proc->thread stage 4: rework the VFS and DEVICE subsystems to take thread
[dragonfly.git] / sys / vfs / ufs / ufs_lookup.c
CommitLineData
984263bc
MD
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: src/sys/ufs/ufs/ufs_lookup.c,v 1.33.2.7 2001/09/22 19:22:13 iedowse Exp $
dadab5e9 40 * $DragonFly: src/sys/vfs/ufs/ufs_lookup.c,v 1.3 2003/06/25 03:56:12 dillon Exp $
984263bc
MD
41 */
42
43#include "opt_ufs.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/kernel.h>
984263bc
MD
48#include <sys/buf.h>
49#include <sys/proc.h>
dadab5e9 50#include <sys/namei.h>
984263bc
MD
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/quota.h>
60#include <ufs/ufs/inode.h>
61#include <ufs/ufs/dir.h>
62#ifdef UFS_DIRHASH
63#include <ufs/ufs/dirhash.h>
64#endif
65#include <ufs/ufs/ufsmount.h>
66#include <ufs/ufs/ufs_extern.h>
67
68#ifdef DIAGNOSTIC
69int dirchk = 1;
70#else
71int dirchk = 0;
72#endif
73
74SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
75
76/* true if old FS format...*/
77#define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
78
79/*
80 * Convert a component of a pathname into a pointer to a locked inode.
81 * This is a very central and rather complicated routine.
82 * If the file system is not maintained in a strict tree hierarchy,
83 * this can result in a deadlock situation (see comments in code below).
84 *
85 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
86 * on whether the name is to be looked up, created, renamed, or deleted.
87 * When CREATE, RENAME, or DELETE is specified, information usable in
88 * creating, renaming, or deleting a directory entry may be calculated.
89 * If flag has LOCKPARENT or'ed into it and the target of the pathname
90 * exists, lookup returns both the target and its parent directory locked.
91 * When creating or renaming and LOCKPARENT is specified, the target may
92 * not be ".". When deleting and LOCKPARENT is specified, the target may
93 * be "."., but the caller must check to ensure it does an vrele and vput
94 * instead of two vputs.
95 *
96 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
97 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
98 * method.
99 *
100 * vfs_cache_lookup() performs the following for us:
101 * check that it is a directory
102 * check accessibility of directory
103 * check for modification attempts on read-only mounts
104 * if name found in cache
105 * if at end of path and deleting or creating
106 * drop it
107 * else
108 * return name.
109 * return VOP_CACHEDLOOKUP()
110 *
111 * Overall outline of ufs_lookup:
112 *
113 * search for name in directory, to found or notfound
114 * notfound:
115 * if creating, return locked directory, leaving info on available slots
116 * else return error
117 * found:
118 * if at end of path and deleting, return information to allow delete
119 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
120 * inode and return info to allow rewrite
121 * if not at end, add name to cache; if at end and neither creating
122 * nor deleting, add name to cache
123 */
124int
125ufs_lookup(ap)
126 struct vop_cachedlookup_args /* {
127 struct vnode *a_dvp;
128 struct vnode **a_vpp;
129 struct componentname *a_cnp;
130 } */ *ap;
131{
132 register struct vnode *vdp; /* vnode for directory being searched */
133 register struct inode *dp; /* inode for directory being searched */
134 struct buf *bp; /* a buffer of directory entries */
135 struct direct *ep; /* the current directory entry */
136 int entryoffsetinblock; /* offset of ep in bp's buffer */
137 enum {NONE, COMPACT, FOUND} slotstatus;
138 doff_t slotoffset; /* offset of area with free space */
139 int slotsize; /* size of area at slotoffset */
140 int slotfreespace; /* amount of space free in slot */
141 int slotneeded; /* size of the entry we're seeking */
142 int numdirpasses; /* strategy for directory search */
143 doff_t endsearch; /* offset to end directory search */
144 doff_t prevoff; /* prev entry dp->i_offset */
145 struct vnode *pdp; /* saved dp during symlink work */
146 struct vnode *tdp; /* returned by VFS_VGET */
147 doff_t enduseful; /* pointer past last used dir slot */
148 u_long bmask; /* block offset mask */
149 int lockparent; /* 1 => lockparent flag is set */
150 int wantparent; /* 1 => wantparent or lockparent flag */
151 int namlen, error;
152 struct vnode **vpp = ap->a_vpp;
153 struct componentname *cnp = ap->a_cnp;
154 struct ucred *cred = cnp->cn_cred;
155 int flags = cnp->cn_flags;
156 int nameiop = cnp->cn_nameiop;
dadab5e9 157 struct thread *td = cnp->cn_td;
984263bc
MD
158
159 bp = NULL;
160 slotoffset = -1;
161 cnp->cn_flags &= ~PDIRUNLOCK;
162/*
163 * XXX there was a soft-update diff about this I couldn't merge.
164 * I think this was the equiv.
