1/* $NetBSD: ffs_alloc.c,v 1.154 2016/10/30 15:01:46 christos Exp $ */
2
3/*-
4 * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Wasabi Systems, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32/*
33 * Copyright (c) 2002 Networks Associates Technology, Inc.
34 * All rights reserved.
35 *
36 * This software was developed for the FreeBSD Project by Marshall
37 * Kirk McKusick and Network Associates Laboratories, the Security
38 * Research Division of Network Associates, Inc. under DARPA/SPAWAR
39 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
40 * research program
41 *
42 * Copyright (c) 1982, 1986, 1989, 1993
43 * The Regents of the University of California. All rights reserved.
44 *
45 * Redistribution and use in source and binary forms, with or without
46 * modification, are permitted provided that the following conditions
47 * are met:
48 * 1. Redistributions of source code must retain the above copyright
49 * notice, this list of conditions and the following disclaimer.
50 * 2. Redistributions in binary form must reproduce the above copyright
51 * notice, this list of conditions and the following disclaimer in the
52 * documentation and/or other materials provided with the distribution.
53 * 3. Neither the name of the University nor the names of its contributors
54 * may be used to endorse or promote products derived from this software
55 * without specific prior written permission.
56 *
57 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67 * SUCH DAMAGE.
68 *
69 * @(#)ffs_alloc.c 8.19 (Berkeley) 7/13/95
70 */
71
72#include <sys/cdefs.h>
73__KERNEL_RCSID(0, "$NetBSD: ffs_alloc.c,v 1.154 2016/10/30 15:01:46 christos Exp $");
74
75#if defined(_KERNEL_OPT)
76#include "opt_ffs.h"
77#include "opt_quota.h"
78#include "opt_uvm_page_trkown.h"
79#endif
80
81#include <sys/param.h>
82#include <sys/systm.h>
83#include <sys/buf.h>
84#include <sys/cprng.h>
85#include <sys/fstrans.h>
86#include <sys/kauth.h>
87#include <sys/kernel.h>
88#include <sys/mount.h>
89#include <sys/proc.h>
90#include <sys/syslog.h>
91#include <sys/vnode.h>
92#include <sys/wapbl.h>
93#include <sys/cprng.h>
94
95#include <miscfs/specfs/specdev.h>
96#include <ufs/ufs/quota.h>
97#include <ufs/ufs/ufsmount.h>
98#include <ufs/ufs/inode.h>
99#include <ufs/ufs/ufs_extern.h>
100#include <ufs/ufs/ufs_bswap.h>
101#include <ufs/ufs/ufs_wapbl.h>
102
103#include <ufs/ffs/fs.h>
104#include <ufs/ffs/ffs_extern.h>
105
106#ifdef UVM_PAGE_TRKOWN
107#include <uvm/uvm.h>
108#endif
109
110static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int, int, int);
111static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t, int, int);
112static ino_t ffs_dirpref(struct inode *);
113static daddr_t ffs_fragextend(struct inode *, int, daddr_t, int, int);
114static void ffs_fserr(struct fs *, kauth_cred_t, const char *);
115static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int, int, int,
116 daddr_t (*)(struct inode *, int, daddr_t, int, int, int));
117static daddr_t ffs_nodealloccg(struct inode *, int, daddr_t, int, int, int);
118static int32_t ffs_mapsearch(struct fs *, struct cg *,
119 daddr_t, int);
120static void ffs_blkfree_common(struct ufsmount *, struct fs *, dev_t, struct buf *,
121 daddr_t, long, bool);
122static void ffs_freefile_common(struct ufsmount *, struct fs *, dev_t, struct buf *, ino_t,
123 int, bool);
124
125/* if 1, changes in optimalization strategy are logged */
126int ffs_log_changeopt = 0;
127
128/* in ffs_tables.c */
129extern const int inside[], around[];
130extern const u_char * const fragtbl[];
131
132/* Basic consistency check for block allocations */
133static int
134ffs_check_bad_allocation(const char *func, struct fs *fs, daddr_t bno,
135 long size, dev_t dev, ino_t inum)
136{
137 if ((u_int)size > fs->fs_bsize || ffs_fragoff(fs, size) != 0 ||
138 ffs_fragnum(fs, bno) + ffs_numfrags(fs, size) > fs->fs_frag) {
139 panic("%s: bad size: dev = 0x%llx, bno = %" PRId64
140 " bsize = %d, size = %ld, fs = %s", func,
141 (long long)dev, bno, fs->fs_bsize, size, fs->fs_fsmnt);
142 }
143
144 if (bno >= fs->fs_size) {
145 printf("%s: bad block %" PRId64 ", ino %llu\n", func, bno,
146 (unsigned long long)inum);
147 ffs_fserr(fs, NOCRED, "bad block");
148 return EINVAL;
149 }
150 return 0;
151}
152
153/*
154 * Allocate a block in the file system.
155 *
156 * The size of the requested block is given, which must be some
157 * multiple of fs_fsize and <= fs_bsize.
158 * A preference may be optionally specified. If a preference is given
159 * the following hierarchy is used to allocate a block:
160 * 1) allocate the requested block.
161 * 2) allocate a rotationally optimal block in the same cylinder.
162 * 3) allocate a block in the same cylinder group.
163 * 4) quadradically rehash into other cylinder groups, until an
164 * available block is located.
165 * If no block preference is given the following hierarchy is used
166 * to allocate a block:
167 * 1) allocate a block in the cylinder group that contains the
168 * inode for the file.
169 * 2) quadradically rehash into other cylinder groups, until an
170 * available block is located.
171 *
172 * => called with um_lock held
173 * => releases um_lock before returning
174 */
175int
176ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size,
177 int flags, kauth_cred_t cred, daddr_t *bnp)
178{
179 struct ufsmount *ump;
180 struct fs *fs;
181 daddr_t bno;
182 int cg;
183#if defined(QUOTA) || defined(QUOTA2)
184 int error;
185#endif
186
187 fs = ip->i_fs;
188 ump = ip->i_ump;
189
190 KASSERT(mutex_owned(&ump->um_lock));
191
192#ifdef UVM_PAGE_TRKOWN
193
194 /*
195 * Sanity-check that allocations within the file size
196 * do not allow other threads to read the stale contents
197 * of newly allocated blocks.
198 * Usually pages will exist to cover the new allocation.
199 * There is an optimization in ffs_write() where we skip
200 * creating pages if several conditions are met:
201 * - the file must not be mapped (in any user address space).
202 * - the write must cover whole pages and whole blocks.
203 * If those conditions are not met then pages must exist and
204 * be locked by the current thread.
205 */
206
207 if (ITOV(ip)->v_type == VREG &&
208 ffs_lblktosize(fs, (voff_t)lbn) < round_page(ITOV(ip)->v_size)) {
209 struct vm_page *pg;
210 struct vnode *vp = ITOV(ip);
211 struct uvm_object *uobj = &vp->v_uobj;
212 voff_t off = trunc_page(ffs_lblktosize(fs, lbn));
213 voff_t endoff = round_page(ffs_lblktosize(fs, lbn) + size);
214
215 mutex_enter(uobj->vmobjlock);
216 while (off < endoff) {
217 pg = uvm_pagelookup(uobj, off);
218 KASSERT((pg == NULL && (vp->v_vflag & VV_MAPPED) == 0 &&
219 (size & PAGE_MASK) == 0 &&
220 ffs_blkoff(fs, size) == 0) ||
221 (pg != NULL && pg->owner == curproc->p_pid &&
222 pg->lowner == curlwp->l_lid));
223 off += PAGE_SIZE;
224 }
225 mutex_exit(uobj->vmobjlock);
226 }
227#endif
228
229 *bnp = 0;
230#ifdef DIAGNOSTIC
231 if (cred == NOCRED)
232 panic("%s: missing credential", __func__);
233 if ((u_int)size > fs->fs_bsize || ffs_fragoff(fs, size) != 0) {
234 panic("%s: bad size: dev = 0x%llx, bsize = %d, size = %d, "
235 "fs = %s", __func__, (unsigned long long)ip->i_dev,
236 fs->fs_bsize, size, fs->fs_fsmnt);
237 }
238#endif /* DIAGNOSTIC */
239 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
240 goto nospace;
241 if (freespace(fs, fs->fs_minfree) <= 0 &&
242 kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL,
243 NULL, NULL) != 0)
244 goto nospace;
245#if defined(QUOTA) || defined(QUOTA2)
246 mutex_exit(&ump->um_lock);
247 if ((error = chkdq(ip, btodb(size), cred, 0)) != 0)
248 return (error);
249 mutex_enter(&ump->um_lock);
250#endif
251
252 if (bpref >= fs->fs_size)
253 bpref = 0;
254 if (bpref == 0)
255 cg = ino_to_cg(fs, ip->i_number);
256 else
257 cg = dtog(fs, bpref);
258 bno = ffs_hashalloc(ip, cg, bpref, size, 0, flags, ffs_alloccg);
259 if (bno > 0) {
260 DIP_ADD(ip, blocks, btodb(size));
261 ip->i_flag |= IN_CHANGE | IN_UPDATE;
262 *bnp = bno;
263 return (0);
264 }
265#if defined(QUOTA) || defined(QUOTA2)
266 /*
267 * Restore user's disk quota because allocation failed.
268 */
269 (void) chkdq(ip, -btodb(size), cred, FORCE);
270#endif
271 if (flags & B_CONTIG) {
272 /*
273 * XXX ump->um_lock handling is "suspect" at best.
274 * For the case where ffs_hashalloc() fails early
275 * in the B_CONTIG case we reach here with um_lock
276 * already unlocked, so we can't release it again
277 * like in the normal error path. See kern/39206.
278 *
279 *
280 * Fail silently - it's up to our caller to report
281 * errors.
282 */
283 return (ENOSPC);
284 }
285nospace:
286 mutex_exit(&ump->um_lock);
287 ffs_fserr(fs, cred, "file system full");
288 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
289 return (ENOSPC);
290}
291
292/*
293 * Reallocate a fragment to a bigger size
294 *
295 * The number and size of the old block is given, and a preference
296 * and new size is also specified. The allocator attempts to extend
297 * the original block. Failing that, the regular block allocator is
298 * invoked to get an appropriate block.
