]>
Commit | Line | Data |
---|---|---|
748de673 AF |
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> | |
5a0e3ad6 | 18 | #include <linux/slab.h> |
748de673 AF |
19 | #include "ext4_jbd2.h" |
20 | #include "ext4_extents.h" | |
21 | #include "ext4.h" | |
22 | ||
e8505970 AF |
23 | /** |
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 | */ | |
33 | static inline int | |
34 | get_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; | |
347fa6f1 AF |
43 | } else if ((*path)[ext_depth(inode)].p_ext == NULL) |
44 | ret = -ENODATA; | |
45 | ||
e8505970 AF |
46 | return ret; |
47 | } | |
748de673 AF |
48 | |
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 | */ | |
55 | static void | |
56 | copy_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 | */ | |
77 | static int | |
78 | mext_next_extent(struct inode *inode, struct ext4_ext_path *path, | |
79 | struct ext4_extent **extent) | |
80 | { | |
fc04cb49 | 81 | struct ext4_extent_header *eh; |
748de673 AF |
82 | 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; | |
fc04cb49 | 88 | path[ppos].p_block = ext_pblock(path[ppos].p_ext); |
748de673 AF |
89 | 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 | ||
fc04cb49 AF |
125 | 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 | ||
748de673 AF |
132 | /* leaf block */ |
133 | path[leaf_ppos].p_ext = *extent = | |
134 | EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr); | |
fc04cb49 AF |
135 | path[leaf_ppos].p_block = |
136 | ext_pblock(path[leaf_ppos].p_ext); | |
748de673 AF |
137 | return 0; |
138 | } | |
139 | } | |
140 | /* We found the last extent */ | |
141 | return 1; | |
142 | } | |
143 | ||
2147b1a6 AF |
144 | /** |
145 | * mext_check_null_inode - NULL check for two inodes | |
146 | * | |
147 | * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0. | |
148 | */ | |
149 | static int | |
150 | mext_check_null_inode(struct inode *inode1, struct inode *inode2, | |
c398eda0 | 151 | const char *function, unsigned int line) |
2147b1a6 AF |
152 | { |
153 | int ret = 0; | |
154 | ||
155 | if (inode1 == NULL) { | |
c398eda0 | 156 | __ext4_error(inode2->i_sb, function, line, |
2147b1a6 AF |
157 | "Both inodes should not be NULL: " |
158 | "inode1 NULL inode2 %lu", inode2->i_ino); | |
159 | ret = -EIO; | |
160 | } else if (inode2 == NULL) { | |
c398eda0 | 161 | __ext4_error(inode1->i_sb, function, line, |
2147b1a6 AF |
162 | "Both inodes should not be NULL: " |
163 | "inode1 %lu inode2 NULL", inode1->i_ino); | |
164 | ret = -EIO; | |
165 | } | |
166 | return ret; | |
167 | } | |
168 | ||
748de673 | 169 | /** |
fc04cb49 | 170 | * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem |
748de673 AF |
171 | * |
172 | * @orig_inode: original inode structure | |
173 | * @donor_inode: donor inode structure | |
fc04cb49 AF |
174 | * Acquire write lock of i_data_sem of the two inodes (orig and donor) by |
175 | * i_ino order. | |
748de673 AF |
176 | */ |
177 | static void | |
fc04cb49 | 178 | double_down_write_data_sem(struct inode *orig_inode, struct inode *donor_inode) |
748de673 AF |
179 | { |
180 | struct inode *first = orig_inode, *second = donor_inode; | |
181 | ||
748de673 AF |
182 | /* |
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); | |
49bd22bc | 193 | down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING); |
748de673 AF |
194 | } |
195 | ||
196 | /** | |
fc04cb49 | 197 | * double_up_write_data_sem - Release two inodes' write lock of i_data_sem |
748de673 AF |
198 | * |
199 | * @orig_inode: original inode structure to be released its lock first | |
200 | * @donor_inode: donor inode structure to be released its lock second | |
fc04cb49 | 201 | * Release write lock of i_data_sem of two inodes (orig and donor). |
748de673 AF |
202 | */ |
203 | static void | |
fc04cb49 | 204 | double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode) |
748de673 | 205 | { |
748de673 AF |
206 | 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 | */ | |
224 | static int | |
225 | mext_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; | |
5fd5249a | 256 | eblock = le32_to_cpu(start_ext->ee_block); |
748de673 AF |
257 | 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; | |
5fd5249a | 267 | eblock = le32_to_cpu(start_ext->ee_block); |
748de673 AF |
