]>
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> | |
18 | #include "ext4_jbd2.h" | |
19 | #include "ext4_extents.h" | |
20 | #include "ext4.h" | |
21 | ||
e8505970 AF |
22 | /** |
23 | * get_ext_path - Find an extent path for designated logical block number. | |
24 | * | |
25 | * @inode: an inode which is searched | |
26 | * @lblock: logical block number to find an extent path | |
27 | * @path: pointer to an extent path pointer (for output) | |
28 | * | |
29 | * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value | |
30 | * on failure. | |
31 | */ | |
32 | static inline int | |
33 | get_ext_path(struct inode *inode, ext4_lblk_t lblock, | |
34 | struct ext4_ext_path **path) | |
35 | { | |
36 | int ret = 0; | |
37 | ||
38 | *path = ext4_ext_find_extent(inode, lblock, *path); | |
39 | if (IS_ERR(*path)) { | |
40 | ret = PTR_ERR(*path); | |
41 | *path = NULL; | |
347fa6f1 AF |
42 | } else if ((*path)[ext_depth(inode)].p_ext == NULL) |
43 | ret = -ENODATA; | |
44 | ||
e8505970 AF |
45 | return ret; |
46 | } | |
748de673 AF |
47 | |
48 | /** | |
49 | * copy_extent_status - Copy the extent's initialization status | |
50 | * | |
51 | * @src: an extent for getting initialize status | |
52 | * @dest: an extent to be set the status | |
53 | */ | |
54 | static void | |
55 | copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest) | |
56 | { | |
57 | if (ext4_ext_is_uninitialized(src)) | |
58 | ext4_ext_mark_uninitialized(dest); | |
59 | else | |
60 | dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest)); | |
61 | } | |
62 | ||
63 | /** | |
64 | * mext_next_extent - Search for the next extent and set it to "extent" | |
65 | * | |
66 | * @inode: inode which is searched | |
67 | * @path: this will obtain data for the next extent | |
68 | * @extent: pointer to the next extent we have just gotten | |
69 | * | |
70 | * Search the next extent in the array of ext4_ext_path structure (@path) | |
71 | * and set it to ext4_extent structure (@extent). In addition, the member of | |
72 | * @path (->p_ext) also points the next extent. Return 0 on success, 1 if | |
73 | * ext4_ext_path structure refers to the last extent, or a negative error | |
74 | * value on failure. | |
75 | */ | |
76 | static int | |
77 | mext_next_extent(struct inode *inode, struct ext4_ext_path *path, | |
78 | struct ext4_extent **extent) | |
79 | { | |
fc04cb49 | 80 | struct ext4_extent_header *eh; |
748de673 AF |
81 | int ppos, leaf_ppos = path->p_depth; |
82 | ||
83 | ppos = leaf_ppos; | |
84 | if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) { | |
85 | /* leaf block */ | |
86 | *extent = ++path[ppos].p_ext; | |
fc04cb49 | 87 | path[ppos].p_block = ext_pblock(path[ppos].p_ext); |
748de673 AF |
88 | return 0; |
89 | } | |
90 | ||
91 | while (--ppos >= 0) { | |
92 | if (EXT_LAST_INDEX(path[ppos].p_hdr) > | |
93 | path[ppos].p_idx) { | |
94 | int cur_ppos = ppos; | |
95 | ||
96 | /* index block */ | |
97 | path[ppos].p_idx++; | |
98 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); | |
99 | if (path[ppos+1].p_bh) | |
100 | brelse(path[ppos+1].p_bh); | |
101 | path[ppos+1].p_bh = | |
102 | sb_bread(inode->i_sb, path[ppos].p_block); | |
103 | if (!path[ppos+1].p_bh) | |
104 | return -EIO; | |
105 | path[ppos+1].p_hdr = | |
106 | ext_block_hdr(path[ppos+1].p_bh); | |
107 | ||
108 | /* Halfway index block */ | |
109 | while (++cur_ppos < leaf_ppos) { | |
110 | path[cur_ppos].p_idx = | |
111 | EXT_FIRST_INDEX(path[cur_ppos].p_hdr); | |
112 | path[cur_ppos].p_block = | |
113 | idx_pblock(path[cur_ppos].p_idx); | |
114 | if (path[cur_ppos+1].p_bh) | |
115 | brelse(path[cur_ppos+1].p_bh); | |
116 | path[cur_ppos+1].p_bh = sb_bread(inode->i_sb, | |
117 | path[cur_ppos].p_block); | |
118 | if (!path[cur_ppos+1].p_bh) | |
119 | return -EIO; | |
120 | path[cur_ppos+1].p_hdr = | |
121 | ext_block_hdr(path[cur_ppos+1].p_bh); | |
122 | } | |
123 | ||
fc04cb49 AF |
124 | path[leaf_ppos].p_ext = *extent = NULL; |
125 | ||
126 | eh = path[leaf_ppos].p_hdr; | |
127 | if (le16_to_cpu(eh->eh_entries) == 0) | |
128 | /* empty leaf is found */ | |
129 | return -ENODATA; | |
130 | ||
748de673 AF |
131 | /* leaf block */ |
132 | path[leaf_ppos].p_ext = *extent = | |
133 | EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr); | |
fc04cb49 AF |
134 | path[leaf_ppos].p_block = |
135 | ext_pblock(path[leaf_ppos].p_ext); | |
748de673 AF |
136 | return 0; |
137 | } | |
138 | } | |
139 | /* We found the last extent */ | |
140 | return 1; | |
141 | } | |
142 | ||
2147b1a6 AF |
143 | /** |
144 | * mext_check_null_inode - NULL check for two inodes | |
145 | * | |
146 | * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0. | |
147 | */ | |
148 | static int | |
149 | mext_check_null_inode(struct inode *inode1, struct inode *inode2, | |
150 | const char *function) | |
151 | { | |
152 | int ret = 0; | |
153 | ||
154 | if (inode1 == NULL) { | |
12062ddd | 155 | __ext4_error(inode2->i_sb, function, |
2147b1a6 AF |
156 | "Both inodes should not be NULL: " |
157 | "inode1 NULL inode2 %lu", inode2->i_ino); | |
158 | ret = -EIO; | |
159 | } else if (inode2 == NULL) { | |
12062ddd | 160 | __ext4_error(inode1->i_sb, function, |
2147b1a6 AF |
161 | "Both inodes should not be NULL: " |
162 | "inode1 %lu inode2 NULL", inode1->i_ino); | |
163 | ret = -EIO; | |
164 | } | |
165 | return ret; | |
166 | } | |
167 | ||
748de673 | 168 | /** |
fc04cb49 | 169 | * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem |
748de673 AF |
170 | * |
171 | * @orig_inode: original inode structure | |
172 | * @donor_inode: donor inode structure | |
fc04cb49 AF |
173 | * Acquire write lock of i_data_sem of the two inodes (orig and donor) by |
174 | * i_ino order. | |
748de673 AF |
175 | */ |
176 | static void | |
fc04cb49 | 177 | double_down_write_data_sem(struct inode *orig_inode, struct inode *donor_inode) |
748de673 AF |
178 | { |
179 | struct inode *first = orig_inode, *second = donor_inode; | |
180 | ||
748de673 AF |
181 | /* |
182 | * Use the inode number to provide the stable locking order instead | |
183 | * of its address, because the C language doesn't guarantee you can | |
184 | * compare pointers that don't come from the same array. | |
185 | */ | |
186 | if (donor_inode->i_ino < orig_inode->i_ino) { | |
187 | first = donor_inode; | |
188 | second = orig_inode; | |
189 | } | |
190 | ||
191 | down_write(&EXT4_I(first)->i_data_sem); | |
49bd22bc | 192 | down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING); |
748de673 AF |
193 | } |
194 | ||
195 | /** | |
fc04cb49 | 196 | * double_up_write_data_sem - Release two inodes' write lock of i_data_sem |
748de673 AF |
197 | * |
198 | * @orig_inode: original inode structure to be released its lock first | |
199 | * @donor_inode: donor inode structure to be released its lock second | |
fc04cb49 | 200 | * Release write lock of i_data_sem of two inodes (orig and donor). |
748de673 AF |
201 | */ |
202 | static void | |
