Simplify vn_lock(), VOP_LOCK(), and VOP_UNLOCK() by removing the thread_t
[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 $
ca466bae 40 * $DragonFly: src/sys/vfs/ufs/ufs_lookup.c,v 1.24 2006/05/05 21:15:10 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
1f2de5d4
MD
59#include "quota.h"
60#include "inode.h"
61#include "dir.h"
984263bc 62#ifdef UFS_DIRHASH
1f2de5d4 63#include "dirhash.h"
984263bc 64#endif
1f2de5d4
MD
65#include "ufsmount.h"
66#include "ufs_extern.h"
984263bc
MD
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.
f719c866 82 * If the filesystem is not maintained in a strict tree hierarchy,
984263bc
MD
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 *
984263bc
MD
96 * Overall outline of ufs_lookup:
97 *
98 * search for name in directory, to found or notfound
99 * notfound:
100 * if creating, return locked directory, leaving info on available slots
101 * else return error
102 * found:
103 * if at end of path and deleting, return information to allow delete
104 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
105 * inode and return info to allow rewrite
106 * if not at end, add name to cache; if at end and neither creating
107 * nor deleting, add name to cache
0973c589
CP
108 *
109 * ufs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
110 * struct componentname *a_cnp)
984263bc
MD
111 */
112int
e62afb5f 113ufs_lookup(struct vop_old_lookup_args *ap)
984263bc 114{
3ff2135f
RG
115 struct vnode *vdp; /* vnode for directory being searched */
116 struct inode *dp; /* inode for directory being searched */
984263bc
MD
117 struct buf *bp; /* a buffer of directory entries */
118 struct direct *ep; /* the current directory entry */
119 int entryoffsetinblock; /* offset of ep in bp's buffer */
120 enum {NONE, COMPACT, FOUND} slotstatus;
121 doff_t slotoffset; /* offset of area with free space */
122 int slotsize; /* size of area at slotoffset */
123 int slotfreespace; /* amount of space free in slot */
124 int slotneeded; /* size of the entry we're seeking */
125 int numdirpasses; /* strategy for directory search */
126 doff_t endsearch; /* offset to end directory search */
127 doff_t prevoff; /* prev entry dp->i_offset */
128 struct vnode *pdp; /* saved dp during symlink work */
129 struct vnode *tdp; /* returned by VFS_VGET */
130 doff_t enduseful; /* pointer past last used dir slot */
131 u_long bmask; /* block offset mask */
132 int lockparent; /* 1 => lockparent flag is set */
133 int wantparent; /* 1 => wantparent or lockparent flag */
134 int namlen, error;
135 struct vnode **vpp = ap->a_vpp;
136 struct componentname *cnp = ap->a_cnp;
137 struct ucred *cred = cnp->cn_cred;
138 int flags = cnp->cn_flags;
139 int nameiop = cnp->cn_nameiop;
24e51f36 140 globaldata_t gd = mycpu;
984263bc
MD
141
142 bp = NULL;
143 slotoffset = -1;
2b69e610 144 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
145/*
146 * XXX there was a soft-update diff about this I couldn't merge.
147 * I think this was the equiv.
148 */
149 *vpp = NULL;
150
151 vdp = ap->a_dvp;
152 dp = VTOI(vdp);
2b69e610
MD
153 lockparent = flags & CNP_LOCKPARENT;
154 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
984263bc
MD
155
156 /*
157 * We now have a segment name to search for, and a directory to search.
158 *
159 * Suppress search for slots unless creating
160 * file and at end of pathname, in which case
161 * we watch for a place to put the new file in
162 * case it doesn't already exist.
163 */
164 slotstatus = FOUND;
165 slotfreespace = slotsize = slotneeded = 0;
fad57d0e 166 if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
984263bc
MD
167 slotstatus = NONE;
168 slotneeded = DIRECTSIZ(cnp->cn_namelen);
169 }
170 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
171
172#ifdef UFS_DIRHASH
173 /*
174 * Use dirhash for fast operations on large directories. The logic
175 * to determine whether to hash the directory is contained within
176 * ufsdirhash_build(); a zero return means that it decided to hash
177 * this directory and it successfully built up the hash table.
