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) quadradically 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) quadradically 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, 0, "EXT2FS filesystem");
192 static int doasyncfree = 1;
194 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0,
195 "Use asychronous writes to update block pointers when freeing blocks");
197 static int doreallocblks = 0;
199 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
202 ext2_reallocblks(struct vop_reallocblks_args *ap)
207 struct buf *sbp, *ebp;
208 uint32_t *bap, *sbap, *ebap;
209 struct ext2mount *ump;
210 struct cluster_save *buflist;
211 struct indir start_ap[EXT2_NIADDR + 1], end_ap[EXT2_NIADDR + 1], *idp;
212 e2fs_lbn_t start_lbn, end_lbn;
214 e2fs_daddr_t newblk, blkno;
215 int i, len, start_lvl, end_lvl, pref, ssize;
217 if (doreallocblks == 0)
225 if (fs->e2fs_contigsumsize <= 0 || ip->i_flag & IN_E4EXTENTS)
228 buflist = ap->a_buflist;
229 len = buflist->bs_nchildren;
230 start_lbn = buflist->bs_children[0]->b_lblkno;
231 end_lbn = start_lbn + len - 1;
233 for (i = 1; i < len; i++)
234 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
235 panic("ext2_reallocblks: non-cluster");
238 * If the cluster crosses the boundary for the first indirect
239 * block, leave space for the indirect block. Indirect blocks
240 * are initially laid out in a position after the last direct
241 * block. Block reallocation would usually destroy locality by
242 * moving the indirect block out of the way to make room for
243 * data blocks if we didn't compensate here. We should also do
244 * this for other indirect block boundaries, but it is only
245 * important for the first one.
247 if (start_lbn < EXT2_NDADDR && end_lbn >= EXT2_NDADDR)
250 * If the latest allocation is in a new cylinder group, assume that
251 * the filesystem has decided to move and do not force it back to
252 * the previous cylinder group.
254 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
255 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
257 if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
258 ext2_getlbns(vp, end_lbn, end_ap, &end_lvl))
261 * Get the starting offset and block map for the first block.
263 if (start_lvl == 0) {
267 idp = &start_ap[start_lvl - 1];
268 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) {
272 sbap = (u_int *)sbp->b_data;
276 * If the block range spans two block maps, get the second map.
279 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
283 if (start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
284 panic("ext2_reallocblks: start == end");
286 ssize = len - (idp->in_off + 1);
287 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp))
289 ebap = (u_int *)ebp->b_data;
292 * Find the preferred location for the cluster.
295 pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0);
297 * Search the block map looking for an allocation of the desired size.
299 if ((newblk = (e2fs_daddr_t)ext2_hashalloc(ip, dtog(fs, pref), pref,
300 len, ext2_clusteralloc)) == 0) {
305 * We have found a new contiguous block.
307 * First we have to replace the old block pointers with the new
308 * block pointers in the inode and indirect blocks associated
311 SDT_PROBE3(ext2fs, , alloc, ext2_reallocblks_realloc,
312 ip->i_number, start_lbn, end_lbn);
314 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
320 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap))
321 panic("ext2_reallocblks: alloc mismatch");
323 SDT_PROBE1(ext2fs, , alloc, ext2_reallocblks_bap, *bap);
327 * Next we must write out the modified inode and indirect blocks.
328 * For strict correctness, the writes should be synchronous since
329 * the old block values may have been written to disk. In practise
330 * they are almost never written, but if we are concerned about
331 * strict correctness, the `doasyncfree' flag should be set to zero.
333 * The test on `doasyncfree' should be changed to test a flag
334 * that shows whether the associated buffers and inodes have
335 * been written. The flag should be set when the cluster is
336 * started and cleared whenever the buffer or inode is flushed.
337 * We can then check below to see if it is set, and do the
338 * synchronous write only when it has been cleared.
