[linux.git] / fs / xfs / xfs_rename.c

/*
 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
 * All Rights Reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_types.h"
#include "xfs_log.h"
#include "xfs_inum.h"
#include "xfs_trans.h"
#include "xfs_sb.h"
#include "xfs_dir2.h"
#include "xfs_dmapi.h"
#include "xfs_mount.h"
#include "xfs_da_btree.h"
#include "xfs_bmap_btree.h"
#include "xfs_dir2_sf.h"
#include "xfs_attr_sf.h"
#include "xfs_dinode.h"
#include "xfs_inode.h"
#include "xfs_inode_item.h"
#include "xfs_bmap.h"
#include "xfs_error.h"
#include "xfs_quota.h"
#include "xfs_refcache.h"
#include "xfs_utils.h"
#include "xfs_trans_space.h"


/*
 * Given an array of up to 4 inode pointers, unlock the pointed to inodes.
 * If there are fewer than 4 entries in the array, the empty entries will
 * be at the end and will have NULL pointers in them.
 */
STATIC void
xfs_rename_unlock4(
	xfs_inode_t	**i_tab,
	uint		lock_mode)
{
	int	i;

	xfs_iunlock(i_tab[0], lock_mode);
	for (i = 1; i < 4; i++) {
		if (i_tab[i] == NULL) {
			break;
		}
		/*
		 * Watch out for duplicate entries in the table.
		 */
		if (i_tab[i] != i_tab[i-1]) {
			xfs_iunlock(i_tab[i], lock_mode);
		}
	}
}

#ifdef DEBUG
int xfs_rename_skip, xfs_rename_nskip;
#endif

/*
 * 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;
}

/*
 * xfs_rename
 */
int
xfs_rename(
	bhv_desc_t	*src_dir_bdp,
	bhv_vname_t	*src_vname,
	bhv_vnode_t	*target_dir_vp,
	bhv_vname_t	*target_vname,
	cred_t		*credp)
{
	xfs_trans_t	*tp;
	xfs_inode_t	*src_dp, *target_dp, *src_ip, *target_ip;
	xfs_mount_t	*mp;
	int		new_parent;		/* moving to a new dir */
	int		src_is_directory;	/* src_name is a directory */
	int		error;
	xfs_bmap_free_t free_list;
	xfs_fsblock_t   first_block;
	int		cancel_flags;
	int		committed;
	xfs_inode_t	*inodes[4];
	int		target_ip_dropped = 0;	/* dropped target_ip link? */
	bhv_vnode_t	*src_dir_vp;
	int		spaceres;
	int		target_link_zero = 0;
	int		num_inodes;
	char		*src_name = VNAME(src_vname);
	char		*target_name = VNAME(target_vname);
	int		src_namelen = VNAMELEN(src_vname);
	int		target_namelen = VNAMELEN(target_vname);

	src_dir_vp = BHV_TO_VNODE(src_dir_bdp);
	vn_trace_entry(src_dir_vp, "xfs_rename", (inst_t *)__return_address);
	vn_trace_entry(target_dir_vp, "xfs_rename", (inst_t *)__return_address);

	/*
	 * Find the XFS behavior descriptor for the target directory
	 * vnode since it was not handed to us.
	 */
	target_dp = xfs_vtoi(target_dir_vp);
	if (target_dp == NULL) {
		return XFS_ERROR(EXDEV);
	}

	src_dp = XFS_BHVTOI(src_dir_bdp);
	mp = src_dp->i_mount;

	if (DM_EVENT_ENABLED(src_dir_vp->v_vfsp, src_dp, DM_EVENT_RENAME) ||
	    DM_EVENT_ENABLED(target_dir_vp->v_vfsp,
				target_dp, DM_EVENT_RENAME)) {
		error = XFS_SEND_NAMESP(mp, DM_EVENT_RENAME,
					src_dir_vp, DM_RIGHT_NULL,
					target_dir_vp, DM_RIGHT_NULL,
					src_name, target_name,
					0, 0, 0);
		if (error) {
			return error;
		}
	}
	/* Return through std_return after this point. */

	/*
	 * Lock all the participating inodes. Depending upon whether
	 * the target_name exists in the target directory, and
	 * whether the target directory is the same as the source
	 * directory, we can lock from 2 to 4 inodes.
	 * xfs_lock_for_rename() will return ENOENT if src_name
	 * does not exist in the source directory.
	 */
	tp = NULL;
	error = xfs_lock_for_rename(src_dp, target_dp, src_vname,
			target_vname, &src_ip, &target_ip, inodes,
			&num_inodes);

	if (error) {
		/*
		 * We have nothing locked, no inode references, and
		 * no transaction, so just get out.
		 */
		goto std_return;
	}