299 *
300 * => called with um_lock held
301 * => return with um_lock released
302 */
303int
304ffs_realloccg(struct inode *ip, daddr_t lbprev, daddr_t bpref, int osize,
305 int nsize, kauth_cred_t cred, struct buf **bpp, daddr_t *blknop)
306{
307 struct ufsmount *ump;
308 struct fs *fs;
309 struct buf *bp;
310 int cg, request, error;
311 daddr_t bprev, bno;
312
313 fs = ip->i_fs;
314 ump = ip->i_ump;
315
316 KASSERT(mutex_owned(&ump->um_lock));
317
318#ifdef UVM_PAGE_TRKOWN
319
320 /*
321 * Sanity-check that allocations within the file size
322 * do not allow other threads to read the stale contents
323 * of newly allocated blocks.
324 * Unlike in ffs_alloc(), here pages must always exist
325 * for such allocations, because only the last block of a file
326 * can be a fragment and ffs_write() will reallocate the
327 * fragment to the new size using ufs_balloc_range(),
328 * which always creates pages to cover blocks it allocates.
329 */
330
331 if (ITOV(ip)->v_type == VREG) {
332 struct vm_page *pg;
333 struct uvm_object *uobj = &ITOV(ip)->v_uobj;
334 voff_t off = trunc_page(ffs_lblktosize(fs, lbprev));
335 voff_t endoff = round_page(ffs_lblktosize(fs, lbprev) + osize);
336
337 mutex_enter(uobj->vmobjlock);
338 while (off < endoff) {
339 pg = uvm_pagelookup(uobj, off);
340 KASSERT(pg->owner == curproc->p_pid &&
341 pg->lowner == curlwp->l_lid);
342 off += PAGE_SIZE;
343 }
344 mutex_exit(uobj->vmobjlock);
345 }
346#endif
347
348#ifdef DIAGNOSTIC
349 if (cred == NOCRED)
350 panic("%s: missing credential", __func__);
351 if ((u_int)osize > fs->fs_bsize || ffs_fragoff(fs, osize) != 0 ||
352 (u_int)nsize > fs->fs_bsize || ffs_fragoff(fs, nsize) != 0) {
353 panic("%s: bad size: dev = 0x%llx, bsize = %d, osize = %d, "
354 "nsize = %d, fs = %s", __func__,
355 (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
356 fs->fs_fsmnt);
357 }
358#endif /* DIAGNOSTIC */
359 if (freespace(fs, fs->fs_minfree) <= 0 &&
360 kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL,
361 NULL, NULL) != 0) {
362 mutex_exit(&ump->um_lock);
363 goto nospace;
364 }
365 if (fs->fs_magic == FS_UFS2_MAGIC)
366 bprev = ufs_rw64(ip->i_ffs2_db[lbprev], UFS_FSNEEDSWAP(fs));
367 else
368 bprev = ufs_rw32(ip->i_ffs1_db[lbprev], UFS_FSNEEDSWAP(fs));
369
370 if (bprev == 0) {
371 panic("%s: bad bprev: dev = 0x%llx, bsize = %d, bprev = %"
372 PRId64 ", fs = %s", __func__,
373 (unsigned long long)ip->i_dev, fs->fs_bsize, bprev,
374 fs->fs_fsmnt);
375 }
376 mutex_exit(&ump->um_lock);
377
378 /*
379 * Allocate the extra space in the buffer.
380 */
381 if (bpp != NULL &&
382 (error = bread(ITOV(ip), lbprev, osize, 0, &bp)) != 0) {
383 return (error);
384 }
385#if defined(QUOTA) || defined(QUOTA2)
386 if ((error = chkdq(ip, btodb(nsize - osize), cred, 0)) != 0) {
387 if (bpp != NULL) {
388 brelse(bp, 0);
389 }
390 return (error);
391 }
392#endif
393 /*
394 * Check for extension in the existing location.
395 */
396 cg = dtog(fs, bprev);
397 mutex_enter(&ump->um_lock);
398 if ((bno = ffs_fragextend(ip, cg, bprev, osize, nsize)) != 0) {
399 DIP_ADD(ip, blocks, btodb(nsize - osize));
400 ip->i_flag |= IN_CHANGE | IN_UPDATE;
401
402 if (bpp != NULL) {
403 if (bp->b_blkno != FFS_FSBTODB(fs, bno)) {
404 panic("%s: bad blockno %#llx != %#llx",
405 __func__, (unsigned long long) bp->b_blkno,
406 (unsigned long long)FFS_FSBTODB(fs, bno));
407 }
408 allocbuf(bp, nsize, 1);
409 memset((char *)bp->b_data + osize, 0, nsize - osize);
410 mutex_enter(bp->b_objlock);
411 KASSERT(!cv_has_waiters(&bp->b_done));
412 bp->b_oflags |= BO_DONE;
413 mutex_exit(bp->b_objlock);
414 *bpp = bp;
415 }
416 if (blknop != NULL) {
417 *blknop = bno;
418 }
419 return (0);
420 }
421 /*
422 * Allocate a new disk location.
423 */
424 if (bpref >= fs->fs_size)
425 bpref = 0;
426 switch ((int)fs->fs_optim) {
427 case FS_OPTSPACE:
428 /*
429 * Allocate an exact sized fragment. Although this makes
430 * best use of space, we will waste time relocating it if
431 * the file continues to grow. If the fragmentation is
432 * less than half of the minimum free reserve, we choose
433 * to begin optimizing for time.
434 */
435 request = nsize;
436 if (fs->fs_minfree < 5 ||
437 fs->fs_cstotal.cs_nffree >
438 fs->fs_dsize * fs->fs_minfree / (2 * 100))
439 break;
440
441 if (ffs_log_changeopt) {
442 log(LOG_NOTICE,
443 "%s: optimization changed from SPACE to TIME\n",
444 fs->fs_fsmnt);
445 }
446
447 fs->fs_optim = FS_OPTTIME;
448 break;
449 case FS_OPTTIME:
450 /*
451 * At this point we have discovered a file that is trying to
452 * grow a small fragment to a larger fragment. To save time,
453 * we allocate a full sized block, then free the unused portion.
454 * If the file continues to grow, the `ffs_fragextend' call
455 * above will be able to grow it in place without further
456 * copying. If aberrant programs cause disk fragmentation to
457 * grow within 2% of the free reserve, we choose to begin
458 * optimizing for space.
459 */
460 request = fs->fs_bsize;
461 if (fs->fs_cstotal.cs_nffree <
462 fs->fs_dsize * (fs->fs_minfree - 2) / 100)
463 break;
464
465 if (ffs_log_changeopt) {
466 log(LOG_NOTICE,
467 "%s: optimization changed from TIME to SPACE\n",
468 fs->fs_fsmnt);
469 }
470
471 fs->fs_optim = FS_OPTSPACE;
472 break;
473 default:
474 panic("%s: bad optim: dev = 0x%llx, optim = %d, fs = %s",
475 __func__, (unsigned long long)ip->i_dev, fs->fs_optim,
476 fs->fs_fsmnt);
477 /* NOTREACHED */
478 }
479 bno = ffs_hashalloc(ip, cg, bpref, request, nsize, 0, ffs_alloccg);
480 if (bno > 0) {
481 /*
482 * Use forced deallocation registration, we can't handle
483 * failure here. This is safe, as this place is ever hit
484 * maximum once per write operation, when fragment is extended
485 * to longer fragment, or a full block.
486 */
487 if ((ip->i_ump->um_mountp->mnt_wapbl) &&
488 (ITOV(ip)->v_type != VREG)) {
489 /* this should never fail */
490 error = UFS_WAPBL_REGISTER_DEALLOCATION_FORCE(
491 ip->i_ump->um_mountp, FFS_FSBTODB(fs, bprev),
492 osize);
493 if (error)
494 panic("ffs_realloccg: dealloc registration failed");
495 } else {
496 ffs_blkfree(fs, ip->i_devvp, bprev, (long)osize,
497 ip->i_number);
498 }
499 DIP_ADD(ip, blocks, btodb(nsize - osize));
500 ip->i_flag |= IN_CHANGE | IN_UPDATE;
501 if (bpp != NULL) {
502 bp->b_blkno = FFS_FSBTODB(fs, bno);
503 allocbuf(bp, nsize, 1);
504 memset((char *)bp->b_data + osize, 0, (u_int)nsize - osize);
505 mutex_enter(bp->b_objlock);
506 KASSERT(!cv_has_waiters(&bp->b_done));
507 bp->b_oflags |= BO_DONE;
508 mutex_exit(bp->b_objlock);
509 *bpp = bp;
510 }
511 if (blknop != NULL) {
512 *blknop = bno;
513 }
514 return (0);
515 }
516 mutex_exit(&ump->um_lock);
517
518#if defined(QUOTA) || defined(QUOTA2)
519 /*
520 * Restore user's disk quota because allocation failed.
521 */
522 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
523#endif
524 if (bpp != NULL) {
525 brelse(bp, 0);
526 }
527
528nospace:
529 /*
530 * no space available
531 */
532 ffs_fserr(fs, cred, "file system full");
533 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
534 return (ENOSPC);
535}
536
537/*
538 * Allocate an inode in the file system.
539 *
540 * If allocating a directory, use ffs_dirpref to select the inode.
541 * If allocating in a directory, the following hierarchy is followed:
542 * 1) allocate the preferred inode.
543 * 2) allocate an inode in the same cylinder group.
544 * 3) quadradically rehash into other cylinder groups, until an
545 * available inode is located.
546 * If no inode preference is given the following hierarchy is used
547 * to allocate an inode:
548 * 1) allocate an inode in cylinder group 0.
549 * 2) quadradically rehash into other cylinder groups, until an
550 * available inode is located.
