/*
* Write a directory entry after a call to namei, using the parameters
* that it left in nameidata. The argument dirp is the new directory
* entry contents. Dvp is a pointer to the directory to be written,
* which was left locked by namei. Remaining parameters (dp->i_offset,
* dp->i_count) indicate how the space for the new entry is to be obtained.
* Non-null bp indicates that a directory is being created (for the
* soft dependency code).
*/
int
ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
struct componentname *cnp, struct buf *newdirbp)
{
struct ucred *cr;
struct proc *p;
int newentrysize;
struct inode *dp;
struct buf *bp;
u_int dsize;
struct direct *ep, *nep;
int error, ret, blkoff, loc, spacefree, flags;
char *dirbuf;
error = 0;
cr = cnp->cn_cred;
p = cnp->cn_proc;
dp = VTOI(dvp);
newentrysize = DIRSIZ(FSFMT(dvp), dirp);
if (dp->i_count == 0) {
/*
* If dp->i_count is 0, then namei could find no
* space in the directory. Here, dp->i_offset will
* be on a directory block boundary and we will write the
* new entry into a fresh block.
*/
if (dp->i_offset & (DIRBLKSIZ - 1))
panic("ufs_direnter: newblk");
flags = B_CLRBUF;
if (!DOINGSOFTDEP(dvp))
flags |= B_SYNC;
if ((error = UFS_BUF_ALLOC(dp, (off_t)dp->i_offset, DIRBLKSIZ,
cr, flags, &bp)) != 0) {
if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
bdwrite(newdirbp);
return (error);
}
DIP_ASSIGN(dp, size, dp->i_offset + DIRBLKSIZ);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
uvm_vnp_setsize(dvp, DIP(dp, size));
dirp->d_reclen = DIRBLKSIZ;
blkoff = dp->i_offset &
(VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL) {
ufsdirhash_newblk(dp, dp->i_offset);
ufsdirhash_add(dp, dirp, dp->i_offset);
ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
dp->i_offset);
}
#endif
if (DOINGSOFTDEP(dvp)) {
/*
* Ensure that the entire newly allocated block is a
* valid directory so that future growth within the
* block does not have to ensure that the block is
* written before the inode.
*/
blkoff += DIRBLKSIZ;
while (blkoff < bp->b_bcount) {
((struct direct *)
(bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
blkoff += DIRBLKSIZ;
}
if (softdep_setup_directory_add(bp, dp, dp->i_offset,
dirp->d_ino, newdirbp, 1) == 0) {
bdwrite(bp);
return (UFS_UPDATE(dp, 0));
}
/* We have just allocated a directory block in an
* indirect block. Rather than tracking when it gets
* claimed by the inode, we simply do a VOP_FSYNC
* now to ensure that it is there (in case the user
* does a future fsync). Note that we have to unlock
* the inode for the entry that we just entered, as
* the VOP_FSYNC may need to lock other inodes which
* can lead to deadlock if we also hold a lock on
* the newly entered node.
*/
if ((error = VOP_BWRITE(bp)))
return (error);
if (tvp != NULL)
VOP_UNLOCK(tvp, 0, p);
error = VOP_FSYNC(dvp, p->p_ucred, MNT_WAIT, p);
if (tvp != NULL)
vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
return (error);
}
error = VOP_BWRITE(bp);
ret = UFS_UPDATE(dp, !DOINGSOFTDEP(dvp));
if (error == 0)
return (ret);
return (error);
}
/*
* If dp->i_count is non-zero, then namei found space for the new
* entry in the range dp->i_offset to dp->i_offset + dp->i_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.
*/
/*
* Increase size of directory if entry eats into new space.
* This should never push the size past a new multiple of
* DIRBLKSIZE.
*
* N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
*/
if (dp->i_offset + dp->i_count > DIP(dp, size))
DIP_ASSIGN(dp, size, dp->i_offset + dp->i_count);
/*
* Get the block containing the space for the new directory entry.
*/
if ((error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, &dirbuf, &bp))
!= 0) {
if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
bdwrite(newdirbp);
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 dp->i_offset to
* dp->i_offset + dp->i_count would yield the space.
*/
ep = (struct direct *)dirbuf;
dsize = ep->d_ino ? DIRSIZ(FSFMT(dvp), ep) : 0;
spacefree = ep->d_reclen - dsize;
for (loc = ep->d_reclen; loc < dp->i_count; ) {
nep = (struct direct *)(dirbuf + loc);
/* Trim the existing slot (NB: dsize may be zero). */
ep->d_reclen = dsize;
ep = (struct direct *)((char *)ep + dsize);
/* Read nep->d_reclen now as the bcopy() may clobber it. */
loc += nep->d_reclen;
if (nep->d_ino == 0) {
/*
* A mid-block unused entry. Such entries are
* never created by the kernel, but fsck_ffs
* can create them (and it doesn't fix them).
*
* Add up the free space, and initialise the
* relocated entry since we don't bcopy it.
*/
spacefree += nep->d_reclen;
ep->d_ino = 0;
dsize = 0;
continue;
}
dsize = DIRSIZ(FSFMT(dvp), nep);
spacefree += nep->d_reclen - dsize;
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_move(dp, nep,
dp->i_offset + ((char *)nep - dirbuf),
dp->i_offset + ((char *)ep - dirbuf));
#endif
if (DOINGSOFTDEP(dvp))
softdep_change_directoryentry_offset(dp, dirbuf,
(caddr_t)nep, (caddr_t)ep, dsize);
else
bcopy((caddr_t)nep, (caddr_t)ep, dsize);
}
/*
* Here, `ep' points to a directory entry containing `dsize' in-use
* bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
* then the entry is completely unused (dsize == 0). The value
* of ep->d_reclen is always indeterminate.
*
* Update the pointer fields in the previous entry (if any),
* copy in the new entry, and write out the block.
*/
if (ep->d_ino == 0) {
if (spacefree + dsize < newentrysize)
panic("ufs_direnter: compact1");
dirp->d_reclen = spacefree + dsize;
} else {
if (spacefree < newentrysize)
panic("ufs_direnter: compact2");
dirp->d_reclen = spacefree;
ep->d_reclen = dsize;
ep = (struct direct *)((char *)ep + dsize);
}
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
dirp->d_reclen == spacefree))
ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
#endif
bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_checkblock(dp, dirbuf -
(dp->i_offset & (DIRBLKSIZ - 1)),
dp->i_offset & ~(DIRBLKSIZ - 1));
#endif
if (DOINGSOFTDEP(dvp)) {
(void)softdep_setup_directory_add(bp, dp,
dp->i_offset + (caddr_t)ep - dirbuf,
dirp->d_ino, newdirbp, 0);
bdwrite(bp);
} else {
error = VOP_BWRITE(bp);
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* If all went well, and the directory can be shortened, proceed
* with the truncation. Note that we have to unlock the inode for
* the entry that we just entered, as the truncation may need to
* lock other inodes which can lead to deadlock if we also hold a
* lock on the newly entered node.
*/
if (error == 0 && dp->i_endoff && dp->i_endoff < DIP(dp, size)) {
if (tvp != NULL)
VOP_UNLOCK(tvp, 0, p);
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_dirtrunc(dp, dp->i_endoff);
#endif
error = UFS_TRUNCATE(dp, (off_t)dp->i_endoff,
IO_NORMAL | IO_SYNC, cr);
if (tvp != NULL)
vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
}
return (error);
}
/*
* Main code to turn off disk quotas for a filesystem. Does not change
* flags.
*/
static int
quotaoff1(struct thread *td, struct mount *mp, int type)
{
struct vnode *vp;
struct vnode *qvp, *mvp;
struct ufsmount *ump;
struct dquot *dq;
struct inode *ip;
struct ucred *cr;
int error;
ump = VFSTOUFS(mp);
UFS_LOCK(ump);
KASSERT((ump->um_qflags[type] & QTF_CLOSING) != 0,
("quotaoff1: flags are invalid"));
if ((qvp = ump->um_quotas[type]) == NULLVP) {
UFS_UNLOCK(ump);
return (0);
}
cr = ump->um_cred[type];
UFS_UNLOCK(ump);
/*
* Search vnodes associated with this mount point,
* deleting any references to quota file being closed.
*/
again:
MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
if (vp->v_type == VNON) {
VI_UNLOCK(vp);
continue;
}
if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
goto again;
}
ip = VTOI(vp);
dq = ip->i_dquot[type];
ip->i_dquot[type] = NODQUOT;
dqrele(vp, dq);
VOP_UNLOCK(vp, 0);
vrele(vp);
}
error = dqflush(qvp);
if (error != 0)
return (error);
/*
* Clear um_quotas before closing the quota vnode to prevent
* access to the closed vnode from dqget/dqsync
*/
UFS_LOCK(ump);
ump->um_quotas[type] = NULLVP;
ump->um_cred[type] = NOCRED;
UFS_UNLOCK(ump);
vn_lock(qvp, LK_EXCLUSIVE | LK_RETRY);
qvp->v_vflag &= ~VV_SYSTEM;
VOP_UNLOCK(qvp, 0);
error = vn_close(qvp, FREAD|FWRITE, td->td_ucred, td);
crfree(cr);
return (error);
}
/*
* Go through the disk queues to initiate sandbagged IO;
* go through the inodes to write those that have been modified;
* initiate the writing of the super block if it has been modified.
*
* Note: we are always called with the filesystem marked `MPBUSY'.
*/
static int
ext2_sync(struct mount *mp, int waitfor)
{
struct vnode *mvp, *vp;
struct thread *td;
struct inode *ip;
struct ext2mount *ump = VFSTOEXT2(mp);
struct m_ext2fs *fs;
int error, allerror = 0;
td = curthread;
fs = ump->um_e2fs;
if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) { /* XXX */
printf("fs = %s\n", fs->e2fs_fsmnt);
panic("ext2_sync: rofs mod");
}
/*
* Write back each (modified) inode.
*/
loop:
MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
if (vp->v_type == VNON) {
VI_UNLOCK(vp);
continue;
}
ip = VTOI(vp);
if ((ip->i_flag &
(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
(vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
waitfor == MNT_LAZY)) {
VI_UNLOCK(vp);
continue;
}
error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
if (error) {
if (error == ENOENT) {
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
goto loop;
}
continue;
}
if ((error = VOP_FSYNC(vp, waitfor, td)) != 0)
allerror = error;
VOP_UNLOCK(vp, 0);
vrele(vp);
}
/*
* Force stale file system control information to be flushed.
*/
if (waitfor != MNT_LAZY) {
vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
if ((error = VOP_FSYNC(ump->um_devvp, waitfor, td)) != 0)
allerror = error;
VOP_UNLOCK(ump->um_devvp, 0);
}
/*
* Write back modified superblock.
*/
if (fs->e2fs_fmod != 0) {
fs->e2fs_fmod = 0;
fs->e2fs->e2fs_wtime = time_second;
if ((error = ext2_cgupdate(ump, waitfor)) != 0)
allerror = error;
}
return (allerror);
}
/*
* this function handles traditional block mapping
*/
static int
ext2_ind_read(struct vop_read_args *ap)
{
struct vnode *vp;
struct inode *ip;
struct uio *uio;
FS *fs;
struct buf *bp;
daddr_t lbn, nextlbn;
off_t bytesinfile;
long size, xfersize, blkoffset;
int error, orig_resid, seqcount;
seqcount = ap->a_ioflag >> IO_SEQSHIFT;
u_short mode;
vp = ap->a_vp;
ip = VTOI(vp);
mode = ip->i_mode;
uio = ap->a_uio;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_READ)
panic("%s: mode", READ_S);
if (vp->v_type == VLNK) {
if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
panic("%s: short symlink", READ_S);
} else if (vp->v_type != VREG && vp->v_type != VDIR)
panic("%s: type %d", READ_S, vp->v_type);
#endif
orig_resid = uio->uio_resid;
KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
if (orig_resid == 0)
return (0);
KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
fs = ip->I_FS;
if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
return (EOVERFLOW);
for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
break;
lbn = lblkno(fs, uio->uio_offset);
nextlbn = lbn + 1;
size = BLKSIZE(fs, ip, lbn);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->e2fs_fsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (bytesinfile < xfersize)
xfersize = bytesinfile;
if (lblktosize(fs, nextlbn) >= ip->i_size)
error = bread(vp, lbn, size, NOCRED, &bp);
else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0)
error = cluster_read(vp, ip->i_size, lbn, size,
NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
else if (seqcount > 1) {
int nextsize = BLKSIZE(fs, ip, nextlbn);
error = breadn(vp, lbn,
size, &nextlbn, &nextsize, 1, NOCRED, &bp);
} else
error = bread(vp, lbn, size, NOCRED, &bp);
if (error) {
brelse(bp);
bp = NULL;
break;
}
/*
* 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,
(int)xfersize, uio);
if (error)
break;
bqrelse(bp);
}
if (bp != NULL)
bqrelse(bp);
if ((error == 0 || uio->uio_resid != orig_resid) &&
(vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
ip->i_flag |= IN_ACCESS;
return (error);
}
/*
* Last reference to an inode. If necessary, write or delete it.
