blob: 3a6c92ac131c449b61f58194087fe376dbd0f32f [file] [log] [blame]
Akira Fujita748de672009-06-17 19:24:03 -04001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090018#include <linux/slab.h>
Akira Fujita748de672009-06-17 19:24:03 -040019#include "ext4_jbd2.h"
20#include "ext4_extents.h"
21#include "ext4.h"
22
Akira Fujitae8505972009-09-16 13:46:38 -040023/**
24 * get_ext_path - Find an extent path for designated logical block number.
25 *
26 * @inode: an inode which is searched
27 * @lblock: logical block number to find an extent path
28 * @path: pointer to an extent path pointer (for output)
29 *
30 * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value
31 * on failure.
32 */
33static inline int
34get_ext_path(struct inode *inode, ext4_lblk_t lblock,
35 struct ext4_ext_path **path)
36{
37 int ret = 0;
38
39 *path = ext4_ext_find_extent(inode, lblock, *path);
40 if (IS_ERR(*path)) {
41 ret = PTR_ERR(*path);
42 *path = NULL;
Akira Fujita347fa6f2009-09-16 14:25:07 -040043 } else if ((*path)[ext_depth(inode)].p_ext == NULL)
44 ret = -ENODATA;
45
Akira Fujitae8505972009-09-16 13:46:38 -040046 return ret;
47}
Akira Fujita748de672009-06-17 19:24:03 -040048
49/**
50 * copy_extent_status - Copy the extent's initialization status
51 *
52 * @src: an extent for getting initialize status
53 * @dest: an extent to be set the status
54 */
55static void
56copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
57{
58 if (ext4_ext_is_uninitialized(src))
59 ext4_ext_mark_uninitialized(dest);
60 else
61 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
62}
63
64/**
65 * mext_next_extent - Search for the next extent and set it to "extent"
66 *
67 * @inode: inode which is searched
68 * @path: this will obtain data for the next extent
69 * @extent: pointer to the next extent we have just gotten
70 *
71 * Search the next extent in the array of ext4_ext_path structure (@path)
72 * and set it to ext4_extent structure (@extent). In addition, the member of
73 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
74 * ext4_ext_path structure refers to the last extent, or a negative error
75 * value on failure.
76 */
77static int
78mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
79 struct ext4_extent **extent)
80{
Akira Fujitafc04cb42009-11-23 07:24:43 -050081 struct ext4_extent_header *eh;
Akira Fujita748de672009-06-17 19:24:03 -040082 int ppos, leaf_ppos = path->p_depth;
83
84 ppos = leaf_ppos;
85 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
86 /* leaf block */
87 *extent = ++path[ppos].p_ext;
Akira Fujitafc04cb42009-11-23 07:24:43 -050088 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
Akira Fujita748de672009-06-17 19:24:03 -040089 return 0;
90 }
91
92 while (--ppos >= 0) {
93 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
94 path[ppos].p_idx) {
95 int cur_ppos = ppos;
96
97 /* index block */
98 path[ppos].p_idx++;
99 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
100 if (path[ppos+1].p_bh)
101 brelse(path[ppos+1].p_bh);
102 path[ppos+1].p_bh =
103 sb_bread(inode->i_sb, path[ppos].p_block);
104 if (!path[ppos+1].p_bh)
105 return -EIO;
106 path[ppos+1].p_hdr =
107 ext_block_hdr(path[ppos+1].p_bh);
108
109 /* Halfway index block */
110 while (++cur_ppos < leaf_ppos) {
111 path[cur_ppos].p_idx =
112 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
113 path[cur_ppos].p_block =
114 idx_pblock(path[cur_ppos].p_idx);
115 if (path[cur_ppos+1].p_bh)
116 brelse(path[cur_ppos+1].p_bh);
117 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
118 path[cur_ppos].p_block);
119 if (!path[cur_ppos+1].p_bh)
120 return -EIO;
121 path[cur_ppos+1].p_hdr =
122 ext_block_hdr(path[cur_ppos+1].p_bh);
123 }
124
Akira Fujitafc04cb42009-11-23 07:24:43 -0500125 path[leaf_ppos].p_ext = *extent = NULL;
126
127 eh = path[leaf_ppos].p_hdr;
128 if (le16_to_cpu(eh->eh_entries) == 0)
129 /* empty leaf is found */
130 return -ENODATA;
131
Akira Fujita748de672009-06-17 19:24:03 -0400132 /* leaf block */
133 path[leaf_ppos].p_ext = *extent =
134 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
Akira Fujitafc04cb42009-11-23 07:24:43 -0500135 path[leaf_ppos].p_block =
136 ext_pblock(path[leaf_ppos].p_ext);
Akira Fujita748de672009-06-17 19:24:03 -0400137 return 0;
138 }
139 }
140 /* We found the last extent */
141 return 1;
142}
143
144/**
Akira Fujita2147b1a2009-09-16 13:46:35 -0400145 * mext_check_null_inode - NULL check for two inodes
146 *
147 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
148 */
149static int
150mext_check_null_inode(struct inode *inode1, struct inode *inode2,
151 const char *function)
152{
153 int ret = 0;
154
155 if (inode1 == NULL) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500156 __ext4_error(inode2->i_sb, function,
Akira Fujita2147b1a2009-09-16 13:46:35 -0400157 "Both inodes should not be NULL: "
158 "inode1 NULL inode2 %lu", inode2->i_ino);
159 ret = -EIO;
160 } else if (inode2 == NULL) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500161 __ext4_error(inode1->i_sb, function,
Akira Fujita2147b1a2009-09-16 13:46:35 -0400162 "Both inodes should not be NULL: "
163 "inode1 %lu inode2 NULL", inode1->i_ino);
164 ret = -EIO;
165 }
166 return ret;
167}
168
169/**
Akira Fujitafc04cb42009-11-23 07:24:43 -0500170 * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem
Akira Fujita748de672009-06-17 19:24:03 -0400171 *
172 * @orig_inode: original inode structure
173 * @donor_inode: donor inode structure
Akira Fujitafc04cb42009-11-23 07:24:43 -0500174 * Acquire write lock of i_data_sem of the two inodes (orig and donor) by
175 * i_ino order.
Akira Fujita748de672009-06-17 19:24:03 -0400176 */
177static void
Akira Fujitafc04cb42009-11-23 07:24:43 -0500178double_down_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
Akira Fujita748de672009-06-17 19:24:03 -0400179{
180 struct inode *first = orig_inode, *second = donor_inode;
181
Akira Fujita748de672009-06-17 19:24:03 -0400182 /*
183 * Use the inode number to provide the stable locking order instead
184 * of its address, because the C language doesn't guarantee you can
185 * compare pointers that don't come from the same array.
186 */
187 if (donor_inode->i_ino < orig_inode->i_ino) {
188 first = donor_inode;
189 second = orig_inode;
190 }
191
192 down_write(&EXT4_I(first)->i_data_sem);
Akira Fujita49bd22b2009-11-23 07:24:41 -0500193 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
Akira Fujita748de672009-06-17 19:24:03 -0400194}
195
196/**
Akira Fujitafc04cb42009-11-23 07:24:43 -0500197 * double_up_write_data_sem - Release two inodes' write lock of i_data_sem
Akira Fujita748de672009-06-17 19:24:03 -0400198 *
199 * @orig_inode: original inode structure to be released its lock first
200 * @donor_inode: donor inode structure to be released its lock second
Akira Fujitafc04cb42009-11-23 07:24:43 -0500201 * Release write lock of i_data_sem of two inodes (orig and donor).
