int
ext2fs_gop_alloc(struct vnode *vp, off_t off, off_t len, int flags,
kauth_cred_t cred)
{
struct inode *ip = VTOI(vp);
struct m_ext2fs *fs = ip->i_e2fs;
int error, delta, bshift, bsize;
UVMHIST_FUNC("ext2fs_gop_alloc"); UVMHIST_CALLED(ubchist);
bshift = fs->e2fs_bshift;
bsize = 1 << bshift;
delta = off & (bsize - 1);
off -= delta;
len += delta;
while (len > 0) {
bsize = min(bsize, len);
UVMHIST_LOG(ubchist, "off 0x%x len 0x%x bsize 0x%x",
off, len, bsize, 0);
error = ext2fs_balloc(ip, ext2_lblkno(fs, off), bsize, cred,
NULL, flags);
if (error) {
UVMHIST_LOG(ubchist, "error %d", error, 0,0,0);
return error;
}
/*
* increase file size now, ext2fs_balloc() requires that
* EOF be up-to-date before each call.
*/
if (ext2fs_size(ip) < off + bsize) {
UVMHIST_LOG(ubchist, "old 0x%lx%8lx new 0x%lx%8lx",
/* Note that arguments are always cast to u_long. */
ext2fs_size(ip) >> 32,
ext2fs_size(ip) & 0xffffffff,
(off + bsize) >> 32,
(off + bsize) & 0xffffffff);
error = ext2fs_setsize(ip, off + bsize);
if (error) {
UVMHIST_LOG(ubchist, "error %d", error, 0,0,0);
return error;
}
}
off += bsize;
len -= bsize;
}
/*
* Advisory record locking support
*/
int
ext2fs_advlock(void *v)
{
struct vop_advlock_args *ap = v;
struct inode *ip = VTOI(ap->a_vp);
return (lf_advlock(&ip->i_lockf, ext2fs_size(ip), ap->a_id, ap->a_op,
ap->a_fl, ap->a_flags));
}
/*
* ext2fs_rmdired_p: Check whether the directory vp has been rmdired.
*
* vp must be locked and referenced.
*/
static bool
ext2fs_rmdired_p(struct vnode *vp)
{
KASSERT(vp != NULL);
KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
KASSERT(vp->v_type == VDIR);
/* XXX Is this correct? */
return (ext2fs_size(VTOI(vp)) == 0);
}
/* ARGSUSED */
int
ext2fs_getattr(void *v)
{
struct vop_getattr_args *ap = v;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct vattr *vap = ap->a_vap;
struct timeval tv;
getmicrotime(&tv);
EXT2FS_ITIMES(ip, &tv, &tv);
/*
* Copy from inode table
*/
vap->va_fsid = ip->i_dev;
vap->va_fileid = ip->i_number;
vap->va_mode = ip->i_e2fs_mode & ALLPERMS;
vap->va_nlink = ip->i_e2fs_nlink;
vap->va_uid = ip->i_e2fs_uid;
vap->va_gid = ip->i_e2fs_gid;
vap->va_rdev = (dev_t)fs2h32(ip->i_e2din->e2di_rdev);
vap->va_size = ext2fs_size(ip);
vap->va_atime.tv_sec = ip->i_e2fs_atime;
vap->va_atime.tv_nsec = 0;
vap->va_mtime.tv_sec = ip->i_e2fs_mtime;
vap->va_mtime.tv_nsec = 0;
vap->va_ctime.tv_sec = ip->i_e2fs_ctime;
vap->va_ctime.tv_nsec = 0;
#ifdef EXT2FS_SYSTEM_FLAGS
vap->va_flags = (ip->i_e2fs_flags & EXT2_APPEND) ? SF_APPEND : 0;
vap->va_flags |= (ip->i_e2fs_flags & EXT2_IMMUTABLE) ? SF_IMMUTABLE : 0;
#else
vap->va_flags = (ip->i_e2fs_flags & EXT2_APPEND) ? UF_APPEND : 0;
vap->va_flags |= (ip->i_e2fs_flags & EXT2_IMMUTABLE) ? UF_IMMUTABLE : 0;
#endif
vap->va_gen = ip->i_e2fs_gen;
/* this doesn't belong here */
if (vp->v_type == VBLK)
vap->va_blocksize = BLKDEV_IOSIZE;
else if (vp->v_type == VCHR)
vap->va_blocksize = MAXBSIZE;
else
vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
vap->va_bytes = dbtob((u_quad_t)ip->i_e2fs_nblock);
vap->va_type = vp->v_type;
vap->va_filerev = ip->i_modrev;
return (0);
}
/*
* Return target name of a symbolic link
*/
int
ext2fs_readlink(void *v)
{
struct vop_readlink_args *ap = v;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
int isize;
isize = ext2fs_size(ip);
if (isize < vp->v_mount->mnt_maxsymlinklen ||
(vp->v_mount->mnt_maxsymlinklen == 0 && ip->i_e2fs_nblock == 0)) {
return (uiomove((char *)ip->i_e2din->e2di_shortlink, isize,
ap->a_uio));
}
return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
}
/*
* Check if a directory is empty or not.
* Inode supplied must be locked.
*
* Using a struct dirtemplate here is not precisely
* what we want, but better than using a struct ext2fs_direct.
*
* NB: does not handle corrupted directories.
*/
int
ext2fs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
{
off_t off;
struct ext2fs_dirtemplate dbuf;
struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
int error, namlen;
size_t count;
#define MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
error = ufs_bufio(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ,
off, IO_NODELOCKED, cred, &count, NULL);
/*
* Since we read MINDIRSIZ, residual must
* be 0 unless we're at end of file.
*/
if (error || count != 0)
return (0);
/* avoid infinite loops */
if (dp->e2d_reclen == 0)
return (0);
/* skip empty entries */
if (dp->e2d_ino == 0)
continue;
/* accept only "." and ".." */
namlen = dp->e2d_namlen;
if (namlen > 2)
return (0);
if (dp->e2d_name[0] != '.')
return (0);
/*
* At this point namlen must be 1 or 2.
* 1 implies ".", 2 implies ".." if second
* char is also "."