	ASSERT(src_ip != NULL);

	if ((src_ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
		/*
		 * Check for link count overflow on target_dp
		 */
		if (target_ip == NULL && (src_dp != target_dp) &&
		    target_dp->i_d.di_nlink >= XFS_MAXLINK) {
			error = XFS_ERROR(EMLINK);
			xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);
			goto rele_return;
		}
	}

	/*
	 * If we are using project inheritance, we only allow renames
	 * into our tree when the project IDs are the same; else the
	 * tree quota mechanism would be circumvented.
	 */
	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
		     (target_dp->i_d.di_projid != src_ip->i_d.di_projid))) {
		error = XFS_ERROR(EXDEV);
		xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);
		goto rele_return;
	}

	new_parent = (src_dp != target_dp);
	src_is_directory = ((src_ip->i_d.di_mode & S_IFMT) == S_IFDIR);

	/*
	 * Drop the locks on our inodes so that we can start the transaction.
	 */
	xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);

	XFS_BMAP_INIT(&free_list, &first_block);
	tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
	spaceres = XFS_RENAME_SPACE_RES(mp, target_namelen);
	error = xfs_trans_reserve(tp, spaceres, XFS_RENAME_LOG_RES(mp), 0,
			XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
	if (error == ENOSPC) {
		spaceres = 0;
		error = xfs_trans_reserve(tp, 0, XFS_RENAME_LOG_RES(mp), 0,
				XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
	}
	if (error) {
		xfs_trans_cancel(tp, 0);
		goto rele_return;
	}

	/*
	 * Attach the dquots to the inodes
	 */
	if ((error = XFS_QM_DQVOPRENAME(mp, inodes))) {
		xfs_trans_cancel(tp, cancel_flags);
		goto rele_return;
	}

	/*
	 * Reacquire the inode locks we dropped above.
	 */
	xfs_lock_inodes(inodes, num_inodes, 0, XFS_ILOCK_EXCL);

	/*
	 * Join all the inodes to the transaction. From this point on,
	 * we can rely on either trans_commit or trans_cancel to unlock
	 * them.  Note that we need to add a vnode reference to the
	 * directories since trans_commit & trans_cancel will decrement
	 * them when they unlock the inodes.  Also, we need to be careful
	 * not to add an inode to the transaction more than once.
	 */
	VN_HOLD(src_dir_vp);
	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
	if (new_parent) {
		VN_HOLD(target_dir_vp);
		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
	}
	if ((src_ip != src_dp) && (src_ip != target_dp)) {
		xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
	}
	if ((target_ip != NULL) &&
	    (target_ip != src_ip) &&
	    (target_ip != src_dp) &&
	    (target_ip != target_dp)) {
		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
	}

	/*
	 * Set up the target.
	 */
	if (target_ip == NULL) {
		/*
		 * If there's no space reservation, check the entry will
		 * fit before actually inserting it.
		 */
		if (spaceres == 0 &&
		    (error = xfs_dir_canenter(tp, target_dp, target_name,
						target_namelen)))
			goto error_return;
		/*
		 * If target does not exist and the rename crosses
		 * directories, adjust the target directory link count
		 * to account for the ".." reference from the new entry.
		 */
		error = xfs_dir_createname(tp, target_dp, target_name,
					   target_namelen, src_ip->i_ino,
					   &first_block, &free_list, spaceres);
		if (error == ENOSPC)
			goto error_return;
		if (error)
			goto abort_return;
		xfs_ichgtime(target_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);

		if (new_parent && src_is_directory) {
			error = xfs_bumplink(tp, target_dp);
			if (error)
				goto abort_return;
		}
	} else { /* target_ip != NULL */
		/*
		 * If target exists and it's a directory, check that both
		 * target and source are directories and that target can be
		 * destroyed, or that neither is a directory.
		 */
		if ((target_ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
			/*
			 * Make sure target dir is empty.
			 */
			if (!(xfs_dir_isempty(target_ip)) ||
			    (target_ip->i_d.di_nlink > 2)) {
				error = XFS_ERROR(EEXIST);
				goto error_return;
			}
		}

		/*
		 * Link the source inode under the target name.
		 * If the source inode is a directory and we are moving
		 * it across directories, its ".." entry will be
		 * inconsistent until we replace that down below.
		 *
		 * In case there is already an entry with the same
		 * name at the destination directory, remove it first.
		 */
		error = xfs_dir_replace(tp, target_dp, target_name,
					target_namelen, src_ip->i_ino,
					&first_block, &free_list, spaceres);
		if (error)
			goto abort_return;
		xfs_ichgtime(target_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);

		/*
		 * Decrement the link count on the target since the target
		 * dir no longer points to it.
		 */
		error = xfs_droplink(tp, target_ip);
		if (error)
			goto abort_return;
		target_ip_dropped = 1;

		if (src_is_directory) {
			/*
			 * Drop the link from the old "." entry.
			 */
			error = xfs_droplink(tp, target_ip);
			if (error)
				goto abort_return;
		}

		/* Do this test while we still hold the locks */
		target_link_zero = (target_ip)->i_d.di_nlink==0;