340 if (sbap != &ip->i_db[0]) {
346 ip->i_flag |= IN_CHANGE | IN_UPDATE;
357 * Last, free the old blocks and assign the new blocks to the buffers.
359 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
360 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
362 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
363 SDT_PROBE1(ext2fs, , alloc, ext2_reallocblks_blkno, blkno);
371 if (sbap != &ip->i_db[0])
377 * Allocate an inode in the filesystem.
381 ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp)
385 struct ext2mount *ump;
399 if (fs->e2fs->e2fs_ficount == 0)
402 * If it is a directory then obtain a cylinder group based on
403 * ext2_dirpref else obtain it using ino_to_cg. The preferred inode is
404 * always the next inode.
406 if ((mode & IFMT) == IFDIR) {
407 cg = ext2_dirpref(pip);
408 if (fs->e2fs_contigdirs[cg] < 255)
409 fs->e2fs_contigdirs[cg]++;
411 cg = ino_to_cg(fs, pip->i_number);
412 if (fs->e2fs_contigdirs[cg] > 0)
413 fs->e2fs_contigdirs[cg]--;
415 ipref = cg * fs->e2fs->e2fs_ipg + 1;
416 ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg);
421 error = vfs_hash_get(ump->um_mountp, ino, LK_EXCLUSIVE, td, vpp, NULL, NULL);
422 if (error || *vpp != NULL) {
426 ip = malloc(sizeof(struct inode), M_EXT2NODE, M_WAITOK | M_ZERO);
431 /* Allocate a new vnode/inode. */
432 if ((error = getnewvnode("ext2fs", ump->um_mountp, &ext2_vnodeops, &vp)) != 0) {
433 free(ip, M_EXT2NODE);
437 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
440 ip->i_e2fs = fs = ump->um_e2fs;
443 ip->i_block_group = ino_to_cg(fs, ino);
444 ip->i_next_alloc_block = 0;
445 ip->i_next_alloc_goal = 0;
447 error = insmntque(vp, ump->um_mountp);
449 free(ip, M_EXT2NODE);
453 error = vfs_hash_insert(vp, ino, LK_EXCLUSIVE, td, vpp, NULL, NULL);
454 if (error || *vpp != NULL) {
456 free(ip, M_EXT2NODE);
460 if ((error = ext2_vinit(ump->um_mountp, &ext2_fifoops, &vp)) != 0) {
463 free(ip, M_EXT2NODE);
467 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_EXTENTS)
468 && (S_ISREG(mode) || S_ISDIR(mode)))
469 ext4_ext_tree_init(ip);
471 memset(ip->i_data, 0, sizeof(ip->i_data));
475 * Set up a new generation number for this inode.
479 ip->i_gen = arc4random();
480 } while (ip->i_gen == 0);
483 ip->i_birthtime = ts.tv_sec;
484 ip->i_birthnsec = ts.tv_nsec;
492 SDT_PROBE2(ext2fs, , alloc, trace, 1, "out of inodes");
497 * 64-bit compatible getters and setters for struct ext2_gd from ext2fs.h
500 e2fs_gd_get_b_bitmap(struct ext2_gd *gd)
503 return (((uint64_t)(gd->ext4bgd_b_bitmap_hi) << 32) |
504 gd->ext2bgd_b_bitmap);
508 e2fs_gd_get_i_bitmap(struct ext2_gd *gd)
511 return (((uint64_t)(gd->ext4bgd_i_bitmap_hi) << 32) |
512 gd->ext2bgd_i_bitmap);
516 e2fs_gd_get_i_tables(struct ext2_gd *gd)
519 return (((uint64_t)(gd->ext4bgd_i_tables_hi) << 32) |
520 gd->ext2bgd_i_tables);
524 e2fs_gd_get_nbfree(struct ext2_gd *gd)
527 return (((uint32_t)(gd->ext4bgd_nbfree_hi) << 16) |
532 e2fs_gd_set_nbfree(struct ext2_gd *gd, uint32_t val)
535 gd->ext2bgd_nbfree = val & 0xffff;
536 gd->ext4bgd_nbfree_hi = val >> 16;
540 e2fs_gd_get_nifree(struct ext2_gd *gd)
543 return (((uint32_t)(gd->ext4bgd_nifree_hi) << 16) |
548 e2fs_gd_set_nifree(struct ext2_gd *gd, uint32_t val)
551 gd->ext2bgd_nifree = val & 0xffff;
552 gd->ext4bgd_nifree_hi = val >> 16;
556 e2fs_gd_get_ndirs(struct ext2_gd *gd)
559 return (((uint32_t)(gd->ext4bgd_ndirs_hi) << 16) |
564 e2fs_gd_set_ndirs(struct ext2_gd *gd, uint32_t val)
567 gd->ext2bgd_ndirs = val & 0xffff;
568 gd->ext4bgd_ndirs_hi = val >> 16;
