1 /*
2 * Copyright (c) 2002 Networks Associates Technology, Inc.
3 * All rights reserved.
4 *
5 * This software was developed for the FreeBSD Project by Marshall
6 * Kirk McKusick and Network Associates Laboratories, the Security
7 * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
9 * research program
10 *
11 * Copyright (c) 1982, 1986, 1989, 1993
12 * The Regents of the University of California. All rights reserved.
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. All advertising materials mentioning features or use of this software
23 * must display the following acknowledgement:
24 * This product includes software developed by the University of
25 * California, Berkeley and its contributors.
26 * 4. Neither the name of the University nor the names of its contributors
27 * may be used to endorse or promote products derived from this software
28 * without specific prior written permission.
29 *
30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * SUCH DAMAGE.
41 *
42 * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95
43 */
44
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD: releng/5.2/sys/ufs/ffs/ffs_alloc.c 121785 2003-10-31 07:25:06Z truckman $");
47
48 #include "opt_quota.h"
49
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/conf.h>
55 #include <sys/file.h>
56 #include <sys/filedesc.h>
57 #include <sys/proc.h>
58 #include <sys/vnode.h>
59 #include <sys/mount.h>
60 #include <sys/kernel.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63
64 #include <ufs/ufs/extattr.h>
65 #include <ufs/ufs/quota.h>
66 #include <ufs/ufs/inode.h>
67 #include <ufs/ufs/ufs_extern.h>
68 #include <ufs/ufs/ufsmount.h>
69
70 #include <ufs/ffs/fs.h>
71 #include <ufs/ffs/ffs_extern.h>
72
73 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, int cg, ufs2_daddr_t bpref,
74 int size);
75
76 static ufs2_daddr_t ffs_alloccg(struct inode *, int, ufs2_daddr_t, int);
77 static ufs2_daddr_t
78 ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t);
79 #ifdef DIAGNOSTIC
80 static int ffs_checkblk(struct inode *, ufs2_daddr_t, long);
81 #endif
82 static ufs2_daddr_t ffs_clusteralloc(struct inode *, int, ufs2_daddr_t, int);
83 static ino_t ffs_dirpref(struct inode *);
84 static ufs2_daddr_t ffs_fragextend(struct inode *, int, ufs2_daddr_t, int, int);
85 static void ffs_fserr(struct fs *, ino_t, char *);
86 static ufs2_daddr_t ffs_hashalloc
87 (struct inode *, int, ufs2_daddr_t, int, allocfcn_t *);
88 static ufs2_daddr_t ffs_nodealloccg(struct inode *, int, ufs2_daddr_t, int);
89 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int);
90 static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *);
91 static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *);
92
93 /*
94 * Allocate a block in the filesystem.
95 *
96 * The size of the requested block is given, which must be some
97 * multiple of fs_fsize and <= fs_bsize.
98 * A preference may be optionally specified. If a preference is given
99 * the following hierarchy is used to allocate a block:
100 * 1) allocate the requested block.
101 * 2) allocate a rotationally optimal block in the same cylinder.
102 * 3) allocate a block in the same cylinder group.
103 * 4) quadradically rehash into other cylinder groups, until an
104 * available block is located.
105 * If no block preference is given the following heirarchy is used
106 * to allocate a block:
107 * 1) allocate a block in the cylinder group that contains the
108 * inode for the file.
109 * 2) quadradically rehash into other cylinder groups, until an
110 * available block is located.
111 */
112 int
113 ffs_alloc(ip, lbn, bpref, size, cred, bnp)
114 struct inode *ip;
115 ufs2_daddr_t lbn, bpref;
116 int size;
117 struct ucred *cred;
118 ufs2_daddr_t *bnp;
119 {
120 struct fs *fs;
121 ufs2_daddr_t bno;
122 int cg, reclaimed;
123 #ifdef QUOTA
124 int error;
125 #endif
126
127 *bnp = 0;
128 fs = ip->i_fs;
129 #ifdef DIAGNOSTIC
130 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
131 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
132 devtoname(ip->i_dev), (long)fs->fs_bsize, size,
133 fs->fs_fsmnt);
134 panic("ffs_alloc: bad size");
135 }
136 if (cred == NOCRED)
137 panic("ffs_alloc: missing credential");
138 #endif /* DIAGNOSTIC */
139 reclaimed = 0;
140 retry:
141 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
142 goto nospace;
143 if (suser_cred(cred, PRISON_ROOT) &&
144 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
145 goto nospace;
146 #ifdef QUOTA
147 error = chkdq(ip, btodb(size), cred, 0);
148 if (error)
149 return (error);
150 #endif
151 if (bpref >= fs->fs_size)
152 bpref = 0;
153 if (bpref == 0)
154 cg = ino_to_cg(fs, ip->i_number);
155 else
156 cg = dtog(fs, bpref);
157 bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
158 if (bno > 0) {
159 DIP(ip, i_blocks) += btodb(size);
160 ip->i_flag |= IN_CHANGE | IN_UPDATE;
161 *bnp = bno;
162 return (0);
163 }
164 #ifdef QUOTA
165 /*
166 * Restore user's disk quota because allocation failed.
167 */
168 (void) chkdq(ip, -btodb(size), cred, FORCE);
169 #endif
170 nospace:
171 if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
172 reclaimed = 1;
173 softdep_request_cleanup(fs, ITOV(ip));
174 goto retry;
175 }
176 ffs_fserr(fs, ip->i_number, "filesystem full");
177 uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt);
178 return (ENOSPC);
179 }
180
181 /*
182 * Reallocate a fragment to a bigger size
183 *
184 * The number and size of the old block is given, and a preference
185 * and new size is also specified. The allocator attempts to extend
186 * the original block. Failing that, the regular block allocator is
187 * invoked to get an appropriate block.
188 */
189 int
190 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, cred, bpp)
191 struct inode *ip;
192 ufs2_daddr_t lbprev;
193 ufs2_daddr_t bprev;
194 ufs2_daddr_t bpref;
195 int osize, nsize;
196 struct ucred *cred;
197 struct buf **bpp;
198 {
199 struct vnode *vp;
200 struct fs *fs;
201 struct buf *bp;
202 int cg, request, error, reclaimed;
203 ufs2_daddr_t bno;
204
205 *bpp = 0;
206 vp = ITOV(ip);
207 fs = ip->i_fs;
208 #ifdef DIAGNOSTIC
209 if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
210 panic("ffs_realloccg: allocation on suspended filesystem");
211 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
212 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
213 printf(
214 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
215 devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
216 nsize, fs->fs_fsmnt);
217 panic("ffs_realloccg: bad size");
218 }
219 if (cred == NOCRED)
220 panic("ffs_realloccg: missing credential");
221 #endif /* DIAGNOSTIC */
222 reclaimed = 0;
223 retry:
224 if (suser_cred(cred, PRISON_ROOT) &&
225 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
226 goto nospace;
227 if (bprev == 0) {
228 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n",
229 devtoname(ip->i_dev), (long)fs->fs_bsize, (intmax_t)bprev,
230 fs->fs_fsmnt);
231 panic("ffs_realloccg: bad bprev");
232 }
233 /*
234 * Allocate the extra space in the buffer.
235 */
236 error = bread(vp, lbprev, osize, NOCRED, &bp);
237 if (error) {
238 brelse(bp);
239 return (error);
240 }
241
242 if (bp->b_blkno == bp->b_lblkno) {
243 if (lbprev >= NDADDR)
244 panic("ffs_realloccg: lbprev out of range");
245 bp->b_blkno = fsbtodb(fs, bprev);
246 }
247
248 #ifdef QUOTA
249 error = chkdq(ip, btodb(nsize - osize), cred, 0);
250 if (error) {
251 brelse(bp);
252 return (error);
253 }
254 #endif
255 /*
256 * Check for extension in the existing location.
257 */
258 cg = dtog(fs, bprev);
259 bno = ffs_fragextend(ip, cg, bprev, osize, nsize);
260 if (bno) {
261 if (bp->b_blkno != fsbtodb(fs, bno))
262 panic("ffs_realloccg: bad blockno");
263 DIP(ip, i_blocks) += btodb(nsize - osize);
264 ip->i_flag |= IN_CHANGE | IN_UPDATE;
265 allocbuf(bp, nsize);
266 bp->b_flags |= B_DONE;
267 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
268 *bpp = bp;
269 return (0);
270 }
271 /*
272 * Allocate a new disk location.
273 */
274 if (bpref >= fs->fs_size)
275 bpref = 0;
276 switch ((int)fs->fs_optim) {
277 case FS_OPTSPACE:
278 /*
279 * Allocate an exact sized fragment. Although this makes
280 * best use of space, we will waste time relocating it if
281 * the file continues to grow. If the fragmentation is
282 * less than half of the minimum free reserve, we choose
283 * to begin optimizing for time.
284 */
285 request = nsize;
286 if (fs->fs_minfree <= 5 ||
287 fs->fs_cstotal.cs_nffree >
288 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
289 break;
290 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
291 fs->fs_fsmnt);
292 fs->fs_optim = FS_OPTTIME;
293 break;
294 case FS_OPTTIME:
295 /*
296 * At this point we have discovered a file that is trying to
297 * grow a small fragment to a larger fragment. To save time,
298 * we allocate a full sized block, then free the unused portion.
299 * If the file continues to grow, the `ffs_fragextend' call
300 * above will be able to grow it in place without further
301 * copying. If aberrant programs cause disk fragmentation to
302 * grow within 2% of the free reserve, we choose to begin
303 * optimizing for space.
