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/11.0/sys/fs/ext2fs/ext2_alloc.c 301548 2016-06-07 14:37:43Z 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, 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 %ju, lbns %jd-%jd\n\told:",
269 (uintmax_t)ip->i_number, (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_flag = 0;
398 ip->i_size = 0;
399 ip->i_blocks = 0;
400 ip->i_mode = 0;
401 ip->i_flags = 0;
402 /* now we want to make sure that the block pointers are zeroed out */
403 for (i = 0; i < NDADDR; i++)
404 ip->i_db[i] = 0;
405 for (i = 0; i < NIADDR; i++)
406 ip->i_ib[i] = 0;
407
408 /*
409 * Set up a new generation number for this inode.
410 * Avoid zero values.
411 */
412 do {
413 ip->i_gen = arc4random();
414 } while ( ip->i_gen == 0);
415
416 vfs_timestamp(&ts);
417 ip->i_birthtime = ts.tv_sec;
418 ip->i_birthnsec = ts.tv_nsec;
419
420 /*
421 printf("ext2_valloc: allocated inode %d\n", ino);
422 */
423 return (0);
424 noinodes:
425 EXT2_UNLOCK(ump);
426 ext2_fserr(fs, cred->cr_uid, "out of inodes");
427 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->e2fs_fsmnt);
428 return (ENOSPC);
429 }
430
431 /*
432 * Find a cylinder to place a directory.
433 *
434 * The policy implemented by this algorithm is to allocate a
435 * directory inode in the same cylinder group as its parent
436 * directory, but also to reserve space for its files inodes
437 * and data. Restrict the number of directories which may be
438 * allocated one after another in the same cylinder group
439 * without intervening allocation of files.
440 *
441 * If we allocate a first level directory then force allocation
442 * in another cylinder group.
443 *
444 */
445 static u_long
446 ext2_dirpref(struct inode *pip)
447 {
448 struct m_ext2fs *fs;
449 int cg, prefcg, cgsize;
450 u_int avgifree, avgbfree, avgndir, curdirsize;
451 u_int minifree, minbfree, maxndir;
452 u_int mincg, minndir;
453 u_int dirsize, maxcontigdirs;
454
455 mtx_assert(EXT2_MTX(pip->i_ump), MA_OWNED);
456 fs = pip->i_e2fs;
457
458 avgifree = fs->e2fs->e2fs_ficount / fs->e2fs_gcount;
459 avgbfree = fs->e2fs->e2fs_fbcount / fs->e2fs_gcount;
460 avgndir = fs->e2fs_total_dir / fs->e2fs_gcount;
461
462 /*
463 * Force allocation in another cg if creating a first level dir.
464 */
465 ASSERT_VOP_LOCKED(ITOV(pip), "ext2fs_dirpref");
466 if (ITOV(pip)->v_vflag & VV_ROOT) {
467 prefcg = arc4random() % fs->e2fs_gcount;
468 mincg = prefcg;
469 minndir = fs->e2fs_ipg;
470 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
471 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir &&
472 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree &&
473 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) {
474 mincg = cg;
475 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs;
476 }
477 for (cg = 0; cg < prefcg; cg++)
478 if (fs->e2fs_gd[cg].ext2bgd_ndirs < minndir &&
479 fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree &&
480 fs->e2fs_gd[cg].ext2bgd_nbfree >= avgbfree) {
481 mincg = cg;
482 minndir = fs->e2fs_gd[cg].ext2bgd_ndirs;
483 }
484
485 return (mincg);
486 }
487
488 /*
489 * Count various limits which used for
490 * optimal allocation of a directory inode.
491 */
492 maxndir = min(avgndir + fs->e2fs_ipg / 16, fs->e2fs_ipg);
493 minifree = avgifree - avgifree / 4;
494 if (minifree < 1)
495 minifree = 1;
496 minbfree = avgbfree - avgbfree / 4;
497 if (minbfree < 1)
498 minbfree = 1;
499 cgsize = fs->e2fs_fsize * fs->e2fs_fpg;
500 dirsize = AVGDIRSIZE;
501 curdirsize = avgndir ? (cgsize - avgbfree * fs->e2fs_bsize) / avgndir : 0;
502 if (dirsize < curdirsize)
503 dirsize = curdirsize;
504 maxcontigdirs = min((avgbfree * fs->e2fs_bsize) / dirsize, 255);
505 maxcontigdirs = min(maxcontigdirs, fs->e2fs_ipg / AFPDIR);
506 if (maxcontigdirs == 0)
507 maxcontigdirs = 1;
508
509 /*
510 * Limit number of dirs in one cg and reserve space for
511 * regular files, but only if we have no deficit in
512 * inodes or space.