268 | 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) { | |
e8505970 | 295 | err = get_ext_path(orig_inode, eblock, &orig_path); |
347fa6f1 | 296 | if (err) |
748de673 AF |
297 | goto out; |
298 | ||
299 | if (ext4_ext_insert_extent(handle, orig_inode, | |
0031462b | 300 | orig_path, new_ext, 0)) |
748de673 AF |
301 | goto out; |
302 | } | |
303 | ||
304 | if (end_flag) { | |
e8505970 AF |
305 | err = get_ext_path(orig_inode, |
306 | le32_to_cpu(end_ext->ee_block) - 1, &orig_path); | |
347fa6f1 | 307 | if (err) |
748de673 AF |
308 | goto out; |
309 | ||
310 | if (ext4_ext_insert_extent(handle, orig_inode, | |
0031462b | 311 | orig_path, end_ext, 0)) |
748de673 AF |
312 | goto out; |
313 | } | |
314 | out: | |
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 | */ | |
338 | static void | |
339 | mext_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 | */ | |
392 | static int | |
393 | mext_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 | */ | |
473 | static int | |
474 | mext_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; | |
748de673 AF |
481 | int oext_alen, new_ext_alen, end_ext_alen; |
482 | int depth = ext_depth(orig_inode); | |
483 | int ret; | |
484 | ||
c26d0bad | 485 | start_ext.ee_block = end_ext.ee_block = 0; |
748de673 AF |
486 | 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; | |
748de673 AF |
495 | |
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)); | |
5fd5249a | 507 | start_ext.ee_block = oext->ee_block; |
748de673 AF |
508 | 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); | |
5fd5249a | 521 | start_ext.ee_block = oext->ee_block; |
748de673 AF |
522 | 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 | */ | |
2147b1a6 | 532 | if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) { |
24676da4 | 533 | EXT4_ERROR_INODE(orig_inode, |
2147b1a6 AF |
534 | "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 | } | |
748de673 AF |
541 | |
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); | |
2147b1a6 | 564 | out: |
748de673 AF |
565 | 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 | |
92c28159 | 575 | * @max_count: the maximum length of extents |
c40ce3c9 AF |
576 | * |
577 | * Return 0 on success, or a negative error value on failure. | |
748de673 | 578 | */ |
c40ce3c9 | 579 | static int |
748de673 AF |
580 | mext_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 | ||
c40ce3c9 AF |
588 | 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 | ||
748de673 AF |
601 | 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); | |
c40ce3c9 AF |
628 | |
629 | return 0; | |
748de673 AF |
630 | } |
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 | |
f868a48d | 640 | * @err: pointer to save return value |
748de673 AF |
641 | * |
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 | * | |
f868a48d | 651 | * Return replaced block count. |
748de673 AF |
652 | */ |
653 | static int | |
654 | mext_replace_branches(handle_t *handle, struct inode *orig_inode, | |
655 | struct inode *donor_inode, ext4_lblk_t from, | |
f868a48d | 656 | ext4_lblk_t count, int *err) |
748de673 AF |
657 | { |
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; | |
748de673 AF |
663 | int depth; |
664 | int replaced_count = 0; | |
665 | int dext_alen; | |
666 | ||
ac48b0a1 AF |
667 | /* Protect extent trees against block allocations via delalloc */ |
668 | double_down_write_data_sem(orig_inode, donor_inode); | |
669 | ||
748de673 | 670 | /* Get the original extent for the block "orig_off" */ |
f868a48d AF |
671 | *err = get_ext_path(orig_inode, orig_off, &orig_path); |
672 | if (*err) | |
748de673 AF |
673 | goto out; |
674 | ||
675 | /* Get the donor extent for the head */ | |
f868a48d AF |
676 | *err = get_ext_path(donor_inode, donor_off, &donor_path); |
677 | if (*err) | |
748de673 AF |
678 | 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 | ||
f868a48d | 687 | *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, |
748de673 | 688 | donor_off, count); |
f868a48d | 689 | if (*err) |
c40ce3c9 | 690 | goto out; |
748de673 AF |
691 | |
692 | /* Loop for the donor extents */ | |
693 | while (1) { | |
694 | /* The extent for donor must be found. */ | |
2147b1a6 | 695 | if (!dext) { |
24676da4 | 696 | EXT4_ERROR_INODE(donor_inode, |
2147b1a6 | 697 | "The extent for donor must be found"); |