304 */
305 request = fs->fs_bsize;
306 if (fs->fs_cstotal.cs_nffree <
307 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
308 break;
309 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
310 fs->fs_fsmnt);
311 fs->fs_optim = FS_OPTSPACE;
312 break;
313 default:
314 printf("dev = %s, optim = %ld, fs = %s\n",
315 devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
316 panic("ffs_realloccg: bad optim");
317 /* NOTREACHED */
318 }
319 bno = ffs_hashalloc(ip, cg, bpref, request, ffs_alloccg);
320 if (bno > 0) {
321 bp->b_blkno = fsbtodb(fs, bno);
322 if (!DOINGSOFTDEP(vp))
323 ffs_blkfree(fs, ip->i_devvp, bprev, (long)osize,
324 ip->i_number);
325 if (nsize < request)
326 ffs_blkfree(fs, ip->i_devvp, bno + numfrags(fs, nsize),
327 (long)(request - nsize), ip->i_number);
328 DIP(ip, i_blocks) += btodb(nsize - osize);
329 ip->i_flag |= IN_CHANGE | IN_UPDATE;
330 allocbuf(bp, nsize);
331 bp->b_flags |= B_DONE;
332 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
333 *bpp = bp;
334 return (0);
335 }
336 #ifdef QUOTA
337 /*
338 * Restore user's disk quota because allocation failed.
339 */
340 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
341 #endif
342 brelse(bp);
343 nospace:
344 /*
345 * no space available
346 */
347 if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
348 reclaimed = 1;
349 softdep_request_cleanup(fs, vp);
350 goto retry;
351 }
352 ffs_fserr(fs, ip->i_number, "filesystem full");
353 uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt);
354 return (ENOSPC);
355 }
356
357 /*
358 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
359 *
360 * The vnode and an array of buffer pointers for a range of sequential
361 * logical blocks to be made contiguous is given. The allocator attempts
362 * to find a range of sequential blocks starting as close as possible
363 * from the end of the allocation for the logical block immediately
364 * preceding the current range. If successful, the physical block numbers
365 * in the buffer pointers and in the inode are changed to reflect the new
366 * allocation. If unsuccessful, the allocation is left unchanged. The
367 * success in doing the reallocation is returned. Note that the error
368 * return is not reflected back to the user. Rather the previous block
369 * allocation will be used.
370 */
371
372 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
373
374 static int doasyncfree = 1;
375 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
376
377 static int doreallocblks = 1;
378 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
379
380 #ifdef DEBUG
381 static volatile int prtrealloc = 0;
382 #endif
383
384 int
385 ffs_reallocblks(ap)
386 struct vop_reallocblks_args /* {
387 struct vnode *a_vp;
388 struct cluster_save *a_buflist;
389 } */ *ap;
390 {
391
392 if (doreallocblks == 0)
393 return (ENOSPC);
394 if (VTOI(ap->a_vp)->i_ump->um_fstype == UFS1)
395 return (ffs_reallocblks_ufs1(ap));
396 return (ffs_reallocblks_ufs2(ap));
397 }
398
399 static int
400 ffs_reallocblks_ufs1(ap)
401 struct vop_reallocblks_args /* {
402 struct vnode *a_vp;
403 struct cluster_save *a_buflist;
404 } */ *ap;
405 {
406 struct fs *fs;
407 struct inode *ip;
408 struct vnode *vp;
409 struct buf *sbp, *ebp;
410 ufs1_daddr_t *bap, *sbap, *ebap = 0;
411 struct cluster_save *buflist;
412 ufs_lbn_t start_lbn, end_lbn;
413 ufs1_daddr_t soff, newblk, blkno;
414 ufs2_daddr_t pref;
415 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
416 int i, len, start_lvl, end_lvl, ssize;
417
418 vp = ap->a_vp;
419 ip = VTOI(vp);
420 fs = ip->i_fs;
421 if (fs->fs_contigsumsize <= 0)
422 return (ENOSPC);
423 buflist = ap->a_buflist;
424 len = buflist->bs_nchildren;
425 start_lbn = buflist->bs_children[0]->b_lblkno;
426 end_lbn = start_lbn + len - 1;
427 #ifdef DIAGNOSTIC
428 for (i = 0; i < len; i++)
429 if (!ffs_checkblk(ip,
430 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
431 panic("ffs_reallocblks: unallocated block 1");
432 for (i = 1; i < len; i++)
433 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
434 panic("ffs_reallocblks: non-logical cluster");
435 blkno = buflist->bs_children[0]->b_blkno;
436 ssize = fsbtodb(fs, fs->fs_frag);
437 for (i = 1; i < len - 1; i++)
438 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
439 panic("ffs_reallocblks: non-physical cluster %d", i);
440 #endif
441 /*
442 * If the latest allocation is in a new cylinder group, assume that
443 * the filesystem has decided to move and do not force it back to
444 * the previous cylinder group.
445 */
446 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
447 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
448 return (ENOSPC);
449 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
450 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
451 return (ENOSPC);
452 /*
453 * Get the starting offset and block map for the first block.
454 */
455 if (start_lvl == 0) {
456 sbap = &ip->i_din1->di_db[0];
457 soff = start_lbn;
458 } else {
459 idp = &start_ap[start_lvl - 1];
460 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
461 brelse(sbp);
462 return (ENOSPC);
463 }
464 sbap = (ufs1_daddr_t *)sbp->b_data;
465 soff = idp->in_off;
466 }
467 /*
468 * Find the preferred location for the cluster.
469 */
470 pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap);
471 /*
472 * If the block range spans two block maps, get the second map.
473 */
474 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
475 ssize = len;
476 } else {
477 #ifdef DIAGNOSTIC
478 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
479 panic("ffs_reallocblk: start == end");
480 #endif
481 ssize = len - (idp->in_off + 1);
482 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
483 goto fail;
484 ebap = (ufs1_daddr_t *)ebp->b_data;
485 }
486 /*
487 * Search the block map looking for an allocation of the desired size.
488 */
489 if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
490 len, ffs_clusteralloc)) == 0)
491 goto fail;
492 /*
493 * We have found a new contiguous block.
494 *
495 * First we have to replace the old block pointers with the new
496 * block pointers in the inode and indirect blocks associated
497 * with the file.
498 */
499 #ifdef DEBUG
500 if (prtrealloc)
501 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
502 (intmax_t)start_lbn, (intmax_t)end_lbn);
503 #endif
504 blkno = newblk;
505 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
506 if (i == ssize) {
507 bap = ebap;
508 soff = -i;
509 }
510 #ifdef DIAGNOSTIC
511 if (!ffs_checkblk(ip,
512 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
513 panic("ffs_reallocblks: unallocated block 2");
514 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
515 panic("ffs_reallocblks: alloc mismatch");
516 #endif
517 #ifdef DEBUG
518 if (prtrealloc)
519 printf(" %d,", *bap);
520 #endif
521 if (DOINGSOFTDEP(vp)) {
522 if (sbap == &ip->i_din1->di_db[0] && i < ssize)
523 softdep_setup_allocdirect(ip, start_lbn + i,
524 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
525 buflist->bs_children[i]);
526 else
527 softdep_setup_allocindir_page(ip, start_lbn + i,
528 i < ssize ? sbp : ebp, soff + i, blkno,
529 *bap, buflist->bs_children[i]);
530 }
531 *bap++ = blkno;
532 }
533 /*
534 * Next we must write out the modified inode and indirect blocks.
535 * For strict correctness, the writes should be synchronous since
536 * the old block values may have been written to disk. In practise
537 * they are almost never written, but if we are concerned about
538 * strict correctness, the `doasyncfree' flag should be set to zero.
539 *
540 * The test on `doasyncfree' should be changed to test a flag
541 * that shows whether the associated buffers and inodes have
542 * been written. The flag should be set when the cluster is
543 * started and cleared whenever the buffer or inode is flushed.
544 * We can then check below to see if it is set, and do the
545 * synchronous write only when it has been cleared.
546 */
547 if (sbap != &ip->i_din1->di_db[0]) {
548 if (doasyncfree)
549 bdwrite(sbp);
550 else
551 bwrite(sbp);
552 } else {
553 ip->i_flag |= IN_CHANGE | IN_UPDATE;
554 if (!doasyncfree)
555 UFS_UPDATE(vp, 1);
556 }
557 if (ssize < len) {
558 if (doasyncfree)
559 bdwrite(ebp);
560 else
561 bwrite(ebp);
562 }
563 /*
564 * Last, free the old blocks and assign the new blocks to the buffers.
565 */
566 #ifdef DEBUG
567 if (prtrealloc)
568 printf("\n\tnew:");
569 #endif
570 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
571 if (!DOINGSOFTDEP(vp))
572 ffs_blkfree(fs, ip->i_devvp,
573 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
574 fs->fs_bsize, ip->i_number);
575 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
576 #ifdef DIAGNOSTIC
577 if (!ffs_checkblk(ip,
578 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
579 panic("ffs_reallocblks: unallocated block 3");
580 #endif
581 #ifdef DEBUG
582 if (prtrealloc)
583 printf(" %d,", blkno);
584 #endif
585 }
586 #ifdef DEBUG
587 if (prtrealloc) {
588 prtrealloc--;
589 printf("\n");
590 }
591 #endif
592 return (0);
593
594 fail:
595 if (ssize < len)
596 brelse(ebp);
597 if (sbap != &ip->i_din1->di_db[0])
598 brelse(sbp);
599 return (ENOSPC);
600 }
601
602 static int
603 ffs_reallocblks_ufs2(ap)
604 struct vop_reallocblks_args /* {
605 struct vnode *a_vp;
606 struct cluster_save *a_buflist;
607 } */ *ap;
608 {
609 struct fs *fs;
610 struct inode *ip;
611 struct vnode *vp;
612 struct buf *sbp, *ebp;
613 ufs2_daddr_t *bap, *sbap, *ebap = 0;
614 struct cluster_save *buflist;
615 ufs_lbn_t start_lbn, end_lbn;
616 ufs2_daddr_t soff, newblk, blkno, pref;
617 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
618 int i, len, start_lvl, end_lvl, ssize;
619
620 vp = ap->a_vp;
621 ip = VTOI(vp);
622 fs = ip->i_fs;
623 if (fs->fs_contigsumsize <= 0)
624 return (ENOSPC);
625 buflist = ap->a_buflist;
626 len = buflist->bs_nchildren;
627 start_lbn = buflist->bs_children[0]->b_lblkno;
628 end_lbn = start_lbn + len - 1;
629 #ifdef DIAGNOSTIC
630 for (i = 0; i < len; i++)
631 if (!ffs_checkblk(ip,
632 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
633 panic("ffs_reallocblks: unallocated block 1");
634 for (i = 1; i < len; i++)
635 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
636 panic("ffs_reallocblks: non-logical cluster");
637 blkno = buflist->bs_children[0]->b_blkno;
638 ssize = fsbtodb(fs, fs->fs_frag);
639 for (i = 1; i < len - 1; i++)
640 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
641 panic("ffs_reallocblks: non-physical cluster %d", i);
642 #endif
643 /*
644 * If the latest allocation is in a new cylinder group, assume that
645 * the filesystem has decided to move and do not force it back to
646 * the previous cylinder group.
