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 * SPDX-License-Identifier: BSD-3-Clause 9 * 10 * Copyright (c) 1982, 1986, 1989, 1993 11 * The Regents of the University of California. All rights reserved. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)ffs_alloc.c 8.8 (Berkeley) 2/21/94 38 * $FreeBSD$ 39 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/conf.h> 44 #include <sys/vnode.h> 45 #include <sys/stat.h> 46 #include <sys/mount.h> 47 #include <sys/sysctl.h> 48 #include <sys/syslog.h> 49 #include <sys/buf.h> 50 #include <sys/endian.h> 51 52 #include <fs/ext2fs/fs.h> 53 #include <fs/ext2fs/inode.h> 54 #include <fs/ext2fs/ext2_mount.h> 55 #include <fs/ext2fs/ext2fs.h> 56 #include <fs/ext2fs/ext2_extern.h> 57 58 static daddr_t ext2_alloccg(struct inode *, int, daddr_t, int); 59 static daddr_t ext2_clusteralloc(struct inode *, int, daddr_t, int); 60 static u_long ext2_dirpref(struct inode *); 61 static e4fs_daddr_t ext2_hashalloc(struct inode *, int, long, int, 62 daddr_t (*)(struct inode *, int, daddr_t, 63 int)); 64 static daddr_t ext2_nodealloccg(struct inode *, int, daddr_t, int); 65 static daddr_t ext2_mapsearch(struct m_ext2fs *, char *, daddr_t); 66 67 /* 68 * Allocate a block in the filesystem. 69 * 70 * A preference may be optionally specified. If a preference is given 71 * the following hierarchy is used to allocate a block: 72 * 1) allocate the requested block. 73 * 2) allocate a rotationally optimal block in the same cylinder. 74 * 3) allocate a block in the same cylinder group. 75 * 4) quadradically rehash into other cylinder groups, until an 76 * available block is located. 77 * If no block preference is given the following hierarchy is used 78 * to allocate a block: 79 * 1) allocate a block in the cylinder group that contains the 80 * inode for the file. 81 * 2) quadradically rehash into other cylinder groups, until an 82 * available block is located. 83 */ 84 int 85 ext2_alloc(struct inode *ip, daddr_t lbn, e4fs_daddr_t bpref, int size, 86 struct ucred *cred, e4fs_daddr_t *bnp) 87 { 88 struct m_ext2fs *fs; 89 struct ext2mount *ump; 90 e4fs_daddr_t bno; 91 int cg; 92 93 *bnp = 0; 94 fs = ip->i_e2fs; 95 ump = ip->i_ump; 96 mtx_assert(EXT2_MTX(ump), MA_OWNED); 97 #ifdef INVARIANTS 98 if ((u_int)size > fs->e2fs_bsize || blkoff(fs, size) != 0) { 99 vn_printf(ip->i_devvp, "bsize = %lu, size = %d, fs = %s\n", 100 (long unsigned int)fs->e2fs_bsize, size, fs->e2fs_fsmnt); 101 panic("ext2_alloc: bad size"); 102 } 103 if (cred == NOCRED) 104 panic("ext2_alloc: missing credential"); 105 #endif /* INVARIANTS */ 106 if (size == fs->e2fs_bsize && fs->e2fs_fbcount == 0) 107 goto nospace; 108 if (cred->cr_uid != 0 && 109 fs->e2fs_fbcount < fs->e2fs_rbcount) 110 goto nospace; 111 if (bpref >= fs->e2fs_bcount) 112 bpref = 0; 113 if (bpref == 0) 114 cg = ino_to_cg(fs, ip->i_number); 115 else 116 cg = dtog(fs, bpref); 117 bno = (daddr_t)ext2_hashalloc(ip, cg, bpref, fs->e2fs_bsize, 118 ext2_alloccg); 119 if (bno > 0) { 120 /* set next_alloc fields as done in block_getblk */ 121 ip->i_next_alloc_block = lbn; 122 ip->i_next_alloc_goal = bno; 123 124 ip->i_blocks += btodb(fs->e2fs_bsize); 125 ip->i_flag |= IN_CHANGE | IN_UPDATE; 126 *bnp = bno; 127 return (0); 128 } 129 nospace: 130 EXT2_UNLOCK(ump); 131 ext2_fserr(fs, cred->cr_uid, "filesystem full"); 132 uprintf("\n%s: write failed, filesystem is full\n", fs->e2fs_fsmnt); 133 return (ENOSPC); 134 } 135 136 /* 137 * Allocate EA's block for inode. 138 */ 139 e4fs_daddr_t 140 ext2_alloc_meta(struct inode *ip) 141 { 142 struct m_ext2fs *fs; 143 daddr_t blk; 144 145 fs = ip->i_e2fs; 146 147 EXT2_LOCK(ip->i_ump); 148 blk = ext2_hashalloc(ip, ino_to_cg(fs, ip->i_number), 0, fs->e2fs_bsize, 149 ext2_alloccg); 150 if (0 == blk) 151 EXT2_UNLOCK(ip->i_ump); 152 153 return (blk); 154 } 155 156 /* 157 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 158 * 159 * The vnode and an array of buffer pointers for a range of sequential 160 * logical blocks to be made contiguous is given. The allocator attempts 161 * to find a range of sequential blocks starting as close as possible to 162 * an fs_rotdelay offset from the end of the allocation for the logical 163 * block immediately preceding the current range. If successful, the 164 * physical block numbers in the buffer pointers and in the inode are 165 * changed to reflect the new allocation. If unsuccessful, the allocation 166 * is left unchanged. The success in doing the reallocation is returned. 167 * Note that the error return is not reflected back to the user. Rather 168 * the previous block allocation will be used. 169 */ 170 171 static SYSCTL_NODE(_vfs, OID_AUTO, ext2fs, CTLFLAG_RW, 0, "EXT2FS filesystem"); 172 173 static int doasyncfree = 1; 174 175 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, 176 "Use asychronous writes to update block pointers when freeing blocks"); 177 178 static int doreallocblks = 0; 179 180 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); 181 182 int 183 ext2_reallocblks(struct vop_reallocblks_args *ap) 184 { 185 struct m_ext2fs *fs; 186 struct inode *ip; 187 struct vnode *vp; 188 struct buf *sbp, *ebp; 189 uint32_t *bap, *sbap, *ebap; 190 struct ext2mount *ump; 191 struct cluster_save *buflist; 192 struct indir start_ap[EXT2_NIADDR + 1], end_ap[EXT2_NIADDR + 1], *idp; 193 e2fs_lbn_t start_lbn, end_lbn; 194 int soff; 195 e2fs_daddr_t