f868a48d | 698 | *err = -EIO; |
2147b1a6 AF |
699 | goto out; |
700 | } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) { | |
24676da4 | 701 | EXT4_ERROR_INODE(donor_inode, |
2147b1a6 AF |
702 | "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)); | |
f868a48d | 706 | *err = -EIO; |
2147b1a6 AF |
707 | goto out; |
708 | } | |
748de673 AF |
709 | |
710 | /* Set donor extent to orig extent */ | |
f868a48d | 711 | *err = mext_leaf_block(handle, orig_inode, |
748de673 | 712 | orig_path, &tmp_dext, &orig_off); |
f868a48d | 713 | if (*err) |
748de673 AF |
714 | goto out; |
715 | ||
716 | /* Set orig extent to donor extent */ | |
f868a48d | 717 | *err = mext_leaf_block(handle, donor_inode, |
748de673 | 718 | donor_path, &tmp_oext, &donor_off); |
f868a48d | 719 | if (*err) |
748de673 AF |
720 | 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); | |
f868a48d AF |
733 | *err = get_ext_path(orig_inode, orig_off, &orig_path); |
734 | if (*err) | |
748de673 AF |
735 | goto out; |
736 | depth = ext_depth(orig_inode); | |
737 | oext = orig_path[depth].p_ext; | |
748de673 AF |
738 | tmp_oext = *oext; |
739 | ||
740 | if (donor_path) | |
741 | ext4_ext_drop_refs(donor_path); | |
f868a48d AF |
742 | *err = get_ext_path(donor_inode, donor_off, &donor_path); |
743 | if (*err) | |
748de673 AF |
744 | goto out; |
745 | depth = ext_depth(donor_inode); | |
746 | dext = donor_path[depth].p_ext; | |
748de673 AF |
747 | tmp_dext = *dext; |
748 | ||
f868a48d | 749 | *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, |
c40ce3c9 | 750 | donor_off, count - replaced_count); |
f868a48d | 751 | if (*err) |
c40ce3c9 | 752 | goto out; |
748de673 AF |
753 | } |
754 | ||
755 | out: | |
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 | ||
ac48b0a1 AF |
765 | 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 | ||
f868a48d | 770 | return replaced_count; |
748de673 AF |
771 | } |
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 | |
f868a48d | 782 | * @err: pointer to save return value |
748de673 AF |
783 | * |
784 | * Save the data in original inode blocks and replace original inode extents | |
785 | * with donor inode extents by calling mext_replace_branches(). | |
f868a48d AF |
786 | * Finally, write out the saved data in new original inode blocks. Return |
787 | * replaced block count. | |
748de673 AF |
788 | */ |
789 | static int | |
44fc48f7 | 790 | move_extent_per_page(struct file *o_filp, struct inode *donor_inode, |
748de673 | 791 | pgoff_t orig_page_offset, int data_offset_in_page, |
f868a48d | 792 | int block_len_in_page, int uninit, int *err) |
748de673 AF |
793 | { |
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; | |
f868a48d | 804 | unsigned int tmp_data_size, data_size, replaced_size; |
748de673 | 805 | void *fsdata; |
f868a48d AF |
806 | int i, jblocks; |
807 | int err2 = 0; | |
808 | int replaced_count = 0; | |
748de673 AF |
809 | 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)) { | |
f868a48d AF |
818 | *err = PTR_ERR(handle); |
819 | return 0; | |
748de673 AF |
820 | } |
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) { | |
f868a48d AF |
835 | replaced_count = mext_replace_branches(handle, orig_inode, |
836 | donor_inode, orig_blk_offset, | |
837 | block_len_in_page, err); | |
748de673 AF |
838 | goto out2; |
839 | } | |
840 | ||
841 | offs = (long long)orig_blk_offset << orig_inode->i_blkbits; | |
842 | ||
f868a48d | 843 | /* Calculate data_size */ |
748de673 AF |
844 | if ((orig_blk_offset + block_len_in_page - 1) == |
845 | ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { | |
846 | /* Replace the last block */ | |
f868a48d | 847 | tmp_data_size = orig_inode->i_size & (blocksize - 1); |
748de673 | 848 | /* |
f868a48d | 849 | * If data_size equal zero, it shows data_size is multiples of |
748de673 AF |
850 | * blocksize. So we set appropriate value. |
851 | */ | |
f868a48d AF |
852 | if (tmp_data_size == 0) |
853 | tmp_data_size = blocksize; | |
748de673 | 854 | |
f868a48d | 855 | data_size = tmp_data_size + |
748de673 | 856 | ((block_len_in_page - 1) << orig_inode->i_blkbits); |
f868a48d AF |
857 | } else |
858 | data_size = block_len_in_page << orig_inode->i_blkbits; | |
859 | ||
860 | replaced_size = data_size; | |
748de673 | 861 | |
