1/* $NetBSD: ulfs_lookup.c,v 1.39 2016/06/20 02:25:03 dholland Exp $ */
2/* from NetBSD: ufs_lookup.c,v 1.135 2015/07/11 11:04:48 mlelstv */
3
4/*
5 * Copyright (c) 1989, 1993
6 * The Regents of the University of California. All rights reserved.
7 * (c) UNIX System Laboratories, Inc.
8 * All or some portions of this file are derived from material licensed
9 * to the University of California by American Telephone and Telegraph
10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11 * the permission of UNIX System Laboratories, Inc.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * @(#)ufs_lookup.c 8.9 (Berkeley) 8/11/94
38 */
39
40#include <sys/cdefs.h>
41__KERNEL_RCSID(0, "$NetBSD: ulfs_lookup.c,v 1.39 2016/06/20 02:25:03 dholland Exp $");
42
43#ifdef _KERNEL_OPT
44#include "opt_lfs.h"
45#endif
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/namei.h>
50#include <sys/buf.h>
51#include <sys/file.h>
52#include <sys/stat.h>
53#include <sys/mount.h>
54#include <sys/vnode.h>
55#include <sys/kernel.h>
56#include <sys/kauth.h>
57#include <sys/fstrans.h>
58#include <sys/proc.h>
59#include <sys/kmem.h>
60
61#include <ufs/lfs/lfs.h>
62#include <ufs/lfs/lfs_accessors.h>
63#include <ufs/lfs/lfs_extern.h>
64
65#include <ufs/lfs/ulfs_inode.h>
66#ifdef LFS_DIRHASH
67#include <ufs/lfs/ulfs_dirhash.h>
68#endif
69#include <ufs/lfs/ulfsmount.h>
70#include <ufs/lfs/ulfs_extern.h>
71#include <ufs/lfs/ulfs_bswap.h>
72
73#include <miscfs/genfs/genfs.h>
74
75#ifdef DIAGNOSTIC
76int lfs_dirchk = 1;
77#else
78int lfs_dirchk = 0;
79#endif
80
81/*
82 * Convert a component of a pathname into a pointer to a locked inode.
83 * This is a very central and rather complicated routine.
84 * If the file system is not maintained in a strict tree hierarchy,
85 * this can result in a deadlock situation (see comments in code below).
86 *
87 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
88 * on whether the name is to be looked up, created, renamed, or deleted.
89 * When CREATE, RENAME, or DELETE is specified, information usable in
90 * creating, renaming, or deleting a directory entry may be calculated.
91 * If flag has LOCKPARENT or'ed into it and the target of the pathname
92 * exists, lookup returns both the target and its parent directory locked.
93 * When creating or renaming and LOCKPARENT is specified, the target may
94 * not be ".". When deleting and LOCKPARENT is specified, the target may
95 * be "."., but the caller must check to ensure it does an vrele and vput
96 * instead of two vputs.
97 *
98 * Overall outline of ulfs_lookup:
99 *
100 * check accessibility of directory
101 * look for name in cache, if found, then if at end of path
102 * and deleting or creating, drop it, else return name
103 * search for name in directory, to found or notfound
104 * notfound:
105 * if creating, return locked directory, leaving info on available slots
106 * else return error
107 * found:
108 * if at end of path and deleting, return information to allow delete
109 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
110 * inode and return info to allow rewrite
111 * if not at end, add name to cache; if at end and neither creating
112 * nor deleting, add name to cache
113 */
114int
115ulfs_lookup(void *v)
116{
117 struct vop_lookup_v2_args /* {
118 struct vnode *a_dvp;
119 struct vnode **a_vpp;
120 struct componentname *a_cnp;
121 } */ *ap = v;
122 struct vnode *vdp = ap->a_dvp; /* vnode for directory being searched */
123 struct inode *dp = VTOI(vdp); /* inode for directory being searched */
124 struct buf *bp; /* a buffer of directory entries */
125 LFS_DIRHEADER *ep; /* the current directory entry */
126 int entryoffsetinblock; /* offset of ep in bp's buffer */
127 enum {
128 NONE, /* need to search a slot for our new entry */
129 COMPACT, /* a compaction can make a slot in the current
130 DIRBLKSIZ block */
131 FOUND, /* found a slot (or no need to search) */
132 } slotstatus;
133 doff_t slotoffset; /* offset of area with free space.
134 a special value -1 for invalid */
135 int slotsize; /* size of area at slotoffset */
136 int slotfreespace; /* accumulated amount of space free in
137 the current DIRBLKSIZ block */
138 int slotneeded; /* size of the entry we're seeking */
139 int numdirpasses; /* strategy for directory search */
140 doff_t endsearch; /* offset to end directory search */
141 doff_t prevoff; /* previous value of ulr_offset */
142 struct vnode *tdp; /* returned by vcache_get */
143 doff_t enduseful; /* pointer past last used dir slot.
144 used for directory truncation. */
145 u_long bmask; /* block offset mask */
146 int error;
147 struct vnode **vpp = ap->a_vpp;
148 struct componentname *cnp = ap->a_cnp;
149 kauth_cred_t cred = cnp->cn_cred;
150 int flags;
151 int nameiop = cnp->cn_nameiop;
152 struct lfs *fs = dp->i_lfs;
153 int dirblksiz = fs->um_dirblksiz;
154 ino_t foundino;
155 struct ulfs_lookup_results *results;
156 int iswhiteout; /* temp result from cache_lookup() */
157
158 flags = cnp->cn_flags;
159
160 bp = NULL;
161 slotoffset = -1;
162 *vpp = NULL;
163 endsearch = 0; /* silence compiler warning */
164
165 /*
166 * Produce the auxiliary lookup results into i_crap. Increment
167 * its serial number so elsewhere we can tell if we're using
168 * stale results. This should not be done this way. XXX.
