1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
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.
13 *
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
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_types.h"
21#include "xfs_bit.h"
22#include "xfs_log.h"
23#include "xfs_inum.h"
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
30#include "xfs_bmap_btree.h"
31#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h"
33#include "xfs_dir2_sf.h"
34#include "xfs_attr_sf.h"
35#include "xfs_dinode.h"
36#include "xfs_inode.h"
37#include "xfs_inode_item.h"
38#include "xfs_bmap.h"
39#include "xfs_btree.h"
40#include "xfs_ialloc.h"
41#include "xfs_itable.h"
42#include "xfs_dfrag.h"
43#include "xfs_error.h"
44#include "xfs_rw.h"
45
46/*
47 * Syssgi interface for swapext
48 */
49int
50xfs_swapext(
51	xfs_swapext_t	__user *sxu)
52{
53	xfs_swapext_t	*sxp;
54	xfs_inode_t     *ip=NULL, *tip=NULL;
55	xfs_mount_t     *mp;
56	struct file	*fp = NULL, *tfp = NULL;
57	bhv_vnode_t	*vp, *tvp;
58	int		error = 0;
59
60	sxp = kmem_alloc(sizeof(xfs_swapext_t), KM_MAYFAIL);
61	if (!sxp) {
62		error = XFS_ERROR(ENOMEM);
63		goto error0;
64	}
65
66	if (copy_from_user(sxp, sxu, sizeof(xfs_swapext_t))) {
67		error = XFS_ERROR(EFAULT);
68		goto error0;
69	}
70
71	/* Pull information for the target fd */
72	if (((fp = fget((int)sxp->sx_fdtarget)) == NULL) ||
73	    ((vp = vn_from_inode(fp->f_path.dentry->d_inode)) == NULL))  {
74		error = XFS_ERROR(EINVAL);
75		goto error0;
76	}
77
78	ip = xfs_vtoi(vp);
79	if (ip == NULL) {
80		error = XFS_ERROR(EBADF);
81		goto error0;
82	}
83
84	if (((tfp = fget((int)sxp->sx_fdtmp)) == NULL) ||
85	    ((tvp = vn_from_inode(tfp->f_path.dentry->d_inode)) == NULL)) {
86		error = XFS_ERROR(EINVAL);
87		goto error0;
88	}
89
90	tip = xfs_vtoi(tvp);
91	if (tip == NULL) {
92		error = XFS_ERROR(EBADF);
93		goto error0;
94	}
95
96	if (ip->i_mount != tip->i_mount) {
97		error =  XFS_ERROR(EINVAL);
98		goto error0;
99	}
100
101	if (ip->i_ino == tip->i_ino) {
102		error =  XFS_ERROR(EINVAL);
103		goto error0;
104	}
105
106	mp = ip->i_mount;
107
108	if (XFS_FORCED_SHUTDOWN(mp)) {
109		error =  XFS_ERROR(EIO);
110		goto error0;
111	}
112
113	error = XFS_SWAP_EXTENTS(mp, &ip->i_iocore, &tip->i_iocore, sxp);
114
115 error0:
116	if (fp != NULL)
117		fput(fp);
118	if (tfp != NULL)
119		fput(tfp);
120
121	if (sxp != NULL)
122		kmem_free(sxp, sizeof(xfs_swapext_t));
123
124	return error;
125}
126
127int
128xfs_swap_extents(
129	xfs_inode_t	*ip,
130	xfs_inode_t	*tip,
131	xfs_swapext_t	*sxp)
132{
133	xfs_mount_t	*mp;
134	xfs_inode_t	*ips[2];
135	xfs_trans_t	*tp;
136	xfs_bstat_t	*sbp = &sxp->sx_stat;
137	bhv_vnode_t	*vp, *tvp;
138	xfs_ifork_t	*tempifp, *ifp, *tifp;
139	int		ilf_fields, tilf_fields;
140	static uint	lock_flags = XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL;
141	int		error = 0;
142	int		aforkblks = 0;
143	int		taforkblks = 0;
144	__uint64_t	tmp;
145	char		locked = 0;
146
147	mp = ip->i_mount;
148
149	tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
150	if (!tempifp) {
151		error = XFS_ERROR(ENOMEM);
152		goto error0;
153	}
154
155	sbp = &sxp->sx_stat;
156	vp = XFS_ITOV(ip);
157	tvp = XFS_ITOV(tip);
158
