/*
* The following routine will acquire the locks required for a rename
* operation. The code understands the semantics of renames and will
* validate that name1 exists under dp1 & that name2 may or may not
* exist under dp2.
*
* We are renaming dp1/name1 to dp2/name2.
*
* Return ENOENT if dp1 does not exist, other lookup errors, or 0 for success.
*/
STATIC int
xfs_lock_for_rename(
xfs_inode_t *dp1, /* old (source) directory inode */
xfs_inode_t *dp2, /* new (target) directory inode */
bhv_vname_t *vname1,/* old entry name */
bhv_vname_t *vname2,/* new entry name */
xfs_inode_t **ipp1, /* inode of old entry */
xfs_inode_t **ipp2, /* inode of new entry, if it
already exists, NULL otherwise. */
xfs_inode_t **i_tab,/* array of inode returned, sorted */
int *num_inodes) /* number of inodes in array */
{
xfs_inode_t *ip1, *ip2, *temp;
xfs_ino_t inum1, inum2;
int error;
int i, j;
uint lock_mode;
int diff_dirs = (dp1 != dp2);
ip2 = NULL;
/*
* First, find out the current inums of the entries so that we
* can determine the initial locking order. We'll have to
* sanity check stuff after all the locks have been acquired
* to see if we still have the right inodes, directories, etc.
*/
lock_mode = xfs_ilock_map_shared(dp1);
error = xfs_get_dir_entry(vname1, &ip1);
if (error) {
xfs_iunlock_map_shared(dp1, lock_mode);
return error;
}
inum1 = ip1->i_ino;
ASSERT(ip1);
ITRACE(ip1);
/*
* Unlock dp1 and lock dp2 if they are different.
*/
if (diff_dirs) {
xfs_iunlock_map_shared(dp1, lock_mode);
lock_mode = xfs_ilock_map_shared(dp2);
}
error = xfs_dir_lookup_int(XFS_ITOBHV(dp2), lock_mode,
vname2, &inum2, &ip2);
if (error == ENOENT) { /* target does not need to exist. */
inum2 = 0;
} else if (error) {
/*
* If dp2 and dp1 are the same, the next line unlocks dp1.
* Got it?
*/
xfs_iunlock_map_shared(dp2, lock_mode);
IRELE (ip1);
return error;
} else {
ITRACE(ip2);
}
/*
* i_tab contains a list of pointers to inodes. We initialize
* the table here & we'll sort it. We will then use it to
* order the acquisition of the inode locks.
*
* Note that the table may contain duplicates. e.g., dp1 == dp2.
*/
i_tab[0] = dp1;
i_tab[1] = dp2;
i_tab[2] = ip1;
if (inum2 == 0) {
*num_inodes = 3;
i_tab[3] = NULL;
} else {
*num_inodes = 4;
i_tab[3] = ip2;
}
/*
* Sort the elements via bubble sort. (Remember, there are at
* most 4 elements to sort, so this is adequate.)
*/
for (i=0; i < *num_inodes; i++) {
for (j=1; j < *num_inodes; j++) {
if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
temp = i_tab[j];
i_tab[j] = i_tab[j-1];
i_tab[j-1] = temp;
}
}
}
/*
* We have dp2 locked. If it isn't first, unlock it.
* If it is first, tell xfs_lock_inodes so it can skip it
* when locking. if dp1 == dp2, xfs_lock_inodes will skip both
* since they are equal. xfs_lock_inodes needs all these inodes
* so that it can unlock and retry if there might be a dead-lock
* potential with the log.
*/
if (i_tab[0] == dp2 && lock_mode == XFS_ILOCK_SHARED) {
#ifdef DEBUG
xfs_rename_skip++;
#endif
xfs_lock_inodes(i_tab, *num_inodes, 1, XFS_ILOCK_SHARED);
} else {
#ifdef DEBUG
xfs_rename_nskip++;
#endif
xfs_iunlock_map_shared(dp2, lock_mode);
xfs_lock_inodes(i_tab, *num_inodes, 0, XFS_ILOCK_SHARED);
}
/*
* Set the return value. Null out any unused entries in i_tab.
