kernel - Provide descriptions for lwkt.* and debug.* sysctl's
[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 $
8d1d71de 40 * $DragonFly: src/sys/vfs/ufs/ufs_lookup.c,v 1.29 2008/10/15 12:12:51 aggelos 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"
ac690a1d 67#include "ffs_extern.h"
984263bc
MD
68
69#ifdef DIAGNOSTIC
70int dirchk = 1;
71#else
72int dirchk = 0;
73#endif
74
0c52fa62
SG
75SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0,
76 "Enable full validation checks of directory");
984263bc
MD
77
78/* true if old FS format...*/
79#define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
80
81/*
82 * Convert a component of a pathname into a pointer to a locked inode.
83 * This is a very central and rather complicated routine.
f719c866 84 * If the filesystem is not maintained in a strict tree hierarchy,
984263bc
MD
85 * this can result in a deadlock situation (see comments in code below).
86 *
87 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
88 * on whether the name is to be looked up, created, renamed, or deleted.
89 * When CREATE, RENAME, or DELETE is specified, information usable in
90 * creating, renaming, or deleting a directory entry may be calculated.
91 * If flag has LOCKPARENT or'ed into it and the target of the pathname
92 * exists, lookup returns both the target and its parent directory locked.
93 * When creating or renaming and LOCKPARENT is specified, the target may
94 * not be ".". When deleting and LOCKPARENT is specified, the target may
95 * be "."., but the caller must check to ensure it does an vrele and vput
96 * instead of two vputs.
97 *
984263bc
MD
98 * Overall outline of ufs_lookup:
99 *
100 * search for name in directory, to found or notfound
101 * notfound:
102 * if creating, return locked directory, leaving info on available slots
103 * else return error
104 * found:
105 * if at end of path and deleting, return information to allow delete
106 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
107 * inode and return info to allow rewrite
108 * if not at end, add name to cache; if at end and neither creating
109 * nor deleting, add name to cache
0973c589
CP
110 *
111 * ufs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
112 * struct componentname *a_cnp)
984263bc
MD
113 */
114int
e62afb5f 115ufs_lookup(struct vop_old_lookup_args *ap)
984263bc 116{
3ff2135f
RG
117 struct vnode *vdp; /* vnode for directory being searched */
118 struct inode *dp; /* inode for directory being searched */
984263bc
MD
119 struct buf *bp; /* a buffer of directory entries */
120 struct direct *ep; /* the current directory entry */
121 int entryoffsetinblock; /* offset of ep in bp's buffer */
122 enum {NONE, COMPACT, FOUND} slotstatus;
123 doff_t slotoffset; /* offset of area with free space */
124 int slotsize; /* size of area at slotoffset */
125 int slotfreespace; /* amount of space free in slot */
126 int slotneeded; /* size of the entry we're seeking */
127 int numdirpasses; /* strategy for directory search */
128 doff_t endsearch; /* offset to end directory search */
129 doff_t prevoff; /* prev entry dp->i_offset */
130 struct vnode *pdp; /* saved dp during symlink work */
131 struct vnode *tdp; /* returned by VFS_VGET */
132 doff_t enduseful; /* pointer past last used dir slot */
133 u_long bmask; /* block offset mask */
134 int lockparent; /* 1 => lockparent flag is set */
135 int wantparent; /* 1 => wantparent or lockparent flag */
136 int namlen, error;
137 struct vnode **vpp = ap->a_vpp;
138 struct componentname *cnp = ap->a_cnp;
139 struct ucred *cred = cnp->cn_cred;
140 int flags = cnp->cn_flags;
141 int nameiop = cnp->cn_nameiop;
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) &&
ac690a1d 230 (error = ffs_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp)))
984263bc
MD
231 return (error);
232 numdirpasses = 2;
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 =
ac690a1d 247 ffs_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp);
984263bc
MD
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 */
cb66845a 383 error = VOP_EACCESS(vdp, VWRITE, cred);
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) {
a11aaa81 427 vn_unlock(vdp);
2b69e610 428 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
429 }
430 return (EJUSTRETURN);
431 }
984263bc
MD
432 return (ENOENT);
433
434found:
984263bc
MD
435 /*
436 * Check that directory length properly reflects presence
437 * of this entry.
