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.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
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 $
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/namei.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
53 #include <sys/sysctl.h>
56 #include <vm/vm_extern.h>
58 #include <ufs/ufs/quota.h>
59 #include <ufs/ufs/inode.h>
60 #include <ufs/ufs/dir.h>
62 #include <ufs/ufs/dirhash.h>
64 #include <ufs/ufs/ufsmount.h>
65 #include <ufs/ufs/ufs_extern.h>
73 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
75 /* true if old FS format...*/
76 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
79 * Convert a component of a pathname into a pointer to a locked inode.
80 * This is a very central and rather complicated routine.
81 * If the file system is not maintained in a strict tree hierarchy,
82 * this can result in a deadlock situation (see comments in code below).
84 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
85 * on whether the name is to be looked up, created, renamed, or deleted.
86 * When CREATE, RENAME, or DELETE is specified, information usable in
87 * creating, renaming, or deleting a directory entry may be calculated.
88 * If flag has LOCKPARENT or'ed into it and the target of the pathname
89 * exists, lookup returns both the target and its parent directory locked.
90 * When creating or renaming and LOCKPARENT is specified, the target may
91 * not be ".". When deleting and LOCKPARENT is specified, the target may
92 * be "."., but the caller must check to ensure it does an vrele and vput
93 * instead of two vputs.
95 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
96 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
99 * vfs_cache_lookup() performs the following for us:
100 * check that it is a directory
101 * check accessibility of directory
102 * check for modification attempts on read-only mounts
103 * if name found in cache
104 * if at end of path and deleting or creating
108 * return VOP_CACHEDLOOKUP()
110 * Overall outline of ufs_lookup:
112 * search for name in directory, to found or notfound
114 * if creating, return locked directory, leaving info on available slots
117 * if at end of path and deleting, return information to allow delete
118 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
119 * inode and return info to allow rewrite
120 * if not at end, add name to cache; if at end and neither creating
121 * nor deleting, add name to cache
125 struct vop_cachedlookup_args /* {
127 struct vnode **a_vpp;
128 struct componentname *a_cnp;
131 register struct vnode *vdp; /* vnode for directory being searched */
132 register struct inode *dp; /* inode for directory being searched */
133 struct buf *bp; /* a buffer of directory entries */
134 struct direct *ep; /* the current directory entry */
135 int entryoffsetinblock; /* offset of ep in bp's buffer */
136 enum {NONE, COMPACT, FOUND} slotstatus;
137 doff_t slotoffset; /* offset of area with free space */
138 int slotsize; /* size of area at slotoffset */
139 int slotfreespace; /* amount of space free in slot */
140 int slotneeded; /* size of the entry we're seeking */
141 int numdirpasses; /* strategy for directory search */
142 doff_t endsearch; /* offset to end directory search */
143 doff_t prevoff; /* prev entry dp->i_offset */
144 struct vnode *pdp; /* saved dp during symlink work */
145 struct vnode *tdp; /* returned by VFS_VGET */
146 doff_t enduseful; /* pointer past last used dir slot */
147 u_long bmask; /* block offset mask */
148 int lockparent; /* 1 => lockparent flag is set */
149 int wantparent; /* 1 => wantparent or lockparent flag */
151 struct vnode **vpp = ap->a_vpp;
152 struct componentname *cnp = ap->a_cnp;
153 struct ucred *cred = cnp->cn_cred;
154 int flags = cnp->cn_flags;
155 int nameiop = cnp->cn_nameiop;
156 struct proc *p = cnp->cn_proc;
160 cnp->cn_flags &= ~PDIRUNLOCK;
162 * XXX there was a soft-update diff about this I couldn't merge.
163 * I think this was the equiv.
169 lockparent = flags & LOCKPARENT;
170 wantparent = flags & (LOCKPARENT|WANTPARENT);
173 * We now have a segment name to search for, and a directory to search.
175 * Suppress search for slots unless creating
176 * file and at end of pathname, in which case
177 * we watch for a place to put the new file in
178 * case it doesn't already exist.
