1 /*-
2 * modified for Lites 1.1
3 *
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
6 */
7 /*-
8 * Copyright (c) 1982, 1986, 1989, 1993
9 * The Regents of the University of California. All rights reserved.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 * @(#)ffs_alloc.c 8.8 (Berkeley) 2/21/94
36 * $FreeBSD$
37 */
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/conf.h>
42 #include <sys/vnode.h>
43 #include <sys/stat.h>
44 #include <sys/mount.h>
45 #include <sys/sysctl.h>
46 #include <sys/syslog.h>
47 #include <sys/buf.h>
48
49 #include <fs/ext2fs/fs.h>
50 #include <fs/ext2fs/inode.h>
51 #include <fs/ext2fs/ext2_mount.h>
52 #include <fs/ext2fs/ext2fs.h>
53 #include <fs/ext2fs/ext2_extern.h>
54
55 static daddr_t ext2_alloccg(struct inode *, int, daddr_t, int);
56 static daddr_t ext2_clusteralloc(struct inode *, int, daddr_t, int);
57 static u_long ext2_dirpref(struct inode *);
58 static void ext2_fserr(struct m_ext2fs *, uid_t, char *);
59 static u_long ext2_hashalloc(struct inode *, int, long, int,
60 daddr_t (*)(struct inode *, int, daddr_t,
61 int));
62 static daddr_t ext2_nodealloccg(struct inode *, int, daddr_t, int);
63 static daddr_t ext2_mapsearch(struct m_ext2fs *, char *, daddr_t);
64
65 /*
66 * Allocate a block in the filesystem.
67 *
68 * A preference may be optionally specified. If a preference is given
69 * the following hierarchy is used to allocate a block:
70 * 1) allocate the requested block.
71 * 2) allocate a rotationally optimal block in the same cylinder.
72 * 3) allocate a block in the same cylinder group.
73 * 4) quadradically rehash into other cylinder groups, until an
74 * available block is located.
75 * If no block preference is given the following hierarchy is used
76 * to allocate a block:
77 * 1) allocate a block in the cylinder group that contains the
78 * inode for the file.
79 * 2) quadradically rehash into other cylinder groups, until an
80 * available block is located.
81 */
82 int
83 ext2_alloc(struct inode *ip, daddr_t lbn, e4fs_daddr_t bpref, int size,
84 struct ucred *cred, e4fs_daddr_t *bnp)
85 {
86 struct m_ext2fs *fs;
87 struct ext2mount *ump;
88 int32_t bno;
89 int cg;
90 *bnp = 0;
91 fs = ip->i_e2fs;
92 ump = ip->i_ump;
93 mtx_assert(EXT2_MTX(ump), MA_OWNED);
94 #ifdef INVARIANTS
95 if ((u_int)size > fs->e2fs_bsize || blkoff(fs, size) != 0) {
96 vn_printf(ip->i_devvp, "bsize = %lu, size = %d, fs = %s\n",
97 (long unsigned int)fs->e2fs_bsize, size, fs->e2fs_fsmnt);
98 panic("ext2_alloc: bad size");
99 }
100 if (cred == NOCRED)
101 panic("ext2_alloc: missing credential");
102 #endif /* INVARIANTS */
103 if (size == fs->e2fs_bsize && fs->e2fs->e2fs_fbcount == 0)
104 goto nospace;
105 if (cred->cr_uid != 0 &&
106 fs->e2fs->e2fs_fbcount < fs->e2fs->e2fs_rbcount)
107 goto nospace;
108 if (bpref >= fs->e2fs->e2fs_bcount)
109 bpref = 0;
110 if (bpref == 0)
111 cg = ino_to_cg(fs, ip->i_number);
112 else
113 cg = dtog(fs, bpref);
114 bno = (daddr_t)ext2_hashalloc(ip, cg, bpref, fs->e2fs_bsize,
115 ext2_alloccg);
116 if (bno > 0) {
117 /* set next_alloc fields as done in block_getblk */
118 ip->i_next_alloc_block = lbn;
119 ip->i_next_alloc_goal = bno;
120
121 ip->i_blocks += btodb(fs->e2fs_bsize);
122 ip->i_flag |= IN_CHANGE | IN_UPDATE;
123 *bnp = bno;
124 return (0);
125 }
126 nospace:
127 EXT2_UNLOCK(ump);
128 ext2_fserr(fs, cred->cr_uid, "filesystem full");
129 uprintf("\n%s: write failed, filesystem is full\n", fs->e2fs_fsmnt);
130 return (ENOSPC);
131 }
132
133 /*
134 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
135 *
136 * The vnode and an array of buffer pointers for a range of sequential
137 * logical blocks to be made contiguous is given. The allocator attempts
138 * to find a range of sequential blocks starting as close as possible to
139 * an fs_rotdelay offset from the end of the allocation for the logical
140 * block immediately preceding the current range. If successful, the
141 * physical block numbers in the buffer pointers and in the inode are
142 * changed to reflect the new allocation. If unsuccessful, the allocation
143 * is left unchanged. The success in doing the reallocation is returned.
