]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
7 | * of the GNU General Public License v.2. | |
8 | */ | |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
5c676f6d | 15 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 16 | #include <linux/crc32.h> |
b3b94faa DT |
17 | #include <asm/semaphore.h> |
18 | ||
19 | #include "gfs2.h" | |
5c676f6d SW |
20 | #include "lm_interface.h" |
21 | #include "incore.h" | |
b3b94faa DT |
22 | #include "bmap.h" |
23 | #include "glock.h" | |
24 | #include "inode.h" | |
b3b94faa DT |
25 | #include "meta_io.h" |
26 | #include "page.h" | |
27 | #include "quota.h" | |
28 | #include "rgrp.h" | |
29 | #include "trans.h" | |
18ec7d5c | 30 | #include "dir.h" |
5c676f6d | 31 | #include "util.h" |
b3b94faa DT |
32 | |
33 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
34 | * block is 512, so __u16 is fine for that. It saves stack space to | |
35 | * keep it small. | |
36 | */ | |
37 | struct metapath { | |
38 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; | |
39 | }; | |
40 | ||
41 | typedef int (*block_call_t) (struct gfs2_inode *ip, struct buffer_head *dibh, | |
42 | struct buffer_head *bh, uint64_t *top, | |
43 | uint64_t *bottom, unsigned int height, | |
44 | void *data); | |
45 | ||
46 | struct strip_mine { | |
47 | int sm_first; | |
48 | unsigned int sm_height; | |
49 | }; | |
50 | ||
51 | /** | |
52 | * @gfs2_unstuffer_sync - Synchronously unstuff a dinode | |
53 | * @ip: | |
54 | * @dibh: | |
55 | * @block: | |
56 | * @private: | |
57 | * | |
58 | * Cheat and use a metadata buffer instead of a data page. | |
59 | * | |
60 | * Returns: errno | |
61 | */ | |
62 | ||
63 | int gfs2_unstuffer_sync(struct gfs2_inode *ip, struct buffer_head *dibh, | |
64 | uint64_t block, void *private) | |
65 | { | |
66 | struct buffer_head *bh; | |
67 | int error; | |
68 | ||
69 | bh = gfs2_meta_new(ip->i_gl, block); | |
70 | ||
71 | gfs2_buffer_copy_tail(bh, 0, dibh, sizeof(struct gfs2_dinode)); | |
72 | ||
73 | set_buffer_dirty(bh); | |
74 | error = sync_dirty_buffer(bh); | |
75 | ||
76 | brelse(bh); | |
77 | ||
78 | return error; | |
79 | } | |
80 | ||
81 | /** | |
82 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
83 | * @ip: The GFS2 inode to unstuff | |
84 | * @unstuffer: the routine that handles unstuffing a non-zero length file | |
85 | * @private: private data for the unstuffer | |
86 | * | |
87 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
88 | * that the height can be grown in the traditional way. | |
89 | * | |
90 | * Returns: errno | |
91 | */ | |
92 | ||
93 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, gfs2_unstuffer_t unstuffer, | |
94 | void *private) | |
95 | { | |
96 | struct buffer_head *bh, *dibh; | |
97 | uint64_t block = 0; | |
18ec7d5c | 98 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
99 | int error; |
100 | ||
101 | down_write(&ip->i_rw_mutex); | |
102 | ||
103 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
104 | if (error) | |
105 | goto out; | |
106 | ||
107 | if (ip->i_di.di_size) { | |
108 | /* Get a free block, fill it with the stuffed data, | |
109 | and write it out to disk */ | |
110 | ||
18ec7d5c | 111 | if (isdir) { |
b3b94faa DT |
112 | block = gfs2_alloc_meta(ip); |
113 | ||
18ec7d5c | 114 | error = gfs2_dir_get_buffer(ip, block, 1, &bh); |
b3b94faa DT |
115 | if (error) |
116 | goto out_brelse; | |
117 | gfs2_buffer_copy_tail(bh, | |
118 | sizeof(struct gfs2_meta_header), | |
119 | dibh, sizeof(struct gfs2_dinode)); | |
120 | brelse(bh); | |
121 | } else { | |
122 | block = gfs2_alloc_data(ip); | |
123 | ||
124 | error = unstuffer(ip, dibh, block, private); | |
125 | if (error) | |
126 | goto out_brelse; | |
127 | } | |
128 | } | |
129 | ||
130 | /* Set up the pointer to the new block */ | |
131 | ||
d4e9c4c3 | 132 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
133 | |
134 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); | |
135 | ||
136 | if (ip->i_di.di_size) { | |
568f4c96 SW |
137 | *(uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode)) = |
138 | cpu_to_be64(block); | |
b3b94faa DT |
139 | ip->i_di.di_blocks++; |
140 | } | |
141 | ||
142 | ip->i_di.di_height = 1; | |
143 | ||
144 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
