Rename all the functions and structures for the old VOP namespace API
[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 $
e62afb5f 40 * $DragonFly: src/sys/vfs/ufs/ufs_lookup.c,v 1.18 2005/09/14 01:13:48 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;
dadab5e9 140 struct thread *td = cnp->cn_td;
24e51f36 141 globaldata_t gd = mycpu;
984263bc
MD
142
143 bp = NULL;
144 slotoffset = -1;
2b69e610 145 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
146/*
147 * XXX there was a soft-update diff about this I couldn't merge.
148 * I think this was the equiv.
149 */
150 *vpp = NULL;
151
152 vdp = ap->a_dvp;
153 dp = VTOI(vdp);
2b69e610
MD
154 lockparent = flags & CNP_LOCKPARENT;
155 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
984263bc
MD
156
157 /*
158 * We now have a segment name to search for, and a directory to search.
159 *
160 * Suppress search for slots unless creating
161 * file and at end of pathname, in which case
162 * we watch for a place to put the new file in
163 * case it doesn't already exist.
164 */
165 slotstatus = FOUND;
166 slotfreespace = slotsize = slotneeded = 0;
fad57d0e 167 if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
984263bc
MD
168 slotstatus = NONE;
169 slotneeded = DIRECTSIZ(cnp->cn_namelen);
170 }
171 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
172
173#ifdef UFS_DIRHASH
174 /*
175 * Use dirhash for fast operations on large directories. The logic
176 * to determine whether to hash the directory is contained within
177 * ufsdirhash_build(); a zero return means that it decided to hash
178 * this directory and it successfully built up the hash table.
179 */
180 if (ufsdirhash_build(dp) == 0) {
181 /* Look for a free slot if needed. */
182 enduseful = dp->i_size;
183 if (slotstatus != FOUND) {
184 slotoffset = ufsdirhash_findfree(dp, slotneeded,
185 &slotsize);
186 if (slotoffset >= 0) {
187 slotstatus = COMPACT;
188 enduseful = ufsdirhash_enduseful(dp);
189 if (enduseful < 0)
190 enduseful = dp->i_size;
191 }
192 }
193 /* Look up the component. */
194 numdirpasses = 1;
195 entryoffsetinblock = 0; /* silence compiler warning */
196 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
2b69e610 197 &dp->i_offset, &bp, nameiop == NAMEI_DELETE ? &prevoff : NULL)) {
984263bc
MD
198 case 0:
199 ep = (struct direct *)((char *)bp->b_data +
200 (dp->i_offset & bmask));
201 goto foundentry;
202 case ENOENT:
203 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
204 goto notfound;
205 default:
206 /* Something failed; just do a linear search. */
207 break;
208 }
209 }
210#endif /* UFS_DIRHASH */
211 /*
212 * If there is cached information on a previous search of
213 * this directory, pick up where we last left off.
214 * We cache only lookups as these are the most common
215 * and have the greatest payoff. Caching CREATE has little
216 * benefit as it usually must search the entire directory
217 * to determine that the entry does not exist. Caching the
218 * location of the last DELETE or RENAME has not reduced
219 * profiling time and hence has been removed in the interest
220 * of simplicity.
221 */
2b69e610 222 if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
984263bc
MD
223 dp->i_diroff >= dp->i_size) {
224 entryoffsetinblock = 0;
225 dp->i_offset = 0;
226 numdirpasses = 1;
227 } else {
228 dp->i_offset = dp->i_diroff;
229 if ((entryoffsetinblock = dp->i_offset & bmask) &&
230 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
231 return (error);
232 numdirpasses = 2;
24e51f36 233 gd->gd_nchstats->ncs_2passes++;
984263bc
MD
234 }
235 prevoff = dp->i_offset;
236 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
237 enduseful = 0;
238
239searchloop:
240 while (dp->i_offset < endsearch) {
241 /*
242 * If necessary, get the next directory block.
243 */
244 if ((dp->i_offset & bmask) == 0) {
245 if (bp != NULL)
246 brelse(bp);
247 error =
248 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
249 if (error)
250 return (error);
251 entryoffsetinblock = 0;
252 }
253 /*
254 * If still looking for a slot, and at a DIRBLKSIZE
255 * boundary, have to start looking for free space again.
256 */
257 if (slotstatus == NONE &&
258 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
259 slotoffset = -1;
260 slotfreespace = 0;
261 }
262 /*
263 * Get pointer to next entry.