178 */
179 if (ufsdirhash_build(dp) == 0) {
180 /* Look for a free slot if needed. */
181 enduseful = dp->i_size;
182 if (slotstatus != FOUND) {
183 slotoffset = ufsdirhash_findfree(dp, slotneeded,
184 &slotsize);
185 if (slotoffset >= 0) {
186 slotstatus = COMPACT;
187 enduseful = ufsdirhash_enduseful(dp);
188 if (enduseful < 0)
189 enduseful = dp->i_size;
190 }
191 }
192 /* Look up the component. */
193 numdirpasses = 1;
194 entryoffsetinblock = 0; /* silence compiler warning */
195 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
2b69e610 196 &dp->i_offset, &bp, nameiop == NAMEI_DELETE ? &prevoff : NULL)) {
984263bc
MD
197 case 0:
198 ep = (struct direct *)((char *)bp->b_data +
199 (dp->i_offset & bmask));
200 goto foundentry;
201 case ENOENT:
202 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
203 goto notfound;
204 default:
205 /* Something failed; just do a linear search. */
206 break;
207 }
208 }
209#endif /* UFS_DIRHASH */
210 /*
211 * If there is cached information on a previous search of
212 * this directory, pick up where we last left off.
213 * We cache only lookups as these are the most common
214 * and have the greatest payoff. Caching CREATE has little
215 * benefit as it usually must search the entire directory
216 * to determine that the entry does not exist. Caching the
217 * location of the last DELETE or RENAME has not reduced
218 * profiling time and hence has been removed in the interest
219 * of simplicity.
220 */
2b69e610 221 if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
984263bc
MD
222 dp->i_diroff >= dp->i_size) {
223 entryoffsetinblock = 0;
224 dp->i_offset = 0;
225 numdirpasses = 1;
226 } else {
227 dp->i_offset = dp->i_diroff;
228 if ((entryoffsetinblock = dp->i_offset & bmask) &&
229 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
230 return (error);
231 numdirpasses = 2;
24e51f36 232 gd->gd_nchstats->ncs_2passes++;
984263bc
MD
233 }
234 prevoff = dp->i_offset;
235 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
236 enduseful = 0;
237
238searchloop:
239 while (dp->i_offset < endsearch) {
240 /*
241 * If necessary, get the next directory block.
242 */
243 if ((dp->i_offset & bmask) == 0) {
244 if (bp != NULL)
245 brelse(bp);
246 error =
247 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
248 if (error)
249 return (error);
250 entryoffsetinblock = 0;
251 }
252 /*
253 * If still looking for a slot, and at a DIRBLKSIZE
254 * boundary, have to start looking for free space again.
255 */
256 if (slotstatus == NONE &&
257 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
258 slotoffset = -1;
259 slotfreespace = 0;
260 }
261 /*
262 * Get pointer to next entry.
263 * Full validation checks are slow, so we only check
264 * enough to insure forward progress through the
265 * directory. Complete checks can be run by patching
266 * "dirchk" to be true.
267 */
268 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
269 if (ep->d_reclen == 0 || ep->d_reclen >
270 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
271 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
272 int i;
273
274 ufs_dirbad(dp, dp->i_offset, "mangled entry");
275 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
276 dp->i_offset += i;
277 entryoffsetinblock += i;
278 continue;
279 }
280
281 /*
282 * If an appropriate sized slot has not yet been found,
283 * check to see if one is available. Also accumulate space
284 * in the current block so that we can determine if
285 * compaction is viable.
286 */
287 if (slotstatus != FOUND) {
288 int size = ep->d_reclen;
289
290 if (ep->d_ino != 0)
291 size -= DIRSIZ(OFSFMT(vdp), ep);
292 if (size > 0) {
293 if (size >= slotneeded) {
294 slotstatus = FOUND;
295 slotoffset = dp->i_offset;
296 slotsize = ep->d_reclen;
297 } else if (slotstatus == NONE) {
298 slotfreespace += size;
299 if (slotoffset == -1)
300 slotoffset = dp->i_offset;
301 if (slotfreespace >= slotneeded) {
302 slotstatus = COMPACT;
303 slotsize = dp->i_offset +
304 ep->d_reclen - slotoffset;
305 }
306 }
307 }
308 }
309
310 /*
311 * Check for a name match.
312 */
313 if (ep->d_ino) {
314# if (BYTE_ORDER == LITTLE_ENDIAN)
315 if (OFSFMT(vdp))
316 namlen = ep->d_type;
317 else
318 namlen = ep->d_namlen;
319# else
320 namlen = ep->d_namlen;
321# endif
322 if (namlen == cnp->cn_namelen &&
323 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
324 !bcmp(cnp->cn_nameptr, ep->d_name,
325 (unsigned)namlen)) {
326#ifdef UFS_DIRHASH
327foundentry:
328#endif
329 /*
330 * Save directory entry's inode number and
331 * reclen in ndp->ni_ufs area, and release
332 * directory buffer.