169 */
170 results = &dp->i_crap;
171 dp->i_crapcounter++;
172
173 /*
174 * Check accessiblity of directory.
175 */
176 if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
177 return (error);
178
179 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
180 (nameiop == DELETE || nameiop == RENAME))
181 return (EROFS);
182
183 /*
184 * We now have a segment name to search for, and a directory to search.
185 *
186 * Before tediously performing a linear scan of the directory,
187 * check the name cache to see if the directory/name pair
188 * we are looking for is known already.
189 */
190 if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
191 cnp->cn_nameiop, cnp->cn_flags, &iswhiteout, vpp)) {
192 if (iswhiteout) {
193 cnp->cn_flags |= ISWHITEOUT;
194 }
195 return *vpp == NULLVP ? ENOENT : 0;
196 }
197 if (iswhiteout) {
198 /*
199 * The namecache set iswhiteout without finding a
200 * cache entry. As of this writing (20121014), this
201 * can happen if there was a whiteout entry that has
202 * been invalidated by the lookup. It is not clear if
203 * it is correct to set ISWHITEOUT in this case or
204 * not; however, doing so retains the prior behavior,
205 * so we'll go with that until some clearer answer
206 * appears. XXX
207 */
208 cnp->cn_flags |= ISWHITEOUT;
209 }
210
211 fstrans_start(vdp->v_mount, FSTRANS_SHARED);
212
213 /*
214 * Suppress search for slots unless creating
215 * file and at end of pathname, in which case
216 * we watch for a place to put the new file in
217 * case it doesn't already exist.
218 */
219 slotstatus = FOUND;
220 slotfreespace = slotsize = slotneeded = 0;
221 if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
222 slotstatus = NONE;
223 slotneeded = LFS_DIRECTSIZ(fs, cnp->cn_namelen);
224 }
225
226 /*
227 * If there is cached information on a previous search of
228 * this directory, pick up where we last left off.
229 * We cache only lookups as these are the most common
230 * and have the greatest payoff. Caching CREATE has little
231 * benefit as it usually must search the entire directory
232 * to determine that the entry does not exist. Caching the
233 * location of the last DELETE or RENAME has not reduced
234 * profiling time and hence has been removed in the interest
235 * of simplicity.
236 */
237 bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
238
239#ifdef LFS_DIRHASH
240 /*
241 * Use dirhash for fast operations on large directories. The logic
242 * to determine whether to hash the directory is contained within
243 * ulfsdirhash_build(); a zero return means that it decided to hash
244 * this directory and it successfully built up the hash table.
245 */
246 if (ulfsdirhash_build(dp) == 0) {
247 /* Look for a free slot if needed. */
248 enduseful = dp->i_size;
249 if (slotstatus != FOUND) {
250 slotoffset = ulfsdirhash_findfree(dp, slotneeded,
251 &slotsize);
252 if (slotoffset >= 0) {
253 slotstatus = COMPACT;
254 enduseful = ulfsdirhash_enduseful(dp);
255 if (enduseful < 0)
256 enduseful = dp->i_size;
257 }
258 }
259 /* Look up the component. */
260 numdirpasses = 1;
261 entryoffsetinblock = 0; /* silence compiler warning */
262 switch (ulfsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
263 &results->ulr_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
264 case 0:
265 ep = (LFS_DIRHEADER *)((char *)bp->b_data +
266 (results->ulr_offset & bmask));
267 goto foundentry;
268 case ENOENT:
269 results->ulr_offset = roundup(dp->i_size, dirblksiz);
270 goto notfound;
271 default:
272 /* Something failed; just do a linear search. */
273 break;
274 }
275 }
276#endif /* LFS_DIRHASH */
277
278 if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
279 results->ulr_diroff >= dp->i_size) {
280 entryoffsetinblock = 0;
281 results->ulr_offset = 0;
282 numdirpasses = 1;
283 } else {
284 results->ulr_offset = results->ulr_diroff;
285 if ((entryoffsetinblock = results->ulr_offset & bmask) &&
286 (error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
287 NULL, &bp, false)))
288 goto out;
289 numdirpasses = 2;
290 namecache_count_2passes();
291 }
292 prevoff = results->ulr_offset;
293 endsearch = roundup(dp->i_size, dirblksiz);
294 enduseful = 0;
295
296searchloop:
297 while (results->ulr_offset < endsearch) {
298 if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
299 preempt();
300 /*
301 * If necessary, get the next directory block.
302 */
303 if ((results->ulr_offset & bmask) == 0) {
304 if (bp != NULL)
305 brelse(bp, 0);
306 error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
307 NULL, &bp, false);
308 if (error)
309 goto out;
310 entryoffsetinblock = 0;
311 }
312 /*
313 * If still looking for a slot, and at a DIRBLKSIZ
314 * boundary, have to start looking for free space again.
315 */
316 if (slotstatus == NONE &&
317 (entryoffsetinblock & (dirblksiz - 1)) == 0) {
318 slotoffset = -1;
319 slotfreespace = 0;
320 }
321 /*
322 * Get pointer to next entry.
323 * Full validation checks are slow, so we only check
324 * enough to insure forward progress through the
325 * directory. Complete checks can be run by patching
326 * "lfs_dirchk" to be true.
327 */
328 KASSERT(bp != NULL);
329 ep = (LFS_DIRHEADER *)((char *)bp->b_data + entryoffsetinblock);
330 if (lfs_dir_getreclen(fs, ep) == 0 ||
331 (lfs_dirchk && ulfs_dirbadentry(vdp, ep, entryoffsetinblock))) {
332 int i;
333
334 ulfs_dirbad(dp, results->ulr_offset, "mangled entry");
335 i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
336 results->ulr_offset += i;
337 entryoffsetinblock += i;
338 continue;
339 }
340
341 /*
342 * If an appropriate sized slot has not yet been found,
343 * check to see if one is available. Also accumulate space
344 * in the current block so that we can determine if
345 * compaction is viable.