Akira Fujita748de672009-06-17 19:24:03 -0400202 */
203static void
Akira Fujitafc04cb42009-11-23 07:24:43 -0500204double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
Akira Fujita748de672009-06-17 19:24:03 -0400205{
Akira Fujita748de672009-06-17 19:24:03 -0400206 up_write(&EXT4_I(orig_inode)->i_data_sem);
207 up_write(&EXT4_I(donor_inode)->i_data_sem);
208}
209
210/**
211 * mext_insert_across_blocks - Insert extents across leaf block
212 *
213 * @handle: journal handle
214 * @orig_inode: original inode
215 * @o_start: first original extent to be changed
216 * @o_end: last original extent to be changed
217 * @start_ext: first new extent to be inserted
218 * @new_ext: middle of new extent to be inserted
219 * @end_ext: last new extent to be inserted
220 *
221 * Allocate a new leaf block and insert extents into it. Return 0 on success,
222 * or a negative error value on failure.
223 */
224static int
225mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
226 struct ext4_extent *o_start, struct ext4_extent *o_end,
227 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
228 struct ext4_extent *end_ext)
229{
230 struct ext4_ext_path *orig_path = NULL;
231 ext4_lblk_t eblock = 0;
232 int new_flag = 0;
233 int end_flag = 0;
234 int err = 0;
235
236 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
237 if (o_start == o_end) {
238
239 /* start_ext new_ext end_ext
240 * donor |---------|-----------|--------|
241 * orig |------------------------------|
242 */
243 end_flag = 1;
244 } else {
245
246 /* start_ext new_ext end_ext
247 * donor |---------|----------|---------|
248 * orig |---------------|--------------|
249 */
250 o_end->ee_block = end_ext->ee_block;
251 o_end->ee_len = end_ext->ee_len;
252 ext4_ext_store_pblock(o_end, ext_pblock(end_ext));
253 }
254
255 o_start->ee_len = start_ext->ee_len;
Akira Fujita5fd52492010-03-04 00:31:06 -0500256 eblock = le32_to_cpu(start_ext->ee_block);
Akira Fujita748de672009-06-17 19:24:03 -0400257 new_flag = 1;
258
259 } else if (start_ext->ee_len && new_ext->ee_len &&
260 !end_ext->ee_len && o_start == o_end) {
261
262 /* start_ext new_ext
263 * donor |--------------|---------------|
264 * orig |------------------------------|
265 */
266 o_start->ee_len = start_ext->ee_len;
Akira Fujita5fd52492010-03-04 00:31:06 -0500267 eblock = le32_to_cpu(start_ext->ee_block);
Akira Fujita748de672009-06-17 19:24:03 -0400268 new_flag = 1;
269
270 } else if (!start_ext->ee_len && new_ext->ee_len &&
271 end_ext->ee_len && o_start == o_end) {
272
273 /* new_ext end_ext
274 * donor |--------------|---------------|
275 * orig |------------------------------|
276 */
277 o_end->ee_block = end_ext->ee_block;
278 o_end->ee_len = end_ext->ee_len;
279 ext4_ext_store_pblock(o_end, ext_pblock(end_ext));
280
281 /*
282 * Set 0 to the extent block if new_ext was
283 * the first block.
284 */
285 if (new_ext->ee_block)
286 eblock = le32_to_cpu(new_ext->ee_block);
287
288 new_flag = 1;
289 } else {
290 ext4_debug("ext4 move extent: Unexpected insert case\n");
291 return -EIO;
292 }
293
294 if (new_flag) {
Akira Fujitae8505972009-09-16 13:46:38 -0400295 err = get_ext_path(orig_inode, eblock, &orig_path);
Akira Fujita347fa6f2009-09-16 14:25:07 -0400296 if (err)
Akira Fujita748de672009-06-17 19:24:03 -0400297 goto out;
298
299 if (ext4_ext_insert_extent(handle, orig_inode,
Mingming Cao00314622009-09-28 15:49:08 -0400300 orig_path, new_ext, 0))
Akira Fujita748de672009-06-17 19:24:03 -0400301 goto out;
302 }
303
304 if (end_flag) {
Akira Fujitae8505972009-09-16 13:46:38 -0400305 err = get_ext_path(orig_inode,
306 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
Akira Fujita347fa6f2009-09-16 14:25:07 -0400307 if (err)
Akira Fujita748de672009-06-17 19:24:03 -0400308 goto out;
309
310 if (ext4_ext_insert_extent(handle, orig_inode,
Mingming Cao00314622009-09-28 15:49:08 -0400311 orig_path, end_ext, 0))
Akira Fujita748de672009-06-17 19:24:03 -0400312 goto out;
313 }
314out:
315 if (orig_path) {
316 ext4_ext_drop_refs(orig_path);
317 kfree(orig_path);
318 }
319
320 return err;
321
322}
323
324/**
325 * mext_insert_inside_block - Insert new extent to the extent block
326 *
327 * @o_start: first original extent to be moved
328 * @o_end: last original extent to be moved
329 * @start_ext: first new extent to be inserted
330 * @new_ext: middle of new extent to be inserted
331 * @end_ext: last new extent to be inserted
332 * @eh: extent header of target leaf block
333 * @range_to_move: used to decide how to insert extent
334 *
335 * Insert extents into the leaf block. The extent (@o_start) is overwritten
336 * by inserted extents.
337 */
338static void
339mext_insert_inside_block(struct ext4_extent *o_start,
340 struct ext4_extent *o_end,
341 struct ext4_extent *start_ext,
342 struct ext4_extent *new_ext,
343 struct ext4_extent *end_ext,
344 struct ext4_extent_header *eh,
345 int range_to_move)
346{
347 int i = 0;
348 unsigned long len;
349
350 /* Move the existing extents */
351 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
352 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
353 (unsigned long)(o_end + 1);
354 memmove(o_end + 1 + range_to_move, o_end + 1, len);
355 }
356
357 /* Insert start entry */
358 if (start_ext->ee_len)
359 o_start[i++].ee_len = start_ext->ee_len;
360
361 /* Insert new entry */
362 if (new_ext->ee_len) {
363 o_start[i] = *new_ext;
364 ext4_ext_store_pblock(&o_start[i++], ext_pblock(new_ext));
365 }
366
367 /* Insert end entry */
368 if (end_ext->ee_len)
369 o_start[i] = *end_ext;
370
371 /* Increment the total entries counter on the extent block */
372 le16_add_cpu(&eh->eh_entries, range_to_move);
373}
374
375/**
376 * mext_insert_extents - Insert new extent
377 *
378 * @handle: journal handle
379 * @orig_inode: original inode
380 * @orig_path: path indicates first extent to be changed
381 * @o_start: first original extent to be changed
382 * @o_end: last original extent to be changed
383 * @start_ext: first new extent to be inserted
384 * @new_ext: middle of new extent to be inserted
385 * @end_ext: last new extent to be inserted
386 *
387 * Call the function to insert extents. If we cannot add more extents into
388 * the leaf block, we call mext_insert_across_blocks() to create a
389 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
390 * on success, or a negative error value on failure.
391 */
392static int
393mext_insert_extents(handle_t *handle, struct inode *orig_inode,
394 struct ext4_ext_path *orig_path,
395 struct ext4_extent *o_start,
396 struct ext4_extent *o_end,
397 struct ext4_extent *start_ext,
398 struct ext4_extent *new_ext,
399 struct ext4_extent *end_ext)
400{
401 struct ext4_extent_header *eh;
402 unsigned long need_slots, slots_range;
403 int range_to_move, depth, ret;
404
405 /*
406 * The extents need to be inserted
407 * start_extent + new_extent + end_extent.
408 */
409 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
410 (new_ext->ee_len ? 1 : 0);
411
412 /* The number of slots between start and end */
413 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
414 / sizeof(struct ext4_extent);
415
416 /* Range to move the end of extent */
417 range_to_move = need_slots - slots_range;
418 depth = orig_path->p_depth;
419 orig_path += depth;
420 eh = orig_path->p_hdr;
421
422 if (depth) {
423 /* Register to journal */
424 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
425 if (ret)
426 return ret;
427 }
428
429 /* Expansion */
430 if (range_to_move > 0 &&
431 (range_to_move > le16_to_cpu(eh->eh_max)
432 - le16_to_cpu(eh->eh_entries))) {
433
434 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
435 o_end, start_ext, new_ext, end_ext);
436 if (ret < 0)
437 return ret;
438 } else
439 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
440 end_ext, eh, range_to_move);
441
442 if (depth) {
443 ret = ext4_handle_dirty_metadata(handle, orig_inode,
444 orig_path->p_bh);
445 if (ret)
446 return ret;
447 } else {
448 ret = ext4_mark_inode_dirty(handle, orig_inode);
449 if (ret < 0)
450 return ret;
451 }
452
453 return 0;
454}
455
456/**
457 * mext_leaf_block - Move one leaf extent block into the inode.
