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 $
40 * $DragonFly: src/sys/vfs/ufs/ufs_lookup.c,v 1.29 2008/10/15 12:12:51 aggelos Exp $
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
50 #include <sys/namei.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/sysctl.h>
57 #include <vm/vm_extern.h>
66 #include "ufs_extern.h"
67 #include "ffs_extern.h"
75 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
77 /* true if old FS format...*/
78 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
81 * Convert a component of a pathname into a pointer to a locked inode.
82 * This is a very central and rather complicated routine.
83 * If the filesystem is not maintained in a strict tree hierarchy,
84 * this can result in a deadlock situation (see comments in code below).
86 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
87 * on whether the name is to be looked up, created, renamed, or deleted.
88 * When CREATE, RENAME, or DELETE is specified, information usable in
89 * creating, renaming, or deleting a directory entry may be calculated.
90 * If flag has LOCKPARENT or'ed into it and the target of the pathname
91 * exists, lookup returns both the target and its parent directory locked.
92 * When creating or renaming and LOCKPARENT is specified, the target may
93 * not be ".". When deleting and LOCKPARENT is specified, the target may
94 * be "."., but the caller must check to ensure it does an vrele and vput
95 * instead of two vputs.
97 * Overall outline of ufs_lookup:
99 * search for name in directory, to found or notfound
101 * if creating, return locked directory, leaving info on available slots
104 * if at end of path and deleting, return information to allow delete
105 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
106 * inode and return info to allow rewrite
107 * if not at end, add name to cache; if at end and neither creating
108 * nor deleting, add name to cache
110 * ufs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
111 * struct componentname *a_cnp)
114 ufs_lookup(struct vop_old_lookup_args *ap)
116 struct vnode *vdp; /* vnode for directory being searched */
117 struct inode *dp; /* inode for directory being searched */
118 struct buf *bp; /* a buffer of directory entries */
119 struct direct *ep; /* the current directory entry */
120 int entryoffsetinblock; /* offset of ep in bp's buffer */
121 enum {NONE, COMPACT, FOUND} slotstatus;
122 doff_t slotoffset; /* offset of area with free space */
123 int slotsize; /* size of area at slotoffset */
124 int slotfreespace; /* amount of space free in slot */
125 int slotneeded; /* size of the entry we're seeking */
126 int numdirpasses; /* strategy for directory search */
127 doff_t endsearch; /* offset to end directory search */
128 doff_t prevoff; /* prev entry dp->i_offset */
129 struct vnode *pdp; /* saved dp during symlink work */
130 struct vnode *tdp; /* returned by VFS_VGET */
131 doff_t enduseful; /* pointer past last used dir slot */
132 u_long bmask; /* block offset mask */
133 int lockparent; /* 1 => lockparent flag is set */
134 int wantparent; /* 1 => wantparent or lockparent flag */
136 struct vnode **vpp = ap->a_vpp;
137 struct componentname *cnp = ap->a_cnp;
138 struct ucred *cred = cnp->cn_cred;
139 int flags = cnp->cn_flags;
140 int nameiop = cnp->cn_nameiop;
141 globaldata_t gd = mycpu;
145 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
147 * XXX there was a soft-update diff about this I couldn't merge.
148 * I think this was the equiv.
154 lockparent = flags & CNP_LOCKPARENT;
155 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
158 * We now have a segment name to search for, and a directory to search.
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.
166 slotfreespace = slotsize = slotneeded = 0;
167 if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
169 slotneeded = DIRECTSIZ(cnp->cn_namelen);
171 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
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.
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,
186 if (slotoffset >= 0) {
187 slotstatus = COMPACT;
188 enduseful = ufsdirhash_enduseful(dp);
190 enduseful = dp->i_size;
193 /* Look up the component. */
195 entryoffsetinblock = 0; /* silence compiler warning */
196 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
197 &dp->i_offset, &bp, nameiop == NAMEI_DELETE ? &prevoff : NULL)) {
199 ep = (struct direct *)((char *)bp->b_data +
200 (dp->i_offset & bmask));
203 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
206 /* Something failed; just do a linear search. */
210 #endif /* UFS_DIRHASH */
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
222 if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
223 dp->i_diroff >= dp->i_size) {
224 entryoffsetinblock = 0;
228 dp->i_offset = dp->i_diroff;
229 if ((entryoffsetinblock = dp->i_offset & bmask) &&
230 (error = ffs_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp)))
233 gd->gd_nchstats->ncs_2passes++;
235 prevoff = dp->i_offset;
236 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
240 while (dp->i_offset < endsearch) {
242 * If necessary, get the next directory block.
244 if ((dp->i_offset & bmask) == 0) {
248 ffs_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp);
251 entryoffsetinblock = 0;
254 * If still looking for a slot, and at a DIRBLKSIZE
255 * boundary, have to start looking for free space again.
