diff options
author | Akira Fujita <a-fujita@rs.jp.nec.com> | 2009-06-17 19:24:03 -0400 |
---|---|---|
committer | Theodore Ts'o <tytso@mit.edu> | 2009-06-17 19:24:03 -0400 |
commit | 748de6736c1e482e111f9d1b5a5d5b1787600cad (patch) | |
tree | 9e5bd6085cd0eea0453c2a8bbce1422f291f059f /fs/ext4/move_extent.c | |
parent | 8b0f9e8f78bd0a65fa001bf18f2c47eef2893a10 (diff) |
ext4: online defrag -- Add EXT4_IOC_MOVE_EXT ioctl
The EXT4_IOC_MOVE_EXT exchanges the blocks between orig_fd and donor_fd,
and then write the file data of orig_fd to donor_fd.
ext4_mext_move_extent() is the main fucntion of ext4 online defrag,
and this patch includes all functions related to ext4 online defrag.
Signed-off-by: Akira Fujita <a-fujita@rs.jp.nec.com>
Signed-off-by: Takashi Sato <t-sato@yk.jp.nec.com>
Signed-off-by: Kazuya Mio <k-mio@sx.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Diffstat (limited to 'fs/ext4/move_extent.c')
-rw-r--r-- | fs/ext4/move_extent.c | 1320 |
1 files changed, 1320 insertions, 0 deletions
diff --git a/fs/ext4/move_extent.c b/fs/ext4/move_extent.c new file mode 100644 index 000000000000..bbf2dd9404dc --- /dev/null +++ b/fs/ext4/move_extent.c | |||
@@ -0,0 +1,1320 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd. | ||
3 | * Written by Takashi Sato <t-sato@yk.jp.nec.com> | ||
4 | * Akira Fujita <a-fujita@rs.jp.nec.com> | ||
5 | * | ||
6 | * This program is free software; you can redistribute it and/or modify it | ||
7 | * under the terms of version 2.1 of the GNU Lesser General Public License | ||
8 | * as published by the Free Software Foundation. | ||
9 | * | ||
10 | * This program is distributed in the hope that it will be useful, | ||
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
13 | * GNU General Public License for more details. | ||
14 | */ | ||
15 | |||
16 | #include <linux/fs.h> | ||
17 | #include <linux/quotaops.h> | ||
18 | #include "ext4_jbd2.h" | ||
19 | #include "ext4_extents.h" | ||
20 | #include "ext4.h" | ||
21 | |||
22 | #define get_ext_path(path, inode, block, ret) \ | ||
23 | do { \ | ||
24 | path = ext4_ext_find_extent(inode, block, path); \ | ||
25 | if (IS_ERR(path)) { \ | ||
26 | ret = PTR_ERR(path); \ | ||
27 | path = NULL; \ | ||
28 | } \ | ||
29 | } while (0) | ||
30 | |||
31 | /** | ||
32 | * copy_extent_status - Copy the extent's initialization status | ||
33 | * | ||
34 | * @src: an extent for getting initialize status | ||
35 | * @dest: an extent to be set the status | ||
36 | */ | ||
37 | static void | ||
38 | copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest) | ||
39 | { | ||
40 | if (ext4_ext_is_uninitialized(src)) | ||
41 | ext4_ext_mark_uninitialized(dest); | ||
42 | else | ||
43 | dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest)); | ||
44 | } | ||
45 | |||
46 | /** | ||
47 | * mext_next_extent - Search for the next extent and set it to "extent" | ||
48 | * | ||
49 | * @inode: inode which is searched | ||
50 | * @path: this will obtain data for the next extent | ||
51 | * @extent: pointer to the next extent we have just gotten | ||
52 | * | ||
53 | * Search the next extent in the array of ext4_ext_path structure (@path) | ||
54 | * and set it to ext4_extent structure (@extent). In addition, the member of | ||
55 | * @path (->p_ext) also points the next extent. Return 0 on success, 1 if | ||
56 | * ext4_ext_path structure refers to the last extent, or a negative error | ||
57 | * value on failure. | ||
58 | */ | ||
59 | static int | ||
60 | mext_next_extent(struct inode *inode, struct ext4_ext_path *path, | ||
61 | struct ext4_extent **extent) | ||
62 | { | ||
63 | int ppos, leaf_ppos = path->p_depth; | ||
64 | |||
65 | ppos = leaf_ppos; | ||
66 | if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) { | ||
67 | /* leaf block */ | ||
68 | *extent = ++path[ppos].p_ext; | ||
69 | return 0; | ||
70 | } | ||
71 | |||
72 | while (--ppos >= 0) { | ||
73 | if (EXT_LAST_INDEX(path[ppos].p_hdr) > | ||
74 | path[ppos].p_idx) { | ||
75 | int cur_ppos = ppos; | ||
76 | |||
77 | /* index block */ | ||
78 | path[ppos].p_idx++; | ||
79 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); | ||
80 | if (path[ppos+1].p_bh) | ||
81 | brelse(path[ppos+1].p_bh); | ||
82 | path[ppos+1].p_bh = | ||
83 | sb_bread(inode->i_sb, path[ppos].p_block); | ||
84 | if (!path[ppos+1].p_bh) | ||
85 | return -EIO; | ||
86 | path[ppos+1].p_hdr = | ||
87 | ext_block_hdr(path[ppos+1].p_bh); | ||
88 | |||
89 | /* Halfway index block */ | ||
90 | while (++cur_ppos < leaf_ppos) { | ||
91 | path[cur_ppos].p_idx = | ||
92 | EXT_FIRST_INDEX(path[cur_ppos].p_hdr); | ||
93 | path[cur_ppos].p_block = | ||
94 | idx_pblock(path[cur_ppos].p_idx); | ||
95 | if (path[cur_ppos+1].p_bh) | ||
96 | brelse(path[cur_ppos+1].p_bh); | ||
97 | path[cur_ppos+1].p_bh = sb_bread(inode->i_sb, | ||
98 | path[cur_ppos].p_block); | ||
99 | if (!path[cur_ppos+1].p_bh) | ||
100 | return -EIO; | ||
101 | path[cur_ppos+1].p_hdr = | ||
102 | ext_block_hdr(path[cur_ppos+1].p_bh); | ||
103 | } | ||
104 | |||
105 | /* leaf block */ | ||
106 | path[leaf_ppos].p_ext = *extent = | ||
107 | EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr); | ||
108 | return 0; | ||
109 | } | ||
110 | } | ||
111 | /* We found the last extent */ | ||
112 | return 1; | ||
113 | } | ||
114 | |||
115 | /** | ||
116 | * mext_double_down_read - Acquire two inodes' read semaphore | ||
117 | * | ||
118 | * @orig_inode: original inode structure | ||
119 | * @donor_inode: donor inode structure | ||
120 | * Acquire read semaphore of the two inodes (orig and donor) by i_ino order. | ||
121 | */ | ||
122 | static void | ||
123 | mext_double_down_read(struct inode *orig_inode, struct inode *donor_inode) | ||
124 | { | ||
125 | struct inode *first = orig_inode, *second = donor_inode; | ||
126 | |||
127 | BUG_ON(orig_inode == NULL || donor_inode == NULL); | ||
128 | |||
129 | /* | ||
130 | * Use the inode number to provide the stable locking order instead | ||
131 | * of its address, because the C language doesn't guarantee you can | ||
132 | * compare pointers that don't come from the same array. | ||
133 | */ | ||
134 | if (donor_inode->i_ino < orig_inode->i_ino) { | ||
135 | first = donor_inode; | ||
136 | second = orig_inode; | ||
137 | } | ||
138 | |||
139 | down_read(&EXT4_I(first)->i_data_sem); | ||
140 | down_read(&EXT4_I(second)->i_data_sem); | ||
141 | } | ||
142 | |||
143 | /** | ||
144 | * mext_double_down_write - Acquire two inodes' write semaphore | ||
145 | * | ||
146 | * @orig_inode: original inode structure | ||