551 *
552 * => um_lock not held upon entry or return
553 */
554int
555ffs_valloc(struct vnode *pvp, int mode, kauth_cred_t cred, ino_t *inop)
556{
557 struct ufsmount *ump;
558 struct inode *pip;
559 struct fs *fs;
560 ino_t ino, ipref;
561 int cg, error;
562
563 UFS_WAPBL_JUNLOCK_ASSERT(pvp->v_mount);
564
565 pip = VTOI(pvp);
566 fs = pip->i_fs;
567 ump = pip->i_ump;
568
569 error = UFS_WAPBL_BEGIN(pvp->v_mount);
570 if (error) {
571 return error;
572 }
573 mutex_enter(&ump->um_lock);
574 if (fs->fs_cstotal.cs_nifree == 0)
575 goto noinodes;
576
577 if ((mode & IFMT) == IFDIR)
578 ipref = ffs_dirpref(pip);
579 else
580 ipref = pip->i_number;
581 if (ipref >= fs->fs_ncg * fs->fs_ipg)
582 ipref = 0;
583 cg = ino_to_cg(fs, ipref);
584 /*
585 * Track number of dirs created one after another
586 * in a same cg without intervening by files.
587 */
588 if ((mode & IFMT) == IFDIR) {
589 if (fs->fs_contigdirs[cg] < 255)
590 fs->fs_contigdirs[cg]++;
591 } else {
592 if (fs->fs_contigdirs[cg] > 0)
593 fs->fs_contigdirs[cg]--;
594 }
595 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, 0, ffs_nodealloccg);
596 if (ino == 0)
597 goto noinodes;
598 UFS_WAPBL_END(pvp->v_mount);
599 *inop = ino;
600 return 0;
601
602noinodes:
603 mutex_exit(&ump->um_lock);
604 UFS_WAPBL_END(pvp->v_mount);
605 ffs_fserr(fs, cred, "out of inodes");
606 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
607 return ENOSPC;
608}
609
610/*
611 * Find a cylinder group in which to place a directory.
612 *
613 * The policy implemented by this algorithm is to allocate a
614 * directory inode in the same cylinder group as its parent
615 * directory, but also to reserve space for its files inodes
616 * and data. Restrict the number of directories which may be
617 * allocated one after another in the same cylinder group
618 * without intervening allocation of files.
619 *
620 * If we allocate a first level directory then force allocation
621 * in another cylinder group.
622 */
623static ino_t
624ffs_dirpref(struct inode *pip)
625{
626 register struct fs *fs;
627 int cg, prefcg;
628 int64_t dirsize, cgsize, curdsz;
629 int avgifree, avgbfree, avgndir;
630 int minifree, minbfree, maxndir;
631 int mincg, minndir;
632 int maxcontigdirs;
633
634 KASSERT(mutex_owned(&pip->i_ump->um_lock));
635
636 fs = pip->i_fs;
637
638 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
639 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
640 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
641
642 /*
643 * Force allocation in another cg if creating a first level dir.
644 */
645 if (ITOV(pip)->v_vflag & VV_ROOT) {
646 prefcg = cprng_fast32() % fs->fs_ncg;
647 mincg = prefcg;
648 minndir = fs->fs_ipg;
649 for (cg = prefcg; cg < fs->fs_ncg; cg++)
650 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
651 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
652 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
653 mincg = cg;
654 minndir = fs->fs_cs(fs, cg).cs_ndir;
655 }
656 for (cg = 0; cg < prefcg; cg++)
657 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
658 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
659 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
660 mincg = cg;
661 minndir = fs->fs_cs(fs, cg).cs_ndir;
662 }
663 return ((ino_t)(fs->fs_ipg * mincg));
664 }
665
666 /*
667 * Count various limits which used for
668 * optimal allocation of a directory inode.
669 * Try cylinder groups with >75% avgifree and avgbfree.
670 * Avoid cylinder groups with no free blocks or inodes as that
671 * triggers an I/O-expensive cylinder group scan.
672 */
673 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
674 minifree = avgifree - avgifree / 4;
675 if (minifree < 1)
676 minifree = 1;
677 minbfree = avgbfree - avgbfree / 4;
678 if (minbfree < 1)
679 minbfree = 1;
680 cgsize = (int64_t)fs->fs_fsize * fs->fs_fpg;
681 dirsize = (int64_t)fs->fs_avgfilesize * fs->fs_avgfpdir;
682 if (avgndir != 0) {
683 curdsz = (cgsize - (int64_t)avgbfree * fs->fs_bsize) / avgndir;
684 if (dirsize < curdsz)
685 dirsize = curdsz;
686 }
687 if (cgsize < dirsize * 255)
688 maxcontigdirs = (avgbfree * fs->fs_bsize) / dirsize;
689 else
690 maxcontigdirs = 255;
691 if (fs->fs_avgfpdir > 0)
692 maxcontigdirs = min(maxcontigdirs,
693 fs->fs_ipg / fs->fs_avgfpdir);
694 if (maxcontigdirs == 0)
695 maxcontigdirs = 1;
696
697 /*
698 * Limit number of dirs in one cg and reserve space for
699 * regular files, but only if we have no deficit in
700 * inodes or space.
701 */
702 prefcg = ino_to_cg(fs, pip->i_number);
703 for (cg = prefcg; cg < fs->fs_ncg; cg++)
704 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
705 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
706 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
707 if (fs->fs_contigdirs[cg] < maxcontigdirs)
708 return ((ino_t)(fs->fs_ipg * cg));
709 }
710 for (cg = 0; cg < prefcg; cg++)
711 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
712 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
713 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
714 if (fs->fs_contigdirs[cg] < maxcontigdirs)
715 return ((ino_t)(fs->fs_ipg * cg));
716 }
717 /*
718 * This is a backstop when we are deficient in space.
719 */
720 for (cg = prefcg; cg < fs->fs_ncg; cg++)
721 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
722 return ((ino_t)(fs->fs_ipg * cg));
723 for (cg = 0; cg < prefcg; cg++)
724 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
725 break;
726 return ((ino_t)(fs->fs_ipg * cg));
727}
728
729/*
730 * Select the desired position for the next block in a file. The file is
731 * logically divided into sections. The first section is composed of the
732 * direct blocks. Each additional section contains fs_maxbpg blocks.
733 *
734 * If no blocks have been allocated in the first section, the policy is to
735 * request a block in the same cylinder group as the inode that describes
736 * the file. If no blocks have been allocated in any other section, the
737 * policy is to place the section in a cylinder group with a greater than
738 * average number of free blocks. An appropriate cylinder group is found
739 * by using a rotor that sweeps the cylinder groups. When a new group of
740 * blocks is needed, the sweep begins in the cylinder group following the
741 * cylinder group from which the previous allocation was made. The sweep
742 * continues until a cylinder group with greater than the average number
743 * of free blocks is found. If the allocation is for the first block in an
744 * indirect block, the information on the previous allocation is unavailable;
745 * here a best guess is made based upon the logical block number being
746 * allocated.
747 *
748 * If a section is already partially allocated, the policy is to
749 * contiguously allocate fs_maxcontig blocks. The end of one of these
750 * contiguous blocks and the beginning of the next is laid out
751 * contigously if possible.
752 *
753 * => um_lock held on entry and exit
754 */
755daddr_t
756ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int flags,
757 int32_t *bap /* XXX ondisk32 */)
758{
759 struct fs *fs;
760 int cg;
761 int avgbfree, startcg;
762
763 KASSERT(mutex_owned(&ip->i_ump->um_lock));
764
765 fs = ip->i_fs;
766
767 /*
768 * If allocating a contiguous file with B_CONTIG, use the hints
769 * in the inode extentions to return the desired block.
770 *
771 * For metadata (indirect blocks) return the address of where
772 * the first indirect block resides - we'll scan for the next
773 * available slot if we need to allocate more than one indirect
774 * block. For data, return the address of the actual block
775 * relative to the address of the first data block.
776 */
777 if (flags & B_CONTIG) {
778 KASSERT(ip->i_ffs_first_data_blk != 0);
779 KASSERT(ip->i_ffs_first_indir_blk != 0);
780 if (flags & B_METAONLY)
781 return ip->i_ffs_first_indir_blk;
782 else
783 return ip->i_ffs_first_data_blk + ffs_blkstofrags(fs, lbn);
784 }
785
786 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
787 if (lbn < UFS_NDADDR + FFS_NINDIR(fs)) {
788 cg = ino_to_cg(fs, ip->i_number);
789 return (cgbase(fs, cg) + fs->fs_frag);
790 }
791 /*
792 * Find a cylinder with greater than average number of
793 * unused data blocks.
794 */
795 if (indx == 0 || bap[indx - 1] == 0)
796 startcg =
797 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
798 else
799 startcg = dtog(fs,
800 ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
801 startcg %= fs->fs_ncg;
802 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
803 for (cg = startcg; cg < fs->fs_ncg; cg++)
804 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
805 return (cgbase(fs, cg) + fs->fs_frag);
806 }
807 for (cg = 0; cg < startcg; cg++)
808 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
809 return (cgbase(fs, cg) + fs->fs_frag);
810 }
811 return (0);
812 }
813 /*
814 * We just always try to lay things out contiguously.
815 */
816 return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
817}
818
819daddr_t
820ffs_blkpref_ufs2(struct inode *ip, daddr_t lbn, int indx, int flags,
821 int64_t *bap)
822{
823 struct fs *fs;
824 int cg;
825 int avgbfree, startcg;
826
827 KASSERT(mutex_owned(&ip->i_ump->um_lock));
828
829 fs = ip->i_fs;
830
831 /*
832 * If allocating a contiguous file with B_CONTIG, use the hints
833 * in the inode extentions to return the desired block.
834 *
835 * For metadata (indirect blocks) return the address of where
836 * the first indirect block resides - we'll scan for the next
837 * available slot if we need to allocate more than one indirect
838 * block. For data, return the address of the actual block
839 * relative to the address of the first data block.
840 */
841 if (flags & B_CONTIG) {
842 KASSERT(ip->i_ffs_first_data_blk != 0);
843 KASSERT(ip->i_ffs_first_indir_blk != 0);
844 if (flags & B_METAONLY)
845 return ip->i_ffs_first_indir_blk;
846 else
847 return ip->i_ffs_first_data_blk + ffs_blkstofrags(fs, lbn);
848 }
849
850 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
851 if (lbn < UFS_NDADDR + FFS_NINDIR(fs)) {
852 cg = ino_to_cg(fs, ip->i_number);
853 return (cgbase(fs, cg) + fs->fs_frag);
854 }
855 /*
856 * Find a cylinder with greater than average number of
857 * unused data blocks.
