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 | */ | |
08bc2dbc | 62 | #if 0 |
b3b94faa DT |
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 | } | |
08bc2dbc | 80 | #endif /* 0 */ |
b3b94faa DT |
81 | |
82 | /** | |
83 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
84 | * @ip: The GFS2 inode to unstuff | |
85 | * @unstuffer: the routine that handles unstuffing a non-zero length file | |
86 | * @private: private data for the unstuffer | |
87 | * | |
88 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
89 | * that the height can be grown in the traditional way. | |
90 | * | |
91 | * Returns: errno | |
92 | */ | |
93 | ||
94 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, gfs2_unstuffer_t unstuffer, | |
95 | void *private) | |
96 | { | |
97 | struct buffer_head *bh, *dibh; | |
98 | uint64_t block = 0; | |
18ec7d5c | 99 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
100 | int error; |
101 | ||
102 | down_write(&ip->i_rw_mutex); | |
103 | ||
104 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
105 | if (error) | |
106 | goto out; | |
107 | ||
108 | if (ip->i_di.di_size) { | |
109 | /* Get a free block, fill it with the stuffed data, | |
110 | and write it out to disk */ | |
111 | ||
18ec7d5c | 112 | if (isdir) { |
b3b94faa DT |
113 | block = gfs2_alloc_meta(ip); |
114 | ||
61e085a8 | 115 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
116 | if (error) |
117 | goto out_brelse; | |
118 | gfs2_buffer_copy_tail(bh, | |
119 | sizeof(struct gfs2_meta_header), | |
120 | dibh, sizeof(struct gfs2_dinode)); | |
121 | brelse(bh); | |
122 | } else { | |
123 | block = gfs2_alloc_data(ip); | |
124 | ||
125 | error = unstuffer(ip, dibh, block, private); | |
126 | if (error) | |
127 | goto out_brelse; | |
128 | } | |
129 | } | |
130 | ||
131 | /* Set up the pointer to the new block */ | |
132 | ||
d4e9c4c3 | 133 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
134 | |
135 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); | |
136 | ||
137 | if (ip->i_di.di_size) { | |
568f4c96 SW |
138 | *(uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode)) = |
139 | cpu_to_be64(block); | |
b3b94faa DT |
140 | ip->i_di.di_blocks++; |
141 | } | |
142 | ||
143 | ip->i_di.di_height = 1; | |
144 | ||
145 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
146 | ||
147 | out_brelse: | |
148 | brelse(dibh); | |
149 | ||
150 | out: | |
151 | up_write(&ip->i_rw_mutex); | |
152 | ||
153 | return error; | |
154 | } | |
155 | ||
156 | /** | |
157 | * calc_tree_height - Calculate the height of a metadata tree | |
158 | * @ip: The GFS2 inode | |
159 | * @size: The proposed size of the file | |
160 | * | |
161 | * Work out how tall a metadata tree needs to be in order to accommodate a | |
162 | * file of a particular size. If size is less than the current size of | |
163 | * the inode, then the current size of the inode is used instead of the | |
164 | * supplied one. | |
165 | * | |
166 | * Returns: the height the tree should be | |
167 | */ | |
168 | ||
169 | static unsigned int calc_tree_height(struct gfs2_inode *ip, uint64_t size) | |
170 | { | |
171 | struct gfs2_sbd *sdp = ip->i_sbd; | |
172 | uint64_t *arr; | |
173 | unsigned int max, height; | |
174 | ||
175 | if (ip->i_di.di_size > size) | |
176 | size = ip->i_di.di_size; | |
177 | ||
18ec7d5c | 178 | if (gfs2_is_dir(ip)) { |
b3b94faa DT |
179 | arr = sdp->sd_jheightsize; |
180 | max = sdp->sd_max_jheight; | |
181 | } else { | |
182 | arr = sdp->sd_heightsize; | |
