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: releng/9.2/sys/fs/ext2fs/ext2_alloc.c 252234 2013-06-26 04:34:16Z pfg $
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, int32_t lbn, int32_t bpref, int size,
84 struct ucred *cred, int32_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 = 0;
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 = 0;
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 = 0;
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 int32_t soff, newblk, blkno;
170 int i, len, start_lvl, end_lvl, pref, ssize;
171
172 if (doreallocblks == 0)
173 return (ENOSPC);
174
175 vp = ap->a_vp;
176 ip = VTOI(vp);
177 fs = ip->i_e2fs;
178 ump = ip->i_ump;
179
180 if (fs->e2fs_contigsumsize <= 0)
181 return (ENOSPC);
182
183 buflist = ap->a_buflist;
184 len = buflist->bs_nchildren;
185 start_lbn = buflist->bs_children[0]->b_lblkno;
186 end_lbn = start_lbn + len - 1;
187 #ifdef INVARIANTS
188 for (i = 1; i < len; i++)
189 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
190 panic("ext2_reallocblks: non-cluster");
191 #endif
192 /*
193 * If the cluster crosses the boundary for the first indirect
194 * block, leave space for the indirect block. Indirect blocks
195 * are initially laid out in a position after the last direct
196 * block. Block reallocation would usually destroy locality by
197 * moving the indirect block out of the way to make room for
198 * data blocks if we didn't compensate here. We should also do
199 * this for other indirect block boundaries, but it is only
200 * important for the first one.
201 */
202 if (start_lbn < NDADDR && end_lbn >= NDADDR)
203 return (ENOSPC);
204 /*
205 * If the latest allocation is in a new cylinder group, assume that
206 * the filesystem has decided to move and do not force it back to
207 * the previous cylinder group.
208 */
209 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
210 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
211 return (ENOSPC);
212 if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
213 ext2_getlbns(vp, end_lbn, end_ap, &end_lvl))
214 return (ENOSPC);
215 /*
216 * Get the starting offset and block map for the first block.
217 */
218 if (start_lvl == 0) {
219 sbap = &ip->i_db[0];
220 soff = start_lbn;
221 } else {
222 idp = &start_ap[start_lvl - 1];
223 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) {
224 brelse(sbp);
225 return (ENOSPC);
226 }
227 sbap = (u_int *)sbp->b_data;
228 soff = idp->in_off;
229 }
230 /*
231 * If the block range spans two block maps, get the second map.
232 */
233 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
234 ssize = len;
235 } else {
236 #ifdef INVARIANTS
237 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
238 panic("ext2_reallocblks: start == end");
239 #endif
240 ssize = len - (idp->in_off + 1);
241 if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp))
242 goto fail;
243 ebap = (u_int *)ebp->b_data;
244 }
245 /*
246 * Find the preferred location for the cluster.
247 */
248 EXT2_LOCK(ump);
249 pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0);
250 /*
251 * Search the block map looking for an allocation of the desired size.
252 */
253 if ((newblk = (int32_t)ext2_hashalloc(ip, dtog(fs, pref), pref,
254 len, ext2_clusteralloc)) == 0){
255 EXT2_UNLOCK(ump);
256 goto fail;
257 }
258 /*
259 * We have found a new contiguous block.
260 *
261 * First we have to replace the old block pointers with the new
262 * block pointers in the inode and indirect blocks associated
263 * with the file.
264 */
265 #ifdef DEBUG
266 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
267 (intmax_t)start_lbn, (intmax_t)end_lbn);
268 #endif /* DEBUG */
269 blkno = newblk;
270 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
271 if (i == ssize) {
272 bap = ebap;
273 soff = -i;
274 }
275 #ifdef INVARIANTS
276 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap))
277 panic("ext2_reallocblks: alloc mismatch");
278 #endif
279 #ifdef DEBUG
280 printf(" %d,", *bap);
281 #endif /* DEBUG */
282 *bap++ = blkno;
283 }
284 /*
285 * Next we must write out the modified inode and indirect blocks.