165 */
166 *vpp = NULL;
167
168 vdp = ap->a_dvp;
169 dp = VTOI(vdp);
170 lockparent = flags & LOCKPARENT;
171 wantparent = flags & (LOCKPARENT|WANTPARENT);
172
173 /*
174 * We now have a segment name to search for, and a directory to search.
175 *
176 * Suppress search for slots unless creating
177 * file and at end of pathname, in which case
178 * we watch for a place to put the new file in
179 * case it doesn't already exist.
180 */
181 slotstatus = FOUND;
182 slotfreespace = slotsize = slotneeded = 0;
183 if ((nameiop == CREATE || nameiop == RENAME) &&
184 (flags & ISLASTCN)) {
185 slotstatus = NONE;
186 slotneeded = DIRECTSIZ(cnp->cn_namelen);
187 }
188 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
189
190#ifdef UFS_DIRHASH
191 /*
192 * Use dirhash for fast operations on large directories. The logic
193 * to determine whether to hash the directory is contained within
194 * ufsdirhash_build(); a zero return means that it decided to hash
195 * this directory and it successfully built up the hash table.
196 */
197 if (ufsdirhash_build(dp) == 0) {
198 /* Look for a free slot if needed. */
199 enduseful = dp->i_size;
200 if (slotstatus != FOUND) {
201 slotoffset = ufsdirhash_findfree(dp, slotneeded,
202 &slotsize);
203 if (slotoffset >= 0) {
204 slotstatus = COMPACT;
205 enduseful = ufsdirhash_enduseful(dp);
206 if (enduseful < 0)
207 enduseful = dp->i_size;
208 }
209 }
210 /* Look up the component. */
211 numdirpasses = 1;
212 entryoffsetinblock = 0; /* silence compiler warning */
213 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
214 &dp->i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
215 case 0:
216 ep = (struct direct *)((char *)bp->b_data +
217 (dp->i_offset & bmask));
218 goto foundentry;
219 case ENOENT:
220 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
221 goto notfound;
222 default:
223 /* Something failed; just do a linear search. */
224 break;
225 }
226 }
227#endif /* UFS_DIRHASH */
228 /*
229 * If there is cached information on a previous search of
230 * this directory, pick up where we last left off.
231 * We cache only lookups as these are the most common
232 * and have the greatest payoff. Caching CREATE has little
233 * benefit as it usually must search the entire directory
234 * to determine that the entry does not exist. Caching the
235 * location of the last DELETE or RENAME has not reduced
236 * profiling time and hence has been removed in the interest
237 * of simplicity.
238 */
239 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
240 dp->i_diroff >= dp->i_size) {
241 entryoffsetinblock = 0;
242 dp->i_offset = 0;
243 numdirpasses = 1;
244 } else {
245 dp->i_offset = dp->i_diroff;
246 if ((entryoffsetinblock = dp->i_offset & bmask) &&
247 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
248 return (error);
249 numdirpasses = 2;
250 nchstats.ncs_2passes++;
251 }
252 prevoff = dp->i_offset;
253 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
254 enduseful = 0;
255
256searchloop:
257 while (dp->i_offset < endsearch) {
258 /*
259 * If necessary, get the next directory block.
260 */
261 if ((dp->i_offset & bmask) == 0) {
262 if (bp != NULL)
263 brelse(bp);
264 error =
265 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
266 if (error)
267 return (error);
268 entryoffsetinblock = 0;
269 }
270 /*
271 * If still looking for a slot, and at a DIRBLKSIZE
272 * boundary, have to start looking for free space again.
273 */
274 if (slotstatus == NONE &&
275 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
276 slotoffset = -1;
277 slotfreespace = 0;
278 }
279 /*
280 * Get pointer to next entry.
281 * Full validation checks are slow, so we only check
282 * enough to insure forward progress through the
283 * directory. Complete checks can be run by patching
284 * "dirchk" to be true.
285 */
286 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
287 if (ep->d_reclen == 0 || ep->d_reclen >
288 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
289 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
290 int i;
291
292 ufs_dirbad(dp, dp->i_offset, "mangled entry");
293 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
294 dp->i_offset += i;
295 entryoffsetinblock += i;
296 continue;
297 }
298
299 /*
300 * If an appropriate sized slot has not yet been found,
301 * check to see if one is available. Also accumulate space
302 * in the current block so that we can determine if
303 * compaction is viable.
304 */
305 if (slotstatus != FOUND) {
306 int size = ep->d_reclen;
307
308 if (ep->d_ino != 0)
309 size -= DIRSIZ(OFSFMT(vdp), ep);
310 if (size > 0) {
311 if (size >= slotneeded) {
312 slotstatus = FOUND;
313 slotoffset = dp->i_offset;
314 slotsize = ep->d_reclen;
315 } else if (slotstatus == NONE) {
316 slotfreespace += size;
317 if (slotoffset == -1)
318 slotoffset = dp->i_offset;
319 if (slotfreespace >= slotneeded) {
320 slotstatus = COMPACT;
321 slotsize = dp->i_offset +
322 ep->d_reclen - slotoffset;
323 }
324 }
325 }
326 }
327
328 /*
329 * Check for a name match.