f868a48d | 862 | *err = a_ops->write_begin(o_filp, mapping, offs, data_size, w_flags, |
748de673 | 863 | &page, &fsdata); |
f868a48d | 864 | if (unlikely(*err < 0)) |
748de673 AF |
865 | 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 | ||
f868a48d AF |
885 | 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; | |
ac48b0a1 | 893 | } else |
f868a48d AF |
894 | goto out; |
895 | } | |
748de673 | 896 | |
748de673 AF |
897 | 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++) { | |
f868a48d | 905 | *err = ext4_get_block(orig_inode, |
748de673 | 906 | (sector_t)(orig_blk_offset + i), bh, 0); |
f868a48d | 907 | if (*err < 0) |
748de673 AF |
908 | goto out; |
909 | ||
910 | if (bh->b_this_page != NULL) | |
911 | bh = bh->b_this_page; | |
912 | } | |
913 | ||
f868a48d | 914 | *err = a_ops->write_end(o_filp, mapping, offs, data_size, replaced_size, |
748de673 AF |
915 | page, fsdata); |
916 | page = NULL; | |
917 | ||
918 | out: | |
919 | if (unlikely(page)) { | |
920 | if (PageLocked(page)) | |
921 | unlock_page(page); | |
922 | page_cache_release(page); | |
91cc219a | 923 | ext4_journal_stop(handle); |
748de673 AF |
924 | } |
925 | out2: | |
926 | ext4_journal_stop(handle); | |
927 | ||
f868a48d AF |
928 | if (err2) |
929 | *err = err2; | |
930 | ||
931 | return replaced_count; | |
748de673 AF |
932 | } |
933 | ||
934 | /** | |
c437b273 | 935 | * mext_check_arguments - Check whether move extent can be done |
748de673 AF |
936 | * |
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 | |
748de673 AF |
942 | * |
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 | */ | |
947 | static int | |
948 | mext_check_arguments(struct inode *orig_inode, | |
446aaa6e KM |
949 | struct inode *donor_inode, __u64 orig_start, |
950 | __u64 donor_start, __u64 *len) | |
748de673 | 951 | { |
70d5d3dc AF |
952 | ext4_lblk_t orig_blocks, donor_blocks; |
953 | unsigned int blkbits = orig_inode->i_blkbits; | |
954 | unsigned int blocksize = 1 << blkbits; | |
955 | ||
4a58579b AF |
956 | 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 | ||
1f5a81e4 TT |
963 | if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) |
964 | return -EPERM; | |
965 | ||
748de673 AF |
966 | /* Ext4 move extent does not support swapfile */ |
967 | if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { | |
968 | ext4_debug("ext4 move extent: The argument files should " | |
969 | "not be swapfile [ino:orig %lu, donor %lu]\n", | |
970 | orig_inode->i_ino, donor_inode->i_ino); | |
971 | return -EINVAL; | |
972 | } | |
973 | ||
974 | /* Files should be in the same ext4 FS */ | |
975 | if (orig_inode->i_sb != donor_inode->i_sb) { | |
976 | ext4_debug("ext4 move extent: The argument files " | |
977 | "should be in same FS [ino:orig %lu, donor %lu]\n", | |
978 | orig_inode->i_ino, donor_inode->i_ino); | |
979 | return -EINVAL; | |
980 | } | |
981 | ||
748de673 | 982 | /* Ext4 move extent supports only extent based file */ |
12e9b892 | 983 | if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) { |
748de673 AF |
984 | ext4_debug("ext4 move extent: orig file is not extents " |
985 | "based file [ino:orig %lu]\n", orig_inode->i_ino); | |
986 | return -EOPNOTSUPP; | |
12e9b892 | 987 | } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { |
748de673 AF |
988 | ext4_debug("ext4 move extent: donor file is not extents " |
989 | "based file [ino:donor %lu]\n", donor_inode->i_ino); | |
990 | return -EOPNOTSUPP; | |
991 | } | |
992 | ||
993 | if ((!orig_inode->i_size) || (!donor_inode->i_size)) { | |
994 | ext4_debug("ext4 move extent: File size is 0 byte\n"); | |
995 | return -EINVAL; | |
996 | } | |
997 | ||
998 | /* Start offset should be same */ | |
999 | if (orig_start != donor_start) { | |
1000 | ext4_debug("ext4 move extent: orig and donor's start " | |
1001 | "offset are not same [ino:orig %lu, donor %lu]\n", | |
1002 | orig_inode->i_ino, donor_inode->i_ino); | |
1003 | return -EINVAL; | |
1004 | } | |
1005 | ||
0a80e986 ES |
1006 | if ((orig_start > EXT_MAX_BLOCK) || |
1007 | (donor_start > EXT_MAX_BLOCK) || | |
1008 | (*len > EXT_MAX_BLOCK) || | |
1009 | (orig_start + *len > EXT_MAX_BLOCK)) { | |
1010 | ext4_debug("ext4 move extent: Can't handle over [%u] blocks " | |
1011 | "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCK, | |
748de673 AF |
