1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, 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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_ffs_broken_fixme.h"
41 #include "opt_ufs.h"
42 #include "opt_quota.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/namei.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/proc.h>
51 #include <sys/stat.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/sysctl.h>
55
56 #include <vm/vm.h>
57 #include <vm/vm_extern.h>
58
59 #include <ufs/ufs/extattr.h>
60 #include <ufs/ufs/quota.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/dir.h>
63 #ifdef UFS_DIRHASH
64 #include <ufs/ufs/dirhash.h>
65 #endif
66 #include <ufs/ufs/ufsmount.h>
67 #include <ufs/ufs/ufs_extern.h>
68
69 #ifdef DIAGNOSTIC
70 static int dirchk = 1;
71 #else
72 static int dirchk = 0;
73 #endif
74
75 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
76
77 /* true if old FS format...*/
78 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
79
80 /*
81 * Convert a component of a pathname into a pointer to a locked inode.
82 * This is a very central and rather complicated routine.
83 * If the filesystem is not maintained in a strict tree hierarchy,
84 * this can result in a deadlock situation (see comments in code below).
85 *
86 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
87 * on whether the name is to be looked up, created, renamed, or deleted.
88 * When CREATE, RENAME, or DELETE is specified, information usable in
89 * creating, renaming, or deleting a directory entry may be calculated.
90 * If flag has LOCKPARENT or'ed into it and the target of the pathname
91 * exists, lookup returns both the target and its parent directory locked.
92 * When creating or renaming and LOCKPARENT is specified, the target may
93 * not be ".". When deleting and LOCKPARENT is specified, the target may
94 * be "."., but the caller must check to ensure it does an vrele and vput
95 * instead of two vputs.
96 *
97 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
98 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
99 * method.
100 *
101 * vfs_cache_lookup() performs the following for us:
102 * check that it is a directory
103 * check accessibility of directory
104 * check for modification attempts on read-only mounts
105 * if name found in cache
106 * if at end of path and deleting or creating
107 * drop it
108 * else
109 * return name.
110 * return VOP_CACHEDLOOKUP()
111 *
112 * Overall outline of ufs_lookup:
113 *
114 * search for name in directory, to found or notfound
115 * notfound:
116 * if creating, return locked directory, leaving info on available slots
117 * else return error
118 * found:
119 * if at end of path and deleting, return information to allow delete
120 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
121 * inode and return info to allow rewrite
122 * if not at end, add name to cache; if at end and neither creating
123 * nor deleting, add name to cache
124 */
125 int
126 ufs_lookup(ap)
127 struct vop_cachedlookup_args /* {
128 struct vnode *a_dvp;
129 struct vnode **a_vpp;
130 struct componentname *a_cnp;
131 } */ *ap;
132 {
133 struct vnode *vdp; /* vnode for directory being searched */
134 struct inode *dp; /* inode for directory being searched */
135 struct buf *bp; /* a buffer of directory entries */
136 struct direct *ep; /* the current directory entry */
137 int entryoffsetinblock; /* offset of ep in bp's buffer */
138 enum {NONE, COMPACT, FOUND} slotstatus;
139 doff_t slotoffset; /* offset of area with free space */
140 doff_t i_diroff; /* cached i_diroff value. */
141 doff_t i_offset; /* cached i_offset value. */
142 int slotsize; /* size of area at slotoffset */
143 int slotfreespace; /* amount of space free in slot */
144 int slotneeded; /* size of the entry we're seeking */
145 int numdirpasses; /* strategy for directory search */
146 doff_t endsearch; /* offset to end directory search */
147 doff_t prevoff; /* prev entry dp->i_offset */
148 struct vnode *pdp; /* saved dp during symlink work */
149 struct vnode *tdp; /* returned by VFS_VGET */
150 doff_t enduseful; /* pointer past last used dir slot */
151 u_long bmask; /* block offset mask */
152 int namlen, error;
153 struct vnode **vpp = ap->a_vpp;
154 struct componentname *cnp = ap->a_cnp;
155 struct ucred *cred = cnp->cn_cred;
156 int flags = cnp->cn_flags;
157 int nameiop = cnp->cn_nameiop;
158 struct thread *td = cnp->cn_thread;
159 ino_t ino;
160 int ltype;
161
162 bp = NULL;
163 slotoffset = -1;
164 /*
165 * XXX there was a soft-update diff about this I couldn't merge.
166 * I think this was the equiv.
167 */
168 *vpp = NULL;
169
170 vdp = ap->a_dvp;
171 dp = VTOI(vdp);
172
173 /*
174 * Create a vm object if vmiodirenable is enabled.
175 * Alternatively we could call vnode_create_vobject
176 * in VFS_VGET but we could end up creating objects
177 * that are never used.
178 */
179 vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread);
180
181 /*
182 * We now have a segment name to search for, and a directory to search.
183 *
184 * Suppress search for slots unless creating
185 * file and at end of pathname, in which case
186 * we watch for a place to put the new file in
187 * case it doesn't already exist.