	} /* target_ip != NULL */

	/*
	 * Remove the source.
	 */
	if (new_parent && src_is_directory) {
		/*
		 * Rewrite the ".." entry to point to the new
		 * directory.
		 */
		error = xfs_dir_replace(tp, src_ip, "..", 2, target_dp->i_ino,
					&first_block, &free_list, spaceres);
		ASSERT(error != EEXIST);
		if (error)
			goto abort_return;
		xfs_ichgtime(src_ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);

	} else {
		/*
		 * We always want to hit the ctime on the source inode.
		 * We do it in the if clause above for the 'new_parent &&
		 * src_is_directory' case, and here we get all the other
		 * cases.  This isn't strictly required by the standards
		 * since the source inode isn't really being changed,
		 * but old unix file systems did it and some incremental
		 * backup programs won't work without it.
		 */
		xfs_ichgtime(src_ip, XFS_ICHGTIME_CHG);
	}

	/*
	 * Adjust the link count on src_dp.  This is necessary when
	 * renaming a directory, either within one parent when
	 * the target existed, or across two parent directories.
	 */
	if (src_is_directory && (new_parent || target_ip != NULL)) {

		/*
		 * Decrement link count on src_directory since the
		 * entry that's moved no longer points to it.
		 */
		error = xfs_droplink(tp, src_dp);
		if (error)
			goto abort_return;
	}

	error = xfs_dir_removename(tp, src_dp, src_name, src_namelen,
			src_ip->i_ino, &first_block, &free_list, spaceres);
	if (error)
		goto abort_return;
	xfs_ichgtime(src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);

	/*
	 * Update the generation counts on all the directory inodes
	 * that we're modifying.
	 */
	src_dp->i_gen++;
	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);

	if (new_parent) {
		target_dp->i_gen++;
		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
	}

	/*
	 * If there was a target inode, take an extra reference on
	 * it here so that it doesn't go to xfs_inactive() from
	 * within the commit.
	 */
	if (target_ip != NULL) {
		IHOLD(target_ip);
	}

	/*
	 * If this is a synchronous mount, make sure that the
	 * rename transaction goes to disk before returning to
	 * the user.
	 */
	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
		xfs_trans_set_sync(tp);
	}

	/*
	 * Take refs. for vop_link_removed calls below.  No need to worry
	 * about directory refs. because the caller holds them.
	 *
	 * Do holds before the xfs_bmap_finish since it might rele them down
	 * to zero.
	 */

	if (target_ip_dropped)
		IHOLD(target_ip);
	IHOLD(src_ip);

	error = xfs_bmap_finish(&tp, &free_list, &committed);
	if (error) {
		xfs_bmap_cancel(&free_list);
		xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
				 XFS_TRANS_ABORT));
		if (target_ip != NULL) {
			IRELE(target_ip);
		}
		if (target_ip_dropped) {
			IRELE(target_ip);
		}
		IRELE(src_ip);
		goto std_return;
	}

	/*
	 * trans_commit will unlock src_ip, target_ip & decrement
	 * the vnode references.
	 */
	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
	if (target_ip != NULL) {
		xfs_refcache_purge_ip(target_ip);
		IRELE(target_ip);
	}
	/*
	 * Let interposed file systems know about removed links.
	 */
	if (target_ip_dropped) {
		bhv_vop_link_removed(XFS_ITOV(target_ip), target_dir_vp,
					target_link_zero);
		IRELE(target_ip);
	}

	IRELE(src_ip);

	/* Fall through to std_return with error = 0 or errno from
	 * xfs_trans_commit	 */
std_return:
	if (DM_EVENT_ENABLED(src_dir_vp->v_vfsp, src_dp, DM_EVENT_POSTRENAME) ||
	    DM_EVENT_ENABLED(target_dir_vp->v_vfsp,
				target_dp, DM_EVENT_POSTRENAME)) {
		(void) XFS_SEND_NAMESP (mp, DM_EVENT_POSTRENAME,
					src_dir_vp, DM_RIGHT_NULL,
					target_dir_vp, DM_RIGHT_NULL,
					src_name, target_name,
					0, error, 0);
	}
	return error;

 abort_return:
	cancel_flags |= XFS_TRANS_ABORT;
	/* FALLTHROUGH */
 error_return:
	xfs_bmap_cancel(&free_list);
	xfs_trans_cancel(tp, cancel_flags);
	goto std_return;