264 * Full validation checks are slow, so we only check
265 * enough to insure forward progress through the
266 * directory. Complete checks can be run by patching
267 * "dirchk" to be true.
268 */
269 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
270 if (ep->d_reclen == 0 || ep->d_reclen >
271 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
272 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
273 int i;
274
275 ufs_dirbad(dp, dp->i_offset, "mangled entry");
276 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
277 dp->i_offset += i;
278 entryoffsetinblock += i;
279 continue;
280 }
281
282 /*
283 * If an appropriate sized slot has not yet been found,
284 * check to see if one is available. Also accumulate space
285 * in the current block so that we can determine if
286 * compaction is viable.
287 */
288 if (slotstatus != FOUND) {
289 int size = ep->d_reclen;
290
291 if (ep->d_ino != 0)
292 size -= DIRSIZ(OFSFMT(vdp), ep);
293 if (size > 0) {
294 if (size >= slotneeded) {
295 slotstatus = FOUND;
296 slotoffset = dp->i_offset;
297 slotsize = ep->d_reclen;
298 } else if (slotstatus == NONE) {
299 slotfreespace += size;
300 if (slotoffset == -1)
301 slotoffset = dp->i_offset;
302 if (slotfreespace >= slotneeded) {
303 slotstatus = COMPACT;
304 slotsize = dp->i_offset +
305 ep->d_reclen - slotoffset;
306 }
307 }
308 }
309 }
310
311 /*
312 * Check for a name match.
313 */
314 if (ep->d_ino) {
315# if (BYTE_ORDER == LITTLE_ENDIAN)
316 if (OFSFMT(vdp))
317 namlen = ep->d_type;
318 else
319 namlen = ep->d_namlen;
320# else
321 namlen = ep->d_namlen;
322# endif
323 if (namlen == cnp->cn_namelen &&
324 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
325 !bcmp(cnp->cn_nameptr, ep->d_name,
326 (unsigned)namlen)) {
327#ifdef UFS_DIRHASH
328foundentry:
329#endif
330 /*
331 * Save directory entry's inode number and
332 * reclen in ndp->ni_ufs area, and release
333 * directory buffer.
334 */
335 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
336 ep->d_type == DT_WHT) {
337 slotstatus = FOUND;
338 slotoffset = dp->i_offset;
339 slotsize = ep->d_reclen;
340 dp->i_reclen = slotsize;
341 enduseful = dp->i_size;
2b69e610 342 ap->a_cnp->cn_flags |= CNP_ISWHITEOUT;
984263bc
MD
343 numdirpasses--;
344 goto notfound;
345 }
346 dp->i_ino = ep->d_ino;
347 dp->i_reclen = ep->d_reclen;
348 goto found;
349 }
350 }
351 prevoff = dp->i_offset;
352 dp->i_offset += ep->d_reclen;
353 entryoffsetinblock += ep->d_reclen;
354 if (ep->d_ino)
355 enduseful = dp->i_offset;
356 }
357notfound:
358 /*
359 * If we started in the middle of the directory and failed
360 * to find our target, we must check the beginning as well.
361 */
362 if (numdirpasses == 2) {
363 numdirpasses--;
364 dp->i_offset = 0;
365 endsearch = dp->i_diroff;
366 goto searchloop;
367 }
368 if (bp != NULL)
369 brelse(bp);
370 /*
371 * If creating, and at end of pathname and current
372 * directory has not been removed, then can consider
373 * allowing file to be created.
374 */
2b69e610
MD
375 if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME ||
376 (nameiop == NAMEI_DELETE &&
377 (ap->a_cnp->cn_flags & CNP_DOWHITEOUT) &&
378 (ap->a_cnp->cn_flags & CNP_ISWHITEOUT))) &&
fad57d0e 379 dp->i_effnlink != 0) {
984263bc
MD
380 /*
381 * Access for write is interpreted as allowing
382 * creation of files in the directory.
383 */
dadab5e9 384 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
984263bc
MD
385 if (error)
386 return (error);
387 /*
388 * Return an indication of where the new directory
389 * entry should be put. If we didn't find a slot,
390 * then set dp->i_count to 0 indicating
391 * that the new slot belongs at the end of the
392 * directory. If we found a slot, then the new entry
393 * can be put in the range from dp->i_offset to
394 * dp->i_offset + dp->i_count.