145 | ||
146 | out_brelse: | |
147 | brelse(dibh); | |
148 | ||
149 | out: | |
150 | up_write(&ip->i_rw_mutex); | |
151 | ||
152 | return error; | |
153 | } | |
154 | ||
155 | /** | |
156 | * calc_tree_height - Calculate the height of a metadata tree | |
157 | * @ip: The GFS2 inode | |
158 | * @size: The proposed size of the file | |
159 | * | |
160 | * Work out how tall a metadata tree needs to be in order to accommodate a | |
161 | * file of a particular size. If size is less than the current size of | |
162 | * the inode, then the current size of the inode is used instead of the | |
163 | * supplied one. | |
164 | * | |
165 | * Returns: the height the tree should be | |
166 | */ | |
167 | ||
168 | static unsigned int calc_tree_height(struct gfs2_inode *ip, uint64_t size) | |
169 | { | |
170 | struct gfs2_sbd *sdp = ip->i_sbd; | |
171 | uint64_t *arr; | |
172 | unsigned int max, height; | |
173 | ||
174 | if (ip->i_di.di_size > size) | |
175 | size = ip->i_di.di_size; | |
176 | ||
18ec7d5c | 177 | if (gfs2_is_dir(ip)) { |
b3b94faa DT |
178 | arr = sdp->sd_jheightsize; |
179 | max = sdp->sd_max_jheight; | |
180 | } else { | |
181 | arr = sdp->sd_heightsize; | |
182 | max = sdp->sd_max_height; | |
183 | } | |
184 | ||
185 | for (height = 0; height < max; height++) | |
186 | if (arr[height] >= size) | |
187 | break; | |
188 | ||
189 | return height; | |
190 | } | |
191 | ||
192 | /** | |
193 | * build_height - Build a metadata tree of the requested height | |
194 | * @ip: The GFS2 inode | |
195 | * @height: The height to build to | |
196 | * | |
197 | * This routine makes sure that the metadata tree is tall enough to hold | |
198 | * "size" bytes of data. | |
199 | * | |
200 | * Returns: errno | |
201 | */ | |
202 | ||
203 | static int build_height(struct gfs2_inode *ip, int height) | |
204 | { | |
205 | struct gfs2_sbd *sdp = ip->i_sbd; | |
206 | struct buffer_head *bh, *dibh; | |
207 | uint64_t block = 0, *bp; | |
208 | unsigned int x; | |
209 | int new_block; | |
210 | int error; | |
211 | ||
212 | while (ip->i_di.di_height < height) { | |
213 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
214 | if (error) | |
215 | return error; | |
216 | ||
217 | new_block = 0; | |
218 | bp = (uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode)); | |
219 | for (x = 0; x < sdp->sd_diptrs; x++, bp++) | |
220 | if (*bp) { | |
221 | new_block = 1; | |
222 | break; | |
223 | } | |
224 | ||
225 | if (new_block) { | |
226 | /* Get a new block, fill it with the old direct | |
227 | pointers, and write it out */ | |
228 | ||
229 | block = gfs2_alloc_meta(ip); | |
230 | ||
231 | bh = gfs2_meta_new(ip->i_gl, block); | |
d4e9c4c3 | 232 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
233 | gfs2_metatype_set(bh, |
234 | GFS2_METATYPE_IN, | |
235 | GFS2_FORMAT_IN); | |
236 | gfs2_buffer_copy_tail(bh, | |
237 | sizeof(struct gfs2_meta_header), | |
238 | dibh, sizeof(struct gfs2_dinode)); | |
239 | ||
240 | brelse(bh); | |
241 | } | |
242 | ||
243 | /* Set up the new direct pointer and write it out to disk */ | |
244 | ||
d4e9c4c3 | 245 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
246 | |
247 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); | |
248 | ||
249 | if (new_block) { | |
568f4c96 SW |
250 | *(uint64_t *)(dibh->b_data + |
251 | sizeof(struct gfs2_dinode)) = | |
252 | cpu_to_be64(block); | |
b3b94faa DT |
253 | ip->i_di.di_blocks++; |
254 | } | |
255 | ||
256 | ip->i_di.di_height++; | |
257 | ||
258 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
259 | brelse(dibh); | |
260 | } | |
261 | ||
262 | return 0; | |
263 | } | |
264 | ||
265 | /** | |
266 | * find_metapath - Find path through the metadata tree | |
267 | * @ip: The inode pointer | |
268 | * @mp: The metapath to return the result in | |
269 | * @block: The disk block to look up | |
270 | * | |
271 | * This routine returns a struct metapath structure that defines a path | |
272 | * through the metadata of inode "ip" to get to block "block". | |
273 | * | |
274 | * Example: | |
275 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
276 | * filesystem with a blocksize of 4096. | |
277 | * | |
278 | * find_metapath() would return a struct metapath structure set to: | |