183 | max = sdp->sd_max_height; | |
184 | } | |
185 | ||
186 | for (height = 0; height < max; height++) | |
187 | if (arr[height] >= size) | |
188 | break; | |
189 | ||
190 | return height; | |
191 | } | |
192 | ||
193 | /** | |
194 | * build_height - Build a metadata tree of the requested height | |
195 | * @ip: The GFS2 inode | |
196 | * @height: The height to build to | |
197 | * | |
198 | * This routine makes sure that the metadata tree is tall enough to hold | |
199 | * "size" bytes of data. | |
200 | * | |
201 | * Returns: errno | |
202 | */ | |
203 | ||
204 | static int build_height(struct gfs2_inode *ip, int height) | |
205 | { | |
206 | struct gfs2_sbd *sdp = ip->i_sbd; | |
207 | struct buffer_head *bh, *dibh; | |
208 | uint64_t block = 0, *bp; | |
209 | unsigned int x; | |
210 | int new_block; | |
211 | int error; | |
212 | ||
213 | while (ip->i_di.di_height < height) { | |
214 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
215 | if (error) | |
216 | return error; | |
217 | ||
218 | new_block = 0; | |
219 | bp = (uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode)); | |
220 | for (x = 0; x < sdp->sd_diptrs; x++, bp++) | |
221 | if (*bp) { | |
222 | new_block = 1; | |
223 | break; | |
224 | } | |
225 | ||
226 | if (new_block) { | |
227 | /* Get a new block, fill it with the old direct | |
228 | pointers, and write it out */ | |
229 | ||
230 | block = gfs2_alloc_meta(ip); | |
231 | ||
232 | bh = gfs2_meta_new(ip->i_gl, block); | |
d4e9c4c3 | 233 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
234 | gfs2_metatype_set(bh, |
235 | GFS2_METATYPE_IN, | |
236 | GFS2_FORMAT_IN); | |
237 | gfs2_buffer_copy_tail(bh, | |
238 | sizeof(struct gfs2_meta_header), | |
239 | dibh, sizeof(struct gfs2_dinode)); | |
240 | ||
241 | brelse(bh); | |
242 | } | |
243 | ||
244 | /* Set up the new direct pointer and write it out to disk */ | |
245 | ||
d4e9c4c3 | 246 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
247 | |
248 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); | |
249 | ||
250 | if (new_block) { | |
568f4c96 SW |
251 | *(uint64_t *)(dibh->b_data + |
252 | sizeof(struct gfs2_dinode)) = | |
253 | cpu_to_be64(block); | |
b3b94faa DT |
254 | ip->i_di.di_blocks++; |
255 | } | |
256 | ||
257 | ip->i_di.di_height++; | |
258 | ||
259 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
260 | brelse(dibh); | |
261 | } | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
266 | /** | |
267 | * find_metapath - Find path through the metadata tree | |
268 | * @ip: The inode pointer | |
269 | * @mp: The metapath to return the result in | |
270 | * @block: The disk block to look up | |
271 | * | |
272 | * This routine returns a struct metapath structure that defines a path | |
273 | * through the metadata of inode "ip" to get to block "block". | |
274 | * | |
275 | * Example: | |
276 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
277 | * filesystem with a blocksize of 4096. | |
278 | * | |
279 | * find_metapath() would return a struct metapath structure set to: | |
280 | * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48, | |
281 | * and mp_list[2] = 165. | |
282 | * | |
283 | * That means that in order to get to the block containing the byte at | |
284 | * offset 101342453, we would load the indirect block pointed to by pointer | |
285 | * 0 in the dinode. We would then load the indirect block pointed to by | |
286 | * pointer 48 in that indirect block. We would then load the data block | |