647 */
648 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
649 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
650 return (ENOSPC);
651 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
652 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
653 return (ENOSPC);
654 /*
655 * Get the starting offset and block map for the first block.
656 */
657 if (start_lvl == 0) {
658 sbap = &ip->i_din2->di_db[0];
659 soff = start_lbn;
660 } else {
661 idp = &start_ap[start_lvl - 1];
662 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
663 brelse(sbp);
664 return (ENOSPC);
665 }
666 sbap = (ufs2_daddr_t *)sbp->b_data;
667 soff = idp->in_off;
668 }
669 /*
670 * Find the preferred location for the cluster.
671 */
672 pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap);
673 /*
674 * If the block range spans two block maps, get the second map.
675 */
676 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
677 ssize = len;
678 } else {
679 #ifdef DIAGNOSTIC
680 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
681 panic("ffs_reallocblk: start == end");
682 #endif
683 ssize = len - (idp->in_off + 1);
684 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
685 goto fail;
686 ebap = (ufs2_daddr_t *)ebp->b_data;
687 }
688 /*
689 * Search the block map looking for an allocation of the desired size.
690 */
691 if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
692 len, ffs_clusteralloc)) == 0)
693 goto fail;
694 /*
695 * We have found a new contiguous block.
696 *
697 * First we have to replace the old block pointers with the new
698 * block pointers in the inode and indirect blocks associated
699 * with the file.
700 */
701 #ifdef DEBUG
702 if (prtrealloc)
703 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
704 (intmax_t)start_lbn, (intmax_t)end_lbn);
705 #endif
706 blkno = newblk;
707 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
708 if (i == ssize) {
709 bap = ebap;
710 soff = -i;
711 }
712 #ifdef DIAGNOSTIC
713 if (!ffs_checkblk(ip,
714 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
715 panic("ffs_reallocblks: unallocated block 2");
716 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
717 panic("ffs_reallocblks: alloc mismatch");
718 #endif
719 #ifdef DEBUG
720 if (prtrealloc)
721 printf(" %jd,", (intmax_t)*bap);
722 #endif
723 if (DOINGSOFTDEP(vp)) {
724 if (sbap == &ip->i_din2->di_db[0] && i < ssize)
725 softdep_setup_allocdirect(ip, start_lbn + i,
726 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
727 buflist->bs_children[i]);
728 else
729 softdep_setup_allocindir_page(ip, start_lbn + i,
730 i < ssize ? sbp : ebp, soff + i, blkno,
731 *bap, buflist->bs_children[i]);
732 }
733 *bap++ = blkno;
734 }
735 /*
736 * Next we must write out the modified inode and indirect blocks.
737 * For strict correctness, the writes should be synchronous since
738 * the old block values may have been written to disk. In practise
739 * they are almost never written, but if we are concerned about
740 * strict correctness, the `doasyncfree' flag should be set to zero.
741 *
742 * The test on `doasyncfree' should be changed to test a flag
743 * that shows whether the associated buffers and inodes have
744 * been written. The flag should be set when the cluster is
745 * started and cleared whenever the buffer or inode is flushed.
746 * We can then check below to see if it is set, and do the
747 * synchronous write only when it has been cleared.
748 */
749 if (sbap != &ip->i_din2->di_db[0]) {
750 if (doasyncfree)
751 bdwrite(sbp);
752 else
753 bwrite(sbp);
754 } else {
755 ip->i_flag |= IN_CHANGE | IN_UPDATE;
756 if (!doasyncfree)
757 UFS_UPDATE(vp, 1);
758 }
759 if (ssize < len) {
760 if (doasyncfree)
761 bdwrite(ebp);
762 else
763 bwrite(ebp);
764 }
765 /*
766 * Last, free the old blocks and assign the new blocks to the buffers.
767 */
768 #ifdef DEBUG
769 if (prtrealloc)
770 printf("\n\tnew:");
771 #endif
772 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
773 if (!DOINGSOFTDEP(vp))
774 ffs_blkfree(fs, ip->i_devvp,
775 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
776 fs->fs_bsize, ip->i_number);
777 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
778 #ifdef DIAGNOSTIC
779 if (!ffs_checkblk(ip,
780 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
781 panic("ffs_reallocblks: unallocated block 3");
782 #endif
783 #ifdef DEBUG
784 if (prtrealloc)
785 printf(" %jd,", (intmax_t)blkno);
786 #endif
787 }
788 #ifdef DEBUG
789 if (prtrealloc) {
790 prtrealloc--;
791 printf("\n");
792 }
793 #endif
794 return (0);
795
796 fail:
797 if (ssize < len)
798 brelse(ebp);
799 if (sbap != &ip->i_din2->di_db[0])
800 brelse(sbp);
801 return (ENOSPC);
802 }
803
804 /*
805 * Allocate an inode in the filesystem.
806 *
807 * If allocating a directory, use ffs_dirpref to select the inode.
808 * If allocating in a directory, the following hierarchy is followed:
809 * 1) allocate the preferred inode.
810 * 2) allocate an inode in the same cylinder group.
811 * 3) quadradically rehash into other cylinder groups, until an
812 * available inode is located.
813 * If no inode preference is given the following heirarchy is used
814 * to allocate an inode:
815 * 1) allocate an inode in cylinder group 0.
816 * 2) quadradically rehash into other cylinder groups, until an
817 * available inode is located.
818 */
819 int
820 ffs_valloc(pvp, mode, cred, vpp)
821 struct vnode *pvp;
822 int mode;
823 struct ucred *cred;
824 struct vnode **vpp;
825 {
826 struct inode *pip;
827 struct fs *fs;
828 struct inode *ip;
829 struct timespec ts;
830 ino_t ino, ipref;
831 int cg, error;
832
833 *vpp = NULL;
834 pip = VTOI(pvp);
835 fs = pip->i_fs;
836 if (fs->fs_cstotal.cs_nifree == 0)
837 goto noinodes;
838
839 if ((mode & IFMT) == IFDIR)
840 ipref = ffs_dirpref(pip);
841 else
842 ipref = pip->i_number;
843 if (ipref >= fs->fs_ncg * fs->fs_ipg)
844 ipref = 0;
845 cg = ino_to_cg(fs, ipref);
846 /*
847 * Track number of dirs created one after another
848 * in a same cg without intervening by files.
849 */
850 if ((mode & IFMT) == IFDIR) {
851 if (fs->fs_contigdirs[cg] < 255)
852 fs->fs_contigdirs[cg]++;
853 } else {
854 if (fs->fs_contigdirs[cg] > 0)
855 fs->fs_contigdirs[cg]--;
856 }
857 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode,
858 (allocfcn_t *)ffs_nodealloccg);
859 if (ino == 0)
860 goto noinodes;
861 error = VFS_VGET(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
862 if (error) {
863 UFS_VFREE(pvp, ino, mode);
864 return (error);
865 }
866 ip = VTOI(*vpp);
867 if (ip->i_mode) {
868 printf("mode = 0%o, inum = %lu, fs = %s\n",
869 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
870 panic("ffs_valloc: dup alloc");
871 }
872 if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */
873 printf("free inode %s/%lu had %ld blocks\n",
874 fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks));
875 DIP(ip, i_blocks) = 0;
876 }
877 ip->i_flags = 0;
878 DIP(ip, i_flags) = 0;
879 /*
880 * Set up a new generation number for this inode.
881 */
882 if (ip->i_gen == 0 || ++ip->i_gen == 0)
883 ip->i_gen = arc4random() / 2 + 1;
884 DIP(ip, i_gen) = ip->i_gen;
885 if (fs->fs_magic == FS_UFS2_MAGIC) {
886 vfs_timestamp(&ts);
887 ip->i_din2->di_birthtime = ts.tv_sec;
888 ip->i_din2->di_birthnsec = ts.tv_nsec;
889 }
890 return (0);
891 noinodes:
892 ffs_fserr(fs, pip->i_number, "out of inodes");
893 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
894 return (ENOSPC);
895 }
896
897 /*
898 * Find a cylinder group to place a directory.
899 *
900 * The policy implemented by this algorithm is to allocate a
901 * directory inode in the same cylinder group as its parent
902 * directory, but also to reserve space for its files inodes
903 * and data. Restrict the number of directories which may be
904 * allocated one after another in the same cylinder group
905 * without intervening allocation of files.
906 *
907 * If we allocate a first level directory then force allocation
908 * in another cylinder group.
909 */
910 static ino_t
911 ffs_dirpref(pip)
912 struct inode *pip;
913 {
914 struct fs *fs;
915 int cg, prefcg, dirsize, cgsize;
916 int avgifree, avgbfree, avgndir, curdirsize;
917 int minifree, minbfree, maxndir;
918 int mincg, minndir;
919 int maxcontigdirs;
920
921 fs = pip->i_fs;
922
923 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
924 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
925 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
926
927 /*
928 * Force allocation in another cg if creating a first level dir.
929 */
930 ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref");
931 if (ITOV(pip)->v_vflag & VV_ROOT) {
932 prefcg = arc4random() % fs->fs_ncg;
933 mincg = prefcg;
934 minndir = fs->fs_ipg;
935 for (cg = prefcg; cg < fs->fs_ncg; cg++)
936 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
937 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
938 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
939 mincg = cg;
940 minndir = fs->fs_cs(fs, cg).cs_ndir;
941 }
942 for (cg = 0; cg < prefcg; cg++)
943 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
944 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
945 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
946 mincg = cg;
947 minndir = fs->fs_cs(fs, cg).cs_ndir;
948 }
949 return ((ino_t)(fs->fs_ipg * mincg));
950 }
951
952 /*
953 * Count various limits which used for
954 * optimal allocation of a directory inode.