286 * For strict correctness, the writes should be synchronous since
287 * the old block values may have been written to disk. In practise
288 * they are almost never written, but if we are concerned about
289 * strict correctness, the `doasyncfree' flag should be set to zero.
290 *
291 * The test on `doasyncfree' should be changed to test a flag
292 * that shows whether the associated buffers and inodes have
293 * been written. The flag should be set when the cluster is
294 * started and cleared whenever the buffer or inode is flushed.
295 * We can then check below to see if it is set, and do the
296 * synchronous write only when it has been cleared.
297 */
298 if (sbap != &ip->i_db[0]) {
299 if (doasyncfree)
300 bdwrite(sbp);
301 else
302 bwrite(sbp);
303 } else {
304 ip->i_flag |= IN_CHANGE | IN_UPDATE;
305 if (!doasyncfree)
306 ext2_update(vp, 1);
307 }
308 if (ssize < len) {
309 if (doasyncfree)
310 bdwrite(ebp);
311 else
312 bwrite(ebp);
313 }
314 /*
315 * Last, free the old blocks and assign the new blocks to the buffers.
316 */
317 #ifdef DEBUG
318 printf("\n\tnew:");
319 #endif /* DEBUG */
320 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
321 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
322 fs->e2fs_bsize);
323 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
324 #ifdef DEBUG
325 printf(" %d,", blkno);
326 #endif /* DEBUG */
327 }
328 #ifdef DEBUG
329 printf("\n");
330 #endif /* DEBUG */
331 return (0);
332
333 fail:
334 if (ssize < len)
335 brelse(ebp);
336 if (sbap != &ip->i_db[0])
337 brelse(sbp);
338 return (ENOSPC);
339 }
340
341 /*
342 * Allocate an inode in the filesystem.
343 *
344 */
345 int
346 ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp)
347 {
348 struct timespec ts;
349 struct inode *pip;
350 struct m_ext2fs *fs;
351 struct inode *ip;
352 struct ext2mount *ump;
353 ino_t ino, ipref;
354 int i, error, cg;
355
356 *vpp = NULL;
357 pip = VTOI(pvp);
358 fs = pip->i_e2fs;
359 ump = pip->i_ump;
360
361 EXT2_LOCK(ump);
362 if (fs->e2fs->e2fs_ficount == 0)
363 goto noinodes;
364 /*
365 * If it is a directory then obtain a cylinder group based on
366 * ext2_dirpref else obtain it using ino_to_cg. The preferred inode is
367 * always the next inode.
368 */
369 if ((mode & IFMT) == IFDIR) {
370 cg = ext2_dirpref(pip);
371 if (fs->e2fs_contigdirs[cg] < 255)
372 fs->e2fs_contigdirs[cg]++;
373 } else {
374 cg = ino_to_cg(fs, pip->i_number);
375 if (fs->e2fs_contigdirs[cg] > 0)
376 fs->e2fs_contigdirs[cg]--;
377 }
378 ipref = cg * fs->e2fs->e2fs_ipg + 1;
379 ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg);
380
381 if (ino == 0)
382 goto noinodes;
383 error = VFS_VGET(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
384 if (error) {
385 ext2_vfree(pvp, ino, mode);
386 return (error);
387 }
388 ip = VTOI(*vpp);
389
390 /*
391 * The question is whether using VGET was such good idea at all:
392 * Linux doesn't read the old inode in when it is allocating a
393 * new one. I will set at least i_size and i_blocks to zero.
394 */
395 ip->i_size = 0;
396 ip->i_blocks = 0;
397 ip->i_mode = 0;
398 ip->i_flags = 0;
399 /* now we want to make sure that the block pointers are zeroed out */
400 for (i = 0; i < NDADDR; i++)
401 ip->i_db[i] = 0;
402 for (i = 0; i < NIADDR; i++)
403 ip->i_ib[i] = 0;
404
405 /*
406 * Set up a new generation number for this inode.