*/
*ipp1 = *ipp2 = NULL;
for (i=0; i < *num_inodes; i++) {
if (i_tab[i]->i_ino == inum1) {
*ipp1 = i_tab[i];
}
if (i_tab[i]->i_ino == inum2) {
*ipp2 = i_tab[i];
}
}
for (;i < 4; i++) {
i_tab[i] = NULL;
}
return 0;
}
bhv_desc_t *
xfs_itobhv(xfs_inode_t *ip)
{
return XFS_ITOBHV(ip);
}
/*
* Look up an inode by number in the given file system.
* The inode is looked up in the hash table for the file system
* represented by the mount point parameter mp. Each bucket of
* the hash table is guarded by an individual semaphore.
*
* If the inode is found in the hash table, its corresponding vnode
* is obtained with a call to vn_get(). This call takes care of
* coordination with the reclamation of the inode and vnode. Note
* that the vmap structure is filled in while holding the hash lock.
* This gives us the state of the inode/vnode when we found it and
* is used for coordination in vn_get().
*
* If it is not in core, read it in from the file system's device and
* add the inode into the hash table.
*
* The inode is locked according to the value of the lock_flags parameter.
* This flag parameter indicates how and if the inode's IO lock and inode lock
* should be taken.
*
* mp -- the mount point structure for the current file system. It points
* to the inode hash table.
* tp -- a pointer to the current transaction if there is one. This is
* simply passed through to the xfs_iread() call.
* ino -- the number of the inode desired. This is the unique identifier
* within the file system for the inode being requested.
* lock_flags -- flags indicating how to lock the inode. See the comment
* for xfs_ilock() for a list of valid values.
* bno -- the block number starting the buffer containing the inode,
* if known (as by bulkstat), else 0.
*/
int
xfs_iget(
xfs_mount_t *mp,
xfs_trans_t *tp,
xfs_ino_t ino,
uint flags,
uint lock_flags,
xfs_inode_t **ipp,
xfs_daddr_t bno)
{
xfs_ihash_t *ih;
xfs_inode_t *ip;
xfs_inode_t *iq;
xfs_vnode_t *vp;
ulong version;
int error;
/* REFERENCED */
int newnode;
xfs_chash_t *ch;
xfs_chashlist_t *chl, *chlnew;
vmap_t vmap;
SPLDECL(s);
XFS_STATS_INC(xs_ig_attempts);
ih = XFS_IHASH(mp, ino);
again:
read_lock(&ih->ih_lock);
for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
if (ip->i_ino == ino) {
vp = XFS_ITOV(ip);
VMAP(vp, vmap);
/*
* Inode cache hit: if ip is not at the front of
* its hash chain, move it there now.
* Do this with the lock held for update, but
* do statistics after releasing the lock.
*/
if (ip->i_prevp != &ih->ih_next
&& rwlock_trypromote(&ih->ih_lock)) {
if ((iq = ip->i_next)) {
iq->i_prevp = ip->i_prevp;
}
*ip->i_prevp = iq;
iq = ih->ih_next;
iq->i_prevp = &ip->i_next;
ip->i_next = iq;
ip->i_prevp = &ih->ih_next;
ih->ih_next = ip;
write_unlock(&ih->ih_lock);
} else {
read_unlock(&ih->ih_lock);
}
XFS_STATS_INC(xs_ig_found);
/*
* Get a reference to the vnode/inode.
* vn_get() takes care of coordination with
* the file system inode release and reclaim
* functions. If it returns NULL, the inode
* has been reclaimed so just start the search
* over again. We probably won't find it,
* but we could be racing with another cpu
* looking for the same inode so we have to at
* least look.
*/
if (!(vp = vn_get(vp, &vmap))) {
XFS_STATS_INC(xs_ig_frecycle);
goto again;
}
if (lock_flags != 0) {
ip->i_flags &= ~XFS_IRECLAIM;
xfs_ilock(ip, lock_flags);
}
newnode = (ip->i_d.di_mode == 0);
if (newnode) {
xfs_iocore_inode_reinit(ip);
}
ip->i_flags &= ~XFS_ISTALE;
vn_trace_exit(vp, "xfs_iget.found",
(inst_t *)__return_address);
goto return_ip;
}
}
/*
* Inode cache miss: save the hash chain version stamp and unlock
* the chain, so we don't deadlock in vn_alloc.