438 */
439 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
440 ufs_dirbad(dp, dp->i_offset, "i_size too small");
441 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
442 dp->i_flag |= IN_CHANGE | IN_UPDATE;
443 }
444 brelse(bp);
445
446 /*
447 * Found component in pathname.
448 * If the final component of path name, save information
449 * in the cache as to where the entry was found.
450 */
fad57d0e 451 if (nameiop == NAMEI_LOOKUP)
984263bc
MD
452 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
453
454 /*
455 * If deleting, and at end of pathname, return
456 * parameters which can be used to remove file.
457 * If the wantparent flag isn't set, we return only
458 * the directory (in ndp->ni_dvp), otherwise we go
459 * on and lock the inode, being careful with ".".
460 */
fad57d0e 461 if (nameiop == NAMEI_DELETE) {
984263bc
MD
462 /*
463 * Write access to directory required to delete files.
464 */
cb66845a 465 error = VOP_EACCESS(vdp, VWRITE, cred);
984263bc
MD
466 if (error)
467 return (error);
468 /*
469 * Return pointer to current entry in dp->i_offset,
470 * and distance past previous entry (if there
471 * is a previous entry in this block) in dp->i_count.
472 * Save directory inode pointer in ndp->ni_dvp for dirremove().
473 */
474 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
475 dp->i_count = 0;
476 else
477 dp->i_count = dp->i_offset - prevoff;
478 if (dp->i_number == dp->i_ino) {
597aea93 479 vref(vdp);
984263bc
MD
480 *vpp = vdp;
481 return (0);
482 }
2b69e610 483 if (flags & CNP_ISDOTDOT)
a11aaa81 484 vn_unlock(vdp); /* race to get the inode */
b9b0a6d0 485 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
2b69e610 486 if (flags & CNP_ISDOTDOT) {
ca466bae 487 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY) != 0)
2b69e610 488 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
489 }
490 if (error)
491 return (error);
492 /*
493 * If directory is "sticky", then user must own
494 * the directory, or the file in it, else she
495 * may not delete it (unless she's root). This
496 * implements append-only directories.
497 */
498 if ((dp->i_mode & ISVTX) &&
499 cred->cr_uid != 0 &&
500 cred->cr_uid != dp->i_uid &&
501 VTOI(tdp)->i_uid != cred->cr_uid) {
502 vput(tdp);
503 return (EPERM);
504 }
505 *vpp = tdp;
506 if (!lockparent) {
a11aaa81 507 vn_unlock(vdp);
2b69e610 508 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
509 }
510 return (0);
511 }
512
513 /*
514 * If rewriting (RENAME), return the inode and the
515 * information required to rewrite the present directory
516 * Must get inode of directory entry to verify it's a
517 * regular file, or empty directory.
518 */
fad57d0e 519 if (nameiop == NAMEI_RENAME && wantparent) {
cb66845a 520 if ((error = VOP_EACCESS(vdp, VWRITE, cred)) != 0)
984263bc
MD
521 return (error);
522 /*
523 * Careful about locking second inode.
524 * This can only occur if the target is ".".
525 */
526 if (dp->i_number == dp->i_ino)
527 return (EISDIR);
2b69e610 528 if (flags & CNP_ISDOTDOT)
a11aaa81 529 vn_unlock(vdp); /* race to get the inode */
b9b0a6d0 530 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
2b69e610 531 if (flags & CNP_ISDOTDOT) {
ca466bae 532 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY) != 0)
2b69e610 533 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
534 }
535 if (error)
536 return (error);
537 *vpp = tdp;
984263bc 538 if (!lockparent) {
a11aaa81 539 vn_unlock(vdp);
2b69e610 540 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
541 }
542 return (0);
543 }
544
545 /*
546 * Step through the translation in the name. We do not `vput' the
547 * directory because we may need it again if a symbolic link
548 * is relative to the current directory. Instead we save it
549 * unlocked as "pdp". We must get the target inode before unlocking
550 * the directory to insure that the inode will not be removed
551 * before we get it. We prevent deadlock by always fetching
552 * inodes from the root, moving down the directory tree. Thus
553 * when following backward pointers ".." we must unlock the
554 * parent directory before getting the requested directory.
555 * There is a potential race condition here if both the current
556 * and parent directories are removed before the VFS_VGET for the
557 * inode associated with ".." returns. We hope that this occurs
558 * infrequently since we cannot avoid this race condition without
559 * implementing a sophisticated deadlock detection algorithm.