858 */
859 if (indx == 0 || bap[indx - 1] == 0)
860 startcg =
861 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
862 else
863 startcg = dtog(fs,
864 ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
865 startcg %= fs->fs_ncg;
866 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
867 for (cg = startcg; cg < fs->fs_ncg; cg++)
868 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
869 return (cgbase(fs, cg) + fs->fs_frag);
870 }
871 for (cg = 0; cg < startcg; cg++)
872 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
873 return (cgbase(fs, cg) + fs->fs_frag);
874 }
875 return (0);
876 }
877 /*
878 * We just always try to lay things out contiguously.
879 */
880 return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
881}
882
883
884/*
885 * Implement the cylinder overflow algorithm.
886 *
887 * The policy implemented by this algorithm is:
888 * 1) allocate the block in its requested cylinder group.
889 * 2) quadradically rehash on the cylinder group number.
890 * 3) brute force search for a free block.
891 *
892 * => called with um_lock held
893 * => returns with um_lock released on success, held on failure
894 * (*allocator releases lock on success, retains lock on failure)
895 */
896/*VARARGS5*/
897static daddr_t
898ffs_hashalloc(struct inode *ip, int cg, daddr_t pref,
899 int size /* size for data blocks, mode for inodes */,
900 int realsize,
901 int flags,
902 daddr_t (*allocator)(struct inode *, int, daddr_t, int, int, int))
903{
904 struct fs *fs;
905 daddr_t result;
906 int i, icg = cg;
907
908 fs = ip->i_fs;
909 /*
910 * 1: preferred cylinder group
911 */
912 result = (*allocator)(ip, cg, pref, size, realsize, flags);
913 if (result)
914 return (result);
915
916 if (flags & B_CONTIG)
917 return (result);
918 /*
919 * 2: quadratic rehash
920 */
921 for (i = 1; i < fs->fs_ncg; i *= 2) {
922 cg += i;
923 if (cg >= fs->fs_ncg)
924 cg -= fs->fs_ncg;
925 result = (*allocator)(ip, cg, 0, size, realsize, flags);
926 if (result)
927 return (result);
928 }
929 /*
930 * 3: brute force search
931 * Note that we start at i == 2, since 0 was checked initially,
932 * and 1 is always checked in the quadratic rehash.
933 */
934 cg = (icg + 2) % fs->fs_ncg;
935 for (i = 2; i < fs->fs_ncg; i++) {
936 result = (*allocator)(ip, cg, 0, size, realsize, flags);
937 if (result)
938 return (result);
939 cg++;
940 if (cg == fs->fs_ncg)
941 cg = 0;
942 }
943 return (0);
944}
945
946/*
947 * Determine whether a fragment can be extended.
948 *
949 * Check to see if the necessary fragments are available, and
950 * if they are, allocate them.
951 *
952 * => called with um_lock held
953 * => returns with um_lock released on success, held on failure
954 */
955static daddr_t
956ffs_fragextend(struct inode *ip, int cg, daddr_t bprev, int osize, int nsize)
957{
958 struct ufsmount *ump;
959 struct fs *fs;
960 struct cg *cgp;
961 struct buf *bp;
962 daddr_t bno;
963 int frags, bbase;
964 int i, error;
965 u_int8_t *blksfree;
966
967 fs = ip->i_fs;
968 ump = ip->i_ump;
969
970 KASSERT(mutex_owned(&ump->um_lock));
971
972 if (fs->fs_cs(fs, cg).cs_nffree < ffs_numfrags(fs, nsize - osize))
973 return (0);
974 frags = ffs_numfrags(fs, nsize);
975 bbase = ffs_fragnum(fs, bprev);
976 if (bbase > ffs_fragnum(fs, (bprev + frags - 1))) {
977 /* cannot extend across a block boundary */
978 return (0);
979 }
980 mutex_exit(&ump->um_lock);
981 error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
982 (int)fs->fs_cgsize, B_MODIFY, &bp);
983 if (error)
984 goto fail;
985 cgp = (struct cg *)bp->b_data;
986 if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs)))
987 goto fail;
988 cgp->cg_old_time = ufs_rw32(time_second, UFS_FSNEEDSWAP(fs));
989 if ((fs->fs_magic != FS_UFS1_MAGIC) ||
990 (fs->fs_old_flags & FS_FLAGS_UPDATED))
991 cgp->cg_time = ufs_rw64(time_second, UFS_FSNEEDSWAP(fs));
992 bno = dtogd(fs, bprev);
993 blksfree = cg_blksfree(cgp, UFS_FSNEEDSWAP(fs));
994 for (i = ffs_numfrags(fs, osize); i < frags; i++)
995 if (isclr(blksfree, bno + i))
996 goto fail;
997 /*
998 * the current fragment can be extended
999 * deduct the count on fragment being extended into
1000 * increase the count on the remaining fragment (if any)
1001 * allocate the extended piece
1002 */
1003 for (i = frags; i < fs->fs_frag - bbase; i++)
1004 if (isclr(blksfree, bno + i))
1005 break;
1006 ufs_add32(cgp->cg_frsum[i - ffs_numfrags(fs, osize)], -1, UFS_FSNEEDSWAP(fs));
1007 if (i != frags)
1008 ufs_add32(cgp->cg_frsum[i - frags], 1, UFS_FSNEEDSWAP(fs));
1009 mutex_enter(&ump->um_lock);
1010 for (i = ffs_numfrags(fs, osize); i < frags; i++) {
1011 clrbit(blksfree, bno + i);
1012 ufs_add32(cgp->cg_cs.cs_nffree, -1, UFS_FSNEEDSWAP(fs));
1013 fs->fs_cstotal.cs_nffree--;
1014 fs->fs_cs(fs, cg).cs_nffree--;
1015 }
1016 fs->fs_fmod = 1;
1017 ACTIVECG_CLR(fs, cg);
1018 mutex_exit(&ump->um_lock);
1019 bdwrite(bp);
1020 return (bprev);
1021
1022 fail:
1023 if (bp != NULL)
1024 brelse(bp, 0);
1025 mutex_enter(&ump->um_lock);
1026 return (0);
1027}
1028
1029/*
1030 * Determine whether a block can be allocated.
1031 *
1032 * Check to see if a block of the appropriate size is available,
1033 * and if it is, allocate it.
1034 */
1035static daddr_t
1036ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size, int realsize,
1037 int flags)
1038{
1039 struct ufsmount *ump;
1040 struct fs *fs = ip->i_fs;
1041 struct cg *cgp;
1042 struct buf *bp;
1043 int32_t bno;
1044 daddr_t blkno;
1045 int error, frags, allocsiz, i;
1046 u_int8_t *blksfree;
1047 const int needswap = UFS_FSNEEDSWAP(fs);
1048
1049 ump = ip->i_ump;
1050
1051 KASSERT(mutex_owned(&ump->um_lock));
1052
1053 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
1054 return (0);
1055 mutex_exit(&ump->um_lock);
1056 error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
1057 (int)fs->fs_cgsize, B_MODIFY, &bp);
1058 if (error)
1059 goto fail;
1060 cgp = (struct cg *)bp->b_data;
1061 if (!cg_chkmagic(cgp, needswap) ||
1062 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
1063 goto fail;
1064 cgp->cg_old_time = ufs_rw32(time_second, needswap);
1065 if ((fs->fs_magic != FS_UFS1_MAGIC) ||
1066 (fs->fs_old_flags & FS_FLAGS_UPDATED))
1067 cgp->cg_time = ufs_rw64(time_second, needswap);
1068 if (size == fs->fs_bsize) {
1069 mutex_enter(&ump->um_lock);
1070 blkno = ffs_alloccgblk(ip, bp, bpref, realsize, flags);
1071 ACTIVECG_CLR(fs, cg);
1072 mutex_exit(&ump->um_lock);
1073
1074 /*
1075 * If actually needed size is lower, free the extra blocks now.
1076 * This is safe to call here, there is no outside reference
1077 * to this block yet. It is not necessary to keep um_lock
1078 * locked.
1079 */
1080 if (realsize != 0 && realsize < size) {
1081 ffs_blkfree_common(ip->i_ump, ip->i_fs,
1082 ip->i_devvp->v_rdev,
1083 bp, blkno + ffs_numfrags(fs, realsize),
1084 (long)(size - realsize), false);
1085 }
1086
1087 bdwrite(bp);
1088 return (blkno);
1089 }
1090 /*
1091 * check to see if any fragments are already available
1092 * allocsiz is the size which will be allocated, hacking
1093 * it down to a smaller size if necessary
1094 */
1095 blksfree = cg_blksfree(cgp, needswap);
1096 frags = ffs_numfrags(fs, size);
1097 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
1098 if (cgp->cg_frsum[allocsiz] != 0)
1099 break;
1100 if (allocsiz == fs->fs_frag) {
1101 /*
1102 * no fragments were available, so a block will be
1103 * allocated, and hacked up
1104 */
1105 if (cgp->cg_cs.cs_nbfree == 0)
1106 goto fail;
1107 mutex_enter(&ump->um_lock);
1108 blkno = ffs_alloccgblk(ip, bp, bpref, realsize, flags);
1109 bno = dtogd(fs, blkno);
1110 for (i = frags; i < fs->fs_frag; i++)
1111 setbit(blksfree, bno + i);
1112 i = fs->fs_frag - frags;
1113 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
1114 fs->fs_cstotal.cs_nffree += i;
1115 fs->fs_cs(fs, cg).cs_nffree += i;
1116 fs->fs_fmod = 1;
1117 ufs_add32(cgp->cg_frsum[i], 1, needswap);
1118 ACTIVECG_CLR(fs, cg);
1119 mutex_exit(&ump->um_lock);
1120 bdwrite(bp);
1121 return (blkno);
1122 }
1123 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
1124#if 0
1125 /*
1126 * XXX fvdl mapsearch will panic, and never return -1
1127 * also: returning NULL as daddr_t ?
