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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95
39 * $FreeBSD$
40 */
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/namei.h>
46 #include <sys/buf.h>
47 #include <sys/proc.h>
48 #include <sys/stat.h>
49 #include <sys/mount.h>
50 #include <sys/vnode.h>
51
52 #include <vm/vm.h>
53 #include <vm/vm_extern.h>
54
55 #include <ufs/ufs/quota.h>
56 #include <ufs/ufs/inode.h>
57 #include <ufs/ufs/dir.h>
58 #include <ufs/ufs/ufsmount.h>
59 #include <ufs/ufs/ufs_extern.h>
60
61 #ifdef DIAGNOSTIC
62 int dirchk = 1;
63 #else
64 int dirchk = 0;
65 #endif
66
67 /* true if old FS format...*/
68 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
69
70 /*
71 * Convert a component of a pathname into a pointer to a locked inode.
72 * This is a very central and rather complicated routine.
73 * If the file system is not maintained in a strict tree hierarchy,
74 * this can result in a deadlock situation (see comments in code below).
75 *
76 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
77 * on whether the name is to be looked up, created, renamed, or deleted.
78 * When CREATE, RENAME, or DELETE is specified, information usable in
79 * creating, renaming, or deleting a directory entry may be calculated.
80 * If flag has LOCKPARENT or'ed into it and the target of the pathname
81 * exists, lookup returns both the target and its parent directory locked.
82 * When creating or renaming and LOCKPARENT is specified, the target may
83 * not be ".". When deleting and LOCKPARENT is specified, the target may
84 * be "."., but the caller must check to ensure it does an vrele and vput
85 * instead of two vputs.
86 *
87 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
88 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
89 * method.
90 *
91 * vfs_cache_lookup() performs the following for us:
92 * check that it is a directory
93 * check accessibility of directory
94 * check for modification attempts on read-only mounts
95 * if name found in cache
96 * if at end of path and deleting or creating
97 * drop it
98 * else
99 * return name.
100 * return VOP_CACHEDLOOKUP()
101 *
102 * Overall outline of ufs_lookup:
103 *
104 * search for name in directory, to found or notfound
105 * notfound:
106 * if creating, return locked directory, leaving info on available slots
107 * else return error
108 * found:
109 * if at end of path and deleting, return information to allow delete
110 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
111 * inode and return info to allow rewrite
112 * if not at end, add name to cache; if at end and neither creating
113 * nor deleting, add name to cache
114 */
115 int
116 ufs_lookup(ap)
117 struct vop_cachedlookup_args /* {
118 struct vnode *a_dvp;
119 struct vnode **a_vpp;
120 struct componentname *a_cnp;
121 } */ *ap;
122 {
123 register struct vnode *vdp; /* vnode for directory being searched */
124 register struct inode *dp; /* inode for directory being searched */
125 struct buf *bp; /* a buffer of directory entries */
126 register struct direct *ep; /* the current directory entry */
127 int entryoffsetinblock; /* offset of ep in bp's buffer */
128 enum {NONE, COMPACT, FOUND} slotstatus;
129 doff_t slotoffset; /* offset of area with free space */
130 int slotsize; /* size of area at slotoffset */
131 int slotfreespace; /* amount of space free in slot */
132 int slotneeded; /* size of the entry we're seeking */
133 int numdirpasses; /* strategy for directory search */
134 doff_t endsearch; /* offset to end directory search */
135 doff_t prevoff; /* prev entry dp->i_offset */
136 struct vnode *pdp; /* saved dp during symlink work */
137 struct vnode *tdp; /* returned by VFS_VGET */
138 doff_t enduseful; /* pointer past last used dir slot */
139 u_long bmask; /* block offset mask */
140 int lockparent; /* 1 => lockparent flag is set */
141 int wantparent; /* 1 => wantparent or lockparent flag */
142 int namlen, error;
143 struct vnode **vpp = ap->a_vpp;
144 struct componentname *cnp = ap->a_cnp;
145 struct ucred *cred = cnp->cn_cred;
146 int flags = cnp->cn_flags;
147 int nameiop = cnp->cn_nameiop;
148 struct proc *p = cnp->cn_proc;
149
150 bp = NULL;
151 slotoffset = -1;
152 /*
153 * XXX there was a soft-update diff about this I couldn't merge.