346 */
347 if (slotstatus != FOUND) {
348 int size = lfs_dir_getreclen(fs, ep);
349
350 if (lfs_dir_getino(fs, ep) != 0)
351 size -= LFS_DIRSIZ(fs, ep);
352 if (size > 0) {
353 if (size >= slotneeded) {
354 slotstatus = FOUND;
355 slotoffset = results->ulr_offset;
356 slotsize = lfs_dir_getreclen(fs, ep);
357 } else if (slotstatus == NONE) {
358 slotfreespace += size;
359 if (slotoffset == -1)
360 slotoffset = results->ulr_offset;
361 if (slotfreespace >= slotneeded) {
362 slotstatus = COMPACT;
363 slotsize = results->ulr_offset +
364 lfs_dir_getreclen(fs, ep) -
365 slotoffset;
366 }
367 }
368 }
369 }
370
371 /*
372 * Check for a name match.
373 */
374 if (lfs_dir_getino(fs, ep)) {
375 int namlen;
376
377 namlen = lfs_dir_getnamlen(fs, ep);
378 if (namlen == cnp->cn_namelen &&
379 !memcmp(cnp->cn_nameptr, lfs_dir_nameptr(fs, ep),
380 (unsigned)namlen)) {
381#ifdef LFS_DIRHASH
382foundentry:
383#endif
384 /*
385 * Save directory entry's inode number and
386 * reclen, and release directory buffer.
387 */
388 if (!FSFMT(vdp) && lfs_dir_gettype(fs, ep) == LFS_DT_WHT) {
389 slotstatus = FOUND;
390 slotoffset = results->ulr_offset;
391 slotsize = lfs_dir_getreclen(fs, ep);
392 results->ulr_reclen = slotsize;
393 /*
394 * This is used to set
395 * results->ulr_endoff,
396 * which may be used by ulfs_direnter()
397 * as a length to truncate the
398 * directory to. Therefore, it must
399 * point past the end of the last
400 * non-empty directory entry. We don't
401 * know where that is in this case, so
402 * we effectively disable shrinking by
403 * using the existing size of the
404 * directory.
405 *
406 * Note that we wouldn't expect to
407 * shrink the directory while rewriting
408 * an existing entry anyway.
409 */
410 enduseful = endsearch;
411 cnp->cn_flags |= ISWHITEOUT;
412 numdirpasses--;
413 goto notfound;
414 }
415 foundino = lfs_dir_getino(fs, ep);
416 results->ulr_reclen = lfs_dir_getreclen(fs, ep);
417 goto found;
418 }
419 }
420 prevoff = results->ulr_offset;
421 results->ulr_offset += lfs_dir_getreclen(fs, ep);
422 entryoffsetinblock += lfs_dir_getreclen(fs, ep);
423 if (lfs_dir_getino(fs, ep))
424 enduseful = results->ulr_offset;
425 }
426notfound:
427 /*
428 * If we started in the middle of the directory and failed
429 * to find our target, we must check the beginning as well.
430 */
431 if (numdirpasses == 2) {
432 numdirpasses--;
433 results->ulr_offset = 0;
434 endsearch = results->ulr_diroff;
435 goto searchloop;
436 }
437 if (bp != NULL)
438 brelse(bp, 0);
439 /*
440 * If creating, and at end of pathname and current
441 * directory has not been removed, then can consider
442 * allowing file to be created.
443 */
444 if ((nameiop == CREATE || nameiop == RENAME ||
445 (nameiop == DELETE &&
446 (cnp->cn_flags & DOWHITEOUT) &&
447 (cnp->cn_flags & ISWHITEOUT))) &&
448 (flags & ISLASTCN) && dp->i_nlink != 0) {
449 /*
450 * Access for write is interpreted as allowing
451 * creation of files in the directory.
452 */
453 error = VOP_ACCESS(vdp, VWRITE, cred);
454 if (error)
455 goto out;
456 /*
457 * Return an indication of where the new directory
458 * entry should be put. If we didn't find a slot,
459 * then set results->ulr_count to 0 indicating
460 * that the new slot belongs at the end of the
461 * directory. If we found a slot, then the new entry
462 * can be put in the range from results->ulr_offset to
463 * results->ulr_offset + results->ulr_count.
464 */
465 if (slotstatus == NONE) {
466 results->ulr_offset = roundup(dp->i_size, dirblksiz);
467 results->ulr_count = 0;
468 enduseful = results->ulr_offset;
469 } else if (nameiop == DELETE) {
470 results->ulr_offset = slotoffset;
471 if ((results->ulr_offset & (dirblksiz - 1)) == 0)
472 results->ulr_count = 0;
473 else
474 results->ulr_count =
475 results->ulr_offset - prevoff;
476 } else {
477 results->ulr_offset = slotoffset;
478 results->ulr_count = slotsize;
479 if (enduseful < slotoffset + slotsize)
480 enduseful = slotoffset + slotsize;
481 }
482 results->ulr_endoff = roundup(enduseful, dirblksiz);
483#if 0 /* commented out by dbj. none of the on disk fields changed */
484 dp->i_flag |= IN_CHANGE | IN_UPDATE;
485#endif
486 /*
487 * We return with the directory locked, so that
488 * the parameters we set up above will still be
489 * valid if we actually decide to do a direnter().
490 * We return ni_vp == NULL to indicate that the entry
491 * does not currently exist; we leave a pointer to
492 * the (locked) directory inode in ndp->ni_dvp.
493 *
494 * NB - if the directory is unlocked, then this
495 * information cannot be used.
496 */
497 error = EJUSTRETURN;
498 goto out;
499 }
500 /*
501 * Insert name into cache (as non-existent) if appropriate.
