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: releng/12.0/sys/fs/ext2fs/ext2_alloc.c 338151 2018-08-21 18:39:29Z fsu $
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 static 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 static 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 unsigned long
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 unsigned long
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 static unsigned long
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 /*
906 * Determine whether a block can be allocated.
907 *
908 * Check to see if a block of the appropriate size is available,
909 * and if it is, allocate it.
910 */
911 static daddr_t
912 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
913 {
914 struct m_ext2fs *fs;
915 struct buf *bp;
916 struct ext2mount *ump;
917 daddr_t bno, runstart, runlen;
918 int bit, loc, end, error, start;
919 char *bbp;
920 /* XXX ondisk32 */
921 fs = ip->i_e2fs;
922 ump = ip->i_ump;
923 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0)
924 return (0);
925 EXT2_UNLOCK(ump);
926 error = bread(ip->i_devvp, fsbtodb(fs,
927 e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
928 (int)fs->e2fs_bsize, NOCRED, &bp);
929 if (error) {
930 brelse(bp);
931 EXT2_LOCK(ump);
932 return (0);
933 }
934 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
935 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
936 error = ext2_cg_block_bitmap_init(fs, cg, bp);
937 if (error) {
938 brelse(bp);
939 EXT2_LOCK(ump);
940 return (0);
941 }
942 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
943 }
944 error = ext2_gd_b_bitmap_csum_verify(fs, cg, bp);
945 if (error) {
946 brelse(bp);
947 EXT2_LOCK(ump);
948 return (0);
949 }
950 if (e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) == 0) {
951 /*
952 * Another thread allocated the last block in this
953 * group while we were waiting for the buffer.
954 */
955 brelse(bp);
956 EXT2_LOCK(ump);
957 return (0);
958 }
959 bbp = (char *)bp->b_data;
960
961 if (dtog(fs, bpref) != cg)
962 bpref = 0;
963 if (bpref != 0) {
964 bpref = dtogd(fs, bpref);
965 /*
966 * if the requested block is available, use it
967 */
968 if (isclr(bbp, bpref)) {
969 bno = bpref;
970 goto gotit;
971 }
972 }
973 /*
974 * no blocks in the requested cylinder, so take next
975 * available one in this cylinder group.
976 * first try to get 8 contigous blocks, then fall back to a single
977 * block.
978 */
979 if (bpref)
980 start = dtogd(fs, bpref) / NBBY;
981 else
982 start = 0;
983 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
984 retry:
985 runlen = 0;
986 runstart = 0;
987 for (loc = start; loc < end; loc++) {
988 if (bbp[loc] == (char)0xff) {
989 runlen = 0;
990 continue;
991 }
992
993 /* Start of a run, find the number of high clear bits. */
994 if (runlen == 0) {
995 bit = fls(bbp[loc]);
996 runlen = NBBY - bit;
997 runstart = loc * NBBY + bit;
998 } else if (bbp[loc] == 0) {
999 /* Continue a run. */
1000 runlen += NBBY;
1001 } else {
1002 /*
1003 * Finish the current run. If it isn't long
1004 * enough, start a new one.
1005 */
1006 bit = ffs(bbp[loc]) - 1;
1007 runlen += bit;
1008 if (runlen >= 8) {
1009 bno = runstart;
1010 goto gotit;
1011 }
1012
1013 /* Run was too short, start a new one. */
1014 bit = fls(bbp[loc]);
1015 runlen = NBBY - bit;
1016 runstart = loc * NBBY + bit;
1017 }
1018
1019 /* If the current run is long enough, use it. */
1020 if (runlen >= 8) {
1021 bno = runstart;
1022 goto gotit;
1023 }
1024 }
1025 if (start != 0) {
1026 end = start;
1027 start = 0;
1028 goto retry;
1029 }
1030 bno = ext2_mapsearch(fs, bbp, bpref);
1031 if (bno < 0) {
1032 brelse(bp);
1033 EXT2_LOCK(ump);
1034 return (0);
1035 }
1036 gotit:
1037 #ifdef INVARIANTS
1038 if (isset(bbp, bno)) {
1039 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n",
1040 cg, (intmax_t)bno, fs->e2fs_fsmnt);
1041 panic("ext2fs_alloccg: dup alloc");
1042 }
1043 #endif
1044 setbit(bbp, bno);
1045 EXT2_LOCK(ump);
1046 ext2_clusteracct(fs, bbp, cg, bno, -1);
1047 fs->e2fs_fbcount--;
1048 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1049 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1);
1050 fs->e2fs_fmod = 1;
1051 EXT2_UNLOCK(ump);
1052 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1053 bdwrite(bp);
1054 return (((uint64_t)cg) * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
1055 }
1056
1057 /*
1058 * Determine whether a cluster can be allocated.
