2 * modified for Lites 1.1
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
8 * SPDX-License-Identifier: BSD-3-Clause
10 * Copyright (c) 1982, 1986, 1989, 1993
11 * The Regents of the University of California. All rights reserved.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
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.
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
37 * @(#)ffs_alloc.c 8.8 (Berkeley) 2/21/94
41 #include <sys/param.h>
42 #include <sys/systm.h>
44 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/sysctl.h>
49 #include <sys/syslog.h>
51 #include <sys/endian.h>
53 #include <fs/ext2fs/fs.h>
54 #include <fs/ext2fs/inode.h>
55 #include <fs/ext2fs/ext2_mount.h>
56 #include <fs/ext2fs/ext2fs.h>
57 #include <fs/ext2fs/ext2_extern.h>
59 SDT_PROVIDER_DEFINE(ext2fs);
62 * arg0: verbosity. Higher numbers give more verbose messages
63 * arg1: Textual message
65 SDT_PROBE_DEFINE2(ext2fs, , alloc, trace, "int", "char*");
66 SDT_PROBE_DEFINE3(ext2fs, , alloc, ext2_reallocblks_realloc,
67 "ino_t", "e2fs_lbn_t", "e2fs_lbn_t");
68 SDT_PROBE_DEFINE1(ext2fs, , alloc, ext2_reallocblks_bap, "uint32_t");
69 SDT_PROBE_DEFINE1(ext2fs, , alloc, ext2_reallocblks_blkno, "e2fs_daddr_t");
70 SDT_PROBE_DEFINE2(ext2fs, , alloc, ext2_b_bitmap_validate_error, "char*", "int");
71 SDT_PROBE_DEFINE3(ext2fs, , alloc, ext2_nodealloccg_bmap_corrupted,
72 "int", "daddr_t", "char*");
73 SDT_PROBE_DEFINE2(ext2fs, , alloc, ext2_blkfree_bad_block, "ino_t", "e4fs_daddr_t");
74 SDT_PROBE_DEFINE2(ext2fs, , alloc, ext2_vfree_doublefree, "char*", "ino_t");
76 static daddr_t ext2_alloccg(struct inode *, int, daddr_t, int);
77 static daddr_t ext2_clusteralloc(struct inode *, int, daddr_t, int);
78 static u_long ext2_dirpref(struct inode *);
79 static e4fs_daddr_t ext2_hashalloc(struct inode *, int, long, int,
80 daddr_t (*)(struct inode *, int, daddr_t,
82 static daddr_t ext2_nodealloccg(struct inode *, int, daddr_t, int);
83 static daddr_t ext2_mapsearch(struct m_ext2fs *, char *, daddr_t);
86 * Allocate a block in the filesystem.
88 * A preference may be optionally specified. If a preference is given
89 * the following hierarchy is used to allocate a block:
90 * 1) allocate the requested block.
91 * 2) allocate a rotationally optimal block in the same cylinder.
92 * 3) allocate a block in the same cylinder group.
93 * 4) quadratically rehash into other cylinder groups, until an
94 * available block is located.
95 * If no block preference is given the following hierarchy is used
96 * to allocate a block:
97 * 1) allocate a block in the cylinder group that contains the
99 * 2) quadratically rehash into other cylinder groups, until an
100 * available block is located.
103 ext2_alloc(struct inode *ip, daddr_t lbn, e4fs_daddr_t bpref, int size,
104 struct ucred *cred, e4fs_daddr_t *bnp)
107 struct ext2mount *ump;
114 mtx_assert(EXT2_MTX(ump), MA_OWNED);
116 if ((u_int)size > fs->e2fs_bsize || blkoff(fs, size) != 0) {
117 vn_printf(ip->i_devvp, "bsize = %lu, size = %d, fs = %s\n",
118 (long unsigned int)fs->e2fs_bsize, size, fs->e2fs_fsmnt);
119 panic("ext2_alloc: bad size");
122 panic("ext2_alloc: missing credential");
123 #endif /* INVARIANTS */
124 if (size == fs->e2fs_bsize && fs->e2fs_fbcount == 0)
126 if (cred->cr_uid != 0 &&
127 fs->e2fs_fbcount < fs->e2fs_rbcount)
129 if (bpref >= fs->e2fs_bcount)
132 cg = ino_to_cg(fs, ip->i_number);
134 cg = dtog(fs, bpref);
135 bno = (daddr_t)ext2_hashalloc(ip, cg, bpref, fs->e2fs_bsize,
138 /* set next_alloc fields as done in block_getblk */
139 ip->i_next_alloc_block = lbn;
140 ip->i_next_alloc_goal = bno;
142 ip->i_blocks += btodb(fs->e2fs_bsize);
143 ip->i_flag |= IN_CHANGE | IN_UPDATE;
149 SDT_PROBE2(ext2fs, , alloc, trace, 1, "cannot allocate data block");
154 * Allocate EA's block for inode.