fc04cb49 | 203 | double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode) |
748de673 | 204 | { |
748de673 AF |
205 | up_write(&EXT4_I(orig_inode)->i_data_sem); |
206 | up_write(&EXT4_I(donor_inode)->i_data_sem); | |
207 | } | |
208 | ||
209 | /** | |
210 | * mext_insert_across_blocks - Insert extents across leaf block | |
211 | * | |
212 | * @handle: journal handle | |
213 | * @orig_inode: original inode | |
214 | * @o_start: first original extent to be changed | |
215 | * @o_end: last original extent to be changed | |
216 | * @start_ext: first new extent to be inserted | |
217 | * @new_ext: middle of new extent to be inserted | |
218 | * @end_ext: last new extent to be inserted | |
219 | * | |
220 | * Allocate a new leaf block and insert extents into it. Return 0 on success, | |
221 | * or a negative error value on failure. | |
222 | */ | |
223 | static int | |
224 | mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode, | |
225 | struct ext4_extent *o_start, struct ext4_extent *o_end, | |
226 | struct ext4_extent *start_ext, struct ext4_extent *new_ext, | |
227 | struct ext4_extent *end_ext) | |
228 | { | |
229 | struct ext4_ext_path *orig_path = NULL; | |
230 | ext4_lblk_t eblock = 0; | |
231 | int new_flag = 0; | |
232 | int end_flag = 0; | |
233 | int err = 0; | |
234 | ||
235 | if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) { | |
236 | if (o_start == o_end) { | |
237 | ||
238 | /* start_ext new_ext end_ext | |
239 | * donor |---------|-----------|--------| | |
240 | * orig |------------------------------| | |
241 | */ | |
242 | end_flag = 1; | |
243 | } else { | |
244 | ||
245 | /* start_ext new_ext end_ext | |
246 | * donor |---------|----------|---------| | |
247 | * orig |---------------|--------------| | |
248 | */ | |
249 | o_end->ee_block = end_ext->ee_block; | |
250 | o_end->ee_len = end_ext->ee_len; | |
251 | ext4_ext_store_pblock(o_end, ext_pblock(end_ext)); | |
252 | } | |
253 | ||
254 | o_start->ee_len = start_ext->ee_len; | |
5fd5249a | 255 | eblock = le32_to_cpu(start_ext->ee_block); |
748de673 AF |
256 | new_flag = 1; |
257 | ||
258 | } else if (start_ext->ee_len && new_ext->ee_len && | |
259 | !end_ext->ee_len && o_start == o_end) { | |
260 | ||
261 | /* start_ext new_ext | |
262 | * donor |--------------|---------------| | |
263 | * orig |------------------------------| | |
264 | */ | |
265 | o_start->ee_len = start_ext->ee_len; | |
5fd5249a | 266 | eblock = le32_to_cpu(start_ext->ee_block); |
748de673 AF |
267 | new_flag = 1; |
268 | ||
269 | } else if (!start_ext->ee_len && new_ext->ee_len && | |
270 | end_ext->ee_len && o_start == o_end) { | |
271 | ||
272 | /* new_ext end_ext | |
273 | * donor |--------------|---------------| | |
274 | * orig |------------------------------| | |
275 | */ | |
276 | o_end->ee_block = end_ext->ee_block; | |
277 | o_end->ee_len = end_ext->ee_len; | |
278 | ext4_ext_store_pblock(o_end, ext_pblock(end_ext)); | |
279 | ||
280 | /* | |
281 | * Set 0 to the extent block if new_ext was | |
282 | * the first block. | |
283 | */ | |
284 | if (new_ext->ee_block) | |
285 | eblock = le32_to_cpu(new_ext->ee_block); | |
286 | ||
287 | new_flag = 1; | |
288 | } else { | |
289 | ext4_debug("ext4 move extent: Unexpected insert case\n"); | |
290 | return -EIO; | |
291 | } | |
292 | ||
293 | if (new_flag) { | |
e8505970 | 294 | err = get_ext_path(orig_inode, eblock, &orig_path); |
347fa6f1 | 295 | if (err) |
748de673 AF |
296 | goto out; |
297 | ||
298 | if (ext4_ext_insert_extent(handle, orig_inode, | |
0031462b | 299 | orig_path, new_ext, 0)) |
748de673 AF |
300 | goto out; |
301 | } | |
302 | ||
303 | if (end_flag) { | |
e8505970 AF |
304 | err = get_ext_path(orig_inode, |
305 | le32_to_cpu(end_ext->ee_block) - 1, &orig_path); | |
347fa6f1 | 306 | if (err) |
748de673 AF |
307 | goto out; |
308 | ||
309 | if (ext4_ext_insert_extent(handle, orig_inode, | |
0031462b | 310 | orig_path, end_ext, 0)) |
748de673 AF |
311 | goto out; |
312 | } | |
313 | out: | |
314 | if (orig_path) { | |
315 | ext4_ext_drop_refs(orig_path); | |
316 | kfree(orig_path); | |
317 | } | |
318 | ||
319 | return err; | |
320 | ||
321 | } | |
322 | ||
323 | /** | |
324 | * mext_insert_inside_block - Insert new extent to the extent block | |
325 | * | |
326 | * @o_start: first original extent to be moved | |
327 | * @o_end: last original extent to be moved | |
328 | * @start_ext: first new extent to be inserted | |
329 | * @new_ext: middle of new extent to be inserted | |
330 | * @end_ext: last new extent to be inserted | |
331 | * @eh: extent header of target leaf block | |
332 | * @range_to_move: used to decide how to insert extent | |
333 | * | |
334 | * Insert extents into the leaf block. The extent (@o_start) is overwritten | |
335 | * by inserted extents. | |
336 | */ | |
337 | static void | |
338 | mext_insert_inside_block(struct ext4_extent *o_start, | |
339 | struct ext4_extent *o_end, | |
340 | struct ext4_extent *start_ext, | |
341 | struct ext4_extent *new_ext, | |
342 | struct ext4_extent *end_ext, | |
343 | struct ext4_extent_header *eh, | |
344 | int range_to_move) | |
345 | { | |
346 | int i = 0; | |
347 | unsigned long len; | |
348 | ||
349 | /* Move the existing extents */ | |
350 | if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) { | |
351 | len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) - | |
352 | (unsigned long)(o_end + 1); | |
353 | memmove(o_end + 1 + range_to_move, o_end + 1, len); | |
354 | } | |
355 | ||
356 | /* Insert start entry */ | |
357 | if (start_ext->ee_len) | |
358 | o_start[i++].ee_len = start_ext->ee_len; | |
359 | ||
360 | /* Insert new entry */ | |
361 | if (new_ext->ee_len) { | |
362 | o_start[i] = *new_ext; | |
363 | ext4_ext_store_pblock(&o_start[i++], ext_pblock(new_ext)); | |
364 | } | |
365 | ||
366 | /* Insert end entry */ | |
367 | if (end_ext->ee_len) | |
368 | o_start[i] = *end_ext; | |
369 | ||
370 | /* Increment the total entries counter on the extent block */ | |
371 | le16_add_cpu(&eh->eh_entries, range_to_move); | |
372 | } | |
373 | ||
374 | /** | |
375 | * mext_insert_extents - Insert new extent | |
376 | * | |
377 | * @handle: journal handle | |
378 | * @orig_inode: original inode | |
379 | * @orig_path: path indicates first extent to be changed | |
380 | * @o_start: first original extent to be changed | |
381 | * @o_end: last original extent to be changed | |
382 | * @start_ext: first new extent to be inserted | |
383 | * @new_ext: middle of new extent to be inserted | |
384 | * @end_ext: last new extent to be inserted | |
385 | * | |
386 | * Call the function to insert extents. If we cannot add more extents into | |
387 | * the leaf block, we call mext_insert_across_blocks() to create a | |
388 | * new leaf block. Otherwise call mext_insert_inside_block(). Return 0 | |
389 | * on success, or a negative error value on failure. | |
390 | */ | |
391 | static int | |