955 */
956 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
957 minifree = avgifree - avgifree / 4;
958 if (minifree < 1)
959 minifree = 1;
960 minbfree = avgbfree - avgbfree / 4;
961 if (minbfree < 1)
962 minbfree = 1;
963 cgsize = fs->fs_fsize * fs->fs_fpg;
964 dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir;
965 curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
966 if (dirsize < curdirsize)
967 dirsize = curdirsize;
968 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
969 if (fs->fs_avgfpdir > 0)
970 maxcontigdirs = min(maxcontigdirs,
971 fs->fs_ipg / fs->fs_avgfpdir);
972 if (maxcontigdirs == 0)
973 maxcontigdirs = 1;
974
975 /*
976 * Limit number of dirs in one cg and reserve space for
977 * regular files, but only if we have no deficit in
978 * inodes or space.
979 */
980 prefcg = ino_to_cg(fs, pip->i_number);
981 for (cg = prefcg; cg < fs->fs_ncg; cg++)
982 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
983 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
984 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
985 if (fs->fs_contigdirs[cg] < maxcontigdirs)
986 return ((ino_t)(fs->fs_ipg * cg));
987 }
988 for (cg = 0; cg < prefcg; cg++)
989 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
990 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
991 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
992 if (fs->fs_contigdirs[cg] < maxcontigdirs)
993 return ((ino_t)(fs->fs_ipg * cg));
994 }
995 /*
996 * This is a backstop when we have deficit in space.
997 */
998 for (cg = prefcg; cg < fs->fs_ncg; cg++)
999 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
1000 return ((ino_t)(fs->fs_ipg * cg));
1001 for (cg = 0; cg < prefcg; cg++)
1002 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
1003 break;
1004 return ((ino_t)(fs->fs_ipg * cg));
1005 }
1006
1007 /*
1008 * Select the desired position for the next block in a file. The file is
1009 * logically divided into sections. The first section is composed of the
1010 * direct blocks. Each additional section contains fs_maxbpg blocks.
1011 *
1012 * If no blocks have been allocated in the first section, the policy is to
1013 * request a block in the same cylinder group as the inode that describes
1014 * the file. If no blocks have been allocated in any other section, the
1015 * policy is to place the section in a cylinder group with a greater than
1016 * average number of free blocks. An appropriate cylinder group is found
1017 * by using a rotor that sweeps the cylinder groups. When a new group of
1018 * blocks is needed, the sweep begins in the cylinder group following the
1019 * cylinder group from which the previous allocation was made. The sweep
1020 * continues until a cylinder group with greater than the average number
1021 * of free blocks is found. If the allocation is for the first block in an
1022 * indirect block, the information on the previous allocation is unavailable;
1023 * here a best guess is made based upon the logical block number being
1024 * allocated.
1025 *
1026 * If a section is already partially allocated, the policy is to
1027 * contiguously allocate fs_maxcontig blocks. The end of one of these
1028 * contiguous blocks and the beginning of the next is laid out
1029 * contiguously if possible.
1030 */
1031 ufs2_daddr_t
1032 ffs_blkpref_ufs1(ip, lbn, indx, bap)
1033 struct inode *ip;
1034 ufs_lbn_t lbn;
1035 int indx;
1036 ufs1_daddr_t *bap;
1037 {
1038 struct fs *fs;
1039 int cg;
1040 int avgbfree, startcg;
1041
1042 fs = ip->i_fs;
1043 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
1044 if (lbn < NDADDR + NINDIR(fs)) {
1045 cg = ino_to_cg(fs, ip->i_number);
1046 return (fs->fs_fpg * cg + fs->fs_frag);
1047 }
1048 /*
1049 * Find a cylinder with greater than average number of
1050 * unused data blocks.
1051 */
1052 if (indx == 0 || bap[indx - 1] == 0)
1053 startcg =
1054 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
1055 else
1056 startcg = dtog(fs, bap[indx - 1]) + 1;
1057 startcg %= fs->fs_ncg;
1058 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1059 for (cg = startcg; cg < fs->fs_ncg; cg++)
1060 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1061 fs->fs_cgrotor = cg;
1062 return (fs->fs_fpg * cg + fs->fs_frag);
1063 }
1064 for (cg = 0; cg <= startcg; cg++)
1065 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1066 fs->fs_cgrotor = cg;
1067 return (fs->fs_fpg * cg + fs->fs_frag);
1068 }
1069 return (0);
1070 }
1071 /*
1072 * We just always try to lay things out contiguously.
1073 */
1074 return (bap[indx - 1] + fs->fs_frag);
1075 }
1076
1077 /*
1078 * Same as above, but for UFS2
1079 */
1080 ufs2_daddr_t
1081 ffs_blkpref_ufs2(ip, lbn, indx, bap)
1082 struct inode *ip;
1083 ufs_lbn_t lbn;
1084 int indx;
1085 ufs2_daddr_t *bap;
1086 {
1087 struct fs *fs;
1088 int cg;
1089 int avgbfree, startcg;
1090
1091 fs = ip->i_fs;
1092 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
1093 if (lbn < NDADDR + NINDIR(fs)) {
1094 cg = ino_to_cg(fs, ip->i_number);
1095 return (fs->fs_fpg * cg + fs->fs_frag);
1096 }
1097 /*
1098 * Find a cylinder with greater than average number of
1099 * unused data blocks.
1100 */
1101 if (indx == 0 || bap[indx - 1] == 0)
1102 startcg =
1103 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
1104 else
1105 startcg = dtog(fs, bap[indx - 1]) + 1;
1106 startcg %= fs->fs_ncg;
1107 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1108 for (cg = startcg; cg < fs->fs_ncg; cg++)
1109 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1110 fs->fs_cgrotor = cg;
1111 return (fs->fs_fpg * cg + fs->fs_frag);
1112 }
1113 for (cg = 0; cg <= startcg; cg++)
1114 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1115 fs->fs_cgrotor = cg;
1116 return (fs->fs_fpg * cg + fs->fs_frag);
1117 }
1118 return (0);
1119 }
1120 /*
1121 * We just always try to lay things out contiguously.
1122 */
1123 return (bap[indx - 1] + fs->fs_frag);
1124 }
1125
1126 /*
1127 * Implement the cylinder overflow algorithm.
1128 *
1129 * The policy implemented by this algorithm is:
1130 * 1) allocate the block in its requested cylinder group.
1131 * 2) quadradically rehash on the cylinder group number.
1132 * 3) brute force search for a free block.
1133 */
1134 /*VARARGS5*/
1135 static ufs2_daddr_t
1136 ffs_hashalloc(ip, cg, pref, size, allocator)
1137 struct inode *ip;
1138 int cg;
1139 ufs2_daddr_t pref;
1140 int size; /* size for data blocks, mode for inodes */
1141 allocfcn_t *allocator;
1142 {
1143 struct fs *fs;
1144 ufs2_daddr_t result;
1145 int i, icg = cg;
1146
1147 #ifdef DIAGNOSTIC
1148 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
1149 panic("ffs_hashalloc: allocation on suspended filesystem");
1150 #endif
1151 fs = ip->i_fs;
1152 /*
1153 * 1: preferred cylinder group
1154 */
1155 result = (*allocator)(ip, cg, pref, size);
1156 if (result)
1157 return (result);
1158 /*
1159 * 2: quadratic rehash
1160 */
1161 for (i = 1; i < fs->fs_ncg; i *= 2) {
1162 cg += i;
1163 if (cg >= fs->fs_ncg)
1164 cg -= fs->fs_ncg;
1165 result = (*allocator)(ip, cg, 0, size);
1166 if (result)
1167 return (result);
1168 }
1169 /*
1170 * 3: brute force search
1171 * Note that we start at i == 2, since 0 was checked initially,
1172 * and 1 is always checked in the quadratic rehash.
1173 */
1174 cg = (icg + 2) % fs->fs_ncg;
1175 for (i = 2; i < fs->fs_ncg; i++) {
1176 result = (*allocator)(ip, cg, 0, size);
1177 if (result)
1178 return (result);
1179 cg++;
1180 if (cg == fs->fs_ncg)
1181 cg = 0;
1182 }
1183 return (0);
1184 }
1185
1186 /*
1187 * Determine whether a fragment can be extended.
1188 *
1189 * Check to see if the necessary fragments are available, and
1190 * if they are, allocate them.
1191 */
1192 static ufs2_daddr_t
1193 ffs_fragextend(ip, cg, bprev, osize, nsize)
1194 struct inode *ip;
1195 int cg;
1196 ufs2_daddr_t bprev;
1197 int osize, nsize;
1198 {
1199 struct fs *fs;
1200 struct cg *cgp;
1201 struct buf *bp;
1202 long bno;
1203 int frags, bbase;
1204 int i, error;
1205 u_int8_t *blksfree;
1206
1207 fs = ip->i_fs;
1208 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
1209 return (0);
1210 frags = numfrags(fs, nsize);
1211 bbase = fragnum(fs, bprev);
1212 if (bbase > fragnum(fs, (bprev + frags - 1))) {
1213 /* cannot extend across a block boundary */
1214 return (0);
1215 }
1216 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1217 (int)fs->fs_cgsize, NOCRED, &bp);
1218 if (error) {
1219 brelse(bp);
1220 return (0);
1221 }
1222 cgp = (struct cg *)bp->b_data;
1223 if (!cg_chkmagic(cgp)) {
1224 brelse(bp);
1225 return (0);
1226 }
1227 bp->b_xflags |= BX_BKGRDWRITE;
1228 cgp->cg_old_time = cgp->cg_time = time_second;
1229 bno = dtogd(fs, bprev);
1230 blksfree = cg_blksfree(cgp);
1231 for (i = numfrags(fs, osize); i < frags; i++)
1232 if (isclr(blksfree, bno + i)) {
1233 brelse(bp);
1234 return (0);
1235 }
1236 /*
1237 * the current fragment can be extended
1238 * deduct the count on fragment being extended into
1239 * increase the count on the remaining fragment (if any)
1240 * allocate the extended piece
1241 */
1242 for (i = frags; i < fs->fs_frag - bbase; i++)
1243 if (isclr(blksfree, bno + i))
1244 break;
1245 cgp->cg_frsum[i - numfrags(fs, osize)]--;
1246 if (i != frags)
1247 cgp->cg_frsum[i - frags]++;
1248 for (i = numfrags(fs, osize); i < frags; i++) {
1249 clrbit(blksfree, bno + i);
1250 cgp->cg_cs.cs_nffree--;
1251 fs->fs_cstotal.cs_nffree--;
1252 fs->fs_cs(fs, cg).cs_nffree--;
1253 }
1254 fs->fs_fmod = 1;
1255 if (DOINGSOFTDEP(ITOV(ip)))
1256 softdep_setup_blkmapdep(bp, fs, bprev);
1257 if (fs->fs_active != 0)
1258 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1259 bdwrite(bp);
1260 return (bprev);
1261 }
1262
1263 /*
1264 * Determine whether a block can be allocated.