*/
int
ufs_inactive(void *v)
{
struct vop_inactive_args *ap = v;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct fs *fs = ip->i_fs;
struct proc *p = curproc;
mode_t mode;
int error = 0, logged = 0, truncate_error = 0;
#ifdef DIAGNOSTIC
extern int prtactive;
if (prtactive && vp->v_usecount != 0)
vprint("ufs_inactive: pushing active", vp);
#endif
UFS_WAPBL_JUNLOCK_ASSERT(vp->v_mount);
/*
* Ignore inodes related to stale file handles.
*/
if (ip->i_din1 == NULL || DIP(ip, mode) == 0)
goto out;
if (DIP(ip, nlink) <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
error = UFS_WAPBL_BEGIN(vp->v_mount);
if (error)
goto out;
logged = 1;
if (getinoquota(ip) == 0)
(void)ufs_quota_free_inode(ip, NOCRED);
if (DIP(ip, size) != 0 && vp->v_mount->mnt_wapbl) {
/*
* When journaling, only truncate one indirect block at
* a time.
*/
uint64_t incr = MNINDIR(ip->i_ump) << fs->fs_bshift;
uint64_t base = NDADDR << fs->fs_bshift;
while (!error && DIP(ip, size) > base + incr) {
/*
* round down to next full indirect block
* boundary.
*/
uint64_t nsize = base +
((DIP(ip, size) - base - 1) &
~(incr - 1));
error = UFS_TRUNCATE(ip, nsize, 0, NOCRED);
if (error)
break;
UFS_WAPBL_END(vp->v_mount);
error = UFS_WAPBL_BEGIN(vp->v_mount);
if (error)
goto out;
}
}
if (error == 0) {
truncate_error = UFS_TRUNCATE(ip, (off_t)0, 0, NOCRED);
/* XXX pedro: remove me */
if (truncate_error)
printf("UFS_TRUNCATE()=%d\n", truncate_error);
}
DIP_ASSIGN(ip, rdev, 0);
mode = DIP(ip, mode);
DIP_ASSIGN(ip, mode, 0);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* Setting the mode to zero needs to wait for the inode to be
* written just as does a change to the link count. So, rather
* than creating a new entry point to do the same thing, we
* just use softdep_change_linkcnt(). Also, we can't let
* softdep co-opt us to help on its worklist, as we may end up
* trying to recycle vnodes and getting to this same point a
* couple of times, blowing the kernel stack. However, this
* could be optimized by checking if we are coming from
* vrele(), vput() or vclean() (by checking for VXLOCK) and
* just avoiding the co-opt to happen in the last case.
*/
if (DOINGSOFTDEP(vp))
softdep_change_linkcnt(ip, 1);
UFS_INODE_FREE(ip, ip->i_number, mode);
}
if (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) {
if (!logged++) {
int err;
err = UFS_WAPBL_BEGIN(vp->v_mount);
if (err) {
error = err;
goto out;
}
}
UFS_UPDATE(ip, 0);
}
if (logged)
UFS_WAPBL_END(vp->v_mount);
out:
VOP_UNLOCK(vp, 0);
/*
* If we are done with the inode, reclaim it
* so that it can be reused immediately.
*/
if (error == 0 && truncate_error == 0 &&
(ip->i_din1 == NULL || DIP(ip, mode) == 0))
vrecycle(vp, p);
return (truncate_error ? truncate_error : 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 ufs_lookup:
*
* 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
*
* ext2_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
* struct componentname *a_cnp)
*/
int
ext2_lookup(struct vop_old_lookup_args *ap)
{
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 ext2_dir_entry_2 *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;
int DIRBLKSIZ = VTOI(ap->a_dvp)->i_e2fs->s_blocksize;
bp = NULL;
slotoffset = -1;
*vpp = NULL;
vdp = ap->a_dvp;
dp = VTOI(vdp);
lockparent = flags & CNP_LOCKPARENT;
wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
/*
* We now have a segment name to search for, and a directory to search.
*/
/*
* 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 == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
slotstatus = NONE;
slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
/* was
slotneeded = (sizeof(struct direct) - MAXNAMLEN +
cnp->cn_namelen + 3) &~ 3; */
}
/*
* 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 = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
if (nameiop != NAMEI_LOOKUP || dp->i_diroff == 0 ||
dp->i_diroff > dp->i_size) {
entryoffsetinblock = 0;
dp->i_offset = 0;
numdirpasses = 1;
} else {
dp->i_offset = dp->i_diroff;
if ((entryoffsetinblock = dp->i_offset & bmask) &&
(error = EXT2_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
return (error);
numdirpasses = 2;
}
prevoff = dp->i_offset;
endsearch = roundup(dp->i_size, DIRBLKSIZ);
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);
if ((error =
EXT2_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)) != 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.
*/
ep = (struct ext2_dir_entry_2 *)
((char *)bp->b_data + entryoffsetinblock);
if (ep->rec_len == 0 ||
(dirchk && ext2_dirbadentry(vdp, ep, entryoffsetinblock))) {
int i;
ext2_dirbad(dp, dp->i_offset, "mangled entry");
i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 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 = ep->rec_len;
if (ep->inode != 0)
size -= EXT2_DIR_REC_LEN(ep->name_len);
if (size > 0) {
if (size >= slotneeded) {
slotstatus = FOUND;
slotoffset = dp->i_offset;
slotsize = ep->rec_len;
} else if (slotstatus == NONE) {
slotfreespace += size;
if (slotoffset == -1)
slotoffset = dp->i_offset;
if (slotfreespace >= slotneeded) {
slotstatus = COMPACT;
slotsize = dp->i_offset +
ep->rec_len - slotoffset;
}
}
}
}
/*
* Check for a name match.
*/
if (ep->inode) {
namlen = ep->name_len;
if (namlen == cnp->cn_namelen &&
!bcmp(cnp->cn_nameptr, ep->name,
(unsigned)namlen)) {
/*
* Save directory entry's inode number and
* reclen in ndp->ni_ufs area, and release
* directory buffer.
*/
dp->i_ino = ep->inode;
dp->i_reclen = ep->rec_len;
goto found;
}
}
prevoff = dp->i_offset;
dp->i_offset += ep->rec_len;
entryoffsetinblock += ep->rec_len;
if (ep->inode)
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 == NAMEI_CREATE || nameiop == NAMEI_RENAME) &&
dp->i_nlink != 0) {
/*
* Access for write is interpreted as allowing
* creation of files in the directory.
*/
if ((error = VOP_EACCESS(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(dp->i_size, DIRBLKSIZ);
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, DIRBLKSIZ);
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.
*/
if (!lockparent)
vn_unlock(vdp);
return (EJUSTRETURN);
}
return (ENOENT);
found:
/*
* Check that directory length properly reflects presence
* of this entry.
*/
if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->name_len)
> dp->i_size) {
ext2_dirbad(dp, dp->i_offset, "i_size too small");
dp->i_size = entryoffsetinblock+EXT2_DIR_REC_LEN(ep->name_len);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
}
brelse(bp);
/*
* Found component in pathname.
* If the final component of path name, save information
* in the cache as to where the entry was found.
*/
if (nameiop == NAMEI_LOOKUP)
dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 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 == NAMEI_DELETE) {
/*
* Write access to directory required to delete files.
*/
if ((error = VOP_EACCESS(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 & (DIRBLKSIZ - 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, NULL, 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_mode & ISVTX) &&
cred->cr_uid != 0 &&
cred->cr_uid != dp->i_uid &&
VTOI(tdp)->i_uid != cred->cr_uid) {
vput(tdp);
return (EPERM);
}
*vpp = tdp;
if (!lockparent)
vn_unlock(vdp);
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 == NAMEI_RENAME && wantparent) {
if ((error = VOP_EACCESS(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, NULL, dp->i_ino, &tdp)) != 0)
return (error);
*vpp = tdp;
if (!lockparent)
vn_unlock(vdp);
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 & CNP_ISDOTDOT) {
vn_unlock(pdp); /* race to get the inode */
if ((error = VFS_VGET(vdp->v_mount, NULL, dp->i_ino, &tdp)) != 0) {
vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
return (error);
}
if (lockparent && (error = vn_lock(pdp, LK_EXCLUSIVE))) {
vput(tdp);
return (error);
}
*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, NULL, dp->i_ino, &tdp)) != 0)
return (error);
if (!lockparent)
vn_unlock(pdp);
*vpp = tdp;
}
return (0);
}
int
ffs_fsync(void *v)
{
struct vop_fsync_args /* {
struct vnode *a_vp;
kauth_cred_t a_cred;
int a_flags;
off_t a_offlo;
off_t a_offhi;
struct lwp *a_l;
} */ *ap = v;
struct buf *bp;
int num, error, i;
struct indir ia[NIADDR + 1];
int bsize;
daddr_t blk_high;
struct vnode *vp;
struct mount *mp;
vp = ap->a_vp;
mp = vp->v_mount;
fstrans_start(mp, FSTRANS_LAZY);
if ((ap->a_offlo == 0 && ap->a_offhi == 0) || (vp->v_type != VREG)) {
error = ffs_full_fsync(vp, ap->a_flags);
goto out;
}
bsize = mp->mnt_stat.f_iosize;
blk_high = ap->a_offhi / bsize;
if (ap->a_offhi % bsize != 0)
blk_high++;
/*
* First, flush all pages in range.
*/
mutex_enter(vp->v_interlock);
error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo),
round_page(ap->a_offhi), PGO_CLEANIT |
((ap->a_flags & FSYNC_WAIT) ? PGO_SYNCIO : 0));
if (error) {
goto out;
}
#ifdef WAPBL
KASSERT(vp->v_type == VREG);
if (mp->mnt_wapbl) {
/*
* Don't bother writing out metadata if the syncer is
* making the request. We will let the sync vnode
* write it out in a single burst through a call to
* VFS_SYNC().
*/
if ((ap->a_flags & (FSYNC_DATAONLY | FSYNC_LAZY)) != 0) {
fstrans_done(mp);
return 0;
}
error = 0;
if (vp->v_tag == VT_UFS && VTOI(vp)->i_flag &
(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY |
IN_MODIFIED | IN_ACCESSED)) {
error = UFS_WAPBL_BEGIN(mp);
if (error) {
fstrans_done(mp);
return error;
}
error = ffs_update(vp, NULL, NULL, UPDATE_CLOSE |
((ap->a_flags & FSYNC_WAIT) ? UPDATE_WAIT : 0));
UFS_WAPBL_END(mp);
}
if (error || (ap->a_flags & FSYNC_NOLOG) != 0) {
fstrans_done(mp);
return error;
}
error = wapbl_flush(mp->mnt_wapbl, 0);
fstrans_done(mp);
return error;
}
#endif /* WAPBL */
/*
* Then, flush indirect blocks.
*/
if (blk_high >= NDADDR) {
error = ufs_getlbns(vp, blk_high, ia, &num);
if (error)
goto out;
mutex_enter(&bufcache_lock);
for (i = 0; i < num; i++) {
if ((bp = incore(vp, ia[i].in_lbn)) == NULL)
continue;
if ((bp->b_cflags & BC_BUSY) != 0 ||
(bp->b_oflags & BO_DELWRI) == 0)
continue;
bp->b_cflags |= BC_BUSY | BC_VFLUSH;
mutex_exit(&bufcache_lock);
bawrite(bp);
mutex_enter(&bufcache_lock);
}
mutex_exit(&bufcache_lock);
}
if (ap->a_flags & FSYNC_WAIT) {
mutex_enter(vp->v_interlock);
while (vp->v_numoutput > 0)
cv_wait(&vp->v_cv, vp->v_interlock);
mutex_exit(vp->v_interlock);
}
error = ffs_update(vp, NULL, NULL, UPDATE_CLOSE |
(((ap->a_flags & (FSYNC_WAIT | FSYNC_DATAONLY)) == FSYNC_WAIT)
? UPDATE_WAIT : 0));
if (error == 0 && ap->a_flags & FSYNC_CACHE) {
int l = 0;
VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE,
curlwp->l_cred);
}
out:
fstrans_done(mp);
return error;
}
/*
* Disable logging
*/
int
lqfs_disable(vnode_t *vp, struct fiolog *flp)
{
int error = 0;
inode_t *ip = VTOI(vp);
qfsvfs_t *qfsvfsp = ip->i_qfsvfs;
fs_lqfs_common_t *fs = VFS_FS_PTR(qfsvfsp);
#ifdef LUFS
struct lockfs lf;
struct ulockfs *ulp;
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LUFS */
flp->error = FIOLOG_ENONE;
/*
* Logging is already disabled; done
*/
if (LQFS_GET_LOGBNO(fs) == 0 || LQFS_GET_LOGP(qfsvfsp) == NULL ||
!LQFS_CAPABLE(qfsvfsp)) {
vfs_setmntopt(qfsvfsp->vfs_vfs, MNTOPT_NOLOGGING, NULL, 0);
error = 0;
goto out;
}
#ifdef LUFS
/*
* File system must be write locked to disable logging
*/
error = qfs_fiolfss(vp, &lf);
if (error) {
goto out;
}
if (!LOCKFS_IS_ULOCK(&lf)) {
flp->error = FIOLOG_EULOCK;
error = 0;
goto out;
}
lf.lf_lock = LOCKFS_WLOCK;
lf.lf_flags = 0;
lf.lf_comment = NULL;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_EWLOCK;
error = 0;
goto out;
}
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LUFS */
if (LQFS_GET_LOGP(qfsvfsp) == NULL || LQFS_GET_LOGBNO(fs) == 0) {
goto errout;
}
/*
* WE ARE COMMITTED TO DISABLING LOGGING PAST THIS POINT
*/
/*
* Disable logging:
* Suspend the reclaim thread and force the delete thread to exit.