572 e2fs_gd_get_i_unused(struct ext2_gd *gd)
574 return (((uint32_t)(gd->ext4bgd_i_unused_hi) << 16) |
575 gd->ext4bgd_i_unused);
579 e2fs_gd_set_i_unused(struct ext2_gd *gd, uint32_t val)
582 gd->ext4bgd_i_unused = val & 0xffff;
583 gd->ext4bgd_i_unused_hi = val >> 16;
587 * Find a cylinder to place a directory.
589 * The policy implemented by this algorithm is to allocate a
590 * directory inode in the same cylinder group as its parent
591 * directory, but also to reserve space for its files inodes
592 * and data. Restrict the number of directories which may be
593 * allocated one after another in the same cylinder group
594 * without intervening allocation of files.
596 * If we allocate a first level directory then force allocation
597 * in another cylinder group.
601 ext2_dirpref(struct inode *pip)
604 int cg, prefcg, cgsize;
605 uint64_t avgbfree, minbfree;
606 u_int avgifree, avgndir, curdirsize;
607 u_int minifree, maxndir;
608 u_int mincg, minndir;
609 u_int dirsize, maxcontigdirs;
611 mtx_assert(EXT2_MTX(pip->i_ump), MA_OWNED);
614 avgifree = fs->e2fs->e2fs_ficount / fs->e2fs_gcount;
615 avgbfree = fs->e2fs_fbcount / fs->e2fs_gcount;
616 avgndir = fs->e2fs_total_dir / fs->e2fs_gcount;
619 * Force allocation in another cg if creating a first level dir.
621 ASSERT_VOP_LOCKED(ITOV(pip), "ext2fs_dirpref");
622 if (ITOV(pip)->v_vflag & VV_ROOT) {
623 prefcg = arc4random() % fs->e2fs_gcount;
625 minndir = fs->e2fs_ipg;
626 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
627 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < minndir &&
628 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree &&
629 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= avgbfree) {
631 minndir = e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]);
633 for (cg = 0; cg < prefcg; cg++)
634 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < minndir &&
635 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree &&
636 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= avgbfree) {
638 minndir = e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]);
643 * Count various limits which used for
644 * optimal allocation of a directory inode.
646 maxndir = min(avgndir + fs->e2fs_ipg / 16, fs->e2fs_ipg);
647 minifree = avgifree - avgifree / 4;
650 minbfree = avgbfree - avgbfree / 4;
653 cgsize = fs->e2fs_fsize * fs->e2fs_fpg;
654 dirsize = AVGDIRSIZE;
655 curdirsize = avgndir ? (cgsize - avgbfree * fs->e2fs_bsize) / avgndir : 0;
656 if (dirsize < curdirsize)
657 dirsize = curdirsize;
658 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255);
659 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR);
660 if (maxcontigdirs == 0)
664 * Limit number of dirs in one cg and reserve space for
665 * regular files, but only if we have no deficit in
668 prefcg = ino_to_cg(fs, pip->i_number);
669 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
670 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < maxndir &&
671 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= minifree &&
672 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= minbfree) {
673 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
676 for (cg = 0; cg < prefcg; cg++)
677 if (e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) < maxndir &&
678 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= minifree &&
679 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) >= minbfree) {
680 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
684 * This is a backstop when we have deficit in space.