279 | * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48, | |
280 | * and mp_list[2] = 165. | |
281 | * | |
282 | * That means that in order to get to the block containing the byte at | |
283 | * offset 101342453, we would load the indirect block pointed to by pointer | |
284 | * 0 in the dinode. We would then load the indirect block pointed to by | |
285 | * pointer 48 in that indirect block. We would then load the data block | |
286 | * pointed to by pointer 165 in that indirect block. | |
287 | * | |
288 | * ---------------------------------------- | |
289 | * | Dinode | | | |
290 | * | | 4| | |
291 | * | |0 1 2 3 4 5 9| | |
292 | * | | 6| | |
293 | * ---------------------------------------- | |
294 | * | | |
295 | * | | |
296 | * V | |
297 | * ---------------------------------------- | |
298 | * | Indirect Block | | |
299 | * | 5| | |
300 | * | 4 4 4 4 4 5 5 1| | |
301 | * |0 5 6 7 8 9 0 1 2| | |
302 | * ---------------------------------------- | |
303 | * | | |
304 | * | | |
305 | * V | |
306 | * ---------------------------------------- | |
307 | * | Indirect Block | | |
308 | * | 1 1 1 1 1 5| | |
309 | * | 6 6 6 6 6 1| | |
310 | * |0 3 4 5 6 7 2| | |
311 | * ---------------------------------------- | |
312 | * | | |
313 | * | | |
314 | * V | |
315 | * ---------------------------------------- | |
316 | * | Data block containing offset | | |
317 | * | 101342453 | | |
318 | * | | | |
319 | * | | | |
320 | * ---------------------------------------- | |
321 | * | |
322 | */ | |
323 | ||
568f4c96 SW |
324 | static void find_metapath(struct gfs2_inode *ip, uint64_t block, |
325 | struct metapath *mp) | |
b3b94faa DT |
326 | { |
327 | struct gfs2_sbd *sdp = ip->i_sbd; | |
328 | uint64_t b = block; | |
329 | unsigned int i; | |
330 | ||
331 | for (i = ip->i_di.di_height; i--;) | |
332 | mp->mp_list[i] = (__u16)do_div(b, sdp->sd_inptrs); | |
333 | ||
334 | } | |
335 | ||
336 | /** | |
337 | * metapointer - Return pointer to start of metadata in a buffer | |
338 | * @bh: The buffer | |
339 | * @height: The metadata height (0 = dinode) | |
340 | * @mp: The metapath | |
341 | * | |
342 | * Return a pointer to the block number of the next height of the metadata | |
343 | * tree given a buffer containing the pointer to the current height of the | |
344 | * metadata tree. | |
345 | */ | |
346 | ||
347 | static inline uint64_t *metapointer(struct buffer_head *bh, | |
348 | unsigned int height, struct metapath *mp) | |
349 | { | |
350 | unsigned int head_size = (height > 0) ? | |
351 | sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode); | |
352 | ||
353 | return ((uint64_t *)(bh->b_data + head_size)) + mp->mp_list[height]; | |
354 | } | |
355 | ||
356 | /** | |
357 | * lookup_block - Get the next metadata block in metadata tree | |
358 | * @ip: The GFS2 inode | |
359 | * @bh: Buffer containing the pointers to metadata blocks | |
360 | * @height: The height of the tree (0 = dinode) | |
361 | * @mp: The metapath | |
362 | * @create: Non-zero if we may create a new meatdata block | |
363 | * @new: Used to indicate if we did create a new metadata block | |
364 | * @block: the returned disk block number | |
365 | * | |
366 | * Given a metatree, complete to a particular height, checks to see if the next | |
367 | * height of the tree exists. If not the next height of the tree is created. | |
368 | * The block number of the next height of the metadata tree is returned. | |
369 | * | |
370 | */ | |
371 | ||
372 | static void lookup_block(struct gfs2_inode *ip, struct buffer_head *bh, | |
373 | unsigned int height, struct metapath *mp, int create, | |
374 | int *new, uint64_t *block) | |
375 | { | |
376 | uint64_t *ptr = metapointer(bh, height, mp); | |
377 | ||
378 | if (*ptr) { | |
379 | *block = be64_to_cpu(*ptr); | |
380 | return; | |
381 | } | |
382 | ||
383 | *block = 0; | |
384 | ||
385 | if (!create) | |
386 | return; | |
387 | ||
388 | if (height == ip->i_di.di_height - 1 && | |
18ec7d5c | 389 | !gfs2_is_dir(ip)) |
b3b94faa DT |
390 | *block = gfs2_alloc_data(ip); |
391 | else | |
392 | *block = gfs2_alloc_meta(ip); | |
393 | ||
d4e9c4c3 | 394 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
395 | |
