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 1.11 2004/04/24 04:32:05 drhodus 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"
74 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
76 /* true if old FS format...*/
77 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
80 * Convert a component of a pathname into a pointer to a locked inode.
81 * This is a very central and rather complicated routine.
82 * If the file system is not maintained in a strict tree hierarchy,
83 * this can result in a deadlock situation (see comments in code below).
85 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
86 * on whether the name is to be looked up, created, renamed, or deleted.
87 * When CREATE, RENAME, or DELETE is specified, information usable in
88 * creating, renaming, or deleting a directory entry may be calculated.
89 * If flag has LOCKPARENT or'ed into it and the target of the pathname
90 * exists, lookup returns both the target and its parent directory locked.
91 * When creating or renaming and LOCKPARENT is specified, the target may
92 * not be ".". When deleting and LOCKPARENT is specified, the target may
93 * be "."., but the caller must check to ensure it does an vrele and vput
94 * instead of two vputs.
96 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
97 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
100 * vfs_cache_lookup() performs the following for us:
101 * check that it is a directory
102 * check accessibility of directory
103 * check for modification attempts on read-only mounts
104 * if name found in cache
105 * if at end of path and deleting or creating
109 * return VOP_CACHEDLOOKUP()
111 * Overall outline of ufs_lookup:
113 * search for name in directory, to found or notfound
115 * if creating, return locked directory, leaving info on available slots
118 * if at end of path and deleting, return information to allow delete
119 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
120 * inode and return info to allow rewrite
121 * if not at end, add name to cache; if at end and neither creating
122 * nor deleting, add name to cache
126 struct vop_cachedlookup_args /* {
128 struct vnode **a_vpp;
129 struct componentname *a_cnp;
132 struct vnode *vdp; /* vnode for directory being searched */
133 struct inode *dp; /* inode for directory being searched */
134 struct buf *bp; /* a buffer of directory entries */
135 struct direct *ep; /* the current directory entry */
136 int entryoffsetinblock; /* offset of ep in bp's buffer */
137 enum {NONE, COMPACT, FOUND} slotstatus;
138 doff_t slotoffset; /* offset of area with free space */
139 int slotsize; /* size of area at slotoffset */
140 int slotfreespace; /* amount of space free in slot */
141 int slotneeded; /* size of the entry we're seeking */
142 int numdirpasses; /* strategy for directory search */
143 doff_t endsearch; /* offset to end directory search */
144 doff_t prevoff; /* prev entry dp->i_offset */
145 struct vnode *pdp; /* saved dp during symlink work */
146 struct vnode *tdp; /* returned by VFS_VGET */
147 doff_t enduseful; /* pointer past last used dir slot */
148 u_long bmask; /* block offset mask */
149 int lockparent; /* 1 => lockparent flag is set */
150 int wantparent; /* 1 => wantparent or lockparent flag */
152 struct vnode **vpp = ap->a_vpp;
153 struct componentname *cnp = ap->a_cnp;
154 struct ucred *cred = cnp->cn_cred;
155 int flags = cnp->cn_flags;
156 int nameiop = cnp->cn_nameiop;
157 struct thread *td = cnp->cn_td;
158 globaldata_t gd = mycpu;
162 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
164 * XXX there was a soft-update diff about this I couldn't merge.
165 * I think this was the equiv.
171 lockparent = flags & CNP_LOCKPARENT;
172 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
175 * We now have a segment name to search for, and a directory to search.
177 * Suppress search for slots unless creating
178 * file and at end of pathname, in which case
179 * we watch for a place to put the new file in
180 * case it doesn't already exist.
183 slotfreespace = slotsize = slotneeded = 0;
184 if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) &&
185 (flags & CNP_ISLASTCN)) {
187 slotneeded = DIRECTSIZ(cnp->cn_namelen);
189 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
193 * Use dirhash for fast operations on large directories. The logic
194 * to determine whether to hash the directory is contained within
195 * ufsdirhash_build(); a zero return means that it decided to hash
196 * this directory and it successfully built up the hash table.