188 */
189 ino = 0;
190 i_diroff = dp->i_diroff;
191 slotstatus = FOUND;
192 slotfreespace = slotsize = slotneeded = 0;
193 if ((nameiop == CREATE || nameiop == RENAME) &&
194 (flags & ISLASTCN)) {
195 slotstatus = NONE;
196 slotneeded = DIRECTSIZ(cnp->cn_namelen);
197 }
198 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
199
200 #ifdef UFS_DIRHASH
201 /*
202 * Use dirhash for fast operations on large directories. The logic
203 * to determine whether to hash the directory is contained within
204 * ufsdirhash_build(); a zero return means that it decided to hash
205 * this directory and it successfully built up the hash table.
206 */
207 if (ufsdirhash_build(dp) == 0) {
208 /* Look for a free slot if needed. */
209 enduseful = dp->i_size;
210 if (slotstatus != FOUND) {
211 slotoffset = ufsdirhash_findfree(dp, slotneeded,
212 &slotsize);
213 if (slotoffset >= 0) {
214 slotstatus = COMPACT;
215 enduseful = ufsdirhash_enduseful(dp);
216 if (enduseful < 0)
217 enduseful = dp->i_size;
218 }
219 }
220 /* Look up the component. */
221 numdirpasses = 1;
222 entryoffsetinblock = 0; /* silence compiler warning */
223 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
224 &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
225 case 0:
226 ep = (struct direct *)((char *)bp->b_data +
227 (i_offset & bmask));
228 goto foundentry;
229 case ENOENT:
230 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
231 goto notfound;
232 default:
233 /* Something failed; just do a linear search. */
234 break;
235 }
236 }
237 #endif /* UFS_DIRHASH */
238 /*
239 * If there is cached information on a previous search of
240 * this directory, pick up where we last left off.
241 * We cache only lookups as these are the most common
242 * and have the greatest payoff. Caching CREATE has little
243 * benefit as it usually must search the entire directory
244 * to determine that the entry does not exist. Caching the
245 * location of the last DELETE or RENAME has not reduced
246 * profiling time and hence has been removed in the interest
247 * of simplicity.
248 */
249 if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) {
250 entryoffsetinblock = 0;
251 i_offset = 0;
252 numdirpasses = 1;
253 } else {
254 i_offset = i_diroff;
255 if ((entryoffsetinblock = i_offset & bmask) &&
256 (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp)))
257 return (error);
258 numdirpasses = 2;
259 nchstats.ncs_2passes++;
260 }
261 prevoff = i_offset;
262 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
263 enduseful = 0;
264
265 searchloop:
266 while (i_offset < endsearch) {
267 /*
268 * If necessary, get the next directory block.
269 */
270 if ((i_offset & bmask) == 0) {
271 if (bp != NULL)
272 brelse(bp);
273 error =
274 UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp);
275 if (error)
276 return (error);
277 entryoffsetinblock = 0;
278 }
279 /*
280 * If still looking for a slot, and at a DIRBLKSIZE
281 * boundary, have to start looking for free space again.
282 */
283 if (slotstatus == NONE &&
284 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
285 slotoffset = -1;
286 slotfreespace = 0;
287 }
288 /*
289 * Get pointer to next entry.
290 * Full validation checks are slow, so we only check
291 * enough to insure forward progress through the
292 * directory. Complete checks can be run by patching
293 * "dirchk" to be true.
294 */
295 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
296 if (ep->d_reclen == 0 || ep->d_reclen >
297 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
298 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
299 int i;
300
301 ufs_dirbad(dp, i_offset, "mangled entry");
302 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
303 i_offset += i;
304 entryoffsetinblock += i;
305 continue;
306 }
307
308 /*
309 * If an appropriate sized slot has not yet been found,
310 * check to see if one is available. Also accumulate space
311 * in the current block so that we can determine if
312 * compaction is viable.
313 */
314 if (slotstatus != FOUND) {
315 int size = ep->d_reclen;
316
317 if (ep->d_ino != 0)
318 size -= DIRSIZ(OFSFMT(vdp), ep);
319 if (size > 0) {
320 if (size >= slotneeded) {
321 slotstatus = FOUND;
322 slotoffset = i_offset;
323 slotsize = ep->d_reclen;
324 } else if (slotstatus == NONE) {
325 slotfreespace += size;
326 if (slotoffset == -1)
327 slotoffset = i_offset;
328 if (slotfreespace >= slotneeded) {
329 slotstatus = COMPACT;
330 slotsize = i_offset +
331 ep->d_reclen - slotoffset;
332 }
333 }
334 }
335 }
336
337 /*
338 * Check for a name match.
339 */
340 if (ep->d_ino) {
341 # if (BYTE_ORDER == LITTLE_ENDIAN)
342 if (OFSFMT(vdp))
343 namlen = ep->d_type;
344 else
345 namlen = ep->d_namlen;
346 # else
347 namlen = ep->d_namlen;
348 # endif
349 if (namlen == cnp->cn_namelen &&
350 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
351 !bcmp(cnp->cn_nameptr, ep->d_name,
352 (unsigned)namlen)) {
353 #ifdef UFS_DIRHASH
354 foundentry:
355 #endif
356 /*
357 * Save directory entry's inode number and
358 * reclen in ndp->ni_ufs area, and release
359 * directory buffer.