1265 *
1266 * Check to see if a block of the appropriate size is available,
1267 * and if it is, allocate it.
1268 */
1269 static ufs2_daddr_t
1270 ffs_alloccg(ip, cg, bpref, size)
1271 struct inode *ip;
1272 int cg;
1273 ufs2_daddr_t bpref;
1274 int size;
1275 {
1276 struct fs *fs;
1277 struct cg *cgp;
1278 struct buf *bp;
1279 ufs1_daddr_t bno;
1280 ufs2_daddr_t blkno;
1281 int i, allocsiz, error, frags;
1282 u_int8_t *blksfree;
1283
1284 fs = ip->i_fs;
1285 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
1286 return (0);
1287 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1288 (int)fs->fs_cgsize, NOCRED, &bp);
1289 if (error) {
1290 brelse(bp);
1291 return (0);
1292 }
1293 cgp = (struct cg *)bp->b_data;
1294 if (!cg_chkmagic(cgp) ||
1295 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
1296 brelse(bp);
1297 return (0);
1298 }
1299 bp->b_xflags |= BX_BKGRDWRITE;
1300 cgp->cg_old_time = cgp->cg_time = time_second;
1301 if (size == fs->fs_bsize) {
1302 blkno = ffs_alloccgblk(ip, bp, bpref);
1303 if (fs->fs_active != 0)
1304 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1305 bdwrite(bp);
1306 return (blkno);
1307 }
1308 /*
1309 * check to see if any fragments are already available
1310 * allocsiz is the size which will be allocated, hacking
1311 * it down to a smaller size if necessary
1312 */
1313 blksfree = cg_blksfree(cgp);
1314 frags = numfrags(fs, size);
1315 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
1316 if (cgp->cg_frsum[allocsiz] != 0)
1317 break;
1318 if (allocsiz == fs->fs_frag) {
1319 /*
1320 * no fragments were available, so a block will be
1321 * allocated, and hacked up
1322 */
1323 if (cgp->cg_cs.cs_nbfree == 0) {
1324 brelse(bp);
1325 return (0);
1326 }
1327 blkno = ffs_alloccgblk(ip, bp, bpref);
1328 bno = dtogd(fs, blkno);
1329 for (i = frags; i < fs->fs_frag; i++)
1330 setbit(blksfree, bno + i);
1331 i = fs->fs_frag - frags;
1332 cgp->cg_cs.cs_nffree += i;
1333 fs->fs_cstotal.cs_nffree += i;
1334 fs->fs_cs(fs, cg).cs_nffree += i;
1335 fs->fs_fmod = 1;
1336 cgp->cg_frsum[i]++;
1337 if (fs->fs_active != 0)
1338 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1339 bdwrite(bp);
1340 return (blkno);
1341 }
1342 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
1343 if (bno < 0) {
1344 brelse(bp);
1345 return (0);
1346 }
1347 for (i = 0; i < frags; i++)
1348 clrbit(blksfree, bno + i);
1349 cgp->cg_cs.cs_nffree -= frags;
1350 fs->fs_cstotal.cs_nffree -= frags;
1351 fs->fs_cs(fs, cg).cs_nffree -= frags;
1352 fs->fs_fmod = 1;
1353 cgp->cg_frsum[allocsiz]--;
1354 if (frags != allocsiz)
1355 cgp->cg_frsum[allocsiz - frags]++;
1356 blkno = cg * fs->fs_fpg + bno;
1357 if (DOINGSOFTDEP(ITOV(ip)))
1358 softdep_setup_blkmapdep(bp, fs, blkno);
1359 if (fs->fs_active != 0)
1360 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1361 bdwrite(bp);
1362 return (blkno);
1363 }
1364
1365 /*
1366 * Allocate a block in a cylinder group.
1367 *
1368 * This algorithm implements the following policy:
1369 * 1) allocate the requested block.
1370 * 2) allocate a rotationally optimal block in the same cylinder.
1371 * 3) allocate the next available block on the block rotor for the
1372 * specified cylinder group.
1373 * Note that this routine only allocates fs_bsize blocks; these
1374 * blocks may be fragmented by the routine that allocates them.
1375 */
1376 static ufs2_daddr_t
1377 ffs_alloccgblk(ip, bp, bpref)
1378 struct inode *ip;
1379 struct buf *bp;
1380 ufs2_daddr_t bpref;
1381 {
1382 struct fs *fs;
1383 struct cg *cgp;
1384 ufs1_daddr_t bno;
1385 ufs2_daddr_t blkno;
1386 u_int8_t *blksfree;
1387
1388 fs = ip->i_fs;
1389 cgp = (struct cg *)bp->b_data;
1390 blksfree = cg_blksfree(cgp);
1391 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
1392 bpref = cgp->cg_rotor;
1393 } else {
1394 bpref = blknum(fs, bpref);
1395 bno = dtogd(fs, bpref);
1396 /*
1397 * if the requested block is available, use it
1398 */
1399 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
1400 goto gotit;
1401 }
1402 /*
1403 * Take the next available block in this cylinder group.
1404 */
1405 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1406 if (bno < 0)
1407 return (0);
1408 cgp->cg_rotor = bno;
1409 gotit:
1410 blkno = fragstoblks(fs, bno);
1411 ffs_clrblock(fs, blksfree, (long)blkno);
1412 ffs_clusteracct(fs, cgp, blkno, -1);
1413 cgp->cg_cs.cs_nbfree--;
1414 fs->fs_cstotal.cs_nbfree--;
1415 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1416 fs->fs_fmod = 1;
1417 blkno = cgp->cg_cgx * fs->fs_fpg + bno;
1418 if (DOINGSOFTDEP(ITOV(ip)))
1419 softdep_setup_blkmapdep(bp, fs, blkno);
1420 return (blkno);
1421 }
1422
1423 /*
1424 * Determine whether a cluster can be allocated.
1425 *
1426 * We do not currently check for optimal rotational layout if there
1427 * are multiple choices in the same cylinder group. Instead we just
1428 * take the first one that we find following bpref.
1429 */
1430 static ufs2_daddr_t
1431 ffs_clusteralloc(ip, cg, bpref, len)
1432 struct inode *ip;
1433 int cg;
1434 ufs2_daddr_t bpref;
1435 int len;
1436 {
1437 struct fs *fs;
1438 struct cg *cgp;
1439 struct buf *bp;
1440 int i, run, bit, map, got;
1441 ufs2_daddr_t bno;
1442 u_char *mapp;
1443 int32_t *lp;
1444 u_int8_t *blksfree;
1445
1446 fs = ip->i_fs;
1447 if (fs->fs_maxcluster[cg] < len)
1448 return (0);
1449 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1450 NOCRED, &bp))
1451 goto fail;
1452 cgp = (struct cg *)bp->b_data;
1453 if (!cg_chkmagic(cgp))
1454 goto fail;
1455 bp->b_xflags |= BX_BKGRDWRITE;
1456 /*
1457 * Check to see if a cluster of the needed size (or bigger) is
1458 * available in this cylinder group.
1459 */
1460 lp = &cg_clustersum(cgp)[len];
1461 for (i = len; i <= fs->fs_contigsumsize; i++)
1462 if (*lp++ > 0)
1463 break;
1464 if (i > fs->fs_contigsumsize) {
1465 /*
1466 * This is the first time looking for a cluster in this
1467 * cylinder group. Update the cluster summary information
1468 * to reflect the true maximum sized cluster so that
1469 * future cluster allocation requests can avoid reading
1470 * the cylinder group map only to find no clusters.
1471 */
1472 lp = &cg_clustersum(cgp)[len - 1];
1473 for (i = len - 1; i > 0; i--)
1474 if (*lp-- > 0)
1475 break;
1476 fs->fs_maxcluster[cg] = i;
1477 goto fail;
1478 }
1479 /*
1480 * Search the cluster map to find a big enough cluster.
1481 * We take the first one that we find, even if it is larger
1482 * than we need as we prefer to get one close to the previous
1483 * block allocation. We do not search before the current
1484 * preference point as we do not want to allocate a block
1485 * that is allocated before the previous one (as we will
1486 * then have to wait for another pass of the elevator
1487 * algorithm before it will be read). We prefer to fail and
1488 * be recalled to try an allocation in the next cylinder group.
1489 */
1490 if (dtog(fs, bpref) != cg)
1491 bpref = 0;
1492 else
1493 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1494 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1495 map = *mapp++;
1496 bit = 1 << (bpref % NBBY);
1497 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1498 if ((map & bit) == 0) {
1499 run = 0;
1500 } else {
1501 run++;
1502 if (run == len)
1503 break;
1504 }
1505 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1506 bit <<= 1;
1507 } else {
1508 map = *mapp++;
1509 bit = 1;
1510 }
1511 }
1512 if (got >= cgp->cg_nclusterblks)
1513 goto fail;
1514 /*
1515 * Allocate the cluster that we have found.