newblk, blkno; 196 int i, len, start_lvl, end_lvl, pref, ssize; 197 198 if (doreallocblks == 0) 199 return (ENOSPC); 200 201 vp = ap->a_vp; 202 ip = VTOI(vp); 203 fs = ip->i_e2fs; 204 ump = ip->i_ump; 205 206 if (fs->e2fs_contigsumsize <= 0 || ip->i_flag & IN_E4EXTENTS) 207 return (ENOSPC); 208 209 buflist = ap->a_buflist; 210 len = buflist->bs_nchildren; 211 start_lbn = buflist->bs_children[0]->b_lblkno; 212 end_lbn = start_lbn + len - 1; 213 #ifdef INVARIANTS 214 for (i = 1; i < len; i++) 215 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 216 panic("ext2_reallocblks: non-cluster"); 217 #endif 218 /* 219 * If the cluster crosses the boundary for the first indirect 220 * block, leave space for the indirect block. Indirect blocks 221 * are initially laid out in a position after the last direct 222 * block. Block reallocation would usually destroy locality by 223 * moving the indirect block out of the way to make room for 224 * data blocks if we didn't compensate here. We should also do 225 * this for other indirect block boundaries, but it is only 226 * important for the first one. 227 */ 228 if (start_lbn < EXT2_NDADDR && end_lbn >= EXT2_NDADDR) 229 return (ENOSPC); 230 /* 231 * If the latest allocation is in a new cylinder group, assume that 232 * the filesystem has decided to move and do not force it back to 233 * the previous cylinder group. 234 */ 235 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 236 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 237 return (ENOSPC); 238 if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) || 239 ext2_getlbns(vp, end_lbn, end_ap, &end_lvl)) 240 return (ENOSPC); 241 /* 242 * Get the starting offset and block map for the first block. 243 */ 244 if (start_lvl == 0) { 245 sbap = &ip->i_db[0]; 246 soff = start_lbn; 247 } else { 248 idp = &start_ap[start_lvl - 1]; 249 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) { 250 brelse(sbp); 251 return (ENOSPC); 252 } 253 sbap = (u_int *)sbp->b_data; 254 soff = idp->in_off; 255 } 256 /* 257 * If the block range spans two block maps, get the second map. 258 */ 259 ebap = NULL; 260 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 261 ssize = len; 262 } else { 263 #ifdef INVARIANTS 264 if (start_ap[start_lvl - 1].in_lbn == idp->in_lbn) 265 panic("ext2_reallocblks: start == end"); 266 #endif 267 ssize = len - (idp->in_off + 1); 268 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp)) 269 goto fail; 270 ebap = (u_int *)ebp->b_data; 271 } 272 /* 273 * Find the preferred location for the cluster. 274 */ 275 EXT2_LOCK(ump); 276 pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0); 277 /* 278 * Search the block map looking for an allocation of the desired size. 279 */ 280 if ((newblk = (e2fs_daddr_t)ext2_hashalloc(ip, dtog(fs, pref), pref, 281 len, ext2_clusteralloc)) == 0) { 282 EXT2_UNLOCK(ump); 283 goto fail; 284 } 285 /* 286 * We have found a new contiguous block. 287 * 288 * First we have to replace the old block pointers with the new 289 * block pointers in the inode and indirect blocks associated 290 * with the file. 291 */ 292 #ifdef DEBUG 293 printf("realloc: ino %ju, lbns %jd-%jd\n\told:", 294 (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); 295 #endif /* DEBUG */ 296 blkno = newblk; 297 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) { 298 if (i == ssize) { 299 bap = ebap; 300 soff = -i; 301 } 302 #ifdef INVARIANTS 303 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap)) 304 panic("ext2_reallocblks: alloc mismatch"); 305 #endif 306 #ifdef DEBUG 307 printf(" %d,", *bap); 308 #endif /* DEBUG */ 309 *bap++ = blkno; 310 } 311 /* 312 * Next we must write out the modified inode and indirect blocks. 313 * For strict correctness, the writes should be synchronous since 314 * the old block values may have been written to disk. In practise 315 * they are almost never written, but if we are concerned about 316 * strict correctness, the `doasyncfree' flag should be set to zero. 317 * 318 * The test on `doasyncfree' should be changed to test a flag 319 * that shows whether the associated buffers and inodes have 320 * been written. The flag should be set when the cluster is 321 * started and cleared whenever the buffer or inode is flushed. 322 * We can then check below to see if it is set, and do the 323 * synchronous write only when it has been cleared. 324 */ 325 if (sbap != &ip->i_db[0]) { 326 if (doasyncfree) 327 bdwrite(sbp); 328 else 329 bwrite(sbp); 330 } else { 331 ip->i_flag |= IN_CHANGE | IN_UPDATE; 332 if (!doasyncfree) 333 ext2_update(vp, 1); 334 } 335 if (ssize < len) { 336 if (doasyncfree) 337 bdwrite(ebp); 338 else 339 bwrite(ebp); 340 } 341 /* 342 * Last, free the old blocks and assign the new blocks to the buffers. 343 */ 344 #ifdef DEBUG 345 printf("\n\tnew:"); 346 #endif /* DEBUG */ 347 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) { 348 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), 349 fs->e2fs_bsize); 350 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 351 #ifdef DEBUG 352 printf(" %d,", blkno); 353 #endif /* DEBUG */ 354 } 355 #ifdef DEBUG 356 printf("\n"); 357 #endif /* DEBUG */ 358 return (0); 359 360 fail: 361 if (ssize < len) 362 brelse(ebp); 363 if (sbap != &ip->i_db[0]) 364 brelse(sbp); 365 return (ENOSPC); 366 } 367 368 /* 369 * Allocate an inode in the filesystem. 