147 | * @donor_inode: donor inode structure | ||
148 | * Acquire write semaphore of the two inodes (orig and donor) by i_ino order. | ||
149 | */ | ||
150 | static void | ||
151 | mext_double_down_write(struct inode *orig_inode, struct inode *donor_inode) | ||
152 | { | ||
153 | struct inode *first = orig_inode, *second = donor_inode; | ||
154 | |||
155 | BUG_ON(orig_inode == NULL || donor_inode == NULL); | ||
156 | |||
157 | /* | ||
158 | * Use the inode number to provide the stable locking order instead | ||
159 | * of its address, because the C language doesn't guarantee you can | ||
160 | * compare pointers that don't come from the same array. | ||
161 | */ | ||
162 | if (donor_inode->i_ino < orig_inode->i_ino) { | ||
163 | first = donor_inode; | ||
164 | second = orig_inode; | ||
165 | } | ||
166 | |||
167 | down_write(&EXT4_I(first)->i_data_sem); | ||
168 | down_write(&EXT4_I(second)->i_data_sem); | ||
169 | } | ||
170 | |||
171 | /** | ||
172 | * mext_double_up_read - Release two inodes' read semaphore | ||
173 | * | ||
174 | * @orig_inode: original inode structure to be released its lock first | ||
175 | * @donor_inode: donor inode structure to be released its lock second | ||
176 | * Release read semaphore of two inodes (orig and donor). | ||
177 | */ | ||
178 | static void | ||
179 | mext_double_up_read(struct inode *orig_inode, struct inode *donor_inode) | ||
180 | { | ||
181 | BUG_ON(orig_inode == NULL || donor_inode == NULL); | ||
182 | |||
183 | up_read(&EXT4_I(orig_inode)->i_data_sem); | ||
184 | up_read(&EXT4_I(donor_inode)->i_data_sem); | ||
185 | } | ||
186 | |||
187 | /** | ||
188 | * mext_double_up_write - Release two inodes' write semaphore | ||
189 | * | ||
190 | * @orig_inode: original inode structure to be released its lock first | ||
191 | * @donor_inode: donor inode structure to be released its lock second | ||
192 | * Release write semaphore of two inodes (orig and donor). | ||
193 | */ | ||
194 | static void | ||
195 | mext_double_up_write(struct inode *orig_inode, struct inode *donor_inode) | ||
196 | { | ||
197 | BUG_ON(orig_inode == NULL || donor_inode == NULL); | ||
198 | |||
199 | up_write(&EXT4_I(orig_inode)->i_data_sem); | ||
200 | up_write(&EXT4_I(donor_inode)->i_data_sem); | ||
201 | } | ||
202 | |||
203 | /** | ||
204 | * mext_insert_across_blocks - Insert extents across leaf block | ||
205 | * | ||
206 | * @handle: journal handle | ||
207 | * @orig_inode: original inode | ||
208 | * @o_start: first original extent to be changed | ||
209 | * @o_end: last original extent to be changed | ||
210 | * @start_ext: first new extent to be inserted | ||
211 | * @new_ext: middle of new extent to be inserted | ||
212 | * @end_ext: last new extent to be inserted | ||
213 | * | ||
214 | * Allocate a new leaf block and insert extents into it. Return 0 on success, | ||
215 | * or a negative error value on failure. | ||
216 | */ | ||
217 | static int | ||
218 | mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode, | ||
219 | struct ext4_extent *o_start, struct ext4_extent *o_end, | ||
220 | struct ext4_extent *start_ext, struct ext4_extent *new_ext, | ||
221 | struct ext4_extent *end_ext) | ||
222 | { | ||
223 | struct ext4_ext_path *orig_path = NULL; | ||
224 | ext4_lblk_t eblock = 0; | ||
225 | int new_flag = 0; | ||
226 | int end_flag = 0; | ||
227 | int err = 0; | ||
228 | |||
229 | if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) { | ||
230 | if (o_start == o_end) { | ||
231 | |||
232 | /* start_ext new_ext end_ext | ||
233 | * donor |---------|-----------|--------| | ||
234 | * orig |------------------------------| | ||
235 | */ | ||
236 | end_flag = 1; | ||
237 | } else { | ||
238 | |||
239 | /* start_ext new_ext end_ext | ||
240 | * donor |---------|----------|---------| | ||
241 | * orig |---------------|--------------| | ||
242 | */ | ||
243 | o_end->ee_block = end_ext->ee_block; | ||
244 | o_end->ee_len = end_ext->ee_len; | ||
245 | ext4_ext_store_pblock(o_end, ext_pblock(end_ext)); | ||
246 | } | ||
247 | |||
248 | o_start->ee_len = start_ext->ee_len; | ||
249 | new_flag = 1; | ||
250 | |||
251 | } else if (start_ext->ee_len && new_ext->ee_len && | ||
252 | !end_ext->ee_len && o_start == o_end) { | ||
253 | |||
254 | /* start_ext new_ext | ||
255 | * donor |--------------|---------------| | ||
256 | * orig |------------------------------| | ||
257 | */ | ||
258 | o_start->ee_len = start_ext->ee_len; | ||
259 | new_flag = 1; | ||
260 | |||
261 | } else if (!start_ext->ee_len && new_ext->ee_len && | ||
262 | end_ext->ee_len && o_start == o_end) { | ||
263 | |||
264 | /* new_ext end_ext | ||
265 | * donor |--------------|---------------| | ||
266 | * orig |------------------------------| | ||
267 | */ | ||
268 | o_end->ee_block = end_ext->ee_block; | ||
269 | o_end->ee_len = end_ext->ee_len; | ||
270 | ext4_ext_store_pblock(o_end, ext_pblock(end_ext)); | ||
271 | |||
272 | /* | ||
273 | * Set 0 to the extent block if new_ext was | ||
274 | * the first block. | ||
275 | */ | ||
276 | if (new_ext->ee_block) | ||
277 | eblock = le32_to_cpu(new_ext->ee_block); | ||
278 | |||
279 | new_flag = 1; | ||
280 | } else { | ||
281 | ext4_debug("ext4 move extent: Unexpected insert case\n"); | ||
282 | return -EIO; | ||
283 | } | ||
284 | |||
285 | if (new_flag) { | ||
286 | get_ext_path(orig_path, orig_inode, eblock, err); | ||
287 | if (orig_path == NULL) | ||
288 | goto out; | ||
289 | |||
290 | if (ext4_ext_insert_extent(handle, orig_inode, | ||
291 | orig_path, new_ext)) | ||
292 | goto out; | ||
293 | } | ||
294 | |||
295 | if (end_flag) { | ||
296 | get_ext_path(orig_path, orig_inode, | ||
297 | le32_to_cpu(end_ext->ee_block) - 1, err); | ||
298 | if (orig_path == NULL) | ||
299 | goto out; | ||
300 | |||
301 | if (ext4_ext_insert_extent(handle, orig_inode, | ||
302 | orig_path, end_ext)) | ||
303 | goto out; | ||
304 | } | ||
305 | out: | ||
306 | if (orig_path) { | ||
307 | ext4_ext_drop_refs(orig_path); | ||
308 | kfree(orig_path); | ||
309 | } | ||
310 | |||
311 | return err; | ||
312 | |||
313 | } | ||
314 | |||
315 | /** | ||
316 | * mext_insert_inside_block - Insert new extent to the extent block | ||
317 | * | ||
318 | * @o_start: first original extent to be moved | ||
319 | * @o_end: last original extent to be moved | ||
320 | * @start_ext: first new extent to be inserted | ||
321 | * @new_ext: middle of new extent to be inserted | ||
322 | * @end_ext: last new extent to be inserted | ||
323 | * @eh: extent header of target leaf block | ||
324 | * @range_to_move: used to decide how to insert extent | ||