560 * Note also that this simple deadlock detection scheme will not
f719c866 561 * work if the filesystem has any hard links other than ".."
984263bc
MD
562 * that point backwards in the directory structure.
563 */
564 pdp = vdp;
2b69e610 565 if (flags & CNP_ISDOTDOT) {
a11aaa81 566 vn_unlock(pdp); /* race to get the inode */
2b69e610 567 cnp->cn_flags |= CNP_PDIRUNLOCK;
b9b0a6d0
MD
568 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
569 if (error) {
ca466bae 570 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
2b69e610 571 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
572 return (error);
573 }
fad57d0e 574 if (lockparent) {
ca466bae 575 if ((error = vn_lock(pdp, LK_EXCLUSIVE)) != 0) {
984263bc
MD
576 vput(tdp);
577 return (error);
578 }
2b69e610 579 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
984263bc
MD
580 }
581 *vpp = tdp;
582 } else if (dp->i_number == dp->i_ino) {
597aea93 583 vref(vdp); /* we want ourself, ie "." */
984263bc
MD
584 *vpp = vdp;
585 } else {
b9b0a6d0 586 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
984263bc
MD
587 if (error)
588 return (error);
fad57d0e 589 if (!lockparent) {
a11aaa81 590 vn_unlock(pdp);
2b69e610 591 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
592 }
593 *vpp = tdp;
594 }
984263bc
MD
595 return (0);
596}
597
598void
0973c589 599ufs_dirbad(struct inode *ip, doff_t offset, char *how)
984263bc
MD
600{
601 struct mount *mp;
602
603 mp = ITOV(ip)->v_mount;
086c1d7e 604 (void)kprintf("%s: bad dir ino %lu at offset %ld: %s\n",
75ffff0d 605 mp->mnt_stat.f_mntfromname, (u_long)ip->i_number, (long)offset, how);
984263bc
MD
606 if ((mp->mnt_flag & MNT_RDONLY) == 0)
607 panic("ufs_dirbad: bad dir");
608}
609
610/*
611 * Do consistency checking on a directory entry:
612 * record length must be multiple of 4
613 * entry must fit in rest of its DIRBLKSIZ block
614 * record must be large enough to contain entry
615 * name is not longer than MAXNAMLEN
616 * name must be as long as advertised, and null terminated
617 */
618int
0973c589 619ufs_dirbadentry(struct vnode *dp, struct direct *ep, int entryoffsetinblock)
984263bc 620{
3ff2135f 621 int i;
984263bc
MD
622 int namlen;
623
624# if (BYTE_ORDER == LITTLE_ENDIAN)
625 if (OFSFMT(dp))
626 namlen = ep->d_type;
627 else
628 namlen = ep->d_namlen;
629# else
630 namlen = ep->d_namlen;
631# endif
632 if ((ep->d_reclen & 0x3) != 0 ||
633 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
634 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
635 /*return (1); */
086c1d7e 636 kprintf("First bad\n");
984263bc
MD
637 goto bad;
638 }
639 if (ep->d_ino == 0)
640 return (0);
641 for (i = 0; i < namlen; i++)
642 if (ep->d_name[i] == '\0') {
643 /*return (1); */
086c1d7e 644 kprintf("Second bad\n");
984263bc
MD
645 goto bad;
646 }
647 if (ep->d_name[i])
648 goto bad;
649 return (0);
650bad:
651 return (1);
652}
653
654/*
655 * Construct a new directory entry after a call to namei, using the
656 * parameters that it left in the componentname argument cnp. The
657 * argument ip is the inode to which the new directory entry will refer.
658 */
659void
0973c589
CP
660ufs_makedirentry(struct inode *ip, struct componentname *cnp,
661 struct direct *newdirp)
984263bc
MD
662{
663
984263bc
MD
664 newdirp->d_ino = ip->i_number;
665 newdirp->d_namlen = cnp->cn_namelen;
666 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
667 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
668 newdirp->d_type = IFTODT(ip->i_mode);
669 else {
670 newdirp->d_type = 0;
671# if (BYTE_ORDER == LITTLE_ENDIAN)
672 { u_char tmp = newdirp->d_namlen;
673 newdirp->d_namlen = newdirp->d_type;
674 newdirp->d_type = tmp; }
675# endif
676 }
677}
678
679/*
680 * Write a directory entry after a call to namei, using the parameters
fad57d0e 681 * that it left in the directory inode. The argument dirp is the new directory
984263bc
MD
682 * entry contents. Dvp is a pointer to the directory to be written,
683 * which was left locked by namei. Remaining parameters (dp->i_offset,
684 * dp->i_count) indicate how the space for the new entry is to be obtained.