1059 */
1060 static daddr_t
1061 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len)
1062 {
1063 struct m_ext2fs *fs;
1064 struct ext2mount *ump;
1065 struct buf *bp;
1066 char *bbp;
1067 int bit, error, got, i, loc, run;
1068 int32_t *lp;
1069 daddr_t bno;
1070
1071 fs = ip->i_e2fs;
1072 ump = ip->i_ump;
1073
1074 if (fs->e2fs_maxcluster[cg] < len)
1075 return (0);
1076
1077 EXT2_UNLOCK(ump);
1078 error = bread(ip->i_devvp,
1079 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1080 (int)fs->e2fs_bsize, NOCRED, &bp);
1081 if (error)
1082 goto fail_lock;
1083
1084 bbp = (char *)bp->b_data;
1085 EXT2_LOCK(ump);
1086 /*
1087 * Check to see if a cluster of the needed size (or bigger) is
1088 * available in this cylinder group.
1089 */
1090 lp = &fs->e2fs_clustersum[cg].cs_sum[len];
1091 for (i = len; i <= fs->e2fs_contigsumsize; i++)
1092 if (*lp++ > 0)
1093 break;
1094 if (i > fs->e2fs_contigsumsize) {
1095 /*
1096 * Update the cluster summary information to reflect
1097 * the true maximum-sized cluster so that future cluster
1098 * allocation requests can avoid reading the bitmap only
1099 * to find no cluster.
1100 */
1101 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1];
1102 for (i = len - 1; i > 0; i--)
1103 if (*lp-- > 0)
1104 break;
1105 fs->e2fs_maxcluster[cg] = i;
1106 goto fail;
1107 }
1108 EXT2_UNLOCK(ump);
1109
1110 /* Search the bitmap to find a big enough cluster like in FFS. */
1111 if (dtog(fs, bpref) != cg)
1112 bpref = 0;
1113 if (bpref != 0)
1114 bpref = dtogd(fs, bpref);
1115 loc = bpref / NBBY;
1116 bit = 1 << (bpref % NBBY);
1117 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) {
1118 if ((bbp[loc] & bit) != 0)
1119 run = 0;
1120 else {
1121 run++;
1122 if (run == len)
1123 break;
1124 }
1125 if ((got & (NBBY - 1)) != (NBBY - 1))
1126 bit <<= 1;
1127 else {
1128 loc++;
1129 bit = 1;
1130 }
1131 }
1132
1133 if (got >= fs->e2fs->e2fs_fpg)
1134 goto fail_lock;
1135
1136 /* Allocate the cluster that we found. */
1137 for (i = 1; i < len; i++)
1138 if (!isclr(bbp, got - run + i))
1139 panic("ext2_clusteralloc: map mismatch");
1140
1141 bno = got - run + 1;
1142 if (bno >= fs->e2fs->e2fs_fpg)
1143 panic("ext2_clusteralloc: allocated out of group");
1144
1145 EXT2_LOCK(ump);
1146 for (i = 0; i < len; i += fs->e2fs_fpb) {
1147 setbit(bbp, bno + i);
1148 ext2_clusteracct(fs, bbp, cg, bno + i, -1);
1149 fs->e2fs_fbcount--;
1150 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1151 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) - 1);
1152 }
1153 fs->e2fs_fmod = 1;
1154 EXT2_UNLOCK(ump);
1155
1156 bdwrite(bp);
1157 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
1158
1159 fail_lock:
1160 EXT2_LOCK(ump);
1161 fail:
1162 brelse(bp);
1163 return (0);
1164 }
1165
1166 static int
1167 ext2_zero_inode_table(struct inode *ip, int cg)
1168 {
1169 struct m_ext2fs *fs;
1170 struct buf *bp;
1171 int i, all_blks, used_blks;
1172
1173 fs = ip->i_e2fs;
1174
1175 if (fs->e2fs_gd[cg].ext4bgd_flags & EXT2_BG_INODE_ZEROED)
1176 return (0);
1177
1178 all_blks = fs->e2fs->e2fs_inode_size * fs->e2fs->e2fs_ipg /
1179 fs->e2fs_bsize;
1180
1181 used_blks = howmany(fs->e2fs->e2fs_ipg -
1182 e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]),
1183 fs->e2fs_bsize / EXT2_INODE_SIZE(fs));
1184
1185 for (i = 0; i < all_blks - used_blks; i++) {
1186 bp = getblk(ip->i_devvp, fsbtodb(fs,
1187 e2fs_gd_get_i_tables(&fs->e2fs_gd[cg]) + used_blks + i),
1188 fs->e2fs_bsize, 0, 0, 0);
1189 if (!bp)
1190 return (EIO);
1191
1192 vfs_bio_bzero_buf(bp, 0, fs->e2fs_bsize);
1193 bawrite(bp);
1194 }
1195
1196 fs->e2fs_gd[cg].ext4bgd_flags |= EXT2_BG_INODE_ZEROED;
1197
1198 return (0);
1199 }
1200
1201 /*
1202 * Determine whether an inode can be allocated.
1203 *
1204 * Check to see if an inode is available, and if it is,
1205 * allocate it using tode in the specified cylinder group.
1206 */
1207 static daddr_t
1208 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode)
1209 {
1210 struct m_ext2fs *fs;
1211 struct buf *bp;
1212 struct ext2mount *ump;
1213 int error, start, len, ifree;
1214 char *ibp, *loc;
1215
1216 ipref--; /* to avoid a lot of (ipref -1) */
1217 if (ipref == -1)
1218 ipref = 0;
1219 fs = ip->i_e2fs;
1220 ump = ip->i_ump;
1221 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0)
1222 return (0);
1223 EXT2_UNLOCK(ump);
1224 error = bread(ip->i_devvp, fsbtodb(fs,
1225 e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])),
1226 (int)fs->e2fs_bsize, NOCRED, &bp);
1227 if (error) {
1228 brelse(bp);
1229 EXT2_LOCK(ump);
1230 return (0);
1231 }
1232 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1233 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1234 if (fs->e2fs_gd[cg].ext4bgd_flags & EXT2_BG_INODE_UNINIT) {
1235 memset(bp->b_data, 0, fs->e2fs_bsize);
1236 fs->e2fs_gd[cg].ext4bgd_flags &= ~EXT2_BG_INODE_UNINIT;
1237 }
1238 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1239 error = ext2_zero_inode_table(ip, cg);
1240 if (error) {
1241 brelse(bp);
1242 EXT2_LOCK(ump);
1243 return (0);
1244 }
1245 }
1246 error = ext2_gd_i_bitmap_csum_verify(fs, cg, bp);
1247 if (error) {
1248 brelse(bp);
1249 EXT2_LOCK(ump);
1250 return (0);
1251 }
1252 if (e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) == 0) {
1253 /*
1254 * Another thread allocated the last i-node in this
1255 * group while we were waiting for the buffer.