686 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
687 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree)
689 for (cg = 0; cg < prefcg; cg++)
690 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) >= avgifree)
696 * Select the desired position for the next block in a file.
698 * we try to mimic what Remy does in inode_getblk/block_getblk
700 * we note: blocknr == 0 means that we're about to allocate either
701 * a direct block or a pointer block at the first level of indirection
702 * (In other words, stuff that will go in i_db[] or i_ib[])
704 * blocknr != 0 means that we're allocating a block that is none
705 * of the above. Then, blocknr tells us the number of the block
706 * that will hold the pointer
709 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, e2fs_daddr_t *bap,
710 e2fs_daddr_t blocknr)
717 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
720 * If the next block is actually what we thought it is, then set the
721 * goal to what we thought it should be.
723 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0)
724 return ip->i_next_alloc_goal;
727 * Now check whether we were provided with an array that basically
728 * tells us previous blocks to which we want to stay close.
731 for (tmp = indx - 1; tmp >= 0; tmp--)
736 * Else lets fall back to the blocknr or, if there is none, follow
737 * the rule that a block should be allocated near its inode.
739 return (blocknr ? blocknr :
740 (e2fs_daddr_t)(ip->i_block_group *
741 EXT2_BLOCKS_PER_GROUP(fs)) + fs->e2fs->e2fs_first_dblock);
745 * Implement the cylinder overflow algorithm.
747 * The policy implemented by this algorithm is:
748 * 1) allocate the block in its requested cylinder group.
749 * 2) quadradically rehash on the cylinder group number.
750 * 3) brute force search for a free block.
753 ext2_hashalloc(struct inode *ip, int cg, long pref, int size,
754 daddr_t (*allocator) (struct inode *, int, daddr_t, int))
760 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
763 * 1: preferred cylinder group
765 result = (*allocator)(ip, cg, pref, size);
769 * 2: quadratic rehash
771 for (i = 1; i < fs->e2fs_gcount; i *= 2) {
773 if (cg >= fs->e2fs_gcount)
774 cg -= fs->e2fs_gcount;
775 result = (*allocator)(ip, cg, 0, size);
780 * 3: brute force search
781 * Note that we start at i == 2, since 0 was checked initially,
782 * and 1 is always checked in the quadratic rehash.
784 cg = (icg + 2) % fs->e2fs_gcount;
785 for (i = 2; i < fs->e2fs_gcount; i++) {
786 result = (*allocator)(ip, cg, 0, size);
790 if (cg == fs->e2fs_gcount)
797 ext2_cg_number_gdb_nometa(struct m_ext2fs *fs, int cg)
800 if (!ext2_cg_has_sb(fs, cg))
803 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG))
804 return (fs->e2fs->e3fs_first_meta_bg);
806 return ((fs->e2fs_gcount + EXT2_DESCS_PER_BLOCK(fs) - 1) /
807 EXT2_DESCS_PER_BLOCK(fs));
811 ext2_cg_number_gdb_meta(struct m_ext2fs *fs, int cg)
813 unsigned long metagroup;
816 metagroup = cg / EXT2_DESCS_PER_BLOCK(fs);
817 first = metagroup * EXT2_DESCS_PER_BLOCK(fs);
818 last = first + EXT2_DESCS_PER_BLOCK(fs) - 1;
820 if (cg == first || cg == first + 1 || cg == last)
827 ext2_cg_number_gdb(struct m_ext2fs *fs, int cg)
829 unsigned long first_meta_bg, metagroup;
831 first_meta_bg = fs->e2fs->e3fs_first_meta_bg;
832 metagroup = cg / EXT2_DESCS_PER_BLOCK(fs);
834 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
835 metagroup < first_meta_bg)
836 return (ext2_cg_number_gdb_nometa(fs, cg));