181 slotfreespace = slotsize = slotneeded = 0;
182 if ((nameiop == CREATE || nameiop == RENAME) &&
183 (flags & ISLASTCN)) {
185 slotneeded = DIRECTSIZ(cnp->cn_namelen);
187 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
191 * Use dirhash for fast operations on large directories. The logic
192 * to determine whether to hash the directory is contained within
193 * ufsdirhash_build(); a zero return means that it decided to hash
194 * this directory and it successfully built up the hash table.
196 if (ufsdirhash_build(dp) == 0) {
197 /* Look for a free slot if needed. */
198 enduseful = dp->i_size;
199 if (slotstatus != FOUND) {
200 slotoffset = ufsdirhash_findfree(dp, slotneeded,
202 if (slotoffset >= 0) {
203 slotstatus = COMPACT;
204 enduseful = ufsdirhash_enduseful(dp);
206 enduseful = dp->i_size;
209 /* Look up the component. */
211 entryoffsetinblock = 0; /* silence compiler warning */
212 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
213 &dp->i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
215 ep = (struct direct *)((char *)bp->b_data +
216 (dp->i_offset & bmask));
219 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
222 /* Something failed; just do a linear search. */
226 #endif /* UFS_DIRHASH */
228 * If there is cached information on a previous search of
229 * this directory, pick up where we last left off.
230 * We cache only lookups as these are the most common
231 * and have the greatest payoff. Caching CREATE has little
232 * benefit as it usually must search the entire directory
233 * to determine that the entry does not exist. Caching the
234 * location of the last DELETE or RENAME has not reduced
235 * profiling time and hence has been removed in the interest
238 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
239 dp->i_diroff >= dp->i_size) {
240 entryoffsetinblock = 0;
244 dp->i_offset = dp->i_diroff;
245 if ((entryoffsetinblock = dp->i_offset & bmask) &&
246 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
249 nchstats.ncs_2passes++;
251 prevoff = dp->i_offset;
252 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
256 while (dp->i_offset < endsearch) {
258 * If necessary, get the next directory block.
260 if ((dp->i_offset & bmask) == 0) {
264 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
267 entryoffsetinblock = 0;
270 * If still looking for a slot, and at a DIRBLKSIZE
271 * boundary, have to start looking for free space again.
273 if (slotstatus == NONE &&
274 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
279 * Get pointer to next entry.
280 * Full validation checks are slow, so we only check
281 * enough to insure forward progress through the
282 * directory. Complete checks can be run by patching
283 * "dirchk" to be true.
285 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
286 if (ep->d_reclen == 0 || ep->d_reclen >
287 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
288 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
291 ufs_dirbad(dp, dp->i_offset, "mangled entry");
292 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
294 entryoffsetinblock += i;
299 * If an appropriate sized slot has not yet been found,
300 * check to see if one is available. Also accumulate space
301 * in the current block so that we can determine if
302 * compaction is viable.
304 if (slotstatus != FOUND) {
305 int size = ep->d_reclen;
308 size -= DIRSIZ(OFSFMT(vdp), ep);
310 if (size >= slotneeded) {
312 slotoffset = dp->i_offset;
313 slotsize = ep->d_reclen;
314 } else if (slotstatus == NONE) {
315 slotfreespace += size;
316 if (slotoffset == -1)
317 slotoffset = dp->i_offset;
318 if (slotfreespace >= slotneeded) {
319 slotstatus = COMPACT;
320 slotsize = dp->i_offset +
321 ep->d_reclen - slotoffset;
328 * Check for a name match.
331 # if (BYTE_ORDER == LITTLE_ENDIAN)
335 namlen = ep->d_namlen;
337 namlen = ep->d_namlen;
339 if (namlen == cnp->cn_namelen &&
340 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
341 !bcmp(cnp->cn_nameptr, ep->d_name,
347 * Save directory entry's inode number and
348 * reclen in ndp->ni_ufs area, and release
351 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
352 ep->d_type == DT_WHT) {
354 slotoffset = dp->i_offset;
355 slotsize = ep->d_reclen;
356 dp->i_reclen = slotsize;
357 enduseful = dp->i_size;
358 ap->a_cnp->cn_flags |= ISWHITEOUT;
362 dp->i_ino = ep->d_ino;
363 dp->i_reclen = ep->d_reclen;
367 prevoff = dp->i_offset;
368 dp->i_offset += ep->d_reclen;
369 entryoffsetinblock += ep->d_reclen;
371 enduseful = dp->i_offset;
375 * If we started in the middle of the directory and failed
376 * to find our target, we must check the beginning as well.