396 | *ptr = cpu_to_be64(*block); | |
397 | ip->i_di.di_blocks++; | |
398 | ||
399 | *new = 1; | |
400 | } | |
401 | ||
402 | /** | |
403 | * gfs2_block_map - Map a block from an inode to a disk block | |
404 | * @ip: The GFS2 inode | |
405 | * @lblock: The logical block number | |
406 | * @new: Value/Result argument (1 = may create/did create new blocks) | |
407 | * @dblock: the disk block number of the start of an extent | |
408 | * @extlen: the size of the extent | |
409 | * | |
410 | * Find the block number on the current device which corresponds to an | |
411 | * inode's block. If the block had to be created, "new" will be set. | |
412 | * | |
413 | * Returns: errno | |
414 | */ | |
415 | ||
416 | int gfs2_block_map(struct gfs2_inode *ip, uint64_t lblock, int *new, | |
417 | uint64_t *dblock, uint32_t *extlen) | |
418 | { | |
419 | struct gfs2_sbd *sdp = ip->i_sbd; | |
420 | struct buffer_head *bh; | |
421 | struct metapath mp; | |
422 | int create = *new; | |
423 | unsigned int bsize; | |
424 | unsigned int height; | |
425 | unsigned int end_of_metadata; | |
426 | unsigned int x; | |
427 | int error = 0; | |
428 | ||
429 | *new = 0; | |
430 | *dblock = 0; | |
431 | if (extlen) | |
432 | *extlen = 0; | |
433 | ||
434 | if (create) | |
435 | down_write(&ip->i_rw_mutex); | |
436 | else | |
437 | down_read(&ip->i_rw_mutex); | |
438 | ||
439 | if (gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip))) | |
440 | goto out; | |
441 | ||
18ec7d5c | 442 | bsize = (gfs2_is_dir(ip)) ? sdp->sd_jbsize : sdp->sd_sb.sb_bsize; |
b3b94faa DT |
443 | |
444 | height = calc_tree_height(ip, (lblock + 1) * bsize); | |
445 | if (ip->i_di.di_height < height) { | |
446 | if (!create) | |
447 | goto out; | |
448 | ||
449 | error = build_height(ip, height); | |
450 | if (error) | |
451 | goto out; | |
452 | } | |
453 | ||
454 | find_metapath(ip, lblock, &mp); | |
455 | end_of_metadata = ip->i_di.di_height - 1; | |
456 | ||
457 | error = gfs2_meta_inode_buffer(ip, &bh); | |
458 | if (error) | |
459 | goto out; | |
460 | ||
461 | for (x = 0; x < end_of_metadata; x++) { | |
462 | lookup_block(ip, bh, x, &mp, create, new, dblock); | |
463 | brelse(bh); | |
464 | if (!*dblock) | |
465 | goto out; | |
466 | ||
467 | error = gfs2_meta_indirect_buffer(ip, x+1, *dblock, *new, &bh); | |
468 | if (error) | |
469 | goto out; | |
470 | } | |
471 | ||
472 | lookup_block(ip, bh, end_of_metadata, &mp, create, new, dblock); | |
473 | ||
474 | if (extlen && *dblock) { | |
475 | *extlen = 1; | |
476 | ||
477 | if (!*new) { | |
478 | uint64_t tmp_dblock; | |
479 | int tmp_new; | |
480 | unsigned int nptrs; | |
481 | ||
482 | nptrs = (end_of_metadata) ? sdp->sd_inptrs : | |
483 | sdp->sd_diptrs; | |
484 | ||
485 | while (++mp.mp_list[end_of_metadata] < nptrs) { | |
486 | lookup_block(ip, bh, end_of_metadata, &mp, | |
487 | 0, &tmp_new, &tmp_dblock); | |
488 | ||
489 | if (*dblock + *extlen != tmp_dblock) | |
490 | break; | |
491 | ||
492 | (*extlen)++; | |
493 | } | |
494 | } | |
495 | } | |
496 | ||
497 | brelse(bh); | |
498 | ||
499 | if (*new) { | |
500 | error = gfs2_meta_inode_buffer(ip, &bh); | |
501 | if (!error) { | |
d4e9c4c3 | 502 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
503 | gfs2_dinode_out(&ip->i_di, bh->b_data); |
504 | brelse(bh); | |
505 | } | |
506 | } | |
507 | ||
508 | out: | |
509 | if (create) | |
510 | up_write(&ip->i_rw_mutex); | |
511 | else | |
512 | up_read(&ip->i_rw_mutex); | |
513 | ||
514 | return error; | |
515 | } | |
516 | ||
517 | /** | |
518 | * recursive_scan - recursively scan through the end of a file | |
519 | * @ip: the inode | |
520 | * @dibh: the dinode buffer | |
521 | * @mp: the path through the metadata to the point to start | |
522 | * @height: the height the recursion is at | |
523 | * @block: the indirect block to look at | |
524 | * @first: 1 if this is the first block | |
525 | * @bc: the call to make for each piece of metadata | |
526 | * @data: data opaque to this function to pass to @bc | |
527 | * | |
528 | * When this is first called @height and @block should be zero and | |
529 | * @first should be 1. | |
530 | * | |
531 | * Returns: errno | |
532 | */ | |
533 | ||
534 | static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh, | |