*/
if (namlen == 1)
continue;
if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
continue;
return (0);
}
return (1);
}
/*
* Last reference to an inode. If necessary, write or delete it.
*/
int
ext2fs_inactive(void *v)
{
struct vop_inactive_args *ap = v;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct proc *p = ap->a_p;
struct timespec ts;
int error = 0;
#ifdef DIAGNOSTIC
extern int prtactive;
if (prtactive && vp->v_usecount != 0)
vprint("ext2fs_inactive: pushing active", vp);
#endif
/* Get rid of inodes related to stale file handles. */
if (ip->i_e2din == NULL || ip->i_e2fs_mode == 0 || ip->i_e2fs_dtime)
goto out;
error = 0;
if (ip->i_e2fs_nlink == 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
if (ext2fs_size(ip) != 0) {
error = ext2fs_truncate(ip, (off_t)0, 0, NOCRED);
}
getnanotime(&ts);
ip->i_e2fs_dtime = ts.tv_sec;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
ext2fs_inode_free(ip, ip->i_number, ip->i_e2fs_mode);
}
if (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) {
ext2fs_update(ip, 0);
}
out:
VOP_UNLOCK(vp, p);
/*
* If we are done with the inode, reclaim it
* so that it can be reused immediately.
*/
if (ip->i_e2din == NULL || ip->i_e2fs_dtime != 0)
vrecycle(vp, p);
return (error);
}
/*
* Write a directory entry after a call to namei, using the parameters
* that it left in nameidata. The argument ip is the inode which the new
* directory entry will refer to. Dvp is a pointer to the directory to
* be written, which was left locked by namei. Remaining parameters
* (ulr_offset, ulr_count) indicate how the space for the new
* entry is to be obtained.
*/
int
ext2fs_direnter(struct inode *ip, struct vnode *dvp,
const struct ufs_lookup_results *ulr,
struct componentname *cnp)
{
struct ext2fs_direct *ep, *nep;
struct inode *dp;
struct buf *bp;
struct ext2fs_direct newdir;
struct iovec aiov;
struct uio auio;
u_int dsize;
int error, loc, newentrysize, spacefree;
char *dirbuf;
struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
int dirblksiz = ump->um_dirblksiz;
dp = VTOI(dvp);
newdir.e2d_ino = h2fs32(ip->i_number);
newdir.e2d_namlen = cnp->cn_namelen;
if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
(ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
} else {
newdir.e2d_type = 0;
}
memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
if (ulr->ulr_count == 0) {
/*
* If ulr_count is 0, then namei could find no
* space in the directory. Here, ulr_offset will
* be on a directory block boundary and we will write the
* new entry into a fresh block.
*/
if (ulr->ulr_offset & (dirblksiz - 1))
panic("ext2fs_direnter: newblk");
auio.uio_offset = ulr->ulr_offset;
newdir.e2d_reclen = h2fs16(dirblksiz);
auio.uio_resid = newentrysize;
aiov.iov_len = newentrysize;
aiov.iov_base = (void *)&newdir;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_rw = UIO_WRITE;
UIO_SETUP_SYSSPACE(&auio);
error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize)
/* XXX should grow with balloc() */
panic("ext2fs_direnter: frag size");
else if (!error) {
error = ext2fs_setsize(dp,
roundup(ext2fs_size(dp), dirblksiz));
if (error)
return (error);
dp->i_flag |= IN_CHANGE;
uvm_vnp_setsize(dvp, ext2fs_size(dp));
}
return (error);
}
/*
* If ulr_count is non-zero, then namei found space
* for the new entry in the range ulr_offset to
* ulr_offset + ulr_count in the directory.
* To use this space, we may have to compact the entries located
* there, by copying them together towards the beginning of the
* block, leaving the free space in one usable chunk at the end.
*/
/*
* Get the block containing the space for the new directory entry.
*/
if ((error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp)) != 0)
return (error);
/*
* Find space for the new entry. In the simple case, the entry at
* offset base will have the space. If it does not, then namei
* arranged that compacting the region ulr_offset to
* ulr_offset + ulr_count would yield the
* space.
*/
ep = (struct ext2fs_direct *)dirbuf;
dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
spacefree = fs2h16(ep->e2d_reclen) - dsize;
for (loc = fs2h16(ep->e2d_reclen); loc < ulr->ulr_count; ) {
nep = (struct ext2fs_direct *)(dirbuf + loc);
if (ep->e2d_ino) {
/* trim the existing slot */
ep->e2d_reclen = h2fs16(dsize);
ep = (struct ext2fs_direct *)((char *)ep + dsize);
} else {
/* overwrite; nothing there; header is ours */
spacefree += dsize;
}
dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
spacefree += fs2h16(nep->e2d_reclen) - dsize;
loc += fs2h16(nep->e2d_reclen);
memcpy((void *)ep, (void *)nep, dsize);
}
/*
* Update the pointer fields in the previous entry (if any),
* copy in the new entry, and write out the block.
*/
if (ep->e2d_ino == 0) {
#ifdef DIAGNOSTIC
if (spacefree + dsize < newentrysize)
panic("ext2fs_direnter: compact1");
#endif
newdir.e2d_reclen = h2fs16(spacefree + dsize);
} else {
#ifdef DIAGNOSTIC
if (spacefree < newentrysize) {
printf("ext2fs_direnter: compact2 %u %u",
(u_int)spacefree, (u_int)newentrysize);
panic("ext2fs_direnter: compact2");
}
#endif
newdir.e2d_reclen = h2fs16(spacefree);
ep->e2d_reclen = h2fs16(dsize);
ep = (struct ext2fs_direct *)((char *)ep + dsize);
}
memcpy((void *)ep, (void *)&newdir, (u_int)newentrysize);
error = VOP_BWRITE(bp->b_vp, bp);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
if (!error && ulr->ulr_endoff && ulr->ulr_endoff < ext2fs_size(dp))
error = ext2fs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC,
cnp->cn_cred);
return (error);
}
/*
* Vnode op for reading directories.
*
* Convert the on-disk entries to <sys/dirent.h> entries.