685 * Non-null bp indicates that a directory is being created (for the
686 * soft dependency code).
687 */
688int
0973c589
CP
689ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
690 struct componentname *cnp, struct buf *newdirbp)
984263bc 691{
dadab5e9 692 struct ucred *cred;
984263bc
MD
693 int newentrysize;
694 struct inode *dp;
695 struct buf *bp;
f719c866 696 uint dsize;
984263bc
MD
697 struct direct *ep, *nep;
698 int error, ret, blkoff, loc, spacefree, flags;
699 char *dirbuf;
700
8d1d71de
AE
701 cred = cnp->cn_cred;
702 KKASSERT(cred != NULL);
984263bc
MD
703
704 dp = VTOI(dvp);
705 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
706
707 if (dp->i_count == 0) {
708 /*
709 * If dp->i_count is 0, then namei could find no
710 * space in the directory. Here, dp->i_offset will
711 * be on a directory block boundary and we will write the
712 * new entry into a fresh block.
713 */
714 if (dp->i_offset & (DIRBLKSIZ - 1))
715 panic("ufs_direnter: newblk");
3bb7eedb
MD
716 nvnode_pager_setsize(dvp, dp->i_offset + DIRBLKSIZ,
717 DIRBLKSIZ, -1);
984263bc
MD
718 flags = B_CLRBUF;
719 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
720 flags |= B_SYNC;
721 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
dadab5e9 722 cred, flags, &bp)) != 0) {
984263bc
MD
723 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
724 bdwrite(newdirbp);
725 return (error);
726 }
727 dp->i_size = dp->i_offset + DIRBLKSIZ;
728 dp->i_flag |= IN_CHANGE | IN_UPDATE;
984263bc
MD
729 dirp->d_reclen = DIRBLKSIZ;
730 blkoff = dp->i_offset &
731 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
732 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
733#ifdef UFS_DIRHASH
734 if (dp->i_dirhash != NULL) {
735 ufsdirhash_newblk(dp, dp->i_offset);
736 ufsdirhash_add(dp, dirp, dp->i_offset);
737 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
738 dp->i_offset);
739 }
740#endif
741 if (DOINGSOFTDEP(dvp)) {
742 /*
743 * Ensure that the entire newly allocated block is a
744 * valid directory so that future growth within the
745 * block does not have to ensure that the block is
746 * written before the inode.
747 */
748 blkoff += DIRBLKSIZ;
749 while (blkoff < bp->b_bcount) {
750 ((struct direct *)
751 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
752 blkoff += DIRBLKSIZ;
753 }
754 softdep_setup_directory_add(bp, dp, dp->i_offset,
755 dirp->d_ino, newdirbp);
756 bdwrite(bp);
ac690a1d 757 return (ffs_update(dvp, 0));
984263bc
MD
758 }
759 if (DOINGASYNC(dvp)) {
760 bdwrite(bp);
ac690a1d 761 return (ffs_update(dvp, 0));
984263bc 762 }
62cfda27 763 error = bwrite(bp);
ac690a1d 764 ret = ffs_update(dvp, 1);
984263bc
MD
765 if (error == 0)
766 return (ret);
767 return (error);
768 }
769
770 /*
771 * If dp->i_count is non-zero, then namei found space for the new
772 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
773 * in the directory. To use this space, we may have to compact
774 * the entries located there, by copying them together towards the
775 * beginning of the block, leaving the free space in one usable
776 * chunk at the end.
777 */
778
779 /*
780 * Increase size of directory if entry eats into new space.
781 * This should never push the size past a new multiple of
782 * DIRBLKSIZE.
783 *
784 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
785 */
786 if (dp->i_offset + dp->i_count > dp->i_size)
787 dp->i_size = dp->i_offset + dp->i_count;
788 /*
789 * Get the block containing the space for the new directory entry.