287 | * pointed to by pointer 165 in that indirect block. | |
288 | * | |
289 | * ---------------------------------------- | |
290 | * | Dinode | | | |
291 | * | | 4| | |
292 | * | |0 1 2 3 4 5 9| | |
293 | * | | 6| | |
294 | * ---------------------------------------- | |
295 | * | | |
296 | * | | |
297 | * V | |
298 | * ---------------------------------------- | |
299 | * | Indirect Block | | |
300 | * | 5| | |
301 | * | 4 4 4 4 4 5 5 1| | |
302 | * |0 5 6 7 8 9 0 1 2| | |
303 | * ---------------------------------------- | |
304 | * | | |
305 | * | | |
306 | * V | |
307 | * ---------------------------------------- | |
308 | * | Indirect Block | | |
309 | * | 1 1 1 1 1 5| | |
310 | * | 6 6 6 6 6 1| | |
311 | * |0 3 4 5 6 7 2| | |
312 | * ---------------------------------------- | |
313 | * | | |
314 | * | | |
315 | * V | |
316 | * ---------------------------------------- | |
317 | * | Data block containing offset | | |
318 | * | 101342453 | | |
319 | * | | | |
320 | * | | | |
321 | * ---------------------------------------- | |
322 | * | |
323 | */ | |
324 | ||
568f4c96 SW |
325 | static void find_metapath(struct gfs2_inode *ip, uint64_t block, |
326 | struct metapath *mp) | |
b3b94faa DT |
327 | { |
328 | struct gfs2_sbd *sdp = ip->i_sbd; | |
329 | uint64_t b = block; | |
330 | unsigned int i; | |
331 | ||
332 | for (i = ip->i_di.di_height; i--;) | |
333 | mp->mp_list[i] = (__u16)do_div(b, sdp->sd_inptrs); | |
334 | ||
335 | } | |
336 | ||
337 | /** | |
338 | * metapointer - Return pointer to start of metadata in a buffer | |
339 | * @bh: The buffer | |
340 | * @height: The metadata height (0 = dinode) | |
341 | * @mp: The metapath | |
342 | * | |
343 | * Return a pointer to the block number of the next height of the metadata | |
344 | * tree given a buffer containing the pointer to the current height of the | |
345 | * metadata tree. | |
346 | */ | |
347 | ||
348 | static inline uint64_t *metapointer(struct buffer_head *bh, | |
349 | unsigned int height, struct metapath *mp) | |
350 | { | |
351 | unsigned int head_size = (height > 0) ? | |
352 | sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode); | |
353 | ||
354 | return ((uint64_t *)(bh->b_data + head_size)) + mp->mp_list[height]; | |
355 | } | |
356 | ||
357 | /** | |
358 | * lookup_block - Get the next metadata block in metadata tree | |
359 | * @ip: The GFS2 inode | |
360 | * @bh: Buffer containing the pointers to metadata blocks | |
361 | * @height: The height of the tree (0 = dinode) | |
362 | * @mp: The metapath | |
363 | * @create: Non-zero if we may create a new meatdata block | |
364 | * @new: Used to indicate if we did create a new metadata block | |
365 | * @block: the returned disk block number | |
366 | * | |
367 | * Given a metatree, complete to a particular height, checks to see if the next | |
368 | * height of the tree exists. If not the next height of the tree is created. | |
369 | * The block number of the next height of the metadata tree is returned. | |
370 | * | |
371 | */ | |
372 | ||
373 | static void lookup_block(struct gfs2_inode *ip, struct buffer_head *bh, | |
374 | unsigned int height, struct metapath *mp, int create, | |
375 | int *new, uint64_t *block) | |
376 | { | |
377 | uint64_t *ptr = metapointer(bh, height, mp); | |
378 | ||
379 | if (*ptr) { | |
380 | *block = be64_to_cpu(*ptr); | |
381 | return; | |
382 | } | |
383 | ||
384 | *block = 0; | |
385 | ||
386 | if (!create) | |
387 | return; | |
388 | ||
389 | if (height == ip->i_di.di_height - 1 && | |