* the problem is that the conversion will blow up some entries by four bytes,
* so it can't be done in place. This is too bad. Right now the conversion is
* done entry by entry, the converted entry is sent via uiomove.
*
* XXX allocate a buffer, convert as many entries as possible, then send
* the whole buffer to uiomove
*/
int
ext2fs_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 uio *uio = ap->a_uio;
int error;
size_t e2fs_count, readcnt;
struct vnode *vp = ap->a_vp;
struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
struct ext2fs_direct *dp;
struct dirent *dstd;
struct uio auio;
struct iovec aiov;
void *dirbuf;
off_t off = uio->uio_offset;
off_t *cookies = NULL;
int nc = 0, ncookies = 0;
int e2d_reclen;
if (vp->v_type != VDIR)
return (ENOTDIR);
e2fs_count = uio->uio_resid;
/* Make sure we don't return partial entries. */
e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
if (e2fs_count <= 0)
return (EINVAL);
auio = *uio;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
aiov.iov_len = e2fs_count;
auio.uio_resid = e2fs_count;
UIO_SETUP_SYSSPACE(&auio);
dirbuf = kmem_alloc(e2fs_count, KM_SLEEP);
dstd = kmem_zalloc(sizeof(struct dirent), KM_SLEEP);
if (ap->a_ncookies) {
nc = e2fs_count / _DIRENT_MINSIZE((struct dirent *)0);
ncookies = nc;
cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
*ap->a_cookies = cookies;
}
aiov.iov_base = dirbuf;
error = UFS_BUFRD(ap->a_vp, &auio, 0, ap->a_cred);
if (error == 0) {
readcnt = e2fs_count - auio.uio_resid;
for (dp = (struct ext2fs_direct *)dirbuf;
(char *)dp < (char *)dirbuf + readcnt; ) {
e2d_reclen = fs2h16(dp->e2d_reclen);
if (e2d_reclen == 0) {
error = EIO;
break;
}
ext2fs_dirconv2ffs(dp, dstd);
if(dstd->d_reclen > uio->uio_resid) {
break;
}
error = uiomove(dstd, dstd->d_reclen, uio);
if (error != 0) {
break;
}
off = off + e2d_reclen;
if (cookies != NULL) {
*cookies++ = off;
if (--ncookies <= 0){
break; /* out of cookies */
}
}
/* advance dp */
dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
}
/* we need to correct uio_offset */
uio->uio_offset = off;
}
kmem_free(dirbuf, e2fs_count);
kmem_free(dstd, sizeof(*dstd));
*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
if (ap->a_ncookies) {
if (error) {
free(*ap->a_cookies, M_TEMP);
*ap->a_ncookies = 0;
*ap->a_cookies = NULL;
} else
*ap->a_ncookies = nc - ncookies;
}
return (error);
}
/*
* 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 ext2fs_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
ext2fs_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 ext2fs_direct *ep; /* the current directory entry */
int entryoffsetinblock; /* offset of ep in bp's buffer */
enum {NONE, COMPACT, FOUND} slotstatus;
doff_t slotoffset; /* offset of area with free space */
int slotsize; /* size of area at slotoffset */
int slotfreespace; /* amount of space free in slot */
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; /* prev entry dp->i_offset */
struct vnode *tdp; /* returned by vcache_get */
doff_t enduseful; /* pointer past last used dir slot */
u_long bmask; /* block offset mask */
int namlen, 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 ufsmount *ump = dp->i_ump;
int dirblksiz = ump->um_dirblksiz;
ino_t foundino;
struct ufs_lookup_results *results;
flags = cnp->cn_flags;
bp = NULL;
slotoffset = -1;
*vpp = NULL;
/*
* 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) &&
(cnp->cn_nameiop == DELETE || cnp->cn_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, NULL, vpp)) {
return *vpp == NULLVP ? ENOENT : 0;
}
/*
* 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 = EXT2FS_DIRSIZ(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;
if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
results->ulr_diroff >= ext2fs_size(dp)) {
entryoffsetinblock = 0;
results->ulr_offset = 0;
numdirpasses = 1;
} else {
results->ulr_offset = results->ulr_diroff;
if ((entryoffsetinblock = results->ulr_offset & bmask) &&
(error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL, &bp)))
return (error);
numdirpasses = 2;
namecache_count_2passes();
}
prevoff = results->ulr_offset;
endsearch = roundup(ext2fs_size(dp), 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 = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL,
&bp);
if (error != 0)
return (error);
entryoffsetinblock = 0;
}
/*
* If still looking for a slot, and at a dirblksize
* 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
* "dirchk" to be true.
*/
KASSERT(bp != NULL);
ep = (struct ext2fs_direct *)
((char *)bp->b_data + entryoffsetinblock);
if (ep->e2d_reclen == 0 ||
(dirchk &&
ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
int i;
ufs_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 = fs2h16(ep->e2d_reclen);
if (ep->e2d_ino != 0)
size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
if (size > 0) {
if (size >= slotneeded) {
slotstatus = FOUND;
slotoffset = results->ulr_offset;
slotsize = fs2h16(ep->e2d_reclen);
} else if (slotstatus == NONE) {
slotfreespace += size;
if (slotoffset == -1)
slotoffset = results->ulr_offset;
if (slotfreespace >= slotneeded) {
slotstatus = COMPACT;
slotsize = results->ulr_offset +
fs2h16(ep->e2d_reclen) -
slotoffset;
}
}
}
}
/*
* Check for a name match.
*/
if (ep->e2d_ino) {
namlen = ep->e2d_namlen;
if (namlen == cnp->cn_namelen &&
!memcmp(cnp->cn_nameptr, ep->e2d_name,
(unsigned)namlen)) {
/*
* Save directory entry's inode number and
* reclen in ndp->ni_ufs area, and release
* directory buffer.
*/
foundino = fs2h32(ep->e2d_ino);
results->ulr_reclen = fs2h16(ep->e2d_reclen);
goto found;
}
}
prevoff = results->ulr_offset;
results->ulr_offset += fs2h16(ep->e2d_reclen);
entryoffsetinblock += fs2h16(ep->e2d_reclen);
if (ep->e2d_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) &&
(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
/*
* Access for write is interpreted as allowing
* creation of files in the directory.