513 */
514 prefcg = ino_to_cg(fs, pip->i_number);
515 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
516 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir &&
517 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree &&
518 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) {
519 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
520 return (cg);
521 }
522 for (cg = 0; cg < prefcg; cg++)
523 if (fs->e2fs_gd[cg].ext2bgd_ndirs < maxndir &&
524 fs->e2fs_gd[cg].ext2bgd_nifree >= minifree &&
525 fs->e2fs_gd[cg].ext2bgd_nbfree >= minbfree) {
526 if (fs->e2fs_contigdirs[cg] < maxcontigdirs)
527 return (cg);
528 }
529 /*
530 * This is a backstop when we have deficit in space.
531 */
532 for (cg = prefcg; cg < fs->e2fs_gcount; cg++)
533 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree)
534 return (cg);
535 for (cg = 0; cg < prefcg; cg++)
536 if (fs->e2fs_gd[cg].ext2bgd_nifree >= avgifree)
537 break;
538 return (cg);
539 }
540
541 /*
542 * Select the desired position for the next block in a file.
543 *
544 * we try to mimic what Remy does in inode_getblk/block_getblk
545 *
546 * we note: blocknr == 0 means that we're about to allocate either
547 * a direct block or a pointer block at the first level of indirection
548 * (In other words, stuff that will go in i_db[] or i_ib[])
549 *
550 * blocknr != 0 means that we're allocating a block that is none
551 * of the above. Then, blocknr tells us the number of the block
552 * that will hold the pointer
553 */
554 e4fs_daddr_t
555 ext2_blkpref(struct inode *ip, e2fs_lbn_t lbn, int indx, e2fs_daddr_t *bap,
556 e2fs_daddr_t blocknr)
557 {
558 int tmp;
559 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
560
561 /* if the next block is actually what we thought it is,
562 then set the goal to what we thought it should be
563 */
564 if (ip->i_next_alloc_block == lbn && ip->i_next_alloc_goal != 0)
565 return ip->i_next_alloc_goal;
566
567 /* now check whether we were provided with an array that basically
568 tells us previous blocks to which we want to stay closeby
569 */
570 if (bap)
571 for (tmp = indx - 1; tmp >= 0; tmp--)
572 if (bap[tmp])
573 return bap[tmp];
574
575 /* else let's fall back to the blocknr, or, if there is none,
576 follow the rule that a block should be allocated near its inode
577 */
578 return blocknr ? blocknr :
579 (e2fs_daddr_t)(ip->i_block_group *
580 EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) +
581 ip->i_e2fs->e2fs->e2fs_first_dblock;
582 }
583
584 /*
585 * Implement the cylinder overflow algorithm.
586 *
587 * The policy implemented by this algorithm is:
588 * 1) allocate the block in its requested cylinder group.
589 * 2) quadradically rehash on the cylinder group number.
590 * 3) brute force search for a free block.
591 */
592 static u_long
593 ext2_hashalloc(struct inode *ip, int cg, long pref, int size,
594 daddr_t (*allocator)(struct inode *, int, daddr_t, int))
595 {
596 struct m_ext2fs *fs;
597 ino_t result;
598 int i, icg = cg;
599
600 mtx_assert(EXT2_MTX(ip->i_ump), MA_OWNED);
601 fs = ip->i_e2fs;
602 /*
603 * 1: preferred cylinder group
604 */
605 result = (*allocator)(ip, cg, pref, size);
606 if (result)
607 return (result);
608 /*
609 * 2: quadratic rehash
610 */
611 for (i = 1; i < fs->e2fs_gcount; i *= 2) {
612 cg += i;
613 if (cg >= fs->e2fs_gcount)
614 cg -= fs->e2fs_gcount;
615 result = (*allocator)(ip, cg, 0, size);
616 if (result)
617 return (result);
618 }
619 /*
620 * 3: brute force search
621 * Note that we start at i == 2, since 0 was checked initially,
622 * and 1 is always checked in the quadratic rehash.