157 ext2_alloc_meta(struct inode *ip)
164 EXT2_LOCK(ip->i_ump);
165 blk = ext2_hashalloc(ip, ino_to_cg(fs, ip->i_number), 0, fs->e2fs_bsize,
168 EXT2_UNLOCK(ip->i_ump);
169 SDT_PROBE2(ext2fs, , alloc, trace, 1, "cannot allocate meta block");
176 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
178 * The vnode and an array of buffer pointers for a range of sequential
179 * logical blocks to be made contiguous is given. The allocator attempts
180 * to find a range of sequential blocks starting as close as possible to
181 * an fs_rotdelay offset from the end of the allocation for the logical
182 * block immediately preceding the current range. If successful, the
183 * physical block numbers in the buffer pointers and in the inode are
184 * changed to reflect the new allocation. If unsuccessful, the allocation
185 * is left unchanged. The success in doing the reallocation is returned.
186 * Note that the error return is not reflected back to the user. Rather
187 * the previous block allocation will be used.
190 static SYSCTL_NODE(_vfs, OID_AUTO, ext2fs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
191 "EXT2FS filesystem");
193 static int doasyncfree = 1;
195 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0,
196 "Use asynchronous writes to update block pointers when freeing blocks");
198 static int doreallocblks = 0;
200 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
203 ext2_reallocblks(struct vop_reallocblks_args *ap)
208 struct buf *sbp, *ebp;
209 uint32_t *bap, *sbap, *ebap;
210 struct ext2mount *ump;
211 struct cluster_save *buflist;
212 struct indir start_ap[EXT2_NIADDR + 1], end_ap[EXT2_NIADDR + 1], *idp;
213 e2fs_lbn_t start_lbn, end_lbn;
215 e2fs_daddr_t newblk, blkno;
216 int i, len, start_lvl, end_lvl, pref, ssize;
218 if (doreallocblks == 0)
226 if (fs->e2fs_contigsumsize <= 0 || ip->i_flag & IN_E4EXTENTS)
229 buflist = ap->a_buflist;
230 len = buflist->bs_nchildren;
231 start_lbn = buflist->bs_children[0]->b_lblkno;
232 end_lbn = start_lbn + len - 1;
234 for (i = 1; i < len; i++)
235 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
236 panic("ext2_reallocblks: non-cluster");
239 * If the cluster crosses the boundary for the first indirect
240 * block, leave space for the indirect block. Indirect blocks
241 * are initially laid out in a position after the last direct
242 * block. Block reallocation would usually destroy locality by
243 * moving the indirect block out of the way to make room for
244 * data blocks if we didn't compensate here. We should also do
245 * this for other indirect block boundaries, but it is only
246 * important for the first one.
248 if (start_lbn < EXT2_NDADDR && end_lbn >= EXT2_NDADDR)
251 * If the latest allocation is in a new cylinder group, assume that
252 * the filesystem has decided to move and do not force it back to
253 * the previous cylinder group.
255 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
256 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
258 if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
259 ext2_getlbns(vp, end_lbn, end_ap, &end_lvl))
262 * Get the starting offset and block map for the first block.
264 if (start_lvl == 0) {
268 idp = &start_ap[start_lvl - 1];
269 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) {
273 sbap = (u_int *)sbp->b_data;
277 * If the block range spans two block maps, get the second map.
280 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
284 if (start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
285 panic("ext2_reallocblks: start == end");
287 ssize = len - (idp->in_off + 1);
288 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp))
290 ebap = (u_int *)ebp->b_data;
293 * Find the preferred location for the cluster.
296 pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0);
298 * Search the block map looking for an allocation of the desired size.
300 if ((newblk = (e2fs_daddr_t)ext2_hashalloc(ip, dtog(fs, pref), pref,
301 len, ext2_clusteralloc)) == 0) {
306 * We have found a new contiguous block.
308 * First we have to replace the old block pointers with the new
309 * block pointers in the inode and indirect blocks associated
312 SDT_PROBE3(ext2fs, , alloc, ext2_reallocblks_realloc,
313 ip->i_number, start_lbn, end_lbn);
315 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
321 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap))
322 panic("ext2_reallocblks: alloc mismatch");
324 SDT_PROBE1(ext2fs, , alloc, ext2_reallocblks_bap, *bap);
328 * Next we must write out the modified inode and indirect blocks.
329 * For strict correctness, the writes should be synchronous since
330 * the old block values may have been written to disk. In practise
331 * they are almost never written, but if we are concerned about
332 * strict correctness, the `doasyncfree' flag should be set to zero.
334 * The test on `doasyncfree' should be changed to test a flag
335 * that shows whether the associated buffers and inodes have
336 * been written. The flag should be set when the cluster is
337 * started and cleared whenever the buffer or inode is flushed.
338 * We can then check below to see if it is set, and do the
339 * synchronous write only when it has been cleared.