154 * I think this was the equiv.
155 */
156 *vpp = NULL;
157
158 vdp = ap->a_dvp;
159 dp = VTOI(vdp);
160 lockparent = flags & LOCKPARENT;
161 wantparent = flags & (LOCKPARENT|WANTPARENT);
162
163 /*
164 * We now have a segment name to search for, and a directory to search.
165 *
166 * Suppress search for slots unless creating
167 * file and at end of pathname, in which case
168 * we watch for a place to put the new file in
169 * case it doesn't already exist.
170 */
171 slotstatus = FOUND;
172 slotfreespace = slotsize = slotneeded = 0;
173 if ((nameiop == CREATE || nameiop == RENAME) &&
174 (flags & ISLASTCN)) {
175 slotstatus = NONE;
176 slotneeded = DIRECTSIZ(cnp->cn_namelen);
177 }
178
179 /*
180 * If there is cached information on a previous search of
181 * this directory, pick up where we last left off.
182 * We cache only lookups as these are the most common
183 * and have the greatest payoff. Caching CREATE has little
184 * benefit as it usually must search the entire directory
185 * to determine that the entry does not exist. Caching the
186 * location of the last DELETE or RENAME has not reduced
187 * profiling time and hence has been removed in the interest
188 * of simplicity.
189 */
190 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
191 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
192 dp->i_diroff > dp->i_size) {
193 entryoffsetinblock = 0;
194 dp->i_offset = 0;
195 numdirpasses = 1;
196 } else {
197 dp->i_offset = dp->i_diroff;
198 if ((entryoffsetinblock = dp->i_offset & bmask) &&
199 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
200 return (error);
201 numdirpasses = 2;
202 nchstats.ncs_2passes++;
203 }
204 prevoff = dp->i_offset;
205 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
206 enduseful = 0;
207
208 searchloop:
209 while (dp->i_offset < endsearch) {
210 /*
211 * If necessary, get the next directory block.
212 */
213 if ((dp->i_offset & bmask) == 0) {
214 if (bp != NULL)
215 brelse(bp);
216 error =
217 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
218 if (error)
219 return (error);
220 entryoffsetinblock = 0;
221 }
222 /*
223 * If still looking for a slot, and at a DIRBLKSIZE
224 * boundary, have to start looking for free space again.
225 */
226 if (slotstatus == NONE &&
227 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
228 slotoffset = -1;
229 slotfreespace = 0;
230 }
231 /*
232 * Get pointer to next entry.
233 * Full validation checks are slow, so we only check
234 * enough to insure forward progress through the
235 * directory. Complete checks can be run by patching
236 * "dirchk" to be true.
237 */
238 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
239 if (ep->d_reclen == 0 ||
240 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
241 int i;
242
243 ufs_dirbad(dp, dp->i_offset, "mangled entry");
244 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
245 dp->i_offset += i;
246 entryoffsetinblock += i;
247 continue;
248 }
249
250 /*
251 * If an appropriate sized slot has not yet been found,
252 * check to see if one is available. Also accumulate space
253 * in the current block so that we can determine if
254 * compaction is viable.
255 */
256 if (slotstatus != FOUND) {
257 int size = ep->d_reclen;
258
259 if (ep->d_ino != 0)
260 size -= DIRSIZ(OFSFMT(vdp), ep);
261 if (size > 0) {
262 if (size >= slotneeded) {
263 slotstatus = FOUND;
264 slotoffset = dp->i_offset;
265 slotsize = ep->d_reclen;
266 } else if (slotstatus == NONE) {
267 slotfreespace += size;
268 if (slotoffset == -1)
269 slotoffset = dp->i_offset;
270 if (slotfreespace >= slotneeded) {
271 slotstatus = COMPACT;
272 slotsize = dp->i_offset +
273 ep->d_reclen - slotoffset;
274 }
275 }
276 }
277 }
278
279 /*
280 * Check for a name match.
281 */
282 if (ep->d_ino) {
283 # if (BYTE_ORDER == LITTLE_ENDIAN)
284 if (OFSFMT(vdp))
285 namlen = ep->d_type;
286 else
287 namlen = ep->d_namlen;
288 # else
289 namlen = ep->d_namlen;
290 # endif
291 if (namlen == cnp->cn_namelen &&
292 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
293 !bcmp(cnp->cn_nameptr, ep->d_name,
294 (unsigned)namlen)) {
295 /*
296 * Save directory entry's inode number and
297 * reclen in ndp->ni_ufs area, and release
298 * directory buffer.