502 */
503 if (nameiop != CREATE) {
504 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
505 cnp->cn_flags);
506 }
507 error = ENOENT;
508 goto out;
509
510found:
511 if (numdirpasses == 2)
512 namecache_count_pass2();
513 /*
514 * Check that directory length properly reflects presence
515 * of this entry.
516 */
517 if (results->ulr_offset + LFS_DIRSIZ(fs, ep) > dp->i_size) {
518 ulfs_dirbad(dp, results->ulr_offset, "i_size too small");
519 dp->i_size =
520 results->ulr_offset + LFS_DIRSIZ(fs, ep);
521 DIP_ASSIGN(dp, size, dp->i_size);
522 dp->i_flag |= IN_CHANGE | IN_UPDATE;
523 }
524 brelse(bp, 0);
525
526 /*
527 * Found component in pathname.
528 * If the final component of path name, save information
529 * in the cache as to where the entry was found.
530 */
531 if ((flags & ISLASTCN) && nameiop == LOOKUP)
532 results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
533
534 /*
535 * If deleting, and at end of pathname, return
536 * parameters which can be used to remove file.
537 * Lock the inode, being careful with ".".
538 */
539 if (nameiop == DELETE && (flags & ISLASTCN)) {
540 /*
541 * Return pointer to current entry in results->ulr_offset,
542 * and distance past previous entry (if there
543 * is a previous entry in this block) in results->ulr_count.
544 * Save directory inode pointer in ndp->ni_dvp for dirremove().
545 */
546 if ((results->ulr_offset & (dirblksiz - 1)) == 0)
547 results->ulr_count = 0;
548 else
549 results->ulr_count = results->ulr_offset - prevoff;
550 if (dp->i_number == foundino) {
551 vref(vdp);
552 tdp = vdp;
553 } else {
554 error = vcache_get(vdp->v_mount,
555 &foundino, sizeof(foundino), &tdp);
556 if (error)
557 goto out;
558 }
559 /*
560 * Write access to directory required to delete files.
561 */
562 error = VOP_ACCESS(vdp, VWRITE, cred);
563 if (error) {
564 vrele(tdp);
565 goto out;
566 }
567 /*
568 * If directory is "sticky", then user must own
569 * the directory, or the file in it, else she
570 * may not delete it (unless she's root). This
571 * implements append-only directories.
572 */
573 if (dp->i_mode & ISVTX) {
574 error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
575 tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
576 VTOI(tdp)->i_uid));
577 if (error) {
578 vrele(tdp);
579 error = EPERM;
580 goto out;
581 }
582 }
583 *vpp = tdp;
584 error = 0;
585 goto out;
586 }
587
588 /*
589 * If rewriting (RENAME), return the inode and the
590 * information required to rewrite the present directory
591 * Must get inode of directory entry to verify it's a
592 * regular file, or empty directory.
593 */
594 if (nameiop == RENAME && (flags & ISLASTCN)) {
595 error = VOP_ACCESS(vdp, VWRITE, cred);
596 if (error)
597 goto out;
598 /*
599 * Careful about locking second inode.
600 * This can only occur if the target is ".".
601 */
602 if (dp->i_number == foundino) {
603 error = EISDIR;
604 goto out;
605 }
606 error = vcache_get(vdp->v_mount,
607 &foundino, sizeof(foundino), &tdp);
608 if (error)
609 goto out;
610 *vpp = tdp;
611 error = 0;
612 goto out;
613 }
614
615 if (dp->i_number == foundino) {
616 vref(vdp); /* we want ourself, ie "." */
617 *vpp = vdp;
618 } else {
619 error = vcache_get(vdp->v_mount,
620 &foundino, sizeof(foundino), &tdp);
621 if (error)
622 goto out;
623 *vpp = tdp;
624 }
625
626 /*
627 * Insert name into cache if appropriate.
628 */
629 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
630 error = 0;
631
632out:
633 fstrans_done(vdp->v_mount);
634 return error;
635}
636
637void
638ulfs_dirbad(struct inode *ip, doff_t offset, const char *how)
639{
640 struct mount *mp;
641
642 mp = ITOV(ip)->v_mount;
643 printf("%s: bad dir ino %llu at offset %d: %s\n",
644 mp->mnt_stat.f_mntonname, (unsigned long long)ip->i_number,
645 offset, how);
646 if ((mp->mnt_flag & MNT_RDONLY) == 0)
647 panic("bad dir");
648}
649
650/*
651 * Do consistency checking on a directory entry:
652 * record length must be multiple of 4
653 * entry must fit in rest of its DIRBLKSIZ block
654 * record must be large enough to contain entry
655 * name is not longer than LFS_MAXNAMLEN
656 * name must be as long as advertised, and null terminated
657 */
658int
659ulfs_dirbadentry(struct vnode *dp, LFS_DIRHEADER *ep, int entryoffsetinblock)
660{
661 int i;
662 int namlen;
663 unsigned reclen;
664 struct ulfsmount *ump = VFSTOULFS(dp->v_mount);
665 struct lfs *fs = ump->um_lfs;
666 int dirblksiz = fs->um_dirblksiz;
667 const char *name;
668
669 namlen = lfs_dir_getnamlen(fs, ep);
670 reclen = lfs_dir_getreclen(fs, ep);
671 if ((reclen & 0x3) != 0 ||
672 reclen > dirblksiz - (entryoffsetinblock & (dirblksiz - 1)) ||
673 reclen < LFS_DIRSIZ(fs, ep) || namlen > LFS_MAXNAMLEN) {
674 /*return (1); */
675 printf("First bad, reclen=%#x, DIRSIZ=%lu, namlen=%d, "
676 "flags=%#x, entryoffsetinblock=%d, dirblksiz = %d\n",
677 lfs_dir_getreclen(fs, ep),
678 (u_long)LFS_DIRSIZ(fs, ep),
679 namlen, dp->v_mount->mnt_flag, entryoffsetinblock,
680 dirblksiz);
681 goto bad;
682 }
683 if (lfs_dir_getino(fs, ep) == 0)
684 return (0);
685 name = lfs_dir_nameptr(fs, ep);
686 for (i = 0; i < namlen; i++)
687 if (name[i] == '\0') {
688 /*return (1); */
689 printf("Second bad\n");
690 goto bad;
691 }
692 if (name[i])
693 goto bad;
694 return (0);
695bad:
696 return (1);
697}
698
699/*
700 * Assign the contents of directory entry DIRP, on volume FS.