18ec7d5c | 390 | !gfs2_is_dir(ip)) |
b3b94faa DT |
391 | *block = gfs2_alloc_data(ip); |
392 | else | |
393 | *block = gfs2_alloc_meta(ip); | |
394 | ||
d4e9c4c3 | 395 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
396 | |
397 | *ptr = cpu_to_be64(*block); | |
398 | ip->i_di.di_blocks++; | |
399 | ||
400 | *new = 1; | |
401 | } | |
402 | ||
403 | /** | |
404 | * gfs2_block_map - Map a block from an inode to a disk block | |
405 | * @ip: The GFS2 inode | |
406 | * @lblock: The logical block number | |
407 | * @new: Value/Result argument (1 = may create/did create new blocks) | |
408 | * @dblock: the disk block number of the start of an extent | |
409 | * @extlen: the size of the extent | |
410 | * | |
411 | * Find the block number on the current device which corresponds to an | |
412 | * inode's block. If the block had to be created, "new" will be set. | |
413 | * | |
414 | * Returns: errno | |
415 | */ | |
416 | ||
417 | int gfs2_block_map(struct gfs2_inode *ip, uint64_t lblock, int *new, | |
418 | uint64_t *dblock, uint32_t *extlen) | |
419 | { | |
420 | struct gfs2_sbd *sdp = ip->i_sbd; | |
421 | struct buffer_head *bh; | |
422 | struct metapath mp; | |
423 | int create = *new; | |
424 | unsigned int bsize; | |
425 | unsigned int height; | |
426 | unsigned int end_of_metadata; | |
427 | unsigned int x; | |
428 | int error = 0; | |
429 | ||
430 | *new = 0; | |
431 | *dblock = 0; | |
432 | if (extlen) | |
433 | *extlen = 0; | |
434 | ||
435 | if (create) | |
436 | down_write(&ip->i_rw_mutex); | |
437 | else | |
438 | down_read(&ip->i_rw_mutex); | |
439 | ||
440 | if (gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip))) | |
441 | goto out; | |
442 | ||
18ec7d5c | 443 | bsize = (gfs2_is_dir(ip)) ? sdp->sd_jbsize : sdp->sd_sb.sb_bsize; |
b3b94faa DT |
444 | |
445 | height = calc_tree_height(ip, (lblock + 1) * bsize); | |
446 | if (ip->i_di.di_height < height) { | |
447 | if (!create) | |
448 | goto out; | |
449 | ||
450 | error = build_height(ip, height); | |
451 | if (error) | |
452 | goto out; | |
453 | } | |
454 | ||
455 | find_metapath(ip, lblock, &mp); | |
456 | end_of_metadata = ip->i_di.di_height - 1; | |
457 | ||
458 | error = gfs2_meta_inode_buffer(ip, &bh); | |
459 | if (error) | |
460 | goto out; | |
461 | ||
462 | for (x = 0; x < end_of_metadata; x++) { | |
463 | lookup_block(ip, bh, x, &mp, create, new, dblock); | |
464 | brelse(bh); | |
465 | if (!*dblock) | |
466 | goto out; | |
467 | ||
468 | error = gfs2_meta_indirect_buffer(ip, x+1, *dblock, *new, &bh); | |
469 | if (error) | |
470 | goto out; | |
471 | } | |
472 | ||
473 | lookup_block(ip, bh, end_of_metadata, &mp, create, new, dblock); | |
474 | ||
475 | if (extlen && *dblock) { | |
476 | *extlen = 1; | |
477 | ||
478 | if (!*new) { | |
479 | uint64_t tmp_dblock; | |
480 | int tmp_new; | |
481 | unsigned int nptrs; | |
482 | ||
483 | nptrs = (end_of_metadata) ? sdp->sd_inptrs : | |
484 | sdp->sd_diptrs; | |
485 | ||
486 | while (++mp.mp_list[end_of_metadata] < nptrs) { | |
487 | lookup_block(ip, bh, end_of_metadata, &mp, | |
488 | 0, &tmp_new, &tmp_dblock); | |
489 | ||
490 | if (*dblock + *extlen != tmp_dblock) | |
491 | break; | |
492 | ||
493 | (*extlen)++; | |
494 | } | |
495 | } | |
496 | } | |
497 | ||
498 | brelse(bh); | |
499 | ||
500 | if (*new) { | |
501 | error = gfs2_meta_inode_buffer(ip, &bh); | |
502 | if (!error) { | |
d4e9c4c3 | 503 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
504 | gfs2_dinode_out(&ip->i_di, bh->b_data); |
505 | brelse(bh); | |