392 | mext_insert_extents(handle_t *handle, struct inode *orig_inode, | |
393 | struct ext4_ext_path *orig_path, | |
394 | struct ext4_extent *o_start, | |
395 | struct ext4_extent *o_end, | |
396 | struct ext4_extent *start_ext, | |
397 | struct ext4_extent *new_ext, | |
398 | struct ext4_extent *end_ext) | |
399 | { | |
400 | struct ext4_extent_header *eh; | |
401 | unsigned long need_slots, slots_range; | |
402 | int range_to_move, depth, ret; | |
403 | ||
404 | /* | |
405 | * The extents need to be inserted | |
406 | * start_extent + new_extent + end_extent. | |
407 | */ | |
408 | need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) + | |
409 | (new_ext->ee_len ? 1 : 0); | |
410 | ||
411 | /* The number of slots between start and end */ | |
412 | slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1) | |
413 | / sizeof(struct ext4_extent); | |
414 | ||
415 | /* Range to move the end of extent */ | |
416 | range_to_move = need_slots - slots_range; | |
417 | depth = orig_path->p_depth; | |
418 | orig_path += depth; | |
419 | eh = orig_path->p_hdr; | |
420 | ||
421 | if (depth) { | |
422 | /* Register to journal */ | |
423 | ret = ext4_journal_get_write_access(handle, orig_path->p_bh); | |
424 | if (ret) | |
425 | return ret; | |
426 | } | |
427 | ||
428 | /* Expansion */ | |
429 | if (range_to_move > 0 && | |
430 | (range_to_move > le16_to_cpu(eh->eh_max) | |
431 | - le16_to_cpu(eh->eh_entries))) { | |
432 | ||
433 | ret = mext_insert_across_blocks(handle, orig_inode, o_start, | |
434 | o_end, start_ext, new_ext, end_ext); | |
435 | if (ret < 0) | |
436 | return ret; | |
437 | } else | |
438 | mext_insert_inside_block(o_start, o_end, start_ext, new_ext, | |
439 | end_ext, eh, range_to_move); | |
440 | ||
441 | if (depth) { | |
442 | ret = ext4_handle_dirty_metadata(handle, orig_inode, | |
443 | orig_path->p_bh); | |
444 | if (ret) | |
445 | return ret; | |
446 | } else { | |
447 | ret = ext4_mark_inode_dirty(handle, orig_inode); | |
448 | if (ret < 0) | |
449 | return ret; | |
450 | } | |
451 | ||
452 | return 0; | |
453 | } | |
454 | ||
455 | /** | |
456 | * mext_leaf_block - Move one leaf extent block into the inode. | |
457 | * | |
458 | * @handle: journal handle | |
459 | * @orig_inode: original inode | |
460 | * @orig_path: path indicates first extent to be changed | |
461 | * @dext: donor extent | |
462 | * @from: start offset on the target file | |
463 | * | |
464 | * In order to insert extents into the leaf block, we must divide the extent | |
465 | * in the leaf block into three extents. The one is located to be inserted | |
466 | * extents, and the others are located around it. | |
467 | * | |
468 | * Therefore, this function creates structures to save extents of the leaf | |
469 | * block, and inserts extents by calling mext_insert_extents() with | |
470 | * created extents. Return 0 on success, or a negative error value on failure. | |
471 | */ | |
472 | static int | |
473 | mext_leaf_block(handle_t *handle, struct inode *orig_inode, | |
474 | struct ext4_ext_path *orig_path, struct ext4_extent *dext, | |
475 | ext4_lblk_t *from) | |
476 | { | |
477 | struct ext4_extent *oext, *o_start, *o_end, *prev_ext; | |
478 | struct ext4_extent new_ext, start_ext, end_ext; | |
479 | ext4_lblk_t new_ext_end; | |
748de673 AF |
480 | int oext_alen, new_ext_alen, end_ext_alen; |
481 | int depth = ext_depth(orig_inode); | |
482 | int ret; | |
483 | ||
484 | o_start = o_end = oext = orig_path[depth].p_ext; | |
485 | oext_alen = ext4_ext_get_actual_len(oext); | |
486 | start_ext.ee_len = end_ext.ee_len = 0; | |
487 | ||
488 | new_ext.ee_block = cpu_to_le32(*from); | |
489 | ext4_ext_store_pblock(&new_ext, ext_pblock(dext)); | |
490 | new_ext.ee_len = dext->ee_len; | |
491 | new_ext_alen = ext4_ext_get_actual_len(&new_ext); | |
492 | new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1; | |
748de673 AF |
493 | |
494 | /* | |
495 | * Case: original extent is first | |
496 | * oext |--------| | |
497 | * new_ext |--| | |
498 | * start_ext |--| | |
499 | */ | |
500 | if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) && | |
501 | le32_to_cpu(new_ext.ee_block) < | |
502 | le32_to_cpu(oext->ee_block) + oext_alen) { | |
503 | start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) - | |
504 | le32_to_cpu(oext->ee_block)); | |
5fd5249a | 505 | start_ext.ee_block = oext->ee_block; |
748de673 AF |
506 | copy_extent_status(oext, &start_ext); |
507 | } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) { | |
508 | prev_ext = oext - 1; | |
509 | /* | |
510 | * We can merge new_ext into previous extent, | |
511 | * if these are contiguous and same extent type. | |
512 | */ | |
513 | if (ext4_can_extents_be_merged(orig_inode, prev_ext, | |
514 | &new_ext)) { | |
515 | o_start = prev_ext; | |
516 | start_ext.ee_len = cpu_to_le16( | |
517 | ext4_ext_get_actual_len(prev_ext) + | |
518 | new_ext_alen); | |
5fd5249a | 519 | start_ext.ee_block = oext->ee_block; |
748de673 AF |
520 | copy_extent_status(prev_ext, &start_ext); |
521 | new_ext.ee_len = 0; | |
522 | } | |
523 | } | |
524 | ||
525 | /* | |
526 | * Case: new_ext_end must be less than oext | |
527 | * oext |-----------| | |
528 | * new_ext |-------| | |
529 | */ | |
2147b1a6 | 530 | if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) { |
12062ddd | 531 | ext4_error(orig_inode->i_sb, |
2147b1a6 AF |
532 | "new_ext_end(%u) should be less than or equal to " |
533 | "oext->ee_block(%u) + oext_alen(%d) - 1", | |
534 | new_ext_end, le32_to_cpu(oext->ee_block), | |
535 | oext_alen); | |
536 | ret = -EIO; | |
537 | goto out; | |
538 | } | |
748de673 AF |
539 | |
540 | /* | |
541 | * Case: new_ext is smaller than original extent | |
542 | * oext |---------------| | |
543 | * new_ext |-----------| | |
544 | * end_ext |---| | |
545 | */ | |
546 | if (le32_to_cpu(oext->ee_block) <= new_ext_end && | |
547 | new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) { | |
548 | end_ext.ee_len = | |
549 | cpu_to_le16(le32_to_cpu(oext->ee_block) + | |
550 | oext_alen - 1 - new_ext_end); | |
551 | copy_extent_status(oext, &end_ext); | |
552 | end_ext_alen = ext4_ext_get_actual_len(&end_ext); | |
553 | ext4_ext_store_pblock(&end_ext, | |
554 | (ext_pblock(o_end) + oext_alen - end_ext_alen)); | |
555 | end_ext.ee_block = | |
556 | cpu_to_le32(le32_to_cpu(o_end->ee_block) + | |
557 | oext_alen - end_ext_alen); | |
558 | } | |
559 | ||
560 | ret = mext_insert_extents(handle, orig_inode, orig_path, o_start, | |
561 | o_end, &start_ext, &new_ext, &end_ext); | |
2147b1a6 | 562 | out: |
748de673 AF |
563 | return ret; |
564 | } | |
565 | ||
566 | /** | |
567 | * mext_calc_swap_extents - Calculate extents for extent swapping. | |
568 | * | |
569 | * @tmp_dext: the extent that will belong to the original inode | |
570 | * @tmp_oext: the extent that will belong to the donor inode | |
571 | * @orig_off: block offset of original inode | |
572 | * @donor_off: block offset of donor inode | |