623 */
624 cg = (icg + 2) % fs->e2fs_gcount;
625 for (i = 2; i < fs->e2fs_gcount; i++) {
626 result = (*allocator)(ip, cg, 0, size);
627 if (result)
628 return (result);
629 cg++;
630 if (cg == fs->e2fs_gcount)
631 cg = 0;
632 }
633 return (0);
634 }
635
636 /*
637 * Determine whether a block can be allocated.
638 *
639 * Check to see if a block of the appropriate size is available,
640 * and if it is, allocate it.
641 */
642 static daddr_t
643 ext2_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
644 {
645 struct m_ext2fs *fs;
646 struct buf *bp;
647 struct ext2mount *ump;
648 daddr_t bno, runstart, runlen;
649 int bit, loc, end, error, start;
650 char *bbp;
651 /* XXX ondisk32 */
652 fs = ip->i_e2fs;
653 ump = ip->i_ump;
654 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0)
655 return (0);
656 EXT2_UNLOCK(ump);
657 error = bread(ip->i_devvp, fsbtodb(fs,
658 fs->e2fs_gd[cg].ext2bgd_b_bitmap),
659 (int)fs->e2fs_bsize, NOCRED, &bp);
660 if (error) {
661 brelse(bp);
662 EXT2_LOCK(ump);
663 return (0);
664 }
665 if (fs->e2fs_gd[cg].ext2bgd_nbfree == 0) {
666 /*
667 * Another thread allocated the last block in this
668 * group while we were waiting for the buffer.
669 */
670 brelse(bp);
671 EXT2_LOCK(ump);
672 return (0);
673 }
674 bbp = (char *)bp->b_data;
675
676 if (dtog(fs, bpref) != cg)
677 bpref = 0;
678 if (bpref != 0) {
679 bpref = dtogd(fs, bpref);
680 /*
681 * if the requested block is available, use it
682 */
683 if (isclr(bbp, bpref)) {
684 bno = bpref;
685 goto gotit;
686 }
687 }
688 /*
689 * no blocks in the requested cylinder, so take next
690 * available one in this cylinder group.
691 * first try to get 8 contigous blocks, then fall back to a single
692 * block.
693 */
694 if (bpref)
695 start = dtogd(fs, bpref) / NBBY;
696 else
697 start = 0;
698 end = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
699 retry:
700 runlen = 0;
701 runstart = 0;
702 for (loc = start; loc < end; loc++) {
703 if (bbp[loc] == (char)0xff) {
704 runlen = 0;
705 continue;
706 }
707
708 /* Start of a run, find the number of high clear bits. */
709 if (runlen == 0) {
710 bit = fls(bbp[loc]);
711 runlen = NBBY - bit;
712 runstart = loc * NBBY + bit;
713 } else if (bbp[loc] == 0) {
714 /* Continue a run. */
715 runlen += NBBY;
716 } else {
717 /*
718 * Finish the current run. If it isn't long
719 * enough, start a new one.
720 */
721 bit = ffs(bbp[loc]) - 1;
722 runlen += bit;
723 if (runlen >= 8) {
724 bno = runstart;
725 goto gotit;
726 }
727
728 /* Run was too short, start a new one. */
729 bit = fls(bbp[loc]);
730 runlen = NBBY - bit;
731 runstart = loc * NBBY + bit;
732 }
733
734 /* If the current run is long enough, use it. */
735 if (runlen >= 8) {
736 bno = runstart;
737 goto gotit;
738 }
739 }
740 if (start != 0) {
741 end = start;
742 start = 0;
743 goto retry;
744 }
745
746 bno = ext2_mapsearch(fs, bbp, bpref);
747 if (bno < 0){
748 brelse(bp);
749 EXT2_LOCK(ump);
750 return (0);
751 }
752 gotit:
753 #ifdef INVARIANTS
754 if (isset(bbp, bno)) {
755 printf("ext2fs_alloccgblk: cg=%d bno=%jd fs=%s\n",
756 cg, (intmax_t)bno, fs->e2fs_fsmnt);
757 panic("ext2fs_alloccg: dup alloc");
758 }
759 #endif
760 setbit(bbp, bno);
761 EXT2_LOCK(ump);
762 ext2_clusteracct(fs, bbp, cg, bno, -1);
763 fs->e2fs->e2fs_fbcount--;
764 fs->e2fs_gd[cg].ext2bgd_nbfree--;
765 fs->e2fs_fmod = 1;
766 EXT2_UNLOCK(ump);
767 bdwrite(bp);
768 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
769 }
770
771 /*
772 * Determine whether a cluster can be allocated.