407 * XXX check if this makes sense in ext2
408 */
409 if (ip->i_gen == 0 || ++ip->i_gen == 0)
410 ip->i_gen = random() / 2 + 1;
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, dirsize, cgsize;
446 u_int avgifree, avgbfree, avgndir, curdirsize;
447 u_int minifree, minbfree, maxndir;
448 u_int mincg, minndir;
449 u_int 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 if (dirsize <= 0)
501 maxcontigdirs = 0; /* dirsize overflowed */
502 else
503 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255);
504 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR);
505 if (maxcontigdirs == 0)
506 maxcontigdirs = 1;
507
508 /*
509 * Limit number of dirs in one cg and reserve space for
510 * regular files, but only if we have no deficit in
511 * inodes or space.
512 */
513 prefcg = ino_to_cg(fs, pip->i_number);
514 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
515 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir &&
516 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree &&
517 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) {
518 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
519 return (cg);
520 }
521 for (cg = 0; cg < prefcg; cg++)
522 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir &&
523 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree &&
524 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) {
525 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
526 return (cg);
527 }
528 /*
529 * This is a backstop when we have deficit in space.
530 */
531 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
532 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree)
533 return (cg);
534 for (cg = 0; cg < prefcg; cg++)
535 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree)
536 break;
537 return (cg);
538 }
539
540 /*
541 * Select the desired position for the next block in a file.
542 *
543 * we try to mimic what Remy does in inode_getblk/block_getblk
544 *
545 * we note: blocknr == 0 means that we're about to allocate either
546 * a direct block or a pointer block at the first level of indirection
547 * (In other words, stuff that will go in i_db[] or i_ib[])
548 *
549 * blocknr != 0 means that we're allocating a block that is none
550 * of the above. Then, blocknr tells us the number of the block
551 * that will hold the pointer
552 */
553 int32_t
554 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, int32_t *bap,
555 int32_t blocknr)
556 {
557 int tmp;
558 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
559
560 /* if the next block is actually what we thought it is,
561 then set the goal to what we thought it should be
562 */
563 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0)
564 return ip->i_next_alloc_goal;
565
566 /* now check whether we were provided with an array that basically
567 tells us previous blocks to which we want to stay closeby
568 */
569 if (bap)
570 for (tmp = indx - 1; tmp >= 0; tmp--)
571 if (bap[tmp])
572 return bap[tmp];
573
574 /* else let's fall back to the blocknr, or, if there is none,
575 follow the rule that a block should be allocated near its inode
576 */
577 return blocknr ? blocknr :
578 (int32_t)(ip->i_block_group *
579 EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) +
580 ip->i_e2fs->e2fs->e2fs_first_dblock;
581 }
582
583 /*
584 * Implement the cylinder overflow algorithm.
585 *
586 * The policy implemented by this algorithm is:
587 * 1) allocate the block in its requested cylinder group.
588 * 2) quadradically rehash on the cylinder group number.
589 * 3) brute force search for a free block.
590 */
591 static u_long
592 ext2_hashalloc(struct inode *ip, int cg, long pref, int size,
593 daddr_t (*allocator)(struct inode *, int, daddr_t, int))
594 {
595 struct m_ext2fs *fs;
596 ino_t result;
597 int i, icg = cg;
598
599 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
600 fs = ip->i_e2fs;
601 /*
602 * 1: preferred cylinder group
603 */
604 result = (*allocator)(ip, cg, pref, size);
605 if (result)
606 return (result);
607 /*
608 * 2: quadratic rehash
609 */
610 for (i = 1; i < fs->e2fs_gcount; i *= 2) {
611 cg += i;
612 if (cg >= fs->e2fs_gcount)
613 cg -= fs->e2fs_gcount;
614 result = (*allocator)(ip, cg, 0, size);
615 if (result)
616 return (result);
617 }
618 /*
619 * 3: brute force search
620 * Note that we start at i == 2, since 0 was checked initially,
621 * and 1 is always checked in the quadratic rehash.
622 */
623 cg = (icg + 2) % fs->e2fs_gcount;
624 for (i = 2; i < fs->e2fs_gcount; i++) {
625 result = (*allocator)(ip, cg, 0, size);
626 if (result)
627 return (result);
628 cg++;
629 if (cg == fs->e2fs_gcount)
630 cg = 0;
631 }
632 return (0);
633 }
634
635 /*
636 * Determine whether a block can be allocated.