* When a nologging mount has completed there may still be
* work for reclaim to do so just suspend this thread until
* it's [deadlock-] safe for it to continue. The delete
* thread won't be needed as qfs_iinactive() calls
* qfs_delete() when logging is disabled.
* Freeze and drain reader ops.
* Commit any outstanding reader transactions (lqfs_flush).
* Set the ``unmounted'' bit in the qfstrans struct.
* If debug, remove metadata from matamap.
* Disable matamap processing.
* NULL the trans ops table.
* Free all of the incore structs related to logging.
* Allow reader ops.
*/
#ifdef LUFS
qfs_thread_suspend(&qfsvfsp->vfs_reclaim);
qfs_thread_exit(&qfsvfsp->vfs_delete);
#else
/* QFS doesn't have file reclaim nor i-node delete threads. */
#endif /* LUFS */
vfs_lock_wait(qfsvfsp->vfs_vfs);
#ifdef LQFS_TODO_LOCKFS
ulp = &qfsvfsp->vfs_ulockfs;
mutex_enter(&ulp->ul_lock);
(void) qfs_quiesce(ulp);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
#ifdef LQFS_TODO
(void) qfs_flush(qfsvfsp->vfs_vfs);
#else
(void) lqfs_flush(qfsvfsp);
if (LQFS_GET_LOGP(qfsvfsp)) {
logmap_start_roll(LQFS_GET_LOGP(qfsvfsp));
}
#endif /* LQFS_TODO */
TRANS_MATA_UMOUNT(qfsvfsp);
LQFS_SET_DOMATAMAP(qfsvfsp, 0);
/*
* Free all of the incore structs
* Aquire the ufs_scan_lock before de-linking the mtm data
* structure so that we keep ufs_sync() and ufs_update() away
* when they execute the ufs_scan_inodes() run while we're in
* progress of enabling/disabling logging.
*/
mutex_enter(&qfs_scan_lock);
(void) lqfs_unsnarf(qfsvfsp);
mutex_exit(&qfs_scan_lock);
#ifdef LQFS_TODO_LOCKFS
atomic_add_long(&ufs_quiesce_pend, -1);
mutex_exit(&ulp->ul_lock);
#else
/* QFS doesn't do this yet. */
#endif /* LQFS_TODO_LOCKFS */
vfs_setmntopt(qfsvfsp->vfs_vfs, MNTOPT_NOLOGGING, NULL, 0);
vfs_unlock(qfsvfsp->vfs_vfs);
LQFS_SET_FS_ROLLED(fs, FS_ALL_ROLLED);
LQFS_SET_NOLOG_SI(qfsvfsp, 0);
/*
* Free the log space and mark the superblock as FSACTIVE
*/
(void) lqfs_free(qfsvfsp);
#ifdef LUFS
/*
* Allow the reclaim thread to continue.
*/
qfs_thread_continue(&qfsvfsp->vfs_reclaim);
#else
/* QFS doesn't have a file reclaim thread. */
#endif /* LUFS */
#ifdef LQFS_TODO_LOCKFS
/*
* Unlock the file system
*/
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_ENOULOCK;
}
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_LOCKFS */
error = 0;
goto out;
errout:
#ifdef LQFS_LOCKFS
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
(void) qfs_fiolfs(vp, &lf, 1);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_LOCKFS */
out:
mutex_enter(&ip->mp->ms.m_waitwr_mutex);
ip->mp->mt.fi_status |= FS_LOGSTATE_KNOWN;
mutex_exit(&ip->mp->ms.m_waitwr_mutex);
return (error);
}
/*
* Enable logging
*/
int
lqfs_enable(struct vnode *vp, struct fiolog *flp, cred_t *cr)
{
int error;
inode_t *ip = VTOI(vp);
qfsvfs_t *qfsvfsp = ip->i_qfsvfs;
fs_lqfs_common_t *fs = VFS_FS_PTR(qfsvfsp);
ml_unit_t *ul;
#ifdef LQFS_TODO_LOCKFS
int reclaim = 0;
struct lockfs lf;
struct ulockfs *ulp;
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
vfs_t *vfsp = qfsvfsp->vfs_vfs;
uint64_t tmp_nbytes_actual;
char fsclean;
sam_sblk_t *sblk = qfsvfsp->mi.m_sbp;
/*
* File system is not capable of logging.
*/
if (!LQFS_CAPABLE(qfsvfsp)) {
flp->error = FIOLOG_ENOTSUP;
error = 0;
goto out;
}
if (!SAM_MAGIC_V2A_OR_HIGHER(&sblk->info.sb)) {
cmn_err(CE_WARN, "SAM-QFS: %s: Not enabling logging, "
" file system is not version 2A.", qfsvfsp->mt.fi_name);
cmn_err(CE_WARN, "\tUpgrade file system with samfsck -u "
"first.");
flp->error = FIOLOG_ENOTSUP;
error = 0;
goto out;
}
if (LQFS_GET_LOGBNO(fs)) {
error = lqfs_log_validate(qfsvfsp, flp, cr);
}
/*
* Check if logging is already enabled
*/
if (LQFS_GET_LOGP(qfsvfsp)) {
flp->error = FIOLOG_ETRANS;
/* for root ensure logging option is set */
vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
error = 0;
goto out;
}
/*
* Come back here to recheck if we had to disable the log.
*/
recheck:
error = 0;
flp->error = FIOLOG_ENONE;
/*
* Adjust requested log size
*/
flp->nbytes_actual = flp->nbytes_requested;
if (flp->nbytes_actual == 0) {
tmp_nbytes_actual =
(((uint64_t)FS_SIZE(fs)) / ldl_divisor) << FS_FSHIFT(fs);
flp->nbytes_actual = (uint_t)MIN(tmp_nbytes_actual, INT_MAX);
}
flp->nbytes_actual = MAX(flp->nbytes_actual, ldl_minlogsize);
flp->nbytes_actual = MIN(flp->nbytes_actual, ldl_maxlogsize);
flp->nbytes_actual = blkroundup(fs, flp->nbytes_actual);
/*
* logging is enabled and the log is the right size; done
*/
ul = LQFS_GET_LOGP(qfsvfsp);
if (ul && LQFS_GET_LOGBNO(fs) &&
(flp->nbytes_actual == ul->un_requestsize)) {
vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
error = 0;
goto out;
}
/*
* Readonly file system
*/
if (FS_RDONLY(fs)) {
flp->error = FIOLOG_EROFS;
error = 0;
goto out;
}
#ifdef LQFS_TODO_LOCKFS
/*
* File system must be write locked to enable logging
*/
error = qfs_fiolfss(vp, &lf);
if (error) {
goto out;
}
if (!LOCKFS_IS_ULOCK(&lf)) {
flp->error = FIOLOG_EULOCK;
error = 0;
goto out;
}
lf.lf_lock = LOCKFS_WLOCK;
lf.lf_flags = 0;
lf.lf_comment = NULL;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_EWLOCK;
error = 0;
goto out;
}
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
/*
* Grab appropriate locks to synchronize with the rest
* of the system
*/
vfs_lock_wait(vfsp);
#ifdef LQFS_TODO_LOCKFS
ulp = &ufsvfsp->vfs_ulockfs;
mutex_enter(&ulp->ul_lock);
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
/*
* File system must be fairly consistent to enable logging
*/
fsclean = LQFS_GET_FS_CLEAN(fs);
if (fsclean != FSLOG &&
fsclean != FSACTIVE &&
fsclean != FSSTABLE &&
fsclean != FSCLEAN) {
flp->error = FIOLOG_ECLEAN;
goto unlockout;
}
#ifdef LUFS
/*
* A write-locked file system is only active if there are
* open deleted files; so remember to set FS_RECLAIM later.
*/
if (LQFS_GET_FS_CLEAN(fs) == FSACTIVE) {
reclaim = FS_RECLAIM;
}
#else
/* QFS doesn't have a reclaim file thread. */
#endif /* LUFS */
/*
* Logging is already enabled; must be changing the log's size
*/
if (LQFS_GET_LOGBNO(fs) && LQFS_GET_LOGP(qfsvfsp)) {
#ifdef LQFS_TODO_LOCKFS
/*
* Before we can disable logging, we must give up our
* lock. As a consequence of unlocking and disabling the
* log, the fs structure may change. Because of this, when
* disabling is complete, we will go back to recheck to
* repeat all of the checks that we performed to get to
* this point. Disabling sets fs->fs_logbno to 0, so this
* will not put us into an infinite loop.
*/
mutex_exit(&ulp->ul_lock);
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
vfs_unlock(vfsp);
#ifdef LQFS_TODO_LOCKFS
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_ENOULOCK;
error = 0;
goto out;
}
#else
/* QFS doesn't really support lockfs. */
#endif /* LQFS_TODO_LOCKFS */
error = lqfs_disable(vp, flp);
if (error || (flp->error != FIOLOG_ENONE)) {
error = 0;
goto out;
}
goto recheck;
}
error = lqfs_alloc(qfsvfsp, flp, cr);
if (error) {
goto errout;
}
#ifdef LUFS
#else
if ((error = lqfs_log_validate(qfsvfsp, flp, cr)) != 0) {
goto errout;
}
#endif /* LUFS */
/*
* Create all of the incore structs
*/
error = lqfs_snarf(qfsvfsp, fs, 0);
if (error) {
goto errout;
}
/*
* DON'T ``GOTO ERROUT'' PAST THIS POINT
*/
/*
* Pretend we were just mounted with logging enabled
* freeze and drain the file system of readers
* Get the ops vector
* If debug, record metadata locations with log subsystem
* Start the delete thread
* Start the reclaim thread, if necessary
* Thaw readers
*/
vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
TRANS_DOMATAMAP(qfsvfsp);
TRANS_MATA_MOUNT(qfsvfsp);
TRANS_MATA_SI(qfsvfsp, fs);
#ifdef LUFS
qfs_thread_start(&qfsvfsp->vfs_delete, qfs_thread_delete, vfsp);
if (fs->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) {
fs->fs_reclaim &= ~FS_RECLAIM;
fs->fs_reclaim |= FS_RECLAIMING;
qfs_thread_start(&qfsvfsp->vfs_reclaim,
qfs_thread_reclaim, vfsp);
} else {
fs->fs_reclaim |= reclaim;
}
#else
/* QFS doesn't have file reclaim nor i-node delete threads. */
#endif /* LUFS */
#ifdef LUFS
mutex_exit(&ulp->ul_lock);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LUFS */
vfs_unlock(vfsp);
#ifdef LQFS_TODO_LOCKFS
/*
* Unlock the file system
*/
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
error = qfs_fiolfs(vp, &lf, 1);
if (error) {
flp->error = FIOLOG_ENOULOCK;
error = 0;
goto out;
}
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
/*
* There's nothing in the log yet (we've just allocated it)
* so directly write out the super block.
* Note, we have to force this sb out to disk
* (not just to the log) so that if we crash we know we are logging
*/
VFS_LOCK_MUTEX_ENTER(qfsvfsp);
LQFS_SET_FS_CLEAN(fs, FSLOG);
LQFS_SET_FS_ROLLED(fs, FS_NEED_ROLL); /* Mark the fs as unrolled */
#ifdef LUFS
QFS_BWRITE2(NULL, qfsvfsp->vfs_bufp);
#else
sam_update_sblk(qfsvfsp, 0, 0, TRUE);
#endif /* LUFS */
VFS_LOCK_MUTEX_EXIT(qfsvfsp);
error = 0;
goto out;
errout:
/*
* Aquire the qfs_scan_lock before de-linking the mtm data
* structure so that we keep qfs_sync() and qfs_update() away
* when they execute the ufs_scan_inodes() run while we're in
* progress of enabling/disabling logging.
*/
mutex_enter(&qfs_scan_lock);
(void) lqfs_unsnarf(qfsvfsp);
mutex_exit(&qfs_scan_lock);
(void) lqfs_free(qfsvfsp);
unlockout:
#ifdef LQFS_TODO_LOCKFS
mutex_exit(&ulp->ul_lock);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
vfs_unlock(vfsp);
#ifdef LQFS_TODO_LOCKFS
lf.lf_lock = LOCKFS_ULOCK;
lf.lf_flags = 0;
(void) qfs_fiolfs(vp, &lf, 1);
#else
/* QFS doesn't really support LOCKFS. */
#endif /* LQFS_TODO_LOCKFS */
out:
mutex_enter(&ip->mp->ms.m_waitwr_mutex);
ip->mp->mt.fi_status |= FS_LOGSTATE_KNOWN;
mutex_exit(&ip->mp->ms.m_waitwr_mutex);
return (error);
}
/*
* Real work associated with removing an extended attribute from a vnode.