257 if (slotstatus == NONE &&
258 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
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.
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))) {
275 ufs_dirbad(dp, dp->i_offset, "mangled entry");
276 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
278 entryoffsetinblock += i;
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.
288 if (slotstatus != FOUND) {
289 int size = ep->d_reclen;
292 size -= DIRSIZ(OFSFMT(vdp), ep);
294 if (size >= slotneeded) {
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;
312 * Check for a name match.
315 # if (BYTE_ORDER == LITTLE_ENDIAN)
319 namlen = ep->d_namlen;
321 namlen = ep->d_namlen;
323 if (namlen == cnp->cn_namelen &&
324 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
325 !bcmp(cnp->cn_nameptr, ep->d_name,
331 * Save directory entry's inode number and
332 * reclen in ndp->ni_ufs area, and release
335 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
336 ep->d_type == DT_WHT) {
338 slotoffset = dp->i_offset;
339 slotsize = ep->d_reclen;
340 dp->i_reclen = slotsize;
341 enduseful = dp->i_size;
342 ap->a_cnp->cn_flags |= CNP_ISWHITEOUT;
346 dp->i_ino = ep->d_ino;
347 dp->i_reclen = ep->d_reclen;
351 prevoff = dp->i_offset;
352 dp->i_offset += ep->d_reclen;
353 entryoffsetinblock += ep->d_reclen;
355 enduseful = dp->i_offset;
359 * If we started in the middle of the directory and failed
360 * to find our target, we must check the beginning as well.
362 if (numdirpasses == 2) {
365 endsearch = dp->i_diroff;
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.
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))) &&
379 dp->i_effnlink != 0) {
381 * Access for write is interpreted as allowing
382 * creation of files in the directory.
384 error = VOP_ACCESS(vdp, VWRITE, cred);
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.
396 if (slotstatus == NONE) {
397 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
399 enduseful = dp->i_offset;
400 } else if (nameiop == NAMEI_DELETE) {
401 dp->i_offset = slotoffset;
402 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
405 dp->i_count = dp->i_offset - prevoff;
407 dp->i_offset = slotoffset;
408 dp->i_count = slotsize;
409 if (enduseful < slotoffset + slotsize)
410 enduseful = slotoffset + slotsize;
412 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
413 dp->i_flag |= IN_CHANGE | IN_UPDATE;
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.
424 * NB - if the directory is unlocked, then this
425 * information cannot be used.
429 cnp->cn_flags |= CNP_PDIRUNLOCK;
431 return (EJUSTRETURN);
436 if (numdirpasses == 2)
437 gd->gd_nchstats->ncs_pass2++;
439 * Check that directory length properly reflects presence
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;
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.
454 if (nameiop == NAMEI_LOOKUP)
455 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
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 ".".
464 if (nameiop == NAMEI_DELETE) {
466 * Write access to directory required to delete files.
468 error = VOP_ACCESS(vdp, VWRITE, cred);
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().
477 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
480 dp->i_count = dp->i_offset - prevoff;
481 if (dp->i_number == dp->i_ino) {
486 if (flags & CNP_ISDOTDOT)
487 vn_unlock(vdp); /* race to get the inode */
488 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
489 if (flags & CNP_ISDOTDOT) {
490 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY) != 0)
491 cnp->cn_flags |= CNP_PDIRUNLOCK;
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.
501 if ((dp->i_mode & ISVTX) &&
503 cred->cr_uid != dp->i_uid &&
504 VTOI(tdp)->i_uid != cred->cr_uid) {
511 cnp->cn_flags |= CNP_PDIRUNLOCK;
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.
522 if (nameiop == NAMEI_RENAME && wantparent) {
523 if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
526 * Careful about locking second inode.
527 * This can only occur if the target is ".".
529 if (dp->i_number == dp->i_ino)
531 if (flags & CNP_ISDOTDOT)
532 vn_unlock(vdp); /* race to get the inode */
533 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
534 if (flags & CNP_ISDOTDOT) {
535 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY) != 0)
536 cnp->cn_flags |= CNP_PDIRUNLOCK;
543 cnp->cn_flags |= CNP_PDIRUNLOCK;
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
564 * work if the filesystem has any hard links other than ".."
565 * that point backwards in the directory structure.