144 * Note that the error return is not reflected back to the user. Rather
145 * the previous block allocation will be used.
146 */
147
148 static SYSCTL_NODE(_vfs, OID_AUTO, ext2fs, CTLFLAG_RW, 0, "EXT2FS filesystem");
149
150 static int doasyncfree = 1;
151 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0,
152 "Use asychronous writes to update block pointers when freeing blocks");
153
154 static int doreallocblks = 1;
155 SYSCTL_INT(_vfs_ext2fs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
156
157 int
158 ext2_reallocblks(struct vop_reallocblks_args *ap)
159 {
160 struct m_ext2fs *fs;
161 struct inode *ip;
162 struct vnode *vp;
163 struct buf *sbp, *ebp;
164 uint32_t *bap, *sbap, *ebap;
165 struct ext2mount *ump;
166 struct cluster_save *buflist;
167 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
168 e2fs_lbn_t start_lbn, end_lbn;
169 int soff;
170 e2fs_daddr_t newblk, blkno;
171 int i, len, start_lvl, end_lvl, pref, ssize;
172
173 if (doreallocblks == 0)
174 return (ENOSPC);
175
176 vp = ap->a_vp;
177 ip = VTOI(vp);
178 fs = ip->i_e2fs;
179 ump = ip->i_ump;
180
181 if (fs->e2fs_contigsumsize <= 0)
182 return (ENOSPC);
183
184 buflist = ap->a_buflist;
185 len = buflist->bs_nchildren;
186 start_lbn = buflist->bs_children[0]->b_lblkno;
187 end_lbn = start_lbn + len - 1;
188 #ifdef INVARIANTS
189 for (i = 1; i < len; i++)
190 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
191 panic("ext2_reallocblks: non-cluster");
192 #endif
193 /*
194 * If the cluster crosses the boundary for the first indirect
195 * block, leave space for the indirect block. Indirect blocks
196 * are initially laid out in a position after the last direct
197 * block. Block reallocation would usually destroy locality by
198 * moving the indirect block out of the way to make room for
199 * data blocks if we didn't compensate here. We should also do
200 * this for other indirect block boundaries, but it is only
201 * important for the first one.
202 */
203 if (start_lbn < NDADDR && end_lbn >= NDADDR)
204 return (ENOSPC);
205 /*
206 * If the latest allocation is in a new cylinder group, assume that
207 * the filesystem has decided to move and do not force it back to
208 * the previous cylinder group.
209 */
210 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
211 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
212 return (ENOSPC);
213 if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
214 ext2_getlbns(vp, end_lbn, end_ap, &end_lvl))
215 return (ENOSPC);
216 /*
217 * Get the starting offset and block map for the first block.
218 */
219 if (start_lvl == 0) {
220 sbap = &ip->i_db[0];
221 soff = start_lbn;
222 } else {
223 idp = &start_ap[start_lvl - 1];
224 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) {
225 brelse(sbp);
226 return (ENOSPC);
227 }
228 sbap = (u_int *)sbp->b_data;
229 soff = idp->in_off;
230 }
231 /*
232 * If the block range spans two block maps, get the second map.
233 */
234 ebap = NULL;
235 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
236 ssize = len;
237 } else {
238 #ifdef INVARIANTS
239 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
240 panic("ext2_reallocblks: start == end");
241 #endif
242 ssize = len - (idp->in_off + 1);
243 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp))
244 goto fail;
245 ebap = (u_int *)ebp->b_data;
246 }
247 /*
248 * Find the preferred location for the cluster.
249 */
250 EXT2_LOCK(ump);
251 pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0);
252 /*
253 * Search the block map looking for an allocation of the desired size.
254 */
255 if ((newblk = (e2fs_daddr_t)ext2_hashalloc(ip, dtog(fs, pref), pref,
256 len, ext2_clusteralloc)) == 0){
257 EXT2_UNLOCK(ump);
258 goto fail;
259 }
260 /*
261 * We have found a new contiguous block.
262 *
263 * First we have to replace the old block pointers with the new
264 * block pointers in the inode and indirect blocks associated
265 * with the file.