* Assumes the attribute lock has already been grabbed.
*/
static int
ufs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
struct ucred *cred, struct proc *p)
{
struct ufs_extattr_list_entry *attribute;
struct ufs_extattr_header ueh;
struct iovec local_aiov;
struct uio local_aio;
struct mount *mp = vp->v_mount;
struct ufsmount *ump = VFSTOUFS(mp);
struct inode *ip = VTOI(vp);
off_t base_offset;
int error = 0, ioflag;
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
return (EOPNOTSUPP);
if (!ufs_extattr_valid_attrname(attrnamespace, name))
return (EINVAL);
attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
if (!attribute)
/* XXX: ENOENT here will eventually be ENOATTR. */
return (ENOENT);
if ((error = ufs_extattr_credcheck(vp, attribute, cred, p, IWRITE)))
return (error);
/*
* Find base offset of header in file based on file header size, and
* data header size + maximum data size, indexed by inode number.
*/
base_offset = sizeof(struct ufs_extattr_fileheader) +
ip->i_number * (sizeof(struct ufs_extattr_header) +
attribute->uele_fileheader.uef_size);
/*
* Check to see if currently defined.
*/
bzero(&ueh, sizeof(struct ufs_extattr_header));
local_aiov.iov_base = (caddr_t) &ueh;
local_aiov.iov_len = sizeof(struct ufs_extattr_header);
local_aio.uio_iov = &local_aiov;
local_aio.uio_iovcnt = 1;
local_aio.uio_rw = UIO_READ;
local_aio.uio_segflg = UIO_SYSSPACE;
local_aio.uio_procp = p;
local_aio.uio_offset = base_offset;
local_aio.uio_resid = sizeof(struct ufs_extattr_header);
VOP_LEASE(attribute->uele_backing_vnode, p, cred, LEASE_WRITE);
/*
* Don't need to get the lock on the backing vnode if the vnode we're
* modifying is it, as we already hold the lock.
*/
if (attribute->uele_backing_vnode != vp)
vn_lock(attribute->uele_backing_vnode,
LK_EXCLUSIVE | LK_NOPAUSE | LK_RETRY, p);
error = VOP_READ(attribute->uele_backing_vnode, &local_aio,
IO_NODELOCKED, ump->um_extattr.uepm_ucred);
if (error)
goto vopunlock_exit;
/* Defined? */
if ((ueh.ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0) {
/* XXX: ENOENT here will eventually be ENOATTR. */
error = ENOENT;
goto vopunlock_exit;
}
/* Valid for the current inode generation? */
if (ueh.ueh_i_gen != ip->i_ffs_gen) {
/*
* The inode itself has a different generation number than
* the attribute data. For now, the best solution is to
* coerce this to undefined, and let it get cleaned up by
* the next write or extattrctl clean.
*/
printf("ufs_extattr_rm: inode number inconsistency (%d, %d)\n",
ueh.ueh_i_gen, ip->i_ffs_gen);
/* XXX: ENOENT here will eventually be ENOATTR. */
error = ENOENT;
goto vopunlock_exit;
}
/* Flag it as not in use. */
ueh.ueh_flags = 0;
ueh.ueh_len = 0;
local_aiov.iov_base = (caddr_t) &ueh;
local_aiov.iov_len = sizeof(struct ufs_extattr_header);
local_aio.uio_iov = &local_aiov;
local_aio.uio_iovcnt = 1;
local_aio.uio_rw = UIO_WRITE;
local_aio.uio_segflg = UIO_SYSSPACE;
local_aio.uio_procp = p;
local_aio.uio_offset = base_offset;
local_aio.uio_resid = sizeof(struct ufs_extattr_header);
ioflag = IO_NODELOCKED;
if (ufs_extattr_sync)
ioflag |= IO_SYNC;
error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
ump->um_extattr.uepm_ucred);
if (error)
goto vopunlock_exit;
if (local_aio.uio_resid != 0)
error = ENXIO;
vopunlock_exit:
VOP_UNLOCK(attribute->uele_backing_vnode, 0, p);
return (error);
}
/*
* Real work associated with retrieving a named attribute--assumes that
* the attribute lock has already been grabbed.
*/
static int
ufs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
struct uio *uio, size_t *size, struct ucred *cred, struct proc *p)
{
struct ufs_extattr_list_entry *attribute;
struct ufs_extattr_header ueh;
struct iovec local_aiov;
struct uio local_aio;
struct mount *mp = vp->v_mount;
struct ufsmount *ump = VFSTOUFS(mp);
struct inode *ip = VTOI(vp);
off_t base_offset;
size_t len, old_len;
int error = 0;
if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
return (EOPNOTSUPP);
if (strlen(name) == 0) {
/* XXX retrieve attribute lists. */
/* XXX should probably be checking for name == NULL? */
return (EINVAL);
}
attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
if (!attribute)
/* XXX: ENOENT here will eventually be ENOATTR. */
return (ENOENT);
if ((error = ufs_extattr_credcheck(vp, attribute, cred, p, IREAD)))
return (error);
/*
* Allow only offsets of zero to encourage the read/replace
* extended attribute semantic. Otherwise we can't guarantee
* atomicity, as we don't provide locks for extended attributes.
*/
if (uio != NULL && uio->uio_offset != 0)
return (ENXIO);
/*
* Find base offset of header in file based on file header size, and
* data header size + maximum data size, indexed by inode number.
*/
base_offset = sizeof(struct ufs_extattr_fileheader) +
ip->i_number * (sizeof(struct ufs_extattr_header) +
attribute->uele_fileheader.uef_size);
/*
* Read in the data header to see if the data is defined, and if so
* how much.
*/
bzero(&ueh, sizeof(struct ufs_extattr_header));
local_aiov.iov_base = (caddr_t) &ueh;
local_aiov.iov_len = sizeof(struct ufs_extattr_header);
local_aio.uio_iov = &local_aiov;
local_aio.uio_iovcnt = 1;
local_aio.uio_rw = UIO_READ;
local_aio.uio_segflg = UIO_SYSSPACE;
local_aio.uio_procp = p;
local_aio.uio_offset = base_offset;
local_aio.uio_resid = sizeof(struct ufs_extattr_header);
/*
* Acquire locks.
*/
VOP_LEASE(attribute->uele_backing_vnode, p, cred, LEASE_READ);
/*
* Don't need to get a lock on the backing file if the getattr is
* being applied to the backing file, as the lock is already held.
*/
if (attribute->uele_backing_vnode != vp)
vn_lock(attribute->uele_backing_vnode, LK_SHARED |
LK_NOPAUSE | LK_RETRY, p);
error = VOP_READ(attribute->uele_backing_vnode, &local_aio,
IO_NODELOCKED, ump->um_extattr.uepm_ucred);
if (error)
goto vopunlock_exit;
/* Defined? */
if ((ueh.ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0) {
/* XXX: ENOENT here will eventually be ENOATTR. */
error = ENOENT;
goto vopunlock_exit;
}
/* Valid for the current inode generation? */
if (ueh.ueh_i_gen != ip->i_ffs_gen) {
/*
* The inode itself has a different generation number
* than the attribute data. For now, the best solution
* is to coerce this to undefined, and let it get cleaned
* up by the next write or extattrctl clean.
*/
printf("ufs_extattr_get: inode number inconsistency (%d, %d)\n",
ueh.ueh_i_gen, ip->i_ffs_gen);
/* XXX: ENOENT here will eventually be ENOATTR. */
error = ENOENT;
goto vopunlock_exit;
}
/* Local size consistency check. */
if (ueh.ueh_len > attribute->uele_fileheader.uef_size) {
error = ENXIO;
goto vopunlock_exit;
}
/* Return full data size if caller requested it. */
if (size != NULL)
*size = ueh.ueh_len;
/* Return data if the caller requested it. */
if (uio != NULL) {
/* Allow for offset into the attribute data. */
uio->uio_offset = base_offset + sizeof(struct
ufs_extattr_header);
/*
* Figure out maximum to transfer -- use buffer size and
* local data limit.
*/
len = MIN(uio->uio_resid, ueh.ueh_len);
old_len = uio->uio_resid;
uio->uio_resid = len;
error = VOP_READ(attribute->uele_backing_vnode, uio,
IO_NODELOCKED, ump->um_extattr.uepm_ucred);
if (error)
goto vopunlock_exit;
uio->uio_resid = old_len - (len - uio->uio_resid);
}
vopunlock_exit:
if (uio != NULL)
uio->uio_offset = 0;
if (attribute->uele_backing_vnode != vp)
VOP_UNLOCK(attribute->uele_backing_vnode, 0, p);
return (error);
}
/*
* Real work associated with setting a vnode's extended attributes;
* assumes that the attribute lock has already been grabbed.
*/
static int
ufs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
struct uio *uio, struct ucred *cred, struct proc *p)
{
struct ufs_extattr_list_entry *attribute;
struct ufs_extattr_header ueh;
struct iovec local_aiov;
struct uio local_aio;
struct mount *mp = vp->v_mount;
struct ufsmount *ump = VFSTOUFS(mp);
struct inode *ip = VTOI(vp);
off_t base_offset;
int error = 0, ioflag;
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
return (EOPNOTSUPP);
if (!ufs_extattr_valid_attrname(attrnamespace, name))
return (EINVAL);
attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
if (!attribute)
/* XXX: ENOENT here will eventually be ENOATTR. */
return (ENOENT);
if ((error = ufs_extattr_credcheck(vp, attribute, cred, p, IWRITE)))
return (error);
/*
* Early rejection of invalid offsets/length.
* Reject: any offset but 0 (replace)
* Any size greater than attribute size limit
*/
if (uio->uio_offset != 0 ||
uio->uio_resid > attribute->uele_fileheader.uef_size)
return (ENXIO);
/*
* Find base offset of header in file based on file header size, and
* data header size + maximum data size, indexed by inode number.
*/
base_offset = sizeof(struct ufs_extattr_fileheader) +
ip->i_number * (sizeof(struct ufs_extattr_header) +
attribute->uele_fileheader.uef_size);
/*
* Write out a data header for the data.
*/
ueh.ueh_len = uio->uio_resid;
ueh.ueh_flags = UFS_EXTATTR_ATTR_FLAG_INUSE;
ueh.ueh_i_gen = ip->i_ffs_gen;
local_aiov.iov_base = (caddr_t) &ueh;
local_aiov.iov_len = sizeof(struct ufs_extattr_header);
local_aio.uio_iov = &local_aiov;
local_aio.uio_iovcnt = 1;
local_aio.uio_rw = UIO_WRITE;
local_aio.uio_segflg = UIO_SYSSPACE;
local_aio.uio_procp = p;
local_aio.uio_offset = base_offset;
local_aio.uio_resid = sizeof(struct ufs_extattr_header);
/*
* Acquire locks.
*/
VOP_LEASE(attribute->uele_backing_vnode, p, cred, LEASE_WRITE);
/*
* Don't need to get a lock on the backing file if the setattr is
* being applied to the backing file, as the lock is already held.
*/
if (attribute->uele_backing_vnode != vp)
vn_lock(attribute->uele_backing_vnode,
LK_EXCLUSIVE | LK_NOPAUSE | LK_RETRY, p);
ioflag = IO_NODELOCKED;
if (ufs_extattr_sync)
ioflag |= IO_SYNC;
error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
ump->um_extattr.uepm_ucred);
if (error)
goto vopunlock_exit;
if (local_aio.uio_resid != 0) {
error = ENXIO;
goto vopunlock_exit;
}
/*
* Write out user data.
*/
uio->uio_offset = base_offset + sizeof(struct ufs_extattr_header);
ioflag = IO_NODELOCKED;
if (ufs_extattr_sync)
ioflag |= IO_SYNC;
error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
ump->um_extattr.uepm_ucred);
vopunlock_exit:
uio->uio_offset = 0;
if (attribute->uele_backing_vnode != vp)
VOP_UNLOCK(attribute->uele_backing_vnode, 0, p);
return (error);
}
/*
* Vnode op for reading directories.
*
* This function has to convert directory entries from the on-disk
* format to the format defined by <sys/dirent.h>. Unfortunately, the
* conversion will blow up some entries by four bytes, so it can't be
* done in place. Instead, the conversion is done entry by entry and
* 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
ext2_readdir(struct vop_readdir_args *ap)
{
struct uio *uio = ap->a_uio;
int count, error;
struct ext2fs_direct_2 *edp, *dp;
int ncookies;
struct dirent dstdp;
struct uio auio;
struct iovec aiov;
caddr_t dirbuf;
int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
int readcnt;
off_t startoffset = uio->uio_offset;
count = uio->uio_resid;
/*
* Avoid complications for partial directory entries by adjusting
* the i/o to end at a block boundary. Don't give up (like ufs
* does) if the initial adjustment gives a negative count, since
* many callers don't supply a large enough buffer. The correct
* size is a little larger than DIRBLKSIZ to allow for expansion
* of directory entries, but some callers just use 512.