1012 | orig_inode->i_ino, donor_inode->i_ino); |
1013 | return -EINVAL; | |
1014 | } | |
1015 | ||
1016 | if (orig_inode->i_size > donor_inode->i_size) { | |
70d5d3dc AF |
1017 | donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits; |
1018 | /* TODO: eliminate this artificial restriction */ | |
1019 | if (orig_start >= donor_blocks) { | |
748de673 | 1020 | ext4_debug("ext4 move extent: orig start offset " |
70d5d3dc AF |
1021 | "[%llu] should be less than donor file blocks " |
1022 | "[%u] [ino:orig %lu, donor %lu]\n", | |
1023 | orig_start, donor_blocks, | |
748de673 AF |
1024 | orig_inode->i_ino, donor_inode->i_ino); |
1025 | return -EINVAL; | |
1026 | } | |
1027 | ||
70d5d3dc AF |
1028 | /* TODO: eliminate this artificial restriction */ |
1029 | if (orig_start + *len > donor_blocks) { | |
748de673 | 1030 | ext4_debug("ext4 move extent: End offset [%llu] should " |
70d5d3dc AF |
1031 | "be less than donor file blocks [%u]." |
1032 | "So adjust length from %llu to %llu " | |
748de673 | 1033 | "[ino:orig %lu, donor %lu]\n", |
70d5d3dc AF |
1034 | orig_start + *len, donor_blocks, |
1035 | *len, donor_blocks - orig_start, | |
748de673 | 1036 | orig_inode->i_ino, donor_inode->i_ino); |
70d5d3dc | 1037 | *len = donor_blocks - orig_start; |
748de673 AF |
1038 | } |
1039 | } else { | |
70d5d3dc AF |
1040 | orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits; |
1041 | if (orig_start >= orig_blocks) { | |
748de673 | 1042 | ext4_debug("ext4 move extent: start offset [%llu] " |
70d5d3dc AF |
1043 | "should be less than original file blocks " |
1044 | "[%u] [ino:orig %lu, donor %lu]\n", | |
1045 | orig_start, orig_blocks, | |
748de673 AF |
1046 | orig_inode->i_ino, donor_inode->i_ino); |
1047 | return -EINVAL; | |
1048 | } | |
1049 | ||
70d5d3dc | 1050 | if (orig_start + *len > orig_blocks) { |
748de673 | 1051 | ext4_debug("ext4 move extent: Adjust length " |
70d5d3dc AF |
1052 | "from %llu to %llu. Because it should be " |
1053 | "less than original file blocks " | |
748de673 | 1054 | "[ino:orig %lu, donor %lu]\n", |
70d5d3dc | 1055 | *len, orig_blocks - orig_start, |
748de673 | 1056 | orig_inode->i_ino, donor_inode->i_ino); |
70d5d3dc | 1057 | *len = orig_blocks - orig_start; |
748de673 AF |
1058 | } |
1059 | } | |
1060 | ||
1061 | if (!*len) { | |
92c28159 | 1062 | ext4_debug("ext4 move extent: len should not be 0 " |
748de673 AF |
1063 | "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, |
1064 | donor_inode->i_ino); | |
1065 | return -EINVAL; | |
1066 | } | |
1067 | ||
1068 | return 0; | |
1069 | } | |
1070 | ||
1071 | /** | |
1072 | * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2 | |
1073 | * | |
1074 | * @inode1: the inode structure | |
1075 | * @inode2: the inode structure | |
1076 | * | |
2147b1a6 AF |
1077 | * Lock two inodes' i_mutex by i_ino order. |
1078 | * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0. | |
748de673 | 1079 | */ |
2147b1a6 | 1080 | static int |
748de673 AF |
1081 | mext_inode_double_lock(struct inode *inode1, struct inode *inode2) |
1082 | { | |
2147b1a6 AF |
1083 | int ret = 0; |
1084 | ||
1085 | BUG_ON(inode1 == NULL && inode2 == NULL); | |
1086 | ||
c398eda0 | 1087 | ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__); |
2147b1a6 AF |
1088 | if (ret < 0) |
1089 | goto out; | |
1090 | ||
1091 | if (inode1 == inode2) { | |
1092 | mutex_lock(&inode1->i_mutex); | |
1093 | goto out; | |
748de673 AF |
1094 | } |
1095 | ||
1096 | if (inode1->i_ino < inode2->i_ino) { | |
1097 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT); | |
1098 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD); | |
1099 | } else { | |
1100 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT); | |
1101 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD); | |
1102 | } | |
2147b1a6 AF |
1103 | |
1104 | out: | |
1105 | return ret; | |
748de673 AF |
1106 | } |
1107 | ||
1108 | /** | |
1109 | * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2 | |
1110 | * | |
1111 | * @inode1: the inode that is released first | |
1112 | * @inode2: the inode that is released second | |
1113 | * | |
2147b1a6 | 1114 | * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0. |
748de673 AF |
1115 | */ |
1116 | ||
2147b1a6 | 1117 | static int |
748de673 AF |