325 | * | ||
326 | * Insert extents into the leaf block. The extent (@o_start) is overwritten | ||
327 | * by inserted extents. | ||
328 | */ | ||
329 | static void | ||
330 | mext_insert_inside_block(struct ext4_extent *o_start, | ||
331 | struct ext4_extent *o_end, | ||
332 | struct ext4_extent *start_ext, | ||
333 | struct ext4_extent *new_ext, | ||
334 | struct ext4_extent *end_ext, | ||
335 | struct ext4_extent_header *eh, | ||
336 | int range_to_move) | ||
337 | { | ||
338 | int i = 0; | ||
339 | unsigned long len; | ||
340 | |||
341 | /* Move the existing extents */ | ||
342 | if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) { | ||
343 | len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) - | ||
344 | (unsigned long)(o_end + 1); | ||
345 | memmove(o_end + 1 + range_to_move, o_end + 1, len); | ||
346 | } | ||
347 | |||
348 | /* Insert start entry */ | ||
349 | if (start_ext->ee_len) | ||
350 | o_start[i++].ee_len = start_ext->ee_len; | ||
351 | |||
352 | /* Insert new entry */ | ||
353 | if (new_ext->ee_len) { | ||
354 | o_start[i] = *new_ext; | ||
355 | ext4_ext_store_pblock(&o_start[i++], ext_pblock(new_ext)); | ||
356 | } | ||
357 | |||
358 | /* Insert end entry */ | ||
359 | if (end_ext->ee_len) | ||
360 | o_start[i] = *end_ext; | ||
361 | |||
362 | /* Increment the total entries counter on the extent block */ | ||
363 | le16_add_cpu(&eh->eh_entries, range_to_move); | ||
364 | } | ||
365 | |||
366 | /** | ||
367 | * mext_insert_extents - Insert new extent | ||
368 | * | ||
369 | * @handle: journal handle | ||
370 | * @orig_inode: original inode | ||
371 | * @orig_path: path indicates first extent to be changed | ||
372 | * @o_start: first original extent to be changed | ||
373 | * @o_end: last original extent to be changed | ||
374 | * @start_ext: first new extent to be inserted | ||
375 | * @new_ext: middle of new extent to be inserted | ||
376 | * @end_ext: last new extent to be inserted | ||
377 | * | ||
378 | * Call the function to insert extents. If we cannot add more extents into | ||
379 | * the leaf block, we call mext_insert_across_blocks() to create a | ||
380 | * new leaf block. Otherwise call mext_insert_inside_block(). Return 0 | ||
381 | * on success, or a negative error value on failure. | ||
382 | */ | ||
383 | static int | ||
384 | mext_insert_extents(handle_t *handle, struct inode *orig_inode, | ||
385 | struct ext4_ext_path *orig_path, | ||
386 | struct ext4_extent *o_start, | ||
387 | struct ext4_extent *o_end, | ||
388 | struct ext4_extent *start_ext, | ||
389 | struct ext4_extent *new_ext, | ||
390 | struct ext4_extent *end_ext) | ||
391 | { | ||
392 | struct ext4_extent_header *eh; | ||
393 | unsigned long need_slots, slots_range; | ||
394 | int range_to_move, depth, ret; | ||
395 | |||
396 | /* | ||
397 | * The extents need to be inserted | ||
398 | * start_extent + new_extent + end_extent. | ||
399 | */ | ||
400 | need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) + | ||
401 | (new_ext->ee_len ? 1 : 0); | ||
402 | |||
403 | /* The number of slots between start and end */ | ||
404 | slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1) | ||
405 | / sizeof(struct ext4_extent); | ||
406 | |||
407 | /* Range to move the end of extent */ | ||
408 | range_to_move = need_slots - slots_range; | ||
409 | depth = orig_path->p_depth; | ||
410 | orig_path += depth; | ||
411 | eh = orig_path->p_hdr; | ||
412 | |||
413 | if (depth) { | ||
414 | /* Register to journal */ | ||
415 | ret = ext4_journal_get_write_access(handle, orig_path->p_bh); | ||
416 | if (ret) | ||
417 | return ret; | ||
418 | } | ||
419 | |||
420 | /* Expansion */ | ||
421 | if (range_to_move > 0 && | ||
422 | (range_to_move > le16_to_cpu(eh->eh_max) | ||
423 | - le16_to_cpu(eh->eh_entries))) { | ||
424 | |||
425 | ret = mext_insert_across_blocks(handle, orig_inode, o_start, | ||
426 | o_end, start_ext, new_ext, end_ext); | ||
427 | if (ret < 0) | ||
428 | return ret; | ||
429 | } else | ||
430 | mext_insert_inside_block(o_start, o_end, start_ext, new_ext, | ||
431 | end_ext, eh, range_to_move); | ||
432 | |||
433 | if (depth) { | ||
434 | ret = ext4_handle_dirty_metadata(handle, orig_inode, | ||
435 | orig_path->p_bh); | ||
436 | if (ret) | ||
437 | return ret; | ||
438 | } else { | ||
439 | ret = ext4_mark_inode_dirty(handle, orig_inode); | ||
440 | if (ret < 0) | ||
441 | return ret; | ||
442 | } | ||
443 | |||
444 | return 0; | ||
445 | } | ||
446 | |||
447 | /** | ||
448 | * mext_leaf_block - Move one leaf extent block into the inode. | ||
449 | * | ||
450 | * @handle: journal handle | ||
451 | * @orig_inode: original inode | ||
452 | * @orig_path: path indicates first extent to be changed | ||
453 | * @dext: donor extent | ||
454 | * @from: start offset on the target file | ||
455 | * | ||
456 | * In order to insert extents into the leaf block, we must divide the extent | ||
457 | * in the leaf block into three extents. The one is located to be inserted | ||
458 | * extents, and the others are located around it. | ||
459 | * | ||
460 | * Therefore, this function creates structures to save extents of the leaf | ||
461 | * block, and inserts extents by calling mext_insert_extents() with | ||
462 | * created extents. Return 0 on success, or a negative error value on failure. | ||
463 | */ | ||
464 | static int | ||
465 | mext_leaf_block(handle_t *handle, struct inode *orig_inode, | ||
466 | struct ext4_ext_path *orig_path, struct ext4_extent *dext, | ||
467 | ext4_lblk_t *from) | ||
468 | { | ||
469 | struct ext4_extent *oext, *o_start, *o_end, *prev_ext; | ||
470 | struct ext4_extent new_ext, start_ext, end_ext; | ||
471 | ext4_lblk_t new_ext_end; | ||
472 | ext4_fsblk_t new_phys_end; | ||
473 | int oext_alen, new_ext_alen, end_ext_alen; | ||
474 | int depth = ext_depth(orig_inode); | ||
475 | int ret; | ||
476 | |||
477 | o_start = o_end = oext = orig_path[depth].p_ext; | ||
478 | oext_alen = ext4_ext_get_actual_len(oext); | ||
479 | start_ext.ee_len = end_ext.ee_len = 0; | ||
480 | |||
481 | new_ext.ee_block = cpu_to_le32(*from); | ||
482 | ext4_ext_store_pblock(&new_ext, ext_pblock(dext)); | ||
483 | new_ext.ee_len = dext->ee_len; | ||
484 | new_ext_alen = ext4_ext_get_actual_len(&new_ext); | ||
485 | new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1; | ||
486 | new_phys_end = ext_pblock(&new_ext) + new_ext_alen - 1; | ||
487 | |||
488 | /* | ||
489 | * Case: original extent is first | ||
490 | * oext |--------| | ||
491 | * new_ext |--| | ||
492 | * start_ext |--| | ||