790 */
ac690a1d 791 error = ffs_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
984263bc
MD
792 if (error) {
793 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
794 bdwrite(newdirbp);
795 return (error);
796 }
797 /*
798 * Find space for the new entry. In the simple case, the entry at
799 * offset base will have the space. If it does not, then namei
800 * arranged that compacting the region dp->i_offset to
801 * dp->i_offset + dp->i_count would yield the space.
802 */
803 ep = (struct direct *)dirbuf;
804 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
805 spacefree = ep->d_reclen - dsize;
806 for (loc = ep->d_reclen; loc < dp->i_count; ) {
807 nep = (struct direct *)(dirbuf + loc);
808
809 /* Trim the existing slot (NB: dsize may be zero). */
810 ep->d_reclen = dsize;
811 ep = (struct direct *)((char *)ep + dsize);
812
813 /* Read nep->d_reclen now as the bcopy() may clobber it. */
814 loc += nep->d_reclen;
815 if (nep->d_ino == 0) {
816 /*
817 * A mid-block unused entry. Such entries are
818 * never created by the kernel, but fsck_ffs
819 * can create them (and it doesn't fix them).
820 *
821 * Add up the free space, and initialise the
822 * relocated entry since we don't bcopy it.
823 */
824 spacefree += nep->d_reclen;
825 ep->d_ino = 0;
826 dsize = 0;
827 continue;
828 }
829 dsize = DIRSIZ(OFSFMT(dvp), nep);
830 spacefree += nep->d_reclen - dsize;
831#ifdef UFS_DIRHASH
832 if (dp->i_dirhash != NULL)
833 ufsdirhash_move(dp, nep,
834 dp->i_offset + ((char *)nep - dirbuf),
835 dp->i_offset + ((char *)ep - dirbuf));
836#endif
837 if (DOINGSOFTDEP(dvp))
838 softdep_change_directoryentry_offset(dp, dirbuf,
839 (caddr_t)nep, (caddr_t)ep, dsize);
840 else
841 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
842 }
843 /*
844 * Here, `ep' points to a directory entry containing `dsize' in-use
845 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
846 * then the entry is completely unused (dsize == 0). The value
847 * of ep->d_reclen is always indeterminate.
848 *
849 * Update the pointer fields in the previous entry (if any),
850 * copy in the new entry, and write out the block.
851 */
852 if (ep->d_ino == 0 ||
853 (ep->d_ino == WINO &&
854 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
855 if (spacefree + dsize < newentrysize)
856 panic("ufs_direnter: compact1");
857 dirp->d_reclen = spacefree + dsize;
858 } else {
859 if (spacefree < newentrysize)
860 panic("ufs_direnter: compact2");
861 dirp->d_reclen = spacefree;
862 ep->d_reclen = dsize;
863 ep = (struct direct *)((char *)ep + dsize);
864 }
865#ifdef UFS_DIRHASH
866 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
867 dirp->d_reclen == spacefree))
868 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
869#endif
f719c866 870 bcopy((caddr_t)dirp, (caddr_t)ep, (uint)newentrysize);
984263bc
MD
871#ifdef UFS_DIRHASH
872 if (dp->i_dirhash != NULL)
873 ufsdirhash_checkblock(dp, dirbuf -
874 (dp->i_offset & (DIRBLKSIZ - 1)),
875 dp->i_offset & ~(DIRBLKSIZ - 1));
876#endif
877
878 if (DOINGSOFTDEP(dvp)) {
879 softdep_setup_directory_add(bp, dp,
880 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
881 bdwrite(bp);
882 } else {
883 if (DOINGASYNC(dvp)) {
884 bdwrite(bp);
885 error = 0;
886 } else {
887 error = bowrite(bp);
888 }
889 }
890 dp->i_flag |= IN_CHANGE | IN_UPDATE;
891 /*
892 * If all went well, and the directory can be shortened, proceed
893 * with the truncation. Note that we have to unlock the inode for
894 * the entry that we just entered, as the truncation may need to
895 * lock other inodes which can lead to deadlock if we also hold a
896 * lock on the newly entered node.
