blob: 681be5c93af500f63fc2cd4e744682023b6c7497 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (c) 2000-2002 Silicon Graphics, Inc. All Rights Reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11 *
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
32
33#include "xfs.h"
34#include "xfs_macros.h"
35#include "xfs_types.h"
36#include "xfs_inum.h"
37#include "xfs_log.h"
38#include "xfs_trans.h"
39#include "xfs_sb.h"
40#include "xfs_dir.h"
41#include "xfs_dir2.h"
42#include "xfs_dmapi.h"
43#include "xfs_mount.h"
44#include "xfs_ag.h"
45#include "xfs_alloc_btree.h"
46#include "xfs_bmap_btree.h"
47#include "xfs_ialloc_btree.h"
48#include "xfs_btree.h"
49#include "xfs_attr_sf.h"
50#include "xfs_dir_sf.h"
51#include "xfs_dir2_sf.h"
52#include "xfs_dinode.h"
53#include "xfs_inode_item.h"
54#include "xfs_inode.h"
55#include "xfs_bmap.h"
56#include "xfs_ialloc.h"
57#include "xfs_itable.h"
58#include "xfs_dfrag.h"
59#include "xfs_error.h"
60#include "xfs_mac.h"
61#include "xfs_rw.h"
62
63/*
64 * Syssgi interface for swapext
65 */
66int
67xfs_swapext(
68 xfs_swapext_t __user *sxp)
69{
70 xfs_swapext_t sx;
71 xfs_inode_t *ip=NULL, *tip=NULL, *ips[2];
72 xfs_trans_t *tp;
73 xfs_mount_t *mp;
74 xfs_bstat_t *sbp;
75 struct file *fp = NULL, *tfp = NULL;
76 vnode_t *vp, *tvp;
77 bhv_desc_t *bdp, *tbdp;
78 vn_bhv_head_t *bhp, *tbhp;
79 uint lock_flags=0;
80 int ilf_fields, tilf_fields;
81 int error = 0;
82 xfs_ifork_t tempif, *ifp, *tifp;
83 __uint64_t tmp;
84 int aforkblks = 0;
85 int taforkblks = 0;
86 int locked = 0;
87
88 if (copy_from_user(&sx, sxp, sizeof(sx)))
89 return XFS_ERROR(EFAULT);
90
91 /* Pull information for the target fd */
92 if (((fp = fget((int)sx.sx_fdtarget)) == NULL) ||
93 ((vp = LINVFS_GET_VP(fp->f_dentry->d_inode)) == NULL)) {
94 error = XFS_ERROR(EINVAL);
95 goto error0;
96 }
97
98 bhp = VN_BHV_HEAD(vp);
99 bdp = vn_bhv_lookup(bhp, &xfs_vnodeops);
100 if (bdp == NULL) {
101 error = XFS_ERROR(EBADF);
102 goto error0;
103 } else {
104 ip = XFS_BHVTOI(bdp);
105 }
106
107 if (((tfp = fget((int)sx.sx_fdtmp)) == NULL) ||
108 ((tvp = LINVFS_GET_VP(tfp->f_dentry->d_inode)) == NULL)) {
109 error = XFS_ERROR(EINVAL);
110 goto error0;
111 }
112
113 tbhp = VN_BHV_HEAD(tvp);
114 tbdp = vn_bhv_lookup(tbhp, &xfs_vnodeops);
115 if (tbdp == NULL) {
116 error = XFS_ERROR(EBADF);
117 goto error0;
118 } else {
119 tip = XFS_BHVTOI(tbdp);
120 }
121
122 if (ip->i_mount != tip->i_mount) {
123 error = XFS_ERROR(EINVAL);
124 goto error0;
125 }
126
127 if (ip->i_ino == tip->i_ino) {
128 error = XFS_ERROR(EINVAL);
129 goto error0;
130 }
131
132 mp = ip->i_mount;
133
134 sbp = &sx.sx_stat;
135
136 if (XFS_FORCED_SHUTDOWN(mp)) {
137 error = XFS_ERROR(EIO);
138 goto error0;
139 }
140
141 locked = 1;
142
143 /* Lock in i_ino order */
144 if (ip->i_ino < tip->i_ino) {
145 ips[0] = ip;
146 ips[1] = tip;
147 } else {
148 ips[0] = tip;
149 ips[1] = ip;
150 }
151 lock_flags = XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL;
