static int
mkentry(struct inodesc *idesc)
{
LFS_DIRHEADER *dirp = idesc->id_dirp;
unsigned namlen;
unsigned newreclen, oldreclen;
/* figure the length needed for id_name */
namlen = strlen(idesc->id_name);
newreclen = LFS_DIRECTSIZ(fs, namlen);
/* find the minimum record length for the existing name */
if (lfs_dir_getino(fs, dirp) != 0)
oldreclen = LFS_DIRSIZ(fs, dirp);
else
oldreclen = 0;
/* Can we insert here? */
if (lfs_dir_getreclen(fs, dirp) - oldreclen < newreclen)
return (KEEPON);
/* Divide the record; all but oldreclen goes to the new record */
newreclen = lfs_dir_getreclen(fs, dirp) - oldreclen;
lfs_dir_setreclen(fs, dirp, oldreclen);
/* advance the pointer to the new record */
dirp = LFS_NEXTDIR(fs, dirp);
/* write record; ino to be entered is in id_parent */
lfs_dir_setino(fs, dirp, idesc->id_parent);
lfs_dir_setreclen(fs, dirp, newreclen);
lfs_dir_settype(fs, dirp, typemap[idesc->id_parent]);
lfs_dir_setnamlen(fs, dirp, namlen);
lfs_copydirname(fs, lfs_dir_nameptr(fs, dirp), idesc->id_name,
namlen, newreclen);
return (ALTERED | STOP);
}
/*
* allocate a new directory
*/
int
allocdir(ino_t parent, ino_t request, int mode)
{
ino_t ino;
char *cp;
union lfs_dinode *dp;
struct ubuf *bp;
LFS_DIRHEADER *dirp;
struct uvnode *vp;
ino = allocino(request, LFS_IFDIR | mode);
vp = vget(fs, ino);
dp = VTOD(vp);
bread(vp, lfs_dino_getdb(fs, dp, 0), lfs_sb_getfsize(fs), 0, &bp);
if (bp->b_flags & B_ERROR) {
brelse(bp, 0);
freeino(ino);
return (0);
}
dirp = (LFS_DIRHEADER *)bp->b_data;
/* . */
lfs_dir_setino(fs, dirp, ino);
lfs_dir_setreclen(fs, dirp, LFS_DIRECTSIZ(fs, 1));
lfs_dir_settype(fs, dirp, LFS_DT_DIR);
lfs_dir_setnamlen(fs, dirp, 1);
lfs_copydirname(fs, lfs_dir_nameptr(fs, dirp), ".", 1,
LFS_DIRECTSIZ(fs, 1));
/* .. */
dirp = LFS_NEXTDIR(fs, dirp);
lfs_dir_setino(fs, dirp, parent);
lfs_dir_setreclen(fs, dirp, LFS_DIRBLKSIZ - LFS_DIRECTSIZ(fs, 1));
lfs_dir_settype(fs, dirp, LFS_DT_DIR);
lfs_dir_setnamlen(fs, dirp, 2);
lfs_copydirname(fs, lfs_dir_nameptr(fs, dirp), "..", 2,
LFS_DIRBLKSIZ - LFS_DIRECTSIZ(fs, 1));
for (cp = &bp->b_data[LFS_DIRBLKSIZ];
cp < &bp->b_data[lfs_sb_getfsize(fs)];
cp += LFS_DIRBLKSIZ) {
zerodirblk(cp);
}
VOP_BWRITE(bp);
lfs_dino_setnlink(fs, dp, 2);
inodirty(VTOI(vp));
if (ino == ULFS_ROOTINO) {
lncntp[ino] = lfs_dino_getnlink(fs, dp);
cacheino(dp, ino);
return (ino);
}
if (statemap[parent] != DSTATE && statemap[parent] != DFOUND) {
freeino(ino);
return (0);
}
cacheino(dp, ino);
statemap[ino] = statemap[parent];
if (statemap[ino] == DSTATE) {
lncntp[ino] = lfs_dino_getnlink(fs, dp);
lncntp[parent]++;
}
vp = vget(fs, parent);
dp = VTOD(vp);
lfs_dino_setnlink(fs, dp, lfs_dino_getnlink(fs, dp) + 1);
inodirty(VTOI(vp));
return (ino);
}
/*
* Convert a component of a pathname into a pointer to a locked inode.