333 */
334 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
335 ep->d_type == DT_WHT) {
336 slotstatus = FOUND;
337 slotoffset = dp->i_offset;
338 slotsize = ep->d_reclen;
339 dp->i_reclen = slotsize;
340 enduseful = dp->i_size;
2b69e610 341 ap->a_cnp->cn_flags |= CNP_ISWHITEOUT;
984263bc
MD
342 numdirpasses--;
343 goto notfound;
344 }
345 dp->i_ino = ep->d_ino;
346 dp->i_reclen = ep->d_reclen;
347 goto found;
348 }
349 }
350 prevoff = dp->i_offset;
351 dp->i_offset += ep->d_reclen;
352 entryoffsetinblock += ep->d_reclen;
353 if (ep->d_ino)
354 enduseful = dp->i_offset;
355 }
356notfound:
357 /*
358 * If we started in the middle of the directory and failed
359 * to find our target, we must check the beginning as well.
360 */
361 if (numdirpasses == 2) {
362 numdirpasses--;
363 dp->i_offset = 0;
364 endsearch = dp->i_diroff;
365 goto searchloop;
366 }
367 if (bp != NULL)
368 brelse(bp);
369 /*
370 * If creating, and at end of pathname and current
371 * directory has not been removed, then can consider
372 * allowing file to be created.
373 */
2b69e610
MD
374 if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME ||
375 (nameiop == NAMEI_DELETE &&
376 (ap->a_cnp->cn_flags & CNP_DOWHITEOUT) &&
377 (ap->a_cnp->cn_flags & CNP_ISWHITEOUT))) &&
fad57d0e 378 dp->i_effnlink != 0) {
984263bc
MD
379 /*
380 * Access for write is interpreted as allowing
381 * creation of files in the directory.
382 */
dadab5e9 383 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
984263bc
MD
384 if (error)
385 return (error);
386 /*
387 * Return an indication of where the new directory
388 * entry should be put. If we didn't find a slot,
389 * then set dp->i_count to 0 indicating
390 * that the new slot belongs at the end of the
391 * directory. If we found a slot, then the new entry
392 * can be put in the range from dp->i_offset to
393 * dp->i_offset + dp->i_count.
394 */
395 if (slotstatus == NONE) {
396 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
397 dp->i_count = 0;
398 enduseful = dp->i_offset;
2b69e610 399 } else if (nameiop == NAMEI_DELETE) {
984263bc
MD
400 dp->i_offset = slotoffset;
401 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
402 dp->i_count = 0;
403 else
404 dp->i_count = dp->i_offset - prevoff;
405 } else {
406 dp->i_offset = slotoffset;
407 dp->i_count = slotsize;
408 if (enduseful < slotoffset + slotsize)
409 enduseful = slotoffset + slotsize;
410 }
411 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
412 dp->i_flag |= IN_CHANGE | IN_UPDATE;
413 /*
414 * We return with the directory locked, so that
415 * the parameters we set up above will still be
416 * valid if we actually decide to do a direnter().
417 * We return ni_vp == NULL to indicate that the entry
418 * does not currently exist; we leave a pointer to
419 * the (locked) directory inode in ndp->ni_dvp.
420 * The pathname buffer is saved so that the name
421 * can be obtained later.
422 *
423 * NB - if the directory is unlocked, then this
424 * information cannot be used.
425 */
984263bc 426 if (!lockparent) {
ca466bae 427 VOP_UNLOCK(vdp, 0);
2b69e610 428 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
429 }
430 return (EJUSTRETURN);
431 }
984263bc
MD
432 return (ENOENT);
433
434found:
435 if (numdirpasses == 2)
24e51f36 436 gd->gd_nchstats->ncs_pass2++;
984263bc
MD
437 /*
438 * Check that directory length properly reflects presence
439 * of this entry.
440 */
441 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
442 ufs_dirbad(dp, dp->i_offset, "i_size too small");
443 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
444 dp->i_flag |= IN_CHANGE | IN_UPDATE;
445 }
446 brelse(bp);
447
448 /*
449 * Found component in pathname.
450 * If the final component of path name, save information
451 * in the cache as to where the entry was found.
452 */
fad57d0e 453 if (nameiop == NAMEI_LOOKUP)
984263bc
MD
454 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
455
456 /*
457 * If deleting, and at end of pathname, return
458 * parameters which can be used to remove file.
459 * If the wantparent flag isn't set, we return only
460 * the directory (in ndp->ni_dvp), otherwise we go
461 * on and lock the inode, being careful with ".".
462 */
fad57d0e 463 if (nameiop == NAMEI_DELETE) {
984263bc
MD
464 /*
465 * Write access to directory required to delete files.
466 */
dadab5e9 467 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
984263bc
MD
468 if (error)
469 return (error);
470 /*
471 * Return pointer to current entry in dp->i_offset,
472 * and distance past previous entry (if there
473 * is a previous entry in this block) in dp->i_count.
474 * Save directory inode pointer in ndp->ni_dvp for dirremove().