92c28159 | 573 | * @max_count: the maximum length of extents |
c40ce3c9 AF |
574 | * |
575 | * Return 0 on success, or a negative error value on failure. | |
748de673 | 576 | */ |
c40ce3c9 | 577 | static int |
748de673 AF |
578 | mext_calc_swap_extents(struct ext4_extent *tmp_dext, |
579 | struct ext4_extent *tmp_oext, | |
580 | ext4_lblk_t orig_off, ext4_lblk_t donor_off, | |
581 | ext4_lblk_t max_count) | |
582 | { | |
583 | ext4_lblk_t diff, orig_diff; | |
584 | struct ext4_extent dext_old, oext_old; | |
585 | ||
c40ce3c9 AF |
586 | BUG_ON(orig_off != donor_off); |
587 | ||
588 | /* original and donor extents have to cover the same block offset */ | |
589 | if (orig_off < le32_to_cpu(tmp_oext->ee_block) || | |
590 | le32_to_cpu(tmp_oext->ee_block) + | |
591 | ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off) | |
592 | return -ENODATA; | |
593 | ||
594 | if (orig_off < le32_to_cpu(tmp_dext->ee_block) || | |
595 | le32_to_cpu(tmp_dext->ee_block) + | |
596 | ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off) | |
597 | return -ENODATA; | |
598 | ||
748de673 AF |
599 | dext_old = *tmp_dext; |
600 | oext_old = *tmp_oext; | |
601 | ||
602 | /* When tmp_dext is too large, pick up the target range. */ | |
603 | diff = donor_off - le32_to_cpu(tmp_dext->ee_block); | |
604 | ||
605 | ext4_ext_store_pblock(tmp_dext, ext_pblock(tmp_dext) + diff); | |
606 | tmp_dext->ee_block = | |
607 | cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff); | |
608 | tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff); | |
609 | ||
610 | if (max_count < ext4_ext_get_actual_len(tmp_dext)) | |
611 | tmp_dext->ee_len = cpu_to_le16(max_count); | |
612 | ||
613 | orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block); | |
614 | ext4_ext_store_pblock(tmp_oext, ext_pblock(tmp_oext) + orig_diff); | |
615 | ||
616 | /* Adjust extent length if donor extent is larger than orig */ | |
617 | if (ext4_ext_get_actual_len(tmp_dext) > | |
618 | ext4_ext_get_actual_len(tmp_oext) - orig_diff) | |
619 | tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) - | |
620 | orig_diff); | |
621 | ||
622 | tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext)); | |
623 | ||
624 | copy_extent_status(&oext_old, tmp_dext); | |
625 | copy_extent_status(&dext_old, tmp_oext); | |
c40ce3c9 AF |
626 | |
627 | return 0; | |
748de673 AF |
628 | } |
629 | ||
630 | /** | |
631 | * mext_replace_branches - Replace original extents with new extents | |
632 | * | |
633 | * @handle: journal handle | |
634 | * @orig_inode: original inode | |
635 | * @donor_inode: donor inode | |
636 | * @from: block offset of orig_inode | |
637 | * @count: block count to be replaced | |
f868a48d | 638 | * @err: pointer to save return value |
748de673 AF |
639 | * |
640 | * Replace original inode extents and donor inode extents page by page. | |
641 | * We implement this replacement in the following three steps: | |
642 | * 1. Save the block information of original and donor inodes into | |
643 | * dummy extents. | |
644 | * 2. Change the block information of original inode to point at the | |
645 | * donor inode blocks. | |
646 | * 3. Change the block information of donor inode to point at the saved | |
647 | * original inode blocks in the dummy extents. | |
648 | * | |
f868a48d | 649 | * Return replaced block count. |
748de673 AF |
650 | */ |
651 | static int | |
652 | mext_replace_branches(handle_t *handle, struct inode *orig_inode, | |
653 | struct inode *donor_inode, ext4_lblk_t from, | |
f868a48d | 654 | ext4_lblk_t count, int *err) |
748de673 AF |
655 | { |
656 | struct ext4_ext_path *orig_path = NULL; | |
657 | struct ext4_ext_path *donor_path = NULL; | |
658 | struct ext4_extent *oext, *dext; | |
659 | struct ext4_extent tmp_dext, tmp_oext; | |
660 | ext4_lblk_t orig_off = from, donor_off = from; | |
748de673 AF |
661 | int depth; |
662 | int replaced_count = 0; | |
663 | int dext_alen; | |
664 | ||
ac48b0a1 AF |
665 | /* Protect extent trees against block allocations via delalloc */ |
666 | double_down_write_data_sem(orig_inode, donor_inode); | |
667 | ||
748de673 | 668 | /* Get the original extent for the block "orig_off" */ |
f868a48d AF |
669 | *err = get_ext_path(orig_inode, orig_off, &orig_path); |
670 | if (*err) | |
748de673 AF |
671 | goto out; |
672 | ||
673 | /* Get the donor extent for the head */ | |
f868a48d AF |
674 | *err = get_ext_path(donor_inode, donor_off, &donor_path); |
675 | if (*err) | |
748de673 AF |
676 | goto out; |
677 | depth = ext_depth(orig_inode); | |
678 | oext = orig_path[depth].p_ext; | |
679 | tmp_oext = *oext; | |
680 | ||
681 | depth = ext_depth(donor_inode); | |
682 | dext = donor_path[depth].p_ext; | |
683 | tmp_dext = *dext; | |
684 | ||
f868a48d | 685 | *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, |
748de673 | 686 | donor_off, count); |
f868a48d | 687 | if (*err) |
c40ce3c9 | 688 | goto out; |
748de673 AF |
689 | |
690 | /* Loop for the donor extents */ | |
691 | while (1) { | |
692 | /* The extent for donor must be found. */ | |
2147b1a6 | 693 | if (!dext) { |
12062ddd | 694 | ext4_error(donor_inode->i_sb, |
2147b1a6 | 695 | "The extent for donor must be found"); |
f868a48d | 696 | *err = -EIO; |
2147b1a6 AF |
697 | goto out; |
698 | } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) { | |
12062ddd | 699 | ext4_error(donor_inode->i_sb, |
2147b1a6 AF |
700 | "Donor offset(%u) and the first block of donor " |
701 | "extent(%u) should be equal", | |
702 | donor_off, | |
703 | le32_to_cpu(tmp_dext.ee_block)); | |
f868a48d | 704 | *err = -EIO; |
2147b1a6 AF |
705 | goto out; |
706 | } | |
748de673 AF |
707 | |
708 | /* Set donor extent to orig extent */ | |
f868a48d | 709 | *err = mext_leaf_block(handle, orig_inode, |
748de673 | 710 | orig_path, &tmp_dext, &orig_off); |
f868a48d | 711 | if (*err) |
748de673 AF |
712 | goto out; |
713 | ||
714 | /* Set orig extent to donor extent */ | |
f868a48d | 715 | *err = mext_leaf_block(handle, donor_inode, |
748de673 | 716 | donor_path, &tmp_oext, &donor_off); |
f868a48d | 717 | if (*err) |
748de673 AF |
718 | goto out; |
719 | ||
720 | dext_alen = ext4_ext_get_actual_len(&tmp_dext); | |
721 | replaced_count += dext_alen; | |
722 | donor_off += dext_alen; | |
723 | orig_off += dext_alen; | |
724 | ||
725 | /* Already moved the expected blocks */ | |
726 | if (replaced_count >= count) | |
727 | break; | |
728 | ||
729 | if (orig_path) | |
730 | ext4_ext_drop_refs(orig_path); | |
f868a48d AF |
731 | *err = get_ext_path(orig_inode, orig_off, &orig_path); |
732 | if (*err) | |
748de673 AF |
733 | goto out; |
734 | depth = ext_depth(orig_inode); | |
735 | oext = orig_path[depth].p_ext; | |
748de673 AF |
736 | tmp_oext = *oext; |
737 | ||
738 | if (donor_path) | |
739 | ext4_ext_drop_refs(donor_path); | |
f868a48d AF |
740 | *err = get_ext_path(donor_inode, donor_off, &donor_path); |
741 | if (*err) | |
748de673 AF |
742 | goto out; |