360 */
361 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
362 ep->d_type == DT_WHT) {
363 slotstatus = FOUND;
364 slotoffset = i_offset;
365 slotsize = ep->d_reclen;
366 enduseful = dp->i_size;
367 ap->a_cnp->cn_flags |= ISWHITEOUT;
368 numdirpasses--;
369 goto notfound;
370 }
371 ino = ep->d_ino;
372 goto found;
373 }
374 }
375 prevoff = i_offset;
376 i_offset += ep->d_reclen;
377 entryoffsetinblock += ep->d_reclen;
378 if (ep->d_ino)
379 enduseful = i_offset;
380 }
381 notfound:
382 /*
383 * If we started in the middle of the directory and failed
384 * to find our target, we must check the beginning as well.
385 */
386 if (numdirpasses == 2) {
387 numdirpasses--;
388 i_offset = 0;
389 endsearch = i_diroff;
390 goto searchloop;
391 }
392 if (bp != NULL)
393 brelse(bp);
394 /*
395 * If creating, and at end of pathname and current
396 * directory has not been removed, then can consider
397 * allowing file to be created.
398 */
399 if ((nameiop == CREATE || nameiop == RENAME ||
400 (nameiop == DELETE &&
401 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
402 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
403 (flags & ISLASTCN) && dp->i_effnlink != 0) {
404 /*
405 * Access for write is interpreted as allowing
406 * creation of files in the directory.
407 */
408 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
409 if (error)
410 return (error);
411 /*
412 * Return an indication of where the new directory
413 * entry should be put. If we didn't find a slot,
414 * then set dp->i_count to 0 indicating
415 * that the new slot belongs at the end of the
416 * directory. If we found a slot, then the new entry
417 * can be put in the range from dp->i_offset to
418 * dp->i_offset + dp->i_count.
419 */
420 if (slotstatus == NONE) {
421 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
422 dp->i_count = 0;
423 enduseful = dp->i_offset;
424 } else if (nameiop == DELETE) {
425 dp->i_offset = slotoffset;
426 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
427 dp->i_count = 0;
428 else
429 dp->i_count = dp->i_offset - prevoff;
430 } else {
431 dp->i_offset = slotoffset;
432 dp->i_count = slotsize;
433 if (enduseful < slotoffset + slotsize)
434 enduseful = slotoffset + slotsize;
435 }
436 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
437 /*
438 * We return with the directory locked, so that
439 * the parameters we set up above will still be
440 * valid if we actually decide to do a direnter().
441 * We return ni_vp == NULL to indicate that the entry
442 * does not currently exist; we leave a pointer to
443 * the (locked) directory inode in ndp->ni_dvp.
444 * The pathname buffer is saved so that the name
445 * can be obtained later.
446 *
447 * NB - if the directory is unlocked, then this
448 * information cannot be used.
449 */
450 cnp->cn_flags |= SAVENAME;
451 return (EJUSTRETURN);
452 }
453 /*
454 * Insert name into cache (as non-existent) if appropriate.
455 */
456 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
457 cache_enter(vdp, *vpp, cnp);
458 return (ENOENT);
459
460 found:
461 if (numdirpasses == 2)
462 nchstats.ncs_pass2++;
463 /*
464 * Check that directory length properly reflects presence
465 * of this entry.
466 */
467 if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
468 ufs_dirbad(dp, i_offset, "i_size too small");
469 dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep);
470 DIP_SET(dp, i_size, dp->i_size);
471 dp->i_flag |= IN_CHANGE | IN_UPDATE;
472 }
473 brelse(bp);
474
475 /*
476 * Found component in pathname.
477 * If the final component of path name, save information
478 * in the cache as to where the entry was found.
479 */
480 if ((flags & ISLASTCN) && nameiop == LOOKUP)
481 dp->i_diroff = i_offset &~ (DIRBLKSIZ - 1);
482
483 /*
484 * If deleting, and at end of pathname, return
485 * parameters which can be used to remove file.
486 */
487 if (nameiop == DELETE && (flags & ISLASTCN)) {
488 if (flags & LOCKPARENT)
489 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
490 /*
491 * Write access to directory required to delete files.
492 */
493 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
494 if (error)
495 return (error);
496 /*
497 * Return pointer to current entry in dp->i_offset,
498 * and distance past previous entry (if there
499 * is a previous entry in this block) in dp->i_count.
500 * Save directory inode pointer in ndp->ni_dvp for dirremove().
501 *
502 * Technically we shouldn't be setting these in the
503 * WANTPARENT case (first lookup in rename()), but any
504 * lookups that will result in directory changes will
505 * overwrite these.
506 */
507 dp->i_offset = i_offset;
508 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
509 dp->i_count = 0;
510 else
511 dp->i_count = dp->i_offset - prevoff;
512 if (dp->i_number == ino) {
513 VREF(vdp);
514 *vpp = vdp;
515 return (0);
516 }
517 if ((error = VFS_VGET(vdp->v_mount, ino,
518 LK_EXCLUSIVE, &tdp)) != 0)
519 return (error);
520 /*
521 * If directory is "sticky", then user must own
522 * the directory, or the file in it, else she
523 * may not delete it (unless she's root). This
524 * implements append-only directories.
