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 #include "xfs_vnodeops.h"
46
47 /*
48 * Syssgi interface for swapext
49 */
50 int
xfs_swapext(xfs_swapext_t * sxp)51 xfs_swapext(
52 xfs_swapext_t *sxp)
53 {
54 xfs_inode_t *ip, *tip;
55 struct file *file, *target_file;
56 int error = 0;
57
58 /* Pull information for the target fd */
59 file = fget((int)sxp->sx_fdtarget);
60 if (!file) {
61 error = XFS_ERROR(EINVAL);
62 goto out;
63 }
64
65 if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND)) {
66 error = XFS_ERROR(EBADF);
67 goto out_put_file;
68 }
69
70 target_file = fget((int)sxp->sx_fdtmp);
71 if (!target_file) {
72 error = XFS_ERROR(EINVAL);
73 goto out_put_file;
74 }
75
76 if (!(target_file->f_mode & FMODE_WRITE) ||
77 (target_file->f_flags & O_APPEND)) {
78 error = XFS_ERROR(EBADF);
79 goto out_put_target_file;
80 }
81
82 ip = XFS_I(file->f_path.dentry->d_inode);
83 tip = XFS_I(target_file->f_path.dentry->d_inode);
84
85 if (ip->i_mount != tip->i_mount) {
86 error = XFS_ERROR(EINVAL);
87 goto out_put_target_file;
88 }
89
90 if (ip->i_ino == tip->i_ino) {
91 error = XFS_ERROR(EINVAL);
92 goto out_put_target_file;
93 }
94
95 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
96 error = XFS_ERROR(EIO);
97 goto out_put_target_file;
98 }
99
100 error = xfs_swap_extents(ip, tip, sxp);
101
102 out_put_target_file:
103 fput(target_file);
104 out_put_file:
105 fput(file);
106 out:
107 return error;
108 }
109
110 int
xfs_swap_extents(xfs_inode_t * ip,xfs_inode_t * tip,xfs_swapext_t * sxp)111 xfs_swap_extents(
112 xfs_inode_t *ip,
113 xfs_inode_t *tip,
114 xfs_swapext_t *sxp)
115 {
116 xfs_mount_t *mp;
117 xfs_trans_t *tp;
118 xfs_bstat_t *sbp = &sxp->sx_stat;
119 xfs_ifork_t *tempifp, *ifp, *tifp;
120 int ilf_fields, tilf_fields;
121 static uint lock_flags = XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL;
122 int error = 0;
123 int aforkblks = 0;
124 int taforkblks = 0;
125 __uint64_t tmp;
126 char locked = 0;
127
128 mp = ip->i_mount;
129
130 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
131 if (!tempifp) {
132 error = XFS_ERROR(ENOMEM);
133 goto error0;
134 }
135
136 sbp = &sxp->sx_stat;
137
138 /*
139 * we have to do two separate lock calls here to keep lockdep
140 * happy. If we try to get all the locks in one call, lock will
141 * report false positives when we drop the ILOCK and regain them
142 * below.
143 */
144 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
145 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
146 locked = 1;
147
148 /* Verify that both files have the same format */
149 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
150 error = XFS_ERROR(EINVAL);
151 goto error0;
152 }
153
154 /* Verify both files are either real-time or non-realtime */
155 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
156 error = XFS_ERROR(EINVAL);
157 goto error0;
158 }
159
160 /* Should never get a local format */
161 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
162 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
163 error = XFS_ERROR(EINVAL);
164 goto error0;
165 }
166
167 if (VN_CACHED(VFS_I(tip)) != 0) {
168 xfs_inval_cached_trace(tip, 0, -1, 0, -1);
169 error = xfs_flushinval_pages(tip, 0, -1,
170 FI_REMAPF_LOCKED);
171 if (error)
172 goto error0;
173 }
174
175 /* Verify O_DIRECT for ftmp */
176 if (VN_CACHED(VFS_I(tip)) != 0) {
177 error = XFS_ERROR(EINVAL);
178 goto error0;
179 }
180
181 /* Verify all data are being swapped */
182 if (sxp->sx_offset != 0 ||
183 sxp->sx_length != ip->i_d.di_size ||
184 sxp->sx_length != tip->i_d.di_size) {
185 error = XFS_ERROR(EFAULT);
186 goto error0;
187 }
188
189 /*
190 * If the target has extended attributes, the tmp file
191 * must also in order to ensure the correct data fork
192 * format.
193 */
194 if ( XFS_IFORK_Q(ip) != XFS_IFORK_Q(tip) ) {
195 error = XFS_ERROR(EINVAL);
196 goto error0;
197 }
198
199 /*
200 * Compare the current change & modify times with that
201 * passed in. If they differ, we abort this swap.
202 * This is the mechanism used to ensure the calling
203 * process that the file was not changed out from
204 * under it.