378 if (numdirpasses == 2) {
381 endsearch = dp->i_diroff;
387 * If creating, and at end of pathname and current
388 * directory has not been removed, then can consider
389 * allowing file to be created.
391 if ((nameiop == CREATE || nameiop == RENAME ||
392 (nameiop == DELETE &&
393 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
394 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
395 (flags & ISLASTCN) && dp->i_effnlink != 0) {
397 * Access for write is interpreted as allowing
398 * creation of files in the directory.
400 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
404 * Return an indication of where the new directory
405 * entry should be put. If we didn't find a slot,
406 * then set dp->i_count to 0 indicating
407 * that the new slot belongs at the end of the
408 * directory. If we found a slot, then the new entry
409 * can be put in the range from dp->i_offset to
410 * dp->i_offset + dp->i_count.
412 if (slotstatus == NONE) {
413 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
415 enduseful = dp->i_offset;
416 } else if (nameiop == DELETE) {
417 dp->i_offset = slotoffset;
418 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
421 dp->i_count = dp->i_offset - prevoff;
423 dp->i_offset = slotoffset;
424 dp->i_count = slotsize;
425 if (enduseful < slotoffset + slotsize)
426 enduseful = slotoffset + slotsize;
428 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
429 dp->i_flag |= IN_CHANGE | IN_UPDATE;
431 * We return with the directory locked, so that
432 * the parameters we set up above will still be
433 * valid if we actually decide to do a direnter().
434 * We return ni_vp == NULL to indicate that the entry
435 * does not currently exist; we leave a pointer to
436 * the (locked) directory inode in ndp->ni_dvp.
437 * The pathname buffer is saved so that the name
438 * can be obtained later.
440 * NB - if the directory is unlocked, then this
441 * information cannot be used.
443 cnp->cn_flags |= SAVENAME;
445 VOP_UNLOCK(vdp, 0, p);
446 cnp->cn_flags |= PDIRUNLOCK;
448 return (EJUSTRETURN);
451 * Insert name into cache (as non-existent) if appropriate.
453 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
454 cache_enter(vdp, *vpp, cnp);
458 if (numdirpasses == 2)
459 nchstats.ncs_pass2++;
461 * Check that directory length properly reflects presence
464 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
465 ufs_dirbad(dp, dp->i_offset, "i_size too small");
466 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
467 dp->i_flag |= IN_CHANGE | IN_UPDATE;
472 * Found component in pathname.
473 * If the final component of path name, save information
474 * in the cache as to where the entry was found.
476 if ((flags & ISLASTCN) && nameiop == LOOKUP)
477 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
480 * If deleting, and at end of pathname, return
481 * parameters which can be used to remove file.
482 * If the wantparent flag isn't set, we return only
483 * the directory (in ndp->ni_dvp), otherwise we go
484 * on and lock the inode, being careful with ".".
486 if (nameiop == DELETE && (flags & ISLASTCN)) {
488 * Write access to directory required to delete files.
490 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
494 * Return pointer to current entry in dp->i_offset,
495 * and distance past previous entry (if there
496 * is a previous entry in this block) in dp->i_count.
497 * Save directory inode pointer in ndp->ni_dvp for dirremove().
499 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
502 dp->i_count = dp->i_offset - prevoff;
503 if (dp->i_number == dp->i_ino) {
508 if (flags & ISDOTDOT)
509 VOP_UNLOCK(vdp, 0, p); /* race to get the inode */
510 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
511 if (flags & ISDOTDOT) {
512 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, p) != 0)
513 cnp->cn_flags |= PDIRUNLOCK;
518 * If directory is "sticky", then user must own
519 * the directory, or the file in it, else she
520 * may not delete it (unless she's root). This
521 * implements append-only directories.