838 return ext2_cg_number_gdb_meta(fs, cg);
842 ext2_number_base_meta_blocks(struct m_ext2fs *fs, int cg)
846 number = ext2_cg_has_sb(fs, cg);
848 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_META_BG) ||
849 cg < fs->e2fs->e3fs_first_meta_bg * EXT2_DESCS_PER_BLOCK(fs)) {
851 number += ext2_cg_number_gdb(fs, cg);
852 number += fs->e2fs->e2fs_reserved_ngdb;
855 number += ext2_cg_number_gdb(fs, cg);
862 ext2_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
866 if (start_bit >= end_bit)
869 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
872 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
876 ext2_get_group_number(struct m_ext2fs *fs, e4fs_daddr_t block)
879 return ((block - fs->e2fs->e2fs_first_dblock) / fs->e2fs_bsize);
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 (!(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->e2fs_bpg + fs->e2fs->e2fs_first_dblock;
909 /* Set bits for block and inode bitmaps, and inode table. */
910 tmp = e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg]);
911 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
912 ext2_block_in_group(fs, tmp, cg))
913 setbit(bp->b_data, tmp - start);
915 tmp = e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg]);
916 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
917 ext2_block_in_group(fs, tmp, cg))
918 setbit(bp->b_data, tmp - start);
920 tmp = e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]);
921 inodes_per_block = fs->e2fs_bsize/EXT2_INODE_SIZE(fs);
922 while( tmp < e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]) +
923 fs->e2fs->e2fs_ipg / inodes_per_block ) {
924 if (!EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG) ||
925 ext2_block_in_group(fs, tmp, cg))
926 setbit(bp->b_data, tmp - start);
931 * Also if the number of blocks within the group is less than
932 * the blocksize * 8 ( which is the size of bitmap ), set rest
933 * of the block bitmap to 1
935 ext2_mark_bitmap_end(fs->e2fs->e2fs_bpg, fs->e2fs_bsize * 8,
939 fs->e2fs_gd[cg].ext4bgd_flags &= ~EXT2_BG_BLOCK_UNINIT;
945 ext2_b_bitmap_validate(struct m_ext2fs *fs, struct buf *bp, int cg)
948 uint64_t group_first_block;
949 unsigned int offset, max_bit;
951 if (EXT2_HAS_INCOMPAT_FEATURE(fs, EXT2F_INCOMPAT_FLEX_BG)) {
953 * It is not possible to check block bitmap in case of this feature,
954 * because the inode and block bitmaps and inode table
955 * blocks may not be in the group at all.
956 * So, skip check in this case.
961 gd = &fs->e2fs_gd[cg];
962 max_bit = fs->e2fs_fpg;
963 group_first_block = ((uint64_t)cg) * fs->e2fs->e2fs_fpg +
964 fs->e2fs->e2fs_first_dblock;
966 /* Check block bitmap block number */
967 offset = e2fs_gd_get_b_bitmap(gd) - group_first_block;
968 if (offset >= max_bit || !isset(bp->b_data, offset)) {
969 SDT_PROBE2(ext2fs, , alloc, ext2_b_bitmap_validate_error,
970 "bad block bitmap, group", cg);
974 /* Check inode bitmap block number */
975 offset = e2fs_gd_get_i_bitmap(gd) - group_first_block;
976 if (offset >= max_bit || !isset(bp->b_data, offset)) {
977 SDT_PROBE2(ext2fs, , alloc, ext2_b_bitmap_validate_error,
978 "bad inode bitmap", cg);
982 /* Check inode table */
983 offset = e2fs_gd_get_i_tables(gd) - group_first_block;
984 if (offset >= max_bit || offset + fs->e2fs_itpg >= max_bit) {
985 SDT_PROBE2(ext2fs, , alloc, ext2_b_bitmap_validate_error,
986 "bad inode table, group", cg);
994 * Determine whether a block can be allocated.