266 */
267 #ifdef DEBUG
268 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
269 (intmax_t)start_lbn, (intmax_t)end_lbn);
270 #endif /* DEBUG */
271 blkno = newblk;
272 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
273 if (i == ssize) {
274 bap = ebap;
275 soff = -i;
276 }
277 #ifdef INVARIANTS
278 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap))
279 panic("ext2_reallocblks: alloc mismatch");
280 #endif
281 #ifdef DEBUG
282 printf(" %d,", *bap);
283 #endif /* DEBUG */
284 *bap++ = blkno;
285 }
286 /*
287 * Next we must write out the modified inode and indirect blocks.
288 * For strict correctness, the writes should be synchronous since
289 * the old block values may have been written to disk. In practise
290 * they are almost never written, but if we are concerned about
291 * strict correctness, the `doasyncfree' flag should be set to zero.
292 *
293 * The test on `doasyncfree' should be changed to test a flag
294 * that shows whether the associated buffers and inodes have
295 * been written. The flag should be set when the cluster is
296 * started and cleared whenever the buffer or inode is flushed.
297 * We can then check below to see if it is set, and do the
298 * synchronous write only when it has been cleared.
299 */
300 if (sbap != &ip->i_db[0]) {
301 if (doasyncfree)
302 bdwrite(sbp);
303 else
304 bwrite(sbp);
305 } else {
306 ip->i_flag |= IN_CHANGE | IN_UPDATE;
307 if (!doasyncfree)
308 ext2_update(vp, 1);
309 }
310 if (ssize < len) {
311 if (doasyncfree)
312 bdwrite(ebp);
313 else
314 bwrite(ebp);
315 }
316 /*
317 * Last, free the old blocks and assign the new blocks to the buffers.
318 */
319 #ifdef DEBUG
320 printf("\n\tnew:");
321 #endif /* DEBUG */
322 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
323 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
324 fs->e2fs_bsize);
325 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
326 #ifdef DEBUG
327 printf(" %d,", blkno);
328 #endif /* DEBUG */
329 }
330 #ifdef DEBUG
331 printf("\n");
332 #endif /* DEBUG */
333 return (0);
334
335 fail:
336 if (ssize < len)
337 brelse(ebp);
338 if (sbap != &ip->i_db[0])
339 brelse(sbp);
340 return (ENOSPC);
341 }
342
343 /*
344 * Allocate an inode in the filesystem.
345 *
346 */
347 int
348 ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp)
349 {
350 struct timespec ts;
351 struct inode *pip;
352 struct m_ext2fs *fs;
353 struct inode *ip;
354 struct ext2mount *ump;
355 ino_t ino, ipref;
356 int i, error, cg;
357
358 *vpp = NULL;
359 pip = VTOI(pvp);
360 fs = pip->i_e2fs;
361 ump = pip->i_ump;
362
363 EXT2_LOCK(ump);
364 if (fs->e2fs->e2fs_ficount == 0)
365 goto noinodes;
366 /*
367 * If it is a directory then obtain a cylinder group based on
368 * ext2_dirpref else obtain it using ino_to_cg. The preferred inode is
369 * always the next inode.
370 */
371 if ((mode & IFMT) == IFDIR) {
372 cg = ext2_dirpref(pip);
373 if (fs->e2fs_contigdirs[cg] < 255)
374 fs->e2fs_contigdirs[cg]++;
375 } else {
376 cg = ino_to_cg(fs, pip->i_number);
377 if (fs->e2fs_contigdirs[cg] > 0)
378 fs->e2fs_contigdirs[cg]--;
379 }
380 ipref = cg * fs->e2fs->e2fs_ipg + 1;
381 ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg);
382
383 if (ino == 0)
384 goto noinodes;
385 error = VFS_VGET(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
386 if (error) {
387 ext2_vfree(pvp, ino, mode);
388 return (error);
389 }
390 ip = VTOI(*vpp);
391
392 /*
393 * The question is whether using VGET was such good idea at all:
394 * Linux doesn't read the old inode in when it is allocating a
395 * new one. I will set at least i_size and i_blocks to zero.
396 */
397 ip->i_size = 0;
398 ip->i_blocks = 0;
399 ip->i_mode = 0;
400 ip->i_flags = 0;
401 /* now we want to make sure that the block pointers are zeroed out */
402 for (i = 0; i < NDADDR; i++)
403 ip->i_db[i] = 0;
404 for (i = 0; i < NIADDR; i++)
405 ip->i_ib[i] = 0;
406
407 /*
408 * Set up a new generation number for this inode.