475 */
476 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
477 dp->i_count = 0;
478 else
479 dp->i_count = dp->i_offset - prevoff;
480 if (dp->i_number == dp->i_ino) {
597aea93 481 vref(vdp);
984263bc
MD
482 *vpp = vdp;
483 return (0);
484 }
2b69e610 485 if (flags & CNP_ISDOTDOT)
ca466bae 486 VOP_UNLOCK(vdp, 0); /* race to get the inode */
984263bc 487 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
2b69e610 488 if (flags & CNP_ISDOTDOT) {
ca466bae 489 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY) != 0)
2b69e610 490 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
491 }
492 if (error)
493 return (error);
494 /*
495 * If directory is "sticky", then user must own
496 * the directory, or the file in it, else she
497 * may not delete it (unless she's root). This
498 * implements append-only directories.
499 */
500 if ((dp->i_mode & ISVTX) &&
501 cred->cr_uid != 0 &&
502 cred->cr_uid != dp->i_uid &&
503 VTOI(tdp)->i_uid != cred->cr_uid) {
504 vput(tdp);
505 return (EPERM);
506 }
507 *vpp = tdp;
508 if (!lockparent) {
ca466bae 509 VOP_UNLOCK(vdp, 0);
2b69e610 510 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
511 }
512 return (0);
513 }
514
515 /*
516 * If rewriting (RENAME), return the inode and the
517 * information required to rewrite the present directory
518 * Must get inode of directory entry to verify it's a
519 * regular file, or empty directory.
520 */
fad57d0e 521 if (nameiop == NAMEI_RENAME && wantparent) {
dadab5e9 522 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
984263bc
MD
523 return (error);
524 /*
525 * Careful about locking second inode.
526 * This can only occur if the target is ".".
527 */
528 if (dp->i_number == dp->i_ino)
529 return (EISDIR);
2b69e610 530 if (flags & CNP_ISDOTDOT)
ca466bae 531 VOP_UNLOCK(vdp, 0); /* race to get the inode */
984263bc 532 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
2b69e610 533 if (flags & CNP_ISDOTDOT) {
ca466bae 534 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY) != 0)
2b69e610 535 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
536 }
537 if (error)
538 return (error);
539 *vpp = tdp;
984263bc 540 if (!lockparent) {
ca466bae 541 VOP_UNLOCK(vdp, 0);
2b69e610 542 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
543 }
544 return (0);
545 }
546
547 /*
548 * Step through the translation in the name. We do not `vput' the
549 * directory because we may need it again if a symbolic link
550 * is relative to the current directory. Instead we save it
551 * unlocked as "pdp". We must get the target inode before unlocking
552 * the directory to insure that the inode will not be removed
553 * before we get it. We prevent deadlock by always fetching
554 * inodes from the root, moving down the directory tree. Thus
555 * when following backward pointers ".." we must unlock the
556 * parent directory before getting the requested directory.
557 * There is a potential race condition here if both the current
558 * and parent directories are removed before the VFS_VGET for the
559 * inode associated with ".." returns. We hope that this occurs
560 * infrequently since we cannot avoid this race condition without
561 * implementing a sophisticated deadlock detection algorithm.
562 * Note also that this simple deadlock detection scheme will not
f719c866 563 * work if the filesystem has any hard links other than ".."
984263bc
MD
564 * that point backwards in the directory structure.
565 */
566 pdp = vdp;
2b69e610 567 if (flags & CNP_ISDOTDOT) {
ca466bae 568 VOP_UNLOCK(pdp, 0); /* race to get the inode */
2b69e610 569 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc 570 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
ca466bae 571 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
2b69e610 572 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
573 return (error);
574 }
fad57d0e 575 if (lockparent) {
ca466bae 576 if ((error = vn_lock(pdp, LK_EXCLUSIVE)) != 0) {
984263bc
MD
577 vput(tdp);
578 return (error);
579 }
2b69e610 580 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
581 }
582 *vpp = tdp;
583 } else if (dp->i_number == dp->i_ino) {
597aea93 584 vref(vdp); /* we want ourself, ie "." */
984263bc
MD
585 *vpp = vdp;
586 } else {
587 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
588 if (error)
589 return (error);
fad57d0e 590 if (!lockparent) {
ca466bae 591 VOP_UNLOCK(pdp, 0);
2b69e610 592 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
593 }
594 *vpp = tdp;
595 }
984263bc
MD
596 return (0);
597}
598
599void
0973c589 600ufs_dirbad(struct inode *ip, doff_t offset, char *how)
984263bc
MD
601{
602 struct mount *mp;
603
604 mp = ITOV(ip)->v_mount;
605 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