330 */
331 if (ep->d_ino) {
332# if (BYTE_ORDER == LITTLE_ENDIAN)
333 if (OFSFMT(vdp))
334 namlen = ep->d_type;
335 else
336 namlen = ep->d_namlen;
337# else
338 namlen = ep->d_namlen;
339# endif
340 if (namlen == cnp->cn_namelen &&
341 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
342 !bcmp(cnp->cn_nameptr, ep->d_name,
343 (unsigned)namlen)) {
344#ifdef UFS_DIRHASH
345foundentry:
346#endif
347 /*
348 * Save directory entry's inode number and
349 * reclen in ndp->ni_ufs area, and release
350 * directory buffer.
351 */
352 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
353 ep->d_type == DT_WHT) {
354 slotstatus = FOUND;
355 slotoffset = dp->i_offset;
356 slotsize = ep->d_reclen;
357 dp->i_reclen = slotsize;
358 enduseful = dp->i_size;
359 ap->a_cnp->cn_flags |= ISWHITEOUT;
360 numdirpasses--;
361 goto notfound;
362 }
363 dp->i_ino = ep->d_ino;
364 dp->i_reclen = ep->d_reclen;
365 goto found;
366 }
367 }
368 prevoff = dp->i_offset;
369 dp->i_offset += ep->d_reclen;
370 entryoffsetinblock += ep->d_reclen;
371 if (ep->d_ino)
372 enduseful = dp->i_offset;
373 }
374notfound:
375 /*
376 * If we started in the middle of the directory and failed
377 * to find our target, we must check the beginning as well.
378 */
379 if (numdirpasses == 2) {
380 numdirpasses--;
381 dp->i_offset = 0;
382 endsearch = dp->i_diroff;
383 goto searchloop;
384 }
385 if (bp != NULL)
386 brelse(bp);
387 /*
388 * If creating, and at end of pathname and current
389 * directory has not been removed, then can consider
390 * allowing file to be created.
391 */
392 if ((nameiop == CREATE || nameiop == RENAME ||
393 (nameiop == DELETE &&
394 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
395 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
396 (flags & ISLASTCN) && dp->i_effnlink != 0) {
397 /*
398 * Access for write is interpreted as allowing
399 * creation of files in the directory.
400 */
dadab5e9 401 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
984263bc
MD
402 if (error)
403 return (error);
404 /*
405 * Return an indication of where the new directory
406 * entry should be put. If we didn't find a slot,
407 * then set dp->i_count to 0 indicating
408 * that the new slot belongs at the end of the
409 * directory. If we found a slot, then the new entry
410 * can be put in the range from dp->i_offset to
411 * dp->i_offset + dp->i_count.
412 */
413 if (slotstatus == NONE) {
414 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
415 dp->i_count = 0;
416 enduseful = dp->i_offset;
417 } else if (nameiop == DELETE) {
418 dp->i_offset = slotoffset;
419 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
420 dp->i_count = 0;
421 else
422 dp->i_count = dp->i_offset - prevoff;
423 } else {
424 dp->i_offset = slotoffset;
425 dp->i_count = slotsize;
426 if (enduseful < slotoffset + slotsize)
427 enduseful = slotoffset + slotsize;
428 }
429 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
430 dp->i_flag |= IN_CHANGE | IN_UPDATE;
431 /*
432 * We return with the directory locked, so that
433 * the parameters we set up above will still be
434 * valid if we actually decide to do a direnter().
435 * We return ni_vp == NULL to indicate that the entry
436 * does not currently exist; we leave a pointer to
437 * the (locked) directory inode in ndp->ni_dvp.
438 * The pathname buffer is saved so that the name
439 * can be obtained later.
440 *
441 * NB - if the directory is unlocked, then this
442 * information cannot be used.
443 */
444 cnp->cn_flags |= SAVENAME;
445 if (!lockparent) {
dadab5e9 446 VOP_UNLOCK(vdp, 0, td);
984263bc
MD
447 cnp->cn_flags |= PDIRUNLOCK;
448 }
449 return (EJUSTRETURN);
450 }
451 /*
452 * Insert name into cache (as non-existent) if appropriate.
453 */
454 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
455 cache_enter(vdp, *vpp, cnp);
456 return (ENOENT);
457
458found:
459 if (numdirpasses == 2)
460 nchstats.ncs_pass2++;
461 /*
462 * Check that directory length properly reflects presence
463 * of this entry.
464 */
465 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
466 ufs_dirbad(dp, dp->i_offset, "i_size too small");
467 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
468 dp->i_flag |= IN_CHANGE | IN_UPDATE;
469 }
470 brelse(bp);
471
472 /*
473 * Found component in pathname.
474 * If the final component of path name, save information
475 * in the cache as to where the entry was found.