* This is a very central and rather complicated routine.
* If the file system is not maintained in a strict tree hierarchy,
* this can result in a deadlock situation (see comments in code below).
*
* The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
* on whether the name is to be looked up, created, renamed, or deleted.
* When CREATE, RENAME, or DELETE is specified, information usable in
* creating, renaming, or deleting a directory entry may be calculated.
* If flag has LOCKPARENT or'ed into it and the target of the pathname
* exists, lookup returns both the target and its parent directory locked.
* When creating or renaming and LOCKPARENT is specified, the target may
* not be ".". When deleting and LOCKPARENT is specified, the target may
* be "."., but the caller must check to ensure it does an vrele and vput
* instead of two vputs.
*
* Overall outline of ulfs_lookup:
*
* check accessibility of directory
* look for name in cache, if found, then if at end of path
* and deleting or creating, drop it, else return name
* search for name in directory, to found or notfound
* notfound:
* if creating, return locked directory, leaving info on available slots
* else return error
* found:
* if at end of path and deleting, return information to allow delete
* if at end of path and rewriting (RENAME and LOCKPARENT), lock target
* inode and return info to allow rewrite
* if not at end, add name to cache; if at end and neither creating
* nor deleting, add name to cache
*/
int
ulfs_lookup(void *v)
{
struct vop_lookup_v2_args /* {
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
} */ *ap = v;
struct vnode *vdp = ap->a_dvp; /* vnode for directory being searched */
struct inode *dp = VTOI(vdp); /* inode for directory being searched */
struct buf *bp; /* a buffer of directory entries */
struct lfs_direct *ep; /* the current directory entry */
int entryoffsetinblock; /* offset of ep in bp's buffer */
enum {
NONE, /* need to search a slot for our new entry */
COMPACT, /* a compaction can make a slot in the current
DIRBLKSIZ block */
FOUND, /* found a slot (or no need to search) */
} slotstatus;
doff_t slotoffset; /* offset of area with free space.
a special value -1 for invalid */
int slotsize; /* size of area at slotoffset */
int slotfreespace; /* accumulated amount of space free in
the current DIRBLKSIZ block */
int slotneeded; /* size of the entry we're seeking */
int numdirpasses; /* strategy for directory search */
doff_t endsearch; /* offset to end directory search */
doff_t prevoff; /* previous value of ulr_offset */
struct vnode *pdp; /* saved dp during symlink work */
struct vnode *tdp; /* returned by VFS_VGET */
doff_t enduseful; /* pointer past last used dir slot.
used for directory truncation. */
u_long bmask; /* block offset mask */
int error;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
kauth_cred_t cred = cnp->cn_cred;
int flags;
int nameiop = cnp->cn_nameiop;
struct lfs *fs = dp->i_lfs;
const int needswap = ULFS_MPNEEDSWAP(fs);
int dirblksiz = fs->um_dirblksiz;
ino_t foundino;
struct ulfs_lookup_results *results;
int iswhiteout; /* temp result from cache_lookup() */
flags = cnp->cn_flags;
bp = NULL;
slotoffset = -1;
*vpp = NULL;
endsearch = 0; /* silence compiler warning */
/*
* Produce the auxiliary lookup results into i_crap. Increment
* its serial number so elsewhere we can tell if we're using
* stale results. This should not be done this way. XXX.
*/
results = &dp->i_crap;
dp->i_crapcounter++;
/*
* Check accessiblity of directory.
*/
if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
return (error);
if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
(nameiop == DELETE || nameiop == RENAME))
return (EROFS);
/*
* We now have a segment name to search for, and a directory to search.