1128 */
1129 if (bno < 0)
1130 goto fail;
1131#endif
1132 for (i = 0; i < frags; i++)
1133 clrbit(blksfree, bno + i);
1134 mutex_enter(&ump->um_lock);
1135 ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
1136 fs->fs_cstotal.cs_nffree -= frags;
1137 fs->fs_cs(fs, cg).cs_nffree -= frags;
1138 fs->fs_fmod = 1;
1139 ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
1140 if (frags != allocsiz)
1141 ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
1142 blkno = cgbase(fs, cg) + bno;
1143 ACTIVECG_CLR(fs, cg);
1144 mutex_exit(&ump->um_lock);
1145 bdwrite(bp);
1146 return blkno;
1147
1148 fail:
1149 if (bp != NULL)
1150 brelse(bp, 0);
1151 mutex_enter(&ump->um_lock);
1152 return (0);
1153}
1154
1155/*
1156 * Allocate a block in a cylinder group.
1157 *
1158 * This algorithm implements the following policy:
1159 * 1) allocate the requested block.
1160 * 2) allocate a rotationally optimal block in the same cylinder.
1161 * 3) allocate the next available block on the block rotor for the
1162 * specified cylinder group.
1163 * Note that this routine only allocates fs_bsize blocks; these
1164 * blocks may be fragmented by the routine that allocates them.
1165 */
1166static daddr_t
1167ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref, int realsize,
1168 int flags)
1169{
1170 struct fs *fs = ip->i_fs;
1171 struct cg *cgp;
1172 int cg;
1173 daddr_t blkno;
1174 int32_t bno;
1175 u_int8_t *blksfree;
1176 const int needswap = UFS_FSNEEDSWAP(fs);
1177
1178 KASSERT(mutex_owned(&ip->i_ump->um_lock));
1179
1180 cgp = (struct cg *)bp->b_data;
1181 blksfree = cg_blksfree(cgp, needswap);
1182 if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
1183 bpref = ufs_rw32(cgp->cg_rotor, needswap);
1184 } else {
1185 bpref = ffs_blknum(fs, bpref);
1186 bno = dtogd(fs, bpref);
1187 /*
1188 * if the requested block is available, use it
1189 */
1190 if (ffs_isblock(fs, blksfree, ffs_fragstoblks(fs, bno)))
1191 goto gotit;
1192 /*
1193 * if the requested data block isn't available and we are
1194 * trying to allocate a contiguous file, return an error.
1195 */
1196 if ((flags & (B_CONTIG | B_METAONLY)) == B_CONTIG)
1197 return (0);
1198 }
1199
1200 /*
1201 * Take the next available block in this cylinder group.
1202 */
1203 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1204#if 0
1205 /*
1206 * XXX jdolecek ffs_mapsearch() succeeds or panics
1207 */
1208 if (bno < 0)
1209 return (0);
1210#endif
1211 cgp->cg_rotor = ufs_rw32(bno, needswap);
1212gotit:
1213 blkno = ffs_fragstoblks(fs, bno);
1214 ffs_clrblock(fs, blksfree, blkno);
1215 ffs_clusteracct(fs, cgp, blkno, -1);
1216 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
1217 fs->fs_cstotal.cs_nbfree--;
1218 fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
1219 if ((fs->fs_magic == FS_UFS1_MAGIC) &&
1220 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
1221 int cylno;
1222 cylno = old_cbtocylno(fs, bno);
1223 KASSERT(cylno >= 0);
1224 KASSERT(cylno < fs->fs_old_ncyl);
1225 KASSERT(old_cbtorpos(fs, bno) >= 0);
1226 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bno) < fs->fs_old_nrpos);
1227 ufs_add16(old_cg_blks(fs, cgp, cylno, needswap)[old_cbtorpos(fs, bno)], -1,
1228 needswap);
1229 ufs_add32(old_cg_blktot(cgp, needswap)[cylno], -1, needswap);
1230 }
1231 fs->fs_fmod = 1;
1232 cg = ufs_rw32(cgp->cg_cgx, needswap);
1233 blkno = cgbase(fs, cg) + bno;
1234 return (blkno);
1235}
1236
1237/*
1238 * Determine whether an inode can be allocated.
1239 *
1240 * Check to see if an inode is available, and if it is,
1241 * allocate it using the following policy:
1242 * 1) allocate the requested inode.
1243 * 2) allocate the next available inode after the requested
1244 * inode in the specified cylinder group.
1245 */
1246static daddr_t
1247ffs_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode, int realsize,
1248 int flags)
1249{
1250 struct ufsmount *ump = ip->i_ump;
1251 struct fs *fs = ip->i_fs;
1252 struct cg *cgp;
1253 struct buf *bp, *ibp;
1254 u_int8_t *inosused;
1255 int error, start, len, loc, map, i;
1256 int32_t initediblk;
1257 daddr_t nalloc;
1258 struct ufs2_dinode *dp2;
1259 const int needswap = UFS_FSNEEDSWAP(fs);
1260
1261 KASSERT(mutex_owned(&ump->um_lock));
1262 UFS_WAPBL_JLOCK_ASSERT(ip->i_ump->um_mountp);
1263
1264 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1265 return (0);
1266 mutex_exit(&ump->um_lock);
1267 ibp = NULL;
1268 initediblk = -1;
1269retry:
1270 error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
1271 (int)fs->fs_cgsize, B_MODIFY, &bp);
1272 if (error)
1273 goto fail;
1274 cgp = (struct cg *)bp->b_data;
1275 if (!cg_chkmagic(cgp, needswap) || cgp->cg_cs.cs_nifree == 0)
1276 goto fail;
1277
1278 if (ibp != NULL &&
1279 initediblk != ufs_rw32(cgp->cg_initediblk, needswap)) {
1280 /* Another thread allocated more inodes so we retry the test. */
1281 brelse(ibp, 0);
1282 ibp = NULL;
1283 }
1284 /*
1285 * Check to see if we need to initialize more inodes.
1286 */
1287 if (fs->fs_magic == FS_UFS2_MAGIC && ibp == NULL) {
1288 initediblk = ufs_rw32(cgp->cg_initediblk, needswap);
1289 nalloc = fs->fs_ipg - ufs_rw32(cgp->cg_cs.cs_nifree, needswap);
1290 if (nalloc + FFS_INOPB(fs) > initediblk &&
1291 initediblk < ufs_rw32(cgp->cg_niblk, needswap)) {
1292 /*
1293 * We have to release the cg buffer here to prevent
1294 * a deadlock when reading the inode block will
1295 * run a copy-on-write that might use this cg.
1296 */
1297 brelse(bp, 0);
1298 bp = NULL;
1299 error = ffs_getblk(ip->i_devvp, FFS_FSBTODB(fs,
1300 ino_to_fsba(fs, cg * fs->fs_ipg + initediblk)),
1301 FFS_NOBLK, fs->fs_bsize, false, &ibp);
1302 if (error)
1303 goto fail;
1304 goto retry;
1305 }
1306 }
1307
1308 cgp->cg_old_time = ufs_rw32(time_second, needswap);
1309 if ((fs->fs_magic != FS_UFS1_MAGIC) ||
1310 (fs->fs_old_flags & FS_FLAGS_UPDATED))
1311 cgp->cg_time = ufs_rw64(time_second, needswap);
1312 inosused = cg_inosused(cgp, needswap);
1313 if (ipref) {
1314 ipref %= fs->fs_ipg;
1315 if (isclr(inosused, ipref))
1316 goto gotit;
1317 }
1318 start = ufs_rw32(cgp->cg_irotor, needswap) / NBBY;
1319 len = howmany(fs->fs_ipg - ufs_rw32(cgp->cg_irotor, needswap),
1320 NBBY);
1321 loc = skpc(0xff, len, &inosused[start]);
1322 if (loc == 0) {
1323 len = start + 1;
1324 start = 0;
1325 loc = skpc(0xff, len, &inosused[0]);
1326 if (loc == 0) {
1327 panic("%s: map corrupted: cg=%d, irotor=%d, fs=%s",
1328 __func__, cg, ufs_rw32(cgp->cg_irotor, needswap),
1329 fs->fs_fsmnt);
1330 /* NOTREACHED */
1331 }
1332 }
1333 i = start + len - loc;
1334 map = inosused[i] ^ 0xff;
1335 if (map == 0) {
1336 panic("%s: block not in map: fs=%s", __func__, fs->fs_fsmnt);
1337 }
1338 ipref = i * NBBY + ffs(map) - 1;
1339 cgp->cg_irotor = ufs_rw32(ipref, needswap);
1340gotit:
1341 UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp, cg * fs->fs_ipg + ipref,
1342 mode);
1343 /*
1344 * Check to see if we need to initialize more inodes.
1345 */
1346 if (ibp != NULL) {
1347 KASSERT(initediblk == ufs_rw32(cgp->cg_initediblk, needswap));
1348 memset(ibp->b_data, 0, fs->fs_bsize);
1349 dp2 = (struct ufs2_dinode *)(ibp->b_data);
1350 for (i = 0; i < FFS_INOPB(fs); i++) {
1351 /*
1352 * Don't bother to swap, it's supposed to be
1353 * random, after all.
1354 */
1355 dp2->di_gen = (cprng_fast32() & INT32_MAX) / 2 + 1;
1356 dp2++;
1357 }
1358 initediblk += FFS_INOPB(fs);
1359 cgp->cg_initediblk = ufs_rw32(initediblk, needswap);
1360 }
1361
1362 mutex_enter(&ump->um_lock);
1363 ACTIVECG_CLR(fs, cg);
1364 setbit(inosused, ipref);
1365 ufs_add32(cgp->cg_cs.cs_nifree, -1, needswap);
1366 fs->fs_cstotal.cs_nifree--;
1367 fs->fs_cs(fs, cg).cs_nifree--;
1368 fs->fs_fmod = 1;
1369 if ((mode & IFMT) == IFDIR) {
1370 ufs_add32(cgp->cg_cs.cs_ndir, 1, needswap);
1371 fs->fs_cstotal.cs_ndir++;
1372 fs->fs_cs(fs, cg).cs_ndir++;
1373 }
1374 mutex_exit(&ump->um_lock);
1375 if (ibp != NULL) {
1376 bwrite(bp);
1377 bawrite(ibp);
1378 } else
1379 bdwrite(bp);
1380 return (cg * fs->fs_ipg + ipref);
1381 fail:
1382 if (bp != NULL)
1383 brelse(bp, 0);
1384 if (ibp != NULL)
1385 brelse(ibp, 0);
1386 mutex_enter(&ump->um_lock);
1387 return (0);
1388}
1389
1390/*
1391 * Allocate a block or fragment.