458 *
459 * @handle: journal handle
460 * @orig_inode: original inode
461 * @orig_path: path indicates first extent to be changed
462 * @dext: donor extent
463 * @from: start offset on the target file
464 *
465 * In order to insert extents into the leaf block, we must divide the extent
466 * in the leaf block into three extents. The one is located to be inserted
467 * extents, and the others are located around it.
468 *
469 * Therefore, this function creates structures to save extents of the leaf
470 * block, and inserts extents by calling mext_insert_extents() with
471 * created extents. Return 0 on success, or a negative error value on failure.
472 */
473static int
474mext_leaf_block(handle_t *handle, struct inode *orig_inode,
475 struct ext4_ext_path *orig_path, struct ext4_extent *dext,
476 ext4_lblk_t *from)
477{
478 struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
479 struct ext4_extent new_ext, start_ext, end_ext;
480 ext4_lblk_t new_ext_end;
Akira Fujita748de672009-06-17 19:24:03 -0400481 int oext_alen, new_ext_alen, end_ext_alen;
482 int depth = ext_depth(orig_inode);
483 int ret;
484
Steven Liuc26d0ba2010-05-11 00:00:00 -0400485 start_ext.ee_block = end_ext.ee_block = 0;
Akira Fujita748de672009-06-17 19:24:03 -0400486 o_start = o_end = oext = orig_path[depth].p_ext;
487 oext_alen = ext4_ext_get_actual_len(oext);
488 start_ext.ee_len = end_ext.ee_len = 0;
489
490 new_ext.ee_block = cpu_to_le32(*from);
491 ext4_ext_store_pblock(&new_ext, ext_pblock(dext));
492 new_ext.ee_len = dext->ee_len;
493 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
494 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
Akira Fujita748de672009-06-17 19:24:03 -0400495
496 /*
497 * Case: original extent is first
498 * oext |--------|
499 * new_ext |--|
500 * start_ext |--|
501 */
502 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
503 le32_to_cpu(new_ext.ee_block) <
504 le32_to_cpu(oext->ee_block) + oext_alen) {
505 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
506 le32_to_cpu(oext->ee_block));
Akira Fujita5fd52492010-03-04 00:31:06 -0500507 start_ext.ee_block = oext->ee_block;
Akira Fujita748de672009-06-17 19:24:03 -0400508 copy_extent_status(oext, &start_ext);
509 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
510 prev_ext = oext - 1;
511 /*
512 * We can merge new_ext into previous extent,
513 * if these are contiguous and same extent type.
514 */
515 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
516 &new_ext)) {
517 o_start = prev_ext;
518 start_ext.ee_len = cpu_to_le16(
519 ext4_ext_get_actual_len(prev_ext) +
520 new_ext_alen);
Akira Fujita5fd52492010-03-04 00:31:06 -0500521 start_ext.ee_block = oext->ee_block;
Akira Fujita748de672009-06-17 19:24:03 -0400522 copy_extent_status(prev_ext, &start_ext);
523 new_ext.ee_len = 0;
524 }
525 }
526
527 /*
528 * Case: new_ext_end must be less than oext
529 * oext |-----------|
530 * new_ext |-------|
531 */
Akira Fujita2147b1a2009-09-16 13:46:35 -0400532 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
Theodore Ts'o24676da2010-05-16 21:00:00 -0400533 EXT4_ERROR_INODE(orig_inode,
Akira Fujita2147b1a2009-09-16 13:46:35 -0400534 "new_ext_end(%u) should be less than or equal to "
535 "oext->ee_block(%u) + oext_alen(%d) - 1",
536 new_ext_end, le32_to_cpu(oext->ee_block),
537 oext_alen);
538 ret = -EIO;
539 goto out;
540 }
Akira Fujita748de672009-06-17 19:24:03 -0400541
542 /*
543 * Case: new_ext is smaller than original extent
544 * oext |---------------|
545 * new_ext |-----------|
546 * end_ext |---|
547 */
548 if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
549 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
550 end_ext.ee_len =
551 cpu_to_le16(le32_to_cpu(oext->ee_block) +
552 oext_alen - 1 - new_ext_end);
553 copy_extent_status(oext, &end_ext);
554 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
555 ext4_ext_store_pblock(&end_ext,
556 (ext_pblock(o_end) + oext_alen - end_ext_alen));
557 end_ext.ee_block =
558 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
559 oext_alen - end_ext_alen);
560 }
561
562 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
563 o_end, &start_ext, &new_ext, &end_ext);
Akira Fujita2147b1a2009-09-16 13:46:35 -0400564out:
Akira Fujita748de672009-06-17 19:24:03 -0400565 return ret;
566}
567
568/**
569 * mext_calc_swap_extents - Calculate extents for extent swapping.
570 *
571 * @tmp_dext: the extent that will belong to the original inode
572 * @tmp_oext: the extent that will belong to the donor inode
573 * @orig_off: block offset of original inode
574 * @donor_off: block offset of donor inode
Akira Fujita92c28152009-11-23 07:24:50 -0500575 * @max_count: the maximum length of extents
Akira Fujitac40ce3c2009-09-16 14:25:39 -0400576 *
577 * Return 0 on success, or a negative error value on failure.
Akira Fujita748de672009-06-17 19:24:03 -0400578 */
Akira Fujitac40ce3c2009-09-16 14:25:39 -0400579static int
Akira Fujita748de672009-06-17 19:24:03 -0400580mext_calc_swap_extents(struct ext4_extent *tmp_dext,
581 struct ext4_extent *tmp_oext,
582 ext4_lblk_t orig_off, ext4_lblk_t donor_off,
583 ext4_lblk_t max_count)
584{
585 ext4_lblk_t diff, orig_diff;
586 struct ext4_extent dext_old, oext_old;
587
Akira Fujitac40ce3c2009-09-16 14:25:39 -0400588 BUG_ON(orig_off != donor_off);
589
590 /* original and donor extents have to cover the same block offset */
591 if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
592 le32_to_cpu(tmp_oext->ee_block) +
593 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
594 return -ENODATA;
595
596 if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
597 le32_to_cpu(tmp_dext->ee_block) +
598 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
599 return -ENODATA;
600
Akira Fujita748de672009-06-17 19:24:03 -0400601 dext_old = *tmp_dext;
602 oext_old = *tmp_oext;
603
604 /* When tmp_dext is too large, pick up the target range. */
605 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
606
607 ext4_ext_store_pblock(tmp_dext, ext_pblock(tmp_dext) + diff);
608 tmp_dext->ee_block =
609 cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff);
610 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff);
611
612 if (max_count < ext4_ext_get_actual_len(tmp_dext))
613 tmp_dext->ee_len = cpu_to_le16(max_count);
614
615 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
616 ext4_ext_store_pblock(tmp_oext, ext_pblock(tmp_oext) + orig_diff);
617
618 /* Adjust extent length if donor extent is larger than orig */
619 if (ext4_ext_get_actual_len(tmp_dext) >
620 ext4_ext_get_actual_len(tmp_oext) - orig_diff)
621 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
622 orig_diff);
623
624 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
625
626 copy_extent_status(&oext_old, tmp_dext);
627 copy_extent_status(&dext_old, tmp_oext);
Akira Fujitac40ce3c2009-09-16 14:25:39 -0400628
629 return 0;
Akira Fujita748de672009-06-17 19:24:03 -0400630}
631
632/**
633 * mext_replace_branches - Replace original extents with new extents
634 *
635 * @handle: journal handle
636 * @orig_inode: original inode
637 * @donor_inode: donor inode
638 * @from: block offset of orig_inode
639 * @count: block count to be replaced
Akira Fujitaf868a482009-11-23 07:25:48 -0500640 * @err: pointer to save return value
Akira Fujita748de672009-06-17 19:24:03 -0400641 *
642 * Replace original inode extents and donor inode extents page by page.