299 */
300 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
301 ep->d_type == DT_WHT) {
302 slotstatus = FOUND;
303 slotoffset = dp->i_offset;
304 slotsize = ep->d_reclen;
305 dp->i_reclen = slotsize;
306 enduseful = dp->i_size;
307 ap->a_cnp->cn_flags |= ISWHITEOUT;
308 numdirpasses--;
309 goto notfound;
310 }
311 dp->i_ino = ep->d_ino;
312 dp->i_reclen = ep->d_reclen;
313 goto found;
314 }
315 }
316 prevoff = dp->i_offset;
317 dp->i_offset += ep->d_reclen;
318 entryoffsetinblock += ep->d_reclen;
319 if (ep->d_ino)
320 enduseful = dp->i_offset;
321 }
322 notfound:
323 /*
324 * If we started in the middle of the directory and failed
325 * to find our target, we must check the beginning as well.
326 */
327 if (numdirpasses == 2) {
328 numdirpasses--;
329 dp->i_offset = 0;
330 endsearch = dp->i_diroff;
331 goto searchloop;
332 }
333 if (bp != NULL)
334 brelse(bp);
335 /*
336 * If creating, and at end of pathname and current
337 * directory has not been removed, then can consider
338 * allowing file to be created.
339 */
340 if ((nameiop == CREATE || nameiop == RENAME ||
341 (nameiop == DELETE &&
342 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
343 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
344 (flags & ISLASTCN) && dp->i_effnlink != 0) {
345 /*
346 * Access for write is interpreted as allowing
347 * creation of files in the directory.
348 */
349 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
350 if (error)
351 return (error);
352 /*
353 * Return an indication of where the new directory
354 * entry should be put. If we didn't find a slot,
355 * then set dp->i_count to 0 indicating
356 * that the new slot belongs at the end of the
357 * directory. If we found a slot, then the new entry
358 * can be put in the range from dp->i_offset to
359 * dp->i_offset + dp->i_count.
360 */
361 if (slotstatus == NONE) {
362 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
363 dp->i_count = 0;
364 enduseful = dp->i_offset;
365 } else if (nameiop == DELETE) {
366 dp->i_offset = slotoffset;
367 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
368 dp->i_count = 0;
369 else
370 dp->i_count = dp->i_offset - prevoff;
371 } else {
372 dp->i_offset = slotoffset;
373 dp->i_count = slotsize;
374 if (enduseful < slotoffset + slotsize)
375 enduseful = slotoffset + slotsize;
376 }
377 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
378 dp->i_flag |= IN_CHANGE | IN_UPDATE;
379 /*
380 * We return with the directory locked, so that
381 * the parameters we set up above will still be
382 * valid if we actually decide to do a direnter().
383 * We return ni_vp == NULL to indicate that the entry
384 * does not currently exist; we leave a pointer to
385 * the (locked) directory inode in ndp->ni_dvp.
386 * The pathname buffer is saved so that the name
387 * can be obtained later.
388 *
389 * NB - if the directory is unlocked, then this
390 * information cannot be used.
391 */
392 cnp->cn_flags |= SAVENAME;
393 if (!lockparent)
394 VOP_UNLOCK(vdp, 0, p);
395 return (EJUSTRETURN);
396 }
397 /*
398 * Insert name into cache (as non-existent) if appropriate.
399 */
400 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
401 cache_enter(vdp, *vpp, cnp);
402 return (ENOENT);
403
404 found:
405 if (numdirpasses == 2)
406 nchstats.ncs_pass2++;
407 /*
408 * Check that directory length properly reflects presence
409 * of this entry.
410 */
411 if (entryoffsetinblock + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
412 ufs_dirbad(dp, dp->i_offset, "i_size too small");
413 dp->i_size = entryoffsetinblock + DIRSIZ(OFSFMT(vdp), ep);
414 dp->i_flag |= IN_CHANGE | IN_UPDATE;
415 }
416 brelse(bp);
417
418 /*
419 * Found component in pathname.