743 | depth = ext_depth(donor_inode); | |
744 | dext = donor_path[depth].p_ext; | |
748de673 AF |
745 | tmp_dext = *dext; |
746 | ||
f868a48d | 747 | *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off, |
c40ce3c9 | 748 | donor_off, count - replaced_count); |
f868a48d | 749 | if (*err) |
c40ce3c9 | 750 | goto out; |
748de673 AF |
751 | } |
752 | ||
753 | out: | |
754 | if (orig_path) { | |
755 | ext4_ext_drop_refs(orig_path); | |
756 | kfree(orig_path); | |
757 | } | |
758 | if (donor_path) { | |
759 | ext4_ext_drop_refs(donor_path); | |
760 | kfree(donor_path); | |
761 | } | |
762 | ||
ac48b0a1 AF |
763 | ext4_ext_invalidate_cache(orig_inode); |
764 | ext4_ext_invalidate_cache(donor_inode); | |
765 | ||
766 | double_up_write_data_sem(orig_inode, donor_inode); | |
767 | ||
f868a48d | 768 | return replaced_count; |
748de673 AF |
769 | } |
770 | ||
771 | /** | |
772 | * move_extent_per_page - Move extent data per page | |
773 | * | |
774 | * @o_filp: file structure of original file | |
775 | * @donor_inode: donor inode | |
776 | * @orig_page_offset: page index on original file | |
777 | * @data_offset_in_page: block index where data swapping starts | |
778 | * @block_len_in_page: the number of blocks to be swapped | |
779 | * @uninit: orig extent is uninitialized or not | |
f868a48d | 780 | * @err: pointer to save return value |
748de673 AF |
781 | * |
782 | * Save the data in original inode blocks and replace original inode extents | |
783 | * with donor inode extents by calling mext_replace_branches(). | |
f868a48d AF |
784 | * Finally, write out the saved data in new original inode blocks. Return |
785 | * replaced block count. | |
748de673 AF |
786 | */ |
787 | static int | |
44fc48f7 | 788 | move_extent_per_page(struct file *o_filp, struct inode *donor_inode, |
748de673 | 789 | pgoff_t orig_page_offset, int data_offset_in_page, |
f868a48d | 790 | int block_len_in_page, int uninit, int *err) |
748de673 AF |
791 | { |
792 | struct inode *orig_inode = o_filp->f_dentry->d_inode; | |
793 | struct address_space *mapping = orig_inode->i_mapping; | |
794 | struct buffer_head *bh; | |
795 | struct page *page = NULL; | |
796 | const struct address_space_operations *a_ops = mapping->a_ops; | |
797 | handle_t *handle; | |
798 | ext4_lblk_t orig_blk_offset; | |
799 | long long offs = orig_page_offset << PAGE_CACHE_SHIFT; | |
800 | unsigned long blocksize = orig_inode->i_sb->s_blocksize; | |
801 | unsigned int w_flags = 0; | |
f868a48d | 802 | unsigned int tmp_data_size, data_size, replaced_size; |
748de673 | 803 | void *fsdata; |
f868a48d AF |
804 | int i, jblocks; |
805 | int err2 = 0; | |
806 | int replaced_count = 0; | |
748de673 AF |
807 | int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; |
808 | ||
809 | /* | |
810 | * It needs twice the amount of ordinary journal buffers because | |
811 | * inode and donor_inode may change each different metadata blocks. | |
812 | */ | |
813 | jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; | |
814 | handle = ext4_journal_start(orig_inode, jblocks); | |
815 | if (IS_ERR(handle)) { | |
f868a48d AF |
816 | *err = PTR_ERR(handle); |
817 | return 0; | |
748de673 AF |
818 | } |
819 | ||
820 | if (segment_eq(get_fs(), KERNEL_DS)) | |
821 | w_flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
822 | ||
823 | orig_blk_offset = orig_page_offset * blocks_per_page + | |
824 | data_offset_in_page; | |
825 | ||
826 | /* | |
827 | * If orig extent is uninitialized one, | |
828 | * it's not necessary force the page into memory | |
829 | * and then force it to be written out again. | |
830 | * Just swap data blocks between orig and donor. | |
831 | */ | |
832 | if (uninit) { | |
f868a48d AF |
833 | replaced_count = mext_replace_branches(handle, orig_inode, |
834 | donor_inode, orig_blk_offset, | |
835 | block_len_in_page, err); | |
748de673 AF |
836 | goto out2; |
837 | } | |
838 | ||
839 | offs = (long long)orig_blk_offset << orig_inode->i_blkbits; | |
840 | ||
f868a48d | 841 | /* Calculate data_size */ |
748de673 AF |
842 | if ((orig_blk_offset + block_len_in_page - 1) == |
843 | ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { | |
844 | /* Replace the last block */ | |
f868a48d | 845 | tmp_data_size = orig_inode->i_size & (blocksize - 1); |
748de673 | 846 | /* |
f868a48d | 847 | * If data_size equal zero, it shows data_size is multiples of |
748de673 AF |
848 | * blocksize. So we set appropriate value. |
849 | */ | |
f868a48d AF |
850 | if (tmp_data_size == 0) |
851 | tmp_data_size = blocksize; | |
748de673 | 852 | |
f868a48d | 853 | data_size = tmp_data_size + |
748de673 | 854 | ((block_len_in_page - 1) << orig_inode->i_blkbits); |
f868a48d AF |
855 | } else |
856 | data_size = block_len_in_page << orig_inode->i_blkbits; | |
857 | ||
858 | replaced_size = data_size; | |
748de673 | 859 | |
f868a48d | 860 | *err = a_ops->write_begin(o_filp, mapping, offs, data_size, w_flags, |
748de673 | 861 | &page, &fsdata); |
f868a48d | 862 | if (unlikely(*err < 0)) |
748de673 AF |
863 | goto out; |
864 | ||
865 | if (!PageUptodate(page)) { | |
866 | mapping->a_ops->readpage(o_filp, page); | |
867 | lock_page(page); | |
868 | } | |
869 | ||
870 | /* | |
871 | * try_to_release_page() doesn't call releasepage in writeback mode. | |
872 | * We should care about the order of writing to the same file | |
873 | * by multiple move extent processes. | |
874 | * It needs to call wait_on_page_writeback() to wait for the | |
875 | * writeback of the page. | |
876 | */ | |
877 | if (PageWriteback(page)) | |
878 | wait_on_page_writeback(page); | |
879 | ||
880 | /* Release old bh and drop refs */ | |
881 | try_to_release_page(page, 0); | |
882 | ||
f868a48d AF |
883 | replaced_count = mext_replace_branches(handle, orig_inode, donor_inode, |
884 | orig_blk_offset, block_len_in_page, | |
885 | &err2); | |
886 | if (err2) { | |
887 | if (replaced_count) { | |
888 | block_len_in_page = replaced_count; | |
889 | replaced_size = | |
890 | block_len_in_page << orig_inode->i_blkbits; | |
ac48b0a1 | 891 | } else |
f868a48d AF |
892 | goto out; |
893 | } | |
748de673 | 894 | |
748de673 AF |
895 | if (!page_has_buffers(page)) |
896 | create_empty_buffers(page, 1 << orig_inode->i_blkbits, 0); | |
897 | ||
898 | bh = page_buffers(page); | |
899 | for (i = 0; i < data_offset_in_page; i++) | |
900 | bh = bh->b_this_page; | |
901 | ||
902 | for (i = 0; i < block_len_in_page; i++) { | |
f868a48d | 903 | *err = ext4_get_block(orig_inode, |
748de673 | 904 | (sector_t)(orig_blk_offset + i), bh, 0); |
f868a48d | 905 | if (*err < 0) |
748de673 AF |
906 | goto out; |
907 | ||
908 | if (bh->b_this_page != NULL) | |
909 | bh = bh->b_this_page; | |
910 | } | |
911 | ||
f868a48d | 912 | *err = a_ops->write_end(o_filp, mapping, offs, data_size, replaced_size, |
748de673 AF |
913 | page, fsdata); |
914 | page = NULL; | |
915 | ||
916 | out: | |
917 | if (unlikely(page)) { | |
918 | if (PageLocked(page)) | |