409 */
410 ip->i_gen = arc4random();
411
412 vfs_timestamp(&ts);
413 ip->i_birthtime = ts.tv_sec;
414 ip->i_birthnsec = ts.tv_nsec;
415
416 /*
417 printf("ext2_valloc: allocated inode %d\n", ino);
418 */
419 return (0);
420 noinodes:
421 EXT2_UNLOCK(ump);
422 ext2_fserr(fs, cred->cr_uid, "out of inodes");
423 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->e2fs_fsmnt);
424 return (ENOSPC);
425 }
426
427 /*
428 * Find a cylinder to place a directory.
429 *
430 * The policy implemented by this algorithm is to allocate a
431 * directory inode in the same cylinder group as its parent
432 * directory, but also to reserve space for its files inodes
433 * and data. Restrict the number of directories which may be
434 * allocated one after another in the same cylinder group
435 * without intervening allocation of files.
436 *
437 * If we allocate a first level directory then force allocation
438 * in another cylinder group.
439 *
440 */
441 static u_long
442 ext2_dirpref(struct inode *pip)
443 {
444 struct m_ext2fs *fs;
445 int cg, prefcg, cgsize;
446 u_int avgifree, avgbfree, avgndir, curdirsize;
447 u_int minifree, minbfree, maxndir;
448 u_int mincg, minndir;
449 u_int dirsize, maxcontigdirs;
450
451 mtx_assert(EXT2_MTX(pip->i_ump), MA_OWNED);
452 fs = pip->i_e2fs;
453
454 avgifree = fs->e2fs->e2fs_ficount / fs->e2fs_gcount;
455 avgbfree = fs->e2fs->e2fs_fbcount / fs->e2fs_gcount;
456 avgndir = fs->e2fs_total_dir / fs->e2fs_gcount;
457
458 /*
459 * Force allocation in another cg if creating a first level dir.
460 */
461 ASSERT_VOP_LOCKED(ITOV(pip), "ext2fs_dirpref");
462 if (ITOV(pip)->v_vflag & VV_ROOT) {
463 prefcg = arc4random() % fs->e2fs_gcount;
464 mincg = prefcg;
465 minndir = fs->e2fs_ipg;
466 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
467 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir &&
468 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree &&
469 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) {
470 mincg = cg;
471 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs;
472 }
473 for (cg = 0; cg < prefcg; cg++)
474 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir &&
475 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree &&
476 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) {
477 mincg = cg;
478 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs;
479 }
480
481 return (mincg);
482 }
483
484 /*
485 * Count various limits which used for
486 * optimal allocation of a directory inode.
487 */
488 maxndir = min(avgndir + fs->e2fs_ipg / 16, fs->e2fs_ipg);
489 minifree = avgifree - avgifree / 4;
490 if (minifree < 1)
491 minifree = 1;
492 minbfree = avgbfree - avgbfree / 4;
493 if (minbfree < 1)
494 minbfree = 1;
495 cgsize = fs->e2fs_fsize * fs->e2fs_fpg;
496 dirsize = AVGDIRSIZE;
497 curdirsize = avgndir ? (cgsize - avgbfree * fs->e2fs_bsize) / avgndir : 0;
498 if (dirsize < curdirsize)
499 dirsize = curdirsize;
500 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255);
501 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR);
502 if (maxcontigdirs == 0)
503 maxcontigdirs = 1;
504
505 /*
506 * Limit number of dirs in one cg and reserve space for
507 * regular files, but only if we have no deficit in
508 * inodes or space.
509 */
510 prefcg = ino_to_cg(fs, pip->i_number);
511 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
512 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir &&
513 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree &&
514 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) {
515 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
516 return (cg);
517 }
518 for (cg = 0; cg < prefcg; cg++)
519 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir &&
520 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree &&
521 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) {
522 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
523 return (cg);
524 }
525 /*
526 * This is a backstop when we have deficit in space.
527 */
528 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
529 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree)
530 return (cg);
531 for (cg = 0; cg < prefcg; cg++)
532 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree)
533 break;
534 return (cg);
535 }
536
537 /*
538 * Select the desired position for the next block in a file.
539 *
540 * we try to mimic what Remy does in inode_getblk/block_getblk
541 *
542 * we note: blocknr == 0 means that we're about to allocate either
543 * a direct block or a pointer block at the first level of indirection
544 * (In other words, stuff that will go in i_db[] or i_ib[])
545 *
546 * blocknr != 0 means that we're allocating a block that is none
547 * of the above. Then, blocknr tells us the number of the block
548 * that will hold the pointer
549 */
550 e4fs_daddr_t
551 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, e2fs_daddr_t *bap,
552 e2fs_daddr_t blocknr)
553 {
554 int tmp;
555 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
556
557 /* if the next block is actually what we thought it is,
558 then set the goal to what we thought it should be
559 */
560 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0)
561 return ip->i_next_alloc_goal;
562
563 /* now check whether we were provided with an array that basically
564 tells us previous blocks to which we want to stay closeby
565 */
566 if (bap)
567 for (tmp = indx - 1; tmp >= 0; tmp--)
568 if (bap[tmp])
569 return bap[tmp];
570
571 /* else let's fall back to the blocknr, or, if there is none,
572 follow the rule that a block should be allocated near its inode
573 */
574 return blocknr ? blocknr :
575 (e2fs_daddr_t)(ip->i_block_group *
576 EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) +
577 ip->i_e2fs->e2fs->e2fs_first_dblock;
578 }
579
580 /*
581 * Implement the cylinder overflow algorithm.