*/
XFS_STATS_INC(xs_ig_missed);
version = ih->ih_version;
read_unlock(&ih->ih_lock);
/*
* Read the disk inode attributes into a new inode structure and get
* a new vnode for it. This should also initialize i_ino and i_mount.
*/
error = xfs_iread(mp, tp, ino, &ip, bno);
if (error) {
return error;
}
error = xfs_vn_allocate(mp, ip, &vp);
if (error) {
return error;
}
vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address);
xfs_inode_lock_init(ip, vp);
xfs_iocore_inode_init(ip);
if (lock_flags != 0) {
xfs_ilock(ip, lock_flags);
}
/*
* Put ip on its hash chain, unless someone else hashed a duplicate
* after we released the hash lock.
*/
write_lock(&ih->ih_lock);
if (ih->ih_version != version) {
for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) {
if (iq->i_ino == ino) {
write_unlock(&ih->ih_lock);
xfs_idestroy(ip);
XFS_STATS_INC(xs_ig_dup);
goto again;
}
}
}
/*
* These values _must_ be set before releasing ihlock!
*/
ip->i_hash = ih;
if ((iq = ih->ih_next)) {
iq->i_prevp = &ip->i_next;
}
ip->i_next = iq;
ip->i_prevp = &ih->ih_next;
ih->ih_next = ip;
ip->i_udquot = ip->i_gdquot = NULL;
ih->ih_version++;
write_unlock(&ih->ih_lock);
/*
* put ip on its cluster's hash chain
*/
ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL &&
ip->i_cnext == NULL);
chlnew = NULL;
ch = XFS_CHASH(mp, ip->i_blkno);
chlredo:
s = mutex_spinlock(&ch->ch_lock);
for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
if (chl->chl_blkno == ip->i_blkno) {
/* insert this inode into the doubly-linked list
* where chl points */
if ((iq = chl->chl_ip)) {
ip->i_cprev = iq->i_cprev;
iq->i_cprev->i_cnext = ip;
iq->i_cprev = ip;
ip->i_cnext = iq;
} else {
ip->i_cnext = ip;
ip->i_cprev = ip;
}
chl->chl_ip = ip;
ip->i_chash = chl;
break;
}
}
/* no hash list found for this block; add a new hash list */
if (chl == NULL) {
if (chlnew == NULL) {
mutex_spinunlock(&ch->ch_lock, s);
ASSERT(xfs_chashlist_zone != NULL);
chlnew = (xfs_chashlist_t *)
kmem_zone_alloc(xfs_chashlist_zone,
KM_SLEEP);
ASSERT(chlnew != NULL);
goto chlredo;
} else {
ip->i_cnext = ip;
ip->i_cprev = ip;
ip->i_chash = chlnew;
chlnew->chl_ip = ip;
chlnew->chl_blkno = ip->i_blkno;
chlnew->chl_next = ch->ch_list;
ch->ch_list = chlnew;
chlnew = NULL;
}
} else {
if (chlnew != NULL) {
kmem_zone_free(xfs_chashlist_zone, chlnew);
}
}
mutex_spinunlock(&ch->ch_lock, s);
/*
* Link ip to its mount and thread it on the mount's inode list.
*/
XFS_MOUNT_ILOCK(mp);
if ((iq = mp->m_inodes)) {
ASSERT(iq->i_mprev->i_mnext == iq);
ip->i_mprev = iq->i_mprev;
iq->i_mprev->i_mnext = ip;
iq->i_mprev = ip;
ip->i_mnext = iq;
} else {
ip->i_mnext = ip;
ip->i_mprev = ip;
}
mp->m_inodes = ip;
XFS_MOUNT_IUNLOCK(mp);
newnode = 1;
return_ip:
ASSERT(ip->i_df.if_ext_max ==
XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
*ipp = ip;
/*
* If we have a real type for an on-disk inode, we can set ops(&unlock)
* now. If it's a new inode being created, xfs_ialloc will handle it.
*/
XVFS_INIT_VNODE(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1);
return 0;
}