493 | */ | ||
494 | if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) && | ||
495 | le32_to_cpu(new_ext.ee_block) < | ||
496 | le32_to_cpu(oext->ee_block) + oext_alen) { | ||
497 | start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) - | ||
498 | le32_to_cpu(oext->ee_block)); | ||
499 | copy_extent_status(oext, &start_ext); | ||
500 | } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) { | ||
501 | prev_ext = oext - 1; | ||
502 | /* | ||
503 | * We can merge new_ext into previous extent, | ||
504 | * if these are contiguous and same extent type. | ||
505 | */ | ||
506 | if (ext4_can_extents_be_merged(orig_inode, prev_ext, | ||
507 | &new_ext)) { | ||
508 | o_start = prev_ext; | ||
509 | start_ext.ee_len = cpu_to_le16( | ||
510 | ext4_ext_get_actual_len(prev_ext) + | ||
511 | new_ext_alen); | ||
512 | copy_extent_status(prev_ext, &start_ext); | ||
513 | new_ext.ee_len = 0; | ||
514 | } | ||
515 | } | ||
516 | |||
517 | /* | ||
518 | * Case: new_ext_end must be less than oext | ||
519 | * oext |-----------| | ||
520 | * new_ext |-------| | ||
521 | */ | ||
522 | BUG_ON(le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end); | ||
523 | |||
524 | /* | ||
525 | * Case: new_ext is smaller than original extent | ||
526 | * oext |---------------| | ||
527 | * new_ext |-----------| | ||
528 | * end_ext |---| | ||
529 | */ | ||
530 | if (le32_to_cpu(oext->ee_block) <= new_ext_end && | ||
531 | new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) { | ||
532 | end_ext.ee_len = | ||
533 | cpu_to_le16(le32_to_cpu(oext->ee_block) + | ||
534 | oext_alen - 1 - new_ext_end); | ||
535 | copy_extent_status(oext, &end_ext); | ||
536 | end_ext_alen = ext4_ext_get_actual_len(&end_ext); | ||
537 | ext4_ext_store_pblock(&end_ext, | ||
538 | (ext_pblock(o_end) + oext_alen - end_ext_alen)); | ||
539 | end_ext.ee_block = | ||
540 | cpu_to_le32(le32_to_cpu(o_end->ee_block) + | ||
541 | oext_alen - end_ext_alen); | ||
542 | } | ||
543 | |||
544 | ret = mext_insert_extents(handle, orig_inode, orig_path, o_start, | ||
545 | o_end, &start_ext, &new_ext, &end_ext); | ||
546 | return ret; | ||
547 | } | ||
548 | |||
549 | /** | ||
550 | * mext_calc_swap_extents - Calculate extents for extent swapping. | ||
551 | * | ||
552 | * @tmp_dext: the extent that will belong to the original inode | ||
553 | * @tmp_oext: the extent that will belong to the donor inode | ||
554 | * @orig_off: block offset of original inode | ||
555 | * @donor_off: block offset of donor inode | ||
556 | * @max_count: the maximun length of extents | ||
557 | */ | ||
558 | static void | ||
559 | mext_calc_swap_extents(struct ext4_extent *tmp_dext, | ||
560 | struct ext4_extent *tmp_oext, | ||
561 | ext4_lblk_t orig_off, ext4_lblk_t donor_off, | ||
562 | ext4_lblk_t max_count) | ||
563 | { | ||
564 | ext4_lblk_t diff, orig_diff; | ||
565 | struct ext4_extent dext_old, oext_old; | ||
566 | |||
567 | dext_old = *tmp_dext; | ||
568 | oext_old = *tmp_oext; | ||
569 | |||
570 | /* When tmp_dext is too large, pick up the target range. */ | ||
571 | diff = donor_off - le32_to_cpu(tmp_dext->ee_block); | ||
572 | |||
573 | ext4_ext_store_pblock(tmp_dext, ext_pblock(tmp_dext) + diff); | ||
574 | tmp_dext->ee_block = | ||
575 | cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff); | ||
576 | tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff); | ||
577 | |||
578 | if (max_count < ext4_ext_get_actual_len(tmp_dext)) | ||
579 | tmp_dext->ee_len = cpu_to_le16(max_count); | ||
580 | |||
581 | orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block); | ||
582 | ext4_ext_store_pblock(tmp_oext, ext_pblock(tmp_oext) + orig_diff); | ||
583 | |||
584 | /* Adjust extent length if donor extent is larger than orig */ | ||
585 | if (ext4_ext_get_actual_len(tmp_dext) > | ||
586 | ext4_ext_get_actual_len(tmp_oext) - orig_diff) | ||
587 | tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) - | ||
588 | orig_diff); | ||
589 | |||
590 | tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext)); | ||
591 | |||
592 | copy_extent_status(&oext_old, tmp_dext); | ||
593 | copy_extent_status(&dext_old, tmp_oext); | ||
594 | } | ||
595 | |||
596 | /** | ||
597 | * mext_replace_branches - Replace original extents with new extents | ||
598 | * | ||
599 | * @handle: journal handle | ||
600 | * @orig_inode: original inode | ||
601 | * @donor_inode: donor inode | ||
602 | * @from: block offset of orig_inode | ||
603 | * @count: block count to be replaced | ||
604 | * | ||
605 | * Replace original inode extents and donor inode extents page by page. | ||
606 | * We implement this replacement in the following three steps: | ||
607 | * 1. Save the block information of original and donor inodes into | ||
608 | * dummy extents. | ||
609 | * 2. Change the block information of original inode to point at the | ||
610 | * donor inode blocks. | ||
611 | * 3. Change the block information of donor inode to point at the saved | ||
612 | * original inode blocks in the dummy extents. | ||
613 | * | ||
614 | * Return 0 on success, or a negative error value on failure. | ||
615 | */ | ||
616 | static int | ||
617 | mext_replace_branches(handle_t *handle, struct inode *orig_inode, | ||
618 | struct inode *donor_inode, ext4_lblk_t from, | ||
619 | ext4_lblk_t count) | ||
620 | { | ||
621 | struct ext4_ext_path *orig_path = NULL; | ||
622 | struct ext4_ext_path *donor_path = NULL; | ||
623 | struct ext4_extent *oext, *dext; | ||
624 | struct ext4_extent tmp_dext, tmp_oext; | ||
625 | ext4_lblk_t orig_off = from, donor_off = from; | ||
626 | int err = 0; | ||
627 | int depth; | ||
628 | int replaced_count = 0; | ||
629 | int dext_alen; | ||
630 | |||
631 | mext_double_down_write(orig_inode, donor_inode); | ||
632 | |||
633 | /* Get the original extent for the block "orig_off" */ | ||
634 | get_ext_path(orig_path, orig_inode, orig_off, err); | ||
635 | if (orig_path == NULL) | ||
636 | goto out; | ||
637 | |||
638 | /* Get the donor extent for the head */ | ||
639 | get_ext_path(donor_path, donor_inode, donor_off, err); | ||
640 | if (donor_path == NULL) | ||
641 | goto out; | ||
642 | depth = ext_depth(orig_inode); | ||
643 | oext = orig_path[depth].p_ext; | ||
644 | tmp_oext = *oext; | ||
645 | |||
646 | depth = ext_depth(donor_inode); | ||
647 | dext = donor_path[depth].p_ext; | ||
648 | tmp_dext = *dext; | ||
649 | |||
650 | mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, | ||
651 | donor_off, count); | ||
652 | |||
653 | /* Loop for the donor extents */ | ||
654 | while (1) { | ||