996 * Check to see if a block of the appropriate size is available,
997 * and if it is, allocate it.
1000 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
1002 struct m_ext2fs *fs;
1004 struct ext2mount *ump;
1005 daddr_t bno, runstart, runlen;
1006 int bit, loc, end, error, start;
1011 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0)
1015 error = bread(ip->i_devvp, fsbtodb(fs,
1016 e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1017 (int)fs->e2fs_bsize, NOCRED, &bp);
1021 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1022 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1023 error = ext2_cg_block_bitmap_init(fs, cg, bp);
1027 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1029 error = ext2_gd_b_bitmap_csum_verify(fs, cg, bp);
1033 error = ext2_b_bitmap_validate(fs,bp, cg);
1038 * Check, that another thread did not not allocate the last block in this
1039 * group while we were waiting for the buffer.
1041 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0)
1044 bbp = (char *)bp->b_data;
1046 if (dtog(fs, bpref) != cg)
1049 bpref = dtogd(fs, bpref);
1051 * if the requested block is available, use it
1053 if (isclr(bbp, bpref)) {
1059 * no blocks in the requested cylinder, so take next
1060 * available one in this cylinder group.
1061 * first try to get 8 contigous blocks, then fall back to a single
1065 start = dtogd(fs, bpref) / NBBY;
1068 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
1072 for (loc = start; loc < end; loc++) {
1073 if (bbp[loc] == (char)0xff) {
1078 /* Start of a run, find the number of high clear bits. */
1080 bit = fls(bbp[loc]);
1081 runlen = NBBY - bit;
1082 runstart = loc * NBBY + bit;
1083 } else if (bbp[loc] == 0) {
1084 /* Continue a run. */
1088 * Finish the current run. If it isn't long
1089 * enough, start a new one.
1091 bit = ffs(bbp[loc]) - 1;
1098 /* Run was too short, start a new one. */
1099 bit = fls(bbp[loc]);
1100 runlen = NBBY - bit;
1101 runstart = loc * NBBY + bit;
1104 /* If the current run is long enough, use it. */
1115 bno = ext2_mapsearch(fs, bbp, bpref);
1121 if (isset(bbp, bno)) {
1122 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n",
1123 cg, (intmax_t)bno, fs->e2fs_fsmnt);
1124 panic("ext2fs_alloccg: dup alloc");
1129 ext2_clusteracct(fs, bbp, cg, bno, -1);
1131 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1132 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1);
1135 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1137 return (((uint64_t)cg) * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
1146 * Determine whether a cluster can be allocated.
1149 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len)
1151 struct m_ext2fs *fs;
1152 struct ext2mount *ump;
1155 int bit, error, got, i, loc, run;
1162 if (fs->e2fs_maxcluster[cg] < len)
1166 error = bread(ip->i_devvp,
1167 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1168 (int)fs->e2fs_bsize, NOCRED, &bp);
1172 bbp = (char *)bp->b_data;
1175 * Check to see if a cluster of the needed size (or bigger) is
1176 * available in this cylinder group.
1178 lp = &fs->e2fs_clustersum[cg].cs_sum[len];
1179 for (i = len; i <= fs->e2fs_contigsumsize; i++)
1182 if (i > fs->e2fs_contigsumsize) {
1184 * Update the cluster summary information to reflect
1185 * the true maximum-sized cluster so that future cluster
1186 * allocation requests can avoid reading the bitmap only
1187 * to find no cluster.