773 */
774 static daddr_t
775 ext2_clusteralloc(struct inode *ip, int cg, daddr_t bpref, int len)
776 {
777 struct m_ext2fs *fs;
778 struct ext2mount *ump;
779 struct buf *bp;
780 char *bbp;
781 int bit, error, got, i, loc, run;
782 int32_t *lp;
783 daddr_t bno;
784
785 fs = ip->i_e2fs;
786 ump = ip->i_ump;
787
788 if (fs->e2fs_maxcluster[cg] < len)
789 return (0);
790
791 EXT2_UNLOCK(ump);
792 error = bread(ip->i_devvp,
793 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap),
794 (int)fs->e2fs_bsize, NOCRED, &bp);
795 if (error)
796 goto fail_lock;
797
798 bbp = (char *)bp->b_data;
799 EXT2_LOCK(ump);
800 /*
801 * Check to see if a cluster of the needed size (or bigger) is
802 * available in this cylinder group.
803 */
804 lp = &fs->e2fs_clustersum[cg].cs_sum[len];
805 for (i = len; i <= fs->e2fs_contigsumsize; i++)
806 if (*lp++ > 0)
807 break;
808 if (i > fs->e2fs_contigsumsize) {
809 /*
810 * Update the cluster summary information to reflect
811 * the true maximum-sized cluster so that future cluster
812 * allocation requests can avoid reading the bitmap only
813 * to find no cluster.
814 */
815 lp = &fs->e2fs_clustersum[cg].cs_sum[len - 1];
816 for (i = len - 1; i > 0; i--)
817 if (*lp-- > 0)
818 break;
819 fs->e2fs_maxcluster[cg] = i;
820 goto fail;
821 }
822 EXT2_UNLOCK(ump);
823
824 /* Search the bitmap to find a big enough cluster like in FFS. */
825 if (dtog(fs, bpref) != cg)
826 bpref = 0;
827 if (bpref != 0)
828 bpref = dtogd(fs, bpref);
829 loc = bpref / NBBY;
830 bit = 1 << (bpref % NBBY);
831 for (run = 0, got = bpref; got < fs->e2fs->e2fs_fpg; got++) {
832 if ((bbp[loc] & bit) != 0)
833 run = 0;
834 else {
835 run++;
836 if (run == len)
837 break;
838 }
839 if ((got & (NBBY - 1)) != (NBBY - 1))
840 bit <<= 1;
841 else {
842 loc++;
843 bit = 1;
844 }
845 }
846
847 if (got >= fs->e2fs->e2fs_fpg)
848 goto fail_lock;
849
850 /* Allocate the cluster that we found. */
851 for (i = 1; i < len; i++)
852 if (!isclr(bbp, got - run + i))
853 panic("ext2_clusteralloc: map mismatch");
854
855 bno = got - run + 1;
856 if (bno >= fs->e2fs->e2fs_fpg)
857 panic("ext2_clusteralloc: allocated out of group");
858
859 EXT2_LOCK(ump);
860 for (i = 0; i < len; i += fs->e2fs_fpb) {
861 setbit(bbp, bno + i);
862 ext2_clusteracct(fs, bbp, cg, bno + i, -1);
863 fs->e2fs->e2fs_fbcount--;
864 fs->e2fs_gd[cg].ext2bgd_nbfree--;
865 }
866 fs->e2fs_fmod = 1;
867 EXT2_UNLOCK(ump);
868
869 bdwrite(bp);
870 return (cg * fs->e2fs->e2fs_fpg + fs->e2fs->e2fs_first_dblock + bno);
871
872 fail_lock:
873 EXT2_LOCK(ump);
874 fail:
875 brelse(bp);
876 return (0);
877 }
878
879 /*
880 * Determine whether an inode can be allocated.
881 *
882 * Check to see if an inode is available, and if it is,
883 * allocate it using tode in the specified cylinder group.