420 * If the final component of path name, save information
421 * in the cache as to where the entry was found.
422 */
423 if ((flags & ISLASTCN) && nameiop == LOOKUP)
424 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
425
426 /*
427 * If deleting, and at end of pathname, return
428 * parameters which can be used to remove file.
429 * If the wantparent flag isn't set, we return only
430 * the directory (in ndp->ni_dvp), otherwise we go
431 * on and lock the inode, being careful with ".".
432 */
433 if (nameiop == DELETE && (flags & ISLASTCN)) {
434 /*
435 * Write access to directory required to delete files.
436 */
437 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
438 if (error)
439 return (error);
440 /*
441 * Return pointer to current entry in dp->i_offset,
442 * and distance past previous entry (if there
443 * is a previous entry in this block) in dp->i_count.
444 * Save directory inode pointer in ndp->ni_dvp for dirremove().
445 */
446 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
447 dp->i_count = 0;
448 else
449 dp->i_count = dp->i_offset - prevoff;
450 if (dp->i_number == dp->i_ino) {
451 VREF(vdp);
452 *vpp = vdp;
453 return (0);
454 }
455 if (flags & ISDOTDOT)
456 VOP_UNLOCK(vdp, 0, p); /* race to get the inode */
457 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
458 if (flags & ISDOTDOT)
459 vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, p);
460 if (error)
461 return (error);
462 /*
463 * If directory is "sticky", then user must own
464 * the directory, or the file in it, else she
465 * may not delete it (unless she's root). This
466 * implements append-only directories.
467 */
468 if ((dp->i_mode & ISVTX) &&
469 cred->cr_uid != 0 &&
470 cred->cr_uid != dp->i_uid &&
471 VTOI(tdp)->i_uid != cred->cr_uid) {
472 vput(tdp);
473 return (EPERM);
474 }
475 *vpp = tdp;
476 if (!lockparent)
477 VOP_UNLOCK(vdp, 0, p);
478 return (0);
479 }
480
481 /*
482 * If rewriting (RENAME), return the inode and the
483 * information required to rewrite the present directory
484 * Must get inode of directory entry to verify it's a
485 * regular file, or empty directory.
486 */
487 if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
488 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
489 return (error);
490 /*
491 * Careful about locking second inode.
492 * This can only occur if the target is ".".
493 */
494 if (dp->i_number == dp->i_ino)
495 return (EISDIR);
496 if (flags & ISDOTDOT)
497 VOP_UNLOCK(vdp, 0, p); /* race to get the inode */
498 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
499 if (flags & ISDOTDOT)
500 vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, p);
501 if (error)
502 return (error);
503 *vpp = tdp;
504 cnp->cn_flags |= SAVENAME;
505 if (!lockparent)
506 VOP_UNLOCK(vdp, 0, p);
507 return (0);
508 }
509
510 /*
511 * Step through the translation in the name. We do not `vput' the
512 * directory because we may need it again if a symbolic link
513 * is relative to the current directory. Instead we save it
514 * unlocked as "pdp". We must get the target inode before unlocking
515 * the directory to insure that the inode will not be removed
516 * before we get it. We prevent deadlock by always fetching
517 * inodes from the root, moving down the directory tree. Thus
518 * when following backward pointers ".." we must unlock the
519 * parent directory before getting the requested directory.
520 * There is a potential race condition here if both the current
521 * and parent directories are removed before the VFS_VGET for the
522 * inode associated with ".." returns. We hope that this occurs
523 * infrequently since we cannot avoid this race condition without
524 * implementing a sophisticated deadlock detection algorithm.
525 * Note also that this simple deadlock detection scheme will not
526 * work if the file system has any hard links other than ".."
527 * that point backwards in the directory structure.
528 */
529 pdp = vdp;
530 if (flags & ISDOTDOT) {
531 VOP_UNLOCK(pdp, 0, p); /* race to get the inode */
532 if (error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) {
533 vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p);
534 return (error);
535 }
536 if (lockparent && (flags & ISLASTCN) &&
537 (error = vn_lock(pdp, LK_EXCLUSIVE, p))) {
538 vput(tdp);
539 return (error);
540 }
541 *vpp = tdp;
542 } else if (dp->i_number == dp->i_ino) {
543 VREF(vdp); /* we want ourself, ie "." */
544 *vpp = vdp;
545 } else {
546 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
547 if (error)
548 return (error);
549 if (!lockparent || !(flags & ISLASTCN))
550 VOP_UNLOCK(pdp, 0, p);
551 *vpp = tdp;
552 }
553
554 /*
555 * Insert name into cache if appropriate.