341 if (sbap != &ip->i_db[0]) {
347 ip->i_flag |= IN_CHANGE | IN_UPDATE;
358 * Last, free the old blocks and assign the new blocks to the buffers.
360 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
361 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
363 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
364 SDT_PROBE1(ext2fs, , alloc, ext2_reallocblks_blkno, blkno);
372 if (sbap != &ip->i_db[0])
378 * Allocate an inode in the filesystem.
382 ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp)
386 struct ext2mount *ump;
400 if (fs->e2fs_ficount == 0)
403 * If it is a directory then obtain a cylinder group based on
404 * ext2_dirpref else obtain it using ino_to_cg. The preferred inode is
405 * always the next inode.
407 if ((mode & IFMT) == IFDIR) {
408 cg = ext2_dirpref(pip);
409 if (fs->e2fs_contigdirs[cg] < 255)
410 fs->e2fs_contigdirs[cg]++;
412 cg = ino_to_cg(fs, pip->i_number);
413 if (fs->e2fs_contigdirs[cg] > 0)
414 fs->e2fs_contigdirs[cg]--;
416 ipref = cg * fs->e2fs_ipg + 1;
417 ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg);
422 error = vfs_hash_get(ump->um_mountp, ino, LK_EXCLUSIVE, td, vpp, NULL, NULL);
423 if (error || *vpp != NULL) {
427 ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
429 /* Allocate a new vnode/inode. */
430 if ((error = getnewvnode("ext2fs", ump->um_mountp, &ext2_vnodeops, &vp)) != 0) {
431 free(ip, M_EXT2NODE);
435 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
438 ip->i_e2fs = fs = ump->um_e2fs;
441 ip->i_block_group = ino_to_cg(fs, ino);
442 ip->i_next_alloc_block = 0;
443 ip->i_next_alloc_goal = 0;
445 error = insmntque(vp, ump->um_mountp);
447 free(ip, M_EXT2NODE);
451 error = vfs_hash_insert(vp, ino, LK_EXCLUSIVE, td, vpp, NULL, NULL);
452 if (error || *vpp != NULL) {
454 free(ip, M_EXT2NODE);
458 if ((error = ext2_vinit(ump->um_mountp, &ext2_fifoops, &vp)) != 0) {
461 free(ip, M_EXT2NODE);
465 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_EXTENTS)
466 && (S_ISREG(mode) || S_ISDIR(mode)))
467 ext4_ext_tree_init(ip);
469 memset(ip->i_data, 0, sizeof(ip->i_data));
472 * Set up a new generation number for this inode.
476 ip->i_gen = arc4random();
477 } while (ip->i_gen == 0);
480 ip->i_birthtime = ts.tv_sec;
481 ip->i_birthnsec = ts.tv_nsec;
489 SDT_PROBE2(ext2fs, , alloc, trace, 1, "out of inodes");
494 * 64-bit compatible getters and setters for struct ext2_gd from ext2fs.h
497 e2fs_gd_get_b_bitmap(struct ext2_gd *gd)
500 return (((uint64_t)(le32toh(gd->ext4bgd_b_bitmap_hi)) << 32) |
501 le32toh(gd->ext2bgd_b_bitmap));
505 e2fs_gd_get_i_bitmap(struct ext2_gd *gd)
508 return (((uint64_t)(le32toh(gd->ext4bgd_i_bitmap_hi)) << 32) |
509 le32toh(gd->ext2bgd_i_bitmap));
513 e2fs_gd_get_i_tables(struct ext2_gd *gd)
516 return (((uint64_t)(le32toh(gd->ext4bgd_i_tables_hi)) << 32) |
517 le32toh(gd->ext2bgd_i_tables));
521 e2fs_gd_get_nbfree(struct ext2_gd *gd)
524 return (((uint32_t)(le16toh(gd->ext4bgd_nbfree_hi)) << 16) |
525 le16toh(gd->ext2bgd_nbfree));
529 e2fs_gd_set_nbfree(struct ext2_gd *gd, uint32_t val)
532 gd->ext2bgd_nbfree = htole16(val & 0xffff);
533 gd->ext4bgd_nbfree_hi = htole16(val >> 16);
537 e2fs_gd_get_nifree(struct ext2_gd *gd)
540 return (((uint32_t)(le16toh(gd->ext4bgd_nifree_hi)) << 16) |
541 le16toh(gd->ext2bgd_nifree));
545 e2fs_gd_set_nifree(struct ext2_gd *gd, uint32_t val)
548 gd->ext2bgd_nifree = htole16(val & 0xffff);
549 gd->ext4bgd_nifree_hi = htole16(val >> 16);
553 e2fs_gd_get_ndirs(struct ext2_gd *gd)
556 return (((uint32_t)(le16toh(gd->ext4bgd_ndirs_hi)) << 16) |
557 le16toh(gd->ext2bgd_ndirs));
561 e2fs_gd_set_ndirs(struct ext2_gd *gd, uint32_t val)
564 gd->ext2bgd_ndirs = htole16(val & 0xffff);
565 gd->ext4bgd_ndirs_hi = htole16(val >> 16);
569 e2fs_gd_get_i_unused(struct ext2_gd *gd)
571 return ((uint32_t)(le16toh(gd->ext4bgd_i_unused_hi) << 16) |
572 le16toh(gd->ext4bgd_i_unused));
576 e2fs_gd_set_i_unused(struct ext2_gd *gd, uint32_t val)
579 gd->ext4bgd_i_unused = htole16(val & 0xffff);
580 gd->ext4bgd_i_unused_hi = htole16(val >> 16);
584 * Find a cylinder to place a directory.