198 if (ufsdirhash_build(dp) == 0) {
199 /* Look for a free slot if needed. */
200 enduseful = dp->i_size;
201 if (slotstatus != FOUND) {
202 slotoffset = ufsdirhash_findfree(dp, slotneeded,
204 if (slotoffset >= 0) {
205 slotstatus = COMPACT;
206 enduseful = ufsdirhash_enduseful(dp);
208 enduseful = dp->i_size;
211 /* Look up the component. */
213 entryoffsetinblock = 0; /* silence compiler warning */
214 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
215 &dp->i_offset, &bp, nameiop == NAMEI_DELETE ? &prevoff : NULL)) {
217 ep = (struct direct *)((char *)bp->b_data +
218 (dp->i_offset & bmask));
221 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
224 /* Something failed; just do a linear search. */
228 #endif /* UFS_DIRHASH */
230 * If there is cached information on a previous search of
231 * this directory, pick up where we last left off.
232 * We cache only lookups as these are the most common
233 * and have the greatest payoff. Caching CREATE has little
234 * benefit as it usually must search the entire directory
235 * to determine that the entry does not exist. Caching the
236 * location of the last DELETE or RENAME has not reduced
237 * profiling time and hence has been removed in the interest
240 if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
241 dp->i_diroff >= dp->i_size) {
242 entryoffsetinblock = 0;
246 dp->i_offset = dp->i_diroff;
247 if ((entryoffsetinblock = dp->i_offset & bmask) &&
248 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
251 gd->gd_nchstats->ncs_2passes++;
253 prevoff = dp->i_offset;
254 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
258 while (dp->i_offset < endsearch) {
260 * If necessary, get the next directory block.
262 if ((dp->i_offset & bmask) == 0) {
266 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
269 entryoffsetinblock = 0;
272 * If still looking for a slot, and at a DIRBLKSIZE
273 * boundary, have to start looking for free space again.
275 if (slotstatus == NONE &&
276 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
281 * Get pointer to next entry.
282 * Full validation checks are slow, so we only check
283 * enough to insure forward progress through the
284 * directory. Complete checks can be run by patching
285 * "dirchk" to be true.
287 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
288 if (ep->d_reclen == 0 || ep->d_reclen >
289 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
290 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
293 ufs_dirbad(dp, dp->i_offset, "mangled entry");
294 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
296 entryoffsetinblock += i;
301 * If an appropriate sized slot has not yet been found,
302 * check to see if one is available. Also accumulate space
303 * in the current block so that we can determine if
304 * compaction is viable.
306 if (slotstatus != FOUND) {
307 int size = ep->d_reclen;
310 size -= DIRSIZ(OFSFMT(vdp), ep);
312 if (size >= slotneeded) {
314 slotoffset = dp->i_offset;
315 slotsize = ep->d_reclen;
316 } else if (slotstatus == NONE) {
317 slotfreespace += size;
318 if (slotoffset == -1)
319 slotoffset = dp->i_offset;
320 if (slotfreespace >= slotneeded) {
321 slotstatus = COMPACT;
322 slotsize = dp->i_offset +
323 ep->d_reclen - slotoffset;
330 * Check for a name match.
333 # if (BYTE_ORDER == LITTLE_ENDIAN)
337 namlen = ep->d_namlen;
339 namlen = ep->d_namlen;
341 if (namlen == cnp->cn_namelen &&
342 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
343 !bcmp(cnp->cn_nameptr, ep->d_name,
349 * Save directory entry's inode number and
350 * reclen in ndp->ni_ufs area, and release
353 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
354 ep->d_type == DT_WHT) {
356 slotoffset = dp->i_offset;
357 slotsize = ep->d_reclen;
358 dp->i_reclen = slotsize;
359 enduseful = dp->i_size;
360 ap->a_cnp->cn_flags |= CNP_ISWHITEOUT;
364 dp->i_ino = ep->d_ino;
365 dp->i_reclen = ep->d_reclen;
369 prevoff = dp->i_offset;
370 dp->i_offset += ep->d_reclen;
371 entryoffsetinblock += ep->d_reclen;
373 enduseful = dp->i_offset;
377 * If we started in the middle of the directory and failed
378 * to find our target, we must check the beginning as well.