506 | } | |
507 | } | |
508 | ||
509 | out: | |
510 | if (create) | |
511 | up_write(&ip->i_rw_mutex); | |
512 | else | |
513 | up_read(&ip->i_rw_mutex); | |
514 | ||
515 | return error; | |
516 | } | |
517 | ||
518 | /** | |
519 | * recursive_scan - recursively scan through the end of a file | |
520 | * @ip: the inode | |
521 | * @dibh: the dinode buffer | |
522 | * @mp: the path through the metadata to the point to start | |
523 | * @height: the height the recursion is at | |
524 | * @block: the indirect block to look at | |
525 | * @first: 1 if this is the first block | |
526 | * @bc: the call to make for each piece of metadata | |
527 | * @data: data opaque to this function to pass to @bc | |
528 | * | |
529 | * When this is first called @height and @block should be zero and | |
530 | * @first should be 1. | |
531 | * | |
532 | * Returns: errno | |
533 | */ | |
534 | ||
535 | static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh, | |
536 | struct metapath *mp, unsigned int height, | |
537 | uint64_t block, int first, block_call_t bc, | |
538 | void *data) | |
539 | { | |
540 | struct gfs2_sbd *sdp = ip->i_sbd; | |
541 | struct buffer_head *bh = NULL; | |
542 | uint64_t *top, *bottom; | |
543 | uint64_t bn; | |
544 | int error; | |
545 | int mh_size = sizeof(struct gfs2_meta_header); | |
546 | ||
547 | if (!height) { | |
548 | error = gfs2_meta_inode_buffer(ip, &bh); | |
549 | if (error) | |
550 | return error; | |
551 | dibh = bh; | |
552 | ||
553 | top = (uint64_t *)(bh->b_data + sizeof(struct gfs2_dinode)) + | |
554 | mp->mp_list[0]; | |
555 | bottom = (uint64_t *)(bh->b_data + sizeof(struct gfs2_dinode)) + | |
556 | sdp->sd_diptrs; | |
557 | } else { | |
558 | error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh); | |
559 | if (error) | |
560 | return error; | |
561 | ||
562 | top = (uint64_t *)(bh->b_data + mh_size) + | |
563 | ((first) ? mp->mp_list[height] : 0); | |
564 | ||
565 | bottom = (uint64_t *)(bh->b_data + mh_size) + sdp->sd_inptrs; | |
566 | } | |
567 | ||
568 | error = bc(ip, dibh, bh, top, bottom, height, data); | |
569 | if (error) | |
570 | goto out; | |
571 | ||
572 | if (height < ip->i_di.di_height - 1) | |
573 | for (; top < bottom; top++, first = 0) { | |
574 | if (!*top) | |
575 | continue; | |
576 | ||
577 | bn = be64_to_cpu(*top); | |
578 | ||
579 | error = recursive_scan(ip, dibh, mp, height + 1, bn, | |
580 | first, bc, data); | |
581 | if (error) | |
582 | break; | |
583 | } | |
584 | ||
585 | out: | |
586 | brelse(bh); | |
587 | ||
588 | return error; | |
589 | } | |
590 | ||
591 | /** | |
592 | * do_strip - Look for a layer a particular layer of the file and strip it off | |
593 | * @ip: the inode | |
594 | * @dibh: the dinode buffer | |
595 | * @bh: A buffer of pointers | |
596 | * @top: The first pointer in the buffer | |
597 | * @bottom: One more than the last pointer | |
598 | * @height: the height this buffer is at | |
599 | * @data: a pointer to a struct strip_mine | |
600 | * | |
601 | * Returns: errno | |
602 | */ | |
603 | ||
604 | static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh, | |
605 | struct buffer_head *bh, uint64_t *top, uint64_t *bottom, | |
606 | unsigned int height, void *data) | |
607 | { | |
608 | struct strip_mine *sm = (struct strip_mine *)data; | |
609 | struct gfs2_sbd *sdp = ip->i_sbd; | |
610 | struct gfs2_rgrp_list rlist; | |
611 | uint64_t bn, bstart; | |
612 | uint32_t blen; | |
613 | uint64_t *p; | |
614 | unsigned int rg_blocks = 0; | |