637 *
638 * Check to see if a block of the appropriate size is available,
639 * and if it is, allocate it.
640 */
641 static daddr_t
642 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
643 {
644 struct m_ext2fs *fs;
645 struct buf *bp;
646 struct ext2mount *ump;
647 daddr_t bno, runstart, runlen;
648 int bit, loc, end, error, start;
649 char *bbp;
650 /* XXX ondisk32 */
651 fs = ip->i_e2fs;
652 ump = ip->i_ump;
653 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0)
654 return (0);
655 EXT2_UNLOCK(ump);
656 error = bread(ip->i_devvp, fsbtodb(fs,
657 fs->e2fs_gd[cg].ext2bgd_b_bitmap),
658 (int)fs->e2fs_bsize, NOCRED, &bp);
659 if (error) {
660 brelse(bp);
661 EXT2_LOCK(ump);
662 return (0);
663 }
664 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0) {
665 /*
666 * Another thread allocated the last block in this
667 * group while we were waiting for the buffer.
668 */
669 brelse(bp);
670 EXT2_LOCK(ump);
671 return (0);
672 }
673 bbp = (char *)bp->b_data;
674
675 if (dtog(fs, bpref) != cg)
676 bpref = 0;
677 if (bpref != 0) {
678 bpref = dtogd(fs, bpref);
679 /*
680 * if the requested block is available, use it
681 */
682 if (isclr(bbp, bpref)) {
683 bno = bpref;
684 goto gotit;
685 }
686 }
687 /*
688 * no blocks in the requested cylinder, so take next
689 * available one in this cylinder group.
690 * first try to get 8 contigous blocks, then fall back to a single
691 * block.
692 */
693 if (bpref)
694 start = dtogd(fs, bpref) / NBBY;
695 else
696 start = 0;
697 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
698 retry:
699 runlen = 0;
700 runstart = 0;
701 for (loc = start; loc < end; loc++) {
702 if (bbp[loc] == (char)0xff) {
703 runlen = 0;
704 continue;
705 }
706
707 /* Start of a run, find the number of high clear bits. */
708 if (runlen == 0) {
709 bit = fls(bbp[loc]);
710 runlen = NBBY - bit;
711 runstart = loc * NBBY + bit;
712 } else if (bbp[loc] == 0) {
713 /* Continue a run. */
714 runlen += NBBY;
715 } else {
716 /*
717 * Finish the current run. If it isn't long
718 * enough, start a new one.
719 */
720 bit = ffs(bbp[loc]) - 1;
721 runlen += bit;
722 if (runlen >= 8) {
723 bno = runstart;
724 goto gotit;
725 }
726
727 /* Run was too short, start a new one. */
728 bit = fls(bbp[loc]);
729 runlen = NBBY - bit;
730 runstart = loc * NBBY + bit;
731 }
732
733 /* If the current run is long enough, use it. */
734 if (runlen >= 8) {
735 bno = runstart;
736 goto gotit;
737 }
738 }
739 if (start != 0) {
740 end = start;
741 start = 0;
742 goto retry;
743 }
744
745 bno = ext2_mapsearch(fs, bbp, bpref);
746 if (bno < 0){
747 brelse(bp);
748 EXT2_LOCK(ump);
749 return (0);
750 }
751 gotit:
752 #ifdef INVARIANTS
753 if (isset(bbp, bno)) {
754 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n",
755 cg, (intmax_t)bno, fs->e2fs_fsmnt);
756 panic("ext2fs_alloccg: dup alloc");
757 }
758 #endif
759 setbit(bbp, bno);
760 EXT2_LOCK(ump);
761 ext2_clusteracct(fs, bbp, cg, bno, -1);
762 fs->e2fs->e2fs_fbcount--;
763 fs->e2fs_gd[cg].ext2bgd_nbfree--;
764 fs->e2fs_fmod = 1;
765 EXT2_UNLOCK(ump);
766 bdwrite(bp);
767 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
768 }
769
770 /*
771 * Determine whether a cluster can be allocated.