*/
error = VOP_ACCESS(vdp, VWRITE, cred);
if (error)
return (error);
/*
* 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(ext2fs_size(dp), dirblksiz);
results->ulr_count = 0;
enduseful = results->ulr_offset;
} else {
results->ulr_offset = slotoffset;
results->ulr_count = slotsize;
if (enduseful < slotoffset + slotsize)
enduseful = slotoffset + slotsize;
}
results->ulr_endoff = roundup(enduseful, dirblksiz);
#if 0
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.
*/
return (EJUSTRETURN);
}
/*
* 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);
}
return ENOENT;
found:
if (numdirpasses == 2)
namecache_count_pass2();
/*
* Check that directory length properly reflects presence
* of this entry.
*/
if (results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
ufs_dirbad(dp, results->ulr_offset, "i_size too small");
error = ext2fs_setsize(dp,
results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
if (error) {
brelse(bp, 0);
return (error);
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
uvm_vnp_setsize(vdp, ext2fs_size(dp));
}
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 {
error = vcache_get(vdp->v_mount,
&foundino, sizeof(foundino), &tdp);
if (error)
return (error);
}
/*
* Write access to directory required to delete files.
*/
if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0) {
vrele(tdp);
return (error);
}
/*
* 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_e2fs_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) {
vrele(tdp);
return (EPERM);
}
}
*vpp = tdp;
return (0);
}
/*
* 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)
return (error);
/*
* Careful about locking second inode.
* This can only occur if the target is ".".
*/
if (dp->i_number == foundino)
return (EISDIR);
error = vcache_get(vdp->v_mount,
&foundino, sizeof(foundino), &tdp);
if (error)
return (error);
*vpp = tdp;
return (0);
}
if (dp->i_number == foundino) {
vref(vdp); /* we want ourself, ie "." */
*vpp = vdp;
} else {
error = vcache_get(vdp->v_mount,
&foundino, sizeof(foundino), &tdp);
if (error)
return (error);
*vpp = tdp;
}
/*
* Insert name into cache if appropriate.
*/
cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
return 0;
}
/*
* Vnode op for writing.
*/
int
ext2fs_write(void *v)
{
struct vop_write_args *ap = v;
struct vnode *vp;
struct uio *uio;
struct inode *ip;
struct m_ext2fs *fs;
struct buf *bp;
int32_t lbn;
off_t osize;
int blkoffset, error, flags, ioflag, size, xfersize;
ssize_t resid, overrun;
ioflag = ap->a_ioflag;
uio = ap->a_uio;
vp = ap->a_vp;
ip = VTOI(vp);
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("%s: mode", "ext2fs_write");
#endif
/*
* If writing 0 bytes, succeed and do not change
* update time or file offset (standards compliance)
*/
if (uio->uio_resid == 0)
return (0);
switch (vp->v_type) {
case VREG:
if (ioflag & IO_APPEND)
uio->uio_offset = ext2fs_size(ip);
if ((ip->i_e2fs_flags & EXT2_APPEND) &&
uio->uio_offset != ext2fs_size(ip))
return (EPERM);
/* FALLTHROUGH */
case VLNK:
break;
case VDIR:
if ((ioflag & IO_SYNC) == 0)
panic("%s: nonsync dir write", "ext2fs_write");
break;
default:
panic("%s: type", "ext2fs_write");
}
fs = ip->i_e2fs;
if (uio->uio_offset < 0 ||
(u_int64_t)uio->uio_offset + uio->uio_resid >
((u_int64_t)0x80000000 * fs->e2fs_bsize - 1))
return (EFBIG);
/* do the filesize rlimit check */
if ((error = vn_fsizechk(vp, uio, ioflag, &overrun)))
return (error);
resid = uio->uio_resid;
osize = ext2fs_size(ip);
flags = ioflag & IO_SYNC ? B_SYNC : 0;
for (error = 0; uio->uio_resid > 0;) {
lbn = lblkno(fs, uio->uio_offset);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->e2fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (fs->e2fs_bsize > xfersize)
flags |= B_CLRBUF;
else
flags &= ~B_CLRBUF;
error = ext2fs_buf_alloc(ip,
lbn, blkoffset + xfersize, ap->a_cred, &bp, flags);
if (error)
break;
if (uio->uio_offset + xfersize > ext2fs_size(ip)) {
error = ext2fs_setsize(ip, uio->uio_offset + xfersize);
if (error)
break;
uvm_vnp_setsize(vp, ip->i_e2fs_size);
}
uvm_vnp_uncache(vp);
size = fs->e2fs_bsize - bp->b_resid;
if (size < xfersize)
xfersize = size;
error =
uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
if (ioflag & IO_SYNC)
(void)bwrite(bp);
else if (xfersize + blkoffset == fs->e2fs_bsize) {
if (doclusterwrite)
cluster_write(bp, &ip->i_ci, ext2fs_size(ip));
else
bawrite(bp);
} else
bdwrite(bp);
if (error || xfersize == 0)
break;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
ip->i_e2fs_mode &= ~(ISUID | ISGID);
if (error) {
if (ioflag & IO_UNIT) {
(void)ext2fs_truncate(ip, osize,
ioflag & IO_SYNC, ap->a_cred);
uio->uio_offset -= resid - uio->uio_resid;
uio->uio_resid = resid;
}
} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
error = ext2fs_update(ip, NULL, NULL, 1);
}
/* correct the result for writes clamped by vn_fsizechk() */
uio->uio_resid += overrun;
return (error);
}
/* ARGSUSED */
int
ext2fs_read(void *v)
{
struct vop_read_args *ap = v;
struct vnode *vp;
struct inode *ip;
struct uio *uio;
struct m_ext2fs *fs;
struct buf *bp;
daddr_t lbn, nextlbn;
off_t bytesinfile;
long size, xfersize, blkoffset;
int error;
vp = ap->a_vp;
ip = VTOI(vp);
uio = ap->a_uio;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_READ)
panic("%s: mode", "ext2fs_read");
if (vp->v_type == VLNK) {
if ((int)ext2fs_size(ip) < vp->v_mount->mnt_maxsymlinklen ||
(vp->v_mount->mnt_maxsymlinklen == 0 &&
ip->i_e2fs_nblock == 0))
panic("%s: short symlink", "ext2fs_read");
} else if (vp->v_type != VREG && vp->v_type != VDIR)
panic("%s: type %d", "ext2fs_read", vp->v_type);
#endif
fs = ip->i_e2fs;
if ((u_int64_t)uio->uio_offset >
((u_int64_t)0x80000000 * fs->e2fs_bsize - 1))
return (EFBIG);
if (uio->uio_resid == 0)
return (0);
for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
if ((bytesinfile = ext2fs_size(ip) - uio->uio_offset) <= 0)
break;
lbn = lblkno(fs, uio->uio_offset);
nextlbn = lbn + 1;
size = fs->e2fs_bsize;
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->e2fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (bytesinfile < xfersize)
xfersize = bytesinfile;
if (lblktosize(fs, nextlbn) >= ext2fs_size(ip))
error = bread(vp, lbn, size, &bp);
else if (lbn - 1 == ip->i_ci.ci_lastr) {
int nextsize = fs->e2fs_bsize;
error = breadn(vp, lbn, size, &nextlbn, &nextsize,
1, &bp);
} else
error = bread(vp, lbn, size, &bp);
if (error)
break;
ip->i_ci.ci_lastr = lbn;
/*
* We should only get non-zero b_resid when an I/O error
* has occurred, which should cause us to break above.