884 */
885 static daddr_t
886 ext2_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode)
887 {
888 struct m_ext2fs *fs;
889 struct buf *bp;
890 struct ext2mount *ump;
891 int error, start, len;
892 char *ibp, *loc;
893 ipref--; /* to avoid a lot of (ipref -1) */
894 if (ipref == -1)
895 ipref = 0;
896 fs = ip->i_e2fs;
897 ump = ip->i_ump;
898 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0)
899 return (0);
900 EXT2_UNLOCK(ump);
901 error = bread(ip->i_devvp, fsbtodb(fs,
902 fs->e2fs_gd[cg].ext2bgd_i_bitmap),
903 (int)fs->e2fs_bsize, NOCRED, &bp);
904 if (error) {
905 brelse(bp);
906 EXT2_LOCK(ump);
907 return (0);
908 }
909 if (fs->e2fs_gd[cg].ext2bgd_nifree == 0) {
910 /*
911 * Another thread allocated the last i-node in this
912 * group while we were waiting for the buffer.
913 */
914 brelse(bp);
915 EXT2_LOCK(ump);
916 return (0);
917 }
918 ibp = (char *)bp->b_data;
919 if (ipref) {
920 ipref %= fs->e2fs->e2fs_ipg;
921 if (isclr(ibp, ipref))
922 goto gotit;
923 }
924 start = ipref / NBBY;
925 len = howmany(fs->e2fs->e2fs_ipg - ipref, NBBY);
926 loc = memcchr(&ibp[start], 0xff, len);
927 if (loc == NULL) {
928 len = start + 1;
929 start = 0;
930 loc = memcchr(&ibp[start], 0xff, len);
931 if (loc == NULL) {
932 printf("cg = %d, ipref = %lld, fs = %s\n",
933 cg, (long long)ipref, fs->e2fs_fsmnt);
934 panic("ext2fs_nodealloccg: map corrupted");
935 /* NOTREACHED */
936 }
937 }
938 ipref = (loc - ibp) * NBBY + ffs(~*loc) - 1;
939 gotit:
940 setbit(ibp, ipref);
941 EXT2_LOCK(ump);
942 fs->e2fs_gd[cg].ext2bgd_nifree--;
943 fs->e2fs->e2fs_ficount--;
944 fs->e2fs_fmod = 1;
945 if ((mode & IFMT) == IFDIR) {
946 fs->e2fs_gd[cg].ext2bgd_ndirs++;
947 fs->e2fs_total_dir++;
948 }
949 EXT2_UNLOCK(ump);
950 bdwrite(bp);
951 return (cg * fs->e2fs->e2fs_ipg + ipref +1);
952 }
953
954 /*
955 * Free a block or fragment.
956 *
957 */
958 void
959 ext2_blkfree(struct inode *ip, e4fs_daddr_t bno, long size)
960 {
961 struct m_ext2fs *fs;
962 struct buf *bp;
963 struct ext2mount *ump;
964 int cg, error;
965 char *bbp;
966
967 fs = ip->i_e2fs;
968 ump = ip->i_ump;
969 cg = dtog(fs, bno);
970 if ((u_int)bno >= fs->e2fs->e2fs_bcount) {
971 printf("bad block %lld, ino %ju\n", (long long)bno,
972 (uintmax_t)ip->i_number);
973 ext2_fserr(fs, ip->i_uid, "bad block");
974 return;
975 }
976 error = bread(ip->i_devvp,
977 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_b_bitmap),
978 (int)fs->e2fs_bsize, NOCRED, &bp);
979 if (error) {
980 brelse(bp);
981 return;
982 }
983 bbp = (char *)bp->b_data;
984 bno = dtogd(fs, bno);
985 if (isclr(bbp, bno)) {
986 printf("block = %lld, fs = %s\n",
987 (long long)bno, fs->e2fs_fsmnt);
988 panic("ext2_blkfree: freeing free block");
989 }
990 clrbit(bbp, bno);
991 EXT2_LOCK(ump);
992 ext2_clusteracct(fs, bbp, cg, bno, 1);
993 fs->e2fs->e2fs_fbcount++;
994 fs->e2fs_gd[cg].ext2bgd_nbfree++;
995 fs->e2fs_fmod = 1;
996 EXT2_UNLOCK(ump);
997 bdwrite(bp);
998 }
999
1000 /*
1001 * Free an inode.