582 *
583 * The policy implemented by this algorithm is:
584 * 1) allocate the block in its requested cylinder group.
585 * 2) quadradically rehash on the cylinder group number.
586 * 3) brute force search for a free block.
587 */
588 static u_long
589 ext2_hashalloc(struct inode *ip, int cg, long pref, int size,
590 daddr_t (*allocator)(struct inode *, int, daddr_t, int))
591 {
592 struct m_ext2fs *fs;
593 ino_t result;
594 int i, icg = cg;
595
596 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
597 fs = ip->i_e2fs;
598 /*
599 * 1: preferred cylinder group
600 */
601 result = (*allocator)(ip, cg, pref, size);
602 if (result)
603 return (result);
604 /*
605 * 2: quadratic rehash
606 */
607 for (i = 1; i < fs->e2fs_gcount; i *= 2) {
608 cg += i;
609 if (cg >= fs->e2fs_gcount)
610 cg -= fs->e2fs_gcount;
611 result = (*allocator)(ip, cg, 0, size);
612 if (result)
613 return (result);
614 }
615 /*
616 * 3: brute force search
617 * Note that we start at i == 2, since 0 was checked initially,
618 * and 1 is always checked in the quadratic rehash.
619 */
620 cg = (icg + 2) % fs->e2fs_gcount;
621 for (i = 2; i < fs->e2fs_gcount; i++) {
622 result = (*allocator)(ip, cg, 0, size);
623 if (result)
624 return (result);
625 cg++;
626 if (cg == fs->e2fs_gcount)
627 cg = 0;
628 }
629 return (0);
630 }
631
632 /*
633 * Determine whether a block can be allocated.
634 *
635 * Check to see if a block of the appropriate size is available,
636 * and if it is, allocate it.
637 */
638 static daddr_t
639 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
640 {
641 struct m_ext2fs *fs;
642 struct buf *bp;
643 struct ext2mount *ump;
644 daddr_t bno, runstart, runlen;
645 int bit, loc, end, error, start;
646 char *bbp;
647 /* XXX ondisk32 */
648 fs = ip->i_e2fs;
649 ump = ip->i_ump;
650 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0)
651 return (0);
652 EXT2_UNLOCK(ump);
653 error = bread(ip->i_devvp, fsbtodb(fs,
654 fs->e2fs_gd[cg].ext2bgd_b_bitmap),
655 (int)fs->e2fs_bsize, NOCRED, &bp);
656 if (error) {
657 brelse(bp);
658 EXT2_LOCK(ump);
659 return (0);
660 }
661 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0) {
662 /*
663 * Another thread allocated the last block in this
664 * group while we were waiting for the buffer.
665 */
666 brelse(bp);
667 EXT2_LOCK(ump);
668 return (0);
669 }
670 bbp = (char *)bp->b_data;
671
672 if (dtog(fs, bpref) != cg)
673 bpref = 0;
674 if (bpref != 0) {
675 bpref = dtogd(fs, bpref);
676 /*
677 * if the requested block is available, use it
678 */
679 if (isclr(bbp, bpref)) {
680 bno = bpref;
681 goto gotit;
682 }
683 }
684 /*
685 * no blocks in the requested cylinder, so take next
686 * available one in this cylinder group.
687 * first try to get 8 contigous blocks, then fall back to a single
688 * block.
689 */
690 if (bpref)
691 start = dtogd(fs, bpref) / NBBY;
692 else
693 start = 0;
694 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
695 retry:
696 runlen = 0;
697 runstart = 0;
698 for (loc = start; loc < end; loc++) {
699 if (bbp[loc] == (char)0xff) {
700 runlen = 0;
701 continue;
702 }
703
704 /* Start of a run, find the number of high clear bits. */
705 if (runlen == 0) {
706 bit = fls(bbp[loc]);
707 runlen = NBBY - bit;
708 runstart = loc * NBBY + bit;
709 } else if (bbp[loc] == 0) {
710 /* Continue a run. */
711 runlen += NBBY;
712 } else {
713 /*
714 * Finish the current run. If it isn't long
715 * enough, start a new one.