380 if (numdirpasses == 2) {
383 endsearch = dp->i_diroff;
389 * If creating, and at end of pathname and current
390 * directory has not been removed, then can consider
391 * allowing file to be created.
393 if ((nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME ||
394 (nameiop == NAMEI_DELETE &&
395 (ap->a_cnp->cn_flags & CNP_DOWHITEOUT) &&
396 (ap->a_cnp->cn_flags & CNP_ISWHITEOUT))) &&
397 (flags & CNP_ISLASTCN) && dp->i_effnlink != 0) {
399 * Access for write is interpreted as allowing
400 * creation of files in the directory.
402 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
406 * Return an indication of where the new directory
407 * entry should be put. If we didn't find a slot,
408 * then set dp->i_count to 0 indicating
409 * that the new slot belongs at the end of the
410 * directory. If we found a slot, then the new entry
411 * can be put in the range from dp->i_offset to
412 * dp->i_offset + dp->i_count.
414 if (slotstatus == NONE) {
415 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
417 enduseful = dp->i_offset;
418 } else if (nameiop == NAMEI_DELETE) {
419 dp->i_offset = slotoffset;
420 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
423 dp->i_count = dp->i_offset - prevoff;
425 dp->i_offset = slotoffset;
426 dp->i_count = slotsize;
427 if (enduseful < slotoffset + slotsize)
428 enduseful = slotoffset + slotsize;
430 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
431 dp->i_flag |= IN_CHANGE | IN_UPDATE;
433 * We return with the directory locked, so that
434 * the parameters we set up above will still be
435 * valid if we actually decide to do a direnter().
436 * We return ni_vp == NULL to indicate that the entry
437 * does not currently exist; we leave a pointer to
438 * the (locked) directory inode in ndp->ni_dvp.
439 * The pathname buffer is saved so that the name
440 * can be obtained later.
442 * NB - if the directory is unlocked, then this
443 * information cannot be used.
445 cnp->cn_flags |= CNP_SAVENAME;
447 VOP_UNLOCK(vdp, NULL, 0, td);
448 cnp->cn_flags |= CNP_PDIRUNLOCK;
450 return (EJUSTRETURN);
453 * Insert name into cache (as non-existent) if appropriate.
455 if ((cnp->cn_flags & CNP_MAKEENTRY) && nameiop != NAMEI_CREATE)
456 cache_enter(vdp, NCPNULL, *vpp, cnp);
460 if (numdirpasses == 2)
461 gd->gd_nchstats->ncs_pass2++;
463 * Check that directory length properly reflects presence
466 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
467 ufs_dirbad(dp, dp->i_offset, "i_size too small");
468 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
469 dp->i_flag |= IN_CHANGE | IN_UPDATE;
474 * Found component in pathname.
475 * If the final component of path name, save information
476 * in the cache as to where the entry was found.
478 if ((flags & CNP_ISLASTCN) && nameiop == NAMEI_LOOKUP)
479 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
482 * If deleting, and at end of pathname, return
483 * parameters which can be used to remove file.
484 * If the wantparent flag isn't set, we return only
485 * the directory (in ndp->ni_dvp), otherwise we go
486 * on and lock the inode, being careful with ".".
488 if (nameiop == NAMEI_DELETE && (flags & CNP_ISLASTCN)) {
490 * Write access to directory required to delete files.
492 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
496 * Return pointer to current entry in dp->i_offset,
497 * and distance past previous entry (if there
498 * is a previous entry in this block) in dp->i_count.
499 * Save directory inode pointer in ndp->ni_dvp for dirremove().
501 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
504 dp->i_count = dp->i_offset - prevoff;
505 if (dp->i_number == dp->i_ino) {
510 if (flags & CNP_ISDOTDOT)
511 VOP_UNLOCK(vdp, NULL, 0, td);/* race to get the inode */
512 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
513 if (flags & CNP_ISDOTDOT) {
514 if (vn_lock(vdp, NULL, LK_EXCLUSIVE | LK_RETRY, td) != 0)
515 cnp->cn_flags |= CNP_PDIRUNLOCK;
520 * If directory is "sticky", then user must own
521 * the directory, or the file in it, else she
522 * may not delete it (unless she's root). This
523 * implements append-only directories.