*/
count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
if (count <= 0)
count += DIRBLKSIZ;
auio = *uio;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_resid = count;
auio.uio_segflg = UIO_SYSSPACE;
aiov.iov_len = count;
dirbuf = malloc(count, M_TEMP, M_WAITOK);
aiov.iov_base = dirbuf;
error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
if (error == 0) {
readcnt = count - auio.uio_resid;
edp = (struct ext2fs_direct_2 *)&dirbuf[readcnt];
ncookies = 0;
bzero(&dstdp, offsetof(struct dirent, d_name));
for (dp = (struct ext2fs_direct_2 *)dirbuf;
!error && uio->uio_resid > 0 && dp < edp; ) {
/*-
* "New" ext2fs directory entries differ in 3 ways
* from ufs on-disk ones:
* - the name is not necessarily NUL-terminated.
* - the file type field always exists and always
* follows the name length field.
* - the file type is encoded in a different way.
*
* "Old" ext2fs directory entries need no special
* conversions, since they are binary compatible
* with "new" entries having a file type of 0 (i.e.,
* EXT2_FT_UNKNOWN). Splitting the old name length
* field didn't make a mess like it did in ufs,
* because ext2fs uses a machine-independent disk
* layout.
*/
dstdp.d_fileno = dp->e2d_ino;
dstdp.d_type = FTTODT(dp->e2d_type);
dstdp.d_namlen = dp->e2d_namlen;
dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
bzero(dstdp.d_name + dstdp.d_namlen,
dstdp.d_reclen - offsetof(struct dirent, d_name) -
dstdp.d_namlen);
if (dp->e2d_reclen > 0) {
if(dstdp.d_reclen <= uio->uio_resid) {
/* advance dp */
dp = (struct ext2fs_direct_2 *)
((char *)dp + dp->e2d_reclen);
error =
uiomove(&dstdp, dstdp.d_reclen, uio);
if (!error)
ncookies++;
} else
break;
} else {
error = EIO;
break;
}
}
/* we need to correct uio_offset */
uio->uio_offset = startoffset + (caddr_t)dp - dirbuf;
if (!error && ap->a_ncookies != NULL) {
u_long *cookiep, *cookies, *ecookies;
off_t off;
if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
panic("ext2_readdir: unexpected uio from NFS server");
cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
M_WAITOK);
off = startoffset;
for (dp = (struct ext2fs_direct_2 *)dirbuf,
cookiep = cookies, ecookies = cookies + ncookies;
cookiep < ecookies;
dp = (struct ext2fs_direct_2 *)((caddr_t) dp + dp->e2d_reclen)) {
off += dp->e2d_reclen;
*cookiep++ = (u_long) off;
}
*ap->a_ncookies = ncookies;
*ap->a_cookies = cookies;
}
}
/*
* Remove a directory entry after a call to namei, using
* the parameters which it left in nameidata. The entry
* dp->i_offset contains the offset into the directory of the
* entry to be eliminated. The dp->i_count field contains the
* size of the previous record in the directory. If this
* is 0, the first entry is being deleted, so we need only
* zero the inode number to mark the entry as free. If the
* entry is not the first in the directory, we must reclaim
* the space of the now empty record by adding the record size
* to the size of the previous entry.
*/
int
ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
{
struct inode *dp;
struct direct *ep;
struct buf *bp;
int error;
dp = VTOI(dvp);
if ((error = UFS_BUFATOFF(dp,
(off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
return (error);
#ifdef UFS_DIRHASH
/*
* Remove the dirhash entry. This is complicated by the fact
* that `ep' is the previous entry when dp->i_count != 0.
*/
if (dp->i_dirhash != NULL)
ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
(struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
#endif
if (dp->i_count == 0) {
/*
* First entry in block: set d_ino to zero.
*/
ep->d_ino = 0;
} else {
/*
* Collapse new free space into previous entry.
*/
ep->d_reclen += dp->i_reclen;
}
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_checkblock(dp, (char *)ep -
((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
dp->i_offset & ~(DIRBLKSIZ - 1));
#endif
if (DOINGSOFTDEP(dvp)) {
if (ip) {
ip->i_effnlink--;
softdep_change_linkcnt(ip, 0);
softdep_setup_remove(bp, dp, ip, isrmdir);
}
if (softdep_slowdown(dvp)) {
error = bwrite(bp);
} else {
bdwrite(bp);
error = 0;
}
} else {
if (ip) {
ip->i_effnlink--;
DIP_ADD(ip, nlink, -1);
ip->i_flag |= IN_CHANGE;
}
if (DOINGASYNC(dvp) && dp->i_count != 0) {
bdwrite(bp);
error = 0;
} else
error = bwrite(bp);
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
return (error);
}
/*
* Balloc defines the structure of filesystem storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS1. Below is
* the allocation strategy for UFS2.
*/
int
ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
struct ucred *cred, int flags, struct buf **bpp)
{
struct inode *ip;
struct ufs1_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
ufs1_daddr_t nb;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[NIADDR + 2];
int deallocated, osize, nsize, num, i, error;
ufs2_daddr_t newb;
ufs1_daddr_t *bap, pref;
ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int gbflags, reclaimed;
ip = VTOI(vp);
dp = ip->i_din1;
fs = ip->i_fs;
ump = ip->i_ump;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs1: blk too big");
*bpp = NULL;
if (flags & IO_EXT)
return (EOPNOTSUPP);
if (lbn < 0)
return (EFBIG);
gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lbn, osize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs1(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, -indirs[0].in_off - 1,
(ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
pref = newb + fs->fs_frag;
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs1_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
/*
* If parent indirect has just been allocated, try to cluster
* immediately following it.
*/
if (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, i - num - 1,
(ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
pref = newb + fs->fs_frag;
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
/*
* If allocating metadata at the front of the cylinder
* group and parent indirect block has just been allocated,
* then cluster next to it if it is the first indirect in
* the file. Otherwise it has been allocated in the metadata
* area, so we want to find our own place out in the data area.
*/
if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0))
pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off,
&bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, gbflags, &nbp);
} else {
error = bread_gb(vp, lbn, (int)fs->fs_bsize, NOCRED,
gbflags, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
/*
* Reload all incore data for a filesystem (used after running fsck on
* the root filesystem and finding things to fix). The filesystem must
* be mounted read-only.
*
* Things to do to update the mount:
* 1) invalidate all cached meta-data.
* 2) re-read superblock from disk.
* 3) re-read summary information from disk.
* 4) invalidate all inactive vnodes.
* 5) invalidate all cached file data.
* 6) re-read inode data for all active vnodes.
*/
int
ext2fs_reload(struct mount *mp, kauth_cred_t cred, struct lwp *l)
{
struct vnode *vp, *mvp, *devvp;
struct inode *ip;
struct buf *bp;
struct m_ext2fs *fs;
struct ext2fs *newfs;
int i, error;
void *cp;
struct ufsmount *ump;
if ((mp->mnt_flag & MNT_RDONLY) == 0)
return (EINVAL);
ump = VFSTOUFS(mp);
/*
* Step 1: invalidate all cached meta-data.
*/
devvp = ump->um_devvp;
vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
error = vinvalbuf(devvp, 0, cred, l, 0, 0);
VOP_UNLOCK(devvp);
if (error)
panic("ext2fs_reload: dirty1");
/*
* Step 2: re-read superblock from disk.
*/
error = bread(devvp, SBLOCK, SBSIZE, NOCRED, 0, &bp);
if (error) {
return (error);
}
newfs = (struct ext2fs *)bp->b_data;
error = ext2fs_checksb(newfs, (mp->mnt_flag & MNT_RDONLY) != 0);
if (error) {
brelse(bp, 0);
return (error);
}
fs = ump->um_e2fs;
/*
* copy in new superblock, and compute in-memory values
*/
e2fs_sbload(newfs, &fs->e2fs);
fs->e2fs_ncg =
howmany(fs->e2fs.e2fs_bcount - fs->e2fs.e2fs_first_dblock,
fs->e2fs.e2fs_bpg);
fs->e2fs_fsbtodb = fs->e2fs.e2fs_log_bsize + LOG_MINBSIZE - DEV_BSHIFT;
fs->e2fs_bsize = MINBSIZE << fs->e2fs.e2fs_log_bsize;
fs->e2fs_bshift = LOG_MINBSIZE + fs->e2fs.e2fs_log_bsize;
fs->e2fs_qbmask = fs->e2fs_bsize - 1;
fs->e2fs_bmask = ~fs->e2fs_qbmask;
fs->e2fs_ngdb =
howmany(fs->e2fs_ncg, fs->e2fs_bsize / sizeof(struct ext2_gd));
fs->e2fs_ipb = fs->e2fs_bsize / EXT2_DINODE_SIZE(fs);
fs->e2fs_itpg = fs->e2fs.e2fs_ipg / fs->e2fs_ipb;
brelse(bp, 0);
/*
* Step 3: re-read summary information from disk.
*/
for (i = 0; i < fs->e2fs_ngdb; i++) {
error = bread(devvp ,
EXT2_FSBTODB(fs, fs->e2fs.e2fs_first_dblock +
1 /* superblock */ + i),
fs->e2fs_bsize, NOCRED, 0, &bp);
if (error) {
return (error);
}
e2fs_cgload((struct ext2_gd *)bp->b_data,
&fs->e2fs_gd[i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
fs->e2fs_bsize);
brelse(bp, 0);
}
/* Allocate a marker vnode. */
mvp = vnalloc(mp);
/*
* NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
* and vclean() can be called indirectly
*/
mutex_enter(&mntvnode_lock);
loop:
for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
vmark(mvp, vp);
if (vp->v_mount != mp || vismarker(vp))
continue;
/*
* Step 4: invalidate all inactive vnodes.
*/
if (vrecycle(vp, &mntvnode_lock)) {
mutex_enter(&mntvnode_lock);
(void)vunmark(mvp);
goto loop;
}
/*
* Step 5: invalidate all cached file data.
*/
mutex_enter(vp->v_interlock);
mutex_exit(&mntvnode_lock);
if (vget(vp, LK_EXCLUSIVE)) {
mutex_enter(&mntvnode_lock);
(void)vunmark(mvp);
goto loop;
}
if (vinvalbuf(vp, 0, cred, l, 0, 0))
panic("ext2fs_reload: dirty2");
/*
* Step 6: re-read inode data for all active vnodes.
*/
ip = VTOI(vp);
error = bread(devvp, EXT2_FSBTODB(fs, ino_to_fsba(fs, ip->i_number)),
(int)fs->e2fs_bsize, NOCRED, 0, &bp);
if (error) {
vput(vp);
mutex_enter(&mntvnode_lock);
(void)vunmark(mvp);
break;
}
cp = (char *)bp->b_data +
(ino_to_fsbo(fs, ip->i_number) * EXT2_DINODE_SIZE(fs));
e2fs_iload((struct ext2fs_dinode *)cp, ip->i_din.e2fs_din);
ext2fs_set_inode_guid(ip);
brelse(bp, 0);
vput(vp);
mutex_enter(&mntvnode_lock);
}
mutex_exit(&mntvnode_lock);
vnfree(mvp);
return (error);
}
/*
* this is exactly what we do here - 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
*
* ext2_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
*/
int
ext2_readdir(struct vop_readdir_args *ap)
{
struct uio *uio = ap->a_uio;
int count, error;
struct ext2_dir_entry_2 *edp, *dp;
int ncookies;
struct uio auio;
struct iovec aiov;
caddr_t dirbuf;
int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->s_blocksize;
int readcnt, retval;
off_t startoffset = uio->uio_offset;
if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
return(error);
count = uio->uio_resid;
/*
* Avoid complications for partial directory entries by adjusting
* the i/o to end at a block boundary. Don't give up (like ufs
* does) if the initial adjustment gives a negative count, since
* many callers don't supply a large enough buffer. The correct
* size is a little larger than DIRBLKSIZ to allow for expansion
* of directory entries, but some callers just use 512.
*/
count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
if (count <= 0)
count += DIRBLKSIZ;
if (count > MAXBSIZE) /* limit to a reasonable size */
count = MAXBSIZE;
#ifdef EXT2FS_DEBUG
kprintf("ext2_readdir: uio_offset = %lld, uio_resid = %d, count = %d\n",
uio->uio_offset, uio->uio_resid, count);
#endif
auio = *uio;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_resid = count;
auio.uio_segflg = UIO_SYSSPACE;
aiov.iov_len = count;
dirbuf = kmalloc(count, M_TEMP, M_WAITOK);
aiov.iov_base = dirbuf;
error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
if (error == 0) {
readcnt = count - auio.uio_resid;
edp = (struct ext2_dir_entry_2 *)&dirbuf[readcnt];
ncookies = 0;
for (dp = (struct ext2_dir_entry_2 *)dirbuf;
!error && uio->uio_resid > 0 && dp < edp; ) {
/*-
* "New" ext2fs directory entries differ in 3 ways
* from ufs on-disk ones:
* - the name is not necessarily NUL-terminated.