568 if (flags & CNP_ISDOTDOT) {
569 vn_unlock(pdp); /* race to get the inode */
570 cnp->cn_flags |= CNP_PDIRUNLOCK;
571 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
573 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY) == 0)
574 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
578 if ((error = vn_lock(pdp, LK_EXCLUSIVE)) != 0) {
582 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
585 } else if (dp->i_number == dp->i_ino) {
586 vref(vdp); /* we want ourself, ie "." */
589 error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp);
594 cnp->cn_flags |= CNP_PDIRUNLOCK;
602 ufs_dirbad(struct inode *ip, doff_t offset, char *how)
606 mp = ITOV(ip)->v_mount;
607 (void)kprintf("%s: bad dir ino %lu at offset %ld: %s\n",
608 mp->mnt_stat.f_mntfromname, (u_long)ip->i_number, (long)offset, how);
609 if ((mp->mnt_flag & MNT_RDONLY) == 0)
610 panic("ufs_dirbad: bad dir");
614 * Do consistency checking on a directory entry:
615 * record length must be multiple of 4
616 * entry must fit in rest of its DIRBLKSIZ block
617 * record must be large enough to contain entry
618 * name is not longer than MAXNAMLEN
619 * name must be as long as advertised, and null terminated
622 ufs_dirbadentry(struct vnode *dp, struct direct *ep, int entryoffsetinblock)
627 # if (BYTE_ORDER == LITTLE_ENDIAN)
631 namlen = ep->d_namlen;
633 namlen = ep->d_namlen;
635 if ((ep->d_reclen & 0x3) != 0 ||
636 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
637 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
639 kprintf("First bad\n");
644 for (i = 0; i < namlen; i++)
645 if (ep->d_name[i] == '\0') {
647 kprintf("Second bad\n");
658 * Construct a new directory entry after a call to namei, using the
659 * parameters that it left in the componentname argument cnp. The
660 * argument ip is the inode to which the new directory entry will refer.
663 ufs_makedirentry(struct inode *ip, struct componentname *cnp,
664 struct direct *newdirp)
667 newdirp->d_ino = ip->i_number;
668 newdirp->d_namlen = cnp->cn_namelen;
669 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
670 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
671 newdirp->d_type = IFTODT(ip->i_mode);
674 # if (BYTE_ORDER == LITTLE_ENDIAN)
675 { u_char tmp = newdirp->d_namlen;
676 newdirp->d_namlen = newdirp->d_type;
677 newdirp->d_type = tmp; }
683 * Write a directory entry after a call to namei, using the parameters
684 * that it left in the directory inode. The argument dirp is the new directory
685 * entry contents. Dvp is a pointer to the directory to be written,
686 * which was left locked by namei. Remaining parameters (dp->i_offset,
687 * dp->i_count) indicate how the space for the new entry is to be obtained.
688 * Non-null bp indicates that a directory is being created (for the
689 * soft dependency code).
692 ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
693 struct componentname *cnp, struct buf *newdirbp)
700 struct direct *ep, *nep;
701 int error, ret, blkoff, loc, spacefree, flags;
705 KKASSERT(cred != NULL);
708 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
710 if (dp->i_count == 0) {
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.
717 if (dp->i_offset & (DIRBLKSIZ - 1))
718 panic("ufs_direnter: newblk");
719 vnode_pager_setsize(dvp, dp->i_offset + DIRBLKSIZ);
721 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
723 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
724 cred, flags, &bp)) != 0) {
725 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
729 dp->i_size = dp->i_offset + DIRBLKSIZ;
730 dp->i_flag |= IN_CHANGE | IN_UPDATE;
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);
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,
743 if (DOINGSOFTDEP(dvp)) {
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.
751 while (blkoff < bp->b_bcount) {
753 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
756 softdep_setup_directory_add(bp, dp, dp->i_offset,
757 dirp->d_ino, newdirbp);
759 return (ffs_update(dvp, 0));
761 if (DOINGASYNC(dvp)) {
763 return (ffs_update(dvp, 0));
766 ret = ffs_update(dvp, 1);
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
782 * Increase size of directory if entry eats into new space.
783 * This should never push the size past a new multiple of
786 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
788 if (dp->i_offset + dp->i_count > dp->i_size)
789 dp->i_size = dp->i_offset + dp->i_count;
791 * Get the block containing the space for the new directory entry.
793 error = ffs_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
795 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
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.
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);
811 /* Trim the existing slot (NB: dsize may be zero). */
812 ep->d_reclen = dsize;
813 ep = (struct direct *)((char *)ep + dsize);
815 /* Read nep->d_reclen now as the bcopy() may clobber it. */
816 loc += nep->d_reclen;
817 if (nep->d_ino == 0) {
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).
823 * Add up the free space, and initialise the
824 * relocated entry since we don't bcopy it.
826 spacefree += nep->d_reclen;
831 dsize = DIRSIZ(OFSFMT(dvp), nep);
832 spacefree += nep->d_reclen - dsize;
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));
839 if (DOINGSOFTDEP(dvp))
840 softdep_change_directoryentry_offset(dp, dirbuf,
841 (caddr_t)nep, (caddr_t)ep, dsize);
843 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
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.
851 * Update the pointer fields in the previous entry (if any),
852 * copy in the new entry, and write out the block.