772 */
773 static daddr_t
774 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len)
775 {
776 struct m_ext2fs *fs;
777 struct ext2mount *ump;
778 struct buf *bp;
779 char *bbp;
780 int bit, error, got, i, loc, run;
781 int32_t *lp;
782 daddr_t bno;
783
784 fs = ip->i_e2fs;
785 ump = ip->i_ump;
786
787 if (fs->e2fs_maxcluster[cg] < len)
788 return (0);
789
790 EXT2_UNLOCK(ump);
791 error = bread(ip->i_devvp,
792 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap),
793 (int)fs->e2fs_bsize, NOCRED, &bp);
794 if (error)
795 goto fail_lock;
796
797 bbp = (char *)bp->b_data;
798 bp->b_xflags |= BX_BKGRDWRITE;
799
800 EXT2_LOCK(ump);
801 /*
802 * Check to see if a cluster of the needed size (or bigger) is
803 * available in this cylinder group.
804 */
805 lp = &fs->e2fs_clustersum[cg].cs_sum[len];
806 for (i = len; i <= fs->e2fs_contigsumsize; i++)
807 if (*lp++ > 0)
808 break;
809 if (i > fs->e2fs_contigsumsize) {
810 /*
811 * Update the cluster summary information to reflect
812 * the true maximum-sized cluster so that future cluster
813 * allocation requests can avoid reading the bitmap only
814 * to find no cluster.
815 */
816 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1];
817 for (i = len - 1; i > 0; i--)
818 if (*lp-- > 0)
819 break;
820 fs->e2fs_maxcluster[cg] = i;
821 goto fail;
822 }
823 EXT2_UNLOCK(ump);
824
825 /* Search the bitmap to find a big enough cluster like in FFS. */
826 if (dtog(fs, bpref) != cg)
827 bpref = 0;
828 if (bpref != 0)
829 bpref = dtogd(fs, bpref);
830 loc = bpref / NBBY;
831 bit = 1 << (bpref % NBBY);
832 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) {
833 if ((bbp[loc] & bit) != 0)
834 run = 0;
835 else {
836 run++;
837 if (run == len)
838 break;
839 }
840 if ((got & (NBBY - 1)) != (NBBY - 1))
841 bit <<= 1;
842 else {
843 loc++;
844 bit = 1;
845 }
846 }
847
848 if (got >= fs->e2fs->e2fs_fpg)
849 goto fail_lock;
850
851 /* Allocate the cluster that we found. */
852 for (i = 1; i < len; i++)
853 if (!isclr(bbp, got - run + i))
854 panic("ext2_clusteralloc: map mismatch");
855
856 bno = got - run + 1;
857 if (bno >= fs->e2fs->e2fs_fpg)
858 panic("ext2_clusteralloc: allocated out of group");
859
860 EXT2_LOCK(ump);
861 for (i = 0; i < len; i += fs->e2fs_fpb) {
862 setbit(bbp, bno + i);
863 ext2_clusteracct(fs, bbp, cg, bno + i, -1);
864 fs->e2fs->e2fs_fbcount--;
865 fs->e2fs_gd[cg].ext2bgd_nbfree--;
866 }
867 fs->e2fs_fmod = 1;
868 EXT2_UNLOCK(ump);
869
870 bdwrite(bp);
871 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
872
873 fail_lock:
874 EXT2_LOCK(ump);
875 fail:
876 brelse(bp);
877 return (0);
878 }
879
880 /*
881 * Determine whether an inode can be allocated.
882 *
883 * Check to see if an inode is available, and if it is,
884 * allocate it using tode in the specified cylinder group.
885 */
886 static daddr_t
887 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode)
888 {
889 struct m_ext2fs *fs;
890 struct buf *bp;
891 struct ext2mount *ump;
892 int error, start, len, loc, map, i;
893 char *ibp;
894 ipref--; /* to avoid a lot of (ipref -1) */
895 if (ipref == -1)
896 ipref = 0;
897 fs = ip->i_e2fs;
898 ump = ip->i_ump;
899 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0)
900 return (0);
901 EXT2_UNLOCK(ump);
902 error = bread(ip->i_devvp, fsbtodb(fs,
903 fs->e2fs_gd[cg].ext2bgd_i_bitmap),
904 (int)fs->e2fs_bsize, NOCRED, &bp);
905 if (error) {
906 brelse(bp);
907 EXT2_LOCK(ump);
908 return (0);
909 }
910 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0) {
911 /*
912 * Another thread allocated the last i-node in this
913 * group while we were waiting for the buffer.