1189 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1];
1190 for (i = len - 1; i > 0; i--)
1193 fs->e2fs_maxcluster[cg] = i;
1198 /* Search the bitmap to find a big enough cluster like in FFS. */
1199 if (dtog(fs, bpref) != cg)
1202 bpref = dtogd(fs, bpref);
1204 bit = 1 << (bpref % NBBY);
1205 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) {
1206 if ((bbp[loc] & bit) != 0)
1213 if ((got & (NBBY - 1)) != (NBBY - 1))
1221 if (got >= fs->e2fs->e2fs_fpg)
1224 /* Allocate the cluster that we found. */
1225 for (i = 1; i < len; i++)
1226 if (!isclr(bbp, got - run + i))
1227 panic("ext2_clusteralloc: map mismatch");
1229 bno = got - run + 1;
1230 if (bno >= fs->e2fs->e2fs_fpg)
1231 panic("ext2_clusteralloc: allocated out of group");
1234 for (i = 0; i < len; i += fs->e2fs_fpb) {
1235 setbit(bbp, bno + i);
1236 ext2_clusteracct(fs, bbp, cg, bno + i, -1);
1238 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1239 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1);
1245 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
1255 ext2_zero_inode_table(struct inode *ip, int cg)
1257 struct m_ext2fs *fs;
1259 int i, all_blks, used_blks;
1263 if (fs->e2fs_gd[cg].ext4bgd_flags & EXT2_BG_INODE_ZEROED)
1266 all_blks = fs->e2fs->e2fs_inode_size * fs->e2fs->e2fs_ipg /
1269 used_blks = howmany(fs->e2fs->e2fs_ipg -
1270 e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]),
1271 fs->e2fs_bsize / EXT2_INODE_SIZE(fs));
1273 for (i = 0; i < all_blks - used_blks; i++) {
1274 bp = getblk(ip->i_devvp, fsbtodb(fs,
1275 e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]) + used_blks + i),
1276 fs->e2fs_bsize, 0, 0, 0);
1280 vfs_bio_bzero_buf(bp, 0, fs->e2fs_bsize);
1284 fs->e2fs_gd[cg].ext4bgd_flags |= EXT2_BG_INODE_ZEROED;
1290 * Determine whether an inode can be allocated.
1292 * Check to see if an inode is available, and if it is,
1293 * allocate it using tode in the specified cylinder group.
1296 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode)
1298 struct m_ext2fs *fs;
1300 struct ext2mount *ump;
1301 int error, start, len, ifree;
1304 ipref--; /* to avoid a lot of (ipref -1) */
1309 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0)
1312 error = bread(ip->i_devvp, fsbtodb(fs,
1313 e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])),
1314 (int)fs->e2fs_bsize, NOCRED, &bp);
1320 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1321 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1322 if (fs->e2fs_gd[cg].ext4bgd_flags & EXT2_BG_INODE_UNINIT) {
1323 memset(bp->b_data, 0, fs->e2fs_bsize);
1324 fs->e2fs_gd[cg].ext4bgd_flags &= ~EXT2_BG_INODE_UNINIT;
1326 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1327 error = ext2_zero_inode_table(ip, cg);
1334 error = ext2_gd_i_bitmap_csum_verify(fs, cg, bp);
1340 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0) {
1342 * Another thread allocated the last i-node in this
1343 * group while we were waiting for the buffer.
1349 ibp = (char *)bp->b_data;
1351 ipref %= fs->e2fs->e2fs_ipg;
1352 if (isclr(ibp, ipref))
1355 start = ipref / NBBY;
1356 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY);
1357 loc = memcchr(&ibp[start], 0xff, len);
1361 loc = memcchr(&ibp[start], 0xff, len);
1363 SDT_PROBE3(ext2fs, , alloc, ext2_nodealloccg_bmap_corrupted,
1364 cg, ipref, fs->e2fs_fsmnt);
1370 ipref = (loc - ibp) * NBBY + ffs(~*loc) - 1;
1374 e2fs_gd_set_nifree(&fs->e2fs_gd[cg],
1375 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) - 1);
1376 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1377 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1378 ifree = fs->e2fs->e2fs_ipg - e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]);
1379 if (ipref + 1 > ifree)
1380 e2fs_gd_set_i_unused(&fs->e2fs_gd[cg],
1381 fs->e2fs->e2fs_ipg - (ipref + 1));
1383 fs->e2fs->e2fs_ficount--;
1385 if ((mode & IFMT) == IFDIR) {
1386 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg],
1387 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) + 1);
1388 fs->e2fs_total_dir++;
1391 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1393 return ((uint64_t)cg * fs->e2fs_ipg + ipref + 1);
1397 * Free a block or fragment.