1516 */
1517 blksfree = cg_blksfree(cgp);
1518 for (i = 1; i <= len; i++)
1519 if (!ffs_isblock(fs, blksfree, got - run + i))
1520 panic("ffs_clusteralloc: map mismatch");
1521 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1522 if (dtog(fs, bno) != cg)
1523 panic("ffs_clusteralloc: allocated out of group");
1524 len = blkstofrags(fs, len);
1525 for (i = 0; i < len; i += fs->fs_frag)
1526 if (ffs_alloccgblk(ip, bp, bno + i) != bno + i)
1527 panic("ffs_clusteralloc: lost block");
1528 if (fs->fs_active != 0)
1529 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1530 bdwrite(bp);
1531 return (bno);
1532
1533 fail:
1534 brelse(bp);
1535 return (0);
1536 }
1537
1538 /*
1539 * Determine whether an inode can be allocated.
1540 *
1541 * Check to see if an inode is available, and if it is,
1542 * allocate it using the following policy:
1543 * 1) allocate the requested inode.
1544 * 2) allocate the next available inode after the requested
1545 * inode in the specified cylinder group.
1546 */
1547 static ufs2_daddr_t
1548 ffs_nodealloccg(ip, cg, ipref, mode)
1549 struct inode *ip;
1550 int cg;
1551 ufs2_daddr_t ipref;
1552 int mode;
1553 {
1554 struct fs *fs;
1555 struct cg *cgp;
1556 struct buf *bp, *ibp;
1557 u_int8_t *inosused;
1558 struct ufs2_dinode *dp2;
1559 int error, start, len, loc, map, i;
1560
1561 fs = ip->i_fs;
1562 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1563 return (0);
1564 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1565 (int)fs->fs_cgsize, NOCRED, &bp);
1566 if (error) {
1567 brelse(bp);
1568 return (0);
1569 }
1570 cgp = (struct cg *)bp->b_data;
1571 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1572 brelse(bp);
1573 return (0);
1574 }
1575 bp->b_xflags |= BX_BKGRDWRITE;
1576 cgp->cg_old_time = cgp->cg_time = time_second;
1577 inosused = cg_inosused(cgp);
1578 if (ipref) {
1579 ipref %= fs->fs_ipg;
1580 if (isclr(inosused, ipref))
1581 goto gotit;
1582 }
1583 start = cgp->cg_irotor / NBBY;
1584 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1585 loc = skpc(0xff, len, &inosused[start]);
1586 if (loc == 0) {
1587 len = start + 1;
1588 start = 0;
1589 loc = skpc(0xff, len, &inosused[0]);
1590 if (loc == 0) {
1591 printf("cg = %d, irotor = %ld, fs = %s\n",
1592 cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
1593 panic("ffs_nodealloccg: map corrupted");
1594 /* NOTREACHED */
1595 }
1596 }
1597 i = start + len - loc;
1598 map = inosused[i];
1599 ipref = i * NBBY;
1600 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1601 if ((map & i) == 0) {
1602 cgp->cg_irotor = ipref;
1603 goto gotit;
1604 }
1605 }
1606 printf("fs = %s\n", fs->fs_fsmnt);
1607 panic("ffs_nodealloccg: block not in map");
1608 /* NOTREACHED */
1609 gotit:
1610 if (DOINGSOFTDEP(ITOV(ip)))
1611 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
1612 setbit(inosused, ipref);
1613 cgp->cg_cs.cs_nifree--;
1614 fs->fs_cstotal.cs_nifree--;
1615 fs->fs_cs(fs, cg).cs_nifree--;
1616 fs->fs_fmod = 1;
1617 if ((mode & IFMT) == IFDIR) {
1618 cgp->cg_cs.cs_ndir++;
1619 fs->fs_cstotal.cs_ndir++;
1620 fs->fs_cs(fs, cg).cs_ndir++;
1621 }
1622 /*
1623 * Check to see if we need to initialize more inodes.
1624 */
1625 if (fs->fs_magic == FS_UFS2_MAGIC &&
1626 ipref + INOPB(fs) > cgp->cg_initediblk &&
1627 cgp->cg_initediblk < cgp->cg_niblk) {
1628 ibp = getblk(ip->i_devvp, fsbtodb(fs,
1629 ino_to_fsba(fs, cg * fs->fs_ipg + cgp->cg_initediblk)),
1630 (int)fs->fs_bsize, 0, 0, 0);
1631 bzero(ibp->b_data, (int)fs->fs_bsize);
1632 dp2 = (struct ufs2_dinode *)(ibp->b_data);
1633 for (i = 0; i < INOPB(fs); i++) {
1634 dp2->di_gen = arc4random() / 2 + 1;
1635 dp2++;
1636 }
1637 bawrite(ibp);
1638 cgp->cg_initediblk += INOPB(fs);
1639 }
1640 if (fs->fs_active != 0)
1641 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1642 bdwrite(bp);
1643 return (cg * fs->fs_ipg + ipref);
1644 }
1645
1646 /*
1647 * check if a block is free
1648 */
1649 static int
1650 ffs_isfreeblock(struct fs *fs, u_char *cp, ufs1_daddr_t h)
1651 {
1652
1653 switch ((int)fs->fs_frag) {
1654 case 8:
1655 return (cp[h] == 0);
1656 case 4:
1657 return ((cp[h >> 1] & (0x0f << ((h & 0x1) << 2))) == 0);
1658 case 2:
1659 return ((cp[h >> 2] & (0x03 << ((h & 0x3) << 1))) == 0);
1660 case 1:
1661 return ((cp[h >> 3] & (0x01 << (h & 0x7))) == 0);
1662 default:
1663 panic("ffs_isfreeblock");
1664 }
1665 return (0);
1666 }
1667
1668 /*
1669 * Free a block or fragment.
1670 *
1671 * The specified block or fragment is placed back in the
1672 * free map. If a fragment is deallocated, a possible
1673 * block reassembly is checked.
1674 */
1675 void
1676 ffs_blkfree(fs, devvp, bno, size, inum)
1677 struct fs *fs;
1678 struct vnode *devvp;
1679 ufs2_daddr_t bno;
1680 long size;
1681 ino_t inum;
1682 {
1683 struct cg *cgp;
1684 struct buf *bp;
1685 ufs1_daddr_t fragno, cgbno;
1686 ufs2_daddr_t cgblkno;
1687 int i, cg, blk, frags, bbase;
1688 u_int8_t *blksfree;
1689 dev_t dev;
1690
1691 cg = dtog(fs, bno);
1692 if (devvp->v_type != VCHR) {
1693 /* devvp is a snapshot */
1694 dev = VTOI(devvp)->i_devvp->v_rdev;
1695 cgblkno = fragstoblks(fs, cgtod(fs, cg));
1696 } else {
1697 /* devvp is a normal disk device */
1698 dev = devvp->v_rdev;
1699 cgblkno = fsbtodb(fs, cgtod(fs, cg));
1700 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree");
1701 if ((devvp->v_vflag & VV_COPYONWRITE) &&
1702 ffs_snapblkfree(fs, devvp, bno, size, inum))
1703 return;
1704 VOP_FREEBLKS(devvp, fsbtodb(fs, bno), size);
1705 }
1706 #ifdef DIAGNOSTIC
1707 if (dev->si_mountpoint &&
1708 (dev->si_mountpoint->mnt_kern_flag & MNTK_SUSPENDED))
1709 panic("ffs_blkfree: deallocation on suspended filesystem");
1710 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
1711 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
1712 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n",
1713 devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize,
1714 size, fs->fs_fsmnt);
1715 panic("ffs_blkfree: bad size");
1716 }
1717 #endif
1718 if ((u_int)bno >= fs->fs_size) {
1719 printf("bad block %jd, ino %lu\n", (intmax_t)bno,
1720 (u_long)inum);
1721 ffs_fserr(fs, inum, "bad block");
1722 return;
1723 }
1724 if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) {
1725 brelse(bp);
1726 return;
1727 }
1728 cgp = (struct cg *)bp->b_data;
1729 if (!cg_chkmagic(cgp)) {
1730 brelse(bp);
1731 return;
1732 }
1733 bp->b_xflags |= BX_BKGRDWRITE;
1734 cgp->cg_old_time = cgp->cg_time = time_second;
1735 cgbno = dtogd(fs, bno);
1736 blksfree = cg_blksfree(cgp);
1737 if (size == fs->fs_bsize) {
1738 fragno = fragstoblks(fs, cgbno);
1739 if (!ffs_isfreeblock(fs, blksfree, fragno)) {
1740 if (devvp->v_type != VCHR) {
1741 /* devvp is a snapshot */
1742 brelse(bp);
1743 return;
1744 }
1745 printf("dev = %s, block = %jd, fs = %s\n",
1746 devtoname(dev), (intmax_t)bno, fs->fs_fsmnt);
1747 panic("ffs_blkfree: freeing free block");
1748 }
1749 ffs_setblock(fs, blksfree, fragno);
1750 ffs_clusteracct(fs, cgp, fragno, 1);
1751 cgp->cg_cs.cs_nbfree++;
1752 fs->fs_cstotal.cs_nbfree++;
1753 fs->fs_cs(fs, cg).cs_nbfree++;
1754 } else {
1755 bbase = cgbno - fragnum(fs, cgbno);
1756 /*
1757 * decrement the counts associated with the old frags
1758 */
1759 blk = blkmap(fs, blksfree, bbase);
1760 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1761 /*
1762 * deallocate the fragment
1763 */
1764 frags = numfrags(fs, size);
1765 for (i = 0; i < frags; i++) {
1766 if (isset(blksfree, cgbno + i)) {
1767 printf("dev = %s, block = %jd, fs = %s\n",
1768 devtoname(dev), (intmax_t)(bno + i),
1769 fs->fs_fsmnt);
1770 panic("ffs_blkfree: freeing free frag");
1771 }
1772 setbit(blksfree, cgbno + i);
1773 }
1774 cgp->cg_cs.cs_nffree += i;
1775 fs->fs_cstotal.cs_nffree += i;
1776 fs->fs_cs(fs, cg).cs_nffree += i;
1777 /*
1778 * add back in counts associated with the new frags
1779 */
1780 blk = blkmap(fs, blksfree, bbase);
1781 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1782 /*
1783 * if a complete block has been reassembled, account for it
1784 */
1785 fragno = fragstoblks(fs, bbase);
1786 if (ffs_isblock(fs, blksfree, fragno)) {
1787 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1788 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1789 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1790 ffs_clusteracct(fs, cgp, fragno, 1);
1791 cgp->cg_cs.cs_nbfree++;
1792 fs->fs_cstotal.cs_nbfree++;
1793 fs->fs_cs(fs, cg).cs_nbfree++;
1794 }
1795 }
1796 fs->fs_fmod = 1;
1797 if (fs->fs_active != 0)
1798 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1799 bdwrite(bp);
1800 }
1801
1802 #ifdef DIAGNOSTIC
1803 /*
1804 * Verify allocation of a block or fragment. Returns true if block or
1805 * fragment is allocated, false if it is free.