476 */
477 if ((flags & ISLASTCN) && nameiop == LOOKUP)
478 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
479
480 /*
481 * If deleting, and at end of pathname, return
482 * parameters which can be used to remove file.
483 * If the wantparent flag isn't set, we return only
484 * the directory (in ndp->ni_dvp), otherwise we go
485 * on and lock the inode, being careful with ".".
486 */
487 if (nameiop == DELETE && (flags & ISLASTCN)) {
488 /*
489 * Write access to directory required to delete files.
490 */
dadab5e9 491 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
984263bc
MD
492 if (error)
493 return (error);
494 /*
495 * Return pointer to current entry in dp->i_offset,
496 * and distance past previous entry (if there
497 * is a previous entry in this block) in dp->i_count.
498 * Save directory inode pointer in ndp->ni_dvp for dirremove().
499 */
500 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
501 dp->i_count = 0;
502 else
503 dp->i_count = dp->i_offset - prevoff;
504 if (dp->i_number == dp->i_ino) {
505 VREF(vdp);
506 *vpp = vdp;
507 return (0);
508 }
509 if (flags & ISDOTDOT)
dadab5e9 510 VOP_UNLOCK(vdp, 0, td); /* race to get the inode */
984263bc
MD
511 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
512 if (flags & ISDOTDOT) {
dadab5e9 513 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, td) != 0)
984263bc
MD
514 cnp->cn_flags |= PDIRUNLOCK;
515 }
516 if (error)
517 return (error);
518 /*
519 * If directory is "sticky", then user must own
520 * the directory, or the file in it, else she
521 * may not delete it (unless she's root). This
522 * implements append-only directories.
523 */
524 if ((dp->i_mode & ISVTX) &&
525 cred->cr_uid != 0 &&
526 cred->cr_uid != dp->i_uid &&
527 VTOI(tdp)->i_uid != cred->cr_uid) {
528 vput(tdp);
529 return (EPERM);
530 }
531 *vpp = tdp;
532 if (!lockparent) {
dadab5e9 533 VOP_UNLOCK(vdp, 0, td);
984263bc
MD
534 cnp->cn_flags |= PDIRUNLOCK;
535 }
536 return (0);
537 }
538
539 /*
540 * If rewriting (RENAME), return the inode and the
541 * information required to rewrite the present directory
542 * Must get inode of directory entry to verify it's a
543 * regular file, or empty directory.
544 */
545 if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
dadab5e9 546 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
984263bc
MD
547 return (error);
548 /*
549 * Careful about locking second inode.
550 * This can only occur if the target is ".".
551 */
552 if (dp->i_number == dp->i_ino)
553 return (EISDIR);
554 if (flags & ISDOTDOT)
dadab5e9 555 VOP_UNLOCK(vdp, 0, td); /* race to get the inode */
984263bc
MD
556 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
557 if (flags & ISDOTDOT) {
dadab5e9 558 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, td) != 0)
984263bc
MD
559 cnp->cn_flags |= PDIRUNLOCK;
560 }
561 if (error)
562 return (error);
563 *vpp = tdp;
564 cnp->cn_flags |= SAVENAME;
565 if (!lockparent) {
dadab5e9 566 VOP_UNLOCK(vdp, 0, td);
984263bc
MD
567 cnp->cn_flags |= PDIRUNLOCK;
568 }
569 return (0);
570 }
571
572 /*
573 * Step through the translation in the name. We do not `vput' the
574 * directory because we may need it again if a symbolic link
575 * is relative to the current directory. Instead we save it
576 * unlocked as "pdp". We must get the target inode before unlocking
577 * the directory to insure that the inode will not be removed
578 * before we get it. We prevent deadlock by always fetching
579 * inodes from the root, moving down the directory tree. Thus
580 * when following backward pointers ".." we must unlock the
581 * parent directory before getting the requested directory.
582 * There is a potential race condition here if both the current
583 * and parent directories are removed before the VFS_VGET for the
584 * inode associated with ".." returns. We hope that this occurs
585 * infrequently since we cannot avoid this race condition without
586 * implementing a sophisticated deadlock detection algorithm.
587 * Note also that this simple deadlock detection scheme will not
588 * work if the file system has any hard links other than ".."
589 * that point backwards in the directory structure.
590 */
591 pdp = vdp;
592 if (flags & ISDOTDOT) {
dadab5e9 593 VOP_UNLOCK(pdp, 0, td); /* race to get the inode */
984263bc
MD
594 cnp->cn_flags |= PDIRUNLOCK;
595 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
dadab5e9 596 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, td) == 0)
984263bc
MD
597 cnp->cn_flags &= ~PDIRUNLOCK;
598 return (error);
599 }
600 if (lockparent && (flags & ISLASTCN)) {
dadab5e9 601 if ((error = vn_lock(pdp, LK_EXCLUSIVE, td)) != 0) {
984263bc
MD
602 vput(tdp);
603 return (error);
604 }
605 cnp->cn_flags &= ~PDIRUNLOCK;
606 }
607 *vpp = tdp;
608 } else if (dp->i_number == dp->i_ino) {
609 VREF(vdp); /* we want ourself, ie "." */
610 *vpp = vdp;
611 } else {
612 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
613 if (error)
614 return (error);
615 if (!lockparent || !(flags & ISLASTCN)) {
dadab5e9 616 VOP_UNLOCK(pdp, 0, td);
984263bc
MD
617 cnp->cn_flags |= PDIRUNLOCK;
618 }
619 *vpp = tdp;
620 }
621
622 /*
623 * Insert name into cache if appropriate.