919 | unlock_page(page); | |
920 | page_cache_release(page); | |
91cc219a | 921 | ext4_journal_stop(handle); |
748de673 AF |
922 | } |
923 | out2: | |
924 | ext4_journal_stop(handle); | |
925 | ||
f868a48d AF |
926 | if (err2) |
927 | *err = err2; | |
928 | ||
929 | return replaced_count; | |
748de673 AF |
930 | } |
931 | ||
932 | /** | |
c437b273 | 933 | * mext_check_arguments - Check whether move extent can be done |
748de673 AF |
934 | * |
935 | * @orig_inode: original inode | |
936 | * @donor_inode: donor inode | |
937 | * @orig_start: logical start offset in block for orig | |
938 | * @donor_start: logical start offset in block for donor | |
939 | * @len: the number of blocks to be moved | |
748de673 AF |
940 | * |
941 | * Check the arguments of ext4_move_extents() whether the files can be | |
942 | * exchanged with each other. | |
943 | * Return 0 on success, or a negative error value on failure. | |
944 | */ | |
945 | static int | |
946 | mext_check_arguments(struct inode *orig_inode, | |
446aaa6e KM |
947 | struct inode *donor_inode, __u64 orig_start, |
948 | __u64 donor_start, __u64 *len) | |
748de673 | 949 | { |
70d5d3dc AF |
950 | ext4_lblk_t orig_blocks, donor_blocks; |
951 | unsigned int blkbits = orig_inode->i_blkbits; | |
952 | unsigned int blocksize = 1 << blkbits; | |
953 | ||
4a58579b AF |
954 | if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { |
955 | ext4_debug("ext4 move extent: suid or sgid is set" | |
956 | " to donor file [ino:orig %lu, donor %lu]\n", | |
957 | orig_inode->i_ino, donor_inode->i_ino); | |
958 | return -EINVAL; | |
959 | } | |
960 | ||
748de673 AF |
961 | /* Ext4 move extent does not support swapfile */ |
962 | if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { | |
963 | ext4_debug("ext4 move extent: The argument files should " | |
964 | "not be swapfile [ino:orig %lu, donor %lu]\n", | |
965 | orig_inode->i_ino, donor_inode->i_ino); | |
966 | return -EINVAL; | |
967 | } | |
968 | ||
969 | /* Files should be in the same ext4 FS */ | |
970 | if (orig_inode->i_sb != donor_inode->i_sb) { | |
971 | ext4_debug("ext4 move extent: The argument files " | |
972 | "should be in same FS [ino:orig %lu, donor %lu]\n", | |
973 | orig_inode->i_ino, donor_inode->i_ino); | |
974 | return -EINVAL; | |
975 | } | |
976 | ||
748de673 AF |
977 | /* Ext4 move extent supports only extent based file */ |
978 | if (!(EXT4_I(orig_inode)->i_flags & EXT4_EXTENTS_FL)) { | |
979 | ext4_debug("ext4 move extent: orig file is not extents " | |
980 | "based file [ino:orig %lu]\n", orig_inode->i_ino); | |
981 | return -EOPNOTSUPP; | |
982 | } else if (!(EXT4_I(donor_inode)->i_flags & EXT4_EXTENTS_FL)) { | |
983 | ext4_debug("ext4 move extent: donor file is not extents " | |
984 | "based file [ino:donor %lu]\n", donor_inode->i_ino); | |
985 | return -EOPNOTSUPP; | |
986 | } | |
987 | ||
988 | if ((!orig_inode->i_size) || (!donor_inode->i_size)) { | |
989 | ext4_debug("ext4 move extent: File size is 0 byte\n"); | |
990 | return -EINVAL; | |
991 | } | |
992 | ||
993 | /* Start offset should be same */ | |
994 | if (orig_start != donor_start) { | |
995 | ext4_debug("ext4 move extent: orig and donor's start " | |
996 | "offset are not same [ino:orig %lu, donor %lu]\n", | |
997 | orig_inode->i_ino, donor_inode->i_ino); | |
998 | return -EINVAL; | |
999 | } | |
1000 | ||
0a80e986 ES |
1001 | if ((orig_start > EXT_MAX_BLOCK) || |
1002 | (donor_start > EXT_MAX_BLOCK) || | |
1003 | (*len > EXT_MAX_BLOCK) || | |
1004 | (orig_start + *len > EXT_MAX_BLOCK)) { | |
1005 | ext4_debug("ext4 move extent: Can't handle over [%u] blocks " | |
1006 | "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCK, | |
748de673 AF |
1007 | orig_inode->i_ino, donor_inode->i_ino); |
1008 | return -EINVAL; | |
1009 | } | |
1010 | ||
1011 | if (orig_inode->i_size > donor_inode->i_size) { | |
70d5d3dc AF |
1012 | donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits; |
1013 | /* TODO: eliminate this artificial restriction */ | |
1014 | if (orig_start >= donor_blocks) { | |
748de673 | 1015 | ext4_debug("ext4 move extent: orig start offset " |
70d5d3dc AF |
1016 | "[%llu] should be less than donor file blocks " |
1017 | "[%u] [ino:orig %lu, donor %lu]\n", | |
1018 | orig_start, donor_blocks, | |
748de673 AF |
1019 | orig_inode->i_ino, donor_inode->i_ino); |
1020 | return -EINVAL; | |
1021 | } | |
1022 | ||
70d5d3dc AF |
1023 | /* TODO: eliminate this artificial restriction */ |
1024 | if (orig_start + *len > donor_blocks) { | |
748de673 | 1025 | ext4_debug("ext4 move extent: End offset [%llu] should " |
70d5d3dc AF |
1026 | "be less than donor file blocks [%u]." |
1027 | "So adjust length from %llu to %llu " | |
748de673 | 1028 | "[ino:orig %lu, donor %lu]\n", |
70d5d3dc AF |
1029 | orig_start + *len, donor_blocks, |
1030 | *len, donor_blocks - orig_start, | |
748de673 | 1031 | orig_inode->i_ino, donor_inode->i_ino); |
70d5d3dc | 1032 | *len = donor_blocks - orig_start; |
748de673 AF |
1033 | } |
1034 | } else { | |
70d5d3dc AF |
1035 | orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits; |
1036 | if (orig_start >= orig_blocks) { | |
748de673 | 1037 | ext4_debug("ext4 move extent: start offset [%llu] " |
70d5d3dc AF |
1038 | "should be less than original file blocks " |
1039 | "[%u] [ino:orig %lu, donor %lu]\n", | |
1040 | orig_start, orig_blocks, | |
748de673 AF |
1041 | orig_inode->i_ino, donor_inode->i_ino); |
1042 | return -EINVAL; | |
1043 | } | |
1044 | ||
70d5d3dc | 1045 | if (orig_start + *len > orig_blocks) { |
748de673 | 1046 | ext4_debug("ext4 move extent: Adjust length " |
70d5d3dc AF |
1047 | "from %llu to %llu. Because it should be " |
1048 | "less than original file blocks " | |
748de673 | 1049 | "[ino:orig %lu, donor %lu]\n", |
70d5d3dc | 1050 | *len, orig_blocks - orig_start, |
748de673 | 1051 | orig_inode->i_ino, donor_inode->i_ino); |
70d5d3dc | 1052 | *len = orig_blocks - orig_start; |
748de673 AF |
1053 | } |
1054 | } | |
1055 | ||
1056 | if (!*len) { | |
92c28159 | 1057 | ext4_debug("ext4 move extent: len should not be 0 " |
748de673 AF |
1058 | "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, |
1059 | donor_inode->i_ino); | |
1060 | return -EINVAL; | |
1061 | } | |
1062 | ||
1063 | return 0; | |
1064 | } | |
1065 | ||
1066 | /** | |
1067 | * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2 | |
1068 | * | |
1069 | * @inode1: the inode structure | |
1070 | * @inode2: the inode structure | |
1071 | * | |
2147b1a6 AF |
1072 | * Lock two inodes' i_mutex by i_ino order. |
1073 | * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0. | |
748de673 | 1074 | */ |
2147b1a6 | 1075 | static int |
748de673 AF |
1076 | mext_inode_double_lock(struct inode *inode1, struct inode *inode2) |
1077 | { | |
2147b1a6 AF |
1078 | int ret = 0; |
1079 | ||
1080 | BUG_ON(inode1 == NULL && inode2 == NULL); | |
1081 | ||
1082 | ret = mext_check_null_inode(inode1, inode2, __func__); | |