1392 *
1393 * The specified block or fragment is removed from the
1394 * free map, possibly fragmenting a block in the process.
1395 *
1396 * This implementation should mirror fs_blkfree
1397 *
1398 * => um_lock not held on entry or exit
1399 */
1400int
1401ffs_blkalloc(struct inode *ip, daddr_t bno, long size)
1402{
1403 int error;
1404
1405 error = ffs_check_bad_allocation(__func__, ip->i_fs, bno, size,
1406 ip->i_dev, ip->i_uid);
1407 if (error)
1408 return error;
1409
1410 return ffs_blkalloc_ump(ip->i_ump, bno, size);
1411}
1412
1413int
1414ffs_blkalloc_ump(struct ufsmount *ump, daddr_t bno, long size)
1415{
1416 struct fs *fs = ump->um_fs;
1417 struct cg *cgp;
1418 struct buf *bp;
1419 int32_t fragno, cgbno;
1420 int i, error, cg, blk, frags, bbase;
1421 u_int8_t *blksfree;
1422 const int needswap = UFS_FSNEEDSWAP(fs);
1423
1424 KASSERT((u_int)size <= fs->fs_bsize && ffs_fragoff(fs, size) == 0 &&
1425 ffs_fragnum(fs, bno) + ffs_numfrags(fs, size) <= fs->fs_frag);
1426 KASSERT(bno < fs->fs_size);
1427
1428 cg = dtog(fs, bno);
1429 error = bread(ump->um_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
1430 (int)fs->fs_cgsize, B_MODIFY, &bp);
1431 if (error) {
1432 return error;
1433 }
1434 cgp = (struct cg *)bp->b_data;
1435 if (!cg_chkmagic(cgp, needswap)) {
1436 brelse(bp, 0);
1437 return EIO;
1438 }
1439 cgp->cg_old_time = ufs_rw32(time_second, needswap);
1440 cgp->cg_time = ufs_rw64(time_second, needswap);
1441 cgbno = dtogd(fs, bno);
1442 blksfree = cg_blksfree(cgp, needswap);
1443
1444 mutex_enter(&ump->um_lock);
1445 if (size == fs->fs_bsize) {
1446 fragno = ffs_fragstoblks(fs, cgbno);
1447 if (!ffs_isblock(fs, blksfree, fragno)) {
1448 mutex_exit(&ump->um_lock);
1449 brelse(bp, 0);
1450 return EBUSY;
1451 }
1452 ffs_clrblock(fs, blksfree, fragno);
1453 ffs_clusteracct(fs, cgp, fragno, -1);
1454 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
1455 fs->fs_cstotal.cs_nbfree--;
1456 fs->fs_cs(fs, cg).cs_nbfree--;
1457 } else {
1458 bbase = cgbno - ffs_fragnum(fs, cgbno);
1459
1460 frags = ffs_numfrags(fs, size);
1461 for (i = 0; i < frags; i++) {
1462 if (isclr(blksfree, cgbno + i)) {
1463 mutex_exit(&ump->um_lock);
1464 brelse(bp, 0);
1465 return EBUSY;
1466 }
1467 }
1468 /*
1469 * if a complete block is being split, account for it
1470 */
1471 fragno = ffs_fragstoblks(fs, bbase);
1472 if (ffs_isblock(fs, blksfree, fragno)) {
1473 ufs_add32(cgp->cg_cs.cs_nffree, fs->fs_frag, needswap);
1474 fs->fs_cstotal.cs_nffree += fs->fs_frag;
1475 fs->fs_cs(fs, cg).cs_nffree += fs->fs_frag;
1476 ffs_clusteracct(fs, cgp, fragno, -1);
1477 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
1478 fs->fs_cstotal.cs_nbfree--;
1479 fs->fs_cs(fs, cg).cs_nbfree--;
1480 }
1481 /*
1482 * decrement the counts associated with the old frags
1483 */
1484 blk = blkmap(fs, blksfree, bbase);
1485 ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
1486 /*
1487 * allocate the fragment
1488 */
1489 for (i = 0; i < frags; i++) {
1490 clrbit(blksfree, cgbno + i);
1491 }
1492 ufs_add32(cgp->cg_cs.cs_nffree, -i, needswap);
1493 fs->fs_cstotal.cs_nffree -= i;
1494 fs->fs_cs(fs, cg).cs_nffree -= i;
1495 /*
1496 * add back in counts associated with the new frags
1497 */
1498 blk = blkmap(fs, blksfree, bbase);
1499 ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
1500 }
1501 fs->fs_fmod = 1;
1502 ACTIVECG_CLR(fs, cg);
1503 mutex_exit(&ump->um_lock);
1504 bdwrite(bp);
1505 return 0;
1506}
1507
1508/*
1509 * Free a block or fragment.
1510 *
1511 * The specified block or fragment is placed back in the
1512 * free map. If a fragment is deallocated, a possible
1513 * block reassembly is checked.
1514 *
1515 * => um_lock not held on entry or exit
1516 */
1517static void
1518ffs_blkfree_cg(struct fs *fs, struct vnode *devvp, daddr_t bno, long size)
1519{
1520 struct cg *cgp;
1521 struct buf *bp;
1522 struct ufsmount *ump;
1523 daddr_t cgblkno;
1524 int error, cg;
1525 dev_t dev;
1526 const bool devvp_is_snapshot = (devvp->v_type != VBLK);
1527 const int needswap = UFS_FSNEEDSWAP(fs);
1528
1529 KASSERT(!devvp_is_snapshot);
1530
1531 cg = dtog(fs, bno);
1532 dev = devvp->v_rdev;
1533 ump = VFSTOUFS(spec_node_getmountedfs(devvp));
1534 KASSERT(fs == ump->um_fs);
1535 cgblkno = FFS_FSBTODB(fs, cgtod(fs, cg));
1536
1537 error = bread(devvp, cgblkno, (int)fs->fs_cgsize,
1538 B_MODIFY, &bp);
1539 if (error) {
1540 return;
1541 }
1542 cgp = (struct cg *)bp->b_data;
1543 if (!cg_chkmagic(cgp, needswap)) {
1544 brelse(bp, 0);
1545 return;
1546 }
1547
1548 ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot);
1549
1550 bdwrite(bp);
1551}
1552
1553struct discardopdata {
1554 struct work wk; /* must be first */
1555 struct vnode *devvp;
1556 daddr_t bno;
1557 long size;
1558};
1559
1560struct discarddata {
1561 struct fs *fs;
1562 struct discardopdata *entry;
1563 long maxsize;
1564 kmutex_t entrylk;
1565 struct workqueue *wq;
1566 int wqcnt, wqdraining;
1567 kmutex_t wqlk;
1568 kcondvar_t wqcv;
1569 /* timer for flush? */
1570};
1571
1572static void
1573ffs_blkfree_td(struct fs *fs, struct discardopdata *td)
1574{
1575 struct mount *mp = spec_node_getmountedfs(td->devvp);
1576 long todo;
1577 int error;
1578
1579 while (td->size) {
1580 todo = min(td->size,
1581 ffs_lfragtosize(fs, (fs->fs_frag - ffs_fragnum(fs, td->bno))));
1582 error = UFS_WAPBL_BEGIN(mp);
1583 if (error) {
1584 printf("ffs: failed to begin wapbl transaction"
1585 " for discard: %d\n", error);
1586 break;
1587 }
1588 ffs_blkfree_cg(fs, td->devvp, td->bno, todo);
1589 UFS_WAPBL_END(mp);
1590 td->bno += ffs_numfrags(fs, todo);
1591 td->size -= todo;
1592 }
1593}
1594
1595static void
1596ffs_discardcb(struct work *wk, void *arg)
1597{
1598 struct discardopdata *td = (void *)wk;
1599 struct discarddata *ts = arg;
1600 struct fs *fs = ts->fs;
1601 off_t start, len;
1602#ifdef TRIMDEBUG
1603 int error;
1604#endif
1605
1606/* like FSBTODB but emits bytes; XXX move to fs.h */
1607#ifndef FFS_FSBTOBYTES
1608#define FFS_FSBTOBYTES(fs, b) ((b) << (fs)->fs_fshift)
1609#endif
1610
1611 start = FFS_FSBTOBYTES(fs, td->bno);
1612 len = td->size;
1613#ifdef TRIMDEBUG
1614 error =
1615#endif
1616 VOP_FDISCARD(td->devvp, start, len);
1617#ifdef TRIMDEBUG
1618 printf("trim(%" PRId64 ",%ld):%d\n", td->bno, td->size, error);
1619#endif
1620
1621 ffs_blkfree_td(fs, td);
1622 kmem_free(td, sizeof(*td));
1623 mutex_enter(&ts->wqlk);
1624 ts->wqcnt--;
1625 if (ts->wqdraining && !ts->wqcnt)
1626 cv_signal(&ts->wqcv);
1627 mutex_exit(&ts->wqlk);
1628}
1629
1630void *
1631ffs_discard_init(struct vnode *devvp, struct fs *fs)
1632{
1633 struct discarddata *ts;
1634 int error;
1635
1636 ts = kmem_zalloc(sizeof (*ts), KM_SLEEP);
1637 error = workqueue_create(&ts->wq, "trimwq", ffs_discardcb, ts,
1638 0, 0, 0);
1639 if (error) {
1640 kmem_free(ts, sizeof (*ts));
1641 return NULL;
1642 }
1643 mutex_init(&ts->entrylk, MUTEX_DEFAULT, IPL_NONE);
1644 mutex_init(&ts->wqlk, MUTEX_DEFAULT, IPL_NONE);
1645 cv_init(&ts->wqcv, "trimwqcv");
1646 ts->maxsize = 100*1024; /* XXX */
1647 ts->fs = fs;
1648 return ts;
1649}
1650
1651void
1652ffs_discard_finish(void *vts, int flags)
1653{
1654 struct discarddata *ts = vts;
1655 struct discardopdata *td = NULL;
1656 int res = 0;
1657
1658 /* wait for workqueue to drain */
1659 mutex_enter(&ts->wqlk);
1660 if (ts->wqcnt) {
1661 ts->wqdraining = 1;
1662 res = cv_timedwait(&ts->wqcv, &ts->wqlk, mstohz(5000));
1663 }
1664 mutex_exit(&ts->wqlk);
1665 if (res)
1666 printf("ffs_discarddata drain timeout\n");
1667
1668 mutex_enter(&ts->entrylk);
1669 if (ts->entry) {
1670 td = ts->entry;
1671 ts->entry = NULL;
1672 }
1673 mutex_exit(&ts->entrylk);
1674 if (td) {