523 if ((dp->i_mode & ISVTX) &&
525 cred->cr_uid != dp->i_uid &&
526 VTOI(tdp)->i_uid != cred->cr_uid) {
532 VOP_UNLOCK(vdp, 0, p);
533 cnp->cn_flags |= PDIRUNLOCK;
539 * If rewriting (RENAME), return the inode and the
540 * information required to rewrite the present directory
541 * Must get inode of directory entry to verify it's a
542 * regular file, or empty directory.
544 if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
545 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc)) != 0)
548 * Careful about locking second inode.
549 * This can only occur if the target is ".".
551 if (dp->i_number == dp->i_ino)
553 if (flags & ISDOTDOT)
554 VOP_UNLOCK(vdp, 0, p); /* race to get the inode */
555 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
556 if (flags & ISDOTDOT) {
557 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, p) != 0)
558 cnp->cn_flags |= PDIRUNLOCK;
563 cnp->cn_flags |= SAVENAME;
565 VOP_UNLOCK(vdp, 0, p);
566 cnp->cn_flags |= PDIRUNLOCK;
572 * Step through the translation in the name. We do not `vput' the
573 * directory because we may need it again if a symbolic link
574 * is relative to the current directory. Instead we save it
575 * unlocked as "pdp". We must get the target inode before unlocking
576 * the directory to insure that the inode will not be removed
577 * before we get it. We prevent deadlock by always fetching
578 * inodes from the root, moving down the directory tree. Thus
579 * when following backward pointers ".." we must unlock the
580 * parent directory before getting the requested directory.
581 * There is a potential race condition here if both the current
582 * and parent directories are removed before the VFS_VGET for the
583 * inode associated with ".." returns. We hope that this occurs
584 * infrequently since we cannot avoid this race condition without
585 * implementing a sophisticated deadlock detection algorithm.
586 * Note also that this simple deadlock detection scheme will not
587 * work if the file system has any hard links other than ".."
588 * that point backwards in the directory structure.
591 if (flags & ISDOTDOT) {
592 VOP_UNLOCK(pdp, 0, p); /* race to get the inode */
593 cnp->cn_flags |= PDIRUNLOCK;
594 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
595 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p) == 0)
596 cnp->cn_flags &= ~PDIRUNLOCK;
599 if (lockparent && (flags & ISLASTCN)) {
600 if ((error = vn_lock(pdp, LK_EXCLUSIVE, p)) != 0) {
604 cnp->cn_flags &= ~PDIRUNLOCK;
607 } else if (dp->i_number == dp->i_ino) {
608 VREF(vdp); /* we want ourself, ie "." */
611 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
614 if (!lockparent || !(flags & ISLASTCN)) {
615 VOP_UNLOCK(pdp, 0, p);
616 cnp->cn_flags |= PDIRUNLOCK;
622 * Insert name into cache if appropriate.
624 if (cnp->cn_flags & MAKEENTRY)
625 cache_enter(vdp, *vpp, cnp);
630 ufs_dirbad(ip, offset, how)
637 mp = ITOV(ip)->v_mount;
638 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
639 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
640 if ((mp->mnt_flag & MNT_RDONLY) == 0)
641 panic("ufs_dirbad: bad dir");
645 * Do consistency checking on a directory entry:
646 * record length must be multiple of 4
647 * entry must fit in rest of its DIRBLKSIZ block
648 * record must be large enough to contain entry
649 * name is not longer than MAXNAMLEN
650 * name must be as long as advertised, and null terminated
653 ufs_dirbadentry(dp, ep, entryoffsetinblock)
655 register struct direct *ep;
656 int entryoffsetinblock;
661 # if (BYTE_ORDER == LITTLE_ENDIAN)
665 namlen = ep->d_namlen;
667 namlen = ep->d_namlen;
669 if ((ep->d_reclen & 0x3) != 0 ||
670 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
671 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
673 printf("First bad\n");
678 for (i = 0; i < namlen; i++)
679 if (ep->d_name[i] == '\0') {
681 printf("Second bad\n");
692 * Construct a new directory entry after a call to namei, using the
693 * parameters that it left in the componentname argument cnp. The
694 * argument ip is the inode to which the new directory entry will refer.