624 */
625 if (cnp->cn_flags & MAKEENTRY)
626 cache_enter(vdp, *vpp, cnp);
627 return (0);
628}
629
630void
631ufs_dirbad(ip, offset, how)
632 struct inode *ip;
633 doff_t offset;
634 char *how;
635{
636 struct mount *mp;
637
638 mp = ITOV(ip)->v_mount;
639 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
640 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
641 if ((mp->mnt_flag & MNT_RDONLY) == 0)
642 panic("ufs_dirbad: bad dir");
643}
644
645/*
646 * Do consistency checking on a directory entry:
647 * record length must be multiple of 4
648 * entry must fit in rest of its DIRBLKSIZ block
649 * record must be large enough to contain entry
650 * name is not longer than MAXNAMLEN
651 * name must be as long as advertised, and null terminated
652 */
653int
654ufs_dirbadentry(dp, ep, entryoffsetinblock)
655 struct vnode *dp;
656 register struct direct *ep;
657 int entryoffsetinblock;
658{
659 register int i;
660 int namlen;
661
662# if (BYTE_ORDER == LITTLE_ENDIAN)
663 if (OFSFMT(dp))
664 namlen = ep->d_type;
665 else
666 namlen = ep->d_namlen;
667# else
668 namlen = ep->d_namlen;
669# endif
670 if ((ep->d_reclen & 0x3) != 0 ||
671 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
672 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
673 /*return (1); */
674 printf("First bad\n");
675 goto bad;
676 }
677 if (ep->d_ino == 0)
678 return (0);
679 for (i = 0; i < namlen; i++)
680 if (ep->d_name[i] == '\0') {
681 /*return (1); */
682 printf("Second bad\n");
683 goto bad;
684 }
685 if (ep->d_name[i])
686 goto bad;
687 return (0);
688bad:
689 return (1);
690}
691
692/*
693 * Construct a new directory entry after a call to namei, using the
694 * parameters that it left in the componentname argument cnp. The
695 * argument ip is the inode to which the new directory entry will refer.
696 */
697void
698ufs_makedirentry(ip, cnp, newdirp)
699 struct inode *ip;
700 struct componentname *cnp;
701 struct direct *newdirp;
702{
703
704#ifdef DIAGNOSTIC
705 if ((cnp->cn_flags & SAVENAME) == 0)
706 panic("ufs_makedirentry: missing name");
707#endif
708 newdirp->d_ino = ip->i_number;
709 newdirp->d_namlen = cnp->cn_namelen;
710 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
711 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
712 newdirp->d_type = IFTODT(ip->i_mode);
713 else {
714 newdirp->d_type = 0;
715# if (BYTE_ORDER == LITTLE_ENDIAN)
716 { u_char tmp = newdirp->d_namlen;
717 newdirp->d_namlen = newdirp->d_type;
718 newdirp->d_type = tmp; }
719# endif
720 }
721}
722
723/*
724 * Write a directory entry after a call to namei, using the parameters
725 * that it left in nameidata. The argument dirp is the new directory
726 * entry contents. Dvp is a pointer to the directory to be written,
727 * which was left locked by namei. Remaining parameters (dp->i_offset,
728 * dp->i_count) indicate how the space for the new entry is to be obtained.
729 * Non-null bp indicates that a directory is being created (for the
730 * soft dependency code).
731 */
732int
733ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
734 struct vnode *dvp;
735 struct vnode *tvp;
736 struct direct *dirp;
737 struct componentname *cnp;
738 struct buf *newdirbp;
739{
dadab5e9
MD
740 struct ucred *cred;
741 struct thread *td = curthread; /* XXX */
984263bc
MD
742 int newentrysize;
743 struct inode *dp;
744 struct buf *bp;
745 u_int dsize;
746 struct direct *ep, *nep;
747 int error, ret, blkoff, loc, spacefree, flags;
748 char *dirbuf;
749
dadab5e9
MD
750 KKASSERT(td->td_proc);
751 cred = td->td_proc->p_ucred;
984263bc
MD
752
753 dp = VTOI(dvp);
754 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
755
756 if (dp->i_count == 0) {
757 /*
758 * If dp->i_count is 0, then namei could find no
759 * space in the directory. Here, dp->i_offset will
760 * be on a directory block boundary and we will write the
761 * new entry into a fresh block.