525 */
526 if ((dp->i_mode & ISVTX) &&
527 VOP_ACCESS(vdp, VADMIN, cred, cnp->cn_thread) &&
528 VOP_ACCESS(tdp, VADMIN, cred, cnp->cn_thread)) {
529 vput(tdp);
530 return (EPERM);
531 }
532 *vpp = tdp;
533 return (0);
534 }
535
536 /*
537 * If rewriting (RENAME), return the inode and the
538 * information required to rewrite the present directory
539 * Must get inode of directory entry to verify it's a
540 * regular file, or empty directory.
541 */
542 if (nameiop == RENAME && (flags & ISLASTCN)) {
543 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)))
544 return (error);
545 /*
546 * Careful about locking second inode.
547 * This can only occur if the target is ".".
548 */
549 dp->i_offset = i_offset;
550 if (dp->i_number == ino)
551 return (EISDIR);
552 if ((error = VFS_VGET(vdp->v_mount, ino,
553 LK_EXCLUSIVE, &tdp)) != 0)
554 return (error);
555 *vpp = tdp;
556 cnp->cn_flags |= SAVENAME;
557 return (0);
558 }
559
560 /*
561 * Step through the translation in the name. We do not `vput' the
562 * directory because we may need it again if a symbolic link
563 * is relative to the current directory. Instead we save it
564 * unlocked as "pdp". We must get the target inode before unlocking
565 * the directory to insure that the inode will not be removed
566 * before we get it. We prevent deadlock by always fetching
567 * inodes from the root, moving down the directory tree. Thus
568 * when following backward pointers ".." we must unlock the
569 * parent directory before getting the requested directory.
570 * There is a potential race condition here if both the current
571 * and parent directories are removed before the VFS_VGET for the
572 * inode associated with ".." returns. We hope that this occurs
573 * infrequently since we cannot avoid this race condition without
574 * implementing a sophisticated deadlock detection algorithm.
575 * Note also that this simple deadlock detection scheme will not
576 * work if the filesystem has any hard links other than ".."
577 * that point backwards in the directory structure.
578 */
579 pdp = vdp;
580 if (flags & ISDOTDOT) {
581 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
582 if (error)
583 return (error);
584 *vpp = tdp;
585 } else if (dp->i_number == ino) {
586 VREF(vdp); /* we want ourself, ie "." */
587 /*
588 * When we lookup "." we still can be asked to lock it
589 * differently.
590 */
591 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
592 if (ltype != VOP_ISLOCKED(vdp, td)) {
593 if (ltype == LK_EXCLUSIVE)
594 vn_lock(vdp, LK_UPGRADE | LK_RETRY, td);
595 else /* if (ltype == LK_SHARED) */
596 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY, td);
597 }
598 *vpp = vdp;
599 } else {
600 error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp);
601 if (error)
602 return (error);
603 *vpp = tdp;
604 }
605
606 /*
607 * Insert name into cache if appropriate.
608 */
609 if (cnp->cn_flags & MAKEENTRY)
610 cache_enter(vdp, *vpp, cnp);
611 return (0);
612 }
613
614 void
615 ufs_dirbad(ip, offset, how)
616 struct inode *ip;
617 doff_t offset;
618 char *how;
619 {
620 struct mount *mp;
621
622 mp = ITOV(ip)->v_mount;
623 if ((mp->mnt_flag & MNT_RDONLY) == 0)
624 panic("ufs_dirbad: %s: bad dir ino %lu at offset %ld: %s",
625 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
626 else
627 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
628 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
629 }
630
631 /*
632 * Do consistency checking on a directory entry:
633 * record length must be multiple of 4
634 * entry must fit in rest of its DIRBLKSIZ block
635 * record must be large enough to contain entry
636 * name is not longer than MAXNAMLEN
637 * name must be as long as advertised, and null terminated
638 */
639 int
640 ufs_dirbadentry(dp, ep, entryoffsetinblock)
641 struct vnode *dp;
642 struct direct *ep;
643 int entryoffsetinblock;
644 {
645 int i, namlen;
646
647 # if (BYTE_ORDER == LITTLE_ENDIAN)
648 if (OFSFMT(dp))
649 namlen = ep->d_type;
650 else
651 namlen = ep->d_namlen;
652 # else
653 namlen = ep->d_namlen;
654 # endif
655 if ((ep->d_reclen & 0x3) != 0 ||
656 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
657 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
658 /*return (1); */
659 printf("First bad\n");
660 goto bad;
661 }
662 if (ep->d_ino == 0)
663 return (0);
664 for (i = 0; i < namlen; i++)
665 if (ep->d_name[i] == '\0') {
666 /*return (1); */
667 printf("Second bad\n");
668 goto bad;
669 }
670 if (ep->d_name[i])
671 goto bad;
672 return (0);
673 bad:
674 return (1);
675 }
676
677 /*
678 * Construct a new directory entry after a call to namei, using the
679 * parameters that it left in the componentname argument cnp. The
680 * argument ip is the inode to which the new directory entry will refer.