395 */
396 if (slotstatus == NONE) {
397 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
398 dp->i_count = 0;
399 enduseful = dp->i_offset;
2b69e610 400 } else if (nameiop == NAMEI_DELETE) {
984263bc
MD
401 dp->i_offset = slotoffset;
402 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
403 dp->i_count = 0;
404 else
405 dp->i_count = dp->i_offset - prevoff;
406 } else {
407 dp->i_offset = slotoffset;
408 dp->i_count = slotsize;
409 if (enduseful < slotoffset + slotsize)
410 enduseful = slotoffset + slotsize;
411 }
412 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
413 dp->i_flag |= IN_CHANGE | IN_UPDATE;
414 /*
415 * We return with the directory locked, so that
416 * the parameters we set up above will still be
417 * valid if we actually decide to do a direnter().
418 * We return ni_vp == NULL to indicate that the entry
419 * does not currently exist; we leave a pointer to
420 * the (locked) directory inode in ndp->ni_dvp.
421 * The pathname buffer is saved so that the name
422 * can be obtained later.
423 *
424 * NB - if the directory is unlocked, then this
425 * information cannot be used.
426 */
984263bc 427 if (!lockparent) {
5fd012e0 428 VOP_UNLOCK(vdp, 0, td);
2b69e610 429 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
430 }
431 return (EJUSTRETURN);
432 }
984263bc
MD
433 return (ENOENT);
434
435found:
436 if (numdirpasses == 2)
24e51f36 437 gd->gd_nchstats->ncs_pass2++;
984263bc
MD
438 /*
439 * Check that directory length properly reflects presence
440 * of this entry.
441 */
442 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
443 ufs_dirbad(dp, dp->i_offset, "i_size too small");
444 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
445 dp->i_flag |= IN_CHANGE | IN_UPDATE;
446 }
447 brelse(bp);
448
449 /*
450 * Found component in pathname.
451 * If the final component of path name, save information
452 * in the cache as to where the entry was found.
453 */
fad57d0e 454 if (nameiop == NAMEI_LOOKUP)
984263bc
MD
455 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
456
457 /*
458 * If deleting, and at end of pathname, return
459 * parameters which can be used to remove file.
460 * If the wantparent flag isn't set, we return only
461 * the directory (in ndp->ni_dvp), otherwise we go
462 * on and lock the inode, being careful with ".".
463 */
fad57d0e 464 if (nameiop == NAMEI_DELETE) {
984263bc
MD
465 /*
466 * Write access to directory required to delete files.
467 */
dadab5e9 468 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
984263bc
MD
469 if (error)
470 return (error);
471 /*
472 * Return pointer to current entry in dp->i_offset,
473 * and distance past previous entry (if there
474 * is a previous entry in this block) in dp->i_count.
475 * Save directory inode pointer in ndp->ni_dvp for dirremove().
476 */
477 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
478 dp->i_count = 0;
479 else
480 dp->i_count = dp->i_offset - prevoff;
481 if (dp->i_number == dp->i_ino) {
597aea93 482 vref(vdp);
984263bc
MD
483 *vpp = vdp;
484 return (0);
485 }
2b69e610 486 if (flags & CNP_ISDOTDOT)
5fd012e0 487 VOP_UNLOCK(vdp, 0, td);/* race to get the inode */
984263bc 488 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
2b69e610 489 if (flags & CNP_ISDOTDOT) {
5fd012e0 490 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, td) != 0)
2b69e610 491 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
492 }
493 if (error)
494 return (error);
495 /*
496 * If directory is "sticky", then user must own
497 * the directory, or the file in it, else she
498 * may not delete it (unless she's root). This
499 * implements append-only directories.
500 */
501 if ((dp->i_mode & ISVTX) &&
502 cred->cr_uid != 0 &&
503 cred->cr_uid != dp->i_uid &&
504 VTOI(tdp)->i_uid != cred->cr_uid) {
505 vput(tdp);
506 return (EPERM);
507 }
508 *vpp = tdp;
509 if (!lockparent) {
5fd012e0 510 VOP_UNLOCK(vdp, 0, td);
2b69e610 511 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
512 }
513 return (0);
514 }
515
516 /*
517 * If rewriting (RENAME), return the inode and the
518 * information required to rewrite the present directory
519 * Must get inode of directory entry to verify it's a
520 * regular file, or empty directory.