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;
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);
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));
872 bcopy((caddr_t)dirp, (caddr_t)ep, (uint)newentrysize);
874 if (dp->i_dirhash != NULL)
875 ufsdirhash_checkblock(dp, dirbuf -
876 (dp->i_offset & (DIRBLKSIZ - 1)),
877 dp->i_offset & ~(DIRBLKSIZ - 1));
880 if (DOINGSOFTDEP(dvp)) {
881 softdep_setup_directory_add(bp, dp,
882 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
885 if (DOINGASYNC(dvp)) {
892 dp->i_flag |= IN_CHANGE | IN_UPDATE;
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.
900 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
904 if (dp->i_dirhash != NULL)
905 ufsdirhash_dirtrunc(dp, dp->i_endoff);
907 (void)ffs_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC, cred);
909 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
915 * Remove a directory entry after a call to namei, using
916 * the parameters which it left in the directory inode. The entry
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.
927 ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
936 if (flags & CNP_DOWHITEOUT) {
938 * Whiteout entry: set d_ino to WINO.
941 ffs_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
948 if ((error = ffs_blkatoff(dvp,
949 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
953 * Remove the dirhash entry. This is complicated by the fact
954 * that `ep' is the previous entry when dp->i_count != 0.
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);
960 if (dp->i_count == 0) {
962 * First entry in block: set d_ino to zero.
967 * Collapse new free space into previous entry.
969 ep->d_reclen += dp->i_reclen;
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));
978 if (DOINGSOFTDEP(dvp)) {
981 softdep_change_linkcnt(ip);
982 softdep_setup_remove(bp, dp, ip, isrmdir);
984 if (softdep_slowdown(dvp)) {
994 ip->i_flag |= IN_CHANGE;
996 if (flags & CNP_DOWHITEOUT)
998 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1002 error = bowrite(bp);
1004 dp->i_flag |= IN_CHANGE | IN_UPDATE;
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.
1014 ufs_dirrewrite(struct inode *dp, struct inode *oip, ino_t newinum, int newtype,
1019 struct vnode *vdp = ITOV(dp);
1022 error = ffs_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1025 ep->d_ino = newinum;
1027 ep->d_type = newtype;
1029 if (DOINGSOFTDEP(vdp)) {
1030 softdep_change_linkcnt(oip);
1031 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1035 oip->i_flag |= IN_CHANGE;
1036 if (DOINGASYNC(vdp)) {
1040 error = bowrite(bp);
1043 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1048 * Check if a directory is empty or not.
1049 * Inode supplied must be locked.
1051 * Using a struct dirtemplate here is not precisely
1052 * what we want, but better than using a struct direct.
1054 * NB: does not handle corrupted directories.
1057 ufs_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1060 struct dirtemplate dbuf;
1061 struct direct *dp = (struct direct *)&dbuf;
1062 int error, count, namlen;
1063 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
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,
1067 UIO_SYSSPACE, IO_NODELOCKED, cred, &count);
1069 * Since we read MINDIRSIZ, residual must
1070 * be 0 unless we're at end of file.
1072 if (error || count != 0)
1074 /* avoid infinite loops */
1075 if (dp->d_reclen == 0)
1077 /* skip empty entries */
1078 if (dp->d_ino == 0 || dp->d_ino == WINO)
1080 /* accept only "." and ".." */
1081 # if (BYTE_ORDER == LITTLE_ENDIAN)
1082 if (OFSFMT(ITOV(ip)))
1083 namlen = dp->d_type;
1085 namlen = dp->d_namlen;
1087 namlen = dp->d_namlen;
1091 if (dp->d_name[0] != '.')
1094 * At this point namlen must be 1 or 2.
1095 * 1 implies ".", 2 implies ".." if second
1098 if (namlen == 1 && dp->d_ino == ip->i_number)
1100 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
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.
1113 ufs_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1116 int error, rootino, namlen;
1117 struct dirtemplate dirbuf;
1120 if (target->i_number == source->i_number) {
1126 if (target->i_number == rootino)
1130 if (vp->v_type != VDIR) {
1134 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1135 sizeof (struct dirtemplate), (off_t)0,
1136 UIO_SYSSPACE, IO_NODELOCKED, cred, NULL);
1139 # if (BYTE_ORDER == LITTLE_ENDIAN)
1141 namlen = dirbuf.dotdot_type;
1143 namlen = dirbuf.dotdot_namlen;
1145 namlen = dirbuf.dotdot_namlen;
1148 dirbuf.dotdot_name[0] != '.' ||
1149 dirbuf.dotdot_name[1] != '.') {
1153 if (dirbuf.dotdot_ino == source->i_number) {
1157 if (dirbuf.dotdot_ino == rootino)
1160 error = VFS_VGET(vp->v_mount, NULL, dirbuf.dotdot_ino, &vp);
1168 if (error == ENOTDIR)
1169 kprintf("checkpath: .. not a directory\n");