615 | int metadata; | |
616 | unsigned int revokes = 0; | |
617 | int x; | |
618 | int error; | |
619 | ||
620 | if (!*top) | |
621 | sm->sm_first = 0; | |
622 | ||
623 | if (height != sm->sm_height) | |
624 | return 0; | |
625 | ||
626 | if (sm->sm_first) { | |
627 | top++; | |
628 | sm->sm_first = 0; | |
629 | } | |
630 | ||
18ec7d5c | 631 | metadata = (height != ip->i_di.di_height - 1); |
b3b94faa DT |
632 | if (metadata) |
633 | revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
634 | ||
635 | error = gfs2_rindex_hold(sdp, &ip->i_alloc.al_ri_gh); | |
636 | if (error) | |
637 | return error; | |
638 | ||
639 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); | |
640 | bstart = 0; | |
641 | blen = 0; | |
642 | ||
643 | for (p = top; p < bottom; p++) { | |
644 | if (!*p) | |
645 | continue; | |
646 | ||
647 | bn = be64_to_cpu(*p); | |
648 | ||
649 | if (bstart + blen == bn) | |
650 | blen++; | |
651 | else { | |
652 | if (bstart) | |
653 | gfs2_rlist_add(sdp, &rlist, bstart); | |
654 | ||
655 | bstart = bn; | |
656 | blen = 1; | |
657 | } | |
658 | } | |
659 | ||
660 | if (bstart) | |
661 | gfs2_rlist_add(sdp, &rlist, bstart); | |
662 | else | |
663 | goto out; /* Nothing to do */ | |
664 | ||
665 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0); | |
666 | ||
667 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
668 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 669 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
b3b94faa DT |
670 | rg_blocks += rgd->rd_ri.ri_length; |
671 | } | |
672 | ||
673 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
674 | if (error) | |
675 | goto out_rlist; | |
676 | ||
677 | error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE + | |
678 | RES_INDIRECT + RES_STATFS + RES_QUOTA, | |
679 | revokes); | |
680 | if (error) | |
681 | goto out_rg_gunlock; | |
682 | ||
683 | down_write(&ip->i_rw_mutex); | |
684 | ||
d4e9c4c3 SW |
685 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
686 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
b3b94faa DT |
687 | |
688 | bstart = 0; | |
689 | blen = 0; | |
690 | ||
691 | for (p = top; p < bottom; p++) { | |
692 | if (!*p) | |
693 | continue; | |
694 | ||
695 | bn = be64_to_cpu(*p); | |
696 | ||
697 | if (bstart + blen == bn) | |
698 | blen++; | |
699 | else { | |
700 | if (bstart) { | |
701 | if (metadata) | |
702 | gfs2_free_meta(ip, bstart, blen); | |
703 | else | |
704 | gfs2_free_data(ip, bstart, blen); | |
705 | } | |
706 | ||
707 | bstart = bn; | |
708 | blen = 1; | |
709 | } | |
710 | ||
711 | *p = 0; | |
712 | if (!ip->i_di.di_blocks) | |
713 | gfs2_consist_inode(ip); | |
714 | ip->i_di.di_blocks--; | |
715 | } | |
716 | if (bstart) { | |
717 | if (metadata) | |
718 | gfs2_free_meta(ip, bstart, blen); | |
719 | else | |
720 | gfs2_free_data(ip, bstart, blen); | |
721 | } | |
722 | ||
723 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
724 | ||
725 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
726 | ||
727 | up_write(&ip->i_rw_mutex); | |
728 | ||
729 | gfs2_trans_end(sdp); | |
730 | ||
731 | out_rg_gunlock: | |
732 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
733 | ||
734 | out_rlist: | |
735 | gfs2_rlist_free(&rlist); | |
736 | ||
737 | out: | |
738 | gfs2_glock_dq_uninit(&ip->i_alloc.al_ri_gh); | |
739 | ||
740 | return error; | |
741 | } | |
742 | ||
743 | /** | |
744 | * do_grow - Make a file look bigger than it is | |
745 | * @ip: the inode | |
746 | * @size: the size to set the file to | |