*
* Before tediously performing a linear scan of the directory,
* check the name cache to see if the directory/name pair
* we are looking for is known already.
*/
if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
cnp->cn_nameiop, cnp->cn_flags, &iswhiteout, vpp)) {
if (iswhiteout) {
cnp->cn_flags |= ISWHITEOUT;
}
return *vpp == NULLVP ? ENOENT : 0;
}
if (iswhiteout) {
/*
* The namecache set iswhiteout without finding a
* cache entry. As of this writing (20121014), this
* can happen if there was a whiteout entry that has
* been invalidated by the lookup. It is not clear if
* it is correct to set ISWHITEOUT in this case or
* not; however, doing so retains the prior behavior,
* so we'll go with that until some clearer answer
* appears. XXX
*/
cnp->cn_flags |= ISWHITEOUT;
}
fstrans_start(vdp->v_mount, FSTRANS_SHARED);
/*
* Suppress search for slots unless creating
* file and at end of pathname, in which case
* we watch for a place to put the new file in
* case it doesn't already exist.
*/
slotstatus = FOUND;
slotfreespace = slotsize = slotneeded = 0;
if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
slotstatus = NONE;
slotneeded = LFS_DIRECTSIZ(cnp->cn_namelen);
}
/*
* If there is cached information on a previous search of
* this directory, pick up where we last left off.
* We cache only lookups as these are the most common
* and have the greatest payoff. Caching CREATE has little
* benefit as it usually must search the entire directory
* to determine that the entry does not exist. Caching the
* location of the last DELETE or RENAME has not reduced
* profiling time and hence has been removed in the interest
* of simplicity.
*/
bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
#ifdef LFS_DIRHASH
/*
* Use dirhash for fast operations on large directories. The logic
* to determine whether to hash the directory is contained within
* ulfsdirhash_build(); a zero return means that it decided to hash
* this directory and it successfully built up the hash table.
*/
if (ulfsdirhash_build(dp) == 0) {
/* Look for a free slot if needed. */
enduseful = dp->i_size;
if (slotstatus != FOUND) {
slotoffset = ulfsdirhash_findfree(dp, slotneeded,
&slotsize);
if (slotoffset >= 0) {
slotstatus = COMPACT;
enduseful = ulfsdirhash_enduseful(dp);
if (enduseful < 0)
enduseful = dp->i_size;
}
}
/* Look up the component. */
numdirpasses = 1;
entryoffsetinblock = 0; /* silence compiler warning */
switch (ulfsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
&results->ulr_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
case 0:
ep = (struct lfs_direct *)((char *)bp->b_data +
(results->ulr_offset & bmask));
goto foundentry;
case ENOENT:
results->ulr_offset = roundup(dp->i_size, dirblksiz);
goto notfound;
default:
/* Something failed; just do a linear search. */
break;
}
}
#endif /* LFS_DIRHASH */
if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
results->ulr_diroff >= dp->i_size) {
entryoffsetinblock = 0;
results->ulr_offset = 0;
numdirpasses = 1;
} else {
results->ulr_offset = results->ulr_diroff;
if ((entryoffsetinblock = results->ulr_offset & bmask) &&
(error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
NULL, &bp, false)))
goto out;
numdirpasses = 2;
namecache_count_2passes();
}
prevoff = results->ulr_offset;
endsearch = roundup(dp->i_size, dirblksiz);
enduseful = 0;
searchloop:
while (results->ulr_offset < endsearch) {
if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
preempt();
/*
* If necessary, get the next directory block.
*/
if ((results->ulr_offset & bmask) == 0) {
if (bp != NULL)
brelse(bp, 0);
error = ulfs_blkatoff(vdp, (off_t)results->ulr_offset,
NULL, &bp, false);
if (error)
goto out;
entryoffsetinblock = 0;
}
/*
* If still looking for a slot, and at a DIRBLKSIZ
* boundary, have to start looking for free space again.