1675 /* XXX don't tell disk, its optional */
1676 ffs_blkfree_td(ts->fs, td);
1677#ifdef TRIMDEBUG
1678 printf("finish(%" PRId64 ",%ld)\n", td->bno, td->size);
1679#endif
1680 kmem_free(td, sizeof(*td));
1681 }
1682
1683 cv_destroy(&ts->wqcv);
1684 mutex_destroy(&ts->entrylk);
1685 mutex_destroy(&ts->wqlk);
1686 workqueue_destroy(ts->wq);
1687 kmem_free(ts, sizeof(*ts));
1688}
1689
1690void
1691ffs_blkfree(struct fs *fs, struct vnode *devvp, daddr_t bno, long size,
1692 ino_t inum)
1693{
1694 struct ufsmount *ump;
1695 int error;
1696 dev_t dev;
1697 struct discarddata *ts;
1698 struct discardopdata *td;
1699
1700 dev = devvp->v_rdev;
1701 ump = VFSTOUFS(spec_node_getmountedfs(devvp));
1702 if (ffs_snapblkfree(fs, devvp, bno, size, inum))
1703 return;
1704
1705 error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum);
1706 if (error)
1707 return;
1708
1709 if (!ump->um_discarddata) {
1710 ffs_blkfree_cg(fs, devvp, bno, size);
1711 return;
1712 }
1713
1714#ifdef TRIMDEBUG
1715 printf("blkfree(%" PRId64 ",%ld)\n", bno, size);
1716#endif
1717 ts = ump->um_discarddata;
1718 td = NULL;
1719
1720 mutex_enter(&ts->entrylk);
1721 if (ts->entry) {
1722 td = ts->entry;
1723 /* ffs deallocs backwards, check for prepend only */
1724 if (td->bno == bno + ffs_numfrags(fs, size)
1725 && td->size + size <= ts->maxsize) {
1726 td->bno = bno;
1727 td->size += size;
1728 if (td->size < ts->maxsize) {
1729#ifdef TRIMDEBUG
1730 printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size);
1731#endif
1732 mutex_exit(&ts->entrylk);
1733 return;
1734 }
1735 size = 0; /* mark done */
1736 }
1737 ts->entry = NULL;
1738 }
1739 mutex_exit(&ts->entrylk);
1740
1741 if (td) {
1742#ifdef TRIMDEBUG
1743 printf("enq old(%" PRId64 ",%ld)\n", td->bno, td->size);
1744#endif
1745 mutex_enter(&ts->wqlk);
1746 ts->wqcnt++;
1747 mutex_exit(&ts->wqlk);
1748 workqueue_enqueue(ts->wq, &td->wk, NULL);
1749 }
1750 if (!size)
1751 return;
1752
1753 td = kmem_alloc(sizeof(*td), KM_SLEEP);
1754 td->devvp = devvp;
1755 td->bno = bno;
1756 td->size = size;
1757
1758 if (td->size < ts->maxsize) { /* XXX always the case */
1759 mutex_enter(&ts->entrylk);
1760 if (!ts->entry) { /* possible race? */
1761#ifdef TRIMDEBUG
1762 printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size);
1763#endif
1764 ts->entry = td;
1765 td = NULL;
1766 }
1767 mutex_exit(&ts->entrylk);
1768 }
1769 if (td) {
1770#ifdef TRIMDEBUG
1771 printf("enq new(%" PRId64 ",%ld)\n", td->bno, td->size);
1772#endif
1773 mutex_enter(&ts->wqlk);
1774 ts->wqcnt++;
1775 mutex_exit(&ts->wqlk);
1776 workqueue_enqueue(ts->wq, &td->wk, NULL);
1777 }
1778}
1779
1780/*
1781 * Free a block or fragment from a snapshot cg copy.
1782 *
1783 * The specified block or fragment is placed back in the
1784 * free map. If a fragment is deallocated, a possible
1785 * block reassembly is checked.
1786 *
1787 * => um_lock not held on entry or exit
1788 */
1789void
1790ffs_blkfree_snap(struct fs *fs, struct vnode *devvp, daddr_t bno, long size,
1791 ino_t inum)
1792{
1793 struct cg *cgp;
1794 struct buf *bp;
1795 struct ufsmount *ump;
1796 daddr_t cgblkno;
1797 int error, cg;
1798 dev_t dev;
1799 const bool devvp_is_snapshot = (devvp->v_type != VBLK);
1800 const int needswap = UFS_FSNEEDSWAP(fs);
1801
1802 KASSERT(devvp_is_snapshot);
1803
1804 cg = dtog(fs, bno);
1805 dev = VTOI(devvp)->i_devvp->v_rdev;
1806 ump = VFSTOUFS(devvp->v_mount);
1807 cgblkno = ffs_fragstoblks(fs, cgtod(fs, cg));
1808
1809 error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum);
1810 if (error)
1811 return;
1812
1813 error = bread(devvp, cgblkno, (int)fs->fs_cgsize,
1814 B_MODIFY, &bp);
1815 if (error) {
1816 return;
1817 }
1818 cgp = (struct cg *)bp->b_data;
1819 if (!cg_chkmagic(cgp, needswap)) {
1820 brelse(bp, 0);
1821 return;
1822 }
1823
1824 ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot);
1825
1826 bdwrite(bp);
1827}
1828
1829static void
1830ffs_blkfree_common(struct ufsmount *ump, struct fs *fs, dev_t dev,
1831 struct buf *bp, daddr_t bno, long size, bool devvp_is_snapshot)
1832{
1833 struct cg *cgp;
1834 int32_t fragno, cgbno;
1835 int i, cg, blk, frags, bbase;
1836 u_int8_t *blksfree;
1837 const int needswap = UFS_FSNEEDSWAP(fs);
1838
1839 cg = dtog(fs, bno);
1840 cgp = (struct cg *)bp->b_data;
1841 cgp->cg_old_time = ufs_rw32(time_second, needswap);
1842 if ((fs->fs_magic != FS_UFS1_MAGIC) ||
1843 (fs->fs_old_flags & FS_FLAGS_UPDATED))
1844 cgp->cg_time = ufs_rw64(time_second, needswap);
1845 cgbno = dtogd(fs, bno);
1846 blksfree = cg_blksfree(cgp, needswap);
1847 mutex_enter(&ump->um_lock);
1848 if (size == fs->fs_bsize) {
1849 fragno = ffs_fragstoblks(fs, cgbno);
1850 if (!ffs_isfreeblock(fs, blksfree, fragno)) {
1851 if (devvp_is_snapshot) {
1852 mutex_exit(&ump->um_lock);
1853 return;
1854 }
1855 panic("%s: freeing free block: dev = 0x%llx, block = %"
1856 PRId64 ", fs = %s", __func__,
1857 (unsigned long long)dev, bno, fs->fs_fsmnt);
1858 }
1859 ffs_setblock(fs, blksfree, fragno);
1860 ffs_clusteracct(fs, cgp, fragno, 1);
1861 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
1862 fs->fs_cstotal.cs_nbfree++;
1863 fs->fs_cs(fs, cg).cs_nbfree++;
1864 if ((fs->fs_magic == FS_UFS1_MAGIC) &&
1865 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
1866 i = old_cbtocylno(fs, cgbno);
1867 KASSERT(i >= 0);
1868 KASSERT(i < fs->fs_old_ncyl);
1869 KASSERT(old_cbtorpos(fs, cgbno) >= 0);
1870 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, cgbno) < fs->fs_old_nrpos);
1871 ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs, cgbno)], 1,
1872 needswap);
1873 ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap);
1874 }
1875 } else {
1876 bbase = cgbno - ffs_fragnum(fs, cgbno);
1877 /*
1878 * decrement the counts associated with the old frags
1879 */
1880 blk = blkmap(fs, blksfree, bbase);
1881 ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
1882 /*
1883 * deallocate the fragment
1884 */
1885 frags = ffs_numfrags(fs, size);
1886 for (i = 0; i < frags; i++) {
1887 if (isset(blksfree, cgbno + i)) {
1888 panic("%s: freeing free frag: "
1889 "dev = 0x%llx, block = %" PRId64
1890 ", fs = %s", __func__,
1891 (unsigned long long)dev, bno + i,
1892 fs->fs_fsmnt);
1893 }
1894 setbit(blksfree, cgbno + i);
1895 }
1896 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
1897 fs->fs_cstotal.cs_nffree += i;
1898 fs->fs_cs(fs, cg).cs_nffree += i;
1899 /*
1900 * add back in counts associated with the new frags
1901 */
1902 blk = blkmap(fs, blksfree, bbase);
1903 ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
1904 /*
1905 * if a complete block has been reassembled, account for it
1906 */
1907 fragno = ffs_fragstoblks(fs, bbase);
1908 if (ffs_isblock(fs, blksfree, fragno)) {
1909 ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
1910 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1911 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1912 ffs_clusteracct(fs, cgp, fragno, 1);
1913 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
1914 fs->fs_cstotal.cs_nbfree++;
1915 fs->fs_cs(fs, cg).cs_nbfree++;
1916 if ((fs->fs_magic == FS_UFS1_MAGIC) &&
1917 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
1918 i = old_cbtocylno(fs, bbase);
1919 KASSERT(i >= 0);
1920 KASSERT(i < fs->fs_old_ncyl);
1921 KASSERT(old_cbtorpos(fs, bbase) >= 0);
1922 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bbase) < fs->fs_old_nrpos);
1923 ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs,
1924 bbase)], 1, needswap);
1925 ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap);
1926 }
1927 }
1928 }
1929 fs->fs_fmod = 1;
1930 ACTIVECG_CLR(fs, cg);
1931 mutex_exit(&ump->um_lock);
1932}
1933
1934/*
1935 * Free an inode.