 rele_return:
	IRELE(src_ip);
	if (target_ip != NULL) {
		IRELE(target_ip);
	}
	goto std_return;
}
Commit	Line	Data
1da177e4	1	/*
7b718769 NS	2	* Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
7b718769 NS	3	* All Rights Reserved.
1da177e4	4	*
7b718769 NS	5	* This program is free software; you can redistribute it and/or
7b718769 NS	6	* modify it under the terms of the GNU General Public License as
1da177e4 LT	7	* published by the Free Software Foundation.
1da177e4 LT	8	*
7b718769 NS	9	* This program is distributed in the hope that it would be useful,
	10	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	12	* GNU General Public License for more details.
1da177e4	13	*
7b718769 NS	14	* You should have received a copy of the GNU General Public License
	15	* along with this program; if not, write the Free Software Foundation,
	16	* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
1da177e4	17	*/
1da177e4	18	#include "xfs.h"
a844f451	19	#include "xfs_fs.h"
1da177e4	20	#include "xfs_types.h"
1da177e4	21	#include "xfs_log.h"
a844f451	22	#include "xfs_inum.h"
1da177e4 LT	23	#include "xfs_trans.h"
1da177e4 LT	24	#include "xfs_sb.h"
1da177e4 LT	25	#include "xfs_dir2.h"
	26	#include "xfs_dmapi.h"
	27	#include "xfs_mount.h"
a844f451	28	#include "xfs_da_btree.h"
1da177e4	29	#include "xfs_bmap_btree.h"
1da177e4	30	#include "xfs_dir2_sf.h"
a844f451	31	#include "xfs_attr_sf.h"
1da177e4	32	#include "xfs_dinode.h"
1da177e4	33	#include "xfs_inode.h"
a844f451	34	#include "xfs_inode_item.h"
1da177e4 LT	35	#include "xfs_bmap.h"
	36	#include "xfs_error.h"
	37	#include "xfs_quota.h"
	38	#include "xfs_refcache.h"
	39	#include "xfs_utils.h"
	40	#include "xfs_trans_space.h"
1da177e4 LT	41
	42
	43	/*
	44	* Given an array of up to 4 inode pointers, unlock the pointed to inodes.
	45	* If there are fewer than 4 entries in the array, the empty entries will
	46	* be at the end and will have NULL pointers in them.
	47	*/
	48	STATIC void
	49	xfs_rename_unlock4(
	50	xfs_inode_t **i_tab,
	51	uint lock_mode)
	52	{
	53	int i;
	54
	55	xfs_iunlock(i_tab[0], lock_mode);
	56	for (i = 1; i < 4; i++) {
	57	if (i_tab[i] == NULL) {
	58	break;
	59	}
	60	/*
	61	* Watch out for duplicate entries in the table.
	62	*/
	63	if (i_tab[i] != i_tab[i-1]) {
	64	xfs_iunlock(i_tab[i], lock_mode);
	65	}
	66	}
	67	}
	68
	69	#ifdef DEBUG
	70	int xfs_rename_skip, xfs_rename_nskip;
	71	#endif
	72
	73	/*
	74	* The following routine will acquire the locks required for a rename
	75	* operation. The code understands the semantics of renames and will
	76	* validate that name1 exists under dp1 & that name2 may or may not
	77	* exist under dp2.
	78	*
	79	* We are renaming dp1/name1 to dp2/name2.
	80	*
	81	* Return ENOENT if dp1 does not exist, other lookup errors, or 0 for success.
	82	*/
	83	STATIC int
	84	xfs_lock_for_rename(
	85	xfs_inode_t dp1, / old (source) directory inode */
	86	xfs_inode_t dp2, / new (target) directory inode */
8285fb58 NS	87	bhv_vname_t vname1,/ old entry name */
8285fb58 NS	88	bhv_vname_t vname2,/ new entry name */
1da177e4 LT	89	xfs_inode_t *ipp1, / inode of old entry */
	90	xfs_inode_t *ipp2, / inode of new entry, if it
	91	already exists, NULL otherwise. */
	92	xfs_inode_t *i_tab,/ array of inode returned, sorted */
	93	int num_inodes) / number of inodes in array */
	94	{
	95	xfs_inode_t ip1, ip2, *temp;
	96	xfs_ino_t inum1, inum2;
	97	int error;
	98	int i, j;
	99	uint lock_mode;
	100	int diff_dirs = (dp1 != dp2);
	101
	102	ip2 = NULL;
	103
	104	/*
	105	* First, find out the current inums of the entries so that we
	106	* can determine the initial locking order. We'll have to
	107	* sanity check stuff after all the locks have been acquired
	108	* to see if we still have the right inodes, directories, etc.
	109	*/
	110	lock_mode = xfs_ilock_map_shared(dp1);
	111	error = xfs_get_dir_entry(vname1, &ip1);
	112	if (error) {
	113	xfs_iunlock_map_shared(dp1, lock_mode);
	114	return error;
	115	}
	116
	117	inum1 = ip1->i_ino;