701 *
702 * NAME/NAMLEN is the name, which is not necessarily null terminated.
703 * INUM is the inode number, and DTYPE is the type code (LFS_DT_*).
704 *
705 * Note that these values typically come from:
706 * cnp->cn_nameptr
707 * cnp->cn_namelen
708 * ip->i_number
709 * LFS_IFTODT(ip->i_mode)
710 *
711 * Does not set d_reclen.
712 */
713static void
714ulfs_direntry_assign(struct lfs *fs, LFS_DIRHEADER *dirp,
715 const char *name, size_t namlen,
716 ino_t inum, unsigned dtype)
717{
718 lfs_dir_setino(fs, dirp, inum);
719 lfs_dir_setnamlen(fs, dirp, namlen);
720 lfs_dir_settype(fs, dirp, dtype);
721 memcpy(lfs_dir_nameptr(fs, dirp), name, namlen);
722 lfs_dir_nameptr(fs, dirp)[namlen] = '\0';
723}
724
725/*
726 * Write a directory entry after a call to namei, using the parameters
727 * that ulfs_lookup left in nameidata and in the ulfs_lookup_results.
728 *
729 * DVP is the directory to be updated. It must be locked.
730 * ULR is the ulfs_lookup_results structure from the final lookup step.
731 * TVP is not used. (XXX: why is it here? remove it)
732 * CNP is the componentname from the final lookup step.
733 * INUM is the inode number to insert into the new directory entry.
734 * DTYPE is the type code (LFS_DT_*) to insert into the new directory entry.
735 * NEWDIRBP is not used and (XXX) should be removed. The previous
736 * comment here said it was used by the now-removed softupdates code.
737 *
738 * The link count of the target inode is *not* incremented; the
739 * caller does that.
740 *
741 * If ulr->ulr_count is 0, ulfs_lookup did not find space to insert the
742 * directory entry. ulr_offset, which is the place to put the entry,
743 * should be on a block boundary (and should be at the end of the
744 * directory AFAIK) and a fresh block is allocated to put the new
745 * directory entry in.
746 *
747 * If ulr->ulr_count is not zero, ulfs_lookup found a slot to insert
748 * the entry into. This slot ranges from ulr_offset to ulr_offset +
749 * ulr_count. However, this slot may already be partially populated
750 * requiring compaction. See notes below.
751 *
752 * Furthermore, if ulr_count is not zero and ulr_endoff is not the
753 * same as i_size, the directory is truncated to size ulr_endoff.
754 */
755int
756ulfs_direnter(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
757 struct vnode *tvp,
758 struct componentname *cnp, ino_t inum, unsigned dtype,
759 struct buf *newdirbp)
760{
761 kauth_cred_t cr;
762 int newentrysize;
763 struct inode *dp;
764 struct buf *bp;
765 u_int dsize;
766 LFS_DIRHEADER *ep, *nep;
767 int error, ret, lfs_blkoff, loc, spacefree;
768 char *dirbuf;
769 struct timespec ts;
770 struct ulfsmount *ump = VFSTOULFS(dvp->v_mount);
771 struct lfs *fs = ump->um_lfs;
772 int dirblksiz = fs->um_dirblksiz;
773 const char *name;
774 unsigned namlen, reclen;
775#ifdef LFS_DIRHASH
776 int dohashadd;
777#endif
778
779 error = 0;
780 name = cnp->cn_nameptr; /* note: not null-terminated */
781 namlen = cnp->cn_namelen;
782 cr = cnp->cn_cred;
783
784 dp = VTOI(dvp);
785 newentrysize = LFS_DIRECTSIZ(fs, namlen);
786
787 if (ulr->ulr_count == 0) {
788 /*
789 * If ulr_count is 0, then namei could find no
790 * space in the directory. Here, ulr_offset will
791 * be on a directory block boundary and we will write the
792 * new entry into a fresh block.
793 */
794 if (ulr->ulr_offset & (dirblksiz - 1))
795 panic("ulfs_direnter: newblk");
796 if ((error = lfs_balloc(dvp, (off_t)ulr->ulr_offset, dirblksiz,
797 cr, B_CLRBUF | B_SYNC, &bp)) != 0) {
798 return (error);
799 }
800 dp->i_size = ulr->ulr_offset + dirblksiz;
801 DIP_ASSIGN(dp, size, dp->i_size);
802 dp->i_flag |= IN_CHANGE | IN_UPDATE;
803 uvm_vnp_setsize(dvp, dp->i_size);
804 lfs_blkoff = ulr->ulr_offset & (ump->um_mountp->mnt_stat.f_iosize - 1);
805 ep = (LFS_DIRHEADER *)((char *)bp->b_data + lfs_blkoff);
806 ulfs_direntry_assign(fs, ep, name, namlen, inum, dtype);
807 lfs_dir_setreclen(fs, ep, dirblksiz);
808#ifdef LFS_DIRHASH
809 if (dp->i_dirhash != NULL) {
810 ulfsdirhash_newblk(dp, ulr->ulr_offset);
811 ulfsdirhash_add(dp, ep, ulr->ulr_offset);
812 ulfsdirhash_checkblock(dp, (char *)bp->b_data + lfs_blkoff,
813 ulr->ulr_offset);
814 }
815#endif
816 error = VOP_BWRITE(bp->b_vp, bp);
817 vfs_timestamp(&ts);
818 ret = lfs_update(dvp, &ts, &ts, UPDATE_DIROP);
819 if (error == 0)
820 return (ret);
821 return (error);
822 }
823
824 /*
825 * If ulr_count is non-zero, then namei found space for the new
826 * entry in the range ulr_offset to ulr_offset + ulr_count
827 * in the directory. To use this space, we may have to compact
828 * the entries located there, by copying them together towards the
829 * beginning of the block, leaving the free space in one usable
830 * chunk at the end.