1083 | if (ret < 0) | |
1084 | goto out; | |
1085 | ||
1086 | if (inode1 == inode2) { | |
1087 | mutex_lock(&inode1->i_mutex); | |
1088 | goto out; | |
748de673 AF |
1089 | } |
1090 | ||
1091 | if (inode1->i_ino < inode2->i_ino) { | |
1092 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT); | |
1093 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD); | |
1094 | } else { | |
1095 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT); | |
1096 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD); | |
1097 | } | |
2147b1a6 AF |
1098 | |
1099 | out: | |
1100 | return ret; | |
748de673 AF |
1101 | } |
1102 | ||
1103 | /** | |
1104 | * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2 | |
1105 | * | |
1106 | * @inode1: the inode that is released first | |
1107 | * @inode2: the inode that is released second | |
1108 | * | |
2147b1a6 | 1109 | * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0. |
748de673 AF |
1110 | */ |
1111 | ||
2147b1a6 | 1112 | static int |
748de673 AF |
1113 | mext_inode_double_unlock(struct inode *inode1, struct inode *inode2) |
1114 | { | |
2147b1a6 AF |
1115 | int ret = 0; |
1116 | ||
1117 | BUG_ON(inode1 == NULL && inode2 == NULL); | |
1118 | ||
1119 | ret = mext_check_null_inode(inode1, inode2, __func__); | |
1120 | if (ret < 0) | |
1121 | goto out; | |
1122 | ||
748de673 AF |
1123 | if (inode1) |
1124 | mutex_unlock(&inode1->i_mutex); | |
1125 | ||
1126 | if (inode2 && inode2 != inode1) | |
1127 | mutex_unlock(&inode2->i_mutex); | |
2147b1a6 AF |
1128 | |
1129 | out: | |
1130 | return ret; | |
748de673 AF |
1131 | } |
1132 | ||
1133 | /** | |
1134 | * ext4_move_extents - Exchange the specified range of a file | |
1135 | * | |
1136 | * @o_filp: file structure of the original file | |
1137 | * @d_filp: file structure of the donor file | |
1138 | * @orig_start: start offset in block for orig | |
1139 | * @donor_start: start offset in block for donor | |
1140 | * @len: the number of blocks to be moved | |
1141 | * @moved_len: moved block length | |
1142 | * | |
1143 | * This function returns 0 and moved block length is set in moved_len | |
1144 | * if succeed, otherwise returns error value. | |
1145 | * | |
1146 | * Note: ext4_move_extents() proceeds the following order. | |
1147 | * 1:ext4_move_extents() calculates the last block number of moving extent | |
1148 | * function by the start block number (orig_start) and the number of blocks | |
1149 | * to be moved (len) specified as arguments. | |
1150 | * If the {orig, donor}_start points a hole, the extent's start offset | |
1151 | * pointed by ext_cur (current extent), holecheck_path, orig_path are set | |
1152 | * after hole behind. | |
1153 | * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent | |
1154 | * or the ext_cur exceeds the block_end which is last logical block number. | |
1155 | * 3:To get the length of continues area, call mext_next_extent() | |
1156 | * specified with the ext_cur (initial value is holecheck_path) re-cursive, | |
1157 | * until find un-continuous extent, the start logical block number exceeds | |
1158 | * the block_end or the extent points to the last extent. | |
1159 | * 4:Exchange the original inode data with donor inode data | |
1160 | * from orig_page_offset to seq_end_page. | |
1161 | * The start indexes of data are specified as arguments. | |
1162 | * That of the original inode is orig_page_offset, | |
1163 | * and the donor inode is also orig_page_offset | |
1164 | * (To easily handle blocksize != pagesize case, the offset for the | |
1165 | * donor inode is block unit). | |
1166 | * 5:Update holecheck_path and orig_path to points a next proceeding extent, | |
1167 | * then returns to step 2. | |
1168 | * 6:Release holecheck_path, orig_path and set the len to moved_len | |
1169 | * which shows the number of moved blocks. | |
1170 | * The moved_len is useful for the command to calculate the file offset | |
1171 | * for starting next move extent ioctl. | |
1172 | * 7:Return 0 on success, or a negative error value on failure. | |
1173 | */ | |
1174 | int | |
1175 | ext4_move_extents(struct file *o_filp, struct file *d_filp, | |
1176 | __u64 orig_start, __u64 donor_start, __u64 len, | |
1177 | __u64 *moved_len) | |
1178 | { | |
1179 | struct inode *orig_inode = o_filp->f_dentry->d_inode; | |
1180 | struct inode *donor_inode = d_filp->f_dentry->d_inode; | |
1181 | struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL; | |
1182 | struct ext4_extent *ext_prev, *ext_cur, *ext_dummy; | |
1183 | ext4_lblk_t block_start = orig_start; | |
1184 | ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0; | |
1185 | ext4_lblk_t rest_blocks; | |
1186 | pgoff_t orig_page_offset = 0, seq_end_page; | |
2147b1a6 | 1187 | int ret1, ret2, depth, last_extent = 0; |
748de673 AF |
1188 | int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; |
1189 | int data_offset_in_page; | |
1190 | int block_len_in_page; | |
1191 | int uninit; | |
1192 | ||
f3ce8064 TT |
1193 | /* orig and donor should be different file */ |
1194 | if (orig_inode->i_ino == donor_inode->i_ino) { | |
1195 | ext4_debug("ext4 move extent: The argument files should not " | |
1196 | "be same file [ino:orig %lu, donor %lu]\n", | |
1197 | orig_inode->i_ino, donor_inode->i_ino); | |
1198 | return -EINVAL; | |
1199 | } | |
1200 | ||
7247c0ca AF |
1201 | /* Regular file check */ |
1202 | if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { | |
1203 | ext4_debug("ext4 move extent: The argument files should be " | |
1204 | "regular file [ino:orig %lu, donor %lu]\n", | |
1205 | orig_inode->i_ino, donor_inode->i_ino); | |
1206 | return -EINVAL; | |
1207 | } | |
1208 | ||
fc04cb49 | 1209 | /* Protect orig and donor inodes against a truncate */ |
2147b1a6 AF |
1210 | ret1 = mext_inode_double_lock(orig_inode, donor_inode); |
1211 | if (ret1 < 0) | |
1212 | return ret1; | |
748de673 | 1213 | |
fc04cb49 AF |
1214 | /* Protect extent tree against block allocations via delalloc */ |
1215 | double_down_write_data_sem(orig_inode, donor_inode); | |
748de673 | 1216 | /* Check the filesystem environment whether move_extent can be done */ |
2147b1a6 | 1217 | ret1 = mext_check_arguments(orig_inode, donor_inode, orig_start, |
446aaa6e | 1218 | donor_start, &len); |
2147b1a6 | 1219 | if (ret1) |
347fa6f1 | 1220 | goto out; |
748de673 AF |
1221 | |
1222 | file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits; | |
1223 | block_end = block_start + len - 1; | |
1224 | if (file_end < block_end) | |
1225 | len -= block_end - file_end; | |
1226 | ||
2147b1a6 | 1227 | ret1 = get_ext_path(orig_inode, block_start, &orig_path); |
347fa6f1 AF |
1228 | if (ret1) |
1229 | goto out; | |
748de673 AF |
1230 | |
1231 | /* Get path structure to check the hole */ | |
2147b1a6 | 1232 | ret1 = get_ext_path(orig_inode, block_start, &holecheck_path); |
347fa6f1 | 1233 | if (ret1) |
748de673 AF |
1234 | goto out; |
1235 | ||
1236 | depth = ext_depth(orig_inode); | |