*/
if (slotstatus == NONE &&
(entryoffsetinblock & (dirblksiz - 1)) == 0) {
slotoffset = -1;
slotfreespace = 0;
}
/*
* Get pointer to next entry.
* Full validation checks are slow, so we only check
* enough to insure forward progress through the
* directory. Complete checks can be run by patching
* "lfs_dirchk" to be true.
*/
KASSERT(bp != NULL);
ep = (struct lfs_direct *)((char *)bp->b_data + entryoffsetinblock);
if (ep->d_reclen == 0 ||
(lfs_dirchk && ulfs_dirbadentry(vdp, ep, entryoffsetinblock))) {
int i;
ulfs_dirbad(dp, results->ulr_offset, "mangled entry");
i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
results->ulr_offset += i;
entryoffsetinblock += i;
continue;
}
/*
* If an appropriate sized slot has not yet been found,
* check to see if one is available. Also accumulate space
* in the current block so that we can determine if
* compaction is viable.
*/
if (slotstatus != FOUND) {
int size = ulfs_rw16(ep->d_reclen, needswap);
if (ep->d_ino != 0)
size -= LFS_DIRSIZ(FSFMT(vdp), ep, needswap);
if (size > 0) {
if (size >= slotneeded) {
slotstatus = FOUND;
slotoffset = results->ulr_offset;
slotsize = ulfs_rw16(ep->d_reclen,
needswap);
} else if (slotstatus == NONE) {
slotfreespace += size;
if (slotoffset == -1)
slotoffset = results->ulr_offset;
if (slotfreespace >= slotneeded) {
slotstatus = COMPACT;
slotsize = results->ulr_offset +
ulfs_rw16(ep->d_reclen,
needswap) -
slotoffset;
}
}
}
}
/*
* Check for a name match.
*/
if (ep->d_ino) {
int namlen;
#if (BYTE_ORDER == LITTLE_ENDIAN)
if (FSFMT(vdp) && needswap == 0)
namlen = ep->d_type;
else
namlen = ep->d_namlen;
#else
if (FSFMT(vdp) && needswap != 0)
namlen = ep->d_type;
else
namlen = ep->d_namlen;
#endif
if (namlen == cnp->cn_namelen &&
!memcmp(cnp->cn_nameptr, ep->d_name,
(unsigned)namlen)) {
#ifdef LFS_DIRHASH
foundentry:
#endif
/*
* Save directory entry's inode number and
* reclen, and release directory buffer.
*/
if (!FSFMT(vdp) && ep->d_type == LFS_DT_WHT) {
slotstatus = FOUND;
slotoffset = results->ulr_offset;
slotsize = ulfs_rw16(ep->d_reclen,
needswap);
results->ulr_reclen = slotsize;
/*
* This is used to set
* results->ulr_endoff,
* which may be used by ulfs_direnter()
* as a length to truncate the
* directory to. Therefore, it must
* point past the end of the last
* non-empty directory entry. We don't
* know where that is in this case, so
* we effectively disable shrinking by
* using the existing size of the
* directory.
*
* Note that we wouldn't expect to
* shrink the directory while rewriting
* an existing entry anyway.
*/
enduseful = endsearch;
cnp->cn_flags |= ISWHITEOUT;
numdirpasses--;
goto notfound;
}
foundino = ulfs_rw32(ep->d_ino, needswap);
results->ulr_reclen =
ulfs_rw16(ep->d_reclen, needswap);
goto found;
}
}
prevoff = results->ulr_offset;
results->ulr_offset += ulfs_rw16(ep->d_reclen, needswap);
entryoffsetinblock += ulfs_rw16(ep->d_reclen, needswap);
if (ep->d_ino)
enduseful = results->ulr_offset;
}
notfound:
/*
* If we started in the middle of the directory and failed
* to find our target, we must check the beginning as well.
*/
if (numdirpasses == 2) {
numdirpasses--;
results->ulr_offset = 0;
endsearch = results->ulr_diroff;
goto searchloop;
}
if (bp != NULL)
brelse(bp, 0);
/*
* If creating, and at end of pathname and current
* directory has not been removed, then can consider
* allowing file to be created.