1256 */
1257 brelse(bp);
1258 EXT2_LOCK(ump);
1259 return (0);
1260 }
1261 ibp = (char *)bp->b_data;
1262 if (ipref) {
1263 ipref %= fs->e2fs->e2fs_ipg;
1264 if (isclr(ibp, ipref))
1265 goto gotit;
1266 }
1267 start = ipref / NBBY;
1268 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY);
1269 loc = memcchr(&ibp[start], 0xff, len);
1270 if (loc == NULL) {
1271 len = start + 1;
1272 start = 0;
1273 loc = memcchr(&ibp[start], 0xff, len);
1274 if (loc == NULL) {
1275 printf("cg = %d, ipref = %lld, fs = %s\n",
1276 cg, (long long)ipref, fs->e2fs_fsmnt);
1277 panic("ext2fs_nodealloccg: map corrupted");
1278 /* NOTREACHED */
1279 }
1280 }
1281 ipref = (loc - ibp) * NBBY + ffs(~*loc) - 1;
1282 gotit:
1283 setbit(ibp, ipref);
1284 EXT2_LOCK(ump);
1285 e2fs_gd_set_nifree(&fs->e2fs_gd[cg],
1286 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) - 1);
1287 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_GDT_CSUM) ||
1288 EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1289 ifree = fs->e2fs->e2fs_ipg - e2fs_gd_get_i_unused(&fs->e2fs_gd[cg]);
1290 if (ipref + 1 > ifree)
1291 e2fs_gd_set_i_unused(&fs->e2fs_gd[cg],
1292 fs->e2fs->e2fs_ipg - (ipref + 1));
1293 }
1294 fs->e2fs->e2fs_ficount--;
1295 fs->e2fs_fmod = 1;
1296 if ((mode & IFMT) == IFDIR) {
1297 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg],
1298 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) + 1);
1299 fs->e2fs_total_dir++;
1300 }
1301 EXT2_UNLOCK(ump);
1302 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1303 bdwrite(bp);
1304 return ((uint64_t)cg * fs->e2fs_ipg + ipref + 1);
1305 }
1306
1307 /*
1308 * Free a block or fragment.
1309 *
1310 */
1311 void
1312 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size)
1313 {
1314 struct m_ext2fs *fs;
1315 struct buf *bp;
1316 struct ext2mount *ump;
1317 int cg, error;
1318 char *bbp;
1319
1320 fs = ip->i_e2fs;
1321 ump = ip->i_ump;
1322 cg = dtog(fs, bno);
1323 if (bno >= fs->e2fs_bcount) {
1324 printf("bad block %lld, ino %ju\n", (long long)bno,
1325 (uintmax_t)ip->i_number);
1326 ext2_fserr(fs, ip->i_uid, "bad block");
1327 return;
1328 }
1329 error = bread(ip->i_devvp,
1330 fsbtodb(fs, e2fs_gd_get_b_bitmap(&fs->e2fs_gd[cg])),
1331 (int)fs->e2fs_bsize, NOCRED, &bp);
1332 if (error) {
1333 brelse(bp);
1334 return;
1335 }
1336 bbp = (char *)bp->b_data;
1337 bno = dtogd(fs, bno);
1338 if (isclr(bbp, bno)) {
1339 printf("block = %lld, fs = %s\n",
1340 (long long)bno, fs->e2fs_fsmnt);
1341 panic("ext2_blkfree: freeing free block");
1342 }
1343 clrbit(bbp, bno);
1344 EXT2_LOCK(ump);
1345 ext2_clusteracct(fs, bbp, cg, bno, 1);
1346 fs->e2fs_fbcount++;
1347 e2fs_gd_set_nbfree(&fs->e2fs_gd[cg],
1348 e2fs_gd_get_nbfree(&fs->e2fs_gd[cg]) + 1);
1349 fs->e2fs_fmod = 1;
1350 EXT2_UNLOCK(ump);
1351 ext2_gd_b_bitmap_csum_set(fs, cg, bp);
1352 bdwrite(bp);
1353 }
1354
1355 /*
1356 * Free an inode.