370 * 371 */ 372 int 373 ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp) 374 { 375 struct timespec ts; 376 struct inode *pip; 377 struct m_ext2fs *fs; 378 struct inode *ip; 379 struct ext2mount *ump; 380 ino_t ino, ipref; 381 int error, cg; 382 383 *vpp = NULL; 384 pip = VTOI(pvp); 385 fs = pip->i_e2fs; 386 ump = pip->i_ump; 387 388 EXT2_LOCK(ump); 389 if (fs->e2fs->e2fs_ficount == 0) 390 goto noinodes; 391 /* 392 * If it is a directory then obtain a cylinder group based on 393 * ext2_dirpref else obtain it using ino_to_cg. The preferred inode is 394 * always the next inode. 395 */ 396 if ((mode & IFMT) == IFDIR) { 397 cg = ext2_dirpref(pip); 398 if (fs->e2fs_contigdirs[cg] < 255) 399 fs->e2fs_contigdirs[cg]++; 400 } else { 401 cg = ino_to_cg(fs, pip->i_number); 402 if (fs->e2fs_contigdirs[cg] > 0) 403 fs->e2fs_contigdirs[cg]--; 404 } 405 ipref = cg * fs->e2fs->e2fs_ipg + 1; 406 ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg); 407 408 if (ino == 0) 409 goto noinodes; 410 error = VFS_VGET(pvp->v_mount, ino, LK_EXCLUSIVE, vpp); 411 if (error) { 412 ext2_vfree(pvp, ino, mode); 413 return (error); 414 } 415 ip = VTOI(*vpp); 416 417 /* 418 * The question is whether using VGET was such good idea at all: 419 * Linux doesn't read the old inode in when it is allocating a 420 * new one. I will set at least i_size and i_blocks to zero. 421 */ 422 ip->i_flag = 0; 423 ip->i_size = 0; 424 ip->i_blocks = 0; 425 ip->i_mode = 0; 426 ip->i_flags = 0; 427 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_EXTENTS) 428 && (S_ISREG(mode) || S_ISDIR(mode))) 429 ext4_ext_tree_init(ip); 430 else 431 memset(ip->i_data, 0, sizeof(ip->i_data)); 432 433 434 /* 435 * Set up a new generation number for this inode. 436 * Avoid zero values. 437 */ 438 do { 439 ip->i_gen = arc4random(); 440 } while (ip->i_gen == 0); 441 442 vfs_timestamp(&ts); 443 ip->i_birthtime = ts.tv_sec; 444 ip->i_birthnsec = ts.tv_nsec; 445 446 /* 447 printf("ext2_valloc: allocated inode %d\n", ino); 448 */ 449 return (0); 450 noinodes: 451 EXT2_UNLOCK(ump); 452 ext2_fserr(fs, cred->cr_uid, "out of inodes"); 453 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->e2fs_fsmnt); 454 return (ENOSPC); 455 } 456 457 /* 458 * 64-bit compatible getters and setters for struct ext2_gd from ext2fs.h 459 */ 460 uint64_t 461 e2fs_gd_get_b_bitmap(struct ext2_gd *gd) 462 { 463 464 return (((uint64_t)(gd->ext4bgd_b_bitmap_hi) << 32) | 465 gd->ext2bgd_b_bitmap); 466 } 467 468 uint64_t 469 e2fs_gd_get_i_bitmap(struct ext2_gd *gd) 470 { 471 472 return (((uint64_t)(gd->ext4bgd_i_bitmap_hi) << 32) | 473 gd->ext2bgd_i_bitmap); 474 } 475 476 uint64_t 477 e2fs_gd_get_i_tables(struct ext2_gd *gd) 478 { 479 480 return (((uint64_t)(gd->ext4bgd_i_tables_hi) << 32) | 481 gd->ext2bgd_i_tables); 482 } 483 484 static uint32_t 485 e2fs_gd_get_nbfree(struct ext2_gd *gd) 486 { 487 488 return (((uint32_t)(gd->ext4bgd_nbfree_hi) << 16) | 489 gd->ext2bgd_nbfree); 490 } 491 492 static void 493 e2fs_gd_set_nbfree(struct ext2_gd *gd, uint32_t val) 494 { 495 496 gd->ext2bgd_nbfree = val & 0xffff; 497 gd->ext4bgd_nbfree_hi = val >> 16; 498 } 499 500 static uint32_t 501 e2fs_gd_get_nifree(struct ext2_gd *gd) 502 { 503 504 return (((uint32_t)(gd->ext4bgd_nifree_hi) << 16) | 505 gd->ext2bgd_nifree); 506 } 507 508 static void 509 e2fs_gd_set_nifree(struct ext2_gd *gd, uint32_t val) 510 { 511 512 gd->ext2bgd_nifree = val & 0xffff; 513 gd->ext4bgd_nifree_hi = val >> 16; 514 } 515 516 uint32_t 517 e2fs_gd_get_ndirs(struct ext2_gd *gd) 518 { 519 520 return (((uint32_t)(gd->ext4bgd_ndirs_hi) << 16) | 521 gd->ext2bgd_ndirs); 522 } 523 524 static void 525 e2fs_gd_set_ndirs(struct ext2_gd *gd, uint32_t val) 526 { 527 528 gd->ext2bgd_ndirs = val & 0xffff; 529 gd->ext4bgd_ndirs_hi = val >> 16; 530 } 531 532 static uint32_t 533 e2fs_gd_get_i_unused(struct ext2_gd *gd) 534 { 535 return (((uint32_t)(gd->ext4bgd_i_unused_hi) << 16) | 536 gd->ext4bgd_i_unused); 537 } 538 539 static void 540 e2fs_gd_set_i_unused(struct ext2_gd *gd, uint32_t val) 541 { 542 543 gd->ext4bgd_i_unused = val & 0xffff; 544 gd->ext4bgd_i_unused_hi = val >> 16; 545 } 546 547 /* 548 * Find a cylinder to place a directory. 549 * 550 * The policy implemented by this algorithm is to allocate a 551 * directory inode in the same cylinder group as its parent 552 * directory, but also to reserve space for its files inodes 553 * and data. Restrict the number of directories which may be 554 * allocated one after another in the same cylinder group 555 * without intervening allocation of files. 556 * 557 * If we allocate a first level directory then force allocation 558 * in another cylinder group. 