525 if ((dp->i_mode & ISVTX) &&
527 cred->cr_uid != dp->i_uid &&
528 VTOI(tdp)->i_uid != cred->cr_uid) {
534 VOP_UNLOCK(vdp, NULL, 0, td);
535 cnp->cn_flags |= CNP_PDIRUNLOCK;
541 * If rewriting (RENAME), return the inode and the
542 * information required to rewrite the present directory
543 * Must get inode of directory entry to verify it's a
544 * regular file, or empty directory.
546 if (nameiop == NAMEI_RENAME && wantparent && (flags & CNP_ISLASTCN)) {
547 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
550 * Careful about locking second inode.
551 * This can only occur if the target is ".".
553 if (dp->i_number == dp->i_ino)
555 if (flags & CNP_ISDOTDOT)
556 VOP_UNLOCK(vdp, NULL, 0, td); /* race to get the inode */
557 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
558 if (flags & CNP_ISDOTDOT) {
559 if (vn_lock(vdp, NULL, LK_EXCLUSIVE | LK_RETRY, td) != 0)
560 cnp->cn_flags |= CNP_PDIRUNLOCK;
565 cnp->cn_flags |= CNP_SAVENAME;
567 VOP_UNLOCK(vdp, NULL, 0, td);
568 cnp->cn_flags |= CNP_PDIRUNLOCK;
574 * Step through the translation in the name. We do not `vput' the
575 * directory because we may need it again if a symbolic link
576 * is relative to the current directory. Instead we save it
577 * unlocked as "pdp". We must get the target inode before unlocking
578 * the directory to insure that the inode will not be removed
579 * before we get it. We prevent deadlock by always fetching
580 * inodes from the root, moving down the directory tree. Thus
581 * when following backward pointers ".." we must unlock the
582 * parent directory before getting the requested directory.
583 * There is a potential race condition here if both the current
584 * and parent directories are removed before the VFS_VGET for the
585 * inode associated with ".." returns. We hope that this occurs
586 * infrequently since we cannot avoid this race condition without
587 * implementing a sophisticated deadlock detection algorithm.
588 * Note also that this simple deadlock detection scheme will not
589 * work if the file system has any hard links other than ".."
590 * that point backwards in the directory structure.
593 if (flags & CNP_ISDOTDOT) {
594 VOP_UNLOCK(pdp, NULL, 0, td); /* race to get the inode */
595 cnp->cn_flags |= CNP_PDIRUNLOCK;
596 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
597 if (vn_lock(pdp, NULL, LK_EXCLUSIVE | LK_RETRY, td) == 0)
598 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
601 if (lockparent && (flags & CNP_ISLASTCN)) {
602 if ((error = vn_lock(pdp, NULL, LK_EXCLUSIVE, td)) != 0) {
606 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
609 } else if (dp->i_number == dp->i_ino) {
610 vref(vdp); /* we want ourself, ie "." */
613 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
616 if (!lockparent || !(flags & CNP_ISLASTCN)) {
617 VOP_UNLOCK(pdp, NULL, 0, td);
618 cnp->cn_flags |= CNP_PDIRUNLOCK;
624 * Insert name into cache if appropriate.
626 if (cnp->cn_flags & CNP_MAKEENTRY)
627 cache_enter(vdp, NCPNULL, *vpp, cnp);
632 ufs_dirbad(ip, offset, how)
639 mp = ITOV(ip)->v_mount;
640 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
641 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
642 if ((mp->mnt_flag & MNT_RDONLY) == 0)
643 panic("ufs_dirbad: bad dir");
647 * Do consistency checking on a directory entry:
648 * record length must be multiple of 4
649 * entry must fit in rest of its DIRBLKSIZ block
650 * record must be large enough to contain entry
651 * name is not longer than MAXNAMLEN
652 * name must be as long as advertised, and null terminated
655 ufs_dirbadentry(dp, ep, entryoffsetinblock)
658 int entryoffsetinblock;
663 # if (BYTE_ORDER == LITTLE_ENDIAN)
667 namlen = ep->d_namlen;
669 namlen = ep->d_namlen;
671 if ((ep->d_reclen & 0x3) != 0 ||
672 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
673 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
675 printf("First bad\n");
680 for (i = 0; i < namlen; i++)
681 if (ep->d_name[i] == '\0') {
683 printf("Second bad\n");
694 * Construct a new directory entry after a call to namei, using the
695 * parameters that it left in the componentname argument cnp. The
696 * argument ip is the inode to which the new directory entry will refer.