655 | /* The extent for donor must be found. */ | ||
656 | BUG_ON(!dext || donor_off != le32_to_cpu(tmp_dext.ee_block)); | ||
657 | |||
658 | /* Set donor extent to orig extent */ | ||
659 | err = mext_leaf_block(handle, orig_inode, | ||
660 | orig_path, &tmp_dext, &orig_off); | ||
661 | if (err < 0) | ||
662 | goto out; | ||
663 | |||
664 | /* Set orig extent to donor extent */ | ||
665 | err = mext_leaf_block(handle, donor_inode, | ||
666 | donor_path, &tmp_oext, &donor_off); | ||
667 | if (err < 0) | ||
668 | goto out; | ||
669 | |||
670 | dext_alen = ext4_ext_get_actual_len(&tmp_dext); | ||
671 | replaced_count += dext_alen; | ||
672 | donor_off += dext_alen; | ||
673 | orig_off += dext_alen; | ||
674 | |||
675 | /* Already moved the expected blocks */ | ||
676 | if (replaced_count >= count) | ||
677 | break; | ||
678 | |||
679 | if (orig_path) | ||
680 | ext4_ext_drop_refs(orig_path); | ||
681 | get_ext_path(orig_path, orig_inode, orig_off, err); | ||
682 | if (orig_path == NULL) | ||
683 | goto out; | ||
684 | depth = ext_depth(orig_inode); | ||
685 | oext = orig_path[depth].p_ext; | ||
686 | if (le32_to_cpu(oext->ee_block) + | ||
687 | ext4_ext_get_actual_len(oext) <= orig_off) { | ||
688 | err = 0; | ||
689 | goto out; | ||
690 | } | ||
691 | tmp_oext = *oext; | ||
692 | |||
693 | if (donor_path) | ||
694 | ext4_ext_drop_refs(donor_path); | ||
695 | get_ext_path(donor_path, donor_inode, | ||
696 | donor_off, err); | ||
697 | if (donor_path == NULL) | ||
698 | goto out; | ||
699 | depth = ext_depth(donor_inode); | ||
700 | dext = donor_path[depth].p_ext; | ||
701 | if (le32_to_cpu(dext->ee_block) + | ||
702 | ext4_ext_get_actual_len(dext) <= donor_off) { | ||
703 | err = 0; | ||
704 | goto out; | ||
705 | } | ||
706 | tmp_dext = *dext; | ||
707 | |||
708 | mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, | ||
709 | donor_off, | ||
710 | count - replaced_count); | ||
711 | } | ||
712 | |||
713 | out: | ||
714 | if (orig_path) { | ||
715 | ext4_ext_drop_refs(orig_path); | ||
716 | kfree(orig_path); | ||
717 | } | ||
718 | if (donor_path) { | ||
719 | ext4_ext_drop_refs(donor_path); | ||
720 | kfree(donor_path); | ||
721 | } | ||
722 | |||
723 | mext_double_up_write(orig_inode, donor_inode); | ||
724 | return err; | ||
725 | } | ||
726 | |||
727 | /** | ||
728 | * move_extent_per_page - Move extent data per page | ||
729 | * | ||
730 | * @o_filp: file structure of original file | ||
731 | * @donor_inode: donor inode | ||
732 | * @orig_page_offset: page index on original file | ||
733 | * @data_offset_in_page: block index where data swapping starts | ||
734 | * @block_len_in_page: the number of blocks to be swapped | ||
735 | * @uninit: orig extent is uninitialized or not | ||
736 | * | ||
737 | * Save the data in original inode blocks and replace original inode extents | ||
738 | * with donor inode extents by calling mext_replace_branches(). | ||
739 | * Finally, write out the saved data in new original inode blocks. Return 0 | ||
740 | * on success, or a negative error value on failure. | ||
741 | */ | ||
742 | static int | ||
743 | move_extent_par_page(struct file *o_filp, struct inode *donor_inode, | ||
744 | pgoff_t orig_page_offset, int data_offset_in_page, | ||
745 | int block_len_in_page, int uninit) | ||
746 | { | ||
747 | struct inode *orig_inode = o_filp->f_dentry->d_inode; | ||
748 | struct address_space *mapping = orig_inode->i_mapping; | ||
749 | struct buffer_head *bh; | ||
750 | struct page *page = NULL; | ||
751 | const struct address_space_operations *a_ops = mapping->a_ops; | ||
752 | handle_t *handle; | ||
753 | ext4_lblk_t orig_blk_offset; | ||
754 | long long offs = orig_page_offset << PAGE_CACHE_SHIFT; | ||
755 | unsigned long blocksize = orig_inode->i_sb->s_blocksize; | ||
756 | unsigned int w_flags = 0; | ||
757 | unsigned int tmp_data_len, data_len; | ||
758 | void *fsdata; | ||
759 | int ret, i, jblocks; | ||
760 | int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; | ||
761 | |||
762 | /* | ||
763 | * It needs twice the amount of ordinary journal buffers because | ||
764 | * inode and donor_inode may change each different metadata blocks. | ||
765 | */ | ||
766 | jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; | ||
767 | handle = ext4_journal_start(orig_inode, jblocks); | ||
768 | if (IS_ERR(handle)) { | ||
769 | ret = PTR_ERR(handle); | ||
770 | return ret; | ||
771 | } | ||
772 | |||
773 | if (segment_eq(get_fs(), KERNEL_DS)) | ||
774 | w_flags |= AOP_FLAG_UNINTERRUPTIBLE; | ||
775 | |||
776 | orig_blk_offset = orig_page_offset * blocks_per_page + | ||
777 | data_offset_in_page; | ||
778 | |||
779 | /* | ||
780 | * If orig extent is uninitialized one, | ||
781 | * it's not necessary force the page into memory | ||
782 | * and then force it to be written out again. | ||
783 | * Just swap data blocks between orig and donor. | ||
784 | */ | ||
785 | if (uninit) { | ||
786 | ret = mext_replace_branches(handle, orig_inode, | ||
787 | donor_inode, orig_blk_offset, | ||
788 | block_len_in_page); | ||
789 | |||
790 | /* Clear the inode cache not to refer to the old data */ | ||
791 | ext4_ext_invalidate_cache(orig_inode); | ||
792 | ext4_ext_invalidate_cache(donor_inode); | ||
793 | goto out2; | ||
794 | } | ||
795 | |||
796 | offs = (long long)orig_blk_offset << orig_inode->i_blkbits; | ||
797 | |||
798 | /* Calculate data_len */ | ||
799 | if ((orig_blk_offset + block_len_in_page - 1) == | ||
800 | ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { | ||
801 | /* Replace the last block */ | ||
802 | tmp_data_len = orig_inode->i_size & (blocksize - 1); | ||
803 | /* | ||
804 | * If data_len equal zero, it shows data_len is multiples of | ||
805 | * blocksize. So we set appropriate value. | ||
806 | */ | ||
807 | if (tmp_data_len == 0) | ||
808 | tmp_data_len = blocksize; | ||
809 | |||
810 | data_len = tmp_data_len + | ||
811 | ((block_len_in_page - 1) << orig_inode->i_blkbits); | ||
812 | } else { | ||
813 | data_len = block_len_in_page << orig_inode->i_blkbits; | ||
814 | } | ||
815 | |||
816 | ret = a_ops->write_begin(o_filp, mapping, offs, data_len, w_flags, | ||
817 | &page, &fsdata); | ||
818 | if (unlikely(ret < 0)) | ||
819 | goto out; | ||
820 | |||
821 | if (!PageUptodate(page)) { | ||
822 | mapping->a_ops->readpage(o_filp, page); | ||
823 | lock_page(page); | ||
824 | } | ||
825 | |||
826 | /* | ||
827 | * try_to_release_page() doesn't call releasepage in writeback mode. | ||
828 | * We should care about the order of writing to the same file | ||