681 */
682 void
683 ufs_makedirentry(ip, cnp, newdirp)
684 struct inode *ip;
685 struct componentname *cnp;
686 struct direct *newdirp;
687 {
688
689 #ifdef INVARIANTS
690 if ((cnp->cn_flags & SAVENAME) == 0)
691 panic("ufs_makedirentry: missing name");
692 #endif
693 newdirp->d_ino = ip->i_number;
694 newdirp->d_namlen = cnp->cn_namelen;
695 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
696 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
697 newdirp->d_type = IFTODT(ip->i_mode);
698 else {
699 newdirp->d_type = 0;
700 # if (BYTE_ORDER == LITTLE_ENDIAN)
701 { u_char tmp = newdirp->d_namlen;
702 newdirp->d_namlen = newdirp->d_type;
703 newdirp->d_type = tmp; }
704 # endif
705 }
706 }
707
708 /*
709 * Write a directory entry after a call to namei, using the parameters
710 * that it left in nameidata. The argument dirp is the new directory
711 * entry contents. Dvp is a pointer to the directory to be written,
712 * which was left locked by namei. Remaining parameters (dp->i_offset,
713 * dp->i_count) indicate how the space for the new entry is to be obtained.
714 * Non-null bp indicates that a directory is being created (for the
715 * soft dependency code).
716 */
717 int
718 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
719 struct vnode *dvp;
720 struct vnode *tvp;
721 struct direct *dirp;
722 struct componentname *cnp;
723 struct buf *newdirbp;
724 {
725 struct ucred *cr;
726 struct thread *td;
727 int newentrysize;
728 struct inode *dp;
729 struct buf *bp;
730 u_int dsize;
731 struct direct *ep, *nep;
732 int error, ret, blkoff, loc, spacefree, flags, namlen;
733 char *dirbuf;
734
735 td = curthread; /* XXX */
736 cr = td->td_ucred;
737
738 dp = VTOI(dvp);
739 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
740
741 if (dp->i_count == 0) {
742 /*
743 * If dp->i_count is 0, then namei could find no
744 * space in the directory. Here, dp->i_offset will
745 * be on a directory block boundary and we will write the
746 * new entry into a fresh block.
747 */
748 if (dp->i_offset & (DIRBLKSIZ - 1))
749 panic("ufs_direnter: newblk");
750 flags = BA_CLRBUF;
751 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
752 flags |= IO_SYNC;
753 #ifdef QUOTA
754 if ((error = getinoquota(dp)) != 0) {
755 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
756 bdwrite(newdirbp);
757 return (error);
758 }
759 #endif
760 if ((error = UFS_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
761 cr, flags, &bp)) != 0) {
762 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
763 bdwrite(newdirbp);
764 return (error);
765 }
766 dp->i_size = dp->i_offset + DIRBLKSIZ;
767 DIP_SET(dp, i_size, dp->i_size);
768 dp->i_flag |= IN_CHANGE | IN_UPDATE;
769 vnode_pager_setsize(dvp, (u_long)dp->i_size);
770 dirp->d_reclen = DIRBLKSIZ;
771 blkoff = dp->i_offset &
772 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
773 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
774 #ifdef UFS_DIRHASH
775 if (dp->i_dirhash != NULL) {
776 ufsdirhash_newblk(dp, dp->i_offset);
777 ufsdirhash_add(dp, dirp, dp->i_offset);
778 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
779 dp->i_offset);
780 }
781 #endif
782 if (DOINGSOFTDEP(dvp)) {
783 /*
784 * Ensure that the entire newly allocated block is a
785 * valid directory so that future growth within the
786 * block does not have to ensure that the block is
787 * written before the inode.
788 */
789 blkoff += DIRBLKSIZ;
790 while (blkoff < bp->b_bcount) {
791 ((struct direct *)
792 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
793 blkoff += DIRBLKSIZ;
794 }
795 if (softdep_setup_directory_add(bp, dp, dp->i_offset,
796 dirp->d_ino, newdirbp, 1) == 0) {
797 bdwrite(bp);
798 return (UFS_UPDATE(dvp, 0));
799 }
800 /* We have just allocated a directory block in an
801 * indirect block. Rather than tracking when it gets
802 * claimed by the inode, we simply do a VOP_FSYNC
803 * now to ensure that it is there (in case the user
804 * does a future fsync). Note that we have to unlock
805 * the inode for the entry that we just entered, as
806 * the VOP_FSYNC may need to lock other inodes which
807 * can lead to deadlock if we also hold a lock on
808 * the newly entered node.