699 ufs_makedirentry(ip, cnp, newdirp)
701 struct componentname *cnp;
702 struct direct *newdirp;
706 if ((cnp->cn_flags & CNP_SAVENAME) == 0)
707 panic("ufs_makedirentry: missing name");
709 newdirp->d_ino = ip->i_number;
710 newdirp->d_namlen = cnp->cn_namelen;
711 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
712 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
713 newdirp->d_type = IFTODT(ip->i_mode);
716 # if (BYTE_ORDER == LITTLE_ENDIAN)
717 { u_char tmp = newdirp->d_namlen;
718 newdirp->d_namlen = newdirp->d_type;
719 newdirp->d_type = tmp; }
725 * Write a directory entry after a call to namei, using the parameters
726 * that it left in nameidata. The argument dirp is the new directory
727 * entry contents. Dvp is a pointer to the directory to be written,
728 * which was left locked by namei. Remaining parameters (dp->i_offset,
729 * dp->i_count) indicate how the space for the new entry is to be obtained.
730 * Non-null bp indicates that a directory is being created (for the
731 * soft dependency code).
734 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
738 struct componentname *cnp;
739 struct buf *newdirbp;
742 struct thread *td = curthread; /* XXX */
747 struct direct *ep, *nep;
748 int error, ret, blkoff, loc, spacefree, flags;
751 KKASSERT(td->td_proc); /* YYY use/require cred passed in cnp? */
752 cred = td->td_proc->p_ucred;
755 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
757 if (dp->i_count == 0) {
759 * If dp->i_count is 0, then namei could find no
760 * space in the directory. Here, dp->i_offset will
761 * be on a directory block boundary and we will write the
762 * new entry into a fresh block.
764 if (dp->i_offset & (DIRBLKSIZ - 1))
765 panic("ufs_direnter: newblk");
767 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
769 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
770 cred, flags, &bp)) != 0) {
771 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
775 dp->i_size = dp->i_offset + DIRBLKSIZ;
776 dp->i_flag |= IN_CHANGE | IN_UPDATE;
777 vnode_pager_setsize(dvp, (u_long)dp->i_size);
778 dirp->d_reclen = DIRBLKSIZ;
779 blkoff = dp->i_offset &
780 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
781 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
783 if (dp->i_dirhash != NULL) {
784 ufsdirhash_newblk(dp, dp->i_offset);
785 ufsdirhash_add(dp, dirp, dp->i_offset);
786 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
790 if (DOINGSOFTDEP(dvp)) {
792 * Ensure that the entire newly allocated block is a
793 * valid directory so that future growth within the
794 * block does not have to ensure that the block is
795 * written before the inode.
798 while (blkoff < bp->b_bcount) {
800 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
803 softdep_setup_directory_add(bp, dp, dp->i_offset,
804 dirp->d_ino, newdirbp);
806 return (UFS_UPDATE(dvp, 0));
808 if (DOINGASYNC(dvp)) {
810 return (UFS_UPDATE(dvp, 0));
812 error = VOP_BWRITE(bp->b_vp, bp);
813 ret = UFS_UPDATE(dvp, 1);
820 * If dp->i_count is non-zero, then namei found space for the new
821 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
822 * in the directory. To use this space, we may have to compact
823 * the entries located there, by copying them together towards the
824 * beginning of the block, leaving the free space in one usable
829 * Increase size of directory if entry eats into new space.
830 * This should never push the size past a new multiple of
833 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
835 if (dp->i_offset + dp->i_count > dp->i_size)
836 dp->i_size = dp->i_offset + dp->i_count;
838 * Get the block containing the space for the new directory entry.
840 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
842 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
847 * Find space for the new entry. In the simple case, the entry at
848 * offset base will have the space. If it does not, then namei
849 * arranged that compacting the region dp->i_offset to
850 * dp->i_offset + dp->i_count would yield the space.