762 */
763 if (dp->i_offset & (DIRBLKSIZ - 1))
764 panic("ufs_direnter: newblk");
765 flags = B_CLRBUF;
766 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
767 flags |= B_SYNC;
768 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
dadab5e9 769 cred, flags, &bp)) != 0) {
984263bc
MD
770 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
771 bdwrite(newdirbp);
772 return (error);
773 }
774 dp->i_size = dp->i_offset + DIRBLKSIZ;
775 dp->i_flag |= IN_CHANGE | IN_UPDATE;
776 vnode_pager_setsize(dvp, (u_long)dp->i_size);
777 dirp->d_reclen = DIRBLKSIZ;
778 blkoff = dp->i_offset &
779 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
780 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
781#ifdef UFS_DIRHASH
782 if (dp->i_dirhash != NULL) {
783 ufsdirhash_newblk(dp, dp->i_offset);
784 ufsdirhash_add(dp, dirp, dp->i_offset);
785 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
786 dp->i_offset);
787 }
788#endif
789 if (DOINGSOFTDEP(dvp)) {
790 /*
791 * Ensure that the entire newly allocated block is a
792 * valid directory so that future growth within the
793 * block does not have to ensure that the block is
794 * written before the inode.
795 */
796 blkoff += DIRBLKSIZ;
797 while (blkoff < bp->b_bcount) {
798 ((struct direct *)
799 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
800 blkoff += DIRBLKSIZ;
801 }
802 softdep_setup_directory_add(bp, dp, dp->i_offset,
803 dirp->d_ino, newdirbp);
804 bdwrite(bp);
805 return (UFS_UPDATE(dvp, 0));
806 }
807 if (DOINGASYNC(dvp)) {
808 bdwrite(bp);
809 return (UFS_UPDATE(dvp, 0));
810 }
811 error = VOP_BWRITE(bp->b_vp, bp);
812 ret = UFS_UPDATE(dvp, 1);
813 if (error == 0)
814 return (ret);
815 return (error);
816 }
817
818 /*
819 * If dp->i_count is non-zero, then namei found space for the new
820 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
821 * in the directory. To use this space, we may have to compact
822 * the entries located there, by copying them together towards the
823 * beginning of the block, leaving the free space in one usable
824 * chunk at the end.
825 */
826
827 /*
828 * Increase size of directory if entry eats into new space.
829 * This should never push the size past a new multiple of
830 * DIRBLKSIZE.
831 *
832 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
833 */
834 if (dp->i_offset + dp->i_count > dp->i_size)
835 dp->i_size = dp->i_offset + dp->i_count;
836 /*
837 * Get the block containing the space for the new directory entry.
838 */
839 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
840 if (error) {
841 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
842 bdwrite(newdirbp);
843 return (error);
844 }
845 /*
846 * Find space for the new entry. In the simple case, the entry at
847 * offset base will have the space. If it does not, then namei
848 * arranged that compacting the region dp->i_offset to
849 * dp->i_offset + dp->i_count would yield the space.
850 */
851 ep = (struct direct *)dirbuf;
852 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
853 spacefree = ep->d_reclen - dsize;
854 for (loc = ep->d_reclen; loc < dp->i_count; ) {
855 nep = (struct direct *)(dirbuf + loc);
856
857 /* Trim the existing slot (NB: dsize may be zero). */
858 ep->d_reclen = dsize;
859 ep = (struct direct *)((char *)ep + dsize);
860
861 /* Read nep->d_reclen now as the bcopy() may clobber it. */
862 loc += nep->d_reclen;
863 if (nep->d_ino == 0) {
864 /*
865 * A mid-block unused entry. Such entries are
866 * never created by the kernel, but fsck_ffs
867 * can create them (and it doesn't fix them).
868 *
869 * Add up the free space, and initialise the
870 * relocated entry since we don't bcopy it.
871 */
872 spacefree += nep->d_reclen;
873 ep->d_ino = 0;
874 dsize = 0;
875 continue;
876 }
877 dsize = DIRSIZ(OFSFMT(dvp), nep);
878 spacefree += nep->d_reclen - dsize;
879#ifdef UFS_DIRHASH
880 if (dp->i_dirhash != NULL)
881 ufsdirhash_move(dp, nep,
882 dp->i_offset + ((char *)nep - dirbuf),
883 dp->i_offset + ((char *)ep - dirbuf));
884#endif
885 if (DOINGSOFTDEP(dvp))
886 softdep_change_directoryentry_offset(dp, dirbuf,
887 (caddr_t)nep, (caddr_t)ep, dsize);
888 else
889 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
890 }
891 /*
892 * Here, `ep' points to a directory entry containing `dsize' in-use
893 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
894 * then the entry is completely unused (dsize == 0). The value
895 * of ep->d_reclen is always indeterminate.
896 *
897 * Update the pointer fields in the previous entry (if any),
898 * copy in the new entry, and write out the block.