1936 */
1937int
1938ffs_vfree(struct vnode *vp, ino_t ino, int mode)
1939{
1940
1941 return ffs_freefile(vp->v_mount, ino, mode);
1942}
1943
1944/*
1945 * Do the actual free operation.
1946 * The specified inode is placed back in the free map.
1947 *
1948 * => um_lock not held on entry or exit
1949 */
1950int
1951ffs_freefile(struct mount *mp, ino_t ino, int mode)
1952{
1953 struct ufsmount *ump = VFSTOUFS(mp);
1954 struct fs *fs = ump->um_fs;
1955 struct vnode *devvp;
1956 struct cg *cgp;
1957 struct buf *bp;
1958 int error, cg;
1959 daddr_t cgbno;
1960 dev_t dev;
1961 const int needswap = UFS_FSNEEDSWAP(fs);
1962
1963 cg = ino_to_cg(fs, ino);
1964 devvp = ump->um_devvp;
1965 dev = devvp->v_rdev;
1966 cgbno = FFS_FSBTODB(fs, cgtod(fs, cg));
1967
1968 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1969 panic("%s: range: dev = 0x%llx, ino = %llu, fs = %s", __func__,
1970 (long long)dev, (unsigned long long)ino, fs->fs_fsmnt);
1971 error = bread(devvp, cgbno, (int)fs->fs_cgsize,
1972 B_MODIFY, &bp);
1973 if (error) {
1974 return (error);
1975 }
1976 cgp = (struct cg *)bp->b_data;
1977 if (!cg_chkmagic(cgp, needswap)) {
1978 brelse(bp, 0);
1979 return (0);
1980 }
1981
1982 ffs_freefile_common(ump, fs, dev, bp, ino, mode, false);
1983
1984 bdwrite(bp);
1985
1986 return 0;
1987}
1988
1989int
1990ffs_freefile_snap(struct fs *fs, struct vnode *devvp, ino_t ino, int mode)
1991{
1992 struct ufsmount *ump;
1993 struct cg *cgp;
1994 struct buf *bp;
1995 int error, cg;
1996 daddr_t cgbno;
1997 dev_t dev;
1998 const int needswap = UFS_FSNEEDSWAP(fs);
1999
2000 KASSERT(devvp->v_type != VBLK);
2001
2002 cg = ino_to_cg(fs, ino);
2003 dev = VTOI(devvp)->i_devvp->v_rdev;
2004 ump = VFSTOUFS(devvp->v_mount);
2005 cgbno = ffs_fragstoblks(fs, cgtod(fs, cg));
2006 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
2007 panic("%s: range: dev = 0x%llx, ino = %llu, fs = %s", __func__,
2008 (unsigned long long)dev, (unsigned long long)ino,
2009 fs->fs_fsmnt);
2010 error = bread(devvp, cgbno, (int)fs->fs_cgsize,
2011 B_MODIFY, &bp);
2012 if (error) {
2013 return (error);
2014 }
2015 cgp = (struct cg *)bp->b_data;
2016 if (!cg_chkmagic(cgp, needswap)) {
2017 brelse(bp, 0);
2018 return (0);
2019 }
2020 ffs_freefile_common(ump, fs, dev, bp, ino, mode, true);
2021
2022 bdwrite(bp);
2023
2024 return 0;
2025}
2026
2027static void
2028ffs_freefile_common(struct ufsmount *ump, struct fs *fs, dev_t dev,
2029 struct buf *bp, ino_t ino, int mode, bool devvp_is_snapshot)
2030{
2031 int cg;
2032 struct cg *cgp;
2033 u_int8_t *inosused;
2034 const int needswap = UFS_FSNEEDSWAP(fs);
2035
2036 cg = ino_to_cg(fs, ino);
2037 cgp = (struct cg *)bp->b_data;
2038 cgp->cg_old_time = ufs_rw32(time_second, needswap);
2039 if ((fs->fs_magic != FS_UFS1_MAGIC) ||
2040 (fs->fs_old_flags & FS_FLAGS_UPDATED))
2041 cgp->cg_time = ufs_rw64(time_second, needswap);
2042 inosused = cg_inosused(cgp, needswap);
2043 ino %= fs->fs_ipg;
2044 if (isclr(inosused, ino)) {
2045 printf("ifree: dev = 0x%llx, ino = %llu, fs = %s\n",
2046 (unsigned long long)dev, (unsigned long long)ino +
2047 cg * fs->fs_ipg, fs->fs_fsmnt);
2048 if (fs->fs_ronly == 0)
2049 panic("%s: freeing free inode", __func__);
2050 }
2051 clrbit(inosused, ino);
2052 if (!devvp_is_snapshot)
2053 UFS_WAPBL_UNREGISTER_INODE(ump->um_mountp,
2054 ino + cg * fs->fs_ipg, mode);
2055 if (ino < ufs_rw32(cgp->cg_irotor, needswap))
2056 cgp->cg_irotor = ufs_rw32(ino, needswap);
2057 ufs_add32(cgp->cg_cs.cs_nifree, 1, needswap);
2058 mutex_enter(&ump->um_lock);
2059 fs->fs_cstotal.cs_nifree++;
2060 fs->fs_cs(fs, cg).cs_nifree++;
2061 if ((mode & IFMT) == IFDIR) {
2062 ufs_add32(cgp->cg_cs.cs_ndir, -1, needswap);
2063 fs->fs_cstotal.cs_ndir--;
2064 fs->fs_cs(fs, cg).cs_ndir--;
2065 }
2066 fs->fs_fmod = 1;
2067 ACTIVECG_CLR(fs, cg);
2068 mutex_exit(&ump->um_lock);
2069}
2070
2071/*
2072 * Check to see if a file is free.
2073 */
2074int
2075ffs_checkfreefile(struct fs *fs, struct vnode *devvp, ino_t ino)
2076{
2077 struct cg *cgp;
2078 struct buf *bp;
2079 daddr_t cgbno;
2080 int ret, cg;
2081 u_int8_t *inosused;
2082 const bool devvp_is_snapshot = (devvp->v_type != VBLK);
2083
2084 KASSERT(devvp_is_snapshot);
2085
2086 cg = ino_to_cg(fs, ino);
2087 if (devvp_is_snapshot)
2088 cgbno = ffs_fragstoblks(fs, cgtod(fs, cg));
2089 else
2090 cgbno = FFS_FSBTODB(fs, cgtod(fs, cg));
2091 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
2092 return 1;
2093 if (bread(devvp, cgbno, (int)fs->fs_cgsize, 0, &bp)) {
2094 return 1;
2095 }
2096 cgp = (struct cg *)bp->b_data;
2097 if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) {
2098 brelse(bp, 0);
2099 return 1;
2100 }
2101 inosused = cg_inosused(cgp, UFS_FSNEEDSWAP(fs));
2102 ino %= fs->fs_ipg;
2103 ret = isclr(inosused, ino);
2104 brelse(bp, 0);
2105 return ret;
2106}
2107
2108/*
2109 * Find a block of the specified size in the specified cylinder group.
2110 *
2111 * It is a panic if a request is made to find a block if none are
2112 * available.
2113 */
2114static int32_t
2115ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
2116{
2117 int32_t bno;
2118 int start, len, loc, i;
2119 int blk, field, subfield, pos;
2120 int ostart, olen;
2121 u_int8_t *blksfree;
2122 const int needswap = UFS_FSNEEDSWAP(fs);
2123
2124 /* KASSERT(mutex_owned(&ump->um_lock)); */
2125
2126 /*
2127 * find the fragment by searching through the free block
2128 * map for an appropriate bit pattern
2129 */
2130 if (bpref)
2131 start = dtogd(fs, bpref) / NBBY;
2132 else
2133 start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
2134 blksfree = cg_blksfree(cgp, needswap);
2135 len = howmany(fs->fs_fpg, NBBY) - start;
2136 ostart = start;
2137 olen = len;
2138 loc = scanc((u_int)len,
2139 (const u_char *)&blksfree[start],
2140 (const u_char *)fragtbl[fs->fs_frag],
2141 (1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1)))));
2142 if (loc == 0) {
2143 len = start + 1;
2144 start = 0;
2145 loc = scanc((u_int)len,
2146 (const u_char *)&blksfree[0],
2147 (const u_char *)fragtbl[fs->fs_frag],
2148 (1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1)))));
2149 if (loc == 0) {
2150 panic("%s: map corrupted: start=%d, len=%d, "
2151 "fs = %s, offset=%d/%ld, cg %d", __func__,
2152 ostart, olen, fs->fs_fsmnt,
2153 ufs_rw32(cgp->cg_freeoff, needswap),
2154 (long)blksfree - (long)cgp, cgp->cg_cgx);
2155 /* NOTREACHED */
2156 }
2157 }
2158 bno = (start + len - loc) * NBBY;
2159 cgp->cg_frotor = ufs_rw32(bno, needswap);
2160 /*
2161 * found the byte in the map
2162 * sift through the bits to find the selected frag
2163 */
2164 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
2165 blk = blkmap(fs, blksfree, bno);
2166 blk <<= 1;
2167 field = around[allocsiz];
2168 subfield = inside[allocsiz];
2169 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
2170 if ((blk & field) == subfield)
2171 return (bno + pos);
2172 field <<= 1;
2173 subfield <<= 1;
2174 }
2175 }
2176 panic("%s: block not in map: bno=%d, fs=%s", __func__,
2177 bno, fs->fs_fsmnt);
2178 /* return (-1); */
2179}
2180
2181/*
2182 * Fserr prints the name of a file system with an error diagnostic.
2183 *
2184 * The form of the error message is:
2185 * fs: error message
2186 */
2187static void
2188ffs_fserr(struct fs *fs, kauth_cred_t cred, const char *cp)
2189{
2190 KASSERT(cred != NULL);
2191
2192 if (cred == NOCRED || cred == FSCRED) {
2193 log(LOG_ERR, "pid %d, command %s, on %s: %s\n",
2194 curproc->p_pid, curproc->p_comm,
2195 fs->fs_fsmnt, cp);
2196 } else {
2197 log(LOG_ERR, "uid %d, pid %d, command %s, on %s: %s\n",
2198 kauth_cred_getuid(cred), curproc->p_pid, curproc->p_comm,
2199 fs->fs_fsmnt, cp);
2200 }
2201}
2202