75ffff0d 606 mp->mnt_stat.f_mntfromname, (u_long)ip->i_number, (long)offset, how);
984263bc
MD
607 if ((mp->mnt_flag & MNT_RDONLY) == 0)
608 panic("ufs_dirbad: bad dir");
609}
610
611/*
612 * Do consistency checking on a directory entry:
613 * record length must be multiple of 4
614 * entry must fit in rest of its DIRBLKSIZ block
615 * record must be large enough to contain entry
616 * name is not longer than MAXNAMLEN
617 * name must be as long as advertised, and null terminated
618 */
619int
0973c589 620ufs_dirbadentry(struct vnode *dp, struct direct *ep, int entryoffsetinblock)
984263bc 621{
3ff2135f 622 int i;
984263bc
MD
623 int namlen;
624
625# if (BYTE_ORDER == LITTLE_ENDIAN)
626 if (OFSFMT(dp))
627 namlen = ep->d_type;
628 else
629 namlen = ep->d_namlen;
630# else
631 namlen = ep->d_namlen;
632# endif
633 if ((ep->d_reclen & 0x3) != 0 ||
634 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
635 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
636 /*return (1); */
637 printf("First bad\n");
638 goto bad;
639 }
640 if (ep->d_ino == 0)
641 return (0);
642 for (i = 0; i < namlen; i++)
643 if (ep->d_name[i] == '\0') {
644 /*return (1); */
645 printf("Second bad\n");
646 goto bad;
647 }
648 if (ep->d_name[i])
649 goto bad;
650 return (0);
651bad:
652 return (1);
653}
654
655/*
656 * Construct a new directory entry after a call to namei, using the
657 * parameters that it left in the componentname argument cnp. The
658 * argument ip is the inode to which the new directory entry will refer.
659 */
660void
0973c589
CP
661ufs_makedirentry(struct inode *ip, struct componentname *cnp,
662 struct direct *newdirp)
984263bc
MD
663{
664
984263bc
MD
665 newdirp->d_ino = ip->i_number;
666 newdirp->d_namlen = cnp->cn_namelen;
667 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
668 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
669 newdirp->d_type = IFTODT(ip->i_mode);
670 else {
671 newdirp->d_type = 0;
672# if (BYTE_ORDER == LITTLE_ENDIAN)
673 { u_char tmp = newdirp->d_namlen;
674 newdirp->d_namlen = newdirp->d_type;
675 newdirp->d_type = tmp; }
676# endif
677 }
678}
679
680/*
681 * Write a directory entry after a call to namei, using the parameters
fad57d0e 682 * that it left in the directory inode. The argument dirp is the new directory
984263bc
MD
683 * entry contents. Dvp is a pointer to the directory to be written,
684 * which was left locked by namei. Remaining parameters (dp->i_offset,
685 * dp->i_count) indicate how the space for the new entry is to be obtained.
686 * Non-null bp indicates that a directory is being created (for the
687 * soft dependency code).
688 */
689int
0973c589
CP
690ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
691 struct componentname *cnp, struct buf *newdirbp)
984263bc 692{
dadab5e9
MD
693 struct ucred *cred;
694 struct thread *td = curthread; /* XXX */
984263bc
MD
695 int newentrysize;
696 struct inode *dp;
697 struct buf *bp;
f719c866 698 uint dsize;
984263bc
MD
699 struct direct *ep, *nep;
700 int error, ret, blkoff, loc, spacefree, flags;
701 char *dirbuf;
702
8a8d5d85 703 KKASSERT(td->td_proc); /* YYY use/require cred passed in cnp? */
dadab5e9 704 cred = td->td_proc->p_ucred;
984263bc
MD
705
706 dp = VTOI(dvp);
707 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
708
709 if (dp->i_count == 0) {
710 /*
711 * If dp->i_count is 0, then namei could find no
712 * space in the directory. Here, dp->i_offset will
713 * be on a directory block boundary and we will write the
714 * new entry into a fresh block.
715 */
716 if (dp->i_offset & (DIRBLKSIZ - 1))
717 panic("ufs_direnter: newblk");
d37b14c8 718 vnode_pager_setsize(dvp, dp->i_offset + DIRBLKSIZ);
984263bc
MD
719 flags = B_CLRBUF;
720 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
721 flags |= B_SYNC;
722 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
dadab5e9 723 cred, flags, &bp)) != 0) {
984263bc
MD
724 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
725 bdwrite(newdirbp);
726 return (error);
727 }
728 dp->i_size = dp->i_offset + DIRBLKSIZ;
729 dp->i_flag |= IN_CHANGE | IN_UPDATE;
984263bc
MD
730 dirp->d_reclen = DIRBLKSIZ;
731 blkoff = dp->i_offset &
732 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
733 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
734#ifdef UFS_DIRHASH
735 if (dp->i_dirhash != NULL) {
736 ufsdirhash_newblk(dp, dp->i_offset);
737 ufsdirhash_add(dp, dirp, dp->i_offset);
738 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
739 dp->i_offset);
740 }
741#endif
742 if (DOINGSOFTDEP(dvp)) {
743 /*
744 * Ensure that the entire newly allocated block is a
745 * valid directory so that future growth within the
746 * block does not have to ensure that the block is
747 * written before the inode.