897 */
898 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
899 if (tvp != NULL)
a11aaa81 900 vn_unlock(tvp);
984263bc
MD
901#ifdef UFS_DIRHASH
902 if (dp->i_dirhash != NULL)
903 ufsdirhash_dirtrunc(dp, dp->i_endoff);
904#endif
ac690a1d 905 (void)ffs_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, cred);
984263bc 906 if (tvp != NULL)
ca466bae 907 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
984263bc
MD
908 }
909 return (error);
910}
911
912/*
913 * Remove a directory entry after a call to namei, using
fad57d0e 914 * the parameters which it left in the directory inode. The entry
984263bc
MD
915 * dp->i_offset contains the offset into the directory of the
916 * entry to be eliminated. The dp->i_count field contains the
917 * size of the previous record in the directory. If this
918 * is 0, the first entry is being deleted, so we need only
919 * zero the inode number to mark the entry as free. If the
920 * entry is not the first in the directory, we must reclaim
921 * the space of the now empty record by adding the record size
922 * to the size of the previous entry.
923 */
924int
0973c589 925ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
984263bc
MD
926{
927 struct inode *dp;
928 struct direct *ep;
929 struct buf *bp;
930 int error;
931
932 dp = VTOI(dvp);
933
2b69e610 934 if (flags & CNP_DOWHITEOUT) {
984263bc
MD
935 /*
936 * Whiteout entry: set d_ino to WINO.
937 */
938 if ((error =
ac690a1d 939 ffs_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
984263bc
MD
940 return (error);
941 ep->d_ino = WINO;
942 ep->d_type = DT_WHT;
943 goto out;
944 }
945
ac690a1d 946 if ((error = ffs_blkatoff(dvp,
984263bc
MD
947 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
948 return (error);
949#ifdef UFS_DIRHASH
950 /*
951 * Remove the dirhash entry. This is complicated by the fact
952 * that `ep' is the previous entry when dp->i_count != 0.
953 */
954 if (dp->i_dirhash != NULL)
955 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
956 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
957#endif
958 if (dp->i_count == 0) {
959 /*
960 * First entry in block: set d_ino to zero.
961 */
962 ep->d_ino = 0;
963 } else {
964 /*
965 * Collapse new free space into previous entry.
966 */
967 ep->d_reclen += dp->i_reclen;
968 }
969#ifdef UFS_DIRHASH
970 if (dp->i_dirhash != NULL)
971 ufsdirhash_checkblock(dp, (char *)ep -
972 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
973 dp->i_offset & ~(DIRBLKSIZ - 1));
974#endif
975out:
976 if (DOINGSOFTDEP(dvp)) {
977 if (ip) {
978 ip->i_effnlink--;
979 softdep_change_linkcnt(ip);
980 softdep_setup_remove(bp, dp, ip, isrmdir);
981 }
982 if (softdep_slowdown(dvp)) {
62cfda27 983 error = bwrite(bp);
984263bc
MD
984 } else {
985 bdwrite(bp);
986 error = 0;
987 }
988 } else {
989 if (ip) {
990 ip->i_effnlink--;
991 ip->i_nlink--;
992 ip->i_flag |= IN_CHANGE;
993 }
2b69e610 994 if (flags & CNP_DOWHITEOUT)
62cfda27 995 error = bwrite(bp);
984263bc
MD
996 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
997 bdwrite(bp);
998 error = 0;
999 } else
1000 error = bowrite(bp);
1001 }
1002 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1003 return (error);
1004}
1005
1006/*
1007 * Rewrite an existing directory entry to point at the inode
1008 * supplied. The parameters describing the directory entry are
1009 * set up by a call to namei.
1010 */
1011int
0973c589
CP
1012ufs_dirrewrite(struct inode *dp, struct inode *oip, ino_t newinum, int newtype,
1013 int isrmdir)
984263bc
MD
1014{
1015 struct buf *bp;
1016 struct direct *ep;
1017 struct vnode *vdp = ITOV(dp);
1018 int error;
1019
ac690a1d 1020 error = ffs_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
984263bc
MD
1021 if (error)
1022 return (error);
1023 ep->d_ino = newinum;
1024 if (!OFSFMT(vdp))
1025 ep->d_type = newtype;
1026 oip->i_effnlink--;
1027 if (DOINGSOFTDEP(vdp)) {
1028 softdep_change_linkcnt(oip);
1029 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1030 bdwrite(bp);
1031 } else {
1032 oip->i_nlink--;
1033 oip->i_flag |= IN_CHANGE;
1034 if (DOINGASYNC(vdp)) {
1035 bdwrite(bp);
1036 error = 0;
1037 } else {
1038 error = bowrite(bp);
1039 }
1040 }
1041 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1042 return (error);
1043}
1044
1045/*
1046 * Check if a directory is empty or not.