	118
	119	ASSERT(ip1);
	120	ITRACE(ip1);
	121
	122	/*
	123	* Unlock dp1 and lock dp2 if they are different.
	124	*/
	125
	126	if (diff_dirs) {
	127	xfs_iunlock_map_shared(dp1, lock_mode);
	128	lock_mode = xfs_ilock_map_shared(dp2);
	129	}
	130
	131	error = xfs_dir_lookup_int(XFS_ITOBHV(dp2), lock_mode,
	132	vname2, &inum2, &ip2);
	133	if (error == ENOENT) { /* target does not need to exist. */
	134	inum2 = 0;
	135	} else if (error) {
	136	/*
	137	* If dp2 and dp1 are the same, the next line unlocks dp1.
	138	* Got it?
	139	*/
	140	xfs_iunlock_map_shared(dp2, lock_mode);
	141	IRELE (ip1);
	142	return error;
	143	} else {
	144	ITRACE(ip2);
	145	}
	146
	147	/*
	148	* i_tab contains a list of pointers to inodes. We initialize
	149	* the table here & we'll sort it. We will then use it to
	150	* order the acquisition of the inode locks.
	151	*
	152	* Note that the table may contain duplicates. e.g., dp1 == dp2.
153	*/
154	i_tab[0] = dp1;
155	i_tab[1] = dp2;
156	i_tab[2] = ip1;
157	if (inum2 == 0) {
158	*num_inodes = 3;
159	i_tab[3] = NULL;
160	} else {
161	*num_inodes = 4;
162	i_tab[3] = ip2;
163	}
164
165	/*
166	* Sort the elements via bubble sort. (Remember, there are at
167	* most 4 elements to sort, so this is adequate.)
168	*/
169	for (i=0; i < *num_inodes; i++) {
170	for (j=1; j < *num_inodes; j++) {
171	if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
172	temp = i_tab[j];
173	i_tab[j] = i_tab[j-1];
174	i_tab[j-1] = temp;
175	}
176	}
177	}
178
179	/*
180	* We have dp2 locked. If it isn't first, unlock it.
181	* If it is first, tell xfs_lock_inodes so it can skip it
182	* when locking. if dp1 == dp2, xfs_lock_inodes will skip both
183	* since they are equal. xfs_lock_inodes needs all these inodes
184	* so that it can unlock and retry if there might be a dead-lock
185	* potential with the log.
186	*/
187
188	if (i_tab[0] == dp2 && lock_mode == XFS_ILOCK_SHARED) {
189	#ifdef DEBUG
190	xfs_rename_skip++;
191	#endif
192	xfs_lock_inodes(i_tab, *num_inodes, 1, XFS_ILOCK_SHARED);
193	} else {
194	#ifdef DEBUG
195	xfs_rename_nskip++;
196	#endif
197	xfs_iunlock_map_shared(dp2, lock_mode);
198	xfs_lock_inodes(i_tab, *num_inodes, 0, XFS_ILOCK_SHARED);
199	}
200
201	/*
202	* Set the return value. Null out any unused entries in i_tab.
203	*/
204	ipp1 = ipp2 = NULL;
205	for (i=0; i < *num_inodes; i++) {
206	if (i_tab[i]->i_ino == inum1) {
207	*ipp1 = i_tab[i];
208	}
209	if (i_tab[i]->i_ino == inum2) {
210	*ipp2 = i_tab[i];
211	}
212	}
213	for (;i < 4; i++) {
214	i_tab[i] = NULL;
215	}
216	return 0;
217	}
218
1da177e4 LT	219	/*
	220	* xfs_rename
	221	*/
	222	int
	223	xfs_rename(
	224	bhv_desc_t *src_dir_bdp,
8285fb58	225	bhv_vname_t *src_vname,
67fcaa73	226	bhv_vnode_t *target_dir_vp,
8285fb58	227	bhv_vname_t *target_vname,
1da177e4 LT	228	cred_t *credp)
	229	{
	230	xfs_trans_t *tp;
	231	xfs_inode_t src_dp, target_dp, src_ip, target_ip;
	232	xfs_mount_t *mp;
	233	int new_parent; /* moving to a new dir */
	234	int src_is_directory; /* src_name is a directory */
	235	int error;
	236	xfs_bmap_free_t free_list;
	237	xfs_fsblock_t first_block;
	238	int cancel_flags;
	239	int committed;
	240	xfs_inode_t *inodes[4];
	241	int target_ip_dropped = 0; /* dropped target_ip link? */
67fcaa73	242	bhv_vnode_t *src_dir_vp;
1da177e4 LT	243	int spaceres;
	244	int target_link_zero = 0;
	245	int num_inodes;
	246	char *src_name = VNAME(src_vname);
	247	char *target_name = VNAME(target_vname);
	248	int src_namelen = VNAMELEN(src_vname);
	249	int target_namelen = VNAMELEN(target_vname);
	250
	251	src_dir_vp = BHV_TO_VNODE(src_dir_bdp);
	252	vn_trace_entry(src_dir_vp, "xfs_rename", (inst_t *)__return_address);
	253	vn_trace_entry(target_dir_vp, "xfs_rename", (inst_t *)__return_address);
	254
	255	/*
	256	* Find the XFS behavior descriptor for the target directory
	257	* vnode since it was not handed to us.
	258	*/
75e17b3c CH	259	target_dp = xfs_vtoi(target_dir_vp);
75e17b3c CH	260	if (target_dp == NULL) {