748 */
749 blkoff += DIRBLKSIZ;
750 while (blkoff < bp->b_bcount) {
751 ((struct direct *)
752 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
753 blkoff += DIRBLKSIZ;
754 }
755 softdep_setup_directory_add(bp, dp, dp->i_offset,
756 dirp->d_ino, newdirbp);
757 bdwrite(bp);
758 return (UFS_UPDATE(dvp, 0));
759 }
760 if (DOINGASYNC(dvp)) {
761 bdwrite(bp);
762 return (UFS_UPDATE(dvp, 0));
763 }
62cfda27 764 error = bwrite(bp);
984263bc
MD
765 ret = UFS_UPDATE(dvp, 1);
766 if (error == 0)
767 return (ret);
768 return (error);
769 }
770
771 /*
772 * If dp->i_count is non-zero, then namei found space for the new
773 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
774 * in the directory. To use this space, we may have to compact
775 * the entries located there, by copying them together towards the
776 * beginning of the block, leaving the free space in one usable
777 * chunk at the end.
778 */
779
780 /*
781 * Increase size of directory if entry eats into new space.
782 * This should never push the size past a new multiple of
783 * DIRBLKSIZE.
784 *
785 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
786 */
787 if (dp->i_offset + dp->i_count > dp->i_size)
788 dp->i_size = dp->i_offset + dp->i_count;
789 /*
790 * Get the block containing the space for the new directory entry.
791 */
792 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
793 if (error) {
794 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
795 bdwrite(newdirbp);
796 return (error);
797 }
798 /*
799 * Find space for the new entry. In the simple case, the entry at
800 * offset base will have the space. If it does not, then namei
801 * arranged that compacting the region dp->i_offset to
802 * dp->i_offset + dp->i_count would yield the space.
803 */
804 ep = (struct direct *)dirbuf;
805 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
806 spacefree = ep->d_reclen - dsize;
807 for (loc = ep->d_reclen; loc < dp->i_count; ) {
808 nep = (struct direct *)(dirbuf + loc);
809
810 /* Trim the existing slot (NB: dsize may be zero). */
811 ep->d_reclen = dsize;
812 ep = (struct direct *)((char *)ep + dsize);
813
814 /* Read nep->d_reclen now as the bcopy() may clobber it. */
815 loc += nep->d_reclen;
816 if (nep->d_ino == 0) {
817 /*
818 * A mid-block unused entry. Such entries are
819 * never created by the kernel, but fsck_ffs
820 * can create them (and it doesn't fix them).
821 *
822 * Add up the free space, and initialise the
823 * relocated entry since we don't bcopy it.
824 */
825 spacefree += nep->d_reclen;
826 ep->d_ino = 0;
827 dsize = 0;
828 continue;
829 }
830 dsize = DIRSIZ(OFSFMT(dvp), nep);
831 spacefree += nep->d_reclen - dsize;
832#ifdef UFS_DIRHASH
833 if (dp->i_dirhash != NULL)
834 ufsdirhash_move(dp, nep,
835 dp->i_offset + ((char *)nep - dirbuf),
836 dp->i_offset + ((char *)ep - dirbuf));
837#endif
838 if (DOINGSOFTDEP(dvp))
839 softdep_change_directoryentry_offset(dp, dirbuf,
840 (caddr_t)nep, (caddr_t)ep, dsize);
841 else
842 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
843 }
844 /*
845 * Here, `ep' points to a directory entry containing `dsize' in-use
846 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
847 * then the entry is completely unused (dsize == 0). The value
848 * of ep->d_reclen is always indeterminate.
849 *
850 * Update the pointer fields in the previous entry (if any),
851 * copy in the new entry, and write out the block.