716 */
717 bit = ffs(bbp[loc]) - 1;
718 runlen += bit;
719 if (runlen >= 8) {
720 bno = runstart;
721 goto gotit;
722 }
723
724 /* Run was too short, start a new one. */
725 bit = fls(bbp[loc]);
726 runlen = NBBY - bit;
727 runstart = loc * NBBY + bit;
728 }
729
730 /* If the current run is long enough, use it. */
731 if (runlen >= 8) {
732 bno = runstart;
733 goto gotit;
734 }
735 }
736 if (start != 0) {
737 end = start;
738 start = 0;
739 goto retry;
740 }
741
742 bno = ext2_mapsearch(fs, bbp, bpref);
743 if (bno < 0){
744 brelse(bp);
745 EXT2_LOCK(ump);
746 return (0);
747 }
748 gotit:
749 #ifdef INVARIANTS
750 if (isset(bbp, bno)) {
751 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n",
752 cg, (intmax_t)bno, fs->e2fs_fsmnt);
753 panic("ext2fs_alloccg: dup alloc");
754 }
755 #endif
756 setbit(bbp, bno);
757 EXT2_LOCK(ump);
758 ext2_clusteracct(fs, bbp, cg, bno, -1);
759 fs->e2fs->e2fs_fbcount--;
760 fs->e2fs_gd[cg].ext2bgd_nbfree--;
761 fs->e2fs_fmod = 1;
762 EXT2_UNLOCK(ump);
763 bdwrite(bp);
764 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
765 }
766
767 /*
768 * Determine whether a cluster can be allocated.
769 */
770 static daddr_t
771 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len)
772 {
773 struct m_ext2fs *fs;
774 struct ext2mount *ump;
775 struct buf *bp;
776 char *bbp;
777 int bit, error, got, i, loc, run;
778 int32_t *lp;
779 daddr_t bno;
780
781 fs = ip->i_e2fs;
782 ump = ip->i_ump;
783
784 if (fs->e2fs_maxcluster[cg] < len)
785 return (0);
786
787 EXT2_UNLOCK(ump);
788 error = bread(ip->i_devvp,
789 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap),
790 (int)fs->e2fs_bsize, NOCRED, &bp);
791 if (error)
792 goto fail_lock;
793
794 bbp = (char *)bp->b_data;
795 bp->b_xflags |= BX_BKGRDWRITE;
796
797 EXT2_LOCK(ump);
798 /*
799 * Check to see if a cluster of the needed size (or bigger) is
800 * available in this cylinder group.
801 */
802 lp = &fs->e2fs_clustersum[cg].cs_sum[len];
803 for (i = len; i <= fs->e2fs_contigsumsize; i++)
804 if (*lp++ > 0)
805 break;
806 if (i > fs->e2fs_contigsumsize) {
807 /*
808 * Update the cluster summary information to reflect
809 * the true maximum-sized cluster so that future cluster
810 * allocation requests can avoid reading the bitmap only
811 * to find no cluster.
812 */
813 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1];
814 for (i = len - 1; i > 0; i--)
815 if (*lp-- > 0)
816 break;
817 fs->e2fs_maxcluster[cg] = i;
818 goto fail;
819 }
820 EXT2_UNLOCK(ump);
821
822 /* Search the bitmap to find a big enough cluster like in FFS. */
823 if (dtog(fs, bpref) != cg)
824 bpref = 0;
825 if (bpref != 0)
826 bpref = dtogd(fs, bpref);
827 loc = bpref / NBBY;
828 bit = 1 << (bpref % NBBY);
829 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) {
830 if ((bbp[loc] & bit) != 0)
831 run = 0;
832 else {
833 run++;
834 if (run == len)
835 break;
836 }
837 if ((got & (NBBY - 1)) != (NBBY - 1))
838 bit <<= 1;
839 else {
840 loc++;
841 bit = 1;
842 }
843 }
844
845 if (got >= fs->e2fs->e2fs_fpg)
846 goto fail_lock;
847
848 /* Allocate the cluster that we found. */
849 for (i = 1; i < len; i++)
850 if (!isclr(bbp, got - run + i))
851 panic("ext2_clusteralloc: map mismatch");
852
853 bno = got - run + 1;
854 if (bno >= fs->e2fs->e2fs_fpg)
855 panic("ext2_clusteralloc: allocated out of group");
856
857 EXT2_LOCK(ump);
858 for (i = 0; i < len; i += fs->e2fs_fpb) {
859 setbit(bbp, bno + i);
860 ext2_clusteracct(fs, bbp, cg, bno + i, -1);
861 fs->e2fs->e2fs_fbcount--;
862 fs->e2fs_gd[cg].ext2bgd_nbfree--;
863 }
864 fs->e2fs_fmod = 1;
865 EXT2_UNLOCK(ump);
866
867 bdwrite(bp);
868 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
869
870 fail_lock:
871 EXT2_LOCK(ump);
872 fail:
873 brelse(bp);
874 return (0);
875 }
876
877 /*
878 * Determine whether an inode can be allocated.