1da177e4 LT	261	return XFS_ERROR(EXDEV);
	262	}
	263
	264	src_dp = XFS_BHVTOI(src_dir_bdp);
1da177e4 LT	265	mp = src_dp->i_mount;
	266
	267	if (DM_EVENT_ENABLED(src_dir_vp->v_vfsp, src_dp, DM_EVENT_RENAME) \|\|
	268	DM_EVENT_ENABLED(target_dir_vp->v_vfsp,
	269	target_dp, DM_EVENT_RENAME)) {
	270	error = XFS_SEND_NAMESP(mp, DM_EVENT_RENAME,
	271	src_dir_vp, DM_RIGHT_NULL,
	272	target_dir_vp, DM_RIGHT_NULL,
	273	src_name, target_name,
	274	0, 0, 0);
	275	if (error) {
	276	return error;
	277	}
	278	}
	279	/* Return through std_return after this point. */
	280
	281	/*
	282	* Lock all the participating inodes. Depending upon whether
	283	* the target_name exists in the target directory, and
	284	* whether the target directory is the same as the source
	285	* directory, we can lock from 2 to 4 inodes.
	286	* xfs_lock_for_rename() will return ENOENT if src_name
	287	* does not exist in the source directory.
	288	*/
	289	tp = NULL;
	290	error = xfs_lock_for_rename(src_dp, target_dp, src_vname,
	291	target_vname, &src_ip, &target_ip, inodes,
	292	&num_inodes);
	293
	294	if (error) {
1da177e4 LT	295	/*
	296	* We have nothing locked, no inode references, and
	297	* no transaction, so just get out.
	298	*/
	299	goto std_return;
	300	}
	301
	302	ASSERT(src_ip != NULL);
	303
	304	if ((src_ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
	305	/*
	306	* Check for link count overflow on target_dp
	307	*/
	308	if (target_ip == NULL && (src_dp != target_dp) &&
	309	target_dp->i_d.di_nlink >= XFS_MAXLINK) {
1da177e4 LT	310	error = XFS_ERROR(EMLINK);
	311	xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);
	312	goto rele_return;
	313	}
	314	}
	315
b1ecdda9 NS	316	/*
	317	* If we are using project inheritance, we only allow renames
	318	* into our tree when the project IDs are the same; else the
	319	* tree quota mechanism would be circumvented.
	320	*/
	321	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
	322	(target_dp->i_d.di_projid != src_ip->i_d.di_projid))) {
	323	error = XFS_ERROR(EXDEV);
	324	xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);
	325	goto rele_return;
	326	}
	327
1da177e4 LT	328	new_parent = (src_dp != target_dp);
	329	src_is_directory = ((src_ip->i_d.di_mode & S_IFMT) == S_IFDIR);
	330
	331	/*
	332	* Drop the locks on our inodes so that we can start the transaction.
	333	*/
	334	xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);
	335
	336	XFS_BMAP_INIT(&free_list, &first_block);
	337	tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
	338	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
	339	spaceres = XFS_RENAME_SPACE_RES(mp, target_namelen);
	340	error = xfs_trans_reserve(tp, spaceres, XFS_RENAME_LOG_RES(mp), 0,
	341	XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
	342	if (error == ENOSPC) {
	343	spaceres = 0;
	344	error = xfs_trans_reserve(tp, 0, XFS_RENAME_LOG_RES(mp), 0,
	345	XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
	346	}
	347	if (error) {
1da177e4 LT	348	xfs_trans_cancel(tp, 0);
	349	goto rele_return;
	350	}
	351
	352	/*
	353	* Attach the dquots to the inodes
	354	*/
	355	if ((error = XFS_QM_DQVOPRENAME(mp, inodes))) {
	356	xfs_trans_cancel(tp, cancel_flags);
1da177e4 LT	357	goto rele_return;
	358	}
	359
	360	/*
	361	* Reacquire the inode locks we dropped above.
	362	*/
	363	xfs_lock_inodes(inodes, num_inodes, 0, XFS_ILOCK_EXCL);
	364
	365	/*
	366	* Join all the inodes to the transaction. From this point on,
	367	* we can rely on either trans_commit or trans_cancel to unlock
	368	* them. Note that we need to add a vnode reference to the
	369	* directories since trans_commit & trans_cancel will decrement
	370	* them when they unlock the inodes. Also, we need to be careful
	371	* not to add an inode to the transaction more than once.
	372	*/
	373	VN_HOLD(src_dir_vp);
	374	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
	375	if (new_parent) {
	376	VN_HOLD(target_dir_vp);
	377	xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