152 xfs_lock_inodes(ips, 2, 0, lock_flags);
153
154 /* Check permissions */
155 error = xfs_iaccess(ip, S_IWUSR, NULL);
156 if (error)
157 goto error0;
158
159 error = xfs_iaccess(tip, S_IWUSR, NULL);
160 if (error)
161 goto error0;
162
163 /* Verify that both files have the same format */
164 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
165 error = XFS_ERROR(EINVAL);
166 goto error0;
167 }
168
169 /* Verify both files are either real-time or non-realtime */
170 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
171 (tip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
172 error = XFS_ERROR(EINVAL);
173 goto error0;
174 }
175
176 /* Should never get a local format */
177 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
178 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
179 error = XFS_ERROR(EINVAL);
180 goto error0;
181 }
182
Christoph Hellwigbd5a8762005-06-21 15:47:39 +1000183 if (VN_CACHED(tvp) != 0) {
184 xfs_inval_cached_trace(&tip->i_iocore, 0, -1, 0, -1);
185 VOP_FLUSHINVAL_PAGES(tvp, 0, -1, FI_REMAPF_LOCKED);
186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 /* Verify O_DIRECT for ftmp */
189 if (VN_CACHED(tvp) != 0) {
190 error = XFS_ERROR(EINVAL);
191 goto error0;
192 }
193
194 /* Verify all data are being swapped */
195 if (sx.sx_offset != 0 ||
196 sx.sx_length != ip->i_d.di_size ||
197 sx.sx_length != tip->i_d.di_size) {
198 error = XFS_ERROR(EFAULT);
199 goto error0;
200 }
201
202 /*
203 * If the target has extended attributes, the tmp file
204 * must also in order to ensure the correct data fork
205 * format.
206 */
207 if ( XFS_IFORK_Q(ip) != XFS_IFORK_Q(tip) ) {
208 error = XFS_ERROR(EINVAL);
209 goto error0;
210 }
211
212 /*
213 * Compare the current change & modify times with that
214 * passed in. If they differ, we abort this swap.
215 * This is the mechanism used to ensure the calling
216 * process that the file was not changed out from
217 * under it.
218 */
219 if ((sbp->bs_ctime.tv_sec != ip->i_d.di_ctime.t_sec) ||
220 (sbp->bs_ctime.tv_nsec != ip->i_d.di_ctime.t_nsec) ||
221 (sbp->bs_mtime.tv_sec != ip->i_d.di_mtime.t_sec) ||
222 (sbp->bs_mtime.tv_nsec != ip->i_d.di_mtime.t_nsec)) {
223 error = XFS_ERROR(EBUSY);
224 goto error0;
225 }
226
227 /* We need to fail if the file is memory mapped. Once we have tossed
228 * all existing pages, the page fault will have no option
229 * but to go to the filesystem for pages. By making the page fault call
230 * VOP_READ (or write in the case of autogrow) they block on the iolock
231 * until we have switched the extents.
232 */
233 if (VN_MAPPED(vp)) {
234 error = XFS_ERROR(EBUSY);
235 goto error0;
236 }
237
238 xfs_iunlock(ip, XFS_ILOCK_EXCL);
239 xfs_iunlock(tip, XFS_ILOCK_EXCL);
240
241 /*
242 * There is a race condition here since we gave up the
243 * ilock. However, the data fork will not change since
244 * we have the iolock (locked for truncation too) so we
245 * are safe. We don't really care if non-io related
246 * fields change.