159	/* Lock in i_ino order */
160	if (ip->i_ino < tip->i_ino) {
161		ips[0] = ip;
162		ips[1] = tip;
163	} else {
164		ips[0] = tip;
165		ips[1] = ip;
166	}
167
168	xfs_lock_inodes(ips, 2, 0, lock_flags);
169	locked = 1;
170
171	/* Check permissions */
172	error = xfs_iaccess(ip, S_IWUSR, NULL);
173	if (error)
174		goto error0;
175
176	error = xfs_iaccess(tip, S_IWUSR, NULL);
177	if (error)
178		goto error0;
179
180	/* Verify that both files have the same format */
181	if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
182		error = XFS_ERROR(EINVAL);
183		goto error0;
184	}
185
186	/* Verify both files are either real-time or non-realtime */
187	if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
188	    (tip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
189		error = XFS_ERROR(EINVAL);
190		goto error0;
191	}
192
193	/* Should never get a local format */
194	if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
195	    tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
196		error = XFS_ERROR(EINVAL);
197		goto error0;
198	}
199
200	if (VN_CACHED(tvp) != 0) {
201		xfs_inval_cached_trace(&tip->i_iocore, 0, -1, 0, -1);
202		error = bhv_vop_flushinval_pages(tvp, 0, -1, FI_REMAPF_LOCKED);
203		if (error)
204			goto error0;
205	}
206
207	/* Verify O_DIRECT for ftmp */
208	if (VN_CACHED(tvp) != 0) {
209		error = XFS_ERROR(EINVAL);
210		goto error0;
211	}
212
213	/* Verify all data are being swapped */
214	if (sxp->sx_offset != 0 ||
215	    sxp->sx_length != ip->i_d.di_size ||
216	    sxp->sx_length != tip->i_d.di_size) {
217		error = XFS_ERROR(EFAULT);
218		goto error0;
219	}
220
221	/*
222	 * If the target has extended attributes, the tmp file
223	 * must also in order to ensure the correct data fork
224	 * format.
225	 */
226	if ( XFS_IFORK_Q(ip) != XFS_IFORK_Q(tip) ) {
227		error = XFS_ERROR(EINVAL);
228		goto error0;
229	}
230
231	/*
232	 * Compare the current change & modify times with that
233	 * passed in.  If they differ, we abort this swap.
234	 * This is the mechanism used to ensure the calling
235	 * process that the file was not changed out from
236	 * under it.
237	 */
238	if ((sbp->bs_ctime.tv_sec != ip->i_d.di_ctime.t_sec) ||
239	    (sbp->bs_ctime.tv_nsec != ip->i_d.di_ctime.t_nsec) ||
240	    (sbp->bs_mtime.tv_sec != ip->i_d.di_mtime.t_sec) ||
241	    (sbp->bs_mtime.tv_nsec != ip->i_d.di_mtime.t_nsec)) {
242		error = XFS_ERROR(EBUSY);
243		goto error0;
244	}
245
246	/* We need to fail if the file is memory mapped.  Once we have tossed
247	 * all existing pages, the page fault will have no option
248	 * but to go to the filesystem for pages. By making the page fault call
249	 * vop_read (or write in the case of autogrow) they block on the iolock
250	 * until we have switched the extents.
251	 */
252	if (VN_MAPPED(vp)) {
253		error = XFS_ERROR(EBUSY);
254		goto error0;
255	}
256
257	xfs_iunlock(ip, XFS_ILOCK_EXCL);
258	xfs_iunlock(tip, XFS_ILOCK_EXCL);
259
260	/*
261	 * There is a race condition here since we gave up the
262	 * ilock.  However, the data fork will not change since
263	 * we have the iolock (locked for truncation too) so we
264	 * are safe.  We don't really care if non-io related
265	 * fields change.