559 * 560 */ 561 static u_long 562 ext2_dirpref(struct inode *pip) 563 { 564 struct m_ext2fs *fs; 565 int cg, prefcg, cgsize; 566 uint64_t avgbfree, minbfree; 567 u_int avgifree, avgndir, curdirsize; 568 u_int minifree, maxndir; 569 u_int mincg, minndir; 570 u_int dirsize, maxcontigdirs; 571 572 mtx_assert(EXT2_MTX(pip->i_ump), MA_OWNED); 573 fs = pip->i_e2fs; 574 575 avgifree = fs->e2fs->e2fs_ficount / fs->e2fs_gcount; 576 avgbfree = fs->e2fs_fbcount / fs->e2fs_gcount; 577 avgndir = fs->e2fs_total_dir / fs->e2fs_gcount; 578 579 /* 580 * Force allocation in another cg if creating a first level dir. 581 */ 582 ASSERT_VOP_LOCKED(ITOV(pip), "ext2fs_dirpref"); 583 if (ITOV(pip)->v_vflag & VV_ROOT) { 584 prefcg = arc4random() % fs->e2fs_gcount; 585 mincg = prefcg; 586 minndir = fs->e2fs_ipg; 587 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 588 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < minndir && 589 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree && 590 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= avgbfree) { 591 mincg = cg; 592 minndir = e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]); 593 } 594 for (cg = 0; cg < prefcg; cg++) 595 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < minndir && 596 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree && 597 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= avgbfree) { 598 mincg = cg; 599 minndir = e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]); 600 } 601 return (mincg); 602 } 603 /* 604 * Count various limits which used for 605 * optimal allocation of a directory inode. 606 */ 607 maxndir = min(avgndir + fs->e2fs_ipg / 16, fs->e2fs_ipg); 608 minifree = avgifree - avgifree / 4; 609 if (minifree < 1) 610 minifree = 1; 611 minbfree = avgbfree - avgbfree / 4; 612 if (minbfree < 1) 613 minbfree = 1; 614 cgsize = fs->e2fs_fsize * fs->e2fs_fpg; 615 dirsize = AVGDIRSIZE; 616 curdirsize = avgndir ? (cgsize - avgbfree * fs->e2fs_bsize) / avgndir : 0; 617 if (dirsize < curdirsize) 618 dirsize = curdirsize; 619 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255); 620 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR); 621 if (maxcontigdirs == 0) 622 maxcontigdirs = 1; 623 624 /* 625 * Limit number of dirs in one cg and reserve space for 626 * regular files, but only if we have no deficit in 627 * inodes or space. 628 */ 629 prefcg = ino_to_cg(fs, pip->i_number); 630 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 631 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < maxndir && 632 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= minifree && 633 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= minbfree) { 634 if (fs->e2fs_contigdirs[cg] < maxcontigdirs) 635 return (cg); 636 } 637 for (cg = 0; cg < prefcg; cg++) 638 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < maxndir && 639 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= minifree && 640 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= minbfree) { 641 if (fs->e2fs_contigdirs[cg] < maxcontigdirs) 642 return (cg); 643 } 644 /* 645 * This is a backstop when we have deficit in space. 646 */ 647 for (cg = prefcg; cg < fs->e2fs_gcount; cg++) 648 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree) 649 return (cg); 650 for (cg = 0; cg < prefcg; cg++) 651 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree) 652 break; 653 return (cg); 654 } 655 656 /* 657 * Select the desired position for the next block in a file. 658 * 659 * we try to mimic what Remy does in inode_getblk/block_getblk 660 * 661 * we note: blocknr == 0 means that we're about to allocate either 662 * a direct block or a pointer block at the first level of indirection 663 * (In other words, stuff that will go in i_db[] or i_ib[]) 664 * 665 * blocknr != 0 means that we're allocating a block that is none 666 * of the above. Then, blocknr tells us the number of the block 667 * that will hold the pointer 668 */ 669 e4fs_daddr_t 670 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, e2fs_daddr_t *bap, 671 e2fs_daddr_t blocknr) 672 { 673 struct m_ext2fs *fs; 674 int tmp; 675 676 fs = ip->i_e2fs; 677 678 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED); 679 680 /* 681 * If the next block is actually what we thought it is, then set the 682 * goal to what we thought it should be. 683 */ 684 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0) 685 return ip->i_next_alloc_goal; 686 687 /* 688 * Now check whether we were provided with an array that basically 689 * tells us previous blocks to which we want to stay close. 690 */ 691 if (bap) 692 for (tmp = indx - 1; tmp >= 0; tmp--) 693 if (bap[tmp]) 694 return bap[tmp]; 695 696 /* 697 * Else lets fall back to the blocknr or, if there is none, follow 698 * the rule that a block should be allocated near its inode. 699 */ 700 return (blocknr ? blocknr : 701 (e2fs_daddr_t)(ip->i_block_group * 702 EXT2_BLOCKS_PER_GROUP(fs)) + fs->e2fs->e2fs_first_dblock); 703 } 704 705 /* 706 * Implement the cylinder overflow algorithm. 707 * 708 * The policy implemented by this algorithm is: 709 * 1) allocate the block in its requested cylinder group. 710 * 2) quadradically rehash on the cylinder group number. 711 * 3) brute force search for a free block. 712 */ 713 static e4fs_daddr_t 714 ext2_hashalloc(struct inode *ip, int cg, long pref, int size, 715 daddr_t (*allocator) (struct inode *, int, daddr_t, int)) 716 { 717 struct m_ext2fs *fs; 718 e4fs_daddr_t result; 719 int i, icg = cg; 720 721 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED); 722 fs = ip->i_e2fs; 723 /* 724 * 1: preferred cylinder group 725 */ 726 result = (*allocator)(ip, cg, pref, size); 727 if (result) 728 return (result); 729 /* 730 * 2: quadratic rehash 731 */ 732 for (i = 1; i < fs->e2fs_gcount; i *= 2) { 733 cg += i; 734 if (cg >= fs->e2fs_gcount) 735 cg -= fs->e2fs_gcount; 736 result = (*allocator)(ip, cg, 0, size); 737 if (result) 738 return (result); 739 } 740 /* 741 * 3: brute force search 742 * Note that we start at i == 2, since 0 was checked initially, 743 * and 1 is always checked in the quadratic rehash. 