809 */
810 if ((error = bwrite(bp)))
811 return (error);
812 if (tvp != NULL)
813 VOP_UNLOCK(tvp, 0, td);
814 error = VOP_FSYNC(dvp, MNT_WAIT, td);
815 if (tvp != NULL)
816 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
817 return (error);
818 }
819 if (DOINGASYNC(dvp)) {
820 bdwrite(bp);
821 return (UFS_UPDATE(dvp, 0));
822 }
823 error = bwrite(bp);
824 ret = UFS_UPDATE(dvp, 1);
825 if (error == 0)
826 return (ret);
827 return (error);
828 }
829
830 /*
831 * If dp->i_count is non-zero, then namei found space for the new
832 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
833 * in the directory. To use this space, we may have to compact
834 * the entries located there, by copying them together towards the
835 * beginning of the block, leaving the free space in one usable
836 * chunk at the end.
837 */
838
839 /*
840 * Increase size of directory if entry eats into new space.
841 * This should never push the size past a new multiple of
842 * DIRBLKSIZE.
843 *
844 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
845 */
846 if (dp->i_offset + dp->i_count > dp->i_size) {
847 dp->i_size = dp->i_offset + dp->i_count;
848 DIP_SET(dp, i_size, dp->i_size);
849 }
850 /*
851 * Get the block containing the space for the new directory entry.
852 */
853 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
854 if (error) {
855 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
856 bdwrite(newdirbp);
857 return (error);
858 }
859 /*
860 * Find space for the new entry. In the simple case, the entry at
861 * offset base will have the space. If it does not, then namei
862 * arranged that compacting the region dp->i_offset to
863 * dp->i_offset + dp->i_count would yield the space.
864 */
865 ep = (struct direct *)dirbuf;
866 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
867 spacefree = ep->d_reclen - dsize;
868 for (loc = ep->d_reclen; loc < dp->i_count; ) {
869 nep = (struct direct *)(dirbuf + loc);
870
871 /* Trim the existing slot (NB: dsize may be zero). */
872 ep->d_reclen = dsize;
873 ep = (struct direct *)((char *)ep + dsize);
874
875 /* Read nep->d_reclen now as the bcopy() may clobber it. */
876 loc += nep->d_reclen;
877 if (nep->d_ino == 0) {
878 /*
879 * A mid-block unused entry. Such entries are
880 * never created by the kernel, but fsck_ffs
881 * can create them (and it doesn't fix them).
882 *
883 * Add up the free space, and initialise the
884 * relocated entry since we don't bcopy it.
885 */
886 spacefree += nep->d_reclen;
887 ep->d_ino = 0;
888 dsize = 0;
889 continue;
890 }
891 dsize = DIRSIZ(OFSFMT(dvp), nep);
892 spacefree += nep->d_reclen - dsize;
893 #ifdef UFS_DIRHASH
894 if (dp->i_dirhash != NULL)
895 ufsdirhash_move(dp, nep,
896 dp->i_offset + ((char *)nep - dirbuf),
897 dp->i_offset + ((char *)ep - dirbuf));
898 #endif
899 if (DOINGSOFTDEP(dvp))
900 softdep_change_directoryentry_offset(dp, dirbuf,
901 (caddr_t)nep, (caddr_t)ep, dsize);
902 else
903 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
904 }
905 /*
906 * Here, `ep' points to a directory entry containing `dsize' in-use
907 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
908 * then the entry is completely unused (dsize == 0). The value
909 * of ep->d_reclen is always indeterminate.
910 *
911 * Update the pointer fields in the previous entry (if any),
912 * copy in the new entry, and write out the block.
913 */
914 # if (BYTE_ORDER == LITTLE_ENDIAN)
915 if (OFSFMT(dvp))
916 namlen = ep->d_type;
917 else
918 namlen = ep->d_namlen;
919 # else
920 namlen = ep->d_namlen;
921 # endif
922 if (ep->d_ino == 0 ||
923 (ep->d_ino == WINO && namlen == dirp->d_namlen &&
924 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
925 if (spacefree + dsize < newentrysize)
926 panic("ufs_direnter: compact1");
927 dirp->d_reclen = spacefree + dsize;
928 } else {
929 if (spacefree < newentrysize)
930 panic("ufs_direnter: compact2");
931 dirp->d_reclen = spacefree;
932 ep->d_reclen = dsize;
933 ep = (struct direct *)((char *)ep + dsize);
934 }
935 #ifdef UFS_DIRHASH
936 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
937 dirp->d_reclen == spacefree))
938 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
939 #endif
940 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
941 #ifdef UFS_DIRHASH
942 if (dp->i_dirhash != NULL)
943 ufsdirhash_checkblock(dp, dirbuf -
944 (dp->i_offset & (DIRBLKSIZ - 1)),
945 dp->i_offset & ~(DIRBLKSIZ - 1));
946 #endif
947
948 if (DOINGSOFTDEP(dvp)) {
949 (void) softdep_setup_directory_add(bp, dp,
950 dp->i_offset + (caddr_t)ep - dirbuf,
951 dirp->d_ino, newdirbp, 0);
952 bdwrite(bp);
953 } else {
954 if (DOINGASYNC(dvp)) {
955 bdwrite(bp);
956 error = 0;
957 } else {
958 error = bwrite(bp);
959 }
960 }
961 dp->i_flag |= IN_CHANGE | IN_UPDATE;
962 /*
963 * If all went well, and the directory can be shortened, proceed
964 * with the truncation. Note that we have to unlock the inode for
965 * the entry that we just entered, as the truncation may need to
966 * lock other inodes which can lead to deadlock if we also hold a
967 * lock on the newly entered node.
