino_t ino, int mode) 1030 { 1031 struct m_ext2fs *fs; 1032 struct inode *pip; 1033 struct buf *bp; 1034 struct ext2mount *ump; 1035 int error, cg; 1036 char *ibp; 1037 1038 pip = VTOI(pvp); 1039 fs = pip->i_e2fs; 1040 ump = pip->i_ump; 1041 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount) 1042 panic("ext2_vfree: range: devvp = %p, ino = %ju, fs = %s", 1043 pip->i_devvp, (uintmax_t)ino, fs->e2fs_fsmnt); 1044 1045 cg = ino_to_cg(fs, ino); 1046 error = bread(pip->i_devvp, 1047 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_i_bitmap), 1048 (int)fs->e2fs_bsize, NOCRED, &bp); 1049 if (error) { 1050 brelse(bp); 1051 return (0); 1052 } 1053 ibp = (char *)bp->b_data; 1054 ino = (ino - 1) % fs->e2fs->e2fs_ipg; 1055 if (isclr(ibp, ino)) { 1056 printf("ino = %llu, fs = %s\n", 1057 (unsigned long long)ino, fs->e2fs_fsmnt); 1058 if (fs->e2fs_ronly == 0) 1059 panic("ext2_vfree: freeing free inode"); 1060 } 1061 clrbit(ibp, ino); 1062 EXT2_LOCK(ump); 1063 fs->e2fs->e2fs_ficount++; 1064 fs->e2fs_gd[cg].ext2bgd_nifree++; 1065 if ((mode & IFMT) == IFDIR) { 1066 fs->e2fs_gd[cg].ext2bgd_ndirs--; 1067 fs->e2fs_total_dir--; 1068 } 1069 fs->e2fs_fmod = 1; 1070 EXT2_UNLOCK(ump); 1071 bdwrite(bp); 1072 return (0); 1073 } 1074 1075 /* 1076 * Find a block in the specified cylinder group. 1077 * 1078 * It is a panic if a request is made to find a block if none are 1079 * available. 1080 */ 1081 static daddr_t 1082 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref) 1083 { 1084 char *loc; 1085 int start, len; 1086 1087 /* 1088 * find the fragment by searching through the free block 1089 * map for an appropriate bit pattern 1090 */ 1091 if (bpref) 1092 start = dtogd(fs, bpref) / NBBY; 1093 else 1094 start = 0; 1095 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start; 1096 loc = memcchr(&bbp[start], 0xff, len); 1097 if (loc == NULL) { 1098 len = start + 1; 1099 start = 0; 1100 loc = memcchr(&bbp[start], 0xff, len); 1101 if (loc == NULL) { 1102 printf("start = %d, len = %d, fs = %s\n", 1103 start, len, fs->e2fs_fsmnt); 1104 panic("ext2_mapsearch: map corrupted"); 1105 /* NOTREACHED */ 1106 } 1107 } 1108 return ((loc - bbp) * NBBY + ffs(~*loc) - 1); 1109 } 1110 1111 /* 1112 * Fserr prints the name of a filesystem with an error diagnostic. 1113 * 1114 * The form of the error message is: 1115 * fs: error message 1116 */ 1117 static void 1118 ext2_fserr(struct m_ext2fs *fs, uid_t uid, char *cp) 1119 { 1120 1121 log(LOG_ERR, "uid %u on %s: %s\n", uid, fs->e2fs_fsmnt, cp); 1122 } 1123 1124 int 1125 cg_has_sb(int i) 1126 { 1127 int a3, a5, a7; 1128 1129 if (i == 0 || i == 1) 1130 return 1; 1131 for (a3 = 3, a5 = 5, a7 = 7; 1132 a3 <= i || a5 <= i || a7 <= i; 1133 a3 *= 3, a5 *= 5, a7 *= 7) 1134 if (i == a3 || i == a5 || i == a7) 1135 return 1; 1136 return 0; 1137 } 1138