852 */
853 if (ep->d_ino == 0 ||
854 (ep->d_ino == WINO &&
855 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
856 if (spacefree + dsize < newentrysize)
857 panic("ufs_direnter: compact1");
858 dirp->d_reclen = spacefree + dsize;
859 } else {
860 if (spacefree < newentrysize)
861 panic("ufs_direnter: compact2");
862 dirp->d_reclen = spacefree;
863 ep->d_reclen = dsize;
864 ep = (struct direct *)((char *)ep + dsize);
865 }
866#ifdef UFS_DIRHASH
867 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
868 dirp->d_reclen == spacefree))
869 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
870#endif
f719c866 871 bcopy((caddr_t)dirp, (caddr_t)ep, (uint)newentrysize);
984263bc
MD
872#ifdef UFS_DIRHASH
873 if (dp->i_dirhash != NULL)
874 ufsdirhash_checkblock(dp, dirbuf -
875 (dp->i_offset & (DIRBLKSIZ - 1)),
876 dp->i_offset & ~(DIRBLKSIZ - 1));
877#endif
878
879 if (DOINGSOFTDEP(dvp)) {
880 softdep_setup_directory_add(bp, dp,
881 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
882 bdwrite(bp);
883 } else {
884 if (DOINGASYNC(dvp)) {
885 bdwrite(bp);
886 error = 0;
887 } else {
888 error = bowrite(bp);
889 }
890 }
891 dp->i_flag |= IN_CHANGE | IN_UPDATE;
892 /*
893 * If all went well, and the directory can be shortened, proceed
894 * with the truncation. Note that we have to unlock the inode for
895 * the entry that we just entered, as the truncation may need to
896 * lock other inodes which can lead to deadlock if we also hold a
897 * lock on the newly entered node.
898 */
899 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
900 if (tvp != NULL)
ca466bae 901 VOP_UNLOCK(tvp, 0);
984263bc
MD
902#ifdef UFS_DIRHASH
903 if (dp->i_dirhash != NULL)
904 ufsdirhash_dirtrunc(dp, dp->i_endoff);
905#endif
f719c866 906 (void)UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cred, td);
984263bc 907 if (tvp != NULL)
ca466bae 908 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
984263bc
MD
909 }
910 return (error);
911}
912
913/*
914 * Remove a directory entry after a call to namei, using
fad57d0e 915 * the parameters which it left in the directory inode. The entry
984263bc
MD
916 * dp->i_offset contains the offset into the directory of the
917 * entry to be eliminated. The dp->i_count field contains the
918 * size of the previous record in the directory. If this
919 * is 0, the first entry is being deleted, so we need only
920 * zero the inode number to mark the entry as free. If the
921 * entry is not the first in the directory, we must reclaim
922 * the space of the now empty record by adding the record size
923 * to the size of the previous entry.
924 */
925int
0973c589 926ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
984263bc
MD
927{
928 struct inode *dp;
929 struct direct *ep;
930 struct buf *bp;
931 int error;
932
933 dp = VTOI(dvp);
934
2b69e610 935 if (flags & CNP_DOWHITEOUT) {
984263bc
MD
936 /*
937 * Whiteout entry: set d_ino to WINO.
938 */
939 if ((error =
940 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
941 return (error);
942 ep->d_ino = WINO;
943 ep->d_type = DT_WHT;
944 goto out;
945 }
946
947 if ((error = UFS_BLKATOFF(dvp,
948 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
949 return (error);
950#ifdef UFS_DIRHASH
951 /*
952 * Remove the dirhash entry. This is complicated by the fact
953 * that `ep' is the previous entry when dp->i_count != 0.
954 */
955 if (dp->i_dirhash != NULL)
956 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
957 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
958#endif
959 if (dp->i_count == 0) {
960 /*
961 * First entry in block: set d_ino to zero.
962 */
963 ep->d_ino = 0;
964 } else {
965 /*
966 * Collapse new free space into previous entry.
967 */
968 ep->d_reclen += dp->i_reclen;
969 }
970#ifdef UFS_DIRHASH
971 if (dp->i_dirhash != NULL)
972 ufsdirhash_checkblock(dp, (char *)ep -
973 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
974 dp->i_offset & ~(DIRBLKSIZ - 1));
975#endif
976out:
977 if (DOINGSOFTDEP(dvp)) {
978 if (ip) {
979 ip->i_effnlink--;
980 softdep_change_linkcnt(ip);
981 softdep_setup_remove(bp, dp, ip, isrmdir);
982 }
983 if (softdep_slowdown(dvp)) {
62cfda27 984 error = bwrite(bp);
984263bc
MD
985 } else {
986 bdwrite(bp);
987 error = 0;
988 }
989 } else {
990 if (ip) {
991 ip->i_effnlink--;
992 ip->i_nlink--;
993 ip->i_flag |= IN_CHANGE;
994 }
2b69e610 995 if (flags & CNP_DOWHITEOUT)
62cfda27 996 error = bwrite(bp);
984263bc
MD
997 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
998 bdwrite(bp);
999 error = 0;
1000 } else
1001 error = bowrite(bp);
1002 }
1003 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1004 return (error);
1005}
1006
1007/*
1008 * Rewrite an existing directory entry to point at the inode
1009 * supplied. The parameters describing the directory entry are
1010 * set up by a call to namei.