968 */
969 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
970 if (tvp != NULL)
971 VOP_UNLOCK(tvp, 0, td);
972 #ifdef UFS_DIRHASH
973 if (dp->i_dirhash != NULL)
974 ufsdirhash_dirtrunc(dp, dp->i_endoff);
975 #endif
976 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff,
977 IO_NORMAL | IO_SYNC, cr, td);
978 if (tvp != NULL)
979 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
980 }
981 return (error);
982 }
983
984 /*
985 * Remove a directory entry after a call to namei, using
986 * the parameters which it left in nameidata. The entry
987 * dp->i_offset contains the offset into the directory of the
988 * entry to be eliminated. The dp->i_count field contains the
989 * size of the previous record in the directory. If this
990 * is 0, the first entry is being deleted, so we need only
991 * zero the inode number to mark the entry as free. If the
992 * entry is not the first in the directory, we must reclaim
993 * the space of the now empty record by adding the record size
994 * to the size of the previous entry.
995 */
996 int
997 ufs_dirremove(dvp, ip, flags, isrmdir)
998 struct vnode *dvp;
999 struct inode *ip;
1000 int flags;
1001 int isrmdir;
1002 {
1003 struct inode *dp;
1004 struct direct *ep, *rep;
1005 struct buf *bp;
1006 int error;
1007
1008 dp = VTOI(dvp);
1009
1010 if (flags & DOWHITEOUT) {
1011 /*
1012 * Whiteout entry: set d_ino to WINO.
1013 */
1014 if ((error =
1015 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
1016 return (error);
1017 ep->d_ino = WINO;
1018 ep->d_type = DT_WHT;
1019 goto out;
1020 }
1021
1022 if ((error = UFS_BLKATOFF(dvp,
1023 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
1024 return (error);
1025
1026 /* Set 'rep' to the entry being removed. */
1027 if (dp->i_count == 0)
1028 rep = ep;
1029 else
1030 rep = (struct direct *)((char *)ep + ep->d_reclen);
1031 #ifdef UFS_DIRHASH
1032 /*
1033 * Remove the dirhash entry. This is complicated by the fact
1034 * that `ep' is the previous entry when dp->i_count != 0.
1035 */
1036 if (dp->i_dirhash != NULL)
1037 ufsdirhash_remove(dp, rep, dp->i_offset);
1038 #endif
1039 if (dp->i_count == 0) {
1040 /*
1041 * First entry in block: set d_ino to zero.
1042 */
1043 ep->d_ino = 0;
1044 } else {
1045 /*
1046 * Collapse new free space into previous entry.
1047 */
1048 ep->d_reclen += rep->d_reclen;
1049 }
1050 #ifdef UFS_DIRHASH
1051 if (dp->i_dirhash != NULL)
1052 ufsdirhash_checkblock(dp, (char *)ep -
1053 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1054 dp->i_offset & ~(DIRBLKSIZ - 1));
1055 #endif
1056 out:
1057 if (DOINGSOFTDEP(dvp)) {
1058 if (ip) {
1059 ip->i_effnlink--;
1060 softdep_change_linkcnt(ip);
1061 softdep_setup_remove(bp, dp, ip, isrmdir);
1062 }
1063 if (softdep_slowdown(dvp)) {
1064 error = bwrite(bp);
1065 } else {
1066 bdwrite(bp);
1067 error = 0;
1068 }
1069 } else {
1070 if (ip) {
1071 ip->i_effnlink--;
1072 ip->i_nlink--;
1073 DIP_SET(ip, i_nlink, ip->i_nlink);
1074 ip->i_flag |= IN_CHANGE;
1075 }
1076 if (flags & DOWHITEOUT)
1077 error = bwrite(bp);
1078 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1079 bdwrite(bp);
1080 error = 0;
1081 } else
1082 error = bwrite(bp);
1083 }
1084 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1085 /*
1086 * If the last named reference to a snapshot goes away,
1087 * drop its snapshot reference so that it will be reclaimed
1088 * when last open reference goes away.
1089 */
1090 #if defined(FFS) || defined(IFS)
1091 if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 && ip->i_effnlink == 0)
1092 ffs_snapgone(ip);
1093 #endif
1094 return (error);
1095 }
1096
1097 /*
1098 * Rewrite an existing directory entry to point at the inode
1099 * supplied. The parameters describing the directory entry are
1100 * set up by a call to namei.