1401 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size)
1403 struct m_ext2fs *fs;
1405 struct ext2mount *ump;
1412 if (bno >= fs->e2fs_bcount) {
1413 SDT_PROBE2(ext2fs, , alloc, ext2_blkfree_bad_block, ip->i_number, bno);
1416 error = bread(ip->i_devvp,
1417 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1418 (int)fs->e2fs_bsize, NOCRED, &bp);
1423 bbp = (char *)bp->b_data;
1424 bno = dtogd(fs, bno);
1425 if (isclr(bbp, bno)) {
1426 panic("ext2_blkfree: freeing free block %lld, fs=%s",
1427 (long long)bno, fs->e2fs_fsmnt);
1431 ext2_clusteracct(fs, bbp, cg, bno, 1);
1433 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1434 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) + 1);
1437 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1446 ext2_vfree(struct vnode *pvp, ino_t ino, int mode)
1448 struct m_ext2fs *fs;
1451 struct ext2mount *ump;
1458 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount)
1459 panic("ext2_vfree: range: devvp = %p, ino = %ju, fs = %s",
1460 pip->i_devvp, (uintmax_t)ino, fs->e2fs_fsmnt);
1462 cg = ino_to_cg(fs, ino);
1463 error = bread(pip->i_devvp,
1464 fsbtodb(fs, e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])),
1465 (int)fs->e2fs_bsize, NOCRED, &bp);
1470 ibp = (char *)bp->b_data;
1471 ino = (ino - 1) % fs->e2fs->e2fs_ipg;
1472 if (isclr(ibp, ino)) {
1473 SDT_PROBE2(ext2fs, , alloc, ext2_vfree_doublefree,
1474 fs->e2fs_fsmnt, ino);
1475 if (fs->e2fs_ronly == 0)
1476 panic("ext2_vfree: freeing free inode");
1480 fs->e2fs->e2fs_ficount++;
1481 e2fs_gd_set_nifree(&fs->e2fs_gd[cg],
1482 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) + 1);
1483 if ((mode & IFMT) == IFDIR) {
1484 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg],
1485 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) - 1);
1486 fs->e2fs_total_dir--;
1490 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1496 * Find a block in the specified cylinder group.
1498 * It is a panic if a request is made to find a block if none are
1502 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref)
1508 * find the fragment by searching through the free block
1509 * map for an appropriate bit pattern
1512 start = dtogd(fs, bpref) / NBBY;
1515 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
1516 loc = memcchr(&bbp[start], 0xff, len);
1520 loc = memcchr(&bbp[start], 0xff, len);
1522 panic("ext2_mapsearch: map corrupted: start=%d, len=%d, fs=%s",
1523 start, len, fs->e2fs_fsmnt);
1527 return ((loc - bbp) * NBBY + ffs(~*loc) - 1);
1531 ext2_cg_has_sb(struct m_ext2fs *fs, int cg)
1538 if (EXT2_HAS_COMPAT_FEATURE(fs, EXT2F_COMPAT_SPARSESUPER2)) {
1539 if (cg == fs->e2fs->e4fs_backup_bgs[0] ||
1540 cg == fs->e2fs->e4fs_backup_bgs[1])
1546 !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_SPARSESUPER))
1552 for (a3 = 3, a5 = 5, a7 = 7;
1553 a3 <= cg || a5 <= cg || a7 <= cg;
1554 a3 *= 3, a5 *= 5, a7 *= 7)
1555 if (cg == a3 || cg == a5 || cg == a7)