829 | * by multiple move extent processes. | ||
830 | * It needs to call wait_on_page_writeback() to wait for the | ||
831 | * writeback of the page. | ||
832 | */ | ||
833 | if (PageWriteback(page)) | ||
834 | wait_on_page_writeback(page); | ||
835 | |||
836 | /* Release old bh and drop refs */ | ||
837 | try_to_release_page(page, 0); | ||
838 | |||
839 | ret = mext_replace_branches(handle, orig_inode, donor_inode, | ||
840 | orig_blk_offset, block_len_in_page); | ||
841 | if (ret < 0) | ||
842 | goto out; | ||
843 | |||
844 | /* Clear the inode cache not to refer to the old data */ | ||
845 | ext4_ext_invalidate_cache(orig_inode); | ||
846 | ext4_ext_invalidate_cache(donor_inode); | ||
847 | |||
848 | if (!page_has_buffers(page)) | ||
849 | create_empty_buffers(page, 1 << orig_inode->i_blkbits, 0); | ||
850 | |||
851 | bh = page_buffers(page); | ||
852 | for (i = 0; i < data_offset_in_page; i++) | ||
853 | bh = bh->b_this_page; | ||
854 | |||
855 | for (i = 0; i < block_len_in_page; i++) { | ||
856 | ret = ext4_get_block(orig_inode, | ||
857 | (sector_t)(orig_blk_offset + i), bh, 0); | ||
858 | if (ret < 0) | ||
859 | goto out; | ||
860 | |||
861 | if (bh->b_this_page != NULL) | ||
862 | bh = bh->b_this_page; | ||
863 | } | ||
864 | |||
865 | ret = a_ops->write_end(o_filp, mapping, offs, data_len, data_len, | ||
866 | page, fsdata); | ||
867 | page = NULL; | ||
868 | |||
869 | out: | ||
870 | if (unlikely(page)) { | ||
871 | if (PageLocked(page)) | ||
872 | unlock_page(page); | ||
873 | page_cache_release(page); | ||
874 | } | ||
875 | out2: | ||
876 | ext4_journal_stop(handle); | ||
877 | |||
878 | return ret < 0 ? ret : 0; | ||
879 | } | ||
880 | |||
881 | /** | ||
882 | * mext_check_argumants - Check whether move extent can be done | ||
883 | * | ||
884 | * @orig_inode: original inode | ||
885 | * @donor_inode: donor inode | ||
886 | * @orig_start: logical start offset in block for orig | ||
887 | * @donor_start: logical start offset in block for donor | ||
888 | * @len: the number of blocks to be moved | ||
889 | * @moved_len: moved block length | ||
890 | * | ||
891 | * Check the arguments of ext4_move_extents() whether the files can be | ||
892 | * exchanged with each other. | ||
893 | * Return 0 on success, or a negative error value on failure. | ||
894 | */ | ||
895 | static int | ||
896 | mext_check_arguments(struct inode *orig_inode, | ||
897 | struct inode *donor_inode, __u64 orig_start, | ||
898 | __u64 donor_start, __u64 *len, __u64 moved_len) | ||
899 | { | ||
900 | /* Regular file check */ | ||
901 | if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { | ||
902 | ext4_debug("ext4 move extent: The argument files should be " | ||
903 | "regular file [ino:orig %lu, donor %lu]\n", | ||
904 | orig_inode->i_ino, donor_inode->i_ino); | ||
905 | return -EINVAL; | ||
906 | } | ||
907 | |||
908 | /* Ext4 move extent does not support swapfile */ | ||
909 | if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { | ||
910 | ext4_debug("ext4 move extent: The argument files should " | ||
911 | "not be swapfile [ino:orig %lu, donor %lu]\n", | ||
912 | orig_inode->i_ino, donor_inode->i_ino); | ||
913 | return -EINVAL; | ||
914 | } | ||
915 | |||
916 | /* Files should be in the same ext4 FS */ | ||
917 | if (orig_inode->i_sb != donor_inode->i_sb) { | ||
918 | ext4_debug("ext4 move extent: The argument files " | ||
919 | "should be in same FS [ino:orig %lu, donor %lu]\n", | ||
920 | orig_inode->i_ino, donor_inode->i_ino); | ||
921 | return -EINVAL; | ||
922 | } | ||
923 | |||
924 | /* orig and donor should be different file */ | ||
925 | if (orig_inode->i_ino == donor_inode->i_ino) { | ||
926 | ext4_debug("ext4 move extent: The argument files should not " | ||
927 | "be same file [ino:orig %lu, donor %lu]\n", | ||
928 | orig_inode->i_ino, donor_inode->i_ino); | ||
929 | return -EINVAL; | ||
930 | } | ||
931 | |||
932 | /* Ext4 move extent supports only extent based file */ | ||
933 | if (!(EXT4_I(orig_inode)->i_flags & EXT4_EXTENTS_FL)) { | ||
934 | ext4_debug("ext4 move extent: orig file is not extents " | ||
935 | "based file [ino:orig %lu]\n", orig_inode->i_ino); | ||
936 | return -EOPNOTSUPP; | ||
937 | } else if (!(EXT4_I(donor_inode)->i_flags & EXT4_EXTENTS_FL)) { | ||
938 | ext4_debug("ext4 move extent: donor file is not extents " | ||
939 | "based file [ino:donor %lu]\n", donor_inode->i_ino); | ||
940 | return -EOPNOTSUPP; | ||
941 | } | ||
942 | |||
943 | if ((!orig_inode->i_size) || (!donor_inode->i_size)) { | ||
944 | ext4_debug("ext4 move extent: File size is 0 byte\n"); | ||
945 | return -EINVAL; | ||
946 | } | ||
947 | |||
948 | /* Start offset should be same */ | ||
949 | if (orig_start != donor_start) { | ||
950 | ext4_debug("ext4 move extent: orig and donor's start " | ||
951 | "offset are not same [ino:orig %lu, donor %lu]\n", | ||
952 | orig_inode->i_ino, donor_inode->i_ino); | ||
953 | return -EINVAL; | ||
954 | } | ||
955 | |||
956 | if (moved_len) { | ||
957 | ext4_debug("ext4 move extent: moved_len should be 0 " | ||
958 | "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, | ||
959 | donor_inode->i_ino); | ||
960 | return -EINVAL; | ||
961 | } | ||
962 | |||
963 | if ((orig_start > MAX_DEFRAG_SIZE) || | ||
964 | (donor_start > MAX_DEFRAG_SIZE) || | ||
965 | (*len > MAX_DEFRAG_SIZE) || | ||
966 | (orig_start + *len > MAX_DEFRAG_SIZE)) { | ||
967 | ext4_debug("ext4 move extent: Can't handle over [%lu] blocks " | ||
968 | "[ino:orig %lu, donor %lu]\n", MAX_DEFRAG_SIZE, | ||
969 | orig_inode->i_ino, donor_inode->i_ino); | ||
970 | return -EINVAL; | ||
971 | } | ||
972 | |||
973 | if (orig_inode->i_size > donor_inode->i_size) { | ||
974 | if (orig_start >= donor_inode->i_size) { | ||
975 | ext4_debug("ext4 move extent: orig start offset " | ||
976 | "[%llu] should be less than donor file size " | ||
977 | "[%lld] [ino:orig %lu, donor_inode %lu]\n", | ||
978 | orig_start, donor_inode->i_size, | ||
979 | orig_inode->i_ino, donor_inode->i_ino); | ||
980 | return -EINVAL; | ||
981 | } | ||
982 | |||
983 | if (orig_start + *len > donor_inode->i_size) { | ||
984 | ext4_debug("ext4 move extent: End offset [%llu] should " | ||
985 | "be less than donor file size [%lld]." | ||
986 | "So adjust length from %llu to %lld " | ||
987 | "[ino:orig %lu, donor %lu]\n", | ||
988 | orig_start + *len, donor_inode->i_size, | ||
989 | *len, donor_inode->i_size - orig_start, | ||
990 | orig_inode->i_ino, donor_inode->i_ino); | ||
991 | *len = donor_inode->i_size - orig_start; | ||
992 | } | ||
993 | } else { | ||