521 */
fad57d0e 522 if (nameiop == NAMEI_RENAME && wantparent) {
dadab5e9 523 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
984263bc
MD
524 return (error);
525 /*
526 * Careful about locking second inode.
527 * This can only occur if the target is ".".
528 */
529 if (dp->i_number == dp->i_ino)
530 return (EISDIR);
2b69e610 531 if (flags & CNP_ISDOTDOT)
5fd012e0 532 VOP_UNLOCK(vdp, 0, td); /* race to get the inode */
984263bc 533 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
2b69e610 534 if (flags & CNP_ISDOTDOT) {
5fd012e0 535 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, td) != 0)
2b69e610 536 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
537 }
538 if (error)
539 return (error);
540 *vpp = tdp;
984263bc 541 if (!lockparent) {
5fd012e0 542 VOP_UNLOCK(vdp, 0, td);
2b69e610 543 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
544 }
545 return (0);
546 }
547
548 /*
549 * Step through the translation in the name. We do not `vput' the
550 * directory because we may need it again if a symbolic link
551 * is relative to the current directory. Instead we save it
552 * unlocked as "pdp". We must get the target inode before unlocking
553 * the directory to insure that the inode will not be removed
554 * before we get it. We prevent deadlock by always fetching
555 * inodes from the root, moving down the directory tree. Thus
556 * when following backward pointers ".." we must unlock the
557 * parent directory before getting the requested directory.
558 * There is a potential race condition here if both the current
559 * and parent directories are removed before the VFS_VGET for the
560 * inode associated with ".." returns. We hope that this occurs
561 * infrequently since we cannot avoid this race condition without
562 * implementing a sophisticated deadlock detection algorithm.
563 * Note also that this simple deadlock detection scheme will not
f719c866 564 * work if the filesystem has any hard links other than ".."
984263bc
MD
565 * that point backwards in the directory structure.
566 */
567 pdp = vdp;
2b69e610 568 if (flags & CNP_ISDOTDOT) {
5fd012e0 569 VOP_UNLOCK(pdp, 0, td); /* race to get the inode */
2b69e610 570 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc 571 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
5fd012e0 572 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, td) == 0)
2b69e610 573 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
574 return (error);
575 }
fad57d0e 576 if (lockparent) {
5fd012e0 577 if ((error = vn_lock(pdp, LK_EXCLUSIVE, td)) != 0) {
984263bc
MD
578 vput(tdp);
579 return (error);
580 }
2b69e610 581 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
582 }
583 *vpp = tdp;
584 } else if (dp->i_number == dp->i_ino) {
597aea93 585 vref(vdp); /* we want ourself, ie "." */
984263bc
MD
586 *vpp = vdp;
587 } else {
588 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
589 if (error)
590 return (error);
fad57d0e 591 if (!lockparent) {
5fd012e0 592 VOP_UNLOCK(pdp, 0, td);
2b69e610 593 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
594 }
595 *vpp = tdp;
596 }
984263bc
MD
597 return (0);
598}
599
600void
0973c589 601ufs_dirbad(struct inode *ip, doff_t offset, char *how)
984263bc
MD
602{
603 struct mount *mp;
604
605 mp = ITOV(ip)->v_mount;
606 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
75ffff0d 607 mp->mnt_stat.f_mntfromname, (u_long)ip->i_number, (long)offset, how);
984263bc
MD
608 if ((mp->mnt_flag & MNT_RDONLY) == 0)
609 panic("ufs_dirbad: bad dir");
610}
611
612/*
613 * Do consistency checking on a directory entry:
614 * record length must be multiple of 4
615 * entry must fit in rest of its DIRBLKSIZ block
616 * record must be large enough to contain entry
617 * name is not longer than MAXNAMLEN
618 * name must be as long as advertised, and null terminated
619 */
620int
0973c589 621ufs_dirbadentry(struct vnode *dp, struct direct *ep, int entryoffsetinblock)
984263bc 622{
3ff2135f 623 int i;
984263bc
MD
624 int namlen;
625
626# if (BYTE_ORDER == LITTLE_ENDIAN)
627 if (OFSFMT(dp))
628 namlen = ep->d_type;
629 else
630 namlen = ep->d_namlen;
631# else
632 namlen = ep->d_namlen;
633# endif
634 if ((ep->d_reclen & 0x3) != 0 ||
635 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
636 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
637 /*return (1); */
638 printf("First bad\n");
639 goto bad;
640 }
641 if (ep->d_ino == 0)
642 return (0);
643 for (i = 0; i < namlen; i++)
644 if (ep->d_name[i] == '\0') {
645 /*return (1); */
646 printf("Second bad\n");
647 goto bad;
648 }
649 if (ep->d_name[i])
650 goto bad;
651 return (0);
652bad:
653 return (1);
654}
655
656/*
657 * Construct a new directory entry after a call to namei, using the
658 * parameters that it left in the componentname argument cnp. The
659 * argument ip is the inode to which the new directory entry will refer.