1011 */
1012int
0973c589
CP
1013ufs_dirrewrite(struct inode *dp, struct inode *oip, ino_t newinum, int newtype,
1014 int isrmdir)
984263bc
MD
1015{
1016 struct buf *bp;
1017 struct direct *ep;
1018 struct vnode *vdp = ITOV(dp);
1019 int error;
1020
1021 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1022 if (error)
1023 return (error);
1024 ep->d_ino = newinum;
1025 if (!OFSFMT(vdp))
1026 ep->d_type = newtype;
1027 oip->i_effnlink--;
1028 if (DOINGSOFTDEP(vdp)) {
1029 softdep_change_linkcnt(oip);
1030 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1031 bdwrite(bp);
1032 } else {
1033 oip->i_nlink--;
1034 oip->i_flag |= IN_CHANGE;
1035 if (DOINGASYNC(vdp)) {
1036 bdwrite(bp);
1037 error = 0;
1038 } else {
1039 error = bowrite(bp);
1040 }
1041 }
1042 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1043 return (error);
1044}
1045
1046/*
1047 * Check if a directory is empty or not.
1048 * Inode supplied must be locked.
1049 *
1050 * Using a struct dirtemplate here is not precisely
1051 * what we want, but better than using a struct direct.
1052 *
1053 * NB: does not handle corrupted directories.
1054 */
1055int
0973c589 1056ufs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
984263bc 1057{
3ff2135f 1058 off_t off;
984263bc 1059 struct dirtemplate dbuf;
3ff2135f 1060 struct direct *dp = (struct direct *)&dbuf;
984263bc
MD
1061 int error, count, namlen;
1062#define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1063
1064 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1065 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
dadab5e9 1066 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
984263bc
MD
1067 /*
1068 * Since we read MINDIRSIZ, residual must
1069 * be 0 unless we're at end of file.
1070 */
1071 if (error || count != 0)
1072 return (0);
1073 /* avoid infinite loops */
1074 if (dp->d_reclen == 0)
1075 return (0);
1076 /* skip empty entries */
1077 if (dp->d_ino == 0 || dp->d_ino == WINO)
1078 continue;
1079 /* accept only "." and ".." */
1080# if (BYTE_ORDER == LITTLE_ENDIAN)
1081 if (OFSFMT(ITOV(ip)))
1082 namlen = dp->d_type;
1083 else
1084 namlen = dp->d_namlen;
1085# else
1086 namlen = dp->d_namlen;
1087# endif
1088 if (namlen > 2)
1089 return (0);
1090 if (dp->d_name[0] != '.')
1091 return (0);
1092 /*
1093 * At this point namlen must be 1 or 2.
1094 * 1 implies ".", 2 implies ".." if second
1095 * char is also "."
1096 */
1097 if (namlen == 1 && dp->d_ino == ip->i_number)
1098 continue;
1099 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1100 continue;
1101 return (0);
1102 }
1103 return (1);
1104}
1105
1106/*
1107 * Check if source directory is in the path of the target directory.
1108 * Target is supplied locked, source is unlocked.
1109 * The target is always vput before returning.
1110 */
1111int
0973c589 1112ufs_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
984263bc
MD
1113{
1114 struct vnode *vp;
1115 int error, rootino, namlen;
1116 struct dirtemplate dirbuf;
1117
1118 vp = ITOV(target);
1119 if (target->i_number == source->i_number) {
1120 error = EEXIST;
1121 goto out;
1122 }
1123 rootino = ROOTINO;
1124 error = 0;
1125 if (target->i_number == rootino)
1126 goto out;
1127
1128 for (;;) {
1129 if (vp->v_type != VDIR) {
1130 error = ENOTDIR;
1131 break;
1132 }
1133 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1134 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
dadab5e9 1135 IO_NODELOCKED, cred, (int *)0, NULL);
984263bc
MD
1136 if (error != 0)
1137 break;
1138# if (BYTE_ORDER == LITTLE_ENDIAN)
1139 if (OFSFMT(vp))
1140 namlen = dirbuf.dotdot_type;
1141 else
1142 namlen = dirbuf.dotdot_namlen;
1143# else
1144 namlen = dirbuf.dotdot_namlen;
1145# endif
1146 if (namlen != 2 ||
1147 dirbuf.dotdot_name[0] != '.' ||
1148 dirbuf.dotdot_name[1] != '.') {
1149 error = ENOTDIR;
1150 break;
1151 }
1152 if (dirbuf.dotdot_ino == source->i_number) {
1153 error = EINVAL;
1154 break;
1155 }
1156 if (dirbuf.dotdot_ino == rootino)
1157 break;
1158 vput(vp);
1159 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1160 if (error) {
1161 vp = NULL;
1162 break;
1163 }
1164 }
1165
1166out:
1167 if (error == ENOTDIR)
1168 printf("checkpath: .. not a directory\n");
1169 if (vp != NULL)
1170 vput(vp);
1171 return (error);
1172}