879 *
880 * Check to see if an inode is available, and if it is,
881 * allocate it using tode in the specified cylinder group.
882 */
883 static daddr_t
884 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode)
885 {
886 struct m_ext2fs *fs;
887 struct buf *bp;
888 struct ext2mount *ump;
889 int error, start, len, loc, map, i;
890 char *ibp;
891 ipref--; /* to avoid a lot of (ipref -1) */
892 if (ipref == -1)
893 ipref = 0;
894 fs = ip->i_e2fs;
895 ump = ip->i_ump;
896 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0)
897 return (0);
898 EXT2_UNLOCK(ump);
899 error = bread(ip->i_devvp, fsbtodb(fs,
900 fs->e2fs_gd[cg].ext2bgd_i_bitmap),
901 (int)fs->e2fs_bsize, NOCRED, &bp);
902 if (error) {
903 brelse(bp);
904 EXT2_LOCK(ump);
905 return (0);
906 }
907 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0) {
908 /*
909 * Another thread allocated the last i-node in this
910 * group while we were waiting for the buffer.
911 */
912 brelse(bp);
913 EXT2_LOCK(ump);
914 return (0);
915 }
916 ibp = (char *)bp->b_data;
917 if (ipref) {
918 ipref %= fs->e2fs->e2fs_ipg;
919 if (isclr(ibp, ipref))
920 goto gotit;
921 }
922 start = ipref / NBBY;
923 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY);
924 loc = skpc(0xff, len, &ibp[start]);
925 if (loc == 0) {
926 len = start + 1;
927 start = 0;
928 loc = skpc(0xff, len, &ibp[0]);
929 if (loc == 0) {
930 printf("cg = %d, ipref = %lld, fs = %s\n",
931 cg, (long long)ipref, fs->e2fs_fsmnt);
932 panic("ext2fs_nodealloccg: map corrupted");
933 /* NOTREACHED */
934 }
935 }
936 i = start + len - loc;
937 map = ibp[i] ^ 0xff;
938 if (map == 0) {
939 printf("fs = %s\n", fs->e2fs_fsmnt);
940 panic("ext2fs_nodealloccg: block not in map");
941 }
942 ipref = i * NBBY + ffs(map) - 1;
943 gotit:
944 setbit(ibp, ipref);
945 EXT2_LOCK(ump);
946 fs->e2fs_gd[cg].ext2bgd_nifree--;
947 fs->e2fs->e2fs_ficount--;
948 fs->e2fs_fmod = 1;
949 if ((mode & IFMT) == IFDIR) {
950 fs->e2fs_gd[cg].ext2bgd_ndirs++;
951 fs->e2fs_total_dir++;
952 }
953 EXT2_UNLOCK(ump);
954 bdwrite(bp);
955 return (cg * fs->e2fs->e2fs_ipg + ipref +1);
956 }
957
958 /*
959 * Free a block or fragment.
960 *
961 */
962 void
963 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size)
964 {
965 struct m_ext2fs *fs;
966 struct buf *bp;
967 struct ext2mount *ump;
968 int cg, error;
969 char *bbp;
970
971 fs = ip->i_e2fs;
972 ump = ip->i_ump;
973 cg = dtog(fs, bno);
974 if ((u_int)bno >= fs->e2fs->e2fs_bcount) {
975 printf("bad block %lld, ino %llu\n", (long long)bno,
976 (unsigned long long)ip->i_number);
977 ext2_fserr(fs, ip->i_uid, "bad block");
978 return;
979 }
980 error = bread(ip->i_devvp,
981 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap),
982 (int)fs->e2fs_bsize, NOCRED, &bp);
983 if (error) {
984 brelse(bp);
985 return;
986 }
987 bbp = (char *)bp->b_data;
988 bno = dtogd(fs, bno);
989 if (isclr(bbp, bno)) {
990 printf("block = %lld, fs = %s\n",
991 (long long)bno, fs->e2fs_fsmnt);
992 panic("ext2_blkfree: freeing free block");
993 }
994 clrbit(bbp, bno);
995 EXT2_LOCK(ump);
996 ext2_clusteracct(fs, bbp, cg, bno, 1);
997 fs->e2fs->e2fs_fbcount++;
998 fs->e2fs_gd[cg].ext2bgd_nbfree++;
999 fs->e2fs_fmod = 1;
1000 EXT2_UNLOCK(ump);
1001 bdwrite(bp);
1002 }
1003
1004 /*
1005 * Free an inode.