	378	}
	379	if ((src_ip != src_dp) && (src_ip != target_dp)) {
	380	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
	381	}
	382	if ((target_ip != NULL) &&
	383	(target_ip != src_ip) &&
	384	(target_ip != src_dp) &&
	385	(target_ip != target_dp)) {
	386	xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
	387	}
	388
	389	/*
	390	* Set up the target.
	391	*/
	392	if (target_ip == NULL) {
	393	/*
	394	* If there's no space reservation, check the entry will
	395	* fit before actually inserting it.
	396	*/
	397	if (spaceres == 0 &&
f6c2d1fa NS	398	(error = xfs_dir_canenter(tp, target_dp, target_name,
f6c2d1fa NS	399	target_namelen)))
1da177e4	400	goto error_return;
1da177e4 LT	401	/*
	402	* If target does not exist and the rename crosses
	403	* directories, adjust the target directory link count
	404	* to account for the ".." reference from the new entry.
	405	*/
f6c2d1fa	406	error = xfs_dir_createname(tp, target_dp, target_name,
1da177e4 LT	407	target_namelen, src_ip->i_ino,
1da177e4 LT	408	&first_block, &free_list, spaceres);
f6c2d1fa	409	if (error == ENOSPC)
1da177e4	410	goto error_return;
f6c2d1fa	411	if (error)
1da177e4	412	goto abort_return;
1da177e4 LT	413	xfs_ichgtime(target_dp, XFS_ICHGTIME_MOD \| XFS_ICHGTIME_CHG);
	414
	415	if (new_parent && src_is_directory) {
	416	error = xfs_bumplink(tp, target_dp);
f6c2d1fa	417	if (error)
1da177e4	418	goto abort_return;
1da177e4 LT	419	}
1da177e4 LT	420	} else { /* target_ip != NULL */
1da177e4 LT	421	/*
	422	* If target exists and it's a directory, check that both
	423	* target and source are directories and that target can be
	424	* destroyed, or that neither is a directory.
	425	*/
	426	if ((target_ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
	427	/*
	428	* Make sure target dir is empty.
	429	*/
f6c2d1fa	430	if (!(xfs_dir_isempty(target_ip)) \|\|
1da177e4 LT	431	(target_ip->i_d.di_nlink > 2)) {
1da177e4 LT	432	error = XFS_ERROR(EEXIST);
1da177e4 LT	433	goto error_return;
	434	}
	435	}
	436
	437	/*
	438	* Link the source inode under the target name.
	439	* If the source inode is a directory and we are moving
	440	* it across directories, its ".." entry will be
	441	* inconsistent until we replace that down below.
	442	*
	443	* In case there is already an entry with the same
	444	* name at the destination directory, remove it first.
	445	*/
f6c2d1fa NS	446	error = xfs_dir_replace(tp, target_dp, target_name,
	447	target_namelen, src_ip->i_ino,
	448	&first_block, &free_list, spaceres);
	449	if (error)
1da177e4	450	goto abort_return;
1da177e4 LT	451	xfs_ichgtime(target_dp, XFS_ICHGTIME_MOD \| XFS_ICHGTIME_CHG);
	452
	453	/*
	454	* Decrement the link count on the target since the target
	455	* dir no longer points to it.
	456	*/
	457	error = xfs_droplink(tp, target_ip);
f6c2d1fa	458	if (error)
1da177e4	459	goto abort_return;
1da177e4 LT	460	target_ip_dropped = 1;
	461
	462	if (src_is_directory) {
	463	/*
	464	* Drop the link from the old "." entry.
	465	*/
	466	error = xfs_droplink(tp, target_ip);
f6c2d1fa	467	if (error)
1da177e4	468	goto abort_return;
1da177e4 LT	469	}
	470
	471	/* Do this test while we still hold the locks */
	472	target_link_zero = (target_ip)->i_d.di_nlink==0;
	473
	474	} /* target_ip != NULL */
	475
	476	/*
	477	* Remove the source.
	478	*/
	479	if (new_parent && src_is_directory) {
1da177e4 LT	480	/*
	481	* Rewrite the ".." entry to point to the new
	482	* directory.
	483	*/
f6c2d1fa NS	484	error = xfs_dir_replace(tp, src_ip, "..", 2, target_dp->i_ino,
f6c2d1fa NS	485	&first_block, &free_list, spaceres);
1da177e4	486	ASSERT(error != EEXIST);
f6c2d1fa	487	if (error)
1da177e4	488	goto abort_return;
1da177e4 LT	489	xfs_ichgtime(src_ip, XFS_ICHGTIME_MOD \| XFS_ICHGTIME_CHG);
	490
	491	} else {
	492	/*
	493	* We always want to hit the ctime on the source inode.
	494	* We do it in the if clause above for the 'new_parent &&
	495	* src_is_directory' case, and here we get all the other