586 * The policy implemented by this algorithm is to allocate a
587 * directory inode in the same cylinder group as its parent
588 * directory, but also to reserve space for its files inodes
589 * and data. Restrict the number of directories which may be
590 * allocated one after another in the same cylinder group
591 * without intervening allocation of files.
593 * If we allocate a first level directory then force allocation
594 * in another cylinder group.
598 ext2_dirpref(struct inode *pip)
601 int cg, prefcg, cgsize;
602 uint64_t avgbfree, minbfree;
603 u_int avgifree, avgndir, curdirsize;
604 u_int minifree, maxndir;
605 u_int mincg, minndir;
606 u_int dirsize, maxcontigdirs;
608 mtx_assert(EXT2_MTX(pip->i_ump), MA_OWNED);
611 avgifree = fs->e2fs_ficount / fs->e2fs_gcount;
612 avgbfree = fs->e2fs_fbcount / fs->e2fs_gcount;
613 avgndir = fs->e2fs_total_dir / fs->e2fs_gcount;
616 * Force allocation in another cg if creating a first level dir.
618 ASSERT_VOP_LOCKED(ITOV(pip), "ext2fs_dirpref");
619 if (ITOV(pip)->v_vflag & VV_ROOT) {
620 prefcg = arc4random() % fs->e2fs_gcount;
622 minndir = fs->e2fs_ipg;
623 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
624 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < minndir &&
625 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree &&
626 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= avgbfree) {
628 minndir = e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]);
630 for (cg = 0; cg < prefcg; cg++)
631 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < minndir &&
632 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree &&
633 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= avgbfree) {
635 minndir = e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]);
640 * Count various limits which used for
641 * optimal allocation of a directory inode.
643 maxndir = min(avgndir + fs->e2fs_ipg / 16, fs->e2fs_ipg);
644 minifree = avgifree - avgifree / 4;
647 minbfree = avgbfree - avgbfree / 4;
650 cgsize = fs->e2fs_fsize * fs->e2fs_fpg;
651 dirsize = AVGDIRSIZE;
652 curdirsize = avgndir ?
653 (cgsize - avgbfree * fs->e2fs_bsize) / avgndir : 0;
654 if (dirsize < curdirsize)
655 dirsize = curdirsize;
656 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255);
657 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR);
658 if (maxcontigdirs == 0)
662 * Limit number of dirs in one cg and reserve space for
663 * regular files, but only if we have no deficit in
666 prefcg = ino_to_cg(fs, pip->i_number);
667 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
668 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < maxndir &&
669 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= minifree &&
670 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= minbfree) {
671 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
674 for (cg = 0; cg < prefcg; cg++)
675 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < maxndir &&
676 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= minifree &&
677 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= minbfree) {
678 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
682 * This is a backstop when we have deficit in space.
684 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
685 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree)
687 for (cg = 0; cg < prefcg; cg++)
688 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree)
694 * Select the desired position for the next block in a file.
696 * we try to mimic what Remy does in inode_getblk/block_getblk
698 * we note: blocknr == 0 means that we're about to allocate either
699 * a direct block or a pointer block at the first level of indirection
700 * (In other words, stuff that will go in i_db[] or i_ib[])
702 * blocknr != 0 means that we're allocating a block that is none
703 * of the above. Then, blocknr tells us the number of the block
704 * that will hold the pointer
707 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, e2fs_daddr_t *bap,
708 e2fs_daddr_t blocknr)
715 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
718 * If the next block is actually what we thought it is, then set the
719 * goal to what we thought it should be.
721 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0)
722 return ip->i_next_alloc_goal;
725 * Now check whether we were provided with an array that basically
726 * tells us previous blocks to which we want to stay close.
729 for (tmp = indx - 1; tmp >= 0; tmp--)
731 return (le32toh(bap[tmp]));
734 * Else lets fall back to the blocknr or, if there is none, follow
735 * the rule that a block should be allocated near its inode.
737 return (blocknr ? blocknr :
738 (e2fs_daddr_t)(ip->i_block_group *
739 EXT2_BLOCKS_PER_GROUP(fs)) + le32toh(fs->e2fs->e2fs_first_dblock));
743 * Implement the cylinder overflow algorithm.