643 * We implement this replacement in the following three steps:
644 * 1. Save the block information of original and donor inodes into
645 * dummy extents.
646 * 2. Change the block information of original inode to point at the
647 * donor inode blocks.
648 * 3. Change the block information of donor inode to point at the saved
649 * original inode blocks in the dummy extents.
650 *
Akira Fujitaf868a482009-11-23 07:25:48 -0500651 * Return replaced block count.
Akira Fujita748de672009-06-17 19:24:03 -0400652 */
653static int
654mext_replace_branches(handle_t *handle, struct inode *orig_inode,
655 struct inode *donor_inode, ext4_lblk_t from,
Akira Fujitaf868a482009-11-23 07:25:48 -0500656 ext4_lblk_t count, int *err)
Akira Fujita748de672009-06-17 19:24:03 -0400657{
658 struct ext4_ext_path *orig_path = NULL;
659 struct ext4_ext_path *donor_path = NULL;
660 struct ext4_extent *oext, *dext;
661 struct ext4_extent tmp_dext, tmp_oext;
662 ext4_lblk_t orig_off = from, donor_off = from;
Akira Fujita748de672009-06-17 19:24:03 -0400663 int depth;
664 int replaced_count = 0;
665 int dext_alen;
666
Akira Fujitaac48b0a2009-11-24 10:31:56 -0500667 /* Protect extent trees against block allocations via delalloc */
668 double_down_write_data_sem(orig_inode, donor_inode);
669
Akira Fujita748de672009-06-17 19:24:03 -0400670 /* Get the original extent for the block "orig_off" */
Akira Fujitaf868a482009-11-23 07:25:48 -0500671 *err = get_ext_path(orig_inode, orig_off, &orig_path);
672 if (*err)
Akira Fujita748de672009-06-17 19:24:03 -0400673 goto out;
674
675 /* Get the donor extent for the head */
Akira Fujitaf868a482009-11-23 07:25:48 -0500676 *err = get_ext_path(donor_inode, donor_off, &donor_path);
677 if (*err)
Akira Fujita748de672009-06-17 19:24:03 -0400678 goto out;
679 depth = ext_depth(orig_inode);
680 oext = orig_path[depth].p_ext;
681 tmp_oext = *oext;
682
683 depth = ext_depth(donor_inode);
684 dext = donor_path[depth].p_ext;
685 tmp_dext = *dext;
686
Akira Fujitaf868a482009-11-23 07:25:48 -0500687 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
Akira Fujita748de672009-06-17 19:24:03 -0400688 donor_off, count);
Akira Fujitaf868a482009-11-23 07:25:48 -0500689 if (*err)
Akira Fujitac40ce3c2009-09-16 14:25:39 -0400690 goto out;
Akira Fujita748de672009-06-17 19:24:03 -0400691
692 /* Loop for the donor extents */
693 while (1) {
694 /* The extent for donor must be found. */
Akira Fujita2147b1a2009-09-16 13:46:35 -0400695 if (!dext) {
Theodore Ts'o24676da2010-05-16 21:00:00 -0400696 EXT4_ERROR_INODE(donor_inode,
Akira Fujita2147b1a2009-09-16 13:46:35 -0400697 "The extent for donor must be found");
Akira Fujitaf868a482009-11-23 07:25:48 -0500698 *err = -EIO;
Akira Fujita2147b1a2009-09-16 13:46:35 -0400699 goto out;
700 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -0400701 EXT4_ERROR_INODE(donor_inode,
Akira Fujita2147b1a2009-09-16 13:46:35 -0400702 "Donor offset(%u) and the first block of donor "
703 "extent(%u) should be equal",
704 donor_off,
705 le32_to_cpu(tmp_dext.ee_block));
Akira Fujitaf868a482009-11-23 07:25:48 -0500706 *err = -EIO;
Akira Fujita2147b1a2009-09-16 13:46:35 -0400707 goto out;
708 }
Akira Fujita748de672009-06-17 19:24:03 -0400709
710 /* Set donor extent to orig extent */
Akira Fujitaf868a482009-11-23 07:25:48 -0500711 *err = mext_leaf_block(handle, orig_inode,
Akira Fujita748de672009-06-17 19:24:03 -0400712 orig_path, &tmp_dext, &orig_off);
Akira Fujitaf868a482009-11-23 07:25:48 -0500713 if (*err)
Akira Fujita748de672009-06-17 19:24:03 -0400714 goto out;
715
716 /* Set orig extent to donor extent */
Akira Fujitaf868a482009-11-23 07:25:48 -0500717 *err = mext_leaf_block(handle, donor_inode,
Akira Fujita748de672009-06-17 19:24:03 -0400718 donor_path, &tmp_oext, &donor_off);
Akira Fujitaf868a482009-11-23 07:25:48 -0500719 if (*err)
Akira Fujita748de672009-06-17 19:24:03 -0400720 goto out;
721
722 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
723 replaced_count += dext_alen;
724 donor_off += dext_alen;
725 orig_off += dext_alen;
726
727 /* Already moved the expected blocks */
728 if (replaced_count >= count)
729 break;
730
731 if (orig_path)
732 ext4_ext_drop_refs(orig_path);
Akira Fujitaf868a482009-11-23 07:25:48 -0500733 *err = get_ext_path(orig_inode, orig_off, &orig_path);
734 if (*err)
Akira Fujita748de672009-06-17 19:24:03 -0400735 goto out;
736 depth = ext_depth(orig_inode);
737 oext = orig_path[depth].p_ext;
Akira Fujita748de672009-06-17 19:24:03 -0400738 tmp_oext = *oext;
739
740 if (donor_path)
741 ext4_ext_drop_refs(donor_path);
Akira Fujitaf868a482009-11-23 07:25:48 -0500742 *err = get_ext_path(donor_inode, donor_off, &donor_path);
743 if (*err)
Akira Fujita748de672009-06-17 19:24:03 -0400744 goto out;
745 depth = ext_depth(donor_inode);
746 dext = donor_path[depth].p_ext;
Akira Fujita748de672009-06-17 19:24:03 -0400747 tmp_dext = *dext;
748
Akira Fujitaf868a482009-11-23 07:25:48 -0500749 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
Akira Fujitac40ce3c2009-09-16 14:25:39 -0400750 donor_off, count - replaced_count);
Akira Fujitaf868a482009-11-23 07:25:48 -0500751 if (*err)
Akira Fujitac40ce3c2009-09-16 14:25:39 -0400752 goto out;
Akira Fujita748de672009-06-17 19:24:03 -0400753 }
754
755out:
756 if (orig_path) {
757 ext4_ext_drop_refs(orig_path);
758 kfree(orig_path);
759 }
760 if (donor_path) {
761 ext4_ext_drop_refs(donor_path);
762 kfree(donor_path);
763 }
764
Akira Fujitaac48b0a2009-11-24 10:31:56 -0500765 ext4_ext_invalidate_cache(orig_inode);
766 ext4_ext_invalidate_cache(donor_inode);
767
768 double_up_write_data_sem(orig_inode, donor_inode);
769
Akira Fujitaf868a482009-11-23 07:25:48 -0500770 return replaced_count;
Akira Fujita748de672009-06-17 19:24:03 -0400771}
772
773/**
774 * move_extent_per_page - Move extent data per page
775 *
776 * @o_filp: file structure of original file
777 * @donor_inode: donor inode
778 * @orig_page_offset: page index on original file
779 * @data_offset_in_page: block index where data swapping starts
780 * @block_len_in_page: the number of blocks to be swapped
781 * @uninit: orig extent is uninitialized or not
Akira Fujitaf868a482009-11-23 07:25:48 -0500782 * @err: pointer to save return value
Akira Fujita748de672009-06-17 19:24:03 -0400783 *
784 * Save the data in original inode blocks and replace original inode extents
785 * with donor inode extents by calling mext_replace_branches().