899 */
900 if (ep->d_ino == 0 ||
901 (ep->d_ino == WINO &&
902 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
903 if (spacefree + dsize < newentrysize)
904 panic("ufs_direnter: compact1");
905 dirp->d_reclen = spacefree + dsize;
906 } else {
907 if (spacefree < newentrysize)
908 panic("ufs_direnter: compact2");
909 dirp->d_reclen = spacefree;
910 ep->d_reclen = dsize;
911 ep = (struct direct *)((char *)ep + dsize);
912 }
913#ifdef UFS_DIRHASH
914 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
915 dirp->d_reclen == spacefree))
916 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
917#endif
918 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
919#ifdef UFS_DIRHASH
920 if (dp->i_dirhash != NULL)
921 ufsdirhash_checkblock(dp, dirbuf -
922 (dp->i_offset & (DIRBLKSIZ - 1)),
923 dp->i_offset & ~(DIRBLKSIZ - 1));
924#endif
925
926 if (DOINGSOFTDEP(dvp)) {
927 softdep_setup_directory_add(bp, dp,
928 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
929 bdwrite(bp);
930 } else {
931 if (DOINGASYNC(dvp)) {
932 bdwrite(bp);
933 error = 0;
934 } else {
935 error = bowrite(bp);
936 }
937 }
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 && dp->i_endoff && dp->i_endoff < dp->i_size) {
947 if (tvp != NULL)
dadab5e9 948 VOP_UNLOCK(tvp, 0, td);
984263bc
MD
949#ifdef UFS_DIRHASH
950 if (dp->i_dirhash != NULL)
951 ufsdirhash_dirtrunc(dp, dp->i_endoff);
952#endif
dadab5e9 953 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cred, td);
984263bc 954 if (tvp != NULL)
dadab5e9 955 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
956 }
957 return (error);
958}
959
960/*
961 * Remove a directory entry after a call to namei, using
962 * the parameters which it left in nameidata. The entry
963 * dp->i_offset contains the offset into the directory of the
964 * entry to be eliminated. The dp->i_count field contains the
965 * size of the previous record in the directory. If this
966 * is 0, the first entry is being deleted, so we need only
967 * zero the inode number to mark the entry as free. If the
968 * entry is not the first in the directory, we must reclaim
969 * the space of the now empty record by adding the record size
970 * to the size of the previous entry.
971 */
972int
973ufs_dirremove(dvp, ip, flags, isrmdir)
974 struct vnode *dvp;
975 struct inode *ip;
976 int flags;
977 int isrmdir;
978{
979 struct inode *dp;
980 struct direct *ep;
981 struct buf *bp;
982 int error;
983
984 dp = VTOI(dvp);
985
986 if (flags & DOWHITEOUT) {
987 /*
988 * Whiteout entry: set d_ino to WINO.
989 */
990 if ((error =
991 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
992 return (error);
993 ep->d_ino = WINO;
994 ep->d_type = DT_WHT;
995 goto out;
996 }
997
998 if ((error = UFS_BLKATOFF(dvp,
999 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
1000 return (error);
1001#ifdef UFS_DIRHASH
1002 /*
1003 * Remove the dirhash entry. This is complicated by the fact
1004 * that `ep' is the previous entry when dp->i_count != 0.
1005 */
1006 if (dp->i_dirhash != NULL)
1007 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
1008 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
1009#endif
1010 if (dp->i_count == 0) {
1011 /*
1012 * First entry in block: set d_ino to zero.
1013 */
1014 ep->d_ino = 0;
1015 } else {
1016 /*
1017 * Collapse new free space into previous entry.
1018 */
1019 ep->d_reclen += dp->i_reclen;
1020 }
1021#ifdef UFS_DIRHASH
1022 if (dp->i_dirhash != NULL)
1023 ufsdirhash_checkblock(dp, (char *)ep -
1024 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1025 dp->i_offset & ~(DIRBLKSIZ - 1));
1026#endif
1027out:
1028 if (DOINGSOFTDEP(dvp)) {
1029 if (ip) {
1030 ip->i_effnlink--;
1031 softdep_change_linkcnt(ip);
1032 softdep_setup_remove(bp, dp, ip, isrmdir);
1033 }
1034 if (softdep_slowdown(dvp)) {
1035 error = VOP_BWRITE(bp->b_vp, bp);
1036 } else {
1037 bdwrite(bp);
1038 error = 0;
1039 }
1040 } else {
1041 if (ip) {
1042 ip->i_effnlink--;
1043 ip->i_nlink--;
1044 ip->i_flag |= IN_CHANGE;
1045 }
1046 if (flags & DOWHITEOUT)
1047 error = VOP_BWRITE(bp->b_vp, bp);
1048 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1049 bdwrite(bp);
1050 error = 0;
1051 } else
1052 error = bowrite(bp);
1053 }
1054 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1055 return (error);
1056}
1057
1058/*
1059 * Rewrite an existing directory entry to point at the inode
1060 * supplied. The parameters describing the directory entry are
1061 * set up by a call to namei.