535 | struct metapath *mp, unsigned int height, | |
536 | uint64_t block, int first, block_call_t bc, | |
537 | void *data) | |
538 | { | |
539 | struct gfs2_sbd *sdp = ip->i_sbd; | |
540 | struct buffer_head *bh = NULL; | |
541 | uint64_t *top, *bottom; | |
542 | uint64_t bn; | |
543 | int error; | |
544 | int mh_size = sizeof(struct gfs2_meta_header); | |
545 | ||
546 | if (!height) { | |
547 | error = gfs2_meta_inode_buffer(ip, &bh); | |
548 | if (error) | |
549 | return error; | |
550 | dibh = bh; | |
551 | ||
552 | top = (uint64_t *)(bh->b_data + sizeof(struct gfs2_dinode)) + | |
553 | mp->mp_list[0]; | |
554 | bottom = (uint64_t *)(bh->b_data + sizeof(struct gfs2_dinode)) + | |
555 | sdp->sd_diptrs; | |
556 | } else { | |
557 | error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh); | |
558 | if (error) | |
559 | return error; | |
560 | ||
561 | top = (uint64_t *)(bh->b_data + mh_size) + | |
562 | ((first) ? mp->mp_list[height] : 0); | |
563 | ||
564 | bottom = (uint64_t *)(bh->b_data + mh_size) + sdp->sd_inptrs; | |
565 | } | |
566 | ||
567 | error = bc(ip, dibh, bh, top, bottom, height, data); | |
568 | if (error) | |
569 | goto out; | |
570 | ||
571 | if (height < ip->i_di.di_height - 1) | |
572 | for (; top < bottom; top++, first = 0) { | |
573 | if (!*top) | |
574 | continue; | |
575 | ||
576 | bn = be64_to_cpu(*top); | |
577 | ||
578 | error = recursive_scan(ip, dibh, mp, height + 1, bn, | |
579 | first, bc, data); | |
580 | if (error) | |
581 | break; | |
582 | } | |
583 | ||
584 | out: | |
585 | brelse(bh); | |
586 | ||
587 | return error; | |
588 | } | |
589 | ||
590 | /** | |
591 | * do_strip - Look for a layer a particular layer of the file and strip it off | |
592 | * @ip: the inode | |
593 | * @dibh: the dinode buffer | |
594 | * @bh: A buffer of pointers | |
595 | * @top: The first pointer in the buffer | |
596 | * @bottom: One more than the last pointer | |
597 | * @height: the height this buffer is at | |
598 | * @data: a pointer to a struct strip_mine | |
599 | * | |
600 | * Returns: errno | |
601 | */ | |
602 | ||
603 | static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh, | |
604 | struct buffer_head *bh, uint64_t *top, uint64_t *bottom, | |
605 | unsigned int height, void *data) | |
606 | { | |
607 | struct strip_mine *sm = (struct strip_mine *)data; | |
608 | struct gfs2_sbd *sdp = ip->i_sbd; | |
609 | struct gfs2_rgrp_list rlist; | |
610 | uint64_t bn, bstart; | |
611 | uint32_t blen; | |
612 | uint64_t *p; | |
613 | unsigned int rg_blocks = 0; | |
614 | int metadata; | |
615 | unsigned int revokes = 0; | |
616 | int x; | |
617 | int error; | |
618 | ||
619 | if (!*top) | |
620 | sm->sm_first = 0; | |
621 | ||
622 | if (height != sm->sm_height) | |
623 | return 0; | |
624 | ||
625 | if (sm->sm_first) { | |
626 | top++; | |
627 | sm->sm_first = 0; | |
628 | } | |
629 | ||
18ec7d5c | 630 | metadata = (height != ip->i_di.di_height - 1); |
b3b94faa DT |
631 | if (metadata) |
632 | revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
633 | ||
634 | error = gfs2_rindex_hold(sdp, &ip->i_alloc.al_ri_gh); | |
635 | if (error) | |
636 | return error; | |
637 | ||
638 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); | |
639 | bstart = 0; | |
640 | blen = 0; | |
641 | ||
642 | for (p = top; p < bottom; p++) { | |
643 | if (!*p) | |
644 | continue; | |
645 | ||
646 | bn = be64_to_cpu(*p); | |
647 | ||
648 | if (bstart + blen == bn) | |
649 | blen++; | |
650 | else { | |
651 | if (bstart) | |
652 | gfs2_rlist_add(sdp, &rlist, bstart); | |
653 | ||
654 | bstart = bn; | |
655 | blen = 1; | |
656 | } | |
657 | } | |
658 | ||
659 | if (bstart) | |
660 | gfs2_rlist_add(sdp, &rlist, bstart); | |
661 | else | |
662 | goto out; /* Nothing to do */ | |
663 | ||
664 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0); | |
665 | ||
666 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
667 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 668 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
b3b94faa DT |
669 | rg_blocks += rgd->rd_ri.ri_length; |
670 | } | |
671 | ||
672 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
673 | if (error) | |
674 | goto out_rlist; | |
675 | ||