* - the file type field always exists and always
* follows the name length field.
* - the file type is encoded in a different way.
*
* "Old" ext2fs directory entries need no special
* conversions, since they binary compatible with
* "new" entries having a file type of 0 (i.e.,
* EXT2_FT_UNKNOWN). Splitting the old name length
* field didn't make a mess like it did in ufs,
* because ext2fs uses a machine-dependent disk
* layout.
*/
if (dp->rec_len <= 0) {
error = EIO;
break;
}
retval = vop_write_dirent(&error, uio, dp->inode,
FTTODT(dp->file_type), dp->name_len, dp->name);
if (retval)
break;
/* advance dp */
dp = (struct ext2_dir_entry_2 *)((char *)dp + dp->rec_len);
if (!error)
ncookies++;
}
/* we need to correct uio_offset */
uio->uio_offset = startoffset + (caddr_t)dp - dirbuf;
if (!error && ap->a_ncookies != NULL) {
off_t *cookiep, *cookies, *ecookies;
off_t off;
if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
panic("ext2fs_readdir: unexpected uio from NFS server");
if (ncookies) {
cookies = kmalloc(ncookies * sizeof(off_t),
M_TEMP, M_WAITOK);
} else {
cookies = kmalloc(sizeof(off_t), M_TEMP,
M_WAITOK);
}
off = startoffset;
for (dp = (struct ext2_dir_entry_2 *)dirbuf,
cookiep = cookies, ecookies = cookies + ncookies;
cookiep < ecookies;
dp = (struct ext2_dir_entry_2 *)((caddr_t) dp + dp->rec_len)) {
off += dp->rec_len;
*cookiep++ = off;
}
*ap->a_ncookies = ncookies;
*ap->a_cookies = cookies;
}
}
kfree(dirbuf, M_TEMP);
if (ap->a_eofflag)
*ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
vn_unlock(ap->a_vp);
return (error);
}
/*
* Go through the disk queues to initiate sandbagged IO;
* go through the inodes to write those that have been modified;
* initiate the writing of the super block if it has been modified.
*
* Note: we are always called with the filesystem marked `MPBUSY'.
*/
int
ext2fs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
{
struct vnode *vp, *mvp;
struct inode *ip;
struct ufsmount *ump = VFSTOUFS(mp);
struct m_ext2fs *fs;
int error, allerror = 0;
fs = ump->um_e2fs;
if (fs->e2fs_fmod != 0 && fs->e2fs_ronly != 0) { /* XXX */
printf("fs = %s\n", fs->e2fs_fsmnt);
panic("update: rofs mod");
}
/* Allocate a marker vnode. */
mvp = vnalloc(mp);
/*
* Write back each (modified) inode.
*/
mutex_enter(&mntvnode_lock);
loop:
/*
* NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
* and vclean() can be called indirectly
*/
for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
vmark(mvp, vp);
if (vp->v_mount != mp || vismarker(vp))
continue;
mutex_enter(vp->v_interlock);
ip = VTOI(vp);
if (ip == NULL || (vp->v_iflag & (VI_XLOCK|VI_CLEAN)) != 0 ||
vp->v_type == VNON ||
((ip->i_flag &
(IN_CHANGE | IN_UPDATE | IN_MODIFIED)) == 0 &&
LIST_EMPTY(&vp->v_dirtyblkhd) &&
UVM_OBJ_IS_CLEAN(&vp->v_uobj)))
{
mutex_exit(vp->v_interlock);
continue;
}
mutex_exit(&mntvnode_lock);
error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT);
if (error) {
mutex_enter(&mntvnode_lock);
if (error == ENOENT) {
mutex_enter(&mntvnode_lock);
(void)vunmark(mvp);
goto loop;
}
continue;
}
if (vp->v_type == VREG && waitfor == MNT_LAZY)
error = ext2fs_update(vp, NULL, NULL, 0);
else
error = VOP_FSYNC(vp, cred,
waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0);
if (error)
allerror = error;
vput(vp);
mutex_enter(&mntvnode_lock);
}
mutex_exit(&mntvnode_lock);
vnfree(mvp);
/*
* Force stale file system control information to be flushed.
*/
if (waitfor != MNT_LAZY) {
vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
if ((error = VOP_FSYNC(ump->um_devvp, cred,
waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0)) != 0)
allerror = error;
VOP_UNLOCK(ump->um_devvp);
}
/*
* Write back modified superblock.
*/
if (fs->e2fs_fmod != 0) {
fs->e2fs_fmod = 0;
fs->e2fs.e2fs_wtime = time_second;
if ((error = ext2fs_cgupdate(ump, waitfor)))
allerror = error;
}
return (allerror);
}
/*
* Write a directory entry after a call to namei, using the parameters
* that it left in the directory inode. 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
* (dp->i_offset, dp->i_count) indicate how the space for the new
* entry is to be obtained.
*/
int
ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
{
struct ext2_dir_entry_2 *ep, *nep;
struct inode *dp;
struct buf *bp;
struct ext2_dir_entry_2 newdir;
struct iovec aiov;
struct uio auio;
u_int dsize;
int error, loc, newentrysize, spacefree;
char *dirbuf;
int DIRBLKSIZ = ip->i_e2fs->s_blocksize;
dp = VTOI(dvp);
newdir.inode = ip->i_number;
newdir.name_len = cnp->cn_namelen;
if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
EXT2_FEATURE_INCOMPAT_FILETYPE))
newdir.file_type = DTTOFT(IFTODT(ip->i_mode));
else
newdir.file_type = EXT2_FT_UNKNOWN;
bcopy(cnp->cn_nameptr, newdir.name, (unsigned)cnp->cn_namelen + 1);
newentrysize = EXT2_DIR_REC_LEN(newdir.name_len);
if (dp->i_count == 0) {
/*
* If dp->i_count is 0, then namei could find no
* space in the directory. Here, dp->i_offset will
* be on a directory block boundary and we will write the
* new entry into a fresh block.
*/
if (dp->i_offset & (DIRBLKSIZ - 1))
panic("ext2_direnter: newblk");
auio.uio_offset = dp->i_offset;
newdir.rec_len = DIRBLKSIZ;
auio.uio_resid = newentrysize;
aiov.iov_len = newentrysize;
aiov.iov_base = (caddr_t)&newdir;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_rw = UIO_WRITE;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_td = NULL;
error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
if (DIRBLKSIZ >
VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
/* XXX should grow with balloc() */
panic("ext2_direnter: frag size");
else if (!error) {
dp->i_size = roundup(dp->i_size, DIRBLKSIZ);
dp->i_flag |= IN_CHANGE;
}
return (error);
}
/*
* If dp->i_count is non-zero, then namei found space
* for the new entry in the range dp->i_offset to
* dp->i_offset + dp->i_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.
*/
/*
* Increase size of directory if entry eats into new space.
* This should never push the size past a new multiple of
* DIRBLKSIZE.
*
* N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
*/
if (dp->i_offset + dp->i_count > dp->i_size)
dp->i_size = dp->i_offset + dp->i_count;
/*
* Get the block containing the space for the new directory entry.
*/
if ((error = EXT2_BLKATOFF(dvp, (off_t)dp->i_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 dp->i_offset to
* dp->i_offset + dp->i_count would yield the
* space.
*/
ep = (struct ext2_dir_entry_2 *)dirbuf;
dsize = EXT2_DIR_REC_LEN(ep->name_len);
spacefree = ep->rec_len - dsize;
for (loc = ep->rec_len; loc < dp->i_count; ) {
nep = (struct ext2_dir_entry_2 *)(dirbuf + loc);
if (ep->inode) {
/* trim the existing slot */
ep->rec_len = dsize;
ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
} else {
/* overwrite; nothing there; header is ours */
spacefree += dsize;
}
dsize = EXT2_DIR_REC_LEN(nep->name_len);
spacefree += nep->rec_len - dsize;
loc += nep->rec_len;
bcopy((caddr_t)nep, (caddr_t)ep, dsize);
}
/*
* Update the pointer fields in the previous entry (if any),
* copy in the new entry, and write out the block.
*/
if (ep->inode == 0) {
if (spacefree + dsize < newentrysize)
panic("ext2_direnter: compact1");
newdir.rec_len = spacefree + dsize;
} else {
if (spacefree < newentrysize)
panic("ext2_direnter: compact2");
newdir.rec_len = spacefree;
ep->rec_len = dsize;
ep = (struct ext2_dir_entry_2 *)((char *)ep + dsize);
}
bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
error = bwrite(bp);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
error = EXT2_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC,
cnp->cn_cred);
return (error);
}
/*
* Set attribute vnode op. called from several syscalls
*/
int
ext2fs_setattr(void *v)
{
struct vop_setattr_args /* {
struct vnode *a_vp;
struct vattr *a_vap;
kauth_cred_t a_cred;
} */ *ap = v;
struct vattr *vap = ap->a_vap;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
kauth_cred_t cred = ap->a_cred;
struct lwp *l = curlwp;
int error;
kauth_action_t action = KAUTH_VNODE_WRITE_FLAGS;
bool changing_sysflags = false;
/*
* Check for unsettable attributes.
*/
if ((vap->va_type != VNON) || (vap->va_nlink != (nlink_t)VNOVAL) ||
(vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
(vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
return (EINVAL);
}
if (vap->va_flags != VNOVAL) {
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
/*
* Check if we're allowed to change the flags.
* If EXT2FS_SYSTEM_FLAGS is set, then the flags are treated
* as system flags, otherwise they're considered to be user
* flags.
*/
#ifdef EXT2FS_SYSTEM_FLAGS
/* Indicate we're changing system flags if we are. */
if ((vap->va_flags & SF_APPEND) ||
(vap->va_flags & SF_IMMUTABLE)) {
action |= KAUTH_VNODE_WRITE_SYSFLAGS;
changing_sysflags = true;
}
/* Indicate the node has system flags if it does. */
if (ip->i_e2fs_flags & (EXT2_APPEND | EXT2_IMMUTABLE)) {
action |= KAUTH_VNODE_HAS_SYSFLAGS;
}
#endif /* EXT2FS_SYSTEM_FLAGS */
error = kauth_authorize_vnode(cred, action, vp, NULL,
genfs_can_chflags(cred, vp->v_type, ip->i_uid,
changing_sysflags));
if (error)
return (error);
#ifdef EXT2FS_SYSTEM_FLAGS
ip->i_e2fs_flags &= ~(EXT2_APPEND | EXT2_IMMUTABLE);
ip->i_e2fs_flags |=
(vap->va_flags & SF_APPEND) ? EXT2_APPEND : 0 |
(vap->va_flags & SF_IMMUTABLE) ? EXT2_IMMUTABLE : 0;
#else
ip->i_e2fs_flags &= ~(EXT2_APPEND | EXT2_IMMUTABLE);
ip->i_e2fs_flags |=
(vap->va_flags & UF_APPEND) ? EXT2_APPEND : 0 |
(vap->va_flags & UF_IMMUTABLE) ? EXT2_IMMUTABLE : 0;
#endif
ip->i_flag |= IN_CHANGE;
if (vap->va_flags & (IMMUTABLE | APPEND))
return (0);
}
if (ip->i_e2fs_flags & (EXT2_APPEND | EXT2_IMMUTABLE))
return (EPERM);
/*
* Go through the fields and update iff not VNOVAL.
*/
if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
error = ext2fs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
if (error)
return (error);
}
if (vap->va_size != VNOVAL) {
/*
* Disallow write attempts on read-only file systems;
* unless the file is a socket, fifo, or a block or
* character device resident on the file system.
*/
switch (vp->v_type) {
case VDIR:
return (EISDIR);
case VLNK:
case VREG:
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
default:
break;
}
error = ext2fs_truncate(vp, vap->va_size, 0, cred);
if (error)
return (error);
}
ip = VTOI(vp);
if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_TIMES, vp,
NULL, genfs_can_chtimes(vp, vap->va_vaflags, ip->i_uid,
cred));
if (error)
return (error);
if (vap->va_atime.tv_sec != VNOVAL)
if (!(vp->v_mount->mnt_flag & MNT_NOATIME))
ip->i_flag |= IN_ACCESS;
if (vap->va_mtime.tv_sec != VNOVAL) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (vp->v_mount->mnt_flag & MNT_RELATIME)
ip->i_flag |= IN_ACCESS;
}
error = ext2fs_update(vp, &vap->va_atime, &vap->va_mtime,
UPDATE_WAIT);
if (error)
return (error);
}
error = 0;
if (vap->va_mode != (mode_t)VNOVAL) {
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
error = ext2fs_chmod(vp, (int)vap->va_mode, cred, l);
}
VN_KNOTE(vp, NOTE_ATTRIB);
return (error);
}
/* ARGSUSED */
int
ffs_full_fsync(struct vnode *vp, int flags)
{
int error, i, uflags;
struct mount *mp;
KASSERT(vp->v_tag == VT_UFS);
KASSERT(VTOI(vp) != NULL);
KASSERT(vp->v_type != VCHR && vp->v_type != VBLK);
error = 0;
uflags = UPDATE_CLOSE | ((flags & FSYNC_WAIT) ? UPDATE_WAIT : 0);
mp = vp->v_mount;
/*
* Flush all dirty data associated with the vnode.