744 */ 745 cg = (icg + 2) % fs->e2fs_gcount; 746 for (i = 2; i < fs->e2fs_gcount; i++) { 747 result = (*allocator)(ip, cg, 0, size); 748 if (result) 749 return (result); 750 cg++; 751 if (cg == fs->e2fs_gcount) 752 cg = 0; 753 } 754 return (0); 755 } 756 757 static uint64_t 758 ext2_cg_number_gdb_nometa(struct m_ext2fs *fs, int cg) 759 { 760 761 if (!ext2_cg_has_sb(fs, cg)) 762 return (0); 763 764 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG)) 765 return (fs->e2fs->e3fs_first_meta_bg); 766 767 return ((fs->e2fs_gcount + EXT2_DESCS_PER_BLOCK(fs) - 1) / 768 EXT2_DESCS_PER_BLOCK(fs)); 769 } 770 771 static uint64_t 772 ext2_cg_number_gdb_meta(struct m_ext2fs *fs, int cg) 773 { 774 unsigned long metagroup; 775 int first, last; 776 777 metagroup = cg / EXT2_DESCS_PER_BLOCK(fs); 778 first = metagroup * EXT2_DESCS_PER_BLOCK(fs); 779 last = first + EXT2_DESCS_PER_BLOCK(fs) - 1; 780 781 if (cg == first || cg == first + 1 || cg == last) 782 return (1); 783 784 return (0); 785 } 786 787 uint64_t 788 ext2_cg_number_gdb(struct m_ext2fs *fs, int cg) 789 { 790 unsigned long first_meta_bg, metagroup; 791 792 first_meta_bg = fs->e2fs->e3fs_first_meta_bg; 793 metagroup = cg / EXT2_DESCS_PER_BLOCK(fs); 794 795 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) || 796 metagroup < first_meta_bg) 797 return (ext2_cg_number_gdb_nometa(fs, cg)); 798 799 return ext2_cg_number_gdb_meta(fs, cg); 800 } 801 802 static int 803 ext2_number_base_meta_blocks(struct m_ext2fs *fs, int cg) 804 { 805 int number; 806 807 number = ext2_cg_has_sb(fs, cg); 808 809 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) || 810 cg < fs->e2fs->e3fs_first_meta_bg * EXT2_DESCS_PER_BLOCK(fs)) { 811 if (number) { 812 number += ext2_cg_number_gdb(fs, cg); 813 number += fs->e2fs->e2fs_reserved_ngdb; 814 } 815 } else { 816 number += ext2_cg_number_gdb(fs, cg); 817 } 818 819 return (number); 820 } 821 822 static void 823 ext2_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) 824 { 825 int i; 826 827 if (start_bit >= end_bit) 828 return; 829 830 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) 831 setbit(bitmap, i); 832 if (i < end_bit) 833 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); 834 } 835 836 static int 837 ext2_get_group_number(struct m_ext2fs *fs, e4fs_daddr_t block) 838 { 839 840 return ((block - fs->e2fs->e2fs_first_dblock) / fs->e2fs_bsize); 841 } 842 843 static int 844 ext2_block_in_group(struct m_ext2fs *fs, e4fs_daddr_t block, int cg) 845 { 846 847 return ((ext2_get_group_number(fs, block) == cg) ? 1 : 0); 848 } 849 850 static int 851 ext2_cg_block_bitmap_init(struct m_ext2fs *fs, int cg, struct buf *bp) 852 { 853 int bit, bit_max, inodes_per_block; 854 uint64_t start, tmp; 855 856 if (!(fs->e2fs_gd[cg].ext4bgd_flags & EXT2_BG_BLOCK_UNINIT)) 857 return (0); 858 859 memset(bp->b_data, 0, fs->e2fs_bsize); 860 861 bit_max = ext2_number_base_meta_blocks(fs, cg); 862 if ((bit_max >> 3) >= fs->e2fs_bsize) 863 return (EINVAL); 864 865 for (bit = 0; bit < bit_max; bit++) 866 setbit(bp->b_data, bit); 867 868 start = (uint64_t)cg * fs->e2fs->e2fs_bpg + fs->e2fs->e2fs_first_dblock; 869 870 /* Set bits for block and inode bitmaps, and inode table. */ 871 tmp = e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg]); 872 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) || 873 ext2_block_in_group(fs, tmp, cg)) 874 setbit(bp->b_data, tmp - start); 875 876 tmp = e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg]); 877 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) || 878 ext2_block_in_group(fs, tmp, cg)) 879 setbit(bp->b_data, tmp - start); 880 881 tmp = e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]); 882 inodes_per_block = fs->e2fs_bsize/EXT2_INODE_SIZE(fs); 883 while( tmp < e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]) + 884 fs->e2fs->e2fs_ipg / inodes_per_block ) { 885 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) || 886 ext2_block_in_group(fs, tmp, cg)) 887 setbit(bp->b_data, tmp - start); 888 tmp++; 889 } 890 891 /* 892 * Also if the number of blocks within the group is less than 893 * the blocksize * 8 ( which is the size of bitmap ), set rest 894 * of the block bitmap to 1 895 */ 896 ext2_mark_bitmap_end(fs->e2fs->e2fs_bpg, fs->e2fs_bsize * 8, 897 bp->b_data); 898 899 /* Clean the flag */ 900 fs->e2fs_gd[cg].ext4bgd_flags &= ~EXT2_BG_BLOCK_UNINIT; 901 902 return (0); 903 } 904 905 static int 906 ext2_b_bitmap_validate(struct m_ext2fs *fs, struct buf *bp, int cg) 907 { 908 struct ext2_gd *gd; 909 uint64_t group_first_block; 910 unsigned int offset, max_bit; 911 912 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG)) { 913 /* 914 * It is not possible to check block bitmap in case of this feature, 915 * because the inode and block bitmaps and inode table 916 * blocks may not be in the group at all. 917 * So, skip check in this case. 918 */ 919 return (0); 920 } 921 922 gd = &fs->e2fs_gd[cg]; 923 max_bit = fs->e2fs_fpg; 924 group_first_block = ((uint64_t)cg) * fs->e2fs->e2fs_fpg + 925 fs->e2fs->e2fs_first_dblock; 926 927 /* Check block bitmap block number */ 928 offset = e2fs_gd_get_b_bitmap(gd) - group_first_block; 929 if (offset >= max_bit || !isset(bp->b_data, offset)) { 930 printf("ext2fs: bad block bitmap, group %d\n", cg); 931 return (EINVAL); 932 } 933 934 /* Check inode bitmap block number */ 935 offset = e2fs_gd_get_i_bitmap(gd) - group_first_block; 936 if (offset >= max_bit || !isset(bp->b_data, offset)) { 937 printf("ext2fs: bad inode bitmap, group %d\n", cg); 938 return (EINVAL); 939 } 940 941 /* Check inode table */ 942 offset = e2fs_gd_get_i_tables(gd) - group_first_block; 943 if (offset >= max_bit || offset + fs->e2fs_itpg >= max_bit) { 944 printf("ext2fs: bad inode table, group %d\n", cg); 945 return (EINVAL); 946 } 947 948 return (0); 949 } 950 951 /* 952 * Determine whether a block can be allocated. 