676 | error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE + | |
677 | RES_INDIRECT + RES_STATFS + RES_QUOTA, | |
678 | revokes); | |
679 | if (error) | |
680 | goto out_rg_gunlock; | |
681 | ||
682 | down_write(&ip->i_rw_mutex); | |
683 | ||
d4e9c4c3 SW |
684 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
685 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
b3b94faa DT |
686 | |
687 | bstart = 0; | |
688 | blen = 0; | |
689 | ||
690 | for (p = top; p < bottom; p++) { | |
691 | if (!*p) | |
692 | continue; | |
693 | ||
694 | bn = be64_to_cpu(*p); | |
695 | ||
696 | if (bstart + blen == bn) | |
697 | blen++; | |
698 | else { | |
699 | if (bstart) { | |
700 | if (metadata) | |
701 | gfs2_free_meta(ip, bstart, blen); | |
702 | else | |
703 | gfs2_free_data(ip, bstart, blen); | |
704 | } | |
705 | ||
706 | bstart = bn; | |
707 | blen = 1; | |
708 | } | |
709 | ||
710 | *p = 0; | |
711 | if (!ip->i_di.di_blocks) | |
712 | gfs2_consist_inode(ip); | |
713 | ip->i_di.di_blocks--; | |
714 | } | |
715 | if (bstart) { | |
716 | if (metadata) | |
717 | gfs2_free_meta(ip, bstart, blen); | |
718 | else | |
719 | gfs2_free_data(ip, bstart, blen); | |
720 | } | |
721 | ||
722 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
723 | ||
724 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
725 | ||
726 | up_write(&ip->i_rw_mutex); | |
727 | ||
728 | gfs2_trans_end(sdp); | |
729 | ||
730 | out_rg_gunlock: | |
731 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
732 | ||
733 | out_rlist: | |
734 | gfs2_rlist_free(&rlist); | |
735 | ||
736 | out: | |
737 | gfs2_glock_dq_uninit(&ip->i_alloc.al_ri_gh); | |
738 | ||
739 | return error; | |
740 | } | |
741 | ||
742 | /** | |
743 | * do_grow - Make a file look bigger than it is | |
744 | * @ip: the inode | |
745 | * @size: the size to set the file to | |
746 | * | |
747 | * Called with an exclusive lock on @ip. | |
748 | * | |
749 | * Returns: errno | |
750 | */ | |
751 | ||
752 | static int do_grow(struct gfs2_inode *ip, uint64_t size) | |
753 | { | |
754 | struct gfs2_sbd *sdp = ip->i_sbd; | |
755 | struct gfs2_alloc *al; | |
756 | struct buffer_head *dibh; | |
757 | unsigned int h; | |
758 | int error; | |
759 | ||
760 | al = gfs2_alloc_get(ip); | |
761 | ||
762 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
763 | if (error) | |
764 | goto out; | |
765 | ||
766 | error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid); | |
767 | if (error) | |
768 | goto out_gunlock_q; | |
769 | ||
770 | al->al_requested = sdp->sd_max_height + RES_DATA; | |
771 | ||
772 | error = gfs2_inplace_reserve(ip); | |
773 | if (error) | |
774 | goto out_gunlock_q; | |
775 | ||
776 | error = gfs2_trans_begin(sdp, | |
777 | sdp->sd_max_height + al->al_rgd->rd_ri.ri_length + | |
778 | RES_JDATA + RES_DINODE + RES_STATFS + RES_QUOTA, 0); | |
779 | if (error) | |
780 | goto out_ipres; | |
781 | ||
782 | if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { | |
783 | if (gfs2_is_stuffed(ip)) { | |
784 | error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_page, | |
785 | NULL); | |
786 | if (error) | |
787 | goto out_end_trans; | |
788 | } | |
789 | ||
790 | h = calc_tree_height(ip, size); | |
791 | if (ip->i_di.di_height < h) { | |
792 | down_write(&ip->i_rw_mutex); | |
793 | error = build_height(ip, h); | |
794 | up_write(&ip->i_rw_mutex); | |
795 | if (error) | |
796 | goto out_end_trans; | |
797 | } | |
798 | } | |
799 | ||
800 | ip->i_di.di_size = size; | |
801 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
802 | ||
803 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
804 | if (error) | |
805 | goto out_end_trans; | |
806 | ||
d4e9c4c3 | 807 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
808 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
809 | brelse(dibh); | |
810 | ||
811 | out_end_trans: | |
812 | gfs2_trans_end(sdp); | |
813 | ||
814 | out_ipres: | |
815 | gfs2_inplace_release(ip); | |
816 | ||
817 | out_gunlock_q: | |
818 | gfs2_quota_unlock(ip); | |
819 | ||
820 | out: | |
821 | gfs2_alloc_put(ip); | |
822 | ||
823 | return error; | |
824 | } | |
825 | ||
aa6a85a9 | 826 | static int trunc_start(struct gfs2_inode *ip, uint64_t size) |
b3b94faa DT |
827 | { |