994 | if (orig_start >= orig_inode->i_size) { | ||
995 | ext4_debug("ext4 move extent: start offset [%llu] " | ||
996 | "should be less than original file size " | ||
997 | "[%lld] [inode:orig %lu, donor %lu]\n", | ||
998 | orig_start, orig_inode->i_size, | ||
999 | orig_inode->i_ino, donor_inode->i_ino); | ||
1000 | return -EINVAL; | ||
1001 | } | ||
1002 | |||
1003 | if (orig_start + *len > orig_inode->i_size) { | ||
1004 | ext4_debug("ext4 move extent: Adjust length " | ||
1005 | "from %llu to %lld. Because it should be " | ||
1006 | "less than original file size " | ||
1007 | "[ino:orig %lu, donor %lu]\n", | ||
1008 | *len, orig_inode->i_size - orig_start, | ||
1009 | orig_inode->i_ino, donor_inode->i_ino); | ||
1010 | *len = orig_inode->i_size - orig_start; | ||
1011 | } | ||
1012 | } | ||
1013 | |||
1014 | if (!*len) { | ||
1015 | ext4_debug("ext4 move extent: len shoudld not be 0 " | ||
1016 | "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, | ||
1017 | donor_inode->i_ino); | ||
1018 | return -EINVAL; | ||
1019 | } | ||
1020 | |||
1021 | return 0; | ||
1022 | } | ||
1023 | |||
1024 | /** | ||
1025 | * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2 | ||
1026 | * | ||
1027 | * @inode1: the inode structure | ||
1028 | * @inode2: the inode structure | ||
1029 | * | ||
1030 | * Lock two inodes' i_mutex by i_ino order. This function is moved from | ||
1031 | * fs/inode.c. | ||
1032 | */ | ||
1033 | static void | ||
1034 | mext_inode_double_lock(struct inode *inode1, struct inode *inode2) | ||
1035 | { | ||
1036 | if (inode1 == NULL || inode2 == NULL || inode1 == inode2) { | ||
1037 | if (inode1) | ||
1038 | mutex_lock(&inode1->i_mutex); | ||
1039 | else if (inode2) | ||
1040 | mutex_lock(&inode2->i_mutex); | ||
1041 | return; | ||
1042 | } | ||
1043 | |||
1044 | if (inode1->i_ino < inode2->i_ino) { | ||
1045 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT); | ||
1046 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD); | ||
1047 | } else { | ||
1048 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT); | ||
1049 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD); | ||
1050 | } | ||
1051 | } | ||
1052 | |||
1053 | /** | ||
1054 | * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2 | ||
1055 | * | ||
1056 | * @inode1: the inode that is released first | ||
1057 | * @inode2: the inode that is released second | ||
1058 | * | ||
1059 | * This function is moved from fs/inode.c. | ||
1060 | */ | ||
1061 | |||
1062 | static void | ||
1063 | mext_inode_double_unlock(struct inode *inode1, struct inode *inode2) | ||
1064 | { | ||
1065 | if (inode1) | ||
1066 | mutex_unlock(&inode1->i_mutex); | ||
1067 | |||
1068 | if (inode2 && inode2 != inode1) | ||
1069 | mutex_unlock(&inode2->i_mutex); | ||
1070 | } | ||
1071 | |||
1072 | /** | ||
1073 | * ext4_move_extents - Exchange the specified range of a file | ||
1074 | * | ||
1075 | * @o_filp: file structure of the original file | ||
1076 | * @d_filp: file structure of the donor file | ||
1077 | * @orig_start: start offset in block for orig | ||
1078 | * @donor_start: start offset in block for donor | ||
1079 | * @len: the number of blocks to be moved | ||
1080 | * @moved_len: moved block length | ||
1081 | * | ||
1082 | * This function returns 0 and moved block length is set in moved_len | ||
1083 | * if succeed, otherwise returns error value. | ||
1084 | * | ||
1085 | * Note: ext4_move_extents() proceeds the following order. | ||
1086 | * 1:ext4_move_extents() calculates the last block number of moving extent | ||
1087 | * function by the start block number (orig_start) and the number of blocks | ||
1088 | * to be moved (len) specified as arguments. | ||
1089 | * If the {orig, donor}_start points a hole, the extent's start offset | ||
1090 | * pointed by ext_cur (current extent), holecheck_path, orig_path are set | ||
1091 | * after hole behind. | ||
1092 | * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent | ||
1093 | * or the ext_cur exceeds the block_end which is last logical block number. | ||
1094 | * 3:To get the length of continues area, call mext_next_extent() | ||
1095 | * specified with the ext_cur (initial value is holecheck_path) re-cursive, | ||
1096 | * until find un-continuous extent, the start logical block number exceeds | ||
1097 | * the block_end or the extent points to the last extent. | ||
1098 | * 4:Exchange the original inode data with donor inode data | ||
1099 | * from orig_page_offset to seq_end_page. | ||
1100 | * The start indexes of data are specified as arguments. | ||
1101 | * That of the original inode is orig_page_offset, | ||
1102 | * and the donor inode is also orig_page_offset | ||
1103 | * (To easily handle blocksize != pagesize case, the offset for the | ||
1104 | * donor inode is block unit). | ||
1105 | * 5:Update holecheck_path and orig_path to points a next proceeding extent, | ||
1106 | * then returns to step 2. | ||
1107 | * 6:Release holecheck_path, orig_path and set the len to moved_len | ||
1108 | * which shows the number of moved blocks. | ||
1109 | * The moved_len is useful for the command to calculate the file offset | ||
1110 | * for starting next move extent ioctl. | ||
1111 | * 7:Return 0 on success, or a negative error value on failure. | ||
1112 | */ | ||
1113 | int | ||
1114 | ext4_move_extents(struct file *o_filp, struct file *d_filp, | ||
1115 | __u64 orig_start, __u64 donor_start, __u64 len, | ||
1116 | __u64 *moved_len) | ||
1117 | { | ||
1118 | struct inode *orig_inode = o_filp->f_dentry->d_inode; | ||
1119 | struct inode *donor_inode = d_filp->f_dentry->d_inode; | ||
1120 | struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL; | ||
1121 | struct ext4_extent *ext_prev, *ext_cur, *ext_dummy; | ||
1122 | ext4_lblk_t block_start = orig_start; | ||
1123 | ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0; | ||
1124 | ext4_lblk_t rest_blocks; | ||
1125 | pgoff_t orig_page_offset = 0, seq_end_page; | ||
1126 | int ret, depth, last_extent = 0; | ||
1127 | int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; | ||
1128 | int data_offset_in_page; | ||
1129 | int block_len_in_page; | ||
1130 | int uninit; | ||
1131 | |||
1132 | /* protect orig and donor against a truncate */ | ||
1133 | mext_inode_double_lock(orig_inode, donor_inode); | ||
1134 | |||
1135 | mext_double_down_read(orig_inode, donor_inode); | ||
1136 | /* Check the filesystem environment whether move_extent can be done */ | ||
1137 | ret = mext_check_arguments(orig_inode, donor_inode, orig_start, | ||
1138 | donor_start, &len, *moved_len); | ||