745 * The policy implemented by this algorithm is:
746 * 1) allocate the block in its requested cylinder group.
747 * 2) quadratically rehash on the cylinder group number.
748 * 3) brute force search for a free block.
751 ext2_hashalloc(struct inode *ip, int cg, long pref, int size,
752 daddr_t (*allocator) (struct inode *, int, daddr_t, int))
758 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
761 * 1: preferred cylinder group
763 result = (*allocator)(ip, cg, pref, size);
767 * 2: quadratic rehash
769 for (i = 1; i < fs->e2fs_gcount; i *= 2) {
771 if (cg >= fs->e2fs_gcount)
772 cg -= fs->e2fs_gcount;
773 result = (*allocator)(ip, cg, 0, size);
778 * 3: brute force search
779 * Note that we start at i == 2, since 0 was checked initially,
780 * and 1 is always checked in the quadratic rehash.
782 cg = (icg + 2) % fs->e2fs_gcount;
783 for (i = 2; i < fs->e2fs_gcount; i++) {
784 result = (*allocator)(ip, cg, 0, size);
788 if (cg == fs->e2fs_gcount)
795 ext2_cg_number_gdb_nometa(struct m_ext2fs *fs, int cg)
798 if (!ext2_cg_has_sb(fs, cg))
801 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG))
802 return (le32toh(fs->e2fs->e3fs_first_meta_bg));
804 return ((fs->e2fs_gcount + EXT2_DESCS_PER_BLOCK(fs) - 1) /
805 EXT2_DESCS_PER_BLOCK(fs));
809 ext2_cg_number_gdb_meta(struct m_ext2fs *fs, int cg)
811 unsigned long metagroup;
814 metagroup = cg / EXT2_DESCS_PER_BLOCK(fs);
815 first = metagroup * EXT2_DESCS_PER_BLOCK(fs);
816 last = first + EXT2_DESCS_PER_BLOCK(fs) - 1;
818 if (cg == first || cg == first + 1 || cg == last)
825 ext2_cg_number_gdb(struct m_ext2fs *fs, int cg)
827 unsigned long first_meta_bg, metagroup;
829 first_meta_bg = le32toh(fs->e2fs->e3fs_first_meta_bg);
830 metagroup = cg / EXT2_DESCS_PER_BLOCK(fs);
832 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
833 metagroup < first_meta_bg)
834 return (ext2_cg_number_gdb_nometa(fs, cg));
836 return ext2_cg_number_gdb_meta(fs, cg);
840 ext2_number_base_meta_blocks(struct m_ext2fs *fs, int cg)
844 number = ext2_cg_has_sb(fs, cg);
846 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
847 cg < le32toh(fs->e2fs->e3fs_first_meta_bg) *
848 EXT2_DESCS_PER_BLOCK(fs)) {
850 number += ext2_cg_number_gdb(fs, cg);
851 number += le16toh(fs->e2fs->e2fs_reserved_ngdb);
854 number += ext2_cg_number_gdb(fs, cg);
861 ext2_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
865 if (start_bit >= end_bit)
868 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
871 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
875 ext2_get_group_number(struct m_ext2fs *fs, e4fs_daddr_t block)
878 return ((block - le32toh(fs->e2fs->e2fs_first_dblock)) /
883 ext2_block_in_group(struct m_ext2fs *fs, e4fs_daddr_t block, int cg)
886 return ((ext2_get_group_number(fs, block) == cg) ? 1 : 0);
890 ext2_cg_block_bitmap_init(struct m_ext2fs *fs, int cg, struct buf *bp)
892 int bit, bit_max, inodes_per_block;
895 if (!(le16toh(fs->e2fs_gd[cg].ext4bgd_flags) & EXT2_BG_BLOCK_UNINIT))
898 memset(bp->b_data, 0, fs->e2fs_bsize);
900 bit_max = ext2_number_base_meta_blocks(fs, cg);
901 if ((bit_max >> 3) >= fs->e2fs_bsize)
904 for (bit = 0; bit < bit_max; bit++)
905 setbit(bp->b_data, bit);
907 start = (uint64_t)cg * fs->e2fs_bpg +
908 le32toh(fs->e2fs->e2fs_first_dblock);
910 /* Set bits for block and inode bitmaps, and inode table. */
911 tmp = e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg]);
912 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
913 ext2_block_in_group(fs, tmp, cg))
914 setbit(bp->b_data, tmp - start);
916 tmp = e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg]);
917 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
918 ext2_block_in_group(fs, tmp, cg))
919 setbit(bp->b_data, tmp - start);
921 tmp = e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]);
922 inodes_per_block = fs->e2fs_bsize/EXT2_INODE_SIZE(fs);
923 while( tmp < e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]) +
924 fs->e2fs_ipg / inodes_per_block ) {
925 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
926 ext2_block_in_group(fs, tmp, cg))
927 setbit(bp->b_data, tmp - start);
932 * Also if the number of blocks within the group is less than
933 * the blocksize * 8 ( which is the size of bitmap ), set rest
934 * of the block bitmap to 1
936 ext2_mark_bitmap_end(fs->e2fs_bpg, fs->e2fs_bsize * 8,
940 fs->e2fs_gd[cg].ext4bgd_flags = htole16(le16toh(
941 fs->e2fs_gd[cg].ext4bgd_flags) & ~EXT2_BG_BLOCK_UNINIT);
947 ext2_b_bitmap_validate(struct m_ext2fs *fs, struct buf *bp, int cg)
950 uint64_t group_first_block;
951 unsigned int offset, max_bit;
953 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG)) {
955 * It is not possible to check block bitmap in case of this
956 * feature, because the inode and block bitmaps and inode table
957 * blocks may not be in the group at all.