852 ep = (struct direct *)dirbuf;
853 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
854 spacefree = ep->d_reclen - dsize;
855 for (loc = ep->d_reclen; loc < dp->i_count; ) {
856 nep = (struct direct *)(dirbuf + loc);
858 /* Trim the existing slot (NB: dsize may be zero). */
859 ep->d_reclen = dsize;
860 ep = (struct direct *)((char *)ep + dsize);
862 /* Read nep->d_reclen now as the bcopy() may clobber it. */
863 loc += nep->d_reclen;
864 if (nep->d_ino == 0) {
866 * A mid-block unused entry. Such entries are
867 * never created by the kernel, but fsck_ffs
868 * can create them (and it doesn't fix them).
870 * Add up the free space, and initialise the
871 * relocated entry since we don't bcopy it.
873 spacefree += nep->d_reclen;
878 dsize = DIRSIZ(OFSFMT(dvp), nep);
879 spacefree += nep->d_reclen - dsize;
881 if (dp->i_dirhash != NULL)
882 ufsdirhash_move(dp, nep,
883 dp->i_offset + ((char *)nep - dirbuf),
884 dp->i_offset + ((char *)ep - dirbuf));
886 if (DOINGSOFTDEP(dvp))
887 softdep_change_directoryentry_offset(dp, dirbuf,
888 (caddr_t)nep, (caddr_t)ep, dsize);
890 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
893 * Here, `ep' points to a directory entry containing `dsize' in-use
894 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
895 * then the entry is completely unused (dsize == 0). The value
896 * of ep->d_reclen is always indeterminate.
898 * Update the pointer fields in the previous entry (if any),
899 * copy in the new entry, and write out the block.
901 if (ep->d_ino == 0 ||
902 (ep->d_ino == WINO &&
903 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
904 if (spacefree + dsize < newentrysize)
905 panic("ufs_direnter: compact1");
906 dirp->d_reclen = spacefree + dsize;
908 if (spacefree < newentrysize)
909 panic("ufs_direnter: compact2");
910 dirp->d_reclen = spacefree;
911 ep->d_reclen = dsize;
912 ep = (struct direct *)((char *)ep + dsize);
915 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
916 dirp->d_reclen == spacefree))
917 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
919 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
921 if (dp->i_dirhash != NULL)
922 ufsdirhash_checkblock(dp, dirbuf -
923 (dp->i_offset & (DIRBLKSIZ - 1)),
924 dp->i_offset & ~(DIRBLKSIZ - 1));
927 if (DOINGSOFTDEP(dvp)) {
928 softdep_setup_directory_add(bp, dp,
929 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
932 if (DOINGASYNC(dvp)) {
939 dp->i_flag |= IN_CHANGE | IN_UPDATE;
941 * If all went well, and the directory can be shortened, proceed
942 * with the truncation. Note that we have to unlock the inode for
943 * the entry that we just entered, as the truncation may need to
944 * lock other inodes which can lead to deadlock if we also hold a
945 * lock on the newly entered node.
947 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
949 VOP_UNLOCK(tvp, NULL, 0, td);
951 if (dp->i_dirhash != NULL)
952 ufsdirhash_dirtrunc(dp, dp->i_endoff);
954 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cred, td);
956 vn_lock(tvp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
962 * Remove a directory entry after a call to namei, using
963 * the parameters which it left in nameidata. The entry
964 * dp->i_offset contains the offset into the directory of the
965 * entry to be eliminated. The dp->i_count field contains the
966 * size of the previous record in the directory. If this
967 * is 0, the first entry is being deleted, so we need only
968 * zero the inode number to mark the entry as free. If the
969 * entry is not the first in the directory, we must reclaim
970 * the space of the now empty record by adding the record size
971 * to the size of the previous entry.
974 ufs_dirremove(dvp, ip, flags, isrmdir)
987 if (flags & CNP_DOWHITEOUT) {
989 * Whiteout entry: set d_ino to WINO.
992 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
999 if ((error = UFS_BLKATOFF(dvp,
1000 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
1004 * Remove the dirhash entry. This is complicated by the fact
1005 * that `ep' is the previous entry when dp->i_count != 0.