* However, if the short read did not cause an error,
* then we want to ensure that we do not uiomove bad
* or uninitialized data.
*/
size -= bp->b_resid;
if (size < xfersize) {
if (size == 0)
break;
xfersize = size;
}
error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
if (error)
break;
brelse(bp);
}
if (bp != NULL)
brelse(bp);
if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) {
ip->i_flag |= IN_ACCESS;
}
return (error);
}
/*
* Vnode op for reading directories.
*
* Convert the on-disk entries to <sys/dirent.h> entries.
* the problem is that the conversion will blow up some entries by four bytes,
* so it can't be done in place. This is too bad. Right now the conversion is
* done entry by entry, the converted entry is sent via uiomove.
*
* XXX allocate a buffer, convert as many entries as possible, then send
* the whole buffer to uiomove
*/
int
ext2fs_readdir(void *v)
{
struct vop_readdir_args *ap = v;
struct uio *uio = ap->a_uio;
int error;
size_t e2fs_count, readcnt, entries;
struct vnode *vp = ap->a_vp;
struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
struct ext2fs_direct *dp;
struct dirent dstd;
struct uio auio;
struct iovec aiov;
caddr_t dirbuf;
off_t off = uio->uio_offset;
u_long *cookies = NULL;
int nc = 0, ncookies = 0;
int e2d_reclen;
if (vp->v_type != VDIR)
return (ENOTDIR);
e2fs_count = uio->uio_resid;
entries = (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize - 1);
/* Make sure we don't return partial entries. */
if (e2fs_count <= entries)
return (EINVAL);
e2fs_count -= entries;
auio = *uio;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_segflg = UIO_SYSSPACE;
aiov.iov_len = e2fs_count;
auio.uio_resid = e2fs_count;
dirbuf = malloc(e2fs_count, M_TEMP, M_WAITOK | M_ZERO);
if (ap->a_ncookies) {
nc = ncookies = e2fs_count / 16;
cookies = malloc(sizeof(*cookies) * ncookies, M_TEMP, M_WAITOK);
*ap->a_cookies = cookies;
}
aiov.iov_base = dirbuf;
error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
if (error == 0) {
readcnt = e2fs_count - auio.uio_resid;
for (dp = (struct ext2fs_direct *)dirbuf;
(char *)dp < (char *)dirbuf + readcnt; ) {
e2d_reclen = fs2h16(dp->e2d_reclen);
if (e2d_reclen == 0) {
error = EIO;
break;
}
ext2fs_dirconv2ffs(dp, &dstd);
if(dstd.d_reclen > uio->uio_resid) {
break;
}
if ((error = uiomove((caddr_t)&dstd, dstd.d_reclen, uio)) != 0) {
break;
}
off = off + e2d_reclen;
if (cookies != NULL) {
*cookies++ = off;
if (--ncookies <= 0){
break; /* out of cookies */
}
}
/* advance dp */
dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
}
/* we need to correct uio_offset */
uio->uio_offset = off;
}
free(dirbuf, M_TEMP);
*ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
if (ap->a_ncookies) {
if (error) {
free(*ap->a_cookies, M_TEMP);
*ap->a_ncookies = 0;
*ap->a_cookies = NULL;
} else
*ap->a_ncookies = nc - ncookies;
}
return (error);
}
/*
* 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 ext2fs_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
ext2fs_lookup(void *v)
{
struct vop_lookup_args *ap = v;
struct vnode *vdp; /* vnode for directory being searched */
struct inode *dp; /* inode for directory being searched */
struct buf *bp; /* a buffer of directory entries */
struct ext2fs_direct *ep; /* the current directory entry */
int entryoffsetinblock; /* offset of ep in bp's buffer */
enum {NONE, COMPACT, FOUND} slotstatus;
doff_t slotoffset; /* offset of area with free space */
int slotsize; /* size of area at slotoffset */
int slotfreespace; /* amount of space free in slot */
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; /* prev entry dp->i_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 */
u_long bmask; /* block offset mask */
int lockparent; /* 1 => lockparent flag is set */
int wantparent; /* 1 => wantparent or lockparent flag */
int namlen, error;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
struct ucred *cred = cnp->cn_cred;
int flags = cnp->cn_flags;
int nameiop = cnp->cn_nameiop;
struct proc *p = cnp->cn_proc;
int dirblksize = VTOI(ap->a_dvp)->i_e2fs->e2fs_bsize;
bp = NULL;
slotoffset = -1;
*vpp = NULL;
vdp = ap->a_dvp;
dp = VTOI(vdp);
lockparent = flags & LOCKPARENT;
wantparent = flags & (LOCKPARENT|WANTPARENT);
/*
* Check accessiblity of directory.