958 * So, skip check in this case.
963 gd = &fs->e2fs_gd[cg];
964 max_bit = fs->e2fs_fpg;
965 group_first_block = ((uint64_t)cg) * fs->e2fs_fpg +
966 le32toh(fs->e2fs->e2fs_first_dblock);
968 /* Check block bitmap block number */
969 offset = e2fs_gd_get_b_bitmap(gd) - group_first_block;
970 if (offset >= max_bit || !isset(bp->b_data, offset)) {
971 SDT_PROBE2(ext2fs, , alloc, ext2_b_bitmap_validate_error,
972 "bad block bitmap, group", cg);
976 /* Check inode bitmap block number */
977 offset = e2fs_gd_get_i_bitmap(gd) - group_first_block;
978 if (offset >= max_bit || !isset(bp->b_data, offset)) {
979 SDT_PROBE2(ext2fs, , alloc, ext2_b_bitmap_validate_error,
980 "bad inode bitmap", cg);
984 /* Check inode table */
985 offset = e2fs_gd_get_i_tables(gd) - group_first_block;
986 if (offset >= max_bit || offset + fs->e2fs_itpg >= max_bit) {
987 SDT_PROBE2(ext2fs, , alloc, ext2_b_bitmap_validate_error,
988 "bad inode table, group", cg);
996 * Determine whether a block can be allocated.
998 * Check to see if a block of the appropriate size is available,
999 * and if it is, allocate it.
1002 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
1004 struct m_ext2fs *fs;
1006 struct ext2mount *ump;
1007 daddr_t bno, runstart, runlen;
1008 int bit, loc, end, error, start;
1013 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0)
1017 error = bread(ip->i_devvp, fsbtodb(fs,
1018 e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1019 (int)fs->e2fs_bsize, NOCRED, &bp);
1023 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1024 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1025 error = ext2_cg_block_bitmap_init(fs, cg, bp);
1029 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1031 error = ext2_gd_b_bitmap_csum_verify(fs, cg, bp);
1035 error = ext2_b_bitmap_validate(fs,bp, cg);
1040 * Check, that another thread did not not allocate the last block in
1041 * this group while we were waiting for the buffer.
1043 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0)
1046 bbp = (char *)bp->b_data;
1048 if (dtog(fs, bpref) != cg)
1051 bpref = dtogd(fs, bpref);
1053 * if the requested block is available, use it
1055 if (isclr(bbp, bpref)) {
1061 * no blocks in the requested cylinder, so take next
1062 * available one in this cylinder group.
1063 * first try to get 8 contigous blocks, then fall back to a single
1067 start = dtogd(fs, bpref) / NBBY;
1070 end = howmany(fs->e2fs_fpg, NBBY) - start;
1074 for (loc = start; loc < end; loc++) {
1075 if (bbp[loc] == (char)0xff) {
1080 /* Start of a run, find the number of high clear bits. */
1082 bit = fls(bbp[loc]);
1083 runlen = NBBY - bit;
1084 runstart = loc * NBBY + bit;
1085 } else if (bbp[loc] == 0) {
1086 /* Continue a run. */
1090 * Finish the current run. If it isn't long
1091 * enough, start a new one.
1093 bit = ffs(bbp[loc]) - 1;
1100 /* Run was too short, start a new one. */
1101 bit = fls(bbp[loc]);
1102 runlen = NBBY - bit;
1103 runstart = loc * NBBY + bit;
1106 /* If the current run is long enough, use it. */
1117 bno = ext2_mapsearch(fs, bbp, bpref);
1123 if (isset(bbp, bno)) {
1124 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n",
1125 cg, (intmax_t)bno, fs->e2fs_fsmnt);
1126 panic("ext2fs_alloccg: dup alloc");
1131 ext2_clusteracct(fs, bbp, cg, bno, -1);
1133 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1134 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1);
1137 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1139 return (((uint64_t)cg) * fs->e2fs_fpg +
1140 le32toh(fs->e2fs->e2fs_first_dblock) + bno);
1149 * Determine whether a cluster can be allocated.