660 */
661void
0973c589
CP
662ufs_makedirentry(struct inode *ip, struct componentname *cnp,
663 struct direct *newdirp)
984263bc
MD
664{
665
984263bc
MD
666 newdirp->d_ino = ip->i_number;
667 newdirp->d_namlen = cnp->cn_namelen;
668 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
669 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
670 newdirp->d_type = IFTODT(ip->i_mode);
671 else {
672 newdirp->d_type = 0;
673# if (BYTE_ORDER == LITTLE_ENDIAN)
674 { u_char tmp = newdirp->d_namlen;
675 newdirp->d_namlen = newdirp->d_type;
676 newdirp->d_type = tmp; }
677# endif
678 }
679}
680
681/*
682 * Write a directory entry after a call to namei, using the parameters
fad57d0e 683 * that it left in the directory inode. The argument dirp is the new directory
984263bc
MD
684 * entry contents. Dvp is a pointer to the directory to be written,
685 * which was left locked by namei. Remaining parameters (dp->i_offset,
686 * dp->i_count) indicate how the space for the new entry is to be obtained.
687 * Non-null bp indicates that a directory is being created (for the
688 * soft dependency code).
689 */
690int
0973c589
CP
691ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
692 struct componentname *cnp, struct buf *newdirbp)
984263bc 693{
dadab5e9
MD
694 struct ucred *cred;
695 struct thread *td = curthread; /* XXX */
984263bc
MD
696 int newentrysize;
697 struct inode *dp;
698 struct buf *bp;
f719c866 699 uint dsize;
984263bc
MD
700 struct direct *ep, *nep;
701 int error, ret, blkoff, loc, spacefree, flags;
702 char *dirbuf;
703
8a8d5d85 704 KKASSERT(td->td_proc); /* YYY use/require cred passed in cnp? */
dadab5e9 705 cred = td->td_proc->p_ucred;
984263bc
MD
706
707 dp = VTOI(dvp);
708 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
709
710 if (dp->i_count == 0) {
711 /*
712 * If dp->i_count is 0, then namei could find no
713 * space in the directory. Here, dp->i_offset will
714 * be on a directory block boundary and we will write the
715 * new entry into a fresh block.
716 */
717 if (dp->i_offset & (DIRBLKSIZ - 1))
718 panic("ufs_direnter: newblk");
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;
730 vnode_pager_setsize(dvp, (u_long)dp->i_size);
731 dirp->d_reclen = DIRBLKSIZ;
732 blkoff = dp->i_offset &
733 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
734 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
735#ifdef UFS_DIRHASH
736 if (dp->i_dirhash != NULL) {
737 ufsdirhash_newblk(dp, dp->i_offset);
738 ufsdirhash_add(dp, dirp, dp->i_offset);
739 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
740 dp->i_offset);
741 }
742#endif
743 if (DOINGSOFTDEP(dvp)) {
744 /*
745 * Ensure that the entire newly allocated block is a
746 * valid directory so that future growth within the
747 * block does not have to ensure that the block is
748 * written before the inode.
749 */
750 blkoff += DIRBLKSIZ;
751 while (blkoff < bp->b_bcount) {
752 ((struct direct *)
753 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
754 blkoff += DIRBLKSIZ;
755 }
756 softdep_setup_directory_add(bp, dp, dp->i_offset,
757 dirp->d_ino, newdirbp);
758 bdwrite(bp);
759 return (UFS_UPDATE(dvp, 0));
760 }
761 if (DOINGASYNC(dvp)) {
762 bdwrite(bp);
763 return (UFS_UPDATE(dvp, 0));
764 }
765 error = VOP_BWRITE(bp->b_vp, bp);
766 ret = UFS_UPDATE(dvp, 1);
767 if (error == 0)
768 return (ret);
769 return (error);
770 }
771
772 /*
773 * If dp->i_count is non-zero, then namei found space for the new
774 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
775 * in the directory. To use this space, we may have to compact
776 * the entries located there, by copying them together towards the
777 * beginning of the block, leaving the free space in one usable
778 * chunk at the end.