556 */
557 if (cnp->cn_flags & MAKEENTRY)
558 cache_enter(vdp, *vpp, cnp);
559 return (0);
560 }
561
562 void
563 ufs_dirbad(ip, offset, how)
564 struct inode *ip;
565 doff_t offset;
566 char *how;
567 {
568 struct mount *mp;
569
570 mp = ITOV(ip)->v_mount;
571 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
572 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
573 if ((mp->mnt_stat.f_flags & MNT_RDONLY) == 0)
574 panic("ufs_dirbad: bad dir");
575 }
576
577 /*
578 * Do consistency checking on a directory entry:
579 * record length must be multiple of 4
580 * entry must fit in rest of its DIRBLKSIZ block
581 * record must be large enough to contain entry
582 * name is not longer than MAXNAMLEN
583 * name must be as long as advertised, and null terminated
584 */
585 int
586 ufs_dirbadentry(dp, ep, entryoffsetinblock)
587 struct vnode *dp;
588 register struct direct *ep;
589 int entryoffsetinblock;
590 {
591 register int i;
592 int namlen;
593
594 # if (BYTE_ORDER == LITTLE_ENDIAN)
595 if (OFSFMT(dp))
596 namlen = ep->d_type;
597 else
598 namlen = ep->d_namlen;
599 # else
600 namlen = ep->d_namlen;
601 # endif
602 if ((ep->d_reclen & 0x3) != 0 ||
603 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
604 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
605 /*return (1); */
606 printf("First bad\n");
607 goto bad;
608 }
609 if (ep->d_ino == 0)
610 return (0);
611 for (i = 0; i < namlen; i++)
612 if (ep->d_name[i] == '\0') {
613 /*return (1); */
614 printf("Second bad\n");
615 goto bad;
616 }
617 if (ep->d_name[i])
618 goto bad;
619 return (0);
620 bad:
621 return (1);
622 }
623
624 /*
625 * Construct a new directory entry after a call to namei, using the
626 * parameters that it left in the componentname argument cnp. The
627 * argument ip is the inode to which the new directory entry will refer.
628 */
629 void
630 ufs_makedirentry(ip, cnp, newdirp)
631 struct inode *ip;
632 struct componentname *cnp;
633 struct direct *newdirp;
634 {
635
636 #ifdef DIAGNOSTIC
637 if ((cnp->cn_flags & SAVENAME) == 0)
638 panic("ufs_makedirentry: missing name");
639 #endif
640 newdirp->d_ino = ip->i_number;
641 newdirp->d_namlen = cnp->cn_namelen;
642 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
643 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
644 newdirp->d_type = IFTODT(ip->i_mode);
645 else {
646 newdirp->d_type = 0;
647 # if (BYTE_ORDER == LITTLE_ENDIAN)
648 { u_char tmp = newdirp->d_namlen;
649 newdirp->d_namlen = newdirp->d_type;
650 newdirp->d_type = tmp; }
651 # endif
652 }
653 }
654
655 /*
656 * Write a directory entry after a call to namei, using the parameters
657 * that it left in nameidata. The argument dirp is the new directory
658 * entry contents. Dvp is a pointer to the directory to be written,
659 * which was left locked by namei. Remaining parameters (dp->i_offset,
660 * dp->i_count) indicate how the space for the new entry is to be obtained.
661 * Non-null bp indicates that a directory is being created (for the
662 * soft dependency code).
663 */
664 int
665 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
666 struct vnode *dvp;
667 struct vnode *tvp;
668 struct direct *dirp;
669 struct componentname *cnp;
670 struct buf *newdirbp;
671 {
672 struct ucred *cr;
673 struct proc *p;
674 int newentrysize;
675 struct inode *dp;
676 struct buf *bp;
677 u_int dsize;
678 struct direct *ep, *nep;
679 int error, ret, blkoff, loc, spacefree, flags;
680 char *dirbuf;
681
682 p = curproc; /* XXX */
683 cr = p->p_ucred;
684
685 dp = VTOI(dvp);
686 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
687
688 if (dp->i_count == 0) {
689 /*
690 * If dp->i_count is 0, then namei could find no
691 * space in the directory. Here, dp->i_offset will
692 * be on a directory block boundary and we will write the
693 * new entry into a fresh block.