*/
if ((nameiop == CREATE || nameiop == RENAME ||
(nameiop == DELETE &&
(cnp->cn_flags & DOWHITEOUT) &&
(cnp->cn_flags & ISWHITEOUT))) &&
(flags & ISLASTCN) && dp->i_nlink != 0) {
/*
* Access for write is interpreted as allowing
* creation of files in the directory.
*/
error = VOP_ACCESS(vdp, VWRITE, cred);
if (error)
goto out;
/*
* Return an indication of where the new directory
* entry should be put. If we didn't find a slot,
* then set results->ulr_count to 0 indicating
* that the new slot belongs at the end of the
* directory. If we found a slot, then the new entry
* can be put in the range from results->ulr_offset to
* results->ulr_offset + results->ulr_count.
*/
if (slotstatus == NONE) {
results->ulr_offset = roundup(dp->i_size, dirblksiz);
results->ulr_count = 0;
enduseful = results->ulr_offset;
} else if (nameiop == DELETE) {
results->ulr_offset = slotoffset;
if ((results->ulr_offset & (dirblksiz - 1)) == 0)
results->ulr_count = 0;
else
results->ulr_count =
results->ulr_offset - prevoff;
} else {
results->ulr_offset = slotoffset;
results->ulr_count = slotsize;
if (enduseful < slotoffset + slotsize)
enduseful = slotoffset + slotsize;
}
results->ulr_endoff = roundup(enduseful, dirblksiz);
#if 0 /* commented out by dbj. none of the on disk fields changed */
dp->i_flag |= IN_CHANGE | IN_UPDATE;
#endif
/*
* We return with the directory locked, so that
* the parameters we set up above will still be
* valid if we actually decide to do a direnter().
* We return ni_vp == NULL to indicate that the entry
* does not currently exist; we leave a pointer to
* the (locked) directory inode in ndp->ni_dvp.
*
* NB - if the directory is unlocked, then this
* information cannot be used.
*/
error = EJUSTRETURN;
goto out;
}
/*
* Insert name into cache (as non-existent) if appropriate.
*/
if (nameiop != CREATE) {
cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
cnp->cn_flags);
}
error = ENOENT;
goto out;
found:
if (numdirpasses == 2)
namecache_count_pass2();
/*
* Check that directory length properly reflects presence
* of this entry.
*/
if (results->ulr_offset + LFS_DIRSIZ(FSFMT(vdp), ep, needswap) > dp->i_size) {
ulfs_dirbad(dp, results->ulr_offset, "i_size too small");
dp->i_size =
results->ulr_offset + LFS_DIRSIZ(FSFMT(vdp), ep, needswap);
DIP_ASSIGN(dp, size, dp->i_size);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
}
brelse(bp, 0);
/*
* Found component in pathname.
* If the final component of path name, save information
* in the cache as to where the entry was found.
*/
if ((flags & ISLASTCN) && nameiop == LOOKUP)
results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
/*
* If deleting, and at end of pathname, return
* parameters which can be used to remove file.
* Lock the inode, being careful with ".".
*/
if (nameiop == DELETE && (flags & ISLASTCN)) {
/*
* Return pointer to current entry in results->ulr_offset,
* and distance past previous entry (if there
* is a previous entry in this block) in results->ulr_count.
* Save directory inode pointer in ndp->ni_dvp for dirremove().
*/
if ((results->ulr_offset & (dirblksiz - 1)) == 0)
results->ulr_count = 0;
else
results->ulr_count = results->ulr_offset - prevoff;
if (dp->i_number == foundino) {
vref(vdp);
tdp = vdp;
} else {
if (flags & ISDOTDOT)
VOP_UNLOCK(vdp); /* race to get the inode */
error = VFS_VGET(vdp->v_mount, foundino, &tdp);
if (flags & ISDOTDOT)
vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
if (error)
goto out;
}
/*
* Write access to directory required to delete files.