779 */
780
781 /*
782 * Increase size of directory if entry eats into new space.
783 * This should never push the size past a new multiple of
784 * DIRBLKSIZE.
785 *
786 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
787 */
788 if (dp->i_offset + dp->i_count > dp->i_size)
789 dp->i_size = dp->i_offset + dp->i_count;
790 /*
791 * Get the block containing the space for the new directory entry.
792 */
793 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
794 if (error) {
795 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
796 bdwrite(newdirbp);
797 return (error);
798 }
799 /*
800 * Find space for the new entry. In the simple case, the entry at
801 * offset base will have the space. If it does not, then namei
802 * arranged that compacting the region dp->i_offset to
803 * dp->i_offset + dp->i_count would yield the space.
804 */
805 ep = (struct direct *)dirbuf;
806 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
807 spacefree = ep->d_reclen - dsize;
808 for (loc = ep->d_reclen; loc < dp->i_count; ) {
809 nep = (struct direct *)(dirbuf + loc);
810
811 /* Trim the existing slot (NB: dsize may be zero). */
812 ep->d_reclen = dsize;
813 ep = (struct direct *)((char *)ep + dsize);
814
815 /* Read nep->d_reclen now as the bcopy() may clobber it. */
816 loc += nep->d_reclen;
817 if (nep->d_ino == 0) {
818 /*
819 * A mid-block unused entry. Such entries are
820 * never created by the kernel, but fsck_ffs
821 * can create them (and it doesn't fix them).
822 *
823 * Add up the free space, and initialise the
824 * relocated entry since we don't bcopy it.
825 */
826 spacefree += nep->d_reclen;
827 ep->d_ino = 0;
828 dsize = 0;
829 continue;
830 }
831 dsize = DIRSIZ(OFSFMT(dvp), nep);
832 spacefree += nep->d_reclen - dsize;
833#ifdef UFS_DIRHASH
834 if (dp->i_dirhash != NULL)
835 ufsdirhash_move(dp, nep,
836 dp->i_offset + ((char *)nep - dirbuf),
837 dp->i_offset + ((char *)ep - dirbuf));
838#endif
839 if (DOINGSOFTDEP(dvp))
840 softdep_change_directoryentry_offset(dp, dirbuf,
841 (caddr_t)nep, (caddr_t)ep, dsize);
842 else
843 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
844 }
845 /*
846 * Here, `ep' points to a directory entry containing `dsize' in-use
847 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
848 * then the entry is completely unused (dsize == 0). The value
849 * of ep->d_reclen is always indeterminate.
850 *
851 * Update the pointer fields in the previous entry (if any),
852 * copy in the new entry, and write out the block.
853 */
854 if (ep->d_ino == 0 ||
855 (ep->d_ino == WINO &&
856 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
857 if (spacefree + dsize < newentrysize)
858 panic("ufs_direnter: compact1");
859 dirp->d_reclen = spacefree + dsize;
860 } else {
861 if (spacefree < newentrysize)
862 panic("ufs_direnter: compact2");
863 dirp->d_reclen = spacefree;
864 ep->d_reclen = dsize;
865 ep = (struct direct *)((char *)ep + dsize);
866 }
867#ifdef UFS_DIRHASH
868 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
869 dirp->d_reclen == spacefree))
870 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
871#endif
f719c866 872 bcopy((caddr_t)dirp, (caddr_t)ep, (uint)newentrysize);
984263bc
MD
873#ifdef UFS_DIRHASH
874 if (dp->i_dirhash != NULL)
875 ufsdirhash_checkblock(dp, dirbuf -
876 (dp->i_offset & (DIRBLKSIZ - 1)),
877 dp->i_offset & ~(DIRBLKSIZ - 1));
878#endif
879
880 if (DOINGSOFTDEP(dvp)) {
881 softdep_setup_directory_add(bp, dp,
882 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
883 bdwrite(bp);
884 } else {
885 if (DOINGASYNC(dvp)) {
886 bdwrite(bp);
887 error = 0;
888 } else {
889 error = bowrite(bp);
890 }
891 }
892 dp->i_flag |= IN_CHANGE | IN_UPDATE;
893 /*
894 * If all went well, and the directory can be shortened, proceed
895 * with the truncation. Note that we have to unlock the inode for
896 * the entry that we just entered, as the truncation may need to
897 * lock other inodes which can lead to deadlock if we also hold a
898 * lock on the newly entered node.