	496	* cases. This isn't strictly required by the standards
	497	* since the source inode isn't really being changed,
	498	* but old unix file systems did it and some incremental
	499	* backup programs won't work without it.
	500	*/
	501	xfs_ichgtime(src_ip, XFS_ICHGTIME_CHG);
	502	}
	503
	504	/*
	505	* Adjust the link count on src_dp. This is necessary when
	506	* renaming a directory, either within one parent when
	507	* the target existed, or across two parent directories.
	508	*/
	509	if (src_is_directory && (new_parent \|\| target_ip != NULL)) {
	510
	511	/*
	512	* Decrement link count on src_directory since the
	513	* entry that's moved no longer points to it.
	514	*/
	515	error = xfs_droplink(tp, src_dp);
f6c2d1fa	516	if (error)
1da177e4	517	goto abort_return;
1da177e4 LT	518	}
1da177e4 LT	519
f6c2d1fa	520	error = xfs_dir_removename(tp, src_dp, src_name, src_namelen,
1da177e4	521	src_ip->i_ino, &first_block, &free_list, spaceres);
f6c2d1fa	522	if (error)
1da177e4	523	goto abort_return;
1da177e4 LT	524	xfs_ichgtime(src_dp, XFS_ICHGTIME_MOD \| XFS_ICHGTIME_CHG);
	525
	526	/*
	527	* Update the generation counts on all the directory inodes
	528	* that we're modifying.
	529	*/
	530	src_dp->i_gen++;
	531	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
	532
	533	if (new_parent) {
	534	target_dp->i_gen++;
	535	xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
	536	}
	537
	538	/*
	539	* If there was a target inode, take an extra reference on
	540	* it here so that it doesn't go to xfs_inactive() from
	541	* within the commit.
	542	*/
	543	if (target_ip != NULL) {
	544	IHOLD(target_ip);
	545	}
	546
	547	/*
	548	* If this is a synchronous mount, make sure that the
	549	* rename transaction goes to disk before returning to
	550	* the user.
	551	*/
	552	if (mp->m_flags & (XFS_MOUNT_WSYNC\|XFS_MOUNT_DIRSYNC)) {
	553	xfs_trans_set_sync(tp);
	554	}
	555
	556	/*
	557	* Take refs. for vop_link_removed calls below. No need to worry
	558	* about directory refs. because the caller holds them.
	559	*
	560	* Do holds before the xfs_bmap_finish since it might rele them down
	561	* to zero.
	562	*/
	563
	564	if (target_ip_dropped)
	565	IHOLD(target_ip);
	566	IHOLD(src_ip);
	567
f7c99b6f	568	error = xfs_bmap_finish(&tp, &free_list, &committed);
1da177e4 LT	569	if (error) {
	570	xfs_bmap_cancel(&free_list);
	571	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES \|
	572	XFS_TRANS_ABORT));
	573	if (target_ip != NULL) {
	574	IRELE(target_ip);
	575	}
	576	if (target_ip_dropped) {
	577	IRELE(target_ip);
	578	}
	579	IRELE(src_ip);
	580	goto std_return;
	581	}
	582
	583	/*
	584	* trans_commit will unlock src_ip, target_ip & decrement
	585	* the vnode references.
	586	*/
1c72bf90	587	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1da177e4 LT	588	if (target_ip != NULL) {
	589	xfs_refcache_purge_ip(target_ip);
	590	IRELE(target_ip);
	591	}
	592	/*
	593	* Let interposed file systems know about removed links.
	594	*/
	595	if (target_ip_dropped) {
67fcaa73	596	bhv_vop_link_removed(XFS_ITOV(target_ip), target_dir_vp,
1da177e4 LT	597	target_link_zero);
	598	IRELE(target_ip);
	599	}
	600
1da177e4 LT	601	IRELE(src_ip);
	602
	603	/* Fall through to std_return with error = 0 or errno from
	604	* xfs_trans_commit */
	605	std_return:
	606	if (DM_EVENT_ENABLED(src_dir_vp->v_vfsp, src_dp, DM_EVENT_POSTRENAME) \|\|
	607	DM_EVENT_ENABLED(target_dir_vp->v_vfsp,
	608	target_dp, DM_EVENT_POSTRENAME)) {
	609	(void) XFS_SEND_NAMESP (mp, DM_EVENT_POSTRENAME,
	610	src_dir_vp, DM_RIGHT_NULL,
	611	target_dir_vp, DM_RIGHT_NULL,
	612	src_name, target_name,
	613	0, error, 0);
	614	}
	615	return error;
	616
	617	abort_return:
	618	cancel_flags \|= XFS_TRANS_ABORT;
	619	/* FALLTHROUGH */
	620	error_return:
	621	xfs_bmap_cancel(&free_list);
	622	xfs_trans_cancel(tp, cancel_flags);
	623	goto std_return;
	624
	625	rele_return:
	626	IRELE(src_ip);
	627	if (target_ip != NULL) {
	628	IRELE(target_ip);
	629	}
	630	goto std_return;
	631	}