*/
error = VOP_ACCESS(vdp, VWRITE, cred);
if (error) {
if (dp->i_number == foundino)
vrele(tdp);
else
vput(tdp);
goto out;
}
/*
* If directory is "sticky", then user must own
* the directory, or the file in it, else she
* may not delete it (unless she's root). This
* implements append-only directories.
*/
if (dp->i_mode & ISVTX) {
error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
VTOI(tdp)->i_uid));
if (error) {
if (dp->i_number == foundino)
vrele(tdp);
else
vput(tdp);
error = EPERM;
goto out;
}
}
*vpp = tdp;
error = 0;
goto out;
}
/*
* If rewriting (RENAME), return the inode and the
* information required to rewrite the present directory
* Must get inode of directory entry to verify it's a
* regular file, or empty directory.
*/
if (nameiop == RENAME && (flags & ISLASTCN)) {
error = VOP_ACCESS(vdp, VWRITE, cred);
if (error)
goto out;
/*
* Careful about locking second inode.
* This can only occur if the target is ".".
*/
if (dp->i_number == foundino) {
error = EISDIR;
goto out;
}
if (flags & ISDOTDOT)
VOP_UNLOCK(vdp); /* race to get the inode */
error = VFS_VGET(vdp->v_mount, foundino, &tdp);
if (flags & ISDOTDOT)
vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
if (error)
goto out;
*vpp = tdp;
error = 0;
goto out;
}
/*
* Step through the translation in the name. We do not `vput' the
* directory because we may need it again if a symbolic link
* is relative to the current directory. Instead we save it
* unlocked as "pdp". We must get the target inode before unlocking
* the directory to insure that the inode will not be removed
* before we get it. We prevent deadlock by always fetching
* inodes from the root, moving down the directory tree. Thus
* when following backward pointers ".." we must unlock the
* parent directory before getting the requested directory.
* There is a potential race condition here if both the current
* and parent directories are removed before the VFS_VGET for the
* inode associated with ".." returns. We hope that this occurs
* infrequently since we cannot avoid this race condition without
* implementing a sophisticated deadlock detection algorithm.
* Note also that this simple deadlock detection scheme will not
* work if the file system has any hard links other than ".."
* that point backwards in the directory structure.
*/
pdp = vdp;
if (flags & ISDOTDOT) {
VOP_UNLOCK(pdp); /* race to get the inode */
error = VFS_VGET(vdp->v_mount, foundino, &tdp);
vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
if (error) {
goto out;
}
*vpp = tdp;
} else if (dp->i_number == foundino) {
vref(vdp); /* we want ourself, ie "." */
*vpp = vdp;
} else {
error = VFS_VGET(vdp->v_mount, foundino, &tdp);
if (error)
goto out;
*vpp = tdp;
}
/*
* Insert name into cache if appropriate.
*/
cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
error = 0;
out:
if (error == 0 && *vpp != vdp)
VOP_UNLOCK(*vpp);
fstrans_done(vdp->v_mount);
return error;
}
/*
* Vnode op for reading directories.
*
* This routine handles converting from the on-disk directory format
* "struct lfs_direct" to the in-memory format "struct dirent" as well as
* byte swapping the entries if necessary.