953 * 954 * Check to see if a block of the appropriate size is available, 955 * and if it is, allocate it. 956 */ 957 static daddr_t 958 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size) 959 { 960 struct m_ext2fs *fs; 961 struct buf *bp; 962 struct ext2mount *ump; 963 daddr_t bno, runstart, runlen; 964 int bit, loc, end, error, start; 965 char *bbp; 966 /* XXX ondisk32 */ 967 fs = ip->i_e2fs; 968 ump = ip->i_ump; 969 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0) 970 return (0); 971 972 EXT2_UNLOCK(ump); 973 error = bread(ip->i_devvp, fsbtodb(fs, 974 e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])), 975 (int)fs->e2fs_bsize, NOCRED, &bp); 976 if (error) 977 goto fail; 978 979 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) || 980 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { 981 error = ext2_cg_block_bitmap_init(fs, cg, bp); 982 if (error) 983 goto fail; 984 985 ext2_gd_b_bitmap_csum_set(fs, cg, bp); 986 } 987 error = ext2_gd_b_bitmap_csum_verify(fs, cg, bp); 988 if (error) 989 goto fail; 990 991 error = ext2_b_bitmap_validate(fs,bp, cg); 992 if (error) 993 goto fail; 994 995 /* 996 * Check, that another thread did not not allocate the last block in this 997 * group while we were waiting for the buffer. 998 */ 999 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0) 1000 goto fail; 1001 1002 bbp = (char *)bp->b_data; 1003 1004 if (dtog(fs, bpref) != cg) 1005 bpref = 0; 1006 if (bpref != 0) { 1007 bpref = dtogd(fs, bpref); 1008 /* 1009 * if the requested block is available, use it 1010 */ 1011 if (isclr(bbp, bpref)) { 1012 bno = bpref; 1013 goto gotit; 1014 } 1015 } 1016 /* 1017 * no blocks in the requested cylinder, so take next 1018 * available one in this cylinder group. 1019 * first try to get 8 contigous blocks, then fall back to a single 1020 * block. 1021 */ 1022 if (bpref) 1023 start = dtogd(fs, bpref) / NBBY; 1024 else 1025 start = 0; 1026 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start; 1027 retry: 1028 runlen = 0; 1029 runstart = 0; 1030 for (loc = start; loc < end; loc++) { 1031 if (bbp[loc] == (char)0xff) { 1032 runlen = 0; 1033 continue; 1034 } 1035 1036 /* Start of a run, find the number of high clear bits. */ 1037 if (runlen == 0) { 1038 bit = fls(bbp[loc]); 1039 runlen = NBBY - bit; 1040 runstart = loc * NBBY + bit; 1041 } else if (bbp[loc] == 0) { 1042 /* Continue a run. */ 1043 runlen += NBBY; 1044 } else { 1045 /* 1046 * Finish the current run. If it isn't long 1047 * enough, start a new one. 1048 */ 1049 bit = ffs(bbp[loc]) - 1; 1050 runlen += bit; 1051 if (runlen >= 8) { 1052 bno = runstart; 1053 goto gotit; 1054 } 1055 1056 /* Run was too short, start a new one. */ 1057 bit = fls(bbp[loc]); 1058 runlen = NBBY - bit; 1059 runstart = loc * NBBY + bit; 1060 } 1061 1062 /* If the current run is long enough, use it. */ 1063 if (runlen >= 8) { 1064 bno = runstart; 1065 goto gotit; 1066 } 1067 } 1068 if (start != 0) { 1069 end = start; 1070 start = 0; 1071 goto retry; 1072 } 1073 bno = ext2_mapsearch(fs, bbp, bpref); 1074 if (bno < 0) 1075 goto fail; 1076 1077 gotit: 1078 #ifdef INVARIANTS 1079 if (isset(bbp, bno)) { 1080 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n", 1081 cg, (intmax_t)bno, fs->e2fs_fsmnt); 1082 panic("ext2fs_alloccg: dup alloc"); 1083 } 1084 #endif 1085 setbit(bbp, bno); 1086 EXT2_LOCK(ump); 1087 ext2_clusteracct(fs, bbp, cg, bno, -1); 1088 fs->e2fs_fbcount--; 1089 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg], 1090 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1); 1091 fs->e2fs_fmod = 1; 1092 EXT2_UNLOCK(ump); 1093 ext2_gd_b_bitmap_csum_set(fs, cg, bp); 1094 bdwrite(bp); 1095 return (((uint64_t)cg) * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno); 1096 1097 fail: 1098 brelse(bp); 1099 EXT2_LOCK(ump); 1100 return (0); 1101 } 1102 1103 /* 1104 * Determine whether a cluster can be allocated. 1105 */ 1106 static daddr_t 1107 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len) 1108 { 1109 struct m_ext2fs *fs; 1110 struct ext2mount *ump; 1111 struct buf *bp; 1112 char *bbp; 1113 int bit, error, got, i, loc, run; 1114 int32_t *lp; 1115 daddr_t bno; 1116 1117 fs = ip->i_e2fs; 1118 ump = ip->i_ump; 1119 1120 if (fs->e2fs_maxcluster[cg] < len) 1121 return (0); 1122 1123 EXT2_UNLOCK(ump); 1124 error = bread(ip->i_devvp, 1125 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])), 1126 (int)fs->e2fs_bsize, NOCRED, &bp); 1127 if (error) 1128 goto fail_lock; 1129 1130 bbp = (char *)bp->b_data; 1131 EXT2_LOCK(ump); 1132 /* 1133 * Check to see if a cluster of the needed size (or bigger) is 1134 * available in this cylinder group. 1135 */ 1136 lp = &fs->e2fs_clustersum[cg].cs_sum[len]; 1137 for (i = len; i <= fs->e2fs_contigsumsize; i++) 1138 if (*lp++ > 0) 1139 break; 1140 if (i > fs->e2fs_contigsumsize) { 1141 /* 1142 * Update the cluster summary information to reflect 1143 * the true maximum-sized cluster so that future cluster 1144 * allocation requests can avoid reading the bitmap only 1145 * to find no cluster. 