1118 | mext_inode_double_unlock(struct inode *inode1, struct inode *inode2) |
1119 | { | |
2147b1a6 AF |
1120 | int ret = 0; |
1121 | ||
1122 | BUG_ON(inode1 == NULL && inode2 == NULL); | |
1123 | ||
c398eda0 | 1124 | ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__); |
2147b1a6 AF |
1125 | if (ret < 0) |
1126 | goto out; | |
1127 | ||
748de673 AF |
1128 | if (inode1) |
1129 | mutex_unlock(&inode1->i_mutex); | |
1130 | ||
1131 | if (inode2 && inode2 != inode1) | |
1132 | mutex_unlock(&inode2->i_mutex); | |
2147b1a6 AF |
1133 | |
1134 | out: | |
1135 | return ret; | |
748de673 AF |
1136 | } |
1137 | ||
1138 | /** | |
1139 | * ext4_move_extents - Exchange the specified range of a file | |
1140 | * | |
1141 | * @o_filp: file structure of the original file | |
1142 | * @d_filp: file structure of the donor file | |
1143 | * @orig_start: start offset in block for orig | |
1144 | * @donor_start: start offset in block for donor | |
1145 | * @len: the number of blocks to be moved | |
1146 | * @moved_len: moved block length | |
1147 | * | |
1148 | * This function returns 0 and moved block length is set in moved_len | |
1149 | * if succeed, otherwise returns error value. | |
1150 | * | |
1151 | * Note: ext4_move_extents() proceeds the following order. | |
1152 | * 1:ext4_move_extents() calculates the last block number of moving extent | |
1153 | * function by the start block number (orig_start) and the number of blocks | |
1154 | * to be moved (len) specified as arguments. | |
1155 | * If the {orig, donor}_start points a hole, the extent's start offset | |
1156 | * pointed by ext_cur (current extent), holecheck_path, orig_path are set | |
1157 | * after hole behind. | |
1158 | * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent | |
1159 | * or the ext_cur exceeds the block_end which is last logical block number. | |
1160 | * 3:To get the length of continues area, call mext_next_extent() | |
1161 | * specified with the ext_cur (initial value is holecheck_path) re-cursive, | |
1162 | * until find un-continuous extent, the start logical block number exceeds | |
1163 | * the block_end or the extent points to the last extent. | |
1164 | * 4:Exchange the original inode data with donor inode data | |
1165 | * from orig_page_offset to seq_end_page. | |
1166 | * The start indexes of data are specified as arguments. | |
1167 | * That of the original inode is orig_page_offset, | |
1168 | * and the donor inode is also orig_page_offset | |
1169 | * (To easily handle blocksize != pagesize case, the offset for the | |
1170 | * donor inode is block unit). | |
1171 | * 5:Update holecheck_path and orig_path to points a next proceeding extent, | |
1172 | * then returns to step 2. | |
1173 | * 6:Release holecheck_path, orig_path and set the len to moved_len | |
1174 | * which shows the number of moved blocks. | |
1175 | * The moved_len is useful for the command to calculate the file offset | |
1176 | * for starting next move extent ioctl. | |
1177 | * 7:Return 0 on success, or a negative error value on failure. | |
1178 | */ | |
1179 | int | |
1180 | ext4_move_extents(struct file *o_filp, struct file *d_filp, | |
1181 | __u64 orig_start, __u64 donor_start, __u64 len, | |
1182 | __u64 *moved_len) | |
1183 | { | |
1184 | struct inode *orig_inode = o_filp->f_dentry->d_inode; | |
1185 | struct inode *donor_inode = d_filp->f_dentry->d_inode; | |
1186 | struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL; | |
1187 | struct ext4_extent *ext_prev, *ext_cur, *ext_dummy; | |
1188 | ext4_lblk_t block_start = orig_start; | |
1189 | ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0; | |
1190 | ext4_lblk_t rest_blocks; | |
1191 | pgoff_t orig_page_offset = 0, seq_end_page; | |
2147b1a6 | 1192 | int ret1, ret2, depth, last_extent = 0; |
748de673 AF |
1193 | int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; |
1194 | int data_offset_in_page; | |
1195 | int block_len_in_page; | |
1196 | int uninit; | |
1197 | ||
f3ce8064 TT |
1198 | /* orig and donor should be different file */ |
1199 | if (orig_inode->i_ino == donor_inode->i_ino) { | |
1200 | ext4_debug("ext4 move extent: The argument files should not " | |
1201 | "be same file [ino:orig %lu, donor %lu]\n", | |
1202 | orig_inode->i_ino, donor_inode->i_ino); | |
1203 | return -EINVAL; | |