831 */
832
833 /*
834 * Increase size of directory if entry eats into new space.
835 * This should never push the size past a new multiple of
836 * DIRBLKSIZ.
837 *
838 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
839 */
840 if (ulr->ulr_offset + ulr->ulr_count > dp->i_size) {
841#ifdef DIAGNOSTIC
842 printf("ulfs_direnter: reached 4.2-only block, "
843 "not supposed to happen\n");
844#endif
845 dp->i_size = ulr->ulr_offset + ulr->ulr_count;
846 DIP_ASSIGN(dp, size, dp->i_size);
847 dp->i_flag |= IN_CHANGE | IN_UPDATE;
848 }
849 /*
850 * Get the block containing the space for the new directory entry.
851 */
852 error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp, true);
853 if (error) {
854 return (error);
855 }
856 /*
857 * Find space for the new entry. In the simple case, the entry at
858 * offset base will have the space. If it does not, then namei
859 * arranged that compacting the region ulr_offset to
860 * ulr_offset + ulr_count would yield the space.
861 */
862 ep = (LFS_DIRHEADER *)dirbuf;
863 dsize = (lfs_dir_getino(fs, ep) != 0) ? LFS_DIRSIZ(fs, ep) : 0;
864 spacefree = lfs_dir_getreclen(fs, ep) - dsize;
865 for (loc = lfs_dir_getreclen(fs, ep); loc < ulr->ulr_count; ) {
866 nep = (LFS_DIRHEADER *)(dirbuf + loc);
867
868 /* Trim the existing slot (NB: dsize may be zero). */
869 lfs_dir_setreclen(fs, ep, dsize);
870 ep = LFS_NEXTDIR(fs, ep);
871
872 reclen = lfs_dir_getreclen(fs, nep);
873 loc += reclen;
874 if (lfs_dir_getino(fs, nep) == 0) {
875 /*
876 * A mid-block unused entry. Such entries are
877 * never created by the kernel, but fsck_ffs
878 * can create them (and it doesn't fix them).
879 *
880 * Add up the free space, and initialise the
881 * relocated entry since we don't memcpy it.
882 */
883 spacefree += reclen;
884 lfs_dir_setino(fs, ep, 0);
885 dsize = 0;
886 continue;
887 }
888 dsize = LFS_DIRSIZ(fs, nep);
889 spacefree += reclen - dsize;
890#ifdef LFS_DIRHASH
891 if (dp->i_dirhash != NULL)
892 ulfsdirhash_move(dp, nep,
893 ulr->ulr_offset + ((char *)nep - dirbuf),
894 ulr->ulr_offset + ((char *)ep - dirbuf));
895#endif
896 memcpy((void *)ep, (void *)nep, dsize);
897 }
898 /*
899 * Here, `ep' points to a directory entry containing `dsize' in-use
900 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
901 * then the entry is completely unused (dsize == 0). The value
902 * of ep->d_reclen is always indeterminate.
903 *
904 * Update the pointer fields in the previous entry (if any),
905 * copy in the new entry, and write out the block.
906 */
907 if (lfs_dir_getino(fs, ep) == 0 ||
908 (lfs_dir_getino(fs, ep) == ULFS_WINO &&
909 memcmp(lfs_dir_nameptr(fs, ep), name, namlen) == 0)) {
910 if (spacefree + dsize < newentrysize)
911 panic("ulfs_direnter: compact1");
912 reclen = spacefree + dsize;
913#ifdef LFS_DIRHASH
914 dohashadd = (lfs_dir_getino(fs, ep) == 0);
915#endif
916 } else {
917 if (spacefree < newentrysize)
918 panic("ulfs_direnter: compact2");
919 reclen = spacefree;
920 lfs_dir_setreclen(fs, ep, dsize);
921 ep = LFS_NEXTDIR(fs, ep);
922#ifdef LFS_DIRHASH
923 dohashadd = 1;
924#endif
925 }
926
927 ulfs_direntry_assign(fs, ep, name, namlen, inum, dtype);
928 lfs_dir_setreclen(fs, ep, reclen);
929#ifdef LFS_DIRHASH
930 if (dp->i_dirhash != NULL && dohashadd)
931 ulfsdirhash_add(dp, ep, ulr->ulr_offset + ((char *)ep - dirbuf));
932 if (dp->i_dirhash != NULL)
933 ulfsdirhash_checkblock(dp, dirbuf -
934 (ulr->ulr_offset & (dirblksiz - 1)),
935 ulr->ulr_offset & ~(dirblksiz - 1));
936#endif
937 error = VOP_BWRITE(bp->b_vp, bp);
938 dp->i_flag |= IN_CHANGE | IN_UPDATE;
939 /*
940 * If all went well, and the directory can be shortened, proceed
941 * with the truncation. Note that we have to unlock the inode for
942 * the entry that we just entered, as the truncation may need to
943 * lock other inodes which can lead to deadlock if we also hold a
944 * lock on the newly entered node.
945 */
946 if (error == 0 && ulr->ulr_endoff && ulr->ulr_endoff < dp->i_size) {
947#ifdef LFS_DIRHASH
948 if (dp->i_dirhash != NULL)
949 ulfsdirhash_dirtrunc(dp, ulr->ulr_endoff);
950#endif
951 (void) lfs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC, cr);
952 }
953 return (error);
954}
955
956/*
957 * Remove a directory entry after a call to namei, using the
958 * parameters that ulfs_lookup left in nameidata and in the
959 * ulfs_lookup_results.