914 */
915 brelse(bp);
916 EXT2_LOCK(ump);
917 return (0);
918 }
919 ibp = (char *)bp->b_data;
920 if (ipref) {
921 ipref %= fs->e2fs->e2fs_ipg;
922 if (isclr(ibp, ipref))
923 goto gotit;
924 }
925 start = ipref / NBBY;
926 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY);
927 loc = skpc(0xff, len, &ibp[start]);
928 if (loc == 0) {
929 len = start + 1;
930 start = 0;
931 loc = skpc(0xff, len, &ibp[0]);
932 if (loc == 0) {
933 printf("cg = %d, ipref = %lld, fs = %s\n",
934 cg, (long long)ipref, fs->e2fs_fsmnt);
935 panic("ext2fs_nodealloccg: map corrupted");
936 /* NOTREACHED */
937 }
938 }
939 i = start + len - loc;
940 map = ibp[i] ^ 0xff;
941 if (map == 0) {
942 printf("fs = %s\n", fs->e2fs_fsmnt);
943 panic("ext2fs_nodealloccg: block not in map");
944 }
945 ipref = i * NBBY + ffs(map) - 1;
946 gotit:
947 setbit(ibp, ipref);
948 EXT2_LOCK(ump);
949 fs->e2fs_gd[cg].ext2bgd_nifree--;
950 fs->e2fs->e2fs_ficount--;
951 fs->e2fs_fmod = 1;
952 if ((mode & IFMT) == IFDIR) {
953 fs->e2fs_gd[cg].ext2bgd_ndirs++;
954 fs->e2fs_total_dir++;
955 }
956 EXT2_UNLOCK(ump);
957 bdwrite(bp);
958 return (cg * fs->e2fs->e2fs_ipg + ipref +1);
959 }
960
961 /*
962 * Free a block or fragment.
963 *
964 */
965 void
966 ext2_blkfree(struct inode *ip, int32_t bno, long size)
967 {
968 struct m_ext2fs *fs;
969 struct buf *bp;
970 struct ext2mount *ump;
971 int cg, error;
972 char *bbp;
973
974 fs = ip->i_e2fs;
975 ump = ip->i_ump;
976 cg = dtog(fs, bno);
977 if ((u_int)bno >= fs->e2fs->e2fs_bcount) {
978 printf("bad block %lld, ino %llu\n", (long long)bno,
979 (unsigned long long)ip->i_number);
980 ext2_fserr(fs, ip->i_uid, "bad block");
981 return;
982 }
983 error = bread(ip->i_devvp,
984 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap),
985 (int)fs->e2fs_bsize, NOCRED, &bp);
986 if (error) {
987 brelse(bp);
988 return;
989 }
990 bbp = (char *)bp->b_data;
991 bno = dtogd(fs, bno);
992 if (isclr(bbp, bno)) {
993 printf("block = %lld, fs = %s\n",
994 (long long)bno, fs->e2fs_fsmnt);
995 panic("ext2_blkfree: freeing free block");
996 }
997 clrbit(bbp, bno);
998 EXT2_LOCK(ump);
999 ext2_clusteracct(fs, bbp, cg, bno, 1);
1000 fs->e2fs->e2fs_fbcount++;
1001 fs->e2fs_gd[cg].ext2bgd_nbfree++;
1002 fs->e2fs_fmod = 1;
1003 EXT2_UNLOCK(ump);
1004 bdwrite(bp);
1005 }
1006
1007 /*
1008 * Free an inode.