*/
int
ulfs_readdir(void *v)
{
struct vop_readdir_args /* {
struct vnode *a_vp;
struct uio *a_uio;
kauth_cred_t a_cred;
int *a_eofflag;
off_t **a_cookies;
int *ncookies;
} */ *ap = v;
struct vnode *vp = ap->a_vp;
LFS_DIRHEADER *cdp, *ecdp;
struct dirent *ndp;
char *cdbuf, *ndbuf, *endp;
struct uio auio, *uio;
struct iovec aiov;
int error;
size_t count, ccount, rcount, cdbufsz, ndbufsz;
off_t off, *ccp;
off_t startoff;
size_t skipbytes;
struct ulfsmount *ump = VFSTOULFS(vp->v_mount);
struct lfs *fs = ump->um_lfs;
uio = ap->a_uio;
count = uio->uio_resid;
rcount = count - ((uio->uio_offset + count) & (fs->um_dirblksiz - 1));
if (rcount < LFS_DIRECTSIZ(fs, 0) || count < _DIRENT_MINSIZE(ndp))
return EINVAL;
startoff = uio->uio_offset & ~(fs->um_dirblksiz - 1);
skipbytes = uio->uio_offset - startoff;
rcount += skipbytes;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = startoff;
auio.uio_resid = rcount;
UIO_SETUP_SYSSPACE(&auio);
auio.uio_rw = UIO_READ;
cdbufsz = rcount;
cdbuf = kmem_alloc(cdbufsz, KM_SLEEP);
aiov.iov_base = cdbuf;
aiov.iov_len = rcount;
error = VOP_READ(vp, &auio, 0, ap->a_cred);
if (error != 0) {
kmem_free(cdbuf, cdbufsz);
return error;
}
rcount -= auio.uio_resid;
cdp = (LFS_DIRHEADER *)(void *)cdbuf;
ecdp = (LFS_DIRHEADER *)(void *)&cdbuf[rcount];
ndbufsz = count;
ndbuf = kmem_alloc(ndbufsz, KM_SLEEP);
ndp = (struct dirent *)(void *)ndbuf;
endp = &ndbuf[count];
off = uio->uio_offset;
if (ap->a_cookies) {
ccount = rcount / _DIRENT_RECLEN(ndp, 1);
ccp = *(ap->a_cookies) = malloc(ccount * sizeof(*ccp),
M_TEMP, M_WAITOK);
} else {
/* XXX: GCC */
ccount = 0;
ccp = NULL;
}
while (cdp < ecdp) {
if (skipbytes > 0) {
if (lfs_dir_getreclen(fs, cdp) <= skipbytes) {
skipbytes -= lfs_dir_getreclen(fs, cdp);
cdp = LFS_NEXTDIR(fs, cdp);
continue;
}
/*
* invalid cookie.
*/
error = EINVAL;
goto out;
}
if (lfs_dir_getreclen(fs, cdp) == 0) {
struct dirent *ondp = ndp;
ndp->d_reclen = _DIRENT_MINSIZE(ndp);
ndp = _DIRENT_NEXT(ndp);
ondp->d_reclen = 0;
cdp = ecdp;
break;
}
ndp->d_type = lfs_dir_gettype(fs, cdp);
ndp->d_namlen = lfs_dir_getnamlen(fs, cdp);
ndp->d_reclen = _DIRENT_RECLEN(ndp, ndp->d_namlen);
if ((char *)(void *)ndp + ndp->d_reclen +
_DIRENT_MINSIZE(ndp) > endp)
break;
ndp->d_fileno = lfs_dir_getino(fs, cdp);
(void)memcpy(ndp->d_name, lfs_dir_nameptr(fs, cdp),
ndp->d_namlen);
memset(&ndp->d_name[ndp->d_namlen], 0,
ndp->d_reclen - _DIRENT_NAMEOFF(ndp) - ndp->d_namlen);
off += lfs_dir_getreclen(fs, cdp);
if (ap->a_cookies) {
KASSERT(ccp - *(ap->a_cookies) < ccount);
*(ccp++) = off;
}
ndp = _DIRENT_NEXT(ndp);
cdp = LFS_NEXTDIR(fs, cdp);
}
count = ((char *)(void *)ndp - ndbuf);
error = uiomove(ndbuf, count, uio);
out:
if (ap->a_cookies) {
if (error) {
free(*(ap->a_cookies), M_TEMP);
*(ap->a_cookies) = NULL;
*(ap->a_ncookies) = 0;
} else {
*ap->a_ncookies = ccp - *(ap->a_cookies);
}
}
uio->uio_offset = off;
kmem_free(ndbuf, ndbufsz);
kmem_free(cdbuf, cdbufsz);
*ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
return error;
}