1101 */
1102 int
1103 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1104 struct inode *dp, *oip;
1105 ino_t newinum;
1106 int newtype;
1107 int isrmdir;
1108 {
1109 struct buf *bp;
1110 struct direct *ep;
1111 struct vnode *vdp = ITOV(dp);
1112 int error;
1113
1114 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1115 if (error)
1116 return (error);
1117 ep->d_ino = newinum;
1118 if (!OFSFMT(vdp))
1119 ep->d_type = newtype;
1120 oip->i_effnlink--;
1121 if (DOINGSOFTDEP(vdp)) {
1122 softdep_change_linkcnt(oip);
1123 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1124 bdwrite(bp);
1125 } else {
1126 oip->i_nlink--;
1127 DIP_SET(oip, i_nlink, oip->i_nlink);
1128 oip->i_flag |= IN_CHANGE;
1129 if (DOINGASYNC(vdp)) {
1130 bdwrite(bp);
1131 error = 0;
1132 } else {
1133 error = bwrite(bp);
1134 }
1135 }
1136 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1137 /*
1138 * If the last named reference to a snapshot goes away,
1139 * drop its snapshot reference so that it will be reclaimed
1140 * when last open reference goes away.
1141 */
1142 #if defined(FFS) || defined(IFS)
1143 if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0)
1144 ffs_snapgone(oip);
1145 #endif
1146 return (error);
1147 }
1148
1149 /*
1150 * Check if a directory is empty or not.
1151 * Inode supplied must be locked.
1152 *
1153 * Using a struct dirtemplate here is not precisely
1154 * what we want, but better than using a struct direct.
1155 *
1156 * NB: does not handle corrupted directories.
1157 */
1158 int
1159 ufs_dirempty(ip, parentino, cred)
1160 struct inode *ip;
1161 ino_t parentino;
1162 struct ucred *cred;
1163 {
1164 doff_t off;
1165 struct dirtemplate dbuf;
1166 struct direct *dp = (struct direct *)&dbuf;
1167 int error, count, namlen;
1168 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1169
1170 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1171 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1172 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1173 NOCRED, &count, (struct thread *)0);
1174 /*
1175 * Since we read MINDIRSIZ, residual must
1176 * be 0 unless we're at end of file.
1177 */
1178 if (error || count != 0)
1179 return (0);
1180 /* avoid infinite loops */
1181 if (dp->d_reclen == 0)
1182 return (0);
1183 /* skip empty entries */
1184 if (dp->d_ino == 0 || dp->d_ino == WINO)
1185 continue;
1186 /* accept only "." and ".." */
1187 # if (BYTE_ORDER == LITTLE_ENDIAN)
1188 if (OFSFMT(ITOV(ip)))
1189 namlen = dp->d_type;
1190 else
1191 namlen = dp->d_namlen;
1192 # else
1193 namlen = dp->d_namlen;
1194 # endif
1195 if (namlen > 2)
1196 return (0);
1197 if (dp->d_name[0] != '.')
1198 return (0);
1199 /*
1200 * At this point namlen must be 1 or 2.
1201 * 1 implies ".", 2 implies ".." if second
1202 * char is also "."
1203 */
1204 if (namlen == 1 && dp->d_ino == ip->i_number)
1205 continue;
1206 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1207 continue;
1208 return (0);
1209 }
1210 return (1);
1211 }
1212
1213 /*
1214 * Check if source directory is in the path of the target directory.
1215 * Target is supplied locked, source is unlocked.
1216 * The target is always vput before returning.
1217 */
1218 int
1219 ufs_checkpath(source, target, cred)
1220 struct inode *source, *target;
1221 struct ucred *cred;
1222 {
1223 struct vnode *vp;
1224 int error, namlen;
1225 ino_t rootino;
1226 struct dirtemplate dirbuf;
1227
1228 vp = ITOV(target);
1229 if (target->i_number == source->i_number) {
1230 error = EEXIST;
1231 goto out;
1232 }
1233 rootino = ROOTINO;
1234 error = 0;
1235 if (target->i_number == rootino)
1236 goto out;
1237
1238 for (;;) {
1239 if (vp->v_type != VDIR) {
1240 error = ENOTDIR;
1241 break;
1242 }
1243 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1244 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1245 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, (int *)0,
1246 (struct thread *)0);
1247 if (error != 0)
1248 break;
1249 # if (BYTE_ORDER == LITTLE_ENDIAN)
1250 if (OFSFMT(vp))
1251 namlen = dirbuf.dotdot_type;
1252 else
1253 namlen = dirbuf.dotdot_namlen;
1254 # else
1255 namlen = dirbuf.dotdot_namlen;
1256 # endif
1257 if (namlen != 2 ||
1258 dirbuf.dotdot_name[0] != '.' ||
1259 dirbuf.dotdot_name[1] != '.') {
1260 error = ENOTDIR;
1261 break;
1262 }
1263 if (dirbuf.dotdot_ino == source->i_number) {
1264 error = EINVAL;
1265 break;
1266 }
1267 if (dirbuf.dotdot_ino == rootino)
1268 break;
1269 vput(vp);
1270 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1271 LK_EXCLUSIVE, &vp);
1272 if (error) {
1273 vp = NULL;
1274 break;
1275 }
1276 }
1277
1278 out:
1279 if (error == ENOTDIR)
1280 printf("checkpath: .. not a directory\n");
1281 if (vp != NULL)
1282 vput(vp);
1283 return (error);
1284 }
Cache object: 23cc37996c8e26e6089bea11569f6e19
|