*/
if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
return (error);
if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == DELETE || cnp->cn_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 ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
return (error);
/*
* 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 = EXT2FS_DIRSIZ(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 = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
if (nameiop != LOOKUP || dp->i_diroff == 0 ||
dp->i_diroff >ext2fs_size(dp)) {
entryoffsetinblock = 0;
dp->i_offset = 0;
numdirpasses = 1;
} else {
dp->i_offset = dp->i_diroff;
if ((entryoffsetinblock = dp->i_offset & bmask) &&
(error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
NULL, &bp)))
return (error);
numdirpasses = 2;
}
prevoff = dp->i_offset;
endsearch = roundup(ext2fs_size(dp), dirblksize);
enduseful = 0;
searchloop:
while (dp->i_offset < endsearch) {
/*
* If necessary, get the next directory block.
*/
if ((dp->i_offset & bmask) == 0) {
if (bp != NULL)
brelse(bp);
error = ext2fs_bufatoff(dp, (off_t)dp->i_offset,
NULL, &bp);
if (error != 0)
return (error);
entryoffsetinblock = 0;
}
/*
* If still looking for a slot, and at a dirblksize
* boundary, have to start looking for free space again.
*/
if (slotstatus == NONE &&
(entryoffsetinblock & (dirblksize - 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
* "dirchk" to be true.
*/
ep = (struct ext2fs_direct *)
((char *)bp->b_data + entryoffsetinblock);
if (ep->e2d_reclen == 0 ||
(dirchk &&
ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
int i;
ufs_dirbad(dp, dp->i_offset, "mangled entry");
i = dirblksize -
(entryoffsetinblock & (dirblksize - 1));
dp->i_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 = fs2h16(ep->e2d_reclen);
if (ep->e2d_ino != 0)
size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
if (size > 0) {
if (size >= slotneeded) {
slotstatus = FOUND;
slotoffset = dp->i_offset;
slotsize = fs2h16(ep->e2d_reclen);
} else if (slotstatus == NONE) {
slotfreespace += size;
if (slotoffset == -1)
slotoffset = dp->i_offset;
if (slotfreespace >= slotneeded) {
slotstatus = COMPACT;
slotsize = dp->i_offset +
fs2h16(ep->e2d_reclen) - slotoffset;
}
}
}
}
/*
* Check for a name match.
*/
if (ep->e2d_ino) {
namlen = ep->e2d_namlen;
if (namlen == cnp->cn_namelen &&
!memcmp(cnp->cn_nameptr, ep->e2d_name,
(unsigned)namlen)) {
/*
* Save directory entry's inode number and
* reclen in ndp->ni_ufs area, and release
* directory buffer.
*/
dp->i_ino = fs2h32(ep->e2d_ino);
dp->i_reclen = fs2h16(ep->e2d_reclen);
brelse(bp);
goto found;
}
}
prevoff = dp->i_offset;
dp->i_offset += fs2h16(ep->e2d_reclen);
entryoffsetinblock += fs2h16(ep->e2d_reclen);
if (ep->e2d_ino)
enduseful = dp->i_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--;
dp->i_offset = 0;
endsearch = dp->i_diroff;
goto searchloop;
}
if (bp != NULL)
brelse(bp);
/*
* 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) &&
(flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
/*
* Creation of files on a read-only mounted file system
* is pointless, so don't proceed any further.
*/
if (vdp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
/*
* Access for write is interpreted as allowing
* creation of files in the directory.
*/
if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
return (error);
/*
* Return an indication of where the new directory
* entry should be put. If we didn't find a slot,
* then set dp->i_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 dp->i_offset to
* dp->i_offset + dp->i_count.
*/
if (slotstatus == NONE) {
dp->i_offset = roundup(ext2fs_size(dp), dirblksize);
dp->i_count = 0;
enduseful = dp->i_offset;
} else {
dp->i_offset = slotoffset;
dp->i_count = slotsize;
if (enduseful < slotoffset + slotsize)
enduseful = slotoffset + slotsize;
}
dp->i_endoff = roundup(enduseful, dirblksize);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* 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.
* The pathname buffer is saved so that the name
* can be obtained later.
*
* NB - if the directory is unlocked, then this
* information cannot be used.
*/
cnp->cn_flags |= SAVENAME;
if (!lockparent) {
VOP_UNLOCK(vdp, 0);
cnp->cn_flags |= PDIRUNLOCK;
}
return (EJUSTRETURN);
}
/*
* Insert name into cache (as non-existent) if appropriate.
*/
if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
cache_enter(vdp, *vpp, cnp);
return (ENOENT);
found:
/*
* Check that directory length properly reflects presence
* of this entry.
*/
if (entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen)
> ext2fs_size(dp)) {
ufs_dirbad(dp, dp->i_offset, "i_size too small");
error = ext2fs_setsize(dp,
entryoffsetinblock + EXT2FS_DIRSIZ(ep->e2d_namlen));
if (error) {
brelse(bp);
return(error);
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* 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)
dp->i_diroff = dp->i_offset &~ (dirblksize - 1);
/*
* If deleting, and at end of pathname, return
* parameters which can be used to remove file.
* If the wantparent flag isn't set, we return only
* the directory (in ndp->ni_dvp), otherwise we go
* on and lock the inode, being careful with ".".
*/
if (nameiop == DELETE && (flags & ISLASTCN)) {
/*
* Write access to directory required to delete files.
*/
if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
return (error);
/*
* Return pointer to current entry in dp->i_offset,
* and distance past previous entry (if there
* is a previous entry in this block) in dp->i_count.
* Save directory inode pointer in ndp->ni_dvp for dirremove().