1139 | mext_double_up_read(orig_inode, donor_inode); | ||
1140 | if (ret) | ||
1141 | goto out2; | ||
1142 | |||
1143 | file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits; | ||
1144 | block_end = block_start + len - 1; | ||
1145 | if (file_end < block_end) | ||
1146 | len -= block_end - file_end; | ||
1147 | |||
1148 | get_ext_path(orig_path, orig_inode, block_start, ret); | ||
1149 | if (orig_path == NULL) | ||
1150 | goto out2; | ||
1151 | |||
1152 | /* Get path structure to check the hole */ | ||
1153 | get_ext_path(holecheck_path, orig_inode, block_start, ret); | ||
1154 | if (holecheck_path == NULL) | ||
1155 | goto out; | ||
1156 | |||
1157 | depth = ext_depth(orig_inode); | ||
1158 | ext_cur = holecheck_path[depth].p_ext; | ||
1159 | if (ext_cur == NULL) { | ||
1160 | ret = -EINVAL; | ||
1161 | goto out; | ||
1162 | } | ||
1163 | |||
1164 | /* | ||
1165 | * Get proper extent whose ee_block is beyond block_start | ||
1166 | * if block_start was within the hole. | ||
1167 | */ | ||
1168 | if (le32_to_cpu(ext_cur->ee_block) + | ||
1169 | ext4_ext_get_actual_len(ext_cur) - 1 < block_start) { | ||
1170 | last_extent = mext_next_extent(orig_inode, | ||
1171 | holecheck_path, &ext_cur); | ||
1172 | if (last_extent < 0) { | ||
1173 | ret = last_extent; | ||
1174 | goto out; | ||
1175 | } | ||
1176 | last_extent = mext_next_extent(orig_inode, orig_path, | ||
1177 | &ext_dummy); | ||
1178 | if (last_extent < 0) { | ||
1179 | ret = last_extent; | ||
1180 | goto out; | ||
1181 | } | ||
1182 | } | ||
1183 | seq_start = block_start; | ||
1184 | |||
1185 | /* No blocks within the specified range. */ | ||
1186 | if (le32_to_cpu(ext_cur->ee_block) > block_end) { | ||
1187 | ext4_debug("ext4 move extent: The specified range of file " | ||
1188 | "may be the hole\n"); | ||
1189 | ret = -EINVAL; | ||
1190 | goto out; | ||
1191 | } | ||
1192 | |||
1193 | /* Adjust start blocks */ | ||
1194 | add_blocks = min(le32_to_cpu(ext_cur->ee_block) + | ||
1195 | ext4_ext_get_actual_len(ext_cur), block_end + 1) - | ||
1196 | max(le32_to_cpu(ext_cur->ee_block), block_start); | ||
1197 | |||
1198 | while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) { | ||
1199 | seq_blocks += add_blocks; | ||
1200 | |||
1201 | /* Adjust tail blocks */ | ||
1202 | if (seq_start + seq_blocks - 1 > block_end) | ||
1203 | seq_blocks = block_end - seq_start + 1; | ||
1204 | |||
1205 | ext_prev = ext_cur; | ||
1206 | last_extent = mext_next_extent(orig_inode, holecheck_path, | ||
1207 | &ext_cur); | ||
1208 | if (last_extent < 0) { | ||
1209 | ret = last_extent; | ||
1210 | break; | ||
1211 | } | ||
1212 | add_blocks = ext4_ext_get_actual_len(ext_cur); | ||
1213 | |||
1214 | /* | ||
1215 | * Extend the length of contiguous block (seq_blocks) | ||
1216 | * if extents are contiguous. | ||
1217 | */ | ||
1218 | if (ext4_can_extents_be_merged(orig_inode, | ||
1219 | ext_prev, ext_cur) && | ||
1220 | block_end >= le32_to_cpu(ext_cur->ee_block) && | ||
1221 | !last_extent) | ||
1222 | continue; | ||
1223 | |||
1224 | /* Is original extent is uninitialized */ | ||
1225 | uninit = ext4_ext_is_uninitialized(ext_prev); | ||
1226 | |||
1227 | data_offset_in_page = seq_start % blocks_per_page; | ||
1228 | |||
1229 | /* | ||
1230 | * Calculate data blocks count that should be swapped | ||
1231 | * at the first page. | ||
1232 | */ | ||
1233 | if (data_offset_in_page + seq_blocks > blocks_per_page) { | ||
1234 | /* Swapped blocks are across pages */ | ||
1235 | block_len_in_page = | ||
1236 | blocks_per_page - data_offset_in_page; | ||
1237 | } else { | ||
1238 | /* Swapped blocks are in a page */ | ||
1239 | block_len_in_page = seq_blocks; | ||
1240 | } | ||
1241 | |||
1242 | orig_page_offset = seq_start >> | ||
1243 | (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); | ||
1244 | seq_end_page = (seq_start + seq_blocks - 1) >> | ||
1245 | (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); | ||
1246 | seq_start = le32_to_cpu(ext_cur->ee_block); | ||
1247 | rest_blocks = seq_blocks; | ||
1248 | |||
1249 | /* Discard preallocations of two inodes */ | ||
1250 | down_write(&EXT4_I(orig_inode)->i_data_sem); | ||
1251 | ext4_discard_preallocations(orig_inode); | ||
1252 | up_write(&EXT4_I(orig_inode)->i_data_sem); | ||
1253 | |||
1254 | down_write(&EXT4_I(donor_inode)->i_data_sem); | ||
1255 | ext4_discard_preallocations(donor_inode); | ||
1256 | up_write(&EXT4_I(donor_inode)->i_data_sem); | ||
1257 | |||
1258 | while (orig_page_offset <= seq_end_page) { | ||
1259 | |||
1260 | /* Swap original branches with new branches */ | ||
1261 | ret = move_extent_par_page(o_filp, donor_inode, | ||
1262 | orig_page_offset, | ||
1263 | data_offset_in_page, | ||
1264 | block_len_in_page, uninit); | ||
1265 | if (ret < 0) | ||
1266 | goto out; | ||
1267 | orig_page_offset++; | ||
1268 | /* Count how many blocks we have exchanged */ | ||
1269 | *moved_len += block_len_in_page; | ||
1270 | BUG_ON(*moved_len > len); | ||
1271 | |||
1272 | data_offset_in_page = 0; | ||
1273 | rest_blocks -= block_len_in_page; | ||
1274 | if (rest_blocks > blocks_per_page) | ||
1275 | block_len_in_page = blocks_per_page; | ||
1276 | else | ||
1277 | block_len_in_page = rest_blocks; | ||
1278 | } | ||
1279 | |||
1280 | /* Decrease buffer counter */ | ||
1281 | if (holecheck_path) | ||
1282 | ext4_ext_drop_refs(holecheck_path); | ||
1283 | get_ext_path(holecheck_path, orig_inode, | ||
1284 | seq_start, ret); | ||
1285 | if (holecheck_path == NULL) | ||
1286 | break; | ||
1287 | depth = holecheck_path->p_depth; | ||
1288 | |||
1289 | /* Decrease buffer counter */ | ||
1290 | if (orig_path) | ||
1291 | ext4_ext_drop_refs(orig_path); | ||
1292 | get_ext_path(orig_path, orig_inode, seq_start, ret); | ||
1293 | if (orig_path == NULL) | ||
1294 | break; | ||
1295 | |||
1296 | ext_cur = holecheck_path[depth].p_ext; | ||
1297 | add_blocks = ext4_ext_get_actual_len(ext_cur); | ||
1298 | seq_blocks = 0; | ||
1299 | |||
1300 | } | ||
1301 | out: | ||
1302 | if (orig_path) { | ||
1303 | ext4_ext_drop_refs(orig_path); | ||
1304 | kfree(orig_path); | ||
1305 | } | ||
1306 | if (holecheck_path) { | ||
1307 | ext4_ext_drop_refs(holecheck_path); | ||
1308 | kfree(holecheck_path); | ||
1309 | } | ||
1310 | out2: | ||
1311 | mext_inode_double_unlock(orig_inode, donor_inode); | ||
1312 | |||
1313 | if (ret) | ||
1314 | return ret; | ||
1315 | |||
1316 | /* All of the specified blocks must be exchanged in succeed */ | ||
1317 | BUG_ON(*moved_len != len); | ||
1318 | |||
1319 | return 0; | ||
1320 | } | ||