266	 */
267
268	bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
269
270	tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
271	if ((error = xfs_trans_reserve(tp, 0,
272				     XFS_ICHANGE_LOG_RES(mp), 0,
273				     0, 0))) {
274		xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
275		xfs_iunlock(tip, XFS_IOLOCK_EXCL);
276		xfs_trans_cancel(tp, 0);
277		locked = 0;
278		goto error0;
279	}
280	xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
281
282	/*
283	 * Count the number of extended attribute blocks
284	 */
285	if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
286	     (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
287		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
288		if (error) {
289			xfs_trans_cancel(tp, 0);
290			goto error0;
291		}
292	}
293	if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
294	     (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
295		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
296			&taforkblks);
297		if (error) {
298			xfs_trans_cancel(tp, 0);
299			goto error0;
300		}
301	}
302
303	/*
304	 * Swap the data forks of the inodes
305	 */
306	ifp = &ip->i_df;
307	tifp = &tip->i_df;
308	*tempifp = *ifp;	/* struct copy */
309	*ifp = *tifp;		/* struct copy */
310	*tifp = *tempifp;	/* struct copy */
311
312	/*
313	 * Fix the on-disk inode values
314	 */
315	tmp = (__uint64_t)ip->i_d.di_nblocks;
316	ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
317	tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
318
319	tmp = (__uint64_t) ip->i_d.di_nextents;
320	ip->i_d.di_nextents = tip->i_d.di_nextents;
321	tip->i_d.di_nextents = tmp;
322
323	tmp = (__uint64_t) ip->i_d.di_format;
324	ip->i_d.di_format = tip->i_d.di_format;
325	tip->i_d.di_format = tmp;
326
327	ilf_fields = XFS_ILOG_CORE;
328
329	switch(ip->i_d.di_format) {
330	case XFS_DINODE_FMT_EXTENTS:
331		/* If the extents fit in the inode, fix the
332		 * pointer.  Otherwise it's already NULL or
333		 * pointing to the extent.
334		 */
335		if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
336			ifp->if_u1.if_extents =
337				ifp->if_u2.if_inline_ext;
338		}
339		ilf_fields |= XFS_ILOG_DEXT;
340		break;
341	case XFS_DINODE_FMT_BTREE:
342		ilf_fields |= XFS_ILOG_DBROOT;
343		break;
344	}
345
346	tilf_fields = XFS_ILOG_CORE;
347
348	switch(tip->i_d.di_format) {
349	case XFS_DINODE_FMT_EXTENTS:
350		/* If the extents fit in the inode, fix the
351		 * pointer.  Otherwise it's already NULL or
352		 * pointing to the extent.
353		 */
354		if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
355			tifp->if_u1.if_extents =
356				tifp->if_u2.if_inline_ext;
357		}
358		tilf_fields |= XFS_ILOG_DEXT;
359		break;
360	case XFS_DINODE_FMT_BTREE:
361		tilf_fields |= XFS_ILOG_DBROOT;
362		break;
363	}
364
365	/*
366	 * Increment vnode ref counts since xfs_trans_commit &
367	 * xfs_trans_cancel will both unlock the inodes and
368	 * decrement the associated ref counts.
369	 */
370	VN_HOLD(vp);
371	VN_HOLD(tvp);
372
373	xfs_trans_ijoin(tp, ip, lock_flags);
374	xfs_trans_ijoin(tp, tip, lock_flags);
375
376	xfs_trans_log_inode(tp, ip,  ilf_fields);
377	xfs_trans_log_inode(tp, tip, tilf_fields);
378
379	/*
380	 * If this is a synchronous mount, make sure that the
381	 * transaction goes to disk before returning to the user.
382	 */
383	if (mp->m_flags & XFS_MOUNT_WSYNC) {
384		xfs_trans_set_sync(tp);
385	}
386
387	error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT);
388	locked = 0;
389
390 error0:
391	if (locked) {
392		xfs_iunlock(ip,  lock_flags);
393		xfs_iunlock(tip, lock_flags);
394	}
395	if (tempifp != NULL)
396		kmem_free(tempifp, sizeof(xfs_ifork_t));
397	return error;
398}
399