*/
if ((dp->i_offset & (dirblksize - 1)) == 0)
dp->i_count = 0;
else
dp->i_count = dp->i_offset - prevoff;
if (dp->i_number == dp->i_ino) {
vref(vdp);
*vpp = vdp;
return (0);
}
if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
return (error);
/*
* 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_e2fs_mode & ISVTX) &&
cred->cr_uid != 0 &&
cred->cr_uid != dp->i_e2fs_uid &&
VTOI(tdp)->i_e2fs_uid != cred->cr_uid) {
vput(tdp);
return (EPERM);
}
*vpp = tdp;
if (!lockparent) {
VOP_UNLOCK(vdp, 0);
cnp->cn_flags |= PDIRUNLOCK;
}
return (0);
}
/*
* 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 && wantparent &&
(flags & ISLASTCN)) {
if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0)
return (error);
/*
* Careful about locking second inode.
* This can only occur if the target is ".".
*/
if (dp->i_number == dp->i_ino)
return (EISDIR);
if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
return (error);
*vpp = tdp;
cnp->cn_flags |= SAVENAME;
if (!lockparent) {
VOP_UNLOCK(vdp, 0);
cnp->cn_flags |= PDIRUNLOCK;
}
return (0);
}
/*
* 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, 0); /* race to get the inode */
cnp->cn_flags |= PDIRUNLOCK;
if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p) == 0)
cnp->cn_flags &= ~PDIRUNLOCK;
return (error);
}
if (lockparent && (flags & ISLASTCN)) {
if ((error = vn_lock(pdp, LK_EXCLUSIVE, p)) != 0) {
vput(tdp);
return (error);
}
cnp->cn_flags &= ~PDIRUNLOCK;
}
*vpp = tdp;
} else if (dp->i_number == dp->i_ino) {
vref(vdp); /* we want ourself, ie "." */
*vpp = vdp;
} else {
if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0)
return (error);
if (!lockparent || !(flags & ISLASTCN)) {
VOP_UNLOCK(pdp, 0);
cnp->cn_flags |= PDIRUNLOCK;
}
*vpp = tdp;
}
/*
* Insert name into cache if appropriate.
*/
if (cnp->cn_flags & MAKEENTRY)
cache_enter(vdp, *vpp, cnp);
return (0);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2fs_truncate(struct inode *oip, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp = ITOV(oip);
int32_t lastblock;
int32_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
int32_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct m_ext2fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, vflags, blocksreleased = 0;
int i;
int aflags, error, allerror;
off_t osize;
if (length < 0)
return (EINVAL);
if (ovp->v_type != VREG &&
ovp->v_type != VDIR &&
ovp->v_type != VLNK)
return (0);
if (ovp->v_type == VLNK && ext2fs_size(oip) < EXT2_MAXSYMLINKLEN) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ext2fs_truncate: partial truncate of symlink");
#endif
memset(&oip->i_e2din->e2di_shortlink, 0, ext2fs_size(oip));
(void)ext2fs_setsize(oip, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2fs_update(oip, 1));
}
if (ext2fs_size(oip) == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2fs_update(oip, 0));
}
fs = oip->i_e2fs;
osize = ext2fs_size(oip);
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
#if 0 /* XXX */
if (length > fs->fs_maxfilesize)
return (EFBIG);
#endif
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2fs_buf_alloc(oip, lbn, offset + 1, cred, &bp,
aflags);
if (error)
return (error);
(void)ext2fs_setsize(oip, length);
uvm_vnp_setsize(ovp, length);
uvm_vnp_uncache(ovp);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2fs_update(oip, 1));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessible again because
* of subsequent file growth.
*/
offset = blkoff(fs, length);
if (offset == 0) {
(void)ext2fs_setsize(oip, length);
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2fs_buf_alloc(oip, lbn, offset, cred, &bp,
aflags);
if (error)
return (error);
(void)ext2fs_setsize(oip, length);
size = fs->e2fs_bsize;
uvm_vnp_setsize(ovp, length);
uvm_vnp_uncache(ovp);
memset(bp->b_data + offset, 0, size - offset);
bp->b_bcount = size;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->e2fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->e2fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2fs_indirtrunc below.
*/
memcpy(oldblks, &oip->i_e2fs_blocks[0], sizeof(oldblks));
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
oip->i_e2fs_blocks[NDADDR + level] = 0;
lastiblock[level] = -1;
}
for (i = NDADDR - 1; i > lastblock; i--)
oip->i_e2fs_blocks[i] = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
if ((error = ext2fs_update(oip, 1)) != 0)
allerror = error;
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
memcpy(newblks, &oip->i_e2fs_blocks[0], sizeof(newblks));
memcpy(&oip->i_e2fs_blocks[0], oldblks, sizeof(oldblks));
(void)ext2fs_setsize(oip, osize);
vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, 0);
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) -1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = letoh32(oip->i_e2fs_blocks[NDADDR + level]);
if (bn != 0) {
error = ext2fs_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_e2fs_blocks[NDADDR + level] = 0;
ext2fs_blkfree(oip, bn);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
bn = letoh32(oip->i_e2fs_blocks[i]);
if (bn == 0)
continue;
oip->i_e2fs_blocks[i] = 0;
ext2fs_blkfree(oip, bn);
blocksreleased += btodb(fs->e2fs_bsize);
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] !=
oip->i_e2fs_blocks[NDADDR + level])
panic("ext2fs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_e2fs_blocks[i])
panic("ext2fs_truncate2");
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) ||
!LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("ext2fs_truncate3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
(void)ext2fs_setsize(oip, length);
if (blocksreleased >= oip->i_e2fs_nblock)
oip->i_e2fs_nblock = 0;
else
oip->i_e2fs_nblock -= blocksreleased;
oip->i_flag |= IN_CHANGE;
return (allerror);
}
/*
* Read an inode from disk and initialize this vnode / inode pair.
* Caller assures no other thread will try to load this inode.