1806 */
1807 static int
1808 ffs_checkblk(ip, bno, size)
1809 struct inode *ip;
1810 ufs2_daddr_t bno;
1811 long size;
1812 {
1813 struct fs *fs;
1814 struct cg *cgp;
1815 struct buf *bp;
1816 ufs1_daddr_t cgbno;
1817 int i, error, frags, free;
1818 u_int8_t *blksfree;
1819
1820 fs = ip->i_fs;
1821 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1822 printf("bsize = %ld, size = %ld, fs = %s\n",
1823 (long)fs->fs_bsize, size, fs->fs_fsmnt);
1824 panic("ffs_checkblk: bad size");
1825 }
1826 if ((u_int)bno >= fs->fs_size)
1827 panic("ffs_checkblk: bad block %jd", (intmax_t)bno);
1828 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1829 (int)fs->fs_cgsize, NOCRED, &bp);
1830 if (error)
1831 panic("ffs_checkblk: cg bread failed");
1832 cgp = (struct cg *)bp->b_data;
1833 if (!cg_chkmagic(cgp))
1834 panic("ffs_checkblk: cg magic mismatch");
1835 bp->b_xflags |= BX_BKGRDWRITE;
1836 blksfree = cg_blksfree(cgp);
1837 cgbno = dtogd(fs, bno);
1838 if (size == fs->fs_bsize) {
1839 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno));
1840 } else {
1841 frags = numfrags(fs, size);
1842 for (free = 0, i = 0; i < frags; i++)
1843 if (isset(blksfree, cgbno + i))
1844 free++;
1845 if (free != 0 && free != frags)
1846 panic("ffs_checkblk: partially free fragment");
1847 }
1848 brelse(bp);
1849 return (!free);
1850 }
1851 #endif /* DIAGNOSTIC */
1852
1853 /*
1854 * Free an inode.
1855 */
1856 int
1857 ffs_vfree(pvp, ino, mode)
1858 struct vnode *pvp;
1859 ino_t ino;
1860 int mode;
1861 {
1862 if (DOINGSOFTDEP(pvp)) {
1863 softdep_freefile(pvp, ino, mode);
1864 return (0);
1865 }
1866 return (ffs_freefile(VTOI(pvp)->i_fs, VTOI(pvp)->i_devvp, ino, mode));
1867 }
1868
1869 /*
1870 * Do the actual free operation.
1871 * The specified inode is placed back in the free map.
1872 */
1873 int
1874 ffs_freefile(fs, devvp, ino, mode)
1875 struct fs *fs;
1876 struct vnode *devvp;
1877 ino_t ino;
1878 int mode;
1879 {
1880 struct cg *cgp;
1881 struct buf *bp;
1882 ufs2_daddr_t cgbno;
1883 int error, cg;
1884 u_int8_t *inosused;
1885 dev_t dev;
1886
1887 cg = ino_to_cg(fs, ino);
1888 if (devvp->v_type != VCHR) {
1889 /* devvp is a snapshot */
1890 dev = VTOI(devvp)->i_devvp->v_rdev;
1891 cgbno = fragstoblks(fs, cgtod(fs, cg));
1892 } else {
1893 /* devvp is a normal disk device */
1894 dev = devvp->v_rdev;
1895 cgbno = fsbtodb(fs, cgtod(fs, cg));
1896 }
1897 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1898 panic("ffs_freefile: range: dev = %s, ino = %lu, fs = %s",
1899 devtoname(dev), (u_long)ino, fs->fs_fsmnt);
1900 if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) {
1901 brelse(bp);
1902 return (error);
1903 }
1904 cgp = (struct cg *)bp->b_data;
1905 if (!cg_chkmagic(cgp)) {
1906 brelse(bp);
1907 return (0);
1908 }
1909 bp->b_xflags |= BX_BKGRDWRITE;
1910 cgp->cg_old_time = cgp->cg_time = time_second;
1911 inosused = cg_inosused(cgp);
1912 ino %= fs->fs_ipg;
1913 if (isclr(inosused, ino)) {
1914 printf("dev = %s, ino = %lu, fs = %s\n", devtoname(dev),
1915 (u_long)ino + cg * fs->fs_ipg, fs->fs_fsmnt);
1916 if (fs->fs_ronly == 0)
1917 panic("ffs_freefile: freeing free inode");
1918 }
1919 clrbit(inosused, ino);
1920 if (ino < cgp->cg_irotor)
1921 cgp->cg_irotor = ino;
1922 cgp->cg_cs.cs_nifree++;
1923 fs->fs_cstotal.cs_nifree++;
1924 fs->fs_cs(fs, cg).cs_nifree++;
1925 if ((mode & IFMT) == IFDIR) {
1926 cgp->cg_cs.cs_ndir--;
1927 fs->fs_cstotal.cs_ndir--;
1928 fs->fs_cs(fs, cg).cs_ndir--;
1929 }
1930 fs->fs_fmod = 1;
1931 if (fs->fs_active != 0)
1932 atomic_clear_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
1933 bdwrite(bp);
1934 return (0);
1935 }
1936
1937 /*
1938 * Check to see if a file is free.
1939 */
1940 int
1941 ffs_checkfreefile(fs, devvp, ino)
1942 struct fs *fs;
1943 struct vnode *devvp;
1944 ino_t ino;
1945 {
1946 struct cg *cgp;
1947 struct buf *bp;
1948 ufs2_daddr_t cgbno;
1949 int ret, cg;
1950 u_int8_t *inosused;
1951
1952 cg = ino_to_cg(fs, ino);
1953 if (devvp->v_type != VCHR) {
1954 /* devvp is a snapshot */
1955 cgbno = fragstoblks(fs, cgtod(fs, cg));
1956 } else {
1957 /* devvp is a normal disk device */
1958 cgbno = fsbtodb(fs, cgtod(fs, cg));
1959 }
1960 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1961 return (1);
1962 if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) {
1963 brelse(bp);
1964 return (1);
1965 }
1966 cgp = (struct cg *)bp->b_data;
1967 if (!cg_chkmagic(cgp)) {
1968 brelse(bp);
1969 return (1);
1970 }
1971 inosused = cg_inosused(cgp);
1972 ino %= fs->fs_ipg;
1973 ret = isclr(inosused, ino);
1974 brelse(bp);
1975 return (ret);
1976 }
1977
1978 /*
1979 * Find a block of the specified size in the specified cylinder group.
1980 *
1981 * It is a panic if a request is made to find a block if none are
1982 * available.
1983 */
1984 static ufs1_daddr_t
1985 ffs_mapsearch(fs, cgp, bpref, allocsiz)
1986 struct fs *fs;
1987 struct cg *cgp;
1988 ufs2_daddr_t bpref;
1989 int allocsiz;
1990 {
1991 ufs1_daddr_t bno;
1992 int start, len, loc, i;
1993 int blk, field, subfield, pos;
1994 u_int8_t *blksfree;
1995
1996 /*
1997 * find the fragment by searching through the free block
1998 * map for an appropriate bit pattern
1999 */
2000 if (bpref)
2001 start = dtogd(fs, bpref) / NBBY;
2002 else
2003 start = cgp->cg_frotor / NBBY;
2004 blksfree = cg_blksfree(cgp);
2005 len = howmany(fs->fs_fpg, NBBY) - start;
2006 loc = scanc((u_int)len, (u_char *)&blksfree[start],
2007 (u_char *)fragtbl[fs->fs_frag],
2008 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
2009 if (loc == 0) {
2010 len = start + 1;
2011 start = 0;
2012 loc = scanc((u_int)len, (u_char *)&blksfree[0],
2013 (u_char *)fragtbl[fs->fs_frag],
2014 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
2015 if (loc == 0) {
2016 printf("start = %d, len = %d, fs = %s\n",
2017 start, len, fs->fs_fsmnt);
2018 panic("ffs_alloccg: map corrupted");
2019 /* NOTREACHED */
2020 }
2021 }
2022 bno = (start + len - loc) * NBBY;
2023 cgp->cg_frotor = bno;
2024 /*
2025 * found the byte in the map
2026 * sift through the bits to find the selected frag
2027 */
2028 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
2029 blk = blkmap(fs, blksfree, bno);
2030 blk <<= 1;
2031 field = around[allocsiz];
2032 subfield = inside[allocsiz];
2033 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
2034 if ((blk & field) == subfield)
2035 return (bno + pos);
2036 field <<= 1;
2037 subfield <<= 1;
2038 }
2039 }
2040 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
2041 panic("ffs_alloccg: block not in map");
2042 return (-1);
2043 }
2044
2045 /*
2046 * Update the cluster map because of an allocation or free.
2047 *
2048 * Cnt == 1 means free; cnt == -1 means allocating.
2049 */
2050 void
2051 ffs_clusteracct(fs, cgp, blkno, cnt)
2052 struct fs *fs;
2053 struct cg *cgp;
2054 ufs1_daddr_t blkno;
2055 int cnt;
2056 {
2057 int32_t *sump;
2058 int32_t *lp;
2059 u_char *freemapp, *mapp;
2060 int i, start, end, forw, back, map, bit;
2061
2062 if (fs->fs_contigsumsize <= 0)
2063 return;
2064 freemapp = cg_clustersfree(cgp);
2065 sump = cg_clustersum(cgp);
2066 /*
2067 * Allocate or clear the actual block.
2068 */
2069 if (cnt > 0)
2070 setbit(freemapp, blkno);
2071 else
2072 clrbit(freemapp, blkno);
2073 /*
2074 * Find the size of the cluster going forward.