1357 *
1358 */
1359 int
1360 ext2_vfree(struct vnode *pvp, ino_t ino, int mode)
1361 {
1362 struct m_ext2fs *fs;
1363 struct inode *pip;
1364 struct buf *bp;
1365 struct ext2mount *ump;
1366 int error, cg;
1367 char *ibp;
1368
1369 pip = VTOI(pvp);
1370 fs = pip->i_e2fs;
1371 ump = pip->i_ump;
1372 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount)
1373 panic("ext2_vfree: range: devvp = %p, ino = %ju, fs = %s",
1374 pip->i_devvp, (uintmax_t)ino, fs->e2fs_fsmnt);
1375
1376 cg = ino_to_cg(fs, ino);
1377 error = bread(pip->i_devvp,
1378 fsbtodb(fs, e2fs_gd_get_i_bitmap(&fs->e2fs_gd[cg])),
1379 (int)fs->e2fs_bsize, NOCRED, &bp);
1380 if (error) {
1381 brelse(bp);
1382 return (0);
1383 }
1384 ibp = (char *)bp->b_data;
1385 ino = (ino - 1) % fs->e2fs->e2fs_ipg;
1386 if (isclr(ibp, ino)) {
1387 printf("ino = %ju, fs = %s\n",
1388 ino, fs->e2fs_fsmnt);
1389 if (fs->e2fs_ronly == 0)
1390 panic("ext2_vfree: freeing free inode");
1391 }
1392 clrbit(ibp, ino);
1393 EXT2_LOCK(ump);
1394 fs->e2fs->e2fs_ficount++;
1395 e2fs_gd_set_nifree(&fs->e2fs_gd[cg],
1396 e2fs_gd_get_nifree(&fs->e2fs_gd[cg]) + 1);
1397 if ((mode & IFMT) == IFDIR) {
1398 e2fs_gd_set_ndirs(&fs->e2fs_gd[cg],
1399 e2fs_gd_get_ndirs(&fs->e2fs_gd[cg]) - 1);
1400 fs->e2fs_total_dir--;
1401 }
1402 fs->e2fs_fmod = 1;
1403 EXT2_UNLOCK(ump);
1404 ext2_gd_i_bitmap_csum_set(fs, cg, bp);
1405 bdwrite(bp);
1406 return (0);
1407 }
1408
1409 /*
1410 * Find a block in the specified cylinder group.
1411 *
1412 * It is a panic if a request is made to find a block if none are
1413 * available.
1414 */
1415 static daddr_t
1416 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref)
1417 {
1418 char *loc;
1419 int start, len;
1420
1421 /*
1422 * find the fragment by searching through the free block
1423 * map for an appropriate bit pattern
1424 */
1425 if (bpref)
1426 start = dtogd(fs, bpref) / NBBY;
1427 else
1428 start = 0;
1429 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
1430 loc = memcchr(&bbp[start], 0xff, len);
1431 if (loc == NULL) {
1432 len = start + 1;
1433 start = 0;
1434 loc = memcchr(&bbp[start], 0xff, len);
1435 if (loc == NULL) {
1436 printf("start = %d, len = %d, fs = %s\n",
1437 start, len, fs->e2fs_fsmnt);
1438 panic("ext2_mapsearch: map corrupted");
1439 /* NOTREACHED */
1440 }
1441 }
1442 return ((loc - bbp) * NBBY + ffs(~*loc) - 1);
1443 }
1444
1445 /*
1446 * Fserr prints the name of a filesystem with an error diagnostic.
1447 *
1448 * The form of the error message is:
1449 * fs: error message
1450 */
1451 void
1452 ext2_fserr(struct m_ext2fs *fs, uid_t uid, char *cp)
1453 {
1454
1455 log(LOG_ERR, "uid %u on %s: %s\n", uid, fs->e2fs_fsmnt, cp);
1456 }
1457
1458 int
1459 ext2_cg_has_sb(struct m_ext2fs *fs, int cg)
1460 {
1461 int a3, a5, a7;
1462
1463 if (cg == 0)
1464 return (1);
1465
1466 if (EXT2_HAS_COMPAT_FEATURE(fs, EXT2F_COMPAT_SPARSESUPER2)) {
1467 if (cg == fs->e2fs->e4fs_backup_bgs[0] ||
1468 cg == fs->e2fs->e4fs_backup_bgs[1])
1469 return (1);
1470 return (0);
1471 }
1472
1473 if ((cg <= 1) ||
1474 !EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_SPARSESUPER))
1475 return (1);
1476
1477 if (!(cg & 1))
1478 return (0);
1479
1480 for (a3 = 3, a5 = 5, a7 = 7;
1481 a3 <= cg || a5 <= cg || a7 <= cg;
1482 a3 *= 3, a5 *= 5, a7 *= 7)
1483 if (cg == a3 || cg == a5 || cg == a7)
1484 return (1);
1485 return (0);
1486 }
Cache object: edc4822a45323c1ef30a5a94b6c74ad3
|