828 | struct gfs2_sbd *sdp = ip->i_sbd; | |
829 | struct buffer_head *dibh; | |
830 | int journaled = gfs2_is_jdata(ip); | |
831 | int error; | |
832 | ||
833 | error = gfs2_trans_begin(sdp, | |
834 | RES_DINODE + ((journaled) ? RES_JDATA : 0), 0); | |
835 | if (error) | |
836 | return error; | |
837 | ||
838 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
839 | if (error) | |
840 | goto out; | |
841 | ||
842 | if (gfs2_is_stuffed(ip)) { | |
843 | ip->i_di.di_size = size; | |
844 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
d4e9c4c3 | 845 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
846 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
847 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + size); | |
848 | error = 1; | |
849 | ||
850 | } else { | |
18ec7d5c | 851 | if (size & (uint64_t)(sdp->sd_sb.sb_bsize - 1)) |
257f9b4e | 852 | error = gfs2_block_truncate_page(ip->i_vnode->i_mapping); |
b3b94faa DT |
853 | |
854 | if (!error) { | |
855 | ip->i_di.di_size = size; | |
856 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
857 | ip->i_di.di_flags |= GFS2_DIF_TRUNC_IN_PROG; | |
d4e9c4c3 | 858 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
859 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
860 | } | |
861 | } | |
862 | ||
863 | brelse(dibh); | |
864 | ||
865 | out: | |
866 | gfs2_trans_end(sdp); | |
867 | ||
868 | return error; | |
869 | } | |
870 | ||
871 | static int trunc_dealloc(struct gfs2_inode *ip, uint64_t size) | |
872 | { | |
873 | unsigned int height = ip->i_di.di_height; | |
874 | uint64_t lblock; | |
875 | struct metapath mp; | |
876 | int error; | |
877 | ||
878 | if (!size) | |
879 | lblock = 0; | |
18ec7d5c | 880 | else |
b3b94faa DT |
881 | lblock = (size - 1) >> ip->i_sbd->sd_sb.sb_bsize_shift; |
882 | ||
883 | find_metapath(ip, lblock, &mp); | |
884 | gfs2_alloc_get(ip); | |
885 | ||
886 | error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
887 | if (error) | |
888 | goto out; | |
889 | ||
890 | while (height--) { | |
891 | struct strip_mine sm; | |
892 | sm.sm_first = !!size; | |
893 | sm.sm_height = height; | |
894 | ||
895 | error = recursive_scan(ip, NULL, &mp, 0, 0, 1, do_strip, &sm); | |
896 | if (error) | |
897 | break; | |
898 | } | |
899 | ||
900 | gfs2_quota_unhold(ip); | |
901 | ||
902 | out: | |
903 | gfs2_alloc_put(ip); | |
904 | return error; | |
905 | } | |
906 | ||
907 | static int trunc_end(struct gfs2_inode *ip) | |
908 | { | |
909 | struct gfs2_sbd *sdp = ip->i_sbd; | |
910 | struct buffer_head *dibh; | |
911 | int error; | |
912 | ||
913 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
914 | if (error) | |
915 | return error; | |
916 | ||
917 | down_write(&ip->i_rw_mutex); | |
918 | ||
919 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
920 | if (error) | |
921 | goto out; | |
922 | ||
923 | if (!ip->i_di.di_size) { | |
924 | ip->i_di.di_height = 0; | |
925 | ip->i_di.di_goal_meta = | |
926 | ip->i_di.di_goal_data = | |
927 | ip->i_num.no_addr; | |
928 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); | |
929 | } | |
930 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
931 | ip->i_di.di_flags &= ~GFS2_DIF_TRUNC_IN_PROG; | |
932 | ||
d4e9c4c3 | 933 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
934 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
935 | brelse(dibh); | |
936 | ||
937 | out: | |
938 | up_write(&ip->i_rw_mutex); | |
939 | ||
940 | gfs2_trans_end(sdp); | |
941 | ||
942 | return error; | |
943 | } | |
944 | ||
945 | /** | |
946 | * do_shrink - make a file smaller | |
947 | * @ip: the inode | |
948 | * @size: the size to make the file | |
949 | * @truncator: function to truncate the last partial block | |
950 | * | |
951 | * Called with an exclusive lock on @ip. | |
952 | * | |
953 | * Returns: errno | |
954 | */ | |
955 | ||
aa6a85a9 | 956 | static int do_shrink(struct gfs2_inode *ip, uint64_t size) |
b3b94faa DT |
957 | { |
958 | int error; | |
959 | ||
aa6a85a9 | 960 | error = trunc_start(ip, size); |
b3b94faa DT |
961 | if (error < 0) |
962 | return error; | |
963 | if (error > 0) | |
964 | return 0; | |
965 | ||
966 | error = trunc_dealloc(ip, size); | |
967 | if (!error) | |
968 | error = trunc_end(ip); | |