697 ufs_makedirentry(ip, cnp, newdirp)
699 struct componentname *cnp;
700 struct direct *newdirp;
704 if ((cnp->cn_flags & SAVENAME) == 0)
705 panic("ufs_makedirentry: missing name");
707 newdirp->d_ino = ip->i_number;
708 newdirp->d_namlen = cnp->cn_namelen;
709 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
710 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
711 newdirp->d_type = IFTODT(ip->i_mode);
714 # if (BYTE_ORDER == LITTLE_ENDIAN)
715 { u_char tmp = newdirp->d_namlen;
716 newdirp->d_namlen = newdirp->d_type;
717 newdirp->d_type = tmp; }
723 * Write a directory entry after a call to namei, using the parameters
724 * that it left in nameidata. The argument dirp is the new directory
725 * entry contents. Dvp is a pointer to the directory to be written,
726 * which was left locked by namei. Remaining parameters (dp->i_offset,
727 * dp->i_count) indicate how the space for the new entry is to be obtained.
728 * Non-null bp indicates that a directory is being created (for the
729 * soft dependency code).
732 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
736 struct componentname *cnp;
737 struct buf *newdirbp;
745 struct direct *ep, *nep;
746 int error, ret, blkoff, loc, spacefree, flags;
749 p = curproc; /* XXX */
753 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
755 if (dp->i_count == 0) {
757 * If dp->i_count is 0, then namei could find no
758 * space in the directory. Here, dp->i_offset will
759 * be on a directory block boundary and we will write the
760 * new entry into a fresh block.
762 if (dp->i_offset & (DIRBLKSIZ - 1))
763 panic("ufs_direnter: newblk");
765 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
767 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
768 cr, flags, &bp)) != 0) {
769 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
773 dp->i_size = dp->i_offset + DIRBLKSIZ;
774 dp->i_flag |= IN_CHANGE | IN_UPDATE;
775 vnode_pager_setsize(dvp, (u_long)dp->i_size);
776 dirp->d_reclen = DIRBLKSIZ;
777 blkoff = dp->i_offset &
778 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
779 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
781 if (dp->i_dirhash != NULL) {
782 ufsdirhash_newblk(dp, dp->i_offset);
783 ufsdirhash_add(dp, dirp, dp->i_offset);
784 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
788 if (DOINGSOFTDEP(dvp)) {
790 * Ensure that the entire newly allocated block is a
791 * valid directory so that future growth within the
792 * block does not have to ensure that the block is
793 * written before the inode.
796 while (blkoff < bp->b_bcount) {
798 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
801 softdep_setup_directory_add(bp, dp, dp->i_offset,
802 dirp->d_ino, newdirbp);
804 return (UFS_UPDATE(dvp, 0));
806 if (DOINGASYNC(dvp)) {
808 return (UFS_UPDATE(dvp, 0));
810 error = VOP_BWRITE(bp->b_vp, bp);
811 ret = UFS_UPDATE(dvp, 1);
818 * If dp->i_count is non-zero, then namei found space for the new
819 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
820 * in the directory. To use this space, we may have to compact
821 * the entries located there, by copying them together towards the
822 * beginning of the block, leaving the free space in one usable
827 * Increase size of directory if entry eats into new space.
828 * This should never push the size past a new multiple of
831 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
833 if (dp->i_offset + dp->i_count > dp->i_size)
834 dp->i_size = dp->i_offset + dp->i_count;
836 * Get the block containing the space for the new directory entry.
838 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
840 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
845 * Find space for the new entry. In the simple case, the entry at
846 * offset base will have the space. If it does not, then namei
847 * arranged that compacting the region dp->i_offset to
848 * dp->i_offset + dp->i_count would yield the space.