960 *
961 * DVP is the directory to be updated. It must be locked.
962 * ULR is the ulfs_lookup_results structure from the final lookup step.
963 * IP, if not null, is the inode being unlinked.
964 * FLAGS may contain DOWHITEOUT.
965 * ISRMDIR is not used and (XXX) should be removed.
966 *
967 * If FLAGS contains DOWHITEOUT the entry is replaced with a whiteout
968 * instead of being cleared.
969 *
970 * ulr->ulr_offset contains the position of the directory entry
971 * to be removed.
972 *
973 * ulr->ulr_reclen contains the size of the directory entry to be
974 * removed.
975 *
976 * ulr->ulr_count contains the size of the *previous* directory
977 * entry. This allows finding it, for free space management. If
978 * ulr_count is 0, the target entry is at the beginning of the
979 * directory. (Does this ever happen? The first entry should be ".",
980 * which should only be removed at rmdir time. Does rmdir come here
981 * to clear out the "." and ".." entries? Perhaps, but I doubt it.)
982 *
983 * The space is marked free by adding it to the record length (not
984 * name length) of the preceding entry. If the first entry becomes
985 * free, it is marked free by setting the inode number to 0.
986 *
987 * The link count of IP is decremented. Note that this is not the
988 * inverse behavior of ulfs_direnter, which does not adjust link
989 * counts. Sigh.
990 */
991int
992ulfs_dirremove(struct vnode *dvp, const struct ulfs_lookup_results *ulr,
993 struct inode *ip, int flags, int isrmdir)
994{
995 struct inode *dp = VTOI(dvp);
996 struct lfs *fs = dp->i_lfs;
997 LFS_DIRHEADER *ep;
998 struct buf *bp;
999 int error;
1000
1001 if (flags & DOWHITEOUT) {
1002 /*
1003 * Whiteout entry: set d_ino to ULFS_WINO.
1004 */
1005 error = ulfs_blkatoff(dvp, (off_t)ulr->ulr_offset, (void *)&ep,
1006 &bp, true);
1007 if (error)
1008 return (error);
1009 lfs_dir_setino(fs, ep, ULFS_WINO);
1010 lfs_dir_settype(fs, ep, LFS_DT_WHT);
1011 goto out;
1012 }
1013
1014 if ((error = ulfs_blkatoff(dvp,
1015 (off_t)(ulr->ulr_offset - ulr->ulr_count), (void *)&ep, &bp, true)) != 0)
1016 return (error);
1017
1018#ifdef LFS_DIRHASH
1019 /*
1020 * Remove the dirhash entry. This is complicated by the fact
1021 * that `ep' is the previous entry when ulr_count != 0.
1022 */
1023 if (dp->i_dirhash != NULL)
1024 ulfsdirhash_remove(dp, (ulr->ulr_count == 0) ? ep :
1025 LFS_NEXTDIR(fs, ep), ulr->ulr_offset);
1026#endif
1027
1028 if (ulr->ulr_count == 0) {
1029 /*
1030 * First entry in block: set d_ino to zero.
1031 */
1032 lfs_dir_setino(fs, ep, 0);
1033 } else {
1034 /*
1035 * Collapse new free space into previous entry.
1036 */
1037 lfs_dir_setreclen(fs, ep,
1038 lfs_dir_getreclen(fs, ep) + ulr->ulr_reclen);
1039 }
1040
1041#ifdef LFS_DIRHASH
1042 if (dp->i_dirhash != NULL) {
1043 int dirblksiz = ip->i_lfs->um_dirblksiz;
1044 ulfsdirhash_checkblock(dp, (char *)ep -
1045 ((ulr->ulr_offset - ulr->ulr_count) & (dirblksiz - 1)),
1046 ulr->ulr_offset & ~(dirblksiz - 1));
1047 }
1048#endif
1049
1050out:
1051 if (ip) {
1052 ip->i_nlink--;
1053 DIP_ASSIGN(ip, nlink, ip->i_nlink);
1054 ip->i_flag |= IN_CHANGE;
1055 }
1056 /*
1057 * XXX did it ever occur to anyone that it might be a good
1058 * idea to restore ip->i_nlink if this fails? Or something?
1059 * Currently on error return from this function the state of
1060 * ip->i_nlink depends on what happened, and callers
1061 * definitely do not take this into account.
1062 */
1063 error = VOP_BWRITE(bp->b_vp, bp);
1064 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1065 /*
1066 * If the last named reference to a snapshot goes away,
1067 * drop its snapshot reference so that it will be reclaimed
1068 * when last open reference goes away.
1069 */
1070 if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 &&
1071 ip->i_nlink == 0)
1072 ulfs_snapgone(ip);
1073 return (error);
1074}
1075
1076/*
1077 * Rewrite an existing directory entry to point at the inode supplied.
1078 *
1079 * DP is the directory to update.
1080 * OFFSET is the position of the entry in question. It may come
1081 * from ulr_offset of a ulfs_lookup_results.
1082 * OIP is the old inode the directory previously pointed to.
1083 * NEWINUM is the number of the new inode.
1084 * NEWTYPE is the new value for the type field of the directory entry.
1085 * (This is ignored if the fs doesn't support that.)
1086 * ISRMDIR is not used and (XXX) should be removed.
1087 * IFLAGS are added to DP's inode flags.
1088 *
1089 * The link count of OIP is decremented. Note that the link count of
1090 * the new inode is *not* incremented. Yay for symmetry.