205 */
206 if ((sbp->bs_ctime.tv_sec != ip->i_d.di_ctime.t_sec) ||
207 (sbp->bs_ctime.tv_nsec != ip->i_d.di_ctime.t_nsec) ||
208 (sbp->bs_mtime.tv_sec != ip->i_d.di_mtime.t_sec) ||
209 (sbp->bs_mtime.tv_nsec != ip->i_d.di_mtime.t_nsec)) {
210 error = XFS_ERROR(EBUSY);
211 goto error0;
212 }
213
214 /* We need to fail if the file is memory mapped. Once we have tossed
215 * all existing pages, the page fault will have no option
216 * but to go to the filesystem for pages. By making the page fault call
217 * vop_read (or write in the case of autogrow) they block on the iolock
218 * until we have switched the extents.
219 */
220 if (VN_MAPPED(VFS_I(ip))) {
221 error = XFS_ERROR(EBUSY);
222 goto error0;
223 }
224
225 xfs_iunlock(ip, XFS_ILOCK_EXCL);
226 xfs_iunlock(tip, XFS_ILOCK_EXCL);
227
228 /*
229 * There is a race condition here since we gave up the
230 * ilock. However, the data fork will not change since
231 * we have the iolock (locked for truncation too) so we
232 * are safe. We don't really care if non-io related
233 * fields change.
234 */
235
236 xfs_tosspages(ip, 0, -1, FI_REMAPF);
237
238 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
239 if ((error = xfs_trans_reserve(tp, 0,
240 XFS_ICHANGE_LOG_RES(mp), 0,
241 0, 0))) {
242 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
243 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
244 xfs_trans_cancel(tp, 0);
245 locked = 0;
246 goto error0;
247 }
248 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
249
250 /*
251 * Count the number of extended attribute blocks
252 */
253 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
254 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
255 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
256 if (error) {
257 xfs_trans_cancel(tp, 0);
258 goto error0;
259 }
260 }
261 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
262 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
263 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
264 &taforkblks);
265 if (error) {
266 xfs_trans_cancel(tp, 0);
267 goto error0;
268 }
269 }
270
271 /*
272 * Swap the data forks of the inodes
273 */
274 ifp = &ip->i_df;
275 tifp = &tip->i_df;
276 *tempifp = *ifp; /* struct copy */
277 *ifp = *tifp; /* struct copy */
278 *tifp = *tempifp; /* struct copy */
279
280 /*
281 * Fix the on-disk inode values
282 */
283 tmp = (__uint64_t)ip->i_d.di_nblocks;
284 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
285 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
286
287 tmp = (__uint64_t) ip->i_d.di_nextents;
288 ip->i_d.di_nextents = tip->i_d.di_nextents;
289 tip->i_d.di_nextents = tmp;
290
291 tmp = (__uint64_t) ip->i_d.di_format;
292 ip->i_d.di_format = tip->i_d.di_format;
293 tip->i_d.di_format = tmp;
294
295 ilf_fields = XFS_ILOG_CORE;
296
297 switch(ip->i_d.di_format) {
298 case XFS_DINODE_FMT_EXTENTS:
299 /* If the extents fit in the inode, fix the
300 * pointer. Otherwise it's already NULL or
301 * pointing to the extent.
302 */
303 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
304 ifp->if_u1.if_extents =
305 ifp->if_u2.if_inline_ext;
306 }
307 ilf_fields |= XFS_ILOG_DEXT;
308 break;
309 case XFS_DINODE_FMT_BTREE:
310 ilf_fields |= XFS_ILOG_DBROOT;
311 break;
312 }
313
314 tilf_fields = XFS_ILOG_CORE;
315
316 switch(tip->i_d.di_format) {
317 case XFS_DINODE_FMT_EXTENTS:
318 /* If the extents fit in the inode, fix the
319 * pointer. Otherwise it's already NULL or
320 * pointing to the extent.
321 */
322 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
323 tifp->if_u1.if_extents =
324 tifp->if_u2.if_inline_ext;
325 }
326 tilf_fields |= XFS_ILOG_DEXT;
327 break;
328 case XFS_DINODE_FMT_BTREE:
329 tilf_fields |= XFS_ILOG_DBROOT;
330 break;
331 }
332
333
334 IHOLD(ip);
335 xfs_trans_ijoin(tp, ip, lock_flags);
336
337 IHOLD(tip);
338 xfs_trans_ijoin(tp, tip, lock_flags);
339
340 xfs_trans_log_inode(tp, ip, ilf_fields);
341 xfs_trans_log_inode(tp, tip, tilf_fields);
342
343 /*
344 * If this is a synchronous mount, make sure that the
345 * transaction goes to disk before returning to the user.
346 */
347 if (mp->m_flags & XFS_MOUNT_WSYNC) {
348 xfs_trans_set_sync(tp);
349 }
350
351 error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT);
352 locked = 0;
353
354 error0:
355 if (locked) {
356 xfs_iunlock(ip, lock_flags);
357 xfs_iunlock(tip, lock_flags);
358 }
359 if (tempifp != NULL)
360 kmem_free(tempifp);
361 return error;
362 }
363