1009 *
1010 */
1011 int
1012 ext2_vfree(struct vnode *pvp, ino_t ino, int mode)
1013 {
1014 struct m_ext2fs *fs;
1015 struct inode *pip;
1016 struct buf *bp;
1017 struct ext2mount *ump;
1018 int error, cg;
1019 char * ibp;
1020
1021 pip = VTOI(pvp);
1022 fs = pip->i_e2fs;
1023 ump = pip->i_ump;
1024 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount)
1025 panic("ext2_vfree: range: devvp = %p, ino = %d, fs = %s",
1026 pip->i_devvp, ino, fs->e2fs_fsmnt);
1027
1028 cg = ino_to_cg(fs, ino);
1029 error = bread(pip->i_devvp,
1030 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_i_bitmap),
1031 (int)fs->e2fs_bsize, NOCRED, &bp);
1032 if (error) {
1033 brelse(bp);
1034 return (0);
1035 }
1036 ibp = (char *)bp->b_data;
1037 ino = (ino - 1) % fs->e2fs->e2fs_ipg;
1038 if (isclr(ibp, ino)) {
1039 printf("ino = %llu, fs = %s\n",
1040 (unsigned long long)ino, fs->e2fs_fsmnt);
1041 if (fs->e2fs_ronly == 0)
1042 panic("ext2_vfree: freeing free inode");
1043 }
1044 clrbit(ibp, ino);
1045 EXT2_LOCK(ump);
1046 fs->e2fs->e2fs_ficount++;
1047 fs->e2fs_gd[cg].ext2bgd_nifree++;
1048 if ((mode & IFMT) == IFDIR) {
1049 fs->e2fs_gd[cg].ext2bgd_ndirs--;
1050 fs->e2fs_total_dir--;
1051 }
1052 fs->e2fs_fmod = 1;
1053 EXT2_UNLOCK(ump);
1054 bdwrite(bp);
1055 return (0);
1056 }
1057
1058 /*
1059 * Find a block in the specified cylinder group.
1060 *
1061 * It is a panic if a request is made to find a block if none are
1062 * available.
1063 */
1064 static daddr_t
1065 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref)
1066 {
1067 int start, len, loc, i, map;
1068
1069 /*
1070 * find the fragment by searching through the free block
1071 * map for an appropriate bit pattern
1072 */
1073 if (bpref)
1074 start = dtogd(fs, bpref) / NBBY;
1075 else
1076 start = 0;
1077 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
1078 loc = skpc(0xff, len, &bbp[start]);
1079 if (loc == 0) {
1080 len = start + 1;
1081 start = 0;
1082 loc = skpc(0xff, len, &bbp[start]);
1083 if (loc == 0) {
1084 printf("start = %d, len = %d, fs = %s\n",
1085 start, len, fs->e2fs_fsmnt);
1086 panic("ext2_mapsearch: map corrupted");
1087 /* NOTREACHED */
1088 }
1089 }
1090 i = start + len - loc;
1091 map = bbp[i] ^ 0xff;
1092 if (map == 0) {
1093 printf("fs = %s\n", fs->e2fs_fsmnt);
1094 panic("ext2fs_mapsearch: block not in map");
1095 }
1096 return (i * NBBY + ffs(map) - 1);
1097 }
1098
1099 /*
1100 * Fserr prints the name of a filesystem with an error diagnostic.
1101 *
1102 * The form of the error message is:
1103 * fs: error message
1104 */
1105 static void
1106 ext2_fserr(struct m_ext2fs *fs, uid_t uid, char *cp)
1107 {
1108
1109 log(LOG_ERR, "uid %u on %s: %s\n", uid, fs->e2fs_fsmnt, cp);
1110 }
1111
1112 int
1113 cg_has_sb(int i)
1114 {
1115 int a3, a5, a7;
1116
1117 if (i == 0 || i == 1)
1118 return 1;
1119 for (a3 = 3, a5 = 5, a7 = 7;
1120 a3 <= i || a5 <= i || a7 <= i;
1121 a3 *= 3, a5 *= 5, a7 *= 7)
1122 if (i == a3 || i == a5 || i == a7)
1123 return 1;
1124 return 0;
1125 }
Cache object: a028d3f0eeaa866b89951e31c0fb6fa3
|