899 */
900 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
901 if (tvp != NULL)
5fd012e0 902 VOP_UNLOCK(tvp, 0, td);
984263bc
MD
903#ifdef UFS_DIRHASH
904 if (dp->i_dirhash != NULL)
905 ufsdirhash_dirtrunc(dp, dp->i_endoff);
906#endif
f719c866 907 (void)UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cred, td);
984263bc 908 if (tvp != NULL)
5fd012e0 909 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
910 }
911 return (error);
912}
913
914/*
915 * Remove a directory entry after a call to namei, using
fad57d0e 916 * the parameters which it left in the directory inode. The entry
984263bc
MD
917 * dp->i_offset contains the offset into the directory of the
918 * entry to be eliminated. The dp->i_count field contains the
919 * size of the previous record in the directory. If this
920 * is 0, the first entry is being deleted, so we need only
921 * zero the inode number to mark the entry as free. If the
922 * entry is not the first in the directory, we must reclaim
923 * the space of the now empty record by adding the record size
924 * to the size of the previous entry.
925 */
926int
0973c589 927ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
984263bc
MD
928{
929 struct inode *dp;
930 struct direct *ep;
931 struct buf *bp;
932 int error;
933
934 dp = VTOI(dvp);
935
2b69e610 936 if (flags & CNP_DOWHITEOUT) {
984263bc
MD
937 /*
938 * Whiteout entry: set d_ino to WINO.
939 */
940 if ((error =
941 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
942 return (error);
943 ep->d_ino = WINO;
944 ep->d_type = DT_WHT;
945 goto out;
946 }
947
948 if ((error = UFS_BLKATOFF(dvp,
949 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
950 return (error);
951#ifdef UFS_DIRHASH
952 /*
953 * Remove the dirhash entry. This is complicated by the fact
954 * that `ep' is the previous entry when dp->i_count != 0.
955 */
956 if (dp->i_dirhash != NULL)
957 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
958 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
959#endif
960 if (dp->i_count == 0) {
961 /*
962 * First entry in block: set d_ino to zero.
963 */
964 ep->d_ino = 0;
965 } else {
966 /*
967 * Collapse new free space into previous entry.
968 */
969 ep->d_reclen += dp->i_reclen;
970 }
971#ifdef UFS_DIRHASH
972 if (dp->i_dirhash != NULL)
973 ufsdirhash_checkblock(dp, (char *)ep -
974 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
975 dp->i_offset & ~(DIRBLKSIZ - 1));
976#endif
977out:
978 if (DOINGSOFTDEP(dvp)) {
979 if (ip) {
980 ip->i_effnlink--;
981 softdep_change_linkcnt(ip);
982 softdep_setup_remove(bp, dp, ip, isrmdir);
983 }
984 if (softdep_slowdown(dvp)) {
985 error = VOP_BWRITE(bp->b_vp, bp);
986 } else {
987 bdwrite(bp);
988 error = 0;
989 }
990 } else {
991 if (ip) {
992 ip->i_effnlink--;
993 ip->i_nlink--;
994 ip->i_flag |= IN_CHANGE;
995 }
2b69e610 996 if (flags & CNP_DOWHITEOUT)
984263bc
MD
997 error = VOP_BWRITE(bp->b_vp, bp);
998 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
999 bdwrite(bp);
1000 error = 0;
1001 } else
1002 error = bowrite(bp);
1003 }
1004 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1005 return (error);
1006}
1007
1008/*
1009 * Rewrite an existing directory entry to point at the inode
1010 * supplied. The parameters describing the directory entry are
1011 * set up by a call to namei.
1012 */
1013int
0973c589
CP
1014ufs_dirrewrite(struct inode *dp, struct inode *oip, ino_t newinum, int newtype,
1015 int isrmdir)
984263bc
MD
1016{
1017 struct buf *bp;
1018 struct direct *ep;
1019 struct vnode *vdp = ITOV(dp);
1020 int error;
1021
1022 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1023 if (error)
1024 return (error);
1025 ep->d_ino = newinum;
1026 if (!OFSFMT(vdp))
1027 ep->d_type = newtype;
1028 oip->i_effnlink--;
1029 if (DOINGSOFTDEP(vdp)) {
1030 softdep_change_linkcnt(oip);
1031 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1032 bdwrite(bp);
1033 } else {
1034 oip->i_nlink--;
1035 oip->i_flag |= IN_CHANGE;
1036 if (DOINGASYNC(vdp)) {
1037 bdwrite(bp);
1038 error = 0;
1039 } else {
1040 error = bowrite(bp);
1041 }
1042 }
1043 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1044 return (error);
1045}
1046
1047/*
1048 * Check if a directory is empty or not.