1152 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len)
1154 struct m_ext2fs *fs;
1155 struct ext2mount *ump;
1158 int bit, error, got, i, loc, run;
1165 if (fs->e2fs_maxcluster[cg] < len)
1169 error = bread(ip->i_devvp,
1170 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1171 (int)fs->e2fs_bsize, NOCRED, &bp);
1175 bbp = (char *)bp->b_data;
1178 * Check to see if a cluster of the needed size (or bigger) is
1179 * available in this cylinder group.
1181 lp = &fs->e2fs_clustersum[cg].cs_sum[len];
1182 for (i = len; i <= fs->e2fs_contigsumsize; i++)
1185 if (i > fs->e2fs_contigsumsize) {
1187 * Update the cluster summary information to reflect
1188 * the true maximum-sized cluster so that future cluster
1189 * allocation requests can avoid reading the bitmap only
1190 * to find no cluster.
1192 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1];
1193 for (i = len - 1; i > 0; i--)
1196 fs->e2fs_maxcluster[cg] = i;
1201 /* Search the bitmap to find a big enough cluster like in FFS. */
1202 if (dtog(fs, bpref) != cg)
1205 bpref = dtogd(fs, bpref);
1207 bit = 1 << (bpref % NBBY);
1208 for (run = 0, got = bpref; got < fs->e2fs_fpg; got++) {
1209 if ((bbp[loc] & bit) != 0)
1216 if ((got & (NBBY - 1)) != (NBBY - 1))
1224 if (got >= fs->e2fs_fpg)
1227 /* Allocate the cluster that we found. */
1228 for (i = 1; i < len; i++)
1229 if (!isclr(bbp, got - run + i))
1230 panic("ext2_clusteralloc: map mismatch");
1232 bno = got - run + 1;
1233 if (bno >= fs->e2fs_fpg)
1234 panic("ext2_clusteralloc: allocated out of group");
1237 for (i = 0; i < len; i += fs->e2fs_fpb) {
1238 setbit(bbp, bno + i);
1239 ext2_clusteracct(fs, bbp, cg, bno + i, -1);
1241 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1242 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1);
1248 return (cg * fs->e2fs_fpg + le32toh(fs->e2fs->e2fs_first_dblock)
1259 ext2_zero_inode_table(struct inode *ip, int cg)
1261 struct m_ext2fs *fs;
1263 int i, all_blks, used_blks;
1267 if (le16toh(fs->e2fs_gd[cg].ext4bgd_flags) & EXT2_BG_INODE_ZEROED)
1270 all_blks = le16toh(fs->e2fs->e2fs_inode_size) * fs->e2fs_ipg /
1273 used_blks = howmany(fs->e2fs_ipg -
1274 e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]),
1275 fs->e2fs_bsize / EXT2_INODE_SIZE(fs));
1277 for (i = 0; i < all_blks - used_blks; i++) {
1278 bp = getblk(ip->i_devvp, fsbtodb(fs,
1279 e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]) + used_blks + i),
1280 fs->e2fs_bsize, 0, 0, 0);
1284 vfs_bio_bzero_buf(bp, 0, fs->e2fs_bsize);
1288 fs->e2fs_gd[cg].ext4bgd_flags = htole16(le16toh(
1289 fs->e2fs_gd[cg].ext4bgd_flags) | EXT2_BG_INODE_ZEROED);
1295 ext2_fix_bitmap_tail(unsigned char *bitmap, int first, int last)
1299 for (i = first; i <= last; i++)
1304 * Determine whether an inode can be allocated.
1306 * Check to see if an inode is available, and if it is,
1307 * allocate it using tode in the specified cylinder group.
1310 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode)
1312 struct m_ext2fs *fs;
1314 struct ext2mount *ump;
1315 int error, start, len, ifree, ibytes;
1318 ipref--; /* to avoid a lot of (ipref -1) */
1323 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0)
1326 error = bread(ip->i_devvp, fsbtodb(fs,
1327 e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])),
1328 (int)fs->e2fs_bsize, NOCRED, &bp);
1333 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1334 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1335 if (le16toh(fs->e2fs_gd[cg].ext4bgd_flags) &
1336 EXT2_BG_INODE_UNINIT) {
1337 ibytes = fs->e2fs_ipg / 8;
1338 memset(bp->b_data, 0, ibytes - 1);
1339 ext2_fix_bitmap_tail(bp->b_data, ibytes,
1340 fs->e2fs_bsize - 1);
1341 fs->e2fs_gd[cg].ext4bgd_flags = htole16(le16toh(
1342 fs->e2fs_gd[cg].ext4bgd_flags) &
1343 ~EXT2_BG_INODE_UNINIT);
1345 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1346 error = ext2_zero_inode_table(ip, cg);
1353 error = ext2_gd_i_bitmap_csum_verify(fs, cg, bp);
1359 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0) {
1361 * Another thread allocated the last i-node in this
1362 * group while we were waiting for the buffer.