247 */
248
249 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
250
251 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
252 if ((error = xfs_trans_reserve(tp, 0,
253 XFS_ICHANGE_LOG_RES(mp), 0,
254 0, 0))) {
255 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
256 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
257 xfs_trans_cancel(tp, 0);
258 return error;
259 }
260 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
261
262 /*
263 * Count the number of extended attribute blocks
264 */
265 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
266 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
267 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
268 if (error) {
269 xfs_iunlock(ip, lock_flags);
270 xfs_iunlock(tip, lock_flags);
271 xfs_trans_cancel(tp, 0);
272 return error;
273 }
274 }
275 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
276 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
277 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
278 &taforkblks);
279 if (error) {
280 xfs_iunlock(ip, lock_flags);
281 xfs_iunlock(tip, lock_flags);
282 xfs_trans_cancel(tp, 0);
283 return error;
284 }
285 }
286
287 /*
288 * Swap the data forks of the inodes
289 */
290 ifp = &ip->i_df;
291 tifp = &tip->i_df;
292 tempif = *ifp; /* struct copy */
293 *ifp = *tifp; /* struct copy */
294 *tifp = tempif; /* struct copy */
295
296 /*
297 * Fix the on-disk inode values
298 */
299 tmp = (__uint64_t)ip->i_d.di_nblocks;
300 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
301 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
302
303 tmp = (__uint64_t) ip->i_d.di_nextents;
304 ip->i_d.di_nextents = tip->i_d.di_nextents;
305 tip->i_d.di_nextents = tmp;
306
307 tmp = (__uint64_t) ip->i_d.di_format;
308 ip->i_d.di_format = tip->i_d.di_format;
309 tip->i_d.di_format = tmp;
310
311 ilf_fields = XFS_ILOG_CORE;
312
313 switch(ip->i_d.di_format) {
314 case XFS_DINODE_FMT_EXTENTS:
315 /* If the extents fit in the inode, fix the
316 * pointer. Otherwise it's already NULL or
317 * pointing to the extent.
318 */
319 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
320 ifp->if_u1.if_extents =
321 ifp->if_u2.if_inline_ext;
322 }
323 ilf_fields |= XFS_ILOG_DEXT;
324 break;
325 case XFS_DINODE_FMT_BTREE:
326 ilf_fields |= XFS_ILOG_DBROOT;
327 break;
328 }
329
330 tilf_fields = XFS_ILOG_CORE;
331
332 switch(tip->i_d.di_format) {
333 case XFS_DINODE_FMT_EXTENTS:
334 /* If the extents fit in the inode, fix the
335 * pointer. Otherwise it's already NULL or
336 * pointing to the extent.
337 */
338 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
339 tifp->if_u1.if_extents =
340 tifp->if_u2.if_inline_ext;
341 }
342 tilf_fields |= XFS_ILOG_DEXT;
343 break;
344 case XFS_DINODE_FMT_BTREE:
345 tilf_fields |= XFS_ILOG_DBROOT;
346 break;
347 }
348
349 /*
350 * Increment vnode ref counts since xfs_trans_commit &
351 * xfs_trans_cancel will both unlock the inodes and
352 * decrement the associated ref counts.
353 */
354 VN_HOLD(vp);
355 VN_HOLD(tvp);
356
357 xfs_trans_ijoin(tp, ip, lock_flags);
358 xfs_trans_ijoin(tp, tip, lock_flags);
359
360 xfs_trans_log_inode(tp, ip, ilf_fields);
361 xfs_trans_log_inode(tp, tip, tilf_fields);
362
363 /*
364 * If this is a synchronous mount, make sure that the
365 * transaction goes to disk before returning to the user.
366 */
367 if (mp->m_flags & XFS_MOUNT_WSYNC) {
368 xfs_trans_set_sync(tp);
369 }
370
371 error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT, NULL);
372
373 fput(fp);
374 fput(tfp);
375
376 return error;
377
378 error0:
379 if (locked) {
380 xfs_iunlock(ip, lock_flags);
381 xfs_iunlock(tip, lock_flags);
382 }
383
384 if (fp != NULL) fput(fp);
385 if (tfp != NULL) fput(tfp);
386
387 return error;
388}