2075 */
2076 start = blkno + 1;
2077 end = start + fs->fs_contigsumsize;
2078 if (end >= cgp->cg_nclusterblks)
2079 end = cgp->cg_nclusterblks;
2080 mapp = &freemapp[start / NBBY];
2081 map = *mapp++;
2082 bit = 1 << (start % NBBY);
2083 for (i = start; i < end; i++) {
2084 if ((map & bit) == 0)
2085 break;
2086 if ((i & (NBBY - 1)) != (NBBY - 1)) {
2087 bit <<= 1;
2088 } else {
2089 map = *mapp++;
2090 bit = 1;
2091 }
2092 }
2093 forw = i - start;
2094 /*
2095 * Find the size of the cluster going backward.
2096 */
2097 start = blkno - 1;
2098 end = start - fs->fs_contigsumsize;
2099 if (end < 0)
2100 end = -1;
2101 mapp = &freemapp[start / NBBY];
2102 map = *mapp--;
2103 bit = 1 << (start % NBBY);
2104 for (i = start; i > end; i--) {
2105 if ((map & bit) == 0)
2106 break;
2107 if ((i & (NBBY - 1)) != 0) {
2108 bit >>= 1;
2109 } else {
2110 map = *mapp--;
2111 bit = 1 << (NBBY - 1);
2112 }
2113 }
2114 back = start - i;
2115 /*
2116 * Account for old cluster and the possibly new forward and
2117 * back clusters.
2118 */
2119 i = back + forw + 1;
2120 if (i > fs->fs_contigsumsize)
2121 i = fs->fs_contigsumsize;
2122 sump[i] += cnt;
2123 if (back > 0)
2124 sump[back] -= cnt;
2125 if (forw > 0)
2126 sump[forw] -= cnt;
2127 /*
2128 * Update cluster summary information.
2129 */
2130 lp = &sump[fs->fs_contigsumsize];
2131 for (i = fs->fs_contigsumsize; i > 0; i--)
2132 if (*lp-- > 0)
2133 break;
2134 fs->fs_maxcluster[cgp->cg_cgx] = i;
2135 }
2136
2137 /*
2138 * Fserr prints the name of a filesystem with an error diagnostic.
2139 *
2140 * The form of the error message is:
2141 * fs: error message
2142 */
2143 static void
2144 ffs_fserr(fs, inum, cp)
2145 struct fs *fs;
2146 ino_t inum;
2147 char *cp;
2148 {
2149 struct thread *td = curthread; /* XXX */
2150 struct proc *p = td->td_proc;
2151
2152 log(LOG_ERR, "pid %d (%s), uid %d inumber %d on %s: %s\n",
2153 p->p_pid, p->p_comm, td->td_ucred->cr_uid, inum, fs->fs_fsmnt, cp);
2154 }
2155
2156 /*
2157 * This function provides the capability for the fsck program to
2158 * update an active filesystem. Six operations are provided:
2159 *
2160 * adjrefcnt(inode, amt) - adjusts the reference count on the
2161 * specified inode by the specified amount. Under normal
2162 * operation the count should always go down. Decrementing
2163 * the count to zero will cause the inode to be freed.
2164 * adjblkcnt(inode, amt) - adjust the number of blocks used to
2165 * by the specifed amount.
2166 * freedirs(inode, count) - directory inodes [inode..inode + count - 1]
2167 * are marked as free. Inodes should never have to be marked
2168 * as in use.
2169 * freefiles(inode, count) - file inodes [inode..inode + count - 1]
2170 * are marked as free. Inodes should never have to be marked
2171 * as in use.
2172 * freeblks(blockno, size) - blocks [blockno..blockno + size - 1]
2173 * are marked as free. Blocks should never have to be marked
2174 * as in use.
2175 * setflags(flags, set/clear) - the fs_flags field has the specified
2176 * flags set (second parameter +1) or cleared (second parameter -1).
2177 */
2178
2179 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS);
2180
2181 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT,
2182 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count");
2183
2184 SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR,
2185 sysctl_ffs_fsck, "Adjust Inode Used Blocks Count");
2186
2187 SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR,
2188 sysctl_ffs_fsck, "Free Range of Directory Inodes");
2189
2190 SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR,
2191 sysctl_ffs_fsck, "Free Range of File Inodes");
2192
2193 SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR,
2194 sysctl_ffs_fsck, "Free Range of Blocks");
2195
2196 SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR,
2197 sysctl_ffs_fsck, "Change Filesystem Flags");
2198
2199 #ifdef DEBUG
2200 static int fsckcmds = 0;
2201 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "");
2202 #endif /* DEBUG */
2203
2204 static int
2205 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS)
2206 {
2207 struct fsck_cmd cmd;
2208 struct ufsmount *ump;
2209 struct vnode *vp;
2210 struct inode *ip;
2211 struct mount *mp;
2212 struct fs *fs;
2213 ufs2_daddr_t blkno;
2214 long blkcnt, blksize;
2215 struct file *fp;
2216 int filetype, error;
2217
2218 if (req->newlen > sizeof cmd)
2219 return (EBADRPC);
2220 if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0)
2221 return (error);
2222 if (cmd.version != FFS_CMD_VERSION)
2223 return (ERPCMISMATCH);
2224 if ((error = getvnode(curproc->p_fd, cmd.handle, &fp)) != 0)
2225 return (error);
2226 vn_start_write(fp->f_data, &mp, V_WAIT);
2227 if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) {
2228 vn_finished_write(mp);
2229 fdrop(fp, curthread);
2230 return (EINVAL);
2231 }
2232 if (mp->mnt_flag & MNT_RDONLY) {
2233 vn_finished_write(mp);
2234 fdrop(fp, curthread);
2235 return (EROFS);
2236 }
2237 ump = VFSTOUFS(mp);
2238 fs = ump->um_fs;
2239 filetype = IFREG;
2240
2241 switch (oidp->oid_number) {
2242
2243 case FFS_SET_FLAGS:
2244 #ifdef DEBUG
2245 if (fsckcmds)
2246 printf("%s: %s flags\n", mp->mnt_stat.f_mntonname,
2247 cmd.size > 0 ? "set" : "clear");
2248 #endif /* DEBUG */
2249 if (cmd.size > 0)
2250 fs->fs_flags |= (long)cmd.value;
2251 else
2252 fs->fs_flags &= ~(long)cmd.value;
2253 break;
2254
2255 case FFS_ADJ_REFCNT:
2256 #ifdef DEBUG
2257 if (fsckcmds) {
2258 printf("%s: adjust inode %jd count by %jd\n",
2259 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2260 (intmax_t)cmd.size);
2261 }
2262 #endif /* DEBUG */
2263 if ((error = VFS_VGET(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
2264 break;
2265 ip = VTOI(vp);
2266 ip->i_nlink += cmd.size;
2267 DIP(ip, i_nlink) = ip->i_nlink;
2268 ip->i_effnlink += cmd.size;
2269 ip->i_flag |= IN_CHANGE;
2270 if (DOINGSOFTDEP(vp))
2271 softdep_change_linkcnt(ip);
2272 vput(vp);
2273 break;
2274
2275 case FFS_ADJ_BLKCNT:
2276 #ifdef DEBUG
2277 if (fsckcmds) {
2278 printf("%s: adjust inode %jd block count by %jd\n",
2279 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2280 (intmax_t)cmd.size);
2281 }
2282 #endif /* DEBUG */
2283 if ((error = VFS_VGET(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
2284 break;
2285 ip = VTOI(vp);
2286 DIP(ip, i_blocks) += cmd.size;
2287 ip->i_flag |= IN_CHANGE;
2288 vput(vp);
2289 break;
2290
2291 case FFS_DIR_FREE:
2292 filetype = IFDIR;
2293 /* fall through */
2294
2295 case FFS_FILE_FREE:
2296 #ifdef DEBUG
2297 if (fsckcmds) {
2298 if (cmd.size == 1)
2299 printf("%s: free %s inode %d\n",
2300 mp->mnt_stat.f_mntonname,
2301 filetype == IFDIR ? "directory" : "file",
2302 (ino_t)cmd.value);
2303 else
2304 printf("%s: free %s inodes %d-%d\n",
2305 mp->mnt_stat.f_mntonname,
2306 filetype == IFDIR ? "directory" : "file",
2307 (ino_t)cmd.value,
2308 (ino_t)(cmd.value + cmd.size - 1));
2309 }
2310 #endif /* DEBUG */
2311 while (cmd.size > 0) {
2312 if ((error = ffs_freefile(fs, ump->um_devvp, cmd.value,
2313 filetype)))
2314 break;
2315 cmd.size -= 1;
2316 cmd.value += 1;
2317 }
2318 break;
2319
2320 case FFS_BLK_FREE:
2321 #ifdef DEBUG
2322 if (fsckcmds) {
2323 if (cmd.size == 1)
2324 printf("%s: free block %jd\n",
2325 mp->mnt_stat.f_mntonname,
2326 (intmax_t)cmd.value);
2327 else
2328 printf("%s: free blocks %jd-%jd\n",
2329 mp->mnt_stat.f_mntonname,
2330 (intmax_t)cmd.value,
2331 (intmax_t)cmd.value + cmd.size - 1);
2332 }
2333 #endif /* DEBUG */
2334 blkno = cmd.value;
2335 blkcnt = cmd.size;
2336 blksize = fs->fs_frag - (blkno % fs->fs_frag);
2337 while (blkcnt > 0) {
2338 if (blksize > blkcnt)
2339 blksize = blkcnt;
2340 ffs_blkfree(fs, ump->um_devvp, blkno,
2341 blksize * fs->fs_fsize, ROOTINO);
2342 blkno += blksize;
2343 blkcnt -= blksize;
2344 blksize = fs->fs_frag;
2345 }
2346 break;
2347
2348 default:
2349 #ifdef DEBUG
2350 if (fsckcmds) {
2351 printf("Invalid request %d from fsck\n",
2352 oidp->oid_number);
2353 }
2354 #endif /* DEBUG */
2355 error = EINVAL;
2356 break;
2357
2358 }
2359 fdrop(fp, curthread);
2360 vn_finished_write(mp);
2361 return (error);
2362 }
Cache object: e79e08ee872cffbc8c92e54289f5ad5f
|