Akira Fujitaf868a482009-11-23 07:25:48 -0500786 * Finally, write out the saved data in new original inode blocks. Return
787 * replaced block count.
Akira Fujita748de672009-06-17 19:24:03 -0400788 */
789static int
Akira Fujita44fc48f2009-09-05 23:12:41 -0400790move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
Akira Fujita748de672009-06-17 19:24:03 -0400791 pgoff_t orig_page_offset, int data_offset_in_page,
Akira Fujitaf868a482009-11-23 07:25:48 -0500792 int block_len_in_page, int uninit, int *err)
Akira Fujita748de672009-06-17 19:24:03 -0400793{
794 struct inode *orig_inode = o_filp->f_dentry->d_inode;
795 struct address_space *mapping = orig_inode->i_mapping;
796 struct buffer_head *bh;
797 struct page *page = NULL;
798 const struct address_space_operations *a_ops = mapping->a_ops;
799 handle_t *handle;
800 ext4_lblk_t orig_blk_offset;
801 long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
802 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
803 unsigned int w_flags = 0;
Akira Fujitaf868a482009-11-23 07:25:48 -0500804 unsigned int tmp_data_size, data_size, replaced_size;
Akira Fujita748de672009-06-17 19:24:03 -0400805 void *fsdata;
Akira Fujitaf868a482009-11-23 07:25:48 -0500806 int i, jblocks;
807 int err2 = 0;
808 int replaced_count = 0;
Akira Fujita748de672009-06-17 19:24:03 -0400809 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
810
811 /*
812 * It needs twice the amount of ordinary journal buffers because
813 * inode and donor_inode may change each different metadata blocks.
814 */
815 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
816 handle = ext4_journal_start(orig_inode, jblocks);
817 if (IS_ERR(handle)) {
Akira Fujitaf868a482009-11-23 07:25:48 -0500818 *err = PTR_ERR(handle);
819 return 0;
Akira Fujita748de672009-06-17 19:24:03 -0400820 }
821
822 if (segment_eq(get_fs(), KERNEL_DS))
823 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
824
825 orig_blk_offset = orig_page_offset * blocks_per_page +
826 data_offset_in_page;
827
828 /*
829 * If orig extent is uninitialized one,
830 * it's not necessary force the page into memory
831 * and then force it to be written out again.
832 * Just swap data blocks between orig and donor.
833 */
834 if (uninit) {
Akira Fujitaf868a482009-11-23 07:25:48 -0500835 replaced_count = mext_replace_branches(handle, orig_inode,
836 donor_inode, orig_blk_offset,
837 block_len_in_page, err);
Akira Fujita748de672009-06-17 19:24:03 -0400838 goto out2;
839 }
840
841 offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
842
Akira Fujitaf868a482009-11-23 07:25:48 -0500843 /* Calculate data_size */
Akira Fujita748de672009-06-17 19:24:03 -0400844 if ((orig_blk_offset + block_len_in_page - 1) ==
845 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
846 /* Replace the last block */
Akira Fujitaf868a482009-11-23 07:25:48 -0500847 tmp_data_size = orig_inode->i_size & (blocksize - 1);
Akira Fujita748de672009-06-17 19:24:03 -0400848 /*
Akira Fujitaf868a482009-11-23 07:25:48 -0500849 * If data_size equal zero, it shows data_size is multiples of
Akira Fujita748de672009-06-17 19:24:03 -0400850 * blocksize. So we set appropriate value.
851 */
Akira Fujitaf868a482009-11-23 07:25:48 -0500852 if (tmp_data_size == 0)
853 tmp_data_size = blocksize;
Akira Fujita748de672009-06-17 19:24:03 -0400854
Akira Fujitaf868a482009-11-23 07:25:48 -0500855 data_size = tmp_data_size +
Akira Fujita748de672009-06-17 19:24:03 -0400856 ((block_len_in_page - 1) << orig_inode->i_blkbits);
Akira Fujitaf868a482009-11-23 07:25:48 -0500857 } else
858 data_size = block_len_in_page << orig_inode->i_blkbits;
Akira Fujita748de672009-06-17 19:24:03 -0400859
Akira Fujitaf868a482009-11-23 07:25:48 -0500860 replaced_size = data_size;
861
862 *err = a_ops->write_begin(o_filp, mapping, offs, data_size, w_flags,
Akira Fujita748de672009-06-17 19:24:03 -0400863 &page, &fsdata);
Akira Fujitaf868a482009-11-23 07:25:48 -0500864 if (unlikely(*err < 0))
Akira Fujita748de672009-06-17 19:24:03 -0400865 goto out;
866
867 if (!PageUptodate(page)) {
868 mapping->a_ops->readpage(o_filp, page);
869 lock_page(page);
870 }
871
872 /*
873 * try_to_release_page() doesn't call releasepage in writeback mode.
874 * We should care about the order of writing to the same file
875 * by multiple move extent processes.
876 * It needs to call wait_on_page_writeback() to wait for the
877 * writeback of the page.
878 */
879 if (PageWriteback(page))
880 wait_on_page_writeback(page);
881
882 /* Release old bh and drop refs */
883 try_to_release_page(page, 0);
884
Akira Fujitaf868a482009-11-23 07:25:48 -0500885 replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
886 orig_blk_offset, block_len_in_page,
887 &err2);
888 if (err2) {
889 if (replaced_count) {
890 block_len_in_page = replaced_count;
891 replaced_size =
892 block_len_in_page << orig_inode->i_blkbits;
Akira Fujitaac48b0a2009-11-24 10:31:56 -0500893 } else
Akira Fujitaf868a482009-11-23 07:25:48 -0500894 goto out;
895 }
Akira Fujita748de672009-06-17 19:24:03 -0400896
Akira Fujita748de672009-06-17 19:24:03 -0400897 if (!page_has_buffers(page))
898 create_empty_buffers(page, 1 << orig_inode->i_blkbits, 0);
899
900 bh = page_buffers(page);
901 for (i = 0; i < data_offset_in_page; i++)
902 bh = bh->b_this_page;
903
904 for (i = 0; i < block_len_in_page; i++) {
Akira Fujitaf868a482009-11-23 07:25:48 -0500905 *err = ext4_get_block(orig_inode,
Akira Fujita748de672009-06-17 19:24:03 -0400906 (sector_t)(orig_blk_offset + i), bh, 0);
Akira Fujitaf868a482009-11-23 07:25:48 -0500907 if (*err < 0)
Akira Fujita748de672009-06-17 19:24:03 -0400908 goto out;
909
910 if (bh->b_this_page != NULL)
911 bh = bh->b_this_page;
912 }
913
Akira Fujitaf868a482009-11-23 07:25:48 -0500914 *err = a_ops->write_end(o_filp, mapping, offs, data_size, replaced_size,
Akira Fujita748de672009-06-17 19:24:03 -0400915 page, fsdata);
916 page = NULL;
917
918out:
919 if (unlikely(page)) {
920 if (PageLocked(page))
921 unlock_page(page);
922 page_cache_release(page);
Peng Tao91cc2192009-08-10 23:05:28 -0400923 ext4_journal_stop(handle);
Akira Fujita748de672009-06-17 19:24:03 -0400924 }
925out2:
926 ext4_journal_stop(handle);
927
Akira Fujitaf868a482009-11-23 07:25:48 -0500928 if (err2)
929 *err = err2;
930
931 return replaced_count;
Akira Fujita748de672009-06-17 19:24:03 -0400932}
933
934/**
Akira Fujitac437b272010-03-04 00:39:24 -0500935 * mext_check_arguments - Check whether move extent can be done
Akira Fujita748de672009-06-17 19:24:03 -0400936 *
937 * @orig_inode: original inode
938 * @donor_inode: donor inode
939 * @orig_start: logical start offset in block for orig
940 * @donor_start: logical start offset in block for donor
941 * @len: the number of blocks to be moved
Akira Fujita748de672009-06-17 19:24:03 -0400942 *
943 * Check the arguments of ext4_move_extents() whether the files can be
944 * exchanged with each other.