1006 *
1007 */
1008 int
1009 ext2_vfree(struct vnode *pvp, ino_t ino, int mode)
1010 {
1011 struct m_ext2fs *fs;
1012 struct inode *pip;
1013 struct buf *bp;
1014 struct ext2mount *ump;
1015 int error, cg;
1016 char * ibp;
1017
1018 pip = VTOI(pvp);
1019 fs = pip->i_e2fs;
1020 ump = pip->i_ump;
1021 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount)
1022 panic("ext2_vfree: range: devvp = %p, ino = %d, fs = %s",
1023 pip->i_devvp, ino, fs->e2fs_fsmnt);
1024
1025 cg = ino_to_cg(fs, ino);
1026 error = bread(pip->i_devvp,
1027 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_i_bitmap),
1028 (int)fs->e2fs_bsize, NOCRED, &bp);
1029 if (error) {
1030 brelse(bp);
1031 return (0);
1032 }
1033 ibp = (char *)bp->b_data;
1034 ino = (ino - 1) % fs->e2fs->e2fs_ipg;
1035 if (isclr(ibp, ino)) {
1036 printf("ino = %llu, fs = %s\n",
1037 (unsigned long long)ino, fs->e2fs_fsmnt);
1038 if (fs->e2fs_ronly == 0)
1039 panic("ext2_vfree: freeing free inode");
1040 }
1041 clrbit(ibp, ino);
1042 EXT2_LOCK(ump);
1043 fs->e2fs->e2fs_ficount++;
1044 fs->e2fs_gd[cg].ext2bgd_nifree++;
1045 if ((mode & IFMT) == IFDIR) {
1046 fs->e2fs_gd[cg].ext2bgd_ndirs--;
1047 fs->e2fs_total_dir--;
1048 }
1049 fs->e2fs_fmod = 1;
1050 EXT2_UNLOCK(ump);
1051 bdwrite(bp);
1052 return (0);
1053 }
1054
1055 /*
1056 * Find a block in the specified cylinder group.
1057 *
1058 * It is a panic if a request is made to find a block if none are
1059 * available.
1060 */
1061 static daddr_t
1062 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref)
1063 {
1064 int start, len, loc, i, map;
1065
1066 /*
1067 * find the fragment by searching through the free block
1068 * map for an appropriate bit pattern
1069 */
1070 if (bpref)
1071 start = dtogd(fs, bpref) / NBBY;
1072 else
1073 start = 0;
1074 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
1075 loc = skpc(0xff, len, &bbp[start]);
1076 if (loc == 0) {
1077 len = start + 1;
1078 start = 0;
1079 loc = skpc(0xff, len, &bbp[start]);
1080 if (loc == 0) {
1081 printf("start = %d, len = %d, fs = %s\n",
1082 start, len, fs->e2fs_fsmnt);
1083 panic("ext2_mapsearch: map corrupted");
1084 /* NOTREACHED */
1085 }
1086 }
1087 i = start + len - loc;
1088 map = bbp[i] ^ 0xff;
1089 if (map == 0) {
1090 printf("fs = %s\n", fs->e2fs_fsmnt);
1091 panic("ext2fs_mapsearch: block not in map");
1092 }
1093 return (i * NBBY + ffs(map) - 1);
1094 }
1095
1096 /*
1097 * Fserr prints the name of a filesystem with an error diagnostic.
1098 *
1099 * The form of the error message is:
1100 * fs: error message
1101 */
1102 static void
1103 ext2_fserr(struct m_ext2fs *fs, uid_t uid, char *cp)
1104 {
1105
1106 log(LOG_ERR, "uid %u on %s: %s\n", uid, fs->e2fs_fsmnt, cp);
1107 }
1108
1109 int
1110 cg_has_sb(int i)
1111 {
1112 int a3, a5, a7;
1113
1114 if (i == 0 || i == 1)
1115 return 1;
1116 for (a3 = 3, a5 = 5, a7 = 7;
1117 a3 <= i || a5 <= i || a7 <= i;
1118 a3 *= 3, a5 *= 5, a7 *= 7)
1119 if (i == a3 || i == a5 || i == a7)
1120 return 1;
1121 return 0;
1122 }
Cache object: 2ec2c0f3e00e822257ca8081a5f1cdc1
|