1007 if (dp->i_dirhash != NULL)
1008 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
1009 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
1011 if (dp->i_count == 0) {
1013 * First entry in block: set d_ino to zero.
1018 * Collapse new free space into previous entry.
1020 ep->d_reclen += dp->i_reclen;
1023 if (dp->i_dirhash != NULL)
1024 ufsdirhash_checkblock(dp, (char *)ep -
1025 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1026 dp->i_offset & ~(DIRBLKSIZ - 1));
1029 if (DOINGSOFTDEP(dvp)) {
1032 softdep_change_linkcnt(ip);
1033 softdep_setup_remove(bp, dp, ip, isrmdir);
1035 if (softdep_slowdown(dvp)) {
1036 error = VOP_BWRITE(bp->b_vp, bp);
1045 ip->i_flag |= IN_CHANGE;
1047 if (flags & CNP_DOWHITEOUT)
1048 error = VOP_BWRITE(bp->b_vp, bp);
1049 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1053 error = bowrite(bp);
1055 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1060 * Rewrite an existing directory entry to point at the inode
1061 * supplied. The parameters describing the directory entry are
1062 * set up by a call to namei.
1065 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1066 struct inode *dp, *oip;
1073 struct vnode *vdp = ITOV(dp);
1076 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1079 ep->d_ino = newinum;
1081 ep->d_type = newtype;
1083 if (DOINGSOFTDEP(vdp)) {
1084 softdep_change_linkcnt(oip);
1085 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1089 oip->i_flag |= IN_CHANGE;
1090 if (DOINGASYNC(vdp)) {
1094 error = bowrite(bp);
1097 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1102 * Check if a directory is empty or not.
1103 * Inode supplied must be locked.
1105 * Using a struct dirtemplate here is not precisely
1106 * what we want, but better than using a struct direct.
1108 * NB: does not handle corrupted directories.
1111 ufs_dirempty(ip, parentino, cred)
1117 struct dirtemplate dbuf;
1118 struct direct *dp = (struct direct *)&dbuf;
1119 int error, count, namlen;
1120 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1122 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1123 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
1124 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
1126 * Since we read MINDIRSIZ, residual must
1127 * be 0 unless we're at end of file.
1129 if (error || count != 0)
1131 /* avoid infinite loops */
1132 if (dp->d_reclen == 0)
1134 /* skip empty entries */
1135 if (dp->d_ino == 0 || dp->d_ino == WINO)
1137 /* accept only "." and ".." */
1138 # if (BYTE_ORDER == LITTLE_ENDIAN)
1139 if (OFSFMT(ITOV(ip)))
1140 namlen = dp->d_type;
1142 namlen = dp->d_namlen;
1144 namlen = dp->d_namlen;
1148 if (dp->d_name[0] != '.')
1151 * At this point namlen must be 1 or 2.
1152 * 1 implies ".", 2 implies ".." if second
1155 if (namlen == 1 && dp->d_ino == ip->i_number)
1157 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1165 * Check if source directory is in the path of the target directory.
1166 * Target is supplied locked, source is unlocked.
1167 * The target is always vput before returning.
1170 ufs_checkpath(source, target, cred)
1171 struct inode *source, *target;
1175 int error, rootino, namlen;
1176 struct dirtemplate dirbuf;
1179 if (target->i_number == source->i_number) {
1185 if (target->i_number == rootino)
1189 if (vp->v_type != VDIR) {
1193 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1194 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1195 IO_NODELOCKED, cred, (int *)0, NULL);
1198 # if (BYTE_ORDER == LITTLE_ENDIAN)
1200 namlen = dirbuf.dotdot_type;
1202 namlen = dirbuf.dotdot_namlen;
1204 namlen = dirbuf.dotdot_namlen;
1207 dirbuf.dotdot_name[0] != '.' ||
1208 dirbuf.dotdot_name[1] != '.') {
1212 if (dirbuf.dotdot_ino == source->i_number) {
1216 if (dirbuf.dotdot_ino == rootino)
1219 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1227 if (error == ENOTDIR)
1228 printf("checkpath: .. not a directory\n");