1204 | } | |
1205 | ||
7247c0ca AF |
1206 | /* Regular file check */ |
1207 | if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { | |
1208 | ext4_debug("ext4 move extent: The argument files should be " | |
1209 | "regular file [ino:orig %lu, donor %lu]\n", | |
1210 | orig_inode->i_ino, donor_inode->i_ino); | |
1211 | return -EINVAL; | |
1212 | } | |
1213 | ||
fc04cb49 | 1214 | /* Protect orig and donor inodes against a truncate */ |
2147b1a6 AF |
1215 | ret1 = mext_inode_double_lock(orig_inode, donor_inode); |
1216 | if (ret1 < 0) | |
1217 | return ret1; | |
748de673 | 1218 | |
fc04cb49 AF |
1219 | /* Protect extent tree against block allocations via delalloc */ |
1220 | double_down_write_data_sem(orig_inode, donor_inode); | |
748de673 | 1221 | /* Check the filesystem environment whether move_extent can be done */ |
2147b1a6 | 1222 | ret1 = mext_check_arguments(orig_inode, donor_inode, orig_start, |
446aaa6e | 1223 | donor_start, &len); |
2147b1a6 | 1224 | if (ret1) |
347fa6f1 | 1225 | goto out; |
748de673 AF |
1226 | |
1227 | file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits; | |
1228 | block_end = block_start + len - 1; | |
1229 | if (file_end < block_end) | |
1230 | len -= block_end - file_end; | |
1231 | ||
2147b1a6 | 1232 | ret1 = get_ext_path(orig_inode, block_start, &orig_path); |
347fa6f1 AF |
1233 | if (ret1) |
1234 | goto out; | |
748de673 AF |
1235 | |
1236 | /* Get path structure to check the hole */ | |
2147b1a6 | 1237 | ret1 = get_ext_path(orig_inode, block_start, &holecheck_path); |
347fa6f1 | 1238 | if (ret1) |
748de673 AF |
1239 | goto out; |
1240 | ||
1241 | depth = ext_depth(orig_inode); | |
1242 | ext_cur = holecheck_path[depth].p_ext; | |
748de673 AF |
1243 | |
1244 | /* | |
c40ce3c9 AF |
1245 | * Get proper starting location of block replacement if block_start was |
1246 | * within the hole. | |
748de673 AF |
1247 | */ |
1248 | if (le32_to_cpu(ext_cur->ee_block) + | |
1249 | ext4_ext_get_actual_len(ext_cur) - 1 < block_start) { | |
c40ce3c9 AF |
1250 | /* |
1251 | * The hole exists between extents or the tail of | |
1252 | * original file. | |
1253 | */ | |
748de673 AF |
1254 | last_extent = mext_next_extent(orig_inode, |
1255 | holecheck_path, &ext_cur); | |
1256 | if (last_extent < 0) { | |
2147b1a6 | 1257 | ret1 = last_extent; |
748de673 AF |
1258 | goto out; |
1259 | } | |
1260 | last_extent = mext_next_extent(orig_inode, orig_path, | |
1261 | &ext_dummy); | |
1262 | if (last_extent < 0) { | |
2147b1a6 | 1263 | ret1 = last_extent; |
748de673 AF |
1264 | goto out; |
1265 | } | |
c40ce3c9 AF |
1266 | seq_start = le32_to_cpu(ext_cur->ee_block); |
1267 | } else if (le32_to_cpu(ext_cur->ee_block) > block_start) | |
1268 | /* The hole exists at the beginning of original file. */ | |
1269 | seq_start = le32_to_cpu(ext_cur->ee_block); | |
1270 | else | |
1271 | seq_start = block_start; | |
748de673 AF |
1272 | |
1273 | /* No blocks within the specified range. */ | |
1274 | if (le32_to_cpu(ext_cur->ee_block) > block_end) { | |
1275 | ext4_debug("ext4 move extent: The specified range of file " | |
1276 | "may be the hole\n"); | |
2147b1a6 | 1277 | ret1 = -EINVAL; |
748de673 AF |
1278 | goto out; |
1279 | } | |
1280 | ||
1281 | /* Adjust start blocks */ | |
1282 | add_blocks = min(le32_to_cpu(ext_cur->ee_block) + | |
1283 | ext4_ext_get_actual_len(ext_cur), block_end + 1) - | |
1284 | max(le32_to_cpu(ext_cur->ee_block), block_start); | |
1285 | ||
1286 | while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) { | |
1287 | seq_blocks += add_blocks; | |
1288 | ||
1289 | /* Adjust tail blocks */ | |
1290 | if (seq_start + seq_blocks - 1 > block_end) | |
1291 | seq_blocks = block_end - seq_start + 1; | |
1292 | ||
1293 | ext_prev = ext_cur; | |
1294 | last_extent = mext_next_extent(orig_inode, holecheck_path, | |
1295 | &ext_cur); | |
1296 | if (last_extent < 0) { | |
2147b1a6 | 1297 | ret1 = last_extent; |
748de673 AF |
1298 | break; |
1299 | } | |
1300 | add_blocks = ext4_ext_get_actual_len(ext_cur); | |
1301 | ||
1302 | /* | |
1303 | * Extend the length of contiguous block (seq_blocks) | |
1304 | * if extents are contiguous. | |
1305 | */ | |
1306 | if (ext4_can_extents_be_merged(orig_inode, | |
1307 | ext_prev, ext_cur) && | |
1308 | block_end >= le32_to_cpu(ext_cur->ee_block) && | |