1368 ibp = (char *)bp->b_data;
1370 ipref %= fs->e2fs_ipg;
1371 if (isclr(ibp, ipref))
1374 start = ipref / NBBY;
1375 len = howmany(fs->e2fs_ipg - ipref, NBBY);
1376 loc = memcchr(&ibp[start], 0xff, len);
1380 loc = memcchr(&ibp[start], 0xff, len);
1382 SDT_PROBE3(ext2fs, , alloc,
1383 ext2_nodealloccg_bmap_corrupted, cg, ipref,
1390 ipref = (loc - ibp) * NBBY + ffs(~*loc) - 1;
1394 e2fs_gd_set_nifree(&fs->e2fs_gd[cg],
1395 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) - 1);
1396 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1397 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1398 ifree = fs->e2fs_ipg - e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]);
1399 if (ipref + 1 > ifree)
1400 e2fs_gd_set_i_unused(&fs->e2fs_gd[cg],
1401 fs->e2fs_ipg - (ipref + 1));
1405 if ((mode & IFMT) == IFDIR) {
1406 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg],
1407 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) + 1);
1408 fs->e2fs_total_dir++;
1411 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1413 return ((uint64_t)cg * fs->e2fs_ipg + ipref + 1);
1417 * Free a block or fragment.
1421 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size)
1423 struct m_ext2fs *fs;
1425 struct ext2mount *ump;
1432 if (bno >= fs->e2fs_bcount) {
1433 SDT_PROBE2(ext2fs, , alloc, ext2_blkfree_bad_block,
1437 error = bread(ip->i_devvp,
1438 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1439 (int)fs->e2fs_bsize, NOCRED, &bp);
1443 bbp = (char *)bp->b_data;
1444 bno = dtogd(fs, bno);
1445 if (isclr(bbp, bno)) {
1446 panic("ext2_blkfree: freeing free block %lld, fs=%s",
1447 (long long)bno, fs->e2fs_fsmnt);
1451 ext2_clusteracct(fs, bbp, cg, bno, 1);
1453 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1454 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) + 1);
1457 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1466 ext2_vfree(struct vnode *pvp, ino_t ino, int mode)
1468 struct m_ext2fs *fs;
1471 struct ext2mount *ump;
1478 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount)
1479 panic("ext2_vfree: range: devvp = %p, ino = %ju, fs = %s",
1480 pip->i_devvp, (uintmax_t)ino, fs->e2fs_fsmnt);
1482 cg = ino_to_cg(fs, ino);
1483 error = bread(pip->i_devvp,
1484 fsbtodb(fs, e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])),
1485 (int)fs->e2fs_bsize, NOCRED, &bp);
1489 ibp = (char *)bp->b_data;
1490 ino = (ino - 1) % fs->e2fs_ipg;
1491 if (isclr(ibp, ino)) {
1492 SDT_PROBE2(ext2fs, , alloc, ext2_vfree_doublefree,
1493 fs->e2fs_fsmnt, ino);
1494 if (fs->e2fs_ronly == 0)
1495 panic("ext2_vfree: freeing free inode");
1500 e2fs_gd_set_nifree(&fs->e2fs_gd[cg],
1501 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) + 1);
1502 if ((mode & IFMT) == IFDIR) {
1503 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg],
1504 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) - 1);
1505 fs->e2fs_total_dir--;
1509 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1515 * Find a block in the specified cylinder group.
1517 * It is a panic if a request is made to find a block if none are
1521 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref)
1527 * find the fragment by searching through the free block
1528 * map for an appropriate bit pattern
1531 start = dtogd(fs, bpref) / NBBY;
1534 len = howmany(fs->e2fs_fpg, NBBY) - start;
1535 loc = memcchr(&bbp[start], 0xff, len);
1539 loc = memcchr(&bbp[start], 0xff, len);
1541 panic("ext2_mapsearch: map corrupted: start=%d, len=%d,"
1542 "fs=%s", start, len, fs->e2fs_fsmnt);
1546 return ((loc - bbp) * NBBY + ffs(~*loc) - 1);
1550 ext2_cg_has_sb(struct m_ext2fs *fs, int cg)
1557 if (EXT2_HAS_COMPAT_FEATURE(fs, EXT2F_COMPAT_SPARSESUPER2)) {
1558 if (cg == le32toh(fs->e2fs->e4fs_backup_bgs[0]) ||
1559 cg == le32toh(fs->e2fs->e4fs_backup_bgs[1]))
1565 !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_SPARSESUPER))
1571 for (a3 = 3, a5 = 5, a7 = 7;
1572 a3 <= cg || a5 <= cg || a7 <= cg;
1573 a3 *= 3, a5 *= 5, a7 *= 7)
1574 if (cg == a3 || cg == a5 || cg == a7)