1237 | ext_cur = holecheck_path[depth].p_ext; | |
748de673 AF |
1238 | |
1239 | /* | |
c40ce3c9 AF |
1240 | * Get proper starting location of block replacement if block_start was |
1241 | * within the hole. | |
748de673 AF |
1242 | */ |
1243 | if (le32_to_cpu(ext_cur->ee_block) + | |
1244 | ext4_ext_get_actual_len(ext_cur) - 1 < block_start) { | |
c40ce3c9 AF |
1245 | /* |
1246 | * The hole exists between extents or the tail of | |
1247 | * original file. | |
1248 | */ | |
748de673 AF |
1249 | last_extent = mext_next_extent(orig_inode, |
1250 | holecheck_path, &ext_cur); | |
1251 | if (last_extent < 0) { | |
2147b1a6 | 1252 | ret1 = last_extent; |
748de673 AF |
1253 | goto out; |
1254 | } | |
1255 | last_extent = mext_next_extent(orig_inode, orig_path, | |
1256 | &ext_dummy); | |
1257 | if (last_extent < 0) { | |
2147b1a6 | 1258 | ret1 = last_extent; |
748de673 AF |
1259 | goto out; |
1260 | } | |
c40ce3c9 AF |
1261 | seq_start = le32_to_cpu(ext_cur->ee_block); |
1262 | } else if (le32_to_cpu(ext_cur->ee_block) > block_start) | |
1263 | /* The hole exists at the beginning of original file. */ | |
1264 | seq_start = le32_to_cpu(ext_cur->ee_block); | |
1265 | else | |
1266 | seq_start = block_start; | |
748de673 AF |
1267 | |
1268 | /* No blocks within the specified range. */ | |
1269 | if (le32_to_cpu(ext_cur->ee_block) > block_end) { | |
1270 | ext4_debug("ext4 move extent: The specified range of file " | |
1271 | "may be the hole\n"); | |
2147b1a6 | 1272 | ret1 = -EINVAL; |
748de673 AF |
1273 | goto out; |
1274 | } | |
1275 | ||
1276 | /* Adjust start blocks */ | |
1277 | add_blocks = min(le32_to_cpu(ext_cur->ee_block) + | |
1278 | ext4_ext_get_actual_len(ext_cur), block_end + 1) - | |
1279 | max(le32_to_cpu(ext_cur->ee_block), block_start); | |
1280 | ||
1281 | while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) { | |
1282 | seq_blocks += add_blocks; | |
1283 | ||
1284 | /* Adjust tail blocks */ | |
1285 | if (seq_start + seq_blocks - 1 > block_end) | |
1286 | seq_blocks = block_end - seq_start + 1; | |
1287 | ||
1288 | ext_prev = ext_cur; | |
1289 | last_extent = mext_next_extent(orig_inode, holecheck_path, | |
1290 | &ext_cur); | |
1291 | if (last_extent < 0) { | |
2147b1a6 | 1292 | ret1 = last_extent; |
748de673 AF |
1293 | break; |
1294 | } | |
1295 | add_blocks = ext4_ext_get_actual_len(ext_cur); | |
1296 | ||
1297 | /* | |
1298 | * Extend the length of contiguous block (seq_blocks) | |
1299 | * if extents are contiguous. | |
1300 | */ | |
1301 | if (ext4_can_extents_be_merged(orig_inode, | |
1302 | ext_prev, ext_cur) && | |
1303 | block_end >= le32_to_cpu(ext_cur->ee_block) && | |
1304 | !last_extent) | |
1305 | continue; | |
1306 | ||
1307 | /* Is original extent is uninitialized */ | |
1308 | uninit = ext4_ext_is_uninitialized(ext_prev); | |
1309 | ||
1310 | data_offset_in_page = seq_start % blocks_per_page; | |
1311 | ||
1312 | /* | |
1313 | * Calculate data blocks count that should be swapped | |
1314 | * at the first page. | |
1315 | */ | |
1316 | if (data_offset_in_page + seq_blocks > blocks_per_page) { | |
1317 | /* Swapped blocks are across pages */ | |
1318 | block_len_in_page = | |
1319 | blocks_per_page - data_offset_in_page; | |
1320 | } else { | |
1321 | /* Swapped blocks are in a page */ | |
1322 | block_len_in_page = seq_blocks; | |
1323 | } | |
1324 | ||
1325 | orig_page_offset = seq_start >> | |
1326 | (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); | |
1327 | seq_end_page = (seq_start + seq_blocks - 1) >> | |
1328 | (PAGE_CACHE_SHIFT - orig_inode->i_blkbits); | |
1329 | seq_start = le32_to_cpu(ext_cur->ee_block); | |
1330 | rest_blocks = seq_blocks; | |
1331 | ||
fc04cb49 AF |
1332 | /* |
1333 | * Up semaphore to avoid following problems: | |
1334 | * a. transaction deadlock among ext4_journal_start, | |
1335 | * ->write_begin via pagefault, and jbd2_journal_commit | |
1336 | * b. racing with ->readpage, ->write_begin, and ext4_get_block | |
1337 | * in move_extent_per_page | |
1338 | */ | |
1339 | double_up_write_data_sem(orig_inode, donor_inode); | |
748de673 AF |
1340 | |
1341 | while (orig_page_offset <= seq_end_page) { | |
1342 | ||
1343 | /* Swap original branches with new branches */ | |
f868a48d AF |
1344 | block_len_in_page = move_extent_per_page( |
1345 | o_filp, donor_inode, | |
748de673 AF |
1346 | orig_page_offset, |
1347 | data_offset_in_page, | |
f868a48d AF |
1348 | block_len_in_page, uninit, |
1349 | &ret1); | |
1350 | ||
748de673 AF |
1351 | /* Count how many blocks we have exchanged */ |
1352 | *moved_len += block_len_in_page; | |
f868a48d | 1353 | if (ret1 < 0) |
fc04cb49 | 1354 | break; |
2147b1a6 | 1355 | if (*moved_len > len) { |
12062ddd | 1356 | ext4_error(orig_inode->i_sb, |
2147b1a6 AF |
1357 | "We replaced blocks too much! " |
1358 | "sum of replaced: %llu requested: %llu", | |
1359 | *moved_len, len); | |
1360 | ret1 = -EIO; | |
fc04cb49 | 1361 | break; |
2147b1a6 | 1362 | } |
748de673 | 1363 | |
f868a48d | 1364 | orig_page_offset++; |
748de673 AF |
1365 | data_offset_in_page = 0; |
1366 | rest_blocks -= block_len_in_page; | |
1367 | if (rest_blocks > blocks_per_page) | |
1368 | block_len_in_page = blocks_per_page; | |
1369 | else | |
1370 | block_len_in_page = rest_blocks; | |
1371 | } | |
1372 | ||
fc04cb49 AF |
1373 | double_down_write_data_sem(orig_inode, donor_inode); |
1374 | if (ret1 < 0) | |
1375 | break; | |
1376 | ||
748de673 AF |
1377 | /* Decrease buffer counter */ |
1378 | if (holecheck_path) | |
1379 | ext4_ext_drop_refs(holecheck_path); | |
2147b1a6 | 1380 | ret1 = get_ext_path(orig_inode, seq_start, &holecheck_path); |
347fa6f1 | 1381 | if (ret1) |
748de673 AF |
1382 | break; |
1383 | depth = holecheck_path->p_depth; | |
1384 | ||
1385 | /* Decrease buffer counter */ | |
1386 | if (orig_path) | |
1387 | ext4_ext_drop_refs(orig_path); | |
2147b1a6 | 1388 | ret1 = get_ext_path(orig_inode, seq_start, &orig_path); |
347fa6f1 | 1389 | if (ret1) |
748de673 AF |
1390 | break; |
1391 | ||
1392 | ext_cur = holecheck_path[depth].p_ext; | |
1393 | add_blocks = ext4_ext_get_actual_len(ext_cur); | |
1394 | seq_blocks = 0; | |
1395 | ||
1396 | } | |
1397 | out: | |
94d7c16c AF |
1398 | if (*moved_len) { |
1399 | ext4_discard_preallocations(orig_inode); | |
1400 | ext4_discard_preallocations(donor_inode); | |
1401 | } | |
1402 | ||
748de673 AF |
1403 | if (orig_path) { |
1404 | ext4_ext_drop_refs(orig_path); | |
1405 | kfree(orig_path); | |
1406 | } | |
1407 | if (holecheck_path) { | |
1408 | ext4_ext_drop_refs(holecheck_path); | |
1409 | kfree(holecheck_path); | |
1410 | } | |
fc04cb49 | 1411 | double_up_write_data_sem(orig_inode, donor_inode); |
2147b1a6 | 1412 | ret2 = mext_inode_double_unlock(orig_inode, donor_inode); |
748de673 | 1413 | |
2147b1a6 AF |
1414 | if (ret1) |
1415 | return ret1; | |
1416 | else if (ret2) | |
1417 | return ret2; | |
748de673 | 1418 | |
748de673 AF |
1419 | return 0; |
1420 | } |