1309 | !last_extent) | |
1310 | continue; | |
1311 | ||
1312 | /* Is original extent is uninitialized */ | |
1313 | uninit = ext4_ext_is_uninitialized(ext_prev); | |
1314 | ||
1315 | data_offset_in_page = seq_start % blocks_per_page; | |
1316 | ||
1317 | /* | |
1318 | * Calculate data blocks count that should be swapped | |
1319 | * at the first page. | |
1320 | */ | |
1321 | if (data_offset_in_page + seq_blocks > blocks_per_page) { | |
1322 | /* Swapped blocks are across pages */ | |
1323 | block_len_in_page = | |
1324 | blocks_per_page - data_offset_in_page; | |
1325 | } else { | |
1326 | /* Swapped blocks are in a page */ | |
1327 | block_len_in_page = seq_blocks; | |
1328 | } | |
1329 | ||
1330 | orig_page_offset = seq_start >> | |
1331 | (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); | |
1332 | seq_end_page = (seq_start + seq_blocks - 1) >> | |
1333 | (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); | |
1334 | seq_start = le32_to_cpu(ext_cur->ee_block); | |
1335 | rest_blocks = seq_blocks; | |
1336 | ||
fc04cb49 AF |
1337 | /* |
1338 | * Up semaphore to avoid following problems: | |
1339 | * a. transaction deadlock among ext4_journal_start, | |
1340 | * ->write_begin via pagefault, and jbd2_journal_commit | |
1341 | * b. racing with ->readpage, ->write_begin, and ext4_get_block | |
1342 | * in move_extent_per_page | |
1343 | */ | |
1344 | double_up_write_data_sem(orig_inode, donor_inode); | |
748de673 AF |
1345 | |
1346 | while (orig_page_offset <= seq_end_page) { | |
1347 | ||
1348 | /* Swap original branches with new branches */ | |
f868a48d AF |
1349 | block_len_in_page = move_extent_per_page( |
1350 | o_filp, donor_inode, | |
748de673 AF |
1351 | orig_page_offset, |
1352 | data_offset_in_page, | |
f868a48d AF |
1353 | block_len_in_page, uninit, |
1354 | &ret1); | |
1355 | ||
748de673 AF |
1356 | /* Count how many blocks we have exchanged */ |
1357 | *moved_len += block_len_in_page; | |
f868a48d | 1358 | if (ret1 < 0) |
fc04cb49 | 1359 | break; |
2147b1a6 | 1360 | if (*moved_len > len) { |
24676da4 | 1361 | EXT4_ERROR_INODE(orig_inode, |
2147b1a6 AF |
1362 | "We replaced blocks too much! " |
1363 | "sum of replaced: %llu requested: %llu", | |
1364 | *moved_len, len); | |
1365 | ret1 = -EIO; | |
fc04cb49 | 1366 | break; |
2147b1a6 | 1367 | } |
748de673 | 1368 | |
f868a48d | 1369 | orig_page_offset++; |
748de673 AF |
1370 | data_offset_in_page = 0; |
1371 | rest_blocks -= block_len_in_page; | |
1372 | if (rest_blocks > blocks_per_page) | |
1373 | block_len_in_page = blocks_per_page; | |
1374 | else | |
1375 | block_len_in_page = rest_blocks; | |
1376 | } | |
1377 | ||
fc04cb49 AF |
1378 | double_down_write_data_sem(orig_inode, donor_inode); |
1379 | if (ret1 < 0) | |
1380 | break; | |
1381 | ||
748de673 AF |
1382 | /* Decrease buffer counter */ |
1383 | if (holecheck_path) | |
1384 | ext4_ext_drop_refs(holecheck_path); | |
2147b1a6 | 1385 | ret1 = get_ext_path(orig_inode, seq_start, &holecheck_path); |
347fa6f1 | 1386 | if (ret1) |
748de673 AF |
1387 | break; |
1388 | depth = holecheck_path->p_depth; | |
1389 | ||
1390 | /* Decrease buffer counter */ | |
1391 | if (orig_path) | |
1392 | ext4_ext_drop_refs(orig_path); | |
2147b1a6 | 1393 | ret1 = get_ext_path(orig_inode, seq_start, &orig_path); |
347fa6f1 | 1394 | if (ret1) |
748de673 AF |
1395 | break; |
1396 | ||
1397 | ext_cur = holecheck_path[depth].p_ext; | |
1398 | add_blocks = ext4_ext_get_actual_len(ext_cur); | |
1399 | seq_blocks = 0; | |
1400 | ||
1401 | } | |
1402 | out: | |
94d7c16c AF |
1403 | if (*moved_len) { |
1404 | ext4_discard_preallocations(orig_inode); | |
1405 | ext4_discard_preallocations(donor_inode); | |
1406 | } | |
1407 | ||
748de673 AF |
1408 | if (orig_path) { |
1409 | ext4_ext_drop_refs(orig_path); | |
1410 | kfree(orig_path); | |
1411 | } | |
1412 | if (holecheck_path) { | |
1413 | ext4_ext_drop_refs(holecheck_path); | |
1414 | kfree(holecheck_path); | |
1415 | } | |
fc04cb49 | 1416 | double_up_write_data_sem(orig_inode, donor_inode); |
2147b1a6 | 1417 | ret2 = mext_inode_double_unlock(orig_inode, donor_inode); |
748de673 | 1418 | |
2147b1a6 AF |
1419 | if (ret1) |
1420 | return ret1; | |
1421 | else if (ret2) | |
1422 | return ret2; | |
748de673 | 1423 | |
748de673 AF |
1424 | return 0; |
1425 | } |