850 ep = (struct direct *)dirbuf;
851 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
852 spacefree = ep->d_reclen - dsize;
853 for (loc = ep->d_reclen; loc < dp->i_count; ) {
854 nep = (struct direct *)(dirbuf + loc);
856 /* Trim the existing slot (NB: dsize may be zero). */
857 ep->d_reclen = dsize;
858 ep = (struct direct *)((char *)ep + dsize);
860 /* Read nep->d_reclen now as the bcopy() may clobber it. */
861 loc += nep->d_reclen;
862 if (nep->d_ino == 0) {
864 * A mid-block unused entry. Such entries are
865 * never created by the kernel, but fsck_ffs
866 * can create them (and it doesn't fix them).
868 * Add up the free space, and initialise the
869 * relocated entry since we don't bcopy it.
871 spacefree += nep->d_reclen;
876 dsize = DIRSIZ(OFSFMT(dvp), nep);
877 spacefree += nep->d_reclen - dsize;
879 if (dp->i_dirhash != NULL)
880 ufsdirhash_move(dp, nep,
881 dp->i_offset + ((char *)nep - dirbuf),
882 dp->i_offset + ((char *)ep - dirbuf));
884 if (DOINGSOFTDEP(dvp))
885 softdep_change_directoryentry_offset(dp, dirbuf,
886 (caddr_t)nep, (caddr_t)ep, dsize);
888 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
891 * Here, `ep' points to a directory entry containing `dsize' in-use
892 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
893 * then the entry is completely unused (dsize == 0). The value
894 * of ep->d_reclen is always indeterminate.
896 * Update the pointer fields in the previous entry (if any),
897 * copy in the new entry, and write out the block.
899 if (ep->d_ino == 0 ||
900 (ep->d_ino == WINO &&
901 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
902 if (spacefree + dsize < newentrysize)
903 panic("ufs_direnter: compact1");
904 dirp->d_reclen = spacefree + dsize;
906 if (spacefree < newentrysize)
907 panic("ufs_direnter: compact2");
908 dirp->d_reclen = spacefree;
909 ep->d_reclen = dsize;
910 ep = (struct direct *)((char *)ep + dsize);
913 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
914 dirp->d_reclen == spacefree))
915 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
917 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
919 if (dp->i_dirhash != NULL)
920 ufsdirhash_checkblock(dp, dirbuf -
921 (dp->i_offset & (DIRBLKSIZ - 1)),
922 dp->i_offset & ~(DIRBLKSIZ - 1));
925 if (DOINGSOFTDEP(dvp)) {
926 softdep_setup_directory_add(bp, dp,
927 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
930 if (DOINGASYNC(dvp)) {
937 dp->i_flag |= IN_CHANGE | IN_UPDATE;
939 * If all went well, and the directory can be shortened, proceed
940 * with the truncation. Note that we have to unlock the inode for
941 * the entry that we just entered, as the truncation may need to
942 * lock other inodes which can lead to deadlock if we also hold a
943 * lock on the newly entered node.
945 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
947 VOP_UNLOCK(tvp, 0, p);
949 if (dp->i_dirhash != NULL)
950 ufsdirhash_dirtrunc(dp, dp->i_endoff);
952 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cr, p);
954 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
960 * Remove a directory entry after a call to namei, using
961 * the parameters which it left in nameidata. The entry
962 * dp->i_offset contains the offset into the directory of the
963 * entry to be eliminated. The dp->i_count field contains the
964 * size of the previous record in the directory. If this
965 * is 0, the first entry is being deleted, so we need only
966 * zero the inode number to mark the entry as free. If the
967 * entry is not the first in the directory, we must reclaim
968 * the space of the now empty record by adding the record size
969 * to the size of the previous entry.
972 ufs_dirremove(dvp, ip, flags, isrmdir)
985 if (flags & DOWHITEOUT) {
987 * Whiteout entry: set d_ino to WINO.
990 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
997 if ((error = UFS_BLKATOFF(dvp,
998 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
1002 * Remove the dirhash entry. This is complicated by the fact
1003 * that `ep' is the previous entry when dp->i_count != 0.