694 */
695 if (dp->i_offset & (DIRBLKSIZ - 1))
696 panic("ufs_direnter: newblk");
697 flags = B_CLRBUF;
698 if (!DOINGSOFTDEP(dvp))
699 flags |= B_SYNC;
700 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
701 cr, flags, &bp)) != 0) {
702 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
703 bdwrite(newdirbp);
704 return (error);
705 }
706 dp->i_size = dp->i_offset + DIRBLKSIZ;
707 dp->i_flag |= IN_CHANGE | IN_UPDATE;
708 vnode_pager_setsize(dvp, (u_long)dp->i_size);
709 dirp->d_reclen = DIRBLKSIZ;
710 blkoff = dp->i_offset &
711 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
712 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
713 if (DOINGSOFTDEP(dvp)) {
714 /*
715 * Ensure that the entire newly allocated block is a
716 * valid directory so that future growth within the
717 * block does not have to ensure that the block is
718 * written before the inode.
719 */
720 blkoff += DIRBLKSIZ;
721 while (blkoff < bp->b_bcount) {
722 ((struct direct *)
723 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
724 blkoff += DIRBLKSIZ;
725 }
726 softdep_setup_directory_add(bp, dp, dp->i_offset,
727 dirp->d_ino, newdirbp);
728 bdwrite(bp);
729 } else {
730 error = VOP_BWRITE(bp);
731 }
732 ret = UFS_UPDATE(dvp, !DOINGSOFTDEP(dvp));
733 if (error == 0)
734 return (ret);
735 return (error);
736 }
737
738 /*
739 * If dp->i_count is non-zero, then namei found space for the new
740 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
741 * in the directory. To use this space, we may have to compact
742 * the entries located there, by copying them together towards the
743 * beginning of the block, leaving the free space in one usable
744 * chunk at the end.
745 */
746
747 /*
748 * Increase size of directory if entry eats into new space.
749 * This should never push the size past a new multiple of
750 * DIRBLKSIZE.
751 *
752 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
753 */
754 if (dp->i_offset + dp->i_count > dp->i_size)
755 dp->i_size = dp->i_offset + dp->i_count;
756 /*
757 * Get the block containing the space for the new directory entry.
758 */
759 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
760 if (error) {
761 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
762 bdwrite(newdirbp);
763 return (error);
764 }
765 /*
766 * Find space for the new entry. In the simple case, the entry at
767 * offset base will have the space. If it does not, then namei
768 * arranged that compacting the region dp->i_offset to
769 * dp->i_offset + dp->i_count would yield the space.
770 */
771 ep = (struct direct *)dirbuf;
772 dsize = DIRSIZ(OFSFMT(dvp), ep);
773 spacefree = ep->d_reclen - dsize;
774 for (loc = ep->d_reclen; loc < dp->i_count; ) {
775 nep = (struct direct *)(dirbuf + loc);
776 if (ep->d_ino) {
777 /* trim the existing slot */
778 ep->d_reclen = dsize;
779 ep = (struct direct *)((char *)ep + dsize);
780 } else {
781 /* overwrite; nothing there; header is ours */
782 spacefree += dsize;
783 }
784 dsize = DIRSIZ(OFSFMT(dvp), nep);
785 spacefree += nep->d_reclen - dsize;
786 loc += nep->d_reclen;
787 if (DOINGSOFTDEP(dvp))
788 softdep_change_directoryentry_offset(dp, dirbuf,
789 (caddr_t)nep, (caddr_t)ep, dsize);
790 else
791 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
792 }
793 /*
794 * Update the pointer fields in the previous entry (if any),
795 * copy in the new entry, and write out the block.