969 | ||
970 | return error; | |
971 | } | |
972 | ||
973 | /** | |
666a2c53 | 974 | * gfs2_truncatei - make a file a given size |
b3b94faa DT |
975 | * @ip: the inode |
976 | * @size: the size to make the file | |
977 | * @truncator: function to truncate the last partial block | |
978 | * | |
979 | * The file size can grow, shrink, or stay the same size. | |
980 | * | |
981 | * Returns: errno | |
982 | */ | |
983 | ||
aa6a85a9 | 984 | int gfs2_truncatei(struct gfs2_inode *ip, uint64_t size) |
b3b94faa DT |
985 | { |
986 | int error; | |
987 | ||
988 | if (gfs2_assert_warn(ip->i_sbd, S_ISREG(ip->i_di.di_mode))) | |
989 | return -EINVAL; | |
990 | ||
991 | if (size > ip->i_di.di_size) | |
992 | error = do_grow(ip, size); | |
993 | else | |
aa6a85a9 | 994 | error = do_shrink(ip, size); |
b3b94faa DT |
995 | |
996 | return error; | |
997 | } | |
998 | ||
999 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1000 | { | |
1001 | int error; | |
1002 | error = trunc_dealloc(ip, ip->i_di.di_size); | |
1003 | if (!error) | |
1004 | error = trunc_end(ip); | |
1005 | return error; | |
1006 | } | |
1007 | ||
1008 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1009 | { | |
1010 | return trunc_dealloc(ip, 0); | |
1011 | } | |
1012 | ||
1013 | /** | |
1014 | * gfs2_write_calc_reserv - calculate number of blocks needed to write to a file | |
1015 | * @ip: the file | |
1016 | * @len: the number of bytes to be written to the file | |
1017 | * @data_blocks: returns the number of data blocks required | |
1018 | * @ind_blocks: returns the number of indirect blocks required | |
1019 | * | |
1020 | */ | |
1021 | ||
1022 | void gfs2_write_calc_reserv(struct gfs2_inode *ip, unsigned int len, | |
1023 | unsigned int *data_blocks, unsigned int *ind_blocks) | |
1024 | { | |
1025 | struct gfs2_sbd *sdp = ip->i_sbd; | |
1026 | unsigned int tmp; | |
1027 | ||
18ec7d5c | 1028 | if (gfs2_is_dir(ip)) { |
5c676f6d | 1029 | *data_blocks = DIV_ROUND_UP(len, sdp->sd_jbsize) + 2; |
b3b94faa DT |
1030 | *ind_blocks = 3 * (sdp->sd_max_jheight - 1); |
1031 | } else { | |
1032 | *data_blocks = (len >> sdp->sd_sb.sb_bsize_shift) + 3; | |
1033 | *ind_blocks = 3 * (sdp->sd_max_height - 1); | |
1034 | } | |
1035 | ||
1036 | for (tmp = *data_blocks; tmp > sdp->sd_diptrs;) { | |
5c676f6d | 1037 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); |
b3b94faa DT |
1038 | *ind_blocks += tmp; |
1039 | } | |
1040 | } | |
1041 | ||
1042 | /** | |
1043 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
1044 | * @ip: the file being written to | |
1045 | * @offset: the offset to write to | |
1046 | * @len: the number of bytes being written | |
1047 | * @alloc_required: set to 1 if an alloc is required, 0 otherwise | |
1048 | * | |
1049 | * Returns: errno | |
1050 | */ | |
1051 | ||
1052 | int gfs2_write_alloc_required(struct gfs2_inode *ip, uint64_t offset, | |
1053 | unsigned int len, int *alloc_required) | |
1054 | { | |
1055 | struct gfs2_sbd *sdp = ip->i_sbd; | |
1056 | uint64_t lblock, lblock_stop, dblock; | |
1057 | uint32_t extlen; | |
1058 | int new = 0; | |
1059 | int error = 0; | |
1060 | ||
1061 | *alloc_required = 0; | |
1062 | ||
1063 | if (!len) | |
1064 | return 0; | |
1065 | ||
1066 | if (gfs2_is_stuffed(ip)) { | |
1067 | if (offset + len > | |
1068 | sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
1069 | *alloc_required = 1; | |
1070 | return 0; | |
1071 | } | |
1072 | ||
18ec7d5c | 1073 | if (gfs2_is_dir(ip)) { |
b3b94faa DT |
1074 | unsigned int bsize = sdp->sd_jbsize; |
1075 | lblock = offset; | |
1076 | do_div(lblock, bsize); | |
1077 | lblock_stop = offset + len + bsize - 1; | |
1078 | do_div(lblock_stop, bsize); | |
1079 | } else { | |
1080 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
1081 | lblock = offset >> shift; | |
1082 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
1083 | } | |
1084 | ||
1085 | for (; lblock < lblock_stop; lblock += extlen) { | |
1086 | error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen); | |
1087 | if (error) | |
1088 | return error; | |
1089 | ||
1090 | if (!dblock) { | |
1091 | *alloc_required = 1; | |
1092 | return 0; | |
1093 | } | |
1094 | } | |
1095 | ||
1096 | return 0; | |
1097 | } | |
1098 |