1049 * Inode supplied must be locked.
1050 *
1051 * Using a struct dirtemplate here is not precisely
1052 * what we want, but better than using a struct direct.
1053 *
1054 * NB: does not handle corrupted directories.
1055 */
1056int
0973c589 1057ufs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
984263bc 1058{
3ff2135f 1059 off_t off;
984263bc 1060 struct dirtemplate dbuf;
3ff2135f 1061 struct direct *dp = (struct direct *)&dbuf;
984263bc
MD
1062 int error, count, namlen;
1063#define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1064
1065 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1066 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
dadab5e9 1067 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
984263bc
MD
1068 /*
1069 * Since we read MINDIRSIZ, residual must
1070 * be 0 unless we're at end of file.
1071 */
1072 if (error || count != 0)
1073 return (0);
1074 /* avoid infinite loops */
1075 if (dp->d_reclen == 0)
1076 return (0);
1077 /* skip empty entries */
1078 if (dp->d_ino == 0 || dp->d_ino == WINO)
1079 continue;
1080 /* accept only "." and ".." */
1081# if (BYTE_ORDER == LITTLE_ENDIAN)
1082 if (OFSFMT(ITOV(ip)))
1083 namlen = dp->d_type;
1084 else
1085 namlen = dp->d_namlen;
1086# else
1087 namlen = dp->d_namlen;
1088# endif
1089 if (namlen > 2)
1090 return (0);
1091 if (dp->d_name[0] != '.')
1092 return (0);
1093 /*
1094 * At this point namlen must be 1 or 2.
1095 * 1 implies ".", 2 implies ".." if second
1096 * char is also "."
1097 */
1098 if (namlen == 1 && dp->d_ino == ip->i_number)
1099 continue;
1100 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1101 continue;
1102 return (0);
1103 }
1104 return (1);
1105}
1106
1107/*
1108 * Check if source directory is in the path of the target directory.
1109 * Target is supplied locked, source is unlocked.
1110 * The target is always vput before returning.
1111 */
1112int
0973c589 1113ufs_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
984263bc
MD
1114{
1115 struct vnode *vp;
1116 int error, rootino, namlen;
1117 struct dirtemplate dirbuf;
1118
1119 vp = ITOV(target);
1120 if (target->i_number == source->i_number) {
1121 error = EEXIST;
1122 goto out;
1123 }
1124 rootino = ROOTINO;
1125 error = 0;
1126 if (target->i_number == rootino)
1127 goto out;
1128
1129 for (;;) {
1130 if (vp->v_type != VDIR) {
1131 error = ENOTDIR;
1132 break;
1133 }
1134 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1135 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
dadab5e9 1136 IO_NODELOCKED, cred, (int *)0, NULL);
984263bc
MD
1137 if (error != 0)
1138 break;
1139# if (BYTE_ORDER == LITTLE_ENDIAN)
1140 if (OFSFMT(vp))
1141 namlen = dirbuf.dotdot_type;
1142 else
1143 namlen = dirbuf.dotdot_namlen;
1144# else
1145 namlen = dirbuf.dotdot_namlen;
1146# endif
1147 if (namlen != 2 ||
1148 dirbuf.dotdot_name[0] != '.' ||
1149 dirbuf.dotdot_name[1] != '.') {
1150 error = ENOTDIR;
1151 break;
1152 }
1153 if (dirbuf.dotdot_ino == source->i_number) {
1154 error = EINVAL;
1155 break;
1156 }
1157 if (dirbuf.dotdot_ino == rootino)
1158 break;
1159 vput(vp);
1160 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1161 if (error) {
1162 vp = NULL;
1163 break;
1164 }
1165 }
1166
1167out:
1168 if (error == ENOTDIR)
1169 printf("checkpath: .. not a directory\n");
1170 if (vp != NULL)
1171 vput(vp);
1172 return (error);
1173}