796 */
797 if (ep->d_ino == 0 ||
798 (ep->d_ino == WINO &&
799 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
800 if (spacefree + dsize < newentrysize)
801 panic("ufs_direnter: compact1");
802 dirp->d_reclen = spacefree + dsize;
803 } else {
804 if (spacefree < newentrysize)
805 panic("ufs_direnter: compact2");
806 dirp->d_reclen = spacefree;
807 ep->d_reclen = dsize;
808 ep = (struct direct *)((char *)ep + dsize);
809 }
810 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
811
812 if (DOINGSOFTDEP(dvp)) {
813 softdep_setup_directory_add(bp, dp,
814 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
815 bdwrite(bp);
816 } else {
817 if (dvp->v_mount->mnt_flag & MNT_ASYNC) {
818 bdwrite(bp);
819 error = 0;
820 } else {
821 error = bowrite(bp);
822 }
823 }
824 dp->i_flag |= IN_CHANGE | IN_UPDATE;
825 /*
826 * If all went well, and the directory can be shortened, proceed
827 * with the truncation. Note that we have to unlock the inode for
828 * the entry that we just entered, as the truncation may need to
829 * lock other inodes which can lead to deadlock if we also hold a
830 * lock on the newly entered node.
831 */
832 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
833 if (tvp != NULL)
834 VOP_UNLOCK(tvp, 0, p);
835 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cr, p);
836 if (tvp != NULL)
837 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
838 }
839 return (error);
840 }
841
842 /*
843 * Remove a directory entry after a call to namei, using
844 * the parameters which it left in nameidata. The entry
845 * dp->i_offset contains the offset into the directory of the
846 * entry to be eliminated. The dp->i_count field contains the
847 * size of the previous record in the directory. If this
848 * is 0, the first entry is being deleted, so we need only
849 * zero the inode number to mark the entry as free. If the
850 * entry is not the first in the directory, we must reclaim
851 * the space of the now empty record by adding the record size
852 * to the size of the previous entry.
853 */
854 int
855 ufs_dirremove(dvp, ip, flags, isrmdir)
856 struct vnode *dvp;
857 struct inode *ip;
858 int flags;
859 int isrmdir;
860 {
861 struct inode *dp;
862 struct direct *ep;
863 struct buf *bp;
864 int error;
865
866 dp = VTOI(dvp);
867
868 if (flags & DOWHITEOUT) {
869 /*
870 * Whiteout entry: set d_ino to WINO.
871 */
872 if (error =
873 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp))
874 return (error);
875 ep->d_ino = WINO;
876 ep->d_type = DT_WHT;
877 goto out;
878 }
879
880 if ((error = UFS_BLKATOFF(dvp,
881 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
882 return (error);
883 if (dp->i_count == 0) {
884 /*
885 * First entry in block: set d_ino to zero.
886 */
887 ep->d_ino = 0;
888 } else {
889 /*
890 * Collapse new free space into previous entry.
891 */
892 ep->d_reclen += dp->i_reclen;
893 }
894 out:
895 if (ip) {
896 ip->i_effnlink--;
897 ip->i_flag |= IN_CHANGE;
898 }
899 if (DOINGSOFTDEP(dvp)) {
900 if (ip)
901 softdep_setup_remove(bp, dp, ip, isrmdir);
902 bdwrite(bp);
903 } else {
904 if (ip)
905 ip->i_nlink--;
906 if (flags & DOWHITEOUT)
907 error = VOP_BWRITE(bp);
908 else if (dvp->v_mount->mnt_flag & MNT_ASYNC
909 && dp->i_count != 0) {
910 bdwrite(bp);
911 error = 0;
912 } else
913 error = bowrite(bp);
914 }
915 dp->i_flag |= IN_CHANGE | IN_UPDATE;
916 return (error);
917 }
918
919 /*
920 * Rewrite an existing directory entry to point at the inode
921 * supplied. The parameters describing the directory entry are
922 * set up by a call to namei.
923 */
924 int
925 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
926 struct inode *dp, *oip;
927 ino_t newinum;
928 int newtype;
929 int isrmdir;
930 {
931 struct buf *bp;
932 struct direct *ep;
933 struct vnode *vdp = ITOV(dp);
934 int error;
935
936 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
937 if (error)
938 return (error);
939 ep->d_ino = newinum;
940 if (!OFSFMT(vdp))
941 ep->d_type = newtype;
942 oip->i_effnlink--;
943 oip->i_flag |= IN_CHANGE;
944 if (DOINGSOFTDEP(vdp)) {
945 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
946 bdwrite(bp);
947 } else {
948 oip->i_nlink--;
949 if (vdp->v_mount->mnt_flag & MNT_ASYNC) {
950 bdwrite(bp);
951 error = 0;
952 } else {
953 error = bowrite(bp);
954 }
955 }
956 dp->i_flag |= IN_CHANGE | IN_UPDATE;
957 return (error);
958 }
959
960 /*
961 * Check if a directory is empty or not.