*/
if (vp->v_type == VREG) {
int pflags = PGO_ALLPAGES | PGO_CLEANIT;
if ((flags & FSYNC_WAIT))
pflags |= PGO_SYNCIO;
if (fstrans_getstate(mp) == FSTRANS_SUSPENDING)
pflags |= PGO_FREE;
mutex_enter(vp->v_interlock);
error = VOP_PUTPAGES(vp, 0, 0, pflags);
if (error)
return error;
}
#ifdef WAPBL
if (mp && mp->mnt_wapbl) {
/*
* Don't bother writing out metadata if the syncer is
* making the request. We will let the sync vnode
* write it out in a single burst through a call to
* VFS_SYNC().
*/
if ((flags & (FSYNC_DATAONLY | FSYNC_LAZY)) != 0)
return 0;
if ((VTOI(vp)->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE
| IN_MODIFY | IN_MODIFIED | IN_ACCESSED)) != 0) {
error = UFS_WAPBL_BEGIN(mp);
if (error)
return error;
error = ffs_update(vp, NULL, NULL, uflags);
UFS_WAPBL_END(mp);
}
if (error || (flags & FSYNC_NOLOG) != 0)
return error;
/*
* Don't flush the log if the vnode being flushed
* contains no dirty buffers that could be in the log.
*/
if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
error = wapbl_flush(mp->mnt_wapbl, 0);
if (error)
return error;
}
if ((flags & FSYNC_WAIT) != 0) {
mutex_enter(vp->v_interlock);
while (vp->v_numoutput != 0)
cv_wait(&vp->v_cv, vp->v_interlock);
mutex_exit(vp->v_interlock);
}
return error;
}
#endif /* WAPBL */
error = vflushbuf(vp, (flags & FSYNC_WAIT) != 0);
if (error == 0)
error = ffs_update(vp, NULL, NULL, uflags);
if (error == 0 && (flags & FSYNC_CACHE) != 0) {
i = 1;
(void)VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &i, FWRITE,
kauth_cred_get());
}
return error;
}
static int
fuse_vnode_cmp(struct vnode *vp, void *nidp)
{
return (VTOI(vp) != *((uint64_t *)nidp));
}
/*
* this function handles ext4 extents block mapping
*/
static int
ext4_ext_read(struct vop_read_args *ap)
{
struct vnode *vp;
struct inode *ip;
struct uio *uio;
struct m_ext2fs *fs;
struct buf *bp;
struct ext4_extent nex, *ep;
struct ext4_extent_header *ehp;
struct ext4_extent_path path;
daddr_t lbn, nextlbn, newblk = 0;
off_t bytesinfile;
u_short mode;
int cache_type;
int orig_resid;
int error = 0;
int depth = 0;
long size, xfersize, blkoffset;
vp = ap->a_vp;
ip = VTOI(vp);
mode = ip->i_mode;
uio = ap->a_uio;
memset(&path, 0, sizeof(path));
orig_resid = uio->uio_resid;
KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
if (orig_resid == 0)
return (0);
KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
fs = ip->I_FS;
if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
return (EOVERFLOW);
for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
break;
lbn = lblkno(fs, uio->uio_offset);
nextlbn = lbn + 1;
size = BLKSIZE(fs, ip, lbn);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->e2fs_fsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (bytesinfile < xfersize)
xfersize = bytesinfile;
/* get block from ext4 extent cache */
cache_type = ext4_ext_in_cache(ip, lbn, &nex);
if (cache_type != 0) {
/* block does not be allocated yet */
if (cache_type == EXT4_EXT_CACHE_GAP)
return (error);
else if (cache_type == EXT4_EXT_CACHE_IN)
newblk = lbn - nex.e_blk +
(nex.e_start_lo | ((daddr_t)(nex.e_start_hi) << 31) << 1);
} else {
ext4_ext_find_extent(fs, ip, lbn, &path);
depth = ((struct ext4_extent_header *)(ip->i_db))->eh_depth;
if (path.ep_ext == NULL && depth != 0)
return (EIO);
ehp = path.ep_header;
ep = path.ep_ext;
if (ep == NULL)
return (EIO);
ext4_ext_put_cache(ip, ep, EXT4_EXT_CACHE_IN);
newblk = lbn - ep->e_blk +
(ep->e_start_lo | ((daddr_t)(ep->e_start_hi) << 31) << 1);
if (path.ep_bp != NULL) {
brelse(path.ep_bp);
path.ep_bp = NULL;
}
}
error = bread(ip->i_devvp, fsbtodb(fs, newblk), size, NOCRED, &bp);
if (error) {
brelse(bp);
bp = NULL;
break;
}
size -= bp->b_resid;
if (size < xfersize) {
if (size == 0)
break;
xfersize = size;
}
error = uiomove((char *)bp->b_data + blkoffset,
(int)xfersize, uio);
if (error)
break;
bqrelse(bp);
}
if (bp != NULL)
bqrelse(bp);
return (error);
}
int
linkup(ino_t orphan, ino_t parentdir)
{
union lfs_dinode *dp;
int lostdir;
ino_t oldlfdir;
struct inodesc idesc;
char tempname[BUFSIZ];
struct uvnode *vp;
memset(&idesc, 0, sizeof(struct inodesc));
vp = vget(fs, orphan);
dp = VTOD(vp);
lostdir = (lfs_dino_getmode(fs, dp) & LFS_IFMT) == LFS_IFDIR;
pwarn("UNREF %s ", lostdir ? "DIR" : "FILE");
pinode(orphan);
if (preen && lfs_dino_getsize(fs, dp) == 0)
return (0);
if (preen)
printf(" (RECONNECTED)\n");
else if (reply("RECONNECT") == 0)
return (0);
if (lfdir == 0) {
dp = ginode(ULFS_ROOTINO);
idesc.id_name = lfname;
idesc.id_type = DATA;
idesc.id_func = findino;
idesc.id_number = ULFS_ROOTINO;
if ((ckinode(dp, &idesc) & FOUND) != 0) {
lfdir = idesc.id_parent;
} else {
pwarn("NO lost+found DIRECTORY");
if (preen || reply("CREATE")) {
lfdir = allocdir(ULFS_ROOTINO, (ino_t) 0, lfmode);
if (lfdir != 0) {
if (makeentry(ULFS_ROOTINO, lfdir, lfname) != 0) {
if (preen)
printf(" (CREATED)\n");
} else {
freedir(lfdir, ULFS_ROOTINO);
lfdir = 0;
if (preen)
printf("\n");
}
}
}
}
if (lfdir == 0) {
pfatal("SORRY. CANNOT CREATE lost+found DIRECTORY");
printf("\n\n");
return (0);
}
}
vp = vget(fs, lfdir);
dp = VTOD(vp);
if ((lfs_dino_getmode(fs, dp) & LFS_IFMT) != LFS_IFDIR) {
pfatal("lost+found IS NOT A DIRECTORY");
if (reply("REALLOCATE") == 0)
return (0);
oldlfdir = lfdir;
if ((lfdir = allocdir(ULFS_ROOTINO, (ino_t) 0, lfmode)) == 0) {
pfatal("SORRY. CANNOT CREATE lost+found DIRECTORY\n\n");
return (0);
}
if ((changeino(ULFS_ROOTINO, lfname, lfdir) & ALTERED) == 0) {
pfatal("SORRY. CANNOT CREATE lost+found DIRECTORY\n\n");
return (0);
}
inodirty(VTOI(vp));
idesc.id_type = ADDR;
idesc.id_func = pass4check;
idesc.id_number = oldlfdir;
adjust(&idesc, lncntp[oldlfdir] + 1);
lncntp[oldlfdir] = 0;
vp = vget(fs, lfdir);
dp = VTOD(vp);
}
if (statemap[lfdir] != DFOUND) {
pfatal("SORRY. NO lost+found DIRECTORY\n\n");
return (0);
}
(void) lftempname(tempname, orphan);
if (makeentry(lfdir, orphan, tempname) == 0) {
pfatal("SORRY. NO SPACE IN lost+found DIRECTORY");
printf("\n\n");
return (0);
}
lncntp[orphan]--;
if (lostdir) {
if ((changeino(orphan, "..", lfdir) & ALTERED) == 0 &&
parentdir != (ino_t) - 1)
(void) makeentry(orphan, lfdir, "..");
vp = vget(fs, lfdir);
lfs_dino_setnlink(fs, VTOI(vp)->i_din,
lfs_dino_getnlink(fs, VTOI(vp)->i_din) + 1);
inodirty(VTOI(vp));
lncntp[lfdir]++;
pwarn("DIR I=%llu CONNECTED. ", (unsigned long long)orphan);
if (parentdir != (ino_t) - 1)
printf("PARENT WAS I=%llu\n",
(unsigned long long)parentdir);
if (preen == 0)
printf("\n");
}
return (1);
}
/*
* Q_QUOTAON - set up a quota file for a particular filesystem.
*/
int
quotaon(struct thread *td, struct mount *mp, int type, void *fname)
{
struct ufsmount *ump;
struct vnode *vp, **vpp;
struct vnode *mvp;
struct dquot *dq;
int error, flags;
struct nameidata nd;
error = priv_check(td, PRIV_UFS_QUOTAON);
if (error != 0) {
vfs_unbusy(mp);
return (error);
}
if ((mp->mnt_flag & MNT_RDONLY) != 0) {
vfs_unbusy(mp);
return (EROFS);
}
ump = VFSTOUFS(mp);
dq = NODQUOT;
NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fname, td);
flags = FREAD | FWRITE;
vfs_ref(mp);
vfs_unbusy(mp);
error = vn_open(&nd, &flags, 0, NULL);
if (error != 0) {
vfs_rel(mp);
return (error);
}
NDFREE(&nd, NDF_ONLY_PNBUF);
vp = nd.ni_vp;
error = vfs_busy(mp, MBF_NOWAIT);
vfs_rel(mp);
if (error == 0) {
if (vp->v_type != VREG) {
error = EACCES;
vfs_unbusy(mp);
}
}
if (error != 0) {
VOP_UNLOCK(vp, 0);
(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
return (error);
}
UFS_LOCK(ump);
if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
UFS_UNLOCK(ump);
VOP_UNLOCK(vp, 0);
(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
vfs_unbusy(mp);
return (EALREADY);
}
ump->um_qflags[type] |= QTF_OPENING|QTF_CLOSING;
UFS_UNLOCK(ump);
if ((error = dqopen(vp, ump, type)) != 0) {
VOP_UNLOCK(vp, 0);
UFS_LOCK(ump);
ump->um_qflags[type] &= ~(QTF_OPENING|QTF_CLOSING);
UFS_UNLOCK(ump);
(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
vfs_unbusy(mp);
return (error);
}
VOP_UNLOCK(vp, 0);
MNT_ILOCK(mp);
mp->mnt_flag |= MNT_QUOTA;
MNT_IUNLOCK(mp);
vpp = &ump->um_quotas[type];
if (*vpp != vp)
quotaoff1(td, mp, type);
/*
* When the directory vnode containing the quota file is
* inactivated, due to the shared lookup of the quota file
* vput()ing the dvp, the qsyncvp() call for the containing
* directory would try to acquire the quota lock exclusive.
* At the same time, lookup already locked the quota vnode
* shared. Mark the quota vnode lock as allowing recursion
* and automatically converting shared locks to exclusive.
*
* Also mark quota vnode as system.
*/
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vp->v_vflag |= VV_SYSTEM;
VN_LOCK_AREC(vp);
VN_LOCK_DSHARE(vp);
VOP_UNLOCK(vp, 0);
*vpp = vp;
/*
* Save the credential of the process that turned on quotas.
* Set up the time limits for this quota.
*/
ump->um_cred[type] = crhold(td->td_ucred);
ump->um_btime[type] = MAX_DQ_TIME;
ump->um_itime[type] = MAX_IQ_TIME;
if (dqget(NULLVP, 0, ump, type, &dq) == 0) {
if (dq->dq_btime > 0)
ump->um_btime[type] = dq->dq_btime;
if (dq->dq_itime > 0)
ump->um_itime[type] = dq->dq_itime;
dqrele(NULLVP, dq);
}
/*
* Allow the getdq from getinoquota below to read the quota
* from file.
*/
UFS_LOCK(ump);
ump->um_qflags[type] &= ~QTF_CLOSING;
UFS_UNLOCK(ump);
/*
* Search vnodes associated with this mount point,
* adding references to quota file being opened.
* NB: only need to add dquot's for inodes being modified.