945 * Return 0 on success, or a negative error value on failure.
946 */
947static int
948mext_check_arguments(struct inode *orig_inode,
Kazuya Mio446aaa62009-11-24 10:28:48 -0500949 struct inode *donor_inode, __u64 orig_start,
950 __u64 donor_start, __u64 *len)
Akira Fujita748de672009-06-17 19:24:03 -0400951{
Akira Fujita70d5d3d2009-09-16 14:28:22 -0400952 ext4_lblk_t orig_blocks, donor_blocks;
953 unsigned int blkbits = orig_inode->i_blkbits;
954 unsigned int blocksize = 1 << blkbits;
955
Akira Fujita4a585792009-12-06 23:38:31 -0500956 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
957 ext4_debug("ext4 move extent: suid or sgid is set"
958 " to donor file [ino:orig %lu, donor %lu]\n",
959 orig_inode->i_ino, donor_inode->i_ino);
960 return -EINVAL;
961 }
962
Akira Fujita748de672009-06-17 19:24:03 -0400963 /* Ext4 move extent does not support swapfile */
964 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
965 ext4_debug("ext4 move extent: The argument files should "
966 "not be swapfile [ino:orig %lu, donor %lu]\n",
967 orig_inode->i_ino, donor_inode->i_ino);
968 return -EINVAL;
969 }
970
971 /* Files should be in the same ext4 FS */
972 if (orig_inode->i_sb != donor_inode->i_sb) {
973 ext4_debug("ext4 move extent: The argument files "
974 "should be in same FS [ino:orig %lu, donor %lu]\n",
975 orig_inode->i_ino, donor_inode->i_ino);
976 return -EINVAL;
977 }
978
Akira Fujita748de672009-06-17 19:24:03 -0400979 /* Ext4 move extent supports only extent based file */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400980 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
Akira Fujita748de672009-06-17 19:24:03 -0400981 ext4_debug("ext4 move extent: orig file is not extents "
982 "based file [ino:orig %lu]\n", orig_inode->i_ino);
983 return -EOPNOTSUPP;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400984 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
Akira Fujita748de672009-06-17 19:24:03 -0400985 ext4_debug("ext4 move extent: donor file is not extents "
986 "based file [ino:donor %lu]\n", donor_inode->i_ino);
987 return -EOPNOTSUPP;
988 }
989
990 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
991 ext4_debug("ext4 move extent: File size is 0 byte\n");
992 return -EINVAL;
993 }
994
995 /* Start offset should be same */
996 if (orig_start != donor_start) {
997 ext4_debug("ext4 move extent: orig and donor's start "
998 "offset are not same [ino:orig %lu, donor %lu]\n",
999 orig_inode->i_ino, donor_inode->i_ino);
1000 return -EINVAL;
1001 }
1002
Eric Sandeen0a80e982009-09-17 11:55:58 -04001003 if ((orig_start > EXT_MAX_BLOCK) ||
1004 (donor_start > EXT_MAX_BLOCK) ||
1005 (*len > EXT_MAX_BLOCK) ||
1006 (orig_start + *len > EXT_MAX_BLOCK)) {
1007 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1008 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCK,
Akira Fujita748de672009-06-17 19:24:03 -04001009 orig_inode->i_ino, donor_inode->i_ino);
1010 return -EINVAL;
1011 }
1012
1013 if (orig_inode->i_size > donor_inode->i_size) {
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001014 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1015 /* TODO: eliminate this artificial restriction */
1016 if (orig_start >= donor_blocks) {
Akira Fujita748de672009-06-17 19:24:03 -04001017 ext4_debug("ext4 move extent: orig start offset "
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001018 "[%llu] should be less than donor file blocks "
1019 "[%u] [ino:orig %lu, donor %lu]\n",
1020 orig_start, donor_blocks,
Akira Fujita748de672009-06-17 19:24:03 -04001021 orig_inode->i_ino, donor_inode->i_ino);
1022 return -EINVAL;
1023 }
1024
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001025 /* TODO: eliminate this artificial restriction */
1026 if (orig_start + *len > donor_blocks) {
Akira Fujita748de672009-06-17 19:24:03 -04001027 ext4_debug("ext4 move extent: End offset [%llu] should "
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001028 "be less than donor file blocks [%u]."
1029 "So adjust length from %llu to %llu "
Akira Fujita748de672009-06-17 19:24:03 -04001030 "[ino:orig %lu, donor %lu]\n",
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001031 orig_start + *len, donor_blocks,
1032 *len, donor_blocks - orig_start,
Akira Fujita748de672009-06-17 19:24:03 -04001033 orig_inode->i_ino, donor_inode->i_ino);
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001034 *len = donor_blocks - orig_start;
Akira Fujita748de672009-06-17 19:24:03 -04001035 }
1036 } else {
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001037 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1038 if (orig_start >= orig_blocks) {
Akira Fujita748de672009-06-17 19:24:03 -04001039 ext4_debug("ext4 move extent: start offset [%llu] "
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001040 "should be less than original file blocks "
1041 "[%u] [ino:orig %lu, donor %lu]\n",
1042 orig_start, orig_blocks,
Akira Fujita748de672009-06-17 19:24:03 -04001043 orig_inode->i_ino, donor_inode->i_ino);
1044 return -EINVAL;
1045 }
1046
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001047 if (orig_start + *len > orig_blocks) {
Akira Fujita748de672009-06-17 19:24:03 -04001048 ext4_debug("ext4 move extent: Adjust length "
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001049 "from %llu to %llu. Because it should be "
1050 "less than original file blocks "
Akira Fujita748de672009-06-17 19:24:03 -04001051 "[ino:orig %lu, donor %lu]\n",
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001052 *len, orig_blocks - orig_start,
Akira Fujita748de672009-06-17 19:24:03 -04001053 orig_inode->i_ino, donor_inode->i_ino);
Akira Fujita70d5d3d2009-09-16 14:28:22 -04001054 *len = orig_blocks - orig_start;
Akira Fujita748de672009-06-17 19:24:03 -04001055 }
1056 }
1057
1058 if (!*len) {
Akira Fujita92c28152009-11-23 07:24:50 -05001059 ext4_debug("ext4 move extent: len should not be 0 "
Akira Fujita748de672009-06-17 19:24:03 -04001060 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1061 donor_inode->i_ino);
1062 return -EINVAL;
1063 }
1064
1065 return 0;
1066}
1067
1068/**
1069 * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1070 *
1071 * @inode1: the inode structure
1072 * @inode2: the inode structure
1073 *
Akira Fujita2147b1a2009-09-16 13:46:35 -04001074 * Lock two inodes' i_mutex by i_ino order.