*/
int
ext2fs_loadvnode(struct mount *mp, struct vnode *vp,
const void *key, size_t key_len, const void **new_key)
{
ino_t ino;
struct m_ext2fs *fs;
struct inode *ip;
struct ufsmount *ump;
struct buf *bp;
dev_t dev;
int error;
KASSERT(key_len == sizeof(ino));
memcpy(&ino, key, key_len);
ump = VFSTOUFS(mp);
dev = ump->um_dev;
fs = ump->um_e2fs;
/* Read in the disk contents for the inode, copy into the inode. */
error = bread(ump->um_devvp, EXT2_FSBTODB(fs, ino_to_fsba(fs, ino)),
(int)fs->e2fs_bsize, 0, &bp);
if (error)
return error;
/* Allocate and initialize inode. */
ip = pool_get(&ext2fs_inode_pool, PR_WAITOK);
memset(ip, 0, sizeof(struct inode));
vp->v_tag = VT_EXT2FS;
vp->v_op = ext2fs_vnodeop_p;
vp->v_vflag |= VV_LOCKSWORK;
vp->v_data = ip;
ip->i_vnode = vp;
ip->i_ump = ump;
ip->i_e2fs = fs;
ip->i_dev = dev;
ip->i_number = ino;
ip->i_e2fs_last_lblk = 0;
ip->i_e2fs_last_blk = 0;
/* Initialize genfs node. */
genfs_node_init(vp, &ext2fs_genfsops);
error = ext2fs_loadvnode_content(fs, ino, bp, ip);
brelse(bp, 0);
if (error)
return error;
/* If the inode was deleted, reset all fields */
if (ip->i_e2fs_dtime != 0) {
ip->i_e2fs_mode = 0;
(void)ext2fs_setsize(ip, 0);
(void)ext2fs_setnblock(ip, 0);
memset(ip->i_e2fs_blocks, 0, sizeof(ip->i_e2fs_blocks));
}
/* Initialize the vnode from the inode. */
ext2fs_vinit(mp, ext2fs_specop_p, ext2fs_fifoop_p, &vp);
/* Finish inode initialization. */
ip->i_devvp = ump->um_devvp;
vref(ip->i_devvp);
/*
* Set up a generation number for this inode if it does not
* already have one. This should only happen on old filesystems.
*/
if (ip->i_e2fs_gen == 0) {
if (++ext2gennumber < (u_long)time_second)
ext2gennumber = time_second;
ip->i_e2fs_gen = ext2gennumber;
if ((mp->mnt_flag & MNT_RDONLY) == 0)
ip->i_flag |= IN_MODIFIED;
}
uvm_vnp_setsize(vp, ext2fs_size(ip));
*new_key = &ip->i_number;
return 0;
}
/*
* Look up a EXT2FS dinode number to find its incore vnode, otherwise read it
* in from disk. If it is in core, wait for the lock bit to clear, then
* return the inode locked. Detection and handling of mount points must be
* done by the calling routine.
*/
int
ext2fs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
{
struct m_ext2fs *fs;
struct inode *ip;
struct ufsmount *ump;
struct buf *bp;
struct vnode *vp;
dev_t dev;
int error;
void *cp;
ump = VFSTOUFS(mp);
dev = ump->um_dev;
retry:
if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
return (0);
/* Allocate a new vnode/inode. */
error = getnewvnode(VT_EXT2FS, mp, ext2fs_vnodeop_p, NULL, &vp);
if (error) {
*vpp = NULL;
return (error);
}
ip = pool_get(&ext2fs_inode_pool, PR_WAITOK);
mutex_enter(&ufs_hashlock);
if ((*vpp = ufs_ihashget(dev, ino, 0)) != NULL) {
mutex_exit(&ufs_hashlock);
ungetnewvnode(vp);
pool_put(&ext2fs_inode_pool, ip);
goto retry;
}
vp->v_vflag |= VV_LOCKSWORK;
memset(ip, 0, sizeof(struct inode));
vp->v_data = ip;
ip->i_vnode = vp;
ip->i_ump = ump;
ip->i_e2fs = fs = ump->um_e2fs;
ip->i_dev = dev;
ip->i_number = ino;
ip->i_e2fs_last_lblk = 0;
ip->i_e2fs_last_blk = 0;
genfs_node_init(vp, &ext2fs_genfsops);
/*
* Put it onto its hash chain and lock it so that other requests for
* this inode will block if they arrive while we are sleeping waiting
* for old data structures to be purged or for the contents of the
* disk portion of this inode to be read.
*/
ufs_ihashins(ip);
mutex_exit(&ufs_hashlock);
/* Read in the disk contents for the inode, copy into the inode. */
error = bread(ump->um_devvp, EXT2_FSBTODB(fs, ino_to_fsba(fs, ino)),
(int)fs->e2fs_bsize, NOCRED, 0, &bp);
if (error) {
/*
* The inode does not contain anything useful, so it would
* be misleading to leave it on its hash chain. With mode
* still zero, it will be unlinked and returned to the free
* list by vput().
*/
vput(vp);
*vpp = NULL;
return (error);
}
cp = (char *)bp->b_data + (ino_to_fsbo(fs, ino) * EXT2_DINODE_SIZE(fs));
ip->i_din.e2fs_din = pool_get(&ext2fs_dinode_pool, PR_WAITOK);
e2fs_iload((struct ext2fs_dinode *)cp, ip->i_din.e2fs_din);
ext2fs_set_inode_guid(ip);
brelse(bp, 0);
/* If the inode was deleted, reset all fields */
if (ip->i_e2fs_dtime != 0) {
ip->i_e2fs_mode = 0;
(void)ext2fs_setsize(ip, 0);
(void)ext2fs_setnblock(ip, 0);
memset(ip->i_e2fs_blocks, 0, sizeof(ip->i_e2fs_blocks));
}
/*
* Initialize the vnode from the inode, check for aliases.
*/
error = ext2fs_vinit(mp, ext2fs_specop_p, ext2fs_fifoop_p, &vp);
if (error) {
vput(vp);
*vpp = NULL;
return (error);
}
/*
* Finish inode initialization now that aliasing has been resolved.
*/
ip->i_devvp = ump->um_devvp;
vref(ip->i_devvp);
/*
* Set up a generation number for this inode if it does not
* already have one. This should only happen on old filesystems.
*/
if (ip->i_e2fs_gen == 0) {
if (++ext2gennumber < (u_long)time_second)
ext2gennumber = time_second;
ip->i_e2fs_gen = ext2gennumber;
if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
ip->i_flag |= IN_MODIFIED;
}
uvm_vnp_setsize(vp, ext2fs_size(ip));
*vpp = vp;
return (0);
}