1062 */
1063int
1064ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1065 struct inode *dp, *oip;
1066 ino_t newinum;
1067 int newtype;
1068 int isrmdir;
1069{
1070 struct buf *bp;
1071 struct direct *ep;
1072 struct vnode *vdp = ITOV(dp);
1073 int error;
1074
1075 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1076 if (error)
1077 return (error);
1078 ep->d_ino = newinum;
1079 if (!OFSFMT(vdp))
1080 ep->d_type = newtype;
1081 oip->i_effnlink--;
1082 if (DOINGSOFTDEP(vdp)) {
1083 softdep_change_linkcnt(oip);
1084 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1085 bdwrite(bp);
1086 } else {
1087 oip->i_nlink--;
1088 oip->i_flag |= IN_CHANGE;
1089 if (DOINGASYNC(vdp)) {
1090 bdwrite(bp);
1091 error = 0;
1092 } else {
1093 error = bowrite(bp);
1094 }
1095 }
1096 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1097 return (error);
1098}
1099
1100/*
1101 * Check if a directory is empty or not.
1102 * Inode supplied must be locked.
1103 *
1104 * Using a struct dirtemplate here is not precisely
1105 * what we want, but better than using a struct direct.
1106 *
1107 * NB: does not handle corrupted directories.
1108 */
1109int
1110ufs_dirempty(ip, parentino, cred)
1111 register struct inode *ip;
1112 ino_t parentino;
1113 struct ucred *cred;
1114{
1115 register off_t off;
1116 struct dirtemplate dbuf;
1117 register struct direct *dp = (struct direct *)&dbuf;
1118 int error, count, namlen;
1119#define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1120
1121 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1122 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
dadab5e9 1123 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
984263bc
MD
1124 /*
1125 * Since we read MINDIRSIZ, residual must
1126 * be 0 unless we're at end of file.
1127 */
1128 if (error || count != 0)
1129 return (0);
1130 /* avoid infinite loops */
1131 if (dp->d_reclen == 0)
1132 return (0);
1133 /* skip empty entries */
1134 if (dp->d_ino == 0 || dp->d_ino == WINO)
1135 continue;
1136 /* accept only "." and ".." */
1137# if (BYTE_ORDER == LITTLE_ENDIAN)
1138 if (OFSFMT(ITOV(ip)))
1139 namlen = dp->d_type;
1140 else
1141 namlen = dp->d_namlen;
1142# else
1143 namlen = dp->d_namlen;
1144# endif
1145 if (namlen > 2)
1146 return (0);
1147 if (dp->d_name[0] != '.')
1148 return (0);
1149 /*
1150 * At this point namlen must be 1 or 2.
1151 * 1 implies ".", 2 implies ".." if second
1152 * char is also "."
1153 */
1154 if (namlen == 1 && dp->d_ino == ip->i_number)
1155 continue;
1156 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1157 continue;
1158 return (0);
1159 }
1160 return (1);
1161}
1162
1163/*
1164 * Check if source directory is in the path of the target directory.
1165 * Target is supplied locked, source is unlocked.
1166 * The target is always vput before returning.
1167 */
1168int
1169ufs_checkpath(source, target, cred)
1170 struct inode *source, *target;
1171 struct ucred *cred;
1172{
1173 struct vnode *vp;
1174 int error, rootino, namlen;
1175 struct dirtemplate dirbuf;
1176
1177 vp = ITOV(target);
1178 if (target->i_number == source->i_number) {
1179 error = EEXIST;
1180 goto out;
1181 }
1182 rootino = ROOTINO;
1183 error = 0;
1184 if (target->i_number == rootino)
1185 goto out;
1186
1187 for (;;) {
1188 if (vp->v_type != VDIR) {
1189 error = ENOTDIR;
1190 break;
1191 }
1192 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1193 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
dadab5e9 1194 IO_NODELOCKED, cred, (int *)0, NULL);
984263bc
MD
1195 if (error != 0)
1196 break;
1197# if (BYTE_ORDER == LITTLE_ENDIAN)
1198 if (OFSFMT(vp))
1199 namlen = dirbuf.dotdot_type;
1200 else
1201 namlen = dirbuf.dotdot_namlen;
1202# else
1203 namlen = dirbuf.dotdot_namlen;
1204# endif
1205 if (namlen != 2 ||
1206 dirbuf.dotdot_name[0] != '.' ||
1207 dirbuf.dotdot_name[1] != '.') {
1208 error = ENOTDIR;
1209 break;
1210 }
1211 if (dirbuf.dotdot_ino == source->i_number) {
1212 error = EINVAL;
1213 break;
1214 }
1215 if (dirbuf.dotdot_ino == rootino)
1216 break;
1217 vput(vp);
1218 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1219 if (error) {
1220 vp = NULL;
1221 break;
1222 }
1223 }
1224
1225out:
1226 if (error == ENOTDIR)
1227 printf("checkpath: .. not a directory\n");
1228 if (vp != NULL)
1229 vput(vp);
1230 return (error);
1231}