1002 *
1003 */
1004 int
1005 ext2_vfree(struct vnode *pvp, ino_t ino, int mode)
1006 {
1007 struct m_ext2fs *fs;
1008 struct inode *pip;
1009 struct buf *bp;
1010 struct ext2mount *ump;
1011 int error, cg;
1012 char * ibp;
1013
1014 pip = VTOI(pvp);
1015 fs = pip->i_e2fs;
1016 ump = pip->i_ump;
1017 if ((u_int)ino > fs->e2fs_ipg * fs->e2fs_gcount)
1018 panic("ext2_vfree: range: devvp = %p, ino = %ju, fs = %s",
1019 pip->i_devvp, (uintmax_t)ino, fs->e2fs_fsmnt);
1020
1021 cg = ino_to_cg(fs, ino);
1022 error = bread(pip->i_devvp,
1023 fsbtodb(fs, fs->e2fs_gd[cg].ext2bgd_i_bitmap),
1024 (int)fs->e2fs_bsize, NOCRED, &bp);
1025 if (error) {
1026 brelse(bp);
1027 return (0);
1028 }
1029 ibp = (char *)bp->b_data;
1030 ino = (ino - 1) % fs->e2fs->e2fs_ipg;
1031 if (isclr(ibp, ino)) {
1032 printf("ino = %llu, fs = %s\n",
1033 (unsigned long long)ino, fs->e2fs_fsmnt);
1034 if (fs->e2fs_ronly == 0)
1035 panic("ext2_vfree: freeing free inode");
1036 }
1037 clrbit(ibp, ino);
1038 EXT2_LOCK(ump);
1039 fs->e2fs->e2fs_ficount++;
1040 fs->e2fs_gd[cg].ext2bgd_nifree++;
1041 if ((mode & IFMT) == IFDIR) {
1042 fs->e2fs_gd[cg].ext2bgd_ndirs--;
1043 fs->e2fs_total_dir--;
1044 }
1045 fs->e2fs_fmod = 1;
1046 EXT2_UNLOCK(ump);
1047 bdwrite(bp);
1048 return (0);
1049 }
1050
1051 /*
1052 * Find a block in the specified cylinder group.
1053 *
1054 * It is a panic if a request is made to find a block if none are
1055 * available.
1056 */
1057 static daddr_t
1058 ext2_mapsearch(struct m_ext2fs *fs, char *bbp, daddr_t bpref)
1059 {
1060 char *loc;
1061 int start, len;
1062
1063 /*
1064 * find the fragment by searching through the free block
1065 * map for an appropriate bit pattern
1066 */
1067 if (bpref)
1068 start = dtogd(fs, bpref) / NBBY;
1069 else
1070 start = 0;
1071 len = howmany(fs->e2fs->e2fs_fpg, NBBY) - start;
1072 loc = memcchr(&bbp[start], 0xff, len);
1073 if (loc == NULL) {
1074 len = start + 1;
1075 start = 0;
1076 loc = memcchr(&bbp[start], 0xff, len);
1077 if (loc == NULL) {
1078 printf("start = %d, len = %d, fs = %s\n",
1079 start, len, fs->e2fs_fsmnt);
1080 panic("ext2_mapsearch: map corrupted");
1081 /* NOTREACHED */
1082 }
1083 }
1084 return ((loc - bbp) * NBBY + ffs(~*loc) - 1);
1085 }
1086
1087 /*
1088 * Fserr prints the name of a filesystem with an error diagnostic.
1089 *
1090 * The form of the error message is:
1091 * fs: error message
1092 */
1093 static void
1094 ext2_fserr(struct m_ext2fs *fs, uid_t uid, char *cp)
1095 {
1096
1097 log(LOG_ERR, "uid %u on %s: %s\n", uid, fs->e2fs_fsmnt, cp);
1098 }
1099
1100 int
1101 cg_has_sb(int i)
1102 {
1103 int a3, a5, a7;
1104
1105 if (i == 0 || i == 1)
1106 return 1;
1107 for (a3 = 3, a5 = 5, a7 = 7;
1108 a3 <= i || a5 <= i || a7 <= i;
1109 a3 *= 3, a5 *= 5, a7 *= 7)
1110 if (i == a3 || i == a5 || i == a7)
1111 return 1;
1112 return 0;
1113 }
Cache object: 27d9468132e186912454029c154df6c9
|