1047 * Inode supplied must be locked.
1048 *
1049 * Using a struct dirtemplate here is not precisely
1050 * what we want, but better than using a struct direct.
1051 *
1052 * NB: does not handle corrupted directories.
1053 */
1054int
0973c589 1055ufs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
984263bc 1056{
3ff2135f 1057 off_t off;
984263bc 1058 struct dirtemplate dbuf;
3ff2135f 1059 struct direct *dp = (struct direct *)&dbuf;
984263bc
MD
1060 int error, count, namlen;
1061#define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1062
1063 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1064 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
87de5057 1065 UIO_SYSSPACE, IO_NODELOCKED, cred, &count);
984263bc
MD
1066 /*
1067 * Since we read MINDIRSIZ, residual must
1068 * be 0 unless we're at end of file.
1069 */
1070 if (error || count != 0)
1071 return (0);
1072 /* avoid infinite loops */
1073 if (dp->d_reclen == 0)
1074 return (0);
1075 /* skip empty entries */
1076 if (dp->d_ino == 0 || dp->d_ino == WINO)
1077 continue;
1078 /* accept only "." and ".." */
1079# if (BYTE_ORDER == LITTLE_ENDIAN)
1080 if (OFSFMT(ITOV(ip)))
1081 namlen = dp->d_type;
1082 else
1083 namlen = dp->d_namlen;
1084# else
1085 namlen = dp->d_namlen;
1086# endif
1087 if (namlen > 2)
1088 return (0);
1089 if (dp->d_name[0] != '.')
1090 return (0);
1091 /*
1092 * At this point namlen must be 1 or 2.
1093 * 1 implies ".", 2 implies ".." if second
1094 * char is also "."
1095 */
1096 if (namlen == 1 && dp->d_ino == ip->i_number)
1097 continue;
1098 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1099 continue;
1100 return (0);
1101 }
1102 return (1);
1103}
1104
1105/*
1106 * Check if source directory is in the path of the target directory.
1107 * Target is supplied locked, source is unlocked.
1108 * The target is always vput before returning.
1109 */
1110int
0973c589 1111ufs_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
984263bc
MD
1112{
1113 struct vnode *vp;
1114 int error, rootino, namlen;
1115 struct dirtemplate dirbuf;
1116
1117 vp = ITOV(target);
1118 if (target->i_number == source->i_number) {
1119 error = EEXIST;
1120 goto out;
1121 }
1122 rootino = ROOTINO;
1123 error = 0;
1124 if (target->i_number == rootino)
1125 goto out;
1126
1127 for (;;) {
1128 if (vp->v_type != VDIR) {
1129 error = ENOTDIR;
1130 break;
1131 }
1132 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
87de5057 1133 sizeof (struct dirtemplate), (off_t)0,
60233e58 1134 UIO_SYSSPACE, IO_NODELOCKED, cred, NULL);
984263bc
MD
1135 if (error != 0)
1136 break;
1137# if (BYTE_ORDER == LITTLE_ENDIAN)
1138 if (OFSFMT(vp))
1139 namlen = dirbuf.dotdot_type;
1140 else
1141 namlen = dirbuf.dotdot_namlen;
1142# else
1143 namlen = dirbuf.dotdot_namlen;
1144# endif
1145 if (namlen != 2 ||
1146 dirbuf.dotdot_name[0] != '.' ||
1147 dirbuf.dotdot_name[1] != '.') {
1148 error = ENOTDIR;
1149 break;
1150 }
1151 if (dirbuf.dotdot_ino == source->i_number) {
1152 error = EINVAL;
1153 break;
1154 }
1155 if (dirbuf.dotdot_ino == rootino)
1156 break;
1157 vput(vp);
b9b0a6d0 1158 error = VFS_VGET(vp->v_mount, NULL, dirbuf.dotdot_ino, &vp);
984263bc
MD
1159 if (error) {
1160 vp = NULL;
1161 break;
1162 }
1163 }
1164
1165out:
1166 if (error == ENOTDIR)
086c1d7e 1167 kprintf("checkpath: .. not a directory\n");
984263bc
MD
1168 if (vp != NULL)
1169 vput(vp);
1170 return (error);
1171}