962 * Inode supplied must be locked.
963 *
964 * Using a struct dirtemplate here is not precisely
965 * what we want, but better than using a struct direct.
966 *
967 * NB: does not handle corrupted directories.
968 */
969 int
970 ufs_dirempty(ip, parentino, cred)
971 register struct inode *ip;
972 ino_t parentino;
973 struct ucred *cred;
974 {
975 register off_t off;
976 struct dirtemplate dbuf;
977 register struct direct *dp = (struct direct *)&dbuf;
978 int error, count, namlen;
979 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
980
981 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
982 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
983 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
984 /*
985 * Since we read MINDIRSIZ, residual must
986 * be 0 unless we're at end of file.
987 */
988 if (error || count != 0)
989 return (0);
990 /* avoid infinite loops */
991 if (dp->d_reclen == 0)
992 return (0);
993 /* skip empty entries */
994 if (dp->d_ino == 0 || dp->d_ino == WINO)
995 continue;
996 /* accept only "." and ".." */
997 # if (BYTE_ORDER == LITTLE_ENDIAN)
998 if (OFSFMT(ITOV(ip)))
999 namlen = dp->d_type;
1000 else
1001 namlen = dp->d_namlen;
1002 # else
1003 namlen = dp->d_namlen;
1004 # endif
1005 if (namlen > 2)
1006 return (0);
1007 if (dp->d_name[0] != '.')
1008 return (0);
1009 /*
1010 * At this point namlen must be 1 or 2.
1011 * 1 implies ".", 2 implies ".." if second
1012 * char is also "."
1013 */
1014 if (namlen == 1 && dp->d_ino == ip->i_number)
1015 continue;
1016 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1017 continue;
1018 return (0);
1019 }
1020 return (1);
1021 }
1022
1023 /*
1024 * Check if source directory is in the path of the target directory.
1025 * Target is supplied locked, source is unlocked.
1026 * The target is always vput before returning.
1027 */
1028 int
1029 ufs_checkpath(source, target, cred)
1030 struct inode *source, *target;
1031 struct ucred *cred;
1032 {
1033 struct vnode *vp;
1034 int error, rootino, namlen;
1035 struct dirtemplate dirbuf;
1036
1037 vp = ITOV(target);
1038 if (target->i_number == source->i_number) {
1039 error = EEXIST;
1040 goto out;
1041 }
1042 rootino = ROOTINO;
1043 error = 0;
1044 if (target->i_number == rootino)
1045 goto out;
1046
1047 for (;;) {
1048 if (vp->v_type != VDIR) {
1049 error = ENOTDIR;
1050 break;
1051 }
1052 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1053 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1054 IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
1055 if (error != 0)
1056 break;
1057 # if (BYTE_ORDER == LITTLE_ENDIAN)
1058 if (OFSFMT(vp))
1059 namlen = dirbuf.dotdot_type;
1060 else
1061 namlen = dirbuf.dotdot_namlen;
1062 # else
1063 namlen = dirbuf.dotdot_namlen;
1064 # endif
1065 if (namlen != 2 ||
1066 dirbuf.dotdot_name[0] != '.' ||
1067 dirbuf.dotdot_name[1] != '.') {
1068 error = ENOTDIR;
1069 break;
1070 }
1071 if (dirbuf.dotdot_ino == source->i_number) {
1072 error = EINVAL;
1073 break;
1074 }
1075 if (dirbuf.dotdot_ino == rootino)
1076 break;
1077 vput(vp);
1078 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1079 if (error) {
1080 vp = NULL;
1081 break;
1082 }
1083 }
1084
1085 out:
1086 if (error == ENOTDIR)
1087 printf("checkpath: .. not a directory\n");
1088 if (vp != NULL)
1089 vput(vp);
1090 return (error);
1091 }
Cache object: b587415b3365468c5e1a01c61fe7dfe9
|