*/
again:
MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
goto again;
}
if (vp->v_type == VNON || vp->v_writecount == 0) {
VOP_UNLOCK(vp, 0);
vrele(vp);
continue;
}
error = getinoquota(VTOI(vp));
VOP_UNLOCK(vp, 0);
vrele(vp);
if (error) {
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
break;
}
}
if (error)
quotaoff_inchange(td, mp, type);
UFS_LOCK(ump);
ump->um_qflags[type] &= ~QTF_OPENING;
KASSERT((ump->um_qflags[type] & QTF_CLOSING) == 0,
("quotaon: leaking flags"));
UFS_UNLOCK(ump);
vfs_unbusy(mp);
return (error);
}
/*
* 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);
}
/*
* Reload all incore data for a filesystem (used after running fsck on
* the root filesystem and finding things to fix). The filesystem must
* be mounted read-only.
*
* Things to do to update the mount:
* 1) invalidate all cached meta-data.
* 2) re-read superblock from disk.
* 3) invalidate all cluster summary information.
* 4) invalidate all inactive vnodes.
* 5) invalidate all cached file data.
* 6) re-read inode data for all active vnodes.
* XXX we are missing some steps, in particular # 3, this has to be reviewed.
*/
static int
ext2_reload(struct mount *mp, struct thread *td)
{
struct vnode *vp, *mvp, *devvp;
struct inode *ip;
struct buf *bp;
struct ext2fs *es;
struct m_ext2fs *fs;
struct csum *sump;
int error, i;
int32_t *lp;
if ((mp->mnt_flag & MNT_RDONLY) == 0)
return (EINVAL);
/*
* Step 1: invalidate all cached meta-data.
*/
devvp = VFSTOEXT2(mp)->um_devvp;
vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
if (vinvalbuf(devvp, 0, 0, 0) != 0)
panic("ext2_reload: dirty1");
VOP_UNLOCK(devvp, 0);
/*
* Step 2: re-read superblock from disk.
* constants have been adjusted for ext2
*/
if ((error = bread(devvp, SBLOCK, SBSIZE, NOCRED, &bp)) != 0)
return (error);
es = (struct ext2fs *)bp->b_data;
if (ext2_check_sb_compat(es, devvp->v_rdev, 0) != 0) {
brelse(bp);
return (EIO); /* XXX needs translation */
}
fs = VFSTOEXT2(mp)->um_e2fs;
bcopy(bp->b_data, fs->e2fs, sizeof(struct ext2fs));
if((error = compute_sb_data(devvp, es, fs)) != 0) {
brelse(bp);
return (error);
}
#ifdef UNKLAR
if (fs->fs_sbsize < SBSIZE)
bp->b_flags |= B_INVAL;
#endif
brelse(bp);
/*
* Step 3: invalidate all cluster summary information.
*/
if (fs->e2fs_contigsumsize > 0) {
lp = fs->e2fs_maxcluster;
sump = fs->e2fs_clustersum;
for (i = 0; i < fs->e2fs_gcount; i++, sump++) {
*lp++ = fs->e2fs_contigsumsize;
sump->cs_init = 0;
bzero(sump->cs_sum, fs->e2fs_contigsumsize + 1);
}
}
loop:
MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
/*
* Step 4: invalidate all cached file data.
*/
if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
goto loop;
}
if (vinvalbuf(vp, 0, 0, 0))
panic("ext2_reload: dirty2");
/*
* Step 5: re-read inode data for all active vnodes.
*/
ip = VTOI(vp);
error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
(int)fs->e2fs_bsize, NOCRED, &bp);
if (error) {
VOP_UNLOCK(vp, 0);
vrele(vp);
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
return (error);
}
ext2_ei2i((struct ext2fs_dinode *) ((char *)bp->b_data +
EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)), ip);
brelse(bp);
VOP_UNLOCK(vp, 0);
vrele(vp);
}
return (0);
}
void
osi_DisableAtimes(struct vnode *avp)
{
struct inode *ip = VTOI(avp);
ip->i_flag &= ~IN_ACCESS;
}
/*
* Retrieve the ACL on a file.
*
* As part of the ACL is stored in the inode, and the rest in an EA,
* assemble both into a final ACL product. Right now this is not done
* very efficiently.
*/
static int
ufs_getacl_posix1e(struct vop_getacl_args *ap)
{
struct inode *ip = VTOI(ap->a_vp);
int error;
struct oldacl *old;
/*
* XXX: If ufs_getacl() should work on file systems not supporting
* ACLs, remove this check.
*/
if ((ap->a_vp->v_mount->mnt_flag & MNT_ACLS) == 0)
return (EINVAL);
old = malloc(sizeof(*old), M_ACL, M_WAITOK | M_ZERO);
/*
* Attempt to retrieve the ACL from the extended attributes.
*/
error = ufs_get_oldacl(ap->a_type, old, ap->a_vp, ap->a_td);
switch (error) {
/*
* XXX: If ufs_getacl() should work on filesystems
* without the EA configured, add case EOPNOTSUPP here.
*/
case ENOATTR:
switch (ap->a_type) {
case ACL_TYPE_ACCESS:
/*
* Legitimately no ACL set on object, purely
* emulate it through the inode. These fields will
* be updated when the ACL is synchronized with
* the inode later.
*/
old->acl_cnt = 3;
old->acl_entry[0].ae_tag = ACL_USER_OBJ;
old->acl_entry[0].ae_id = ACL_UNDEFINED_ID;
old->acl_entry[0].ae_perm = ACL_PERM_NONE;
old->acl_entry[1].ae_tag = ACL_GROUP_OBJ;
old->acl_entry[1].ae_id = ACL_UNDEFINED_ID;
old->acl_entry[1].ae_perm = ACL_PERM_NONE;
old->acl_entry[2].ae_tag = ACL_OTHER;
old->acl_entry[2].ae_id = ACL_UNDEFINED_ID;
old->acl_entry[2].ae_perm = ACL_PERM_NONE;
break;
case ACL_TYPE_DEFAULT:
/*
* Unlike ACL_TYPE_ACCESS, there is no relationship
* between the inode contents and the ACL, and it is
* therefore possible for the request for the ACL
* to fail since the ACL is undefined. In this
* situation, return success and an empty ACL,
* as required by POSIX.1e.
*/
old->acl_cnt = 0;
break;
}
/* FALLTHROUGH */
case 0:
error = acl_copy_oldacl_into_acl(old, ap->a_aclp);
if (error != 0)
break;
if (ap->a_type == ACL_TYPE_ACCESS)
ufs_sync_acl_from_inode(ip, ap->a_aclp);
default:
break;
}
free(old, M_ACL);
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 ufs_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
ufs_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 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;
int nameiop = cnp->cn_nameiop;
struct proc *p = cnp->cn_proc;
cnp->cn_flags &= ~PDIRUNLOCK;
flags = cnp->cn_flags;
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 ((DIP(dp, mode) & IFMT) != IFDIR)
return (ENOTDIR);
if ((error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc)) != 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 = (sizeof(struct direct) - MAXNAMLEN +
cnp->cn_namelen + 3) &~ 3;
}
/*
* 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;
#ifdef UFS_DIRHASH
/*
* Use dirhash for fast operations on large directories. The logic
* to determine whether to hash the directory is contained within
* ufsdirhash_build(); a zero return means that it decided to hash
* this directory and it successfully built up the hash table.
*/
if (ufsdirhash_build(dp) == 0) {
/* Look for a free slot if needed. */
enduseful = DIP(dp, size);
if (slotstatus != FOUND) {
slotoffset = ufsdirhash_findfree(dp, slotneeded,
&slotsize);
if (slotoffset >= 0) {
slotstatus = COMPACT;
enduseful = ufsdirhash_enduseful(dp);
if (enduseful < 0)
enduseful = DIP(dp, size);
}
}
/* Look up the component. */
numdirpasses = 1;
entryoffsetinblock = 0; /* silence compiler warning */
switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
&dp->i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
case 0:
ep = (struct direct *)((char *)bp->b_data +
(dp->i_offset & bmask));
goto foundentry;
case ENOENT:
#define roundup2(x, y) (((x)+((y)-1))&(~((y)-1))) /* if y is powers of two */
dp->i_offset = roundup2(DIP(dp, size), DIRBLKSIZ);
goto notfound;
default:
/* Something failed; just do a linear search. */
break;
}
}
#endif /* UFS_DIRHASH */
if (nameiop != LOOKUP || dp->i_diroff == 0 ||
dp->i_diroff >= DIP(dp, size)) {
entryoffsetinblock = 0;
dp->i_offset = 0;
numdirpasses = 1;
} else {
dp->i_offset = dp->i_diroff;
if ((entryoffsetinblock = dp->i_offset & bmask) &&
(error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, NULL, &bp)))
return (error);
numdirpasses = 2;
nchstats.ncs_2passes++;
}
prevoff = dp->i_offset;
endsearch = roundup(DIP(dp, size), DIRBLKSIZ);
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 = UFS_BUFATOFF(dp, (off_t)dp->i_offset, NULL,
&bp);
if (error)
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.
*/
ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
if (ep->d_reclen == 0 ||
(dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
int i;
ufs_dirbad(dp, dp->i_offset, "mangled entry");
i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 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 = ep->d_reclen;
if (ep->d_ino != 0)
size -= DIRSIZ(FSFMT(vdp), ep);
if (size > 0) {
if (size >= slotneeded) {
slotstatus = FOUND;
slotoffset = dp->i_offset;
slotsize = ep->d_reclen;
} else if (slotstatus == NONE) {
slotfreespace += size;
if (slotoffset == -1)
slotoffset = dp->i_offset;
if (slotfreespace >= slotneeded) {
slotstatus = COMPACT;
slotsize = dp->i_offset +
ep->d_reclen - slotoffset;
}
}
}
}
/*
* Check for a name match.
*/
if (ep->d_ino) {
# if (BYTE_ORDER == LITTLE_ENDIAN)
if (vdp->v_mount->mnt_maxsymlinklen > 0)
namlen = ep->d_namlen;
else
namlen = ep->d_type;
# else
namlen = ep->d_namlen;
# endif
if (namlen == cnp->cn_namelen &&
!bcmp(cnp->cn_nameptr, ep->d_name,
(unsigned)namlen)) {
#ifdef UFS_DIRHASH
foundentry:
#endif
/*
* Save directory entry's inode number and
* reclen in ndp->ni_ufs area, and release
* directory buffer.
*/
dp->i_ino = ep->d_ino;
dp->i_reclen = ep->d_reclen;
goto found;
}
}
prevoff = dp->i_offset;
dp->i_offset += ep->d_reclen;
entryoffsetinblock += ep->d_reclen;
if (ep->d_ino)
enduseful = dp->i_offset;
}
#ifdef UFS_DIRHASH
notfound:
#endif
/*
* 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_effnlink != 0) {
/*
* Access for write is interpreted as allowing
* creation of files in the directory.
*/
error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
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 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(DIP(dp, size), DIRBLKSIZ);
dp->i_count = 0;
enduseful = dp->i_offset;
} else if (nameiop == DELETE) {
dp->i_offset = slotoffset;
if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
dp->i_count = 0;
else
dp->i_count = dp->i_offset - prevoff;
} else {
dp->i_offset = slotoffset;
dp->i_count = slotsize;
if (enduseful < slotoffset + slotsize)
enduseful = slotoffset + slotsize;
}
dp->i_endoff = roundup(enduseful, DIRBLKSIZ);
/*
* 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, p);
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:
if (numdirpasses == 2)
nchstats.ncs_pass2++;
/*
* Check that directory length properly reflects presence
* of this entry.
*/
if (dp->i_offset + DIRSIZ(FSFMT(vdp), ep) > DIP(dp, size)) {
ufs_dirbad(dp, dp->i_offset, "i_ffs_size too small");
DIP_ASSIGN(dp, size, dp->i_offset + DIRSIZ(FSFMT(vdp), ep));
dp->i_flag |= IN_CHANGE | IN_UPDATE;
}
brelse(bp);
/*
* 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 &~ (DIRBLKSIZ - 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.
*/
error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
if (error)
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 & (DIRBLKSIZ - 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);
}
error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
if (error)
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 ((DIP(dp, mode) & ISVTX) &&
VOP_ACCESS(vdp, VADMIN, cred, cnp->cn_proc) &&
VOP_ACCESS(tdp, VADMIN, cred, cnp->cn_proc)) {
vput(tdp);
return (EPERM);
}
*vpp = tdp;
if (!lockparent) {
VOP_UNLOCK(vdp, 0, p);
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)) {
error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_proc);
if (error)
return (error);
/*
* Careful about locking second inode.
* This can only occur if the target is ".".
*/
if (dp->i_number == dp->i_ino)
return (EISDIR);
error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
if (error)
return (error);
*vpp = tdp;
cnp->cn_flags |= SAVENAME;
if (!lockparent) {
VOP_UNLOCK(vdp, 0, p);
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, p); /* race to get the inode */
cnp->cn_flags |= PDIRUNLOCK;
error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
if (error) {
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))) {
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 {
error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
if (error)
return (error);
if (!lockparent || !(flags & ISLASTCN)) {
VOP_UNLOCK(pdp, 0, p);
cnp->cn_flags |= PDIRUNLOCK;
}
*vpp = tdp;
}
/*
* Insert name into cache if appropriate.
*/
if (cnp->cn_flags & MAKEENTRY)
cache_enter(vdp, *vpp, cnp);
return (0);
}