1146 */ 1147 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1]; 1148 for (i = len - 1; i > 0; i--) 1149 if (*lp-- > 0) 1150 break; 1151 fs->e2fs_maxcluster[cg] = i; 1152 goto fail; 1153 } 1154 EXT2_UNLOCK(ump); 1155 1156 /* Search the bitmap to find a big enough cluster like in FFS. */ 1157 if (dtog(fs, bpref) != cg) 1158 bpref = 0; 1159 if (bpref != 0) 1160 bpref = dtogd(fs, bpref); 1161 loc = bpref / NBBY; 1162 bit = 1 << (bpref % NBBY); 1163 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) { 1164 if ((bbp[loc] & bit) != 0) 1165 run = 0; 1166 else { 1167 run++; 1168 if (run == len) 1169 break; 1170 } 1171 if ((got & (NBBY - 1)) != (NBBY - 1)) 1172 bit <<= 1; 1173 else { 1174 loc++; 1175 bit = 1; 1176 } 1177 } 1178 1179 if (got >= fs->e2fs->e2fs_fpg) 1180 goto fail_lock; 1181 1182 /* Allocate the cluster that we found. */ 1183 for (i = 1; i < len; i++) 1184 if (!isclr(bbp, got - run + i)) 1185 panic("ext2_clusteralloc: map mismatch"); 1186 1187 bno = got - run + 1; 1188 if (bno >= fs->e2fs->e2fs_fpg) 1189 panic("ext2_clusteralloc: allocated out of group"); 1190 1191 EXT2_LOCK(ump); 1192 for (i = 0; i < len; i += fs->e2fs_fpb) { 1193 setbit(bbp, bno + i); 1194 ext2_clusteracct(fs, bbp, cg, bno + i, -1); 1195 fs->e2fs_fbcount--; 1196 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg], 1197 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1); 1198 } 1199 fs->e2fs_fmod = 1; 1200 EXT2_UNLOCK(ump); 1201 1202 bdwrite(bp); 1203 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno); 1204 1205 fail_lock: 1206 EXT2_LOCK(ump); 1207 fail: 1208 brelse(bp); 1209 return (0); 1210 } 1211 1212 static int 1213 ext2_zero_inode_table(struct inode *ip, int cg) 1214 { 1215 struct m_ext2fs *fs; 1216 struct buf *bp; 1217 int i, all_blks, used_blks; 1218 1219 fs = ip->i_e2fs; 1220 1221 if (fs->e2fs_gd[cg].ext4bgd_flags & EXT2_BG_INODE_ZEROED) 1222 return (0); 1223 1224 all_blks = fs->e2fs->e2fs_inode_size * fs->e2fs->e2fs_ipg / 1225 fs->e2fs_bsize; 1226 1227 used_blks = howmany(fs->e2fs->e2fs_ipg - 1228 e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]), 1229 fs->e2fs_bsize / EXT2_INODE_SIZE(fs)); 1230 1231 for (i = 0; i < all_blks - used_blks; i++) { 1232 bp = getblk(ip->i_devvp, fsbtodb(fs, 1233 e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]) + used_blks + i), 1234 fs->e2fs_bsize, 0, 0, 0); 1235 if (!bp) 1236 return (EIO); 1237 1238 vfs_bio_bzero_buf(bp, 0, fs->e2fs_bsize); 1239 bawrite(bp); 1240 } 1241 1242 fs->e2fs_gd[cg].ext4bgd_flags |= EXT2_BG_INODE_ZEROED; 1243 1244 return (0); 1245 } 1246 1247 /* 1248 * Determine whether an inode can be allocated. 1249 * 1250 * Check to see if an inode is available, and if it is, 1251 * allocate it using tode in the specified cylinder group. 1252 */ 1253 static daddr_t 1254 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode) 1255 { 1256 struct m_ext2fs *fs; 1257 struct buf *bp; 1258 struct ext2mount *ump; 1259 int error, start, len, ifree; 1260 char *ibp, *loc; 1261 1262 ipref--; /* to avoid a lot of (ipref -1) */ 1263 if (ipref == -1) 1264 ipref = 0; 1265 fs = ip->i_e2fs; 1266 ump = ip->i_ump; 1267 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0) 1268 return (0); 1269 EXT2_UNLOCK(ump); 1270 error = bread(ip->i_devvp, fsbtodb(fs, 1271 e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])), 1272 (int)fs->e2fs_bsize, NOCRED, &bp); 1273 if (error) { 1274 brelse(bp); 1275 EXT2_LOCK(ump); 1276 return (0); 1277 } 1278 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) || 1279 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { 1280 if (fs->e2fs_gd[cg].ext4bgd_flags & EXT2_BG_INODE_UNINIT) { 1281 memset(bp->b_data, 0, fs->e2fs_bsize); 1282 fs->e2fs_gd[cg].ext4bgd_flags &= ~EXT2_BG_INODE_UNINIT; 1283 } 1284 ext2_gd_i_bitmap_csum_set(fs, cg, bp); 1285 error = ext2_zero_inode_table(ip, cg); 1286 if (error) { 1287 brelse(bp); 1288 EXT2_LOCK(ump); 1289 return (0); 1290 } 1291 } 1292 error = ext2_gd_i_bitmap_csum_verify(fs, cg, bp); 1293 if (error) { 1294 brelse(bp); 1295 EXT2_LOCK(ump); 1296 return (0); 1297 } 1298 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0) { 1299 /* 1300 * Another thread allocated the last i-node in this 1301 * group while we were waiting for the buffer. 1302 */ 1303 brelse(bp); 1304 EXT2_LOCK(ump); 1305 return (0); 1306 } 1307 ibp = (char *)bp->b_data; 1308 if (ipref) { 1309 ipref %= fs->e2fs->e2fs_ipg; 1310 if (isclr(ibp, ipref)) 1311 goto gotit; 1312 } 1313 start = ipref / NBBY; 1314 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY); 1315 loc = memcchr(&ibp[start], 0xff, len); 1316 if (loc == NULL) { 1317 len = start + 1; 1318 start = 0; 1319 loc = memcchr(&ibp[start], 0xff, len); 1320 if (loc == NULL) { 1321 printf("ext2fs: inode bitmap corrupted: " 1322 "cg = %d, ipref = %lld, fs = %s - run fsck\n", 1323 cg, (long long)ipref, fs->e2fs_fsmnt); 1324 brelse(bp); 1325 EXT2_LOCK(ump); 1326 return (0); 1327 } 1328 } 1329 ipref = (loc - ibp) * NBBY + ffs(~*loc) - 1; 1330 gotit: 1331 setbit(ibp, ipref); 1332 EXT2_LOCK(ump); 1333 e2fs_gd_set_nifree(&fs->e2fs_gd[cg], 1334 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) - 1); 1335 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) || 1336 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { 1337 ifree = fs->e2fs->e2fs_ipg - e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]); 1338 if (ipref + 1 > ifree) 1339 e2fs_gd_set_i_unused(&fs->e2fs_gd[cg], 1340 fs->e2fs->e2fs_ipg - (ipref + 1)); 1341 } 1342 fs->e2fs->e2fs_ficount--; 1343 fs->e2fs_fmod = 1; 1344 if ((mode & IFMT) == IFDIR) { 1345 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg], 1346 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) + 1); 1347 fs->e2fs_total_dir++; 1348 } 1349 EXT2_UNLOCK(ump); 1350 ext2_gd_i_bitmap_csum_set(fs, cg, bp); 1351 bdwrite(bp); 1352 return ((uint64_t)cg * fs->e2fs_ipg + ipref + 1); 1353 } 1354 1355 /* 1356 * Free a block or fragment. 