1075 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
Akira Fujita748de672009-06-17 19:24:03 -04001076 */
Akira Fujita2147b1a2009-09-16 13:46:35 -04001077static int
Akira Fujita748de672009-06-17 19:24:03 -04001078mext_inode_double_lock(struct inode *inode1, struct inode *inode2)
1079{
Akira Fujita2147b1a2009-09-16 13:46:35 -04001080 int ret = 0;
1081
1082 BUG_ON(inode1 == NULL && inode2 == NULL);
1083
1084 ret = mext_check_null_inode(inode1, inode2, __func__);
1085 if (ret < 0)
1086 goto out;
1087
1088 if (inode1 == inode2) {
1089 mutex_lock(&inode1->i_mutex);
1090 goto out;
Akira Fujita748de672009-06-17 19:24:03 -04001091 }
1092
1093 if (inode1->i_ino < inode2->i_ino) {
1094 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1095 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1096 } else {
1097 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1098 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1099 }
Akira Fujita2147b1a2009-09-16 13:46:35 -04001100
1101out:
1102 return ret;
Akira Fujita748de672009-06-17 19:24:03 -04001103}
1104
1105/**
1106 * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1107 *
1108 * @inode1: the inode that is released first
1109 * @inode2: the inode that is released second
1110 *
Akira Fujita2147b1a2009-09-16 13:46:35 -04001111 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
Akira Fujita748de672009-06-17 19:24:03 -04001112 */
1113
Akira Fujita2147b1a2009-09-16 13:46:35 -04001114static int
Akira Fujita748de672009-06-17 19:24:03 -04001115mext_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1116{
Akira Fujita2147b1a2009-09-16 13:46:35 -04001117 int ret = 0;
1118
1119 BUG_ON(inode1 == NULL && inode2 == NULL);
1120
1121 ret = mext_check_null_inode(inode1, inode2, __func__);
1122 if (ret < 0)
1123 goto out;
1124
Akira Fujita748de672009-06-17 19:24:03 -04001125 if (inode1)
1126 mutex_unlock(&inode1->i_mutex);
1127
1128 if (inode2 && inode2 != inode1)
1129 mutex_unlock(&inode2->i_mutex);
Akira Fujita2147b1a2009-09-16 13:46:35 -04001130
1131out:
1132 return ret;
Akira Fujita748de672009-06-17 19:24:03 -04001133}
1134
1135/**
1136 * ext4_move_extents - Exchange the specified range of a file
1137 *
1138 * @o_filp: file structure of the original file
1139 * @d_filp: file structure of the donor file
1140 * @orig_start: start offset in block for orig
1141 * @donor_start: start offset in block for donor
1142 * @len: the number of blocks to be moved
1143 * @moved_len: moved block length
1144 *
1145 * This function returns 0 and moved block length is set in moved_len
1146 * if succeed, otherwise returns error value.
1147 *
1148 * Note: ext4_move_extents() proceeds the following order.
1149 * 1:ext4_move_extents() calculates the last block number of moving extent
1150 * function by the start block number (orig_start) and the number of blocks
1151 * to be moved (len) specified as arguments.
1152 * If the {orig, donor}_start points a hole, the extent's start offset
1153 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1154 * after hole behind.
1155 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1156 * or the ext_cur exceeds the block_end which is last logical block number.
1157 * 3:To get the length of continues area, call mext_next_extent()
1158 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1159 * until find un-continuous extent, the start logical block number exceeds
1160 * the block_end or the extent points to the last extent.
1161 * 4:Exchange the original inode data with donor inode data
1162 * from orig_page_offset to seq_end_page.
1163 * The start indexes of data are specified as arguments.
1164 * That of the original inode is orig_page_offset,
1165 * and the donor inode is also orig_page_offset
1166 * (To easily handle blocksize != pagesize case, the offset for the
1167 * donor inode is block unit).
1168 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1169 * then returns to step 2.
1170 * 6:Release holecheck_path, orig_path and set the len to moved_len
1171 * which shows the number of moved blocks.
1172 * The moved_len is useful for the command to calculate the file offset
1173 * for starting next move extent ioctl.
1174 * 7:Return 0 on success, or a negative error value on failure.
1175 */
1176int
1177ext4_move_extents(struct file *o_filp, struct file *d_filp,
1178 __u64 orig_start, __u64 donor_start, __u64 len,
1179 __u64 *moved_len)
1180{
1181 struct inode *orig_inode = o_filp->f_dentry->d_inode;
1182 struct inode *donor_inode = d_filp->f_dentry->d_inode;
1183 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1184 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1185 ext4_lblk_t block_start = orig_start;
1186 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1187 ext4_lblk_t rest_blocks;
1188 pgoff_t orig_page_offset = 0, seq_end_page;
Akira Fujita2147b1a2009-09-16 13:46:35 -04001189 int ret1, ret2, depth, last_extent = 0;
Akira Fujita748de672009-06-17 19:24:03 -04001190 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1191 int data_offset_in_page;
1192 int block_len_in_page;
1193 int uninit;
1194
Theodore Ts'of3ce8062009-09-28 15:58:29 -04001195 /* orig and donor should be different file */
1196 if (orig_inode->i_ino == donor_inode->i_ino) {
1197 ext4_debug("ext4 move extent: The argument files should not "
1198 "be same file [ino:orig %lu, donor %lu]\n",
1199 orig_inode->i_ino, donor_inode->i_ino);
1200 return -EINVAL;
1201 }
1202
Akira Fujita7247c0c2010-03-04 00:34:58 -05001203 /* Regular file check */
1204 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1205 ext4_debug("ext4 move extent: The argument files should be "
1206 "regular file [ino:orig %lu, donor %lu]\n",
1207 orig_inode->i_ino, donor_inode->i_ino);
1208 return -EINVAL;
1209 }
1210
Akira Fujitafc04cb42009-11-23 07:24:43 -05001211 /* Protect orig and donor inodes against a truncate */
Akira Fujita2147b1a2009-09-16 13:46:35 -04001212 ret1 = mext_inode_double_lock(orig_inode, donor_inode);
1213 if (ret1 < 0)
1214 return ret1;
Akira Fujita748de672009-06-17 19:24:03 -04001215
Akira Fujitafc04cb42009-11-23 07:24:43 -05001216 /* Protect extent tree against block allocations via delalloc */
1217 double_down_write_data_sem(orig_inode, donor_inode);
Akira Fujita748de672009-06-17 19:24:03 -04001218 /* Check the filesystem environment whether move_extent can be done */
Akira Fujita2147b1a2009-09-16 13:46:35 -04001219 ret1 = mext_check_arguments(orig_inode, donor_inode, orig_start,
Kazuya Mio446aaa62009-11-24 10:28:48 -05001220 donor_start, &len);
Akira Fujita2147b1a2009-09-16 13:46:35 -04001221 if (ret1)
Akira Fujita347fa6f2009-09-16 14:25:07 -04001222 goto out;
Akira Fujita748de672009-06-17 19:24:03 -04001223
1224 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1225 block_end = block_start + len - 1;
1226 if (file_end < block_end)
1227 len -= block_end - file_end;
1228
Akira Fujita2147b1a2009-09-16 13:46:35 -04001229 ret1 = get_ext_path(orig_inode, block_start, &orig_path);
Akira Fujita347fa6f2009-09-16 14:25:07 -04001230 if (ret1)
1231 goto out;
Akira Fujita748de672009-06-17 19:24:03 -04001232
1233 /* Get path structure to check the hole */
Akira Fujita2147b1a2009-09-16 13:46:35 -04001234 ret1 = get_ext_path(orig_inode, block_start, &holecheck_path);
Akira Fujita347fa6f2009-09-16 14:25:07 -04001235 if (ret1)
Akira Fujita748de672009-06-17 19:24:03 -04001236 goto out;
1237
1238 depth = ext_depth(orig_inode);
1239 ext_cur = holecheck_path[depth].p_ext;
Akira Fujita748de672009-06-17 19:24:03 -04001240
1241 /*
Akira Fujitac40ce3c2009-09-16 14:25:39 -04001242 * Get proper starting location of block replacement if block_start was
1243 * within the hole.
Akira Fujita748de672009-06-17 19:24:03 -04001244 */
1245 if (le32_to_cpu(ext_cur->ee_block) +
1246 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
Akira Fujitac40ce3c2009-09-16 14:25:39 -04001247 /*
1248 * The hole exists between extents or the tail of
1249 * original file.