1091 */
1092int
1093ulfs_dirrewrite(struct inode *dp, off_t offset,
1094 struct inode *oip, ino_t newinum, int newtype,
1095 int isrmdir, int iflags)
1096{
1097 struct lfs *fs = dp->i_lfs;
1098 struct buf *bp;
1099 LFS_DIRHEADER *ep;
1100 struct vnode *vdp = ITOV(dp);
1101 int error;
1102
1103 error = ulfs_blkatoff(vdp, offset, (void *)&ep, &bp, true);
1104 if (error)
1105 return (error);
1106 lfs_dir_setino(fs, ep, newinum);
1107 lfs_dir_settype(fs, ep, newtype);
1108 oip->i_nlink--;
1109 DIP_ASSIGN(oip, nlink, oip->i_nlink);
1110 oip->i_flag |= IN_CHANGE;
1111 error = VOP_BWRITE(bp->b_vp, bp);
1112 dp->i_flag |= iflags;
1113 /*
1114 * If the last named reference to a snapshot goes away,
1115 * drop its snapshot reference so that it will be reclaimed
1116 * when last open reference goes away.
1117 */
1118 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_nlink == 0)
1119 ulfs_snapgone(oip);
1120 return (error);
1121}
1122
1123/*
1124 * Check if a directory is empty or not.
1125 * Inode supplied must be locked.
1126 *
1127 * Using a struct lfs_dirtemplate here is not precisely
1128 * what we want, but better than using a struct lfs_direct.
1129 *
1130 * NB: does not handle corrupted directories.
1131 */
1132int
1133ulfs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
1134{
1135 struct lfs *fs = ip->i_lfs;
1136 doff_t off;
1137 union lfs_dirtemplate dbuf;
1138 LFS_DIRHEADER *dp = (LFS_DIRHEADER *)&dbuf;
1139 int error, namlen;
1140 const char *name;
1141 size_t count;
1142/* XXX this should probably use LFS_DIRECTSIZ(fs, 2) */
1143#define MINDIRSIZ (sizeof (struct lfs_dirtemplate64) / 2)
1144
1145 for (off = 0; off < ip->i_size; off += lfs_dir_getreclen(fs, dp)) {
1146 error = ulfs_bufio(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ,
1147 off, IO_NODELOCKED, cred, &count, NULL);
1148 /*
1149 * Since we read MINDIRSIZ, residual must
1150 * be 0 unless we're at end of file.
1151 */
1152 if (error || count != 0)
1153 return (0);
1154 /* avoid infinite loops */
1155 if (lfs_dir_getreclen(fs, dp) == 0)
1156 return (0);
1157 /* skip empty entries */
1158 if (lfs_dir_getino(fs, dp) == 0 ||
1159 lfs_dir_getino(fs, dp) == ULFS_WINO)
1160 continue;
1161 /* accept only "." and ".." */
1162 namlen = lfs_dir_getnamlen(fs, dp);
1163 name = lfs_dir_nameptr(fs, dp);
1164 if (namlen > 2)
1165 return (0);
1166 if (name[0] != '.')
1167 return (0);
1168 /*
1169 * At this point namlen must be 1 or 2.
1170 * 1 implies ".", 2 implies ".." if second
1171 * char is also "."
1172 */
1173 if (namlen == 1 && lfs_dir_getino(fs, dp) == ip->i_number)
1174 continue;
1175 if (name[1] == '.' && lfs_dir_getino(fs, dp) == parentino)
1176 continue;
1177 return (0);
1178 }
1179 return (1);
1180}
1181
1182#define ULFS_DIRRABLKS 0
1183int ulfs_dirrablks = ULFS_DIRRABLKS;
1184
1185/*
1186 * ulfs_blkatoff: Return buffer with the contents of block "offset" from
1187 * the beginning of directory "vp". If "res" is non-NULL, fill it in with
1188 * a pointer to the remaining space in the directory. If the caller intends
1189 * to modify the buffer returned, "modify" must be true.
1190 */
1191
1192int
1193ulfs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp,
1194 bool modify)
1195{
1196 struct inode *ip __diagused;
1197 struct buf *bp;
1198 daddr_t lbn;
1199 const int dirrablks = ulfs_dirrablks;
1200 daddr_t *blks;
1201 int *blksizes;
1202 int run, error;
1203 struct mount *mp = vp->v_mount;
1204 const int bshift = mp->mnt_fs_bshift;
1205 const int bsize = 1 << bshift;
1206 off_t eof;
1207
1208 blks = kmem_alloc((1 + dirrablks) * sizeof(daddr_t), KM_SLEEP);
1209 blksizes = kmem_alloc((1 + dirrablks) * sizeof(int), KM_SLEEP);
1210 ip = VTOI(vp);
1211 KASSERT(vp->v_size == ip->i_size);
1212 GOP_SIZE(vp, vp->v_size, &eof, 0);
1213 lbn = offset >> bshift;
1214
1215 for (run = 0; run <= dirrablks;) {
1216 const off_t curoff = lbn << bshift;
1217 const int size = MIN(eof - curoff, bsize);
1218
1219 if (size == 0) {
1220 break;
1221 }
1222 KASSERT(curoff < eof);
1223 blks[run] = lbn;
1224 blksizes[run] = size;
1225 lbn++;
1226 run++;
1227 if (size != bsize) {
1228 break;
1229 }
1230 }
1231 KASSERT(run >= 1);
1232 error = breadn(vp, blks[0], blksizes[0], &blks[1], &blksizes[1],
1233 run - 1, (modify ? B_MODIFY : 0), &bp);
1234 if (error != 0) {
1235 *bpp = NULL;
1236 goto out;
1237 }
1238 if (res) {
1239 *res = (char *)bp->b_data + (offset & (bsize - 1));
1240 }
1241 *bpp = bp;
1242
1243 out:
1244 kmem_free(blks, (1 + dirrablks) * sizeof(daddr_t));
1245 kmem_free(blksizes, (1 + dirrablks) * sizeof(int));
1246 return error;
1247}
1248