1357 * 1358 */ 1359 void 1360 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size) 1361 { 1362 struct m_ext2fs *fs; 1363 struct buf *bp; 1364 struct ext2mount *ump; 1365 int cg, error; 1366 char *bbp; 1367 1368 fs = ip->i_e2fs; 1369 ump = ip->i_ump; 1370 cg = dtog(fs, bno); 1371 if (bno >= fs->e2fs_bcount) { 1372 printf("bad block %lld, ino %ju\n", (long long)bno, 1373 (uintmax_t)ip->i_number); 1374 ext2_fserr(fs, ip->i_uid, "bad block"); 1375 return; 1376 } 1377 error = bread(ip->i_devvp, 1378 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])), 1379 (int)fs->e2fs_bsize, NOCRED, &bp); 1380 if (error) { 1381 brelse(bp); 1382 return; 1383 } 1384 bbp = (char *)bp->b_data; 1385 bno = dtogd(fs, bno); 1386 if (isclr(bbp, bno)) { 1387 printf("block = %lld, fs = %s\n", 1388 (long long)bno, fs->e2fs_fsmnt); 1389 panic("ext2_blkfree: freeing free block"); 1390 } 1391 clrbit(bbp, bno); 1392 EXT2_LOCK(ump); 1393 ext2_clusteracct(fs, bbp, cg, bno, 1); 1394 fs->e2fs_fbcount++; 1395 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg], 1396 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) + 1); 1397 fs->e2fs_fmod = 1; 1398 EXT2_UNLOCK(ump); 1399 ext2_gd_b_bitmap_csum_set(fs, cg, bp); 1400 bdwrite(bp); 1401 } 1402 1403 /* 1404 * Free an inode. 1405 * 1406 */ 1407 int 1408 ext2_vfree(struct vnode *pvp, ino_t ino, int mode) 1409 { 1410 struct m_ext2fs *fs; 1411 struct inode *pip; 1412 struct buf *bp; 1413 struct ext2mount *ump; 1414 int error, cg; 1415 char *ibp; 1416 1417 pip = VTOI(pvp); 1418 fs = pip->i_e2fs; 1419 ump = pip->i_ump; 1420 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount) 1421 panic("ext2_vfree: range: devvp = %p, ino = %ju, fs = %s", 1422 pip->i_devvp, (uintmax_t)ino, fs->e2fs_fsmnt); 1423 1424 cg = ino_to_cg(fs, ino); 1425 error = bread(pip->i_devvp, 1426 fsbtodb(fs, e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])), 1427 (int)fs->e2fs_bsize, NOCRED, &bp); 1428 if (error) { 1429 brelse(bp); 1430 return (0); 1431 } 1432 ibp = (char *)bp->b_data; 1433 ino = (ino - 1) % fs->e2fs->e2fs_ipg; 1434 if (isclr(ibp, ino)) { 1435 printf("ino = %ju, fs = %s\n", 1436 ino, fs->e2fs_fsmnt); 1437 if (fs->e2fs_ronly == 0) 1438 panic("ext2_vfree: freeing free inode"); 1439 } 1440 clrbit(ibp, ino); 1441 EXT2_LOCK(ump); 1442 fs->e2fs->e2fs_ficount++; 1443 e2fs_gd_set_nifree(&fs->e2fs_gd[cg], 1444 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) + 1); 1445 if ((mode & IFMT) == IFDIR) { 1446 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg], 1447 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) - 1); 1448 fs->e2fs_total_dir--; 1449 } 1450 fs->e2fs_fmod = 1; 1451 EXT2_UNLOCK(ump); 1452 ext2_gd_i_bitmap_csum_set(fs, cg, bp); 1453 bdwrite(bp); 1454 return (0); 1455 } 1456 1457 /* 1458 * Find a block in the specified cylinder group. 1459 * 1460 * It is a panic if a request is made to find a block if none are 1461 * available. 1462 */ 1463 static daddr_t 1464 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref) 1465 { 1466 char *loc; 1467 int start, len; 1468 1469 /* 1470 * find the fragment by searching through the free block 1471 * map for an appropriate bit pattern 1472 */ 1473 if (bpref) 1474 start = dtogd(fs, bpref) / NBBY; 1475 else 1476 start = 0; 1477 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start; 1478 loc = memcchr(&bbp[start], 0xff, len); 1479 if (loc == NULL) { 1480 len = start + 1; 1481 start = 0; 1482 loc = memcchr(&bbp[start], 0xff, len); 1483 if (loc == NULL) { 1484 printf("start = %d, len = %d, fs = %s\n", 1485 start, len, fs->e2fs_fsmnt); 1486 panic("ext2_mapsearch: map corrupted"); 1487 /* NOTREACHED */ 1488 } 1489 } 1490 return ((loc - bbp) * NBBY + ffs(~*loc) - 1); 1491 } 1492 1493 /* 1494 * Fserr prints the name of a filesystem with an error diagnostic. 1495 * 1496 * The form of the error message is: 1497 * fs: error message 1498 */ 1499 void 1500 ext2_fserr(struct m_ext2fs *fs, uid_t uid, char *cp) 1501 { 1502 1503 log(LOG_ERR, "uid %u on %s: %s\n", uid, fs->e2fs_fsmnt, cp); 1504 } 1505 1506 int 1507 ext2_cg_has_sb(struct m_ext2fs *fs, int cg) 1508 { 1509 int a3, a5, a7; 1510 1511 if (cg == 0) 1512 return (1); 1513 1514 if (EXT2_HAS_COMPAT_FEATURE(fs, EXT2F_COMPAT_SPARSESUPER2)) { 1515 if (cg == fs->e2fs->e4fs_backup_bgs[0] || 1516 cg == fs->e2fs->e4fs_backup_bgs[1]) 1517 return (1); 1518 return (0); 1519 } 1520 1521 if ((cg <= 1) || 1522 !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_SPARSESUPER)) 1523 return (1); 1524 1525 if (!(cg & 1)) 1526 return (0); 1527 1528 for (a3 = 3, a5 = 5, a7 = 7; 1529 a3 <= cg || a5 <= cg || a7 <= cg; 1530 a3 *= 3, a5 *= 5, a7 *= 7) 1531 if (cg == a3 || cg == a5 || cg == a7) 1532 return (1); 1533 return (0); 1534 } 1535