1250 */
Akira Fujita748de672009-06-17 19:24:03 -04001251 last_extent = mext_next_extent(orig_inode,
1252 holecheck_path, &ext_cur);
1253 if (last_extent < 0) {
Akira Fujita2147b1a2009-09-16 13:46:35 -04001254 ret1 = last_extent;
Akira Fujita748de672009-06-17 19:24:03 -04001255 goto out;
1256 }
1257 last_extent = mext_next_extent(orig_inode, orig_path,
1258 &ext_dummy);
1259 if (last_extent < 0) {
Akira Fujita2147b1a2009-09-16 13:46:35 -04001260 ret1 = last_extent;
Akira Fujita748de672009-06-17 19:24:03 -04001261 goto out;
1262 }
Akira Fujitac40ce3c2009-09-16 14:25:39 -04001263 seq_start = le32_to_cpu(ext_cur->ee_block);
1264 } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1265 /* The hole exists at the beginning of original file. */
1266 seq_start = le32_to_cpu(ext_cur->ee_block);
1267 else
1268 seq_start = block_start;
Akira Fujita748de672009-06-17 19:24:03 -04001269
1270 /* No blocks within the specified range. */
1271 if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1272 ext4_debug("ext4 move extent: The specified range of file "
1273 "may be the hole\n");
Akira Fujita2147b1a2009-09-16 13:46:35 -04001274 ret1 = -EINVAL;
Akira Fujita748de672009-06-17 19:24:03 -04001275 goto out;
1276 }
1277
1278 /* Adjust start blocks */
1279 add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1280 ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1281 max(le32_to_cpu(ext_cur->ee_block), block_start);
1282
1283 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1284 seq_blocks += add_blocks;
1285
1286 /* Adjust tail blocks */
1287 if (seq_start + seq_blocks - 1 > block_end)
1288 seq_blocks = block_end - seq_start + 1;
1289
1290 ext_prev = ext_cur;
1291 last_extent = mext_next_extent(orig_inode, holecheck_path,
1292 &ext_cur);
1293 if (last_extent < 0) {
Akira Fujita2147b1a2009-09-16 13:46:35 -04001294 ret1 = last_extent;
Akira Fujita748de672009-06-17 19:24:03 -04001295 break;
1296 }
1297 add_blocks = ext4_ext_get_actual_len(ext_cur);
1298
1299 /*
1300 * Extend the length of contiguous block (seq_blocks)
1301 * if extents are contiguous.
1302 */
1303 if (ext4_can_extents_be_merged(orig_inode,
1304 ext_prev, ext_cur) &&
1305 block_end >= le32_to_cpu(ext_cur->ee_block) &&
1306 !last_extent)
1307 continue;
1308
1309 /* Is original extent is uninitialized */
1310 uninit = ext4_ext_is_uninitialized(ext_prev);
1311
1312 data_offset_in_page = seq_start % blocks_per_page;
1313
1314 /*
1315 * Calculate data blocks count that should be swapped
1316 * at the first page.
1317 */
1318 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1319 /* Swapped blocks are across pages */
1320 block_len_in_page =
1321 blocks_per_page - data_offset_in_page;
1322 } else {
1323 /* Swapped blocks are in a page */
1324 block_len_in_page = seq_blocks;
1325 }
1326
1327 orig_page_offset = seq_start >>
1328 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1329 seq_end_page = (seq_start + seq_blocks - 1) >>
1330 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1331 seq_start = le32_to_cpu(ext_cur->ee_block);
1332 rest_blocks = seq_blocks;
1333
Akira Fujitafc04cb42009-11-23 07:24:43 -05001334 /*
1335 * Up semaphore to avoid following problems:
1336 * a. transaction deadlock among ext4_journal_start,
1337 * ->write_begin via pagefault, and jbd2_journal_commit
1338 * b. racing with ->readpage, ->write_begin, and ext4_get_block
1339 * in move_extent_per_page
1340 */
1341 double_up_write_data_sem(orig_inode, donor_inode);
Akira Fujita748de672009-06-17 19:24:03 -04001342
1343 while (orig_page_offset <= seq_end_page) {
1344
1345 /* Swap original branches with new branches */
Akira Fujitaf868a482009-11-23 07:25:48 -05001346 block_len_in_page = move_extent_per_page(
1347 o_filp, donor_inode,
Akira Fujita748de672009-06-17 19:24:03 -04001348 orig_page_offset,
1349 data_offset_in_page,
Akira Fujitaf868a482009-11-23 07:25:48 -05001350 block_len_in_page, uninit,
1351 &ret1);
1352
Akira Fujita748de672009-06-17 19:24:03 -04001353 /* Count how many blocks we have exchanged */
1354 *moved_len += block_len_in_page;
Akira Fujitaf868a482009-11-23 07:25:48 -05001355 if (ret1 < 0)
Akira Fujitafc04cb42009-11-23 07:24:43 -05001356 break;
Akira Fujita2147b1a2009-09-16 13:46:35 -04001357 if (*moved_len > len) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001358 EXT4_ERROR_INODE(orig_inode,
Akira Fujita2147b1a2009-09-16 13:46:35 -04001359 "We replaced blocks too much! "
1360 "sum of replaced: %llu requested: %llu",
1361 *moved_len, len);
1362 ret1 = -EIO;
Akira Fujitafc04cb42009-11-23 07:24:43 -05001363 break;
Akira Fujita2147b1a2009-09-16 13:46:35 -04001364 }
Akira Fujita748de672009-06-17 19:24:03 -04001365
Akira Fujitaf868a482009-11-23 07:25:48 -05001366 orig_page_offset++;
Akira Fujita748de672009-06-17 19:24:03 -04001367 data_offset_in_page = 0;
1368 rest_blocks -= block_len_in_page;
1369 if (rest_blocks > blocks_per_page)
1370 block_len_in_page = blocks_per_page;
1371 else
1372 block_len_in_page = rest_blocks;
1373 }
1374
Akira Fujitafc04cb42009-11-23 07:24:43 -05001375 double_down_write_data_sem(orig_inode, donor_inode);
1376 if (ret1 < 0)
1377 break;
1378
Akira Fujita748de672009-06-17 19:24:03 -04001379 /* Decrease buffer counter */
1380 if (holecheck_path)
1381 ext4_ext_drop_refs(holecheck_path);
Akira Fujita2147b1a2009-09-16 13:46:35 -04001382 ret1 = get_ext_path(orig_inode, seq_start, &holecheck_path);
Akira Fujita347fa6f2009-09-16 14:25:07 -04001383 if (ret1)
Akira Fujita748de672009-06-17 19:24:03 -04001384 break;
1385 depth = holecheck_path->p_depth;
1386
1387 /* Decrease buffer counter */
1388 if (orig_path)
1389 ext4_ext_drop_refs(orig_path);
Akira Fujita2147b1a2009-09-16 13:46:35 -04001390 ret1 = get_ext_path(orig_inode, seq_start, &orig_path);
Akira Fujita347fa6f2009-09-16 14:25:07 -04001391 if (ret1)
Akira Fujita748de672009-06-17 19:24:03 -04001392 break;
1393
1394 ext_cur = holecheck_path[depth].p_ext;
1395 add_blocks = ext4_ext_get_actual_len(ext_cur);
1396 seq_blocks = 0;
1397
1398 }
1399out:
Akira Fujita94d7c162009-11-24 10:19:57 -05001400 if (*moved_len) {
1401 ext4_discard_preallocations(orig_inode);
1402 ext4_discard_preallocations(donor_inode);
1403 }
1404
Akira Fujita748de672009-06-17 19:24:03 -04001405 if (orig_path) {
1406 ext4_ext_drop_refs(orig_path);
1407 kfree(orig_path);
1408 }
1409 if (holecheck_path) {
1410 ext4_ext_drop_refs(holecheck_path);
1411 kfree(holecheck_path);
1412 }
Akira Fujitafc04cb42009-11-23 07:24:43 -05001413 double_up_write_data_sem(orig_inode, donor_inode);
Akira Fujita2147b1a2009-09-16 13:46:35 -04001414 ret2 = mext_inode_double_unlock(orig_inode, donor_inode);
Akira Fujita748de672009-06-17 19:24:03 -04001415
Akira Fujita2147b1a2009-09-16 13:46:35 -04001416 if (ret1)
1417 return ret1;
1418 else if (ret2)
1419 return ret2;
Akira Fujita748de672009-06-17 19:24:03 -04001420
Akira Fujita748de672009-06-17 19:24:03 -04001421 return 0;
1422}