1005 if (dp->i_dirhash != NULL)
1006 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
1007 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
1009 if (dp->i_count == 0) {
1011 * First entry in block: set d_ino to zero.
1016 * Collapse new free space into previous entry.
1018 ep->d_reclen += dp->i_reclen;
1021 if (dp->i_dirhash != NULL)
1022 ufsdirhash_checkblock(dp, (char *)ep -
1023 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1024 dp->i_offset & ~(DIRBLKSIZ - 1));
1027 if (DOINGSOFTDEP(dvp)) {
1030 softdep_change_linkcnt(ip);
1031 softdep_setup_remove(bp, dp, ip, isrmdir);
1033 if (softdep_slowdown(dvp)) {
1034 error = VOP_BWRITE(bp->b_vp, bp);
1043 ip->i_flag |= IN_CHANGE;
1045 if (flags & DOWHITEOUT)
1046 error = VOP_BWRITE(bp->b_vp, bp);
1047 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1051 error = bowrite(bp);
1053 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1058 * Rewrite an existing directory entry to point at the inode
1059 * supplied. The parameters describing the directory entry are
1060 * set up by a call to namei.
1063 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1064 struct inode *dp, *oip;
1071 struct vnode *vdp = ITOV(dp);
1074 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1077 ep->d_ino = newinum;
1079 ep->d_type = newtype;
1081 if (DOINGSOFTDEP(vdp)) {
1082 softdep_change_linkcnt(oip);
1083 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1087 oip->i_flag |= IN_CHANGE;
1088 if (DOINGASYNC(vdp)) {
1092 error = bowrite(bp);
1095 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1100 * Check if a directory is empty or not.
1101 * Inode supplied must be locked.
1103 * Using a struct dirtemplate here is not precisely
1104 * what we want, but better than using a struct direct.
1106 * NB: does not handle corrupted directories.
1109 ufs_dirempty(ip, parentino, cred)
1110 register struct inode *ip;
1115 struct dirtemplate dbuf;
1116 register struct direct *dp = (struct direct *)&dbuf;
1117 int error, count, namlen;
1118 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1120 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1121 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
1122 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
1124 * Since we read MINDIRSIZ, residual must
1125 * be 0 unless we're at end of file.
1127 if (error || count != 0)
1129 /* avoid infinite loops */
1130 if (dp->d_reclen == 0)
1132 /* skip empty entries */
1133 if (dp->d_ino == 0 || dp->d_ino == WINO)
1135 /* accept only "." and ".." */
1136 # if (BYTE_ORDER == LITTLE_ENDIAN)
1137 if (OFSFMT(ITOV(ip)))
1138 namlen = dp->d_type;
1140 namlen = dp->d_namlen;
1142 namlen = dp->d_namlen;
1146 if (dp->d_name[0] != '.')
1149 * At this point namlen must be 1 or 2.
1150 * 1 implies ".", 2 implies ".." if second
1153 if (namlen == 1 && dp->d_ino == ip->i_number)
1155 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1163 * Check if source directory is in the path of the target directory.
1164 * Target is supplied locked, source is unlocked.
1165 * The target is always vput before returning.
1168 ufs_checkpath(source, target, cred)
1169 struct inode *source, *target;
1173 int error, rootino, namlen;
1174 struct dirtemplate dirbuf;
1177 if (target->i_number == source->i_number) {
1183 if (target->i_number == rootino)
1187 if (vp->v_type != VDIR) {
1191 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1192 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1193 IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
1196 # if (BYTE_ORDER == LITTLE_ENDIAN)
1198 namlen = dirbuf.dotdot_type;
1200 namlen = dirbuf.dotdot_namlen;
1202 namlen = dirbuf.dotdot_namlen;
1205 dirbuf.dotdot_name[0] != '.' ||
1206 dirbuf.dotdot_name[1] != '.') {
1210 if (dirbuf.dotdot_ino == source->i_number) {
1214 if (dirbuf.dotdot_ino == rootino)
1217 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1225 if (error == ENOTDIR)
1226 printf("checkpath: .. not a directory\n");