747 | * | |
748 | * Called with an exclusive lock on @ip. | |
749 | * | |
750 | * Returns: errno | |
751 | */ | |
752 | ||
753 | static int do_grow(struct gfs2_inode *ip, uint64_t size) | |
754 | { | |
755 | struct gfs2_sbd *sdp = ip->i_sbd; | |
756 | struct gfs2_alloc *al; | |
757 | struct buffer_head *dibh; | |
758 | unsigned int h; | |
759 | int error; | |
760 | ||
761 | al = gfs2_alloc_get(ip); | |
762 | ||
763 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
764 | if (error) | |
765 | goto out; | |
766 | ||
767 | error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid); | |
768 | if (error) | |
769 | goto out_gunlock_q; | |
770 | ||
771 | al->al_requested = sdp->sd_max_height + RES_DATA; | |
772 | ||
773 | error = gfs2_inplace_reserve(ip); | |
774 | if (error) | |
775 | goto out_gunlock_q; | |
776 | ||
777 | error = gfs2_trans_begin(sdp, | |
778 | sdp->sd_max_height + al->al_rgd->rd_ri.ri_length + | |
779 | RES_JDATA + RES_DINODE + RES_STATFS + RES_QUOTA, 0); | |
780 | if (error) | |
781 | goto out_ipres; | |
782 | ||
783 | if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { | |
784 | if (gfs2_is_stuffed(ip)) { | |
785 | error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_page, | |
786 | NULL); | |
787 | if (error) | |
788 | goto out_end_trans; | |
789 | } | |
790 | ||
791 | h = calc_tree_height(ip, size); | |
792 | if (ip->i_di.di_height < h) { | |
793 | down_write(&ip->i_rw_mutex); | |
794 | error = build_height(ip, h); | |
795 | up_write(&ip->i_rw_mutex); | |
796 | if (error) | |
797 | goto out_end_trans; | |
798 | } | |
799 | } | |
800 | ||
801 | ip->i_di.di_size = size; | |
802 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
803 | ||
804 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
805 | if (error) | |
806 | goto out_end_trans; | |
807 | ||
d4e9c4c3 | 808 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
809 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
810 | brelse(dibh); | |
811 | ||
812 | out_end_trans: | |
813 | gfs2_trans_end(sdp); | |
814 | ||
815 | out_ipres: | |
816 | gfs2_inplace_release(ip); | |
817 | ||
818 | out_gunlock_q: | |
819 | gfs2_quota_unlock(ip); | |
820 | ||
821 | out: | |
822 | gfs2_alloc_put(ip); | |
823 | ||
824 | return error; | |
825 | } | |
826 | ||
aa6a85a9 | 827 | static int trunc_start(struct gfs2_inode *ip, uint64_t size) |
b3b94faa DT |
828 | { |
829 | struct gfs2_sbd *sdp = ip->i_sbd; | |
830 | struct buffer_head *dibh; | |
831 | int journaled = gfs2_is_jdata(ip); | |
832 | int error; | |
833 | ||
834 | error = gfs2_trans_begin(sdp, | |
835 | RES_DINODE + ((journaled) ? RES_JDATA : 0), 0); | |
836 | if (error) | |
837 | return error; | |
838 | ||
839 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
840 | if (error) | |
841 | goto out; | |
842 | ||
843 | if (gfs2_is_stuffed(ip)) { | |
844 | ip->i_di.di_size = size; | |
845 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
d4e9c4c3 | 846 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
847 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
848 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + size); | |
849 | error = 1; | |
850 | ||
851 | } else { | |
18ec7d5c | 852 | if (size & (uint64_t)(sdp->sd_sb.sb_bsize - 1)) |
257f9b4e | 853 | error = gfs2_block_truncate_page(ip->i_vnode->i_mapping); |
b3b94faa DT |
854 | |
855 | if (!error) { | |
856 | ip->i_di.di_size = size; | |
857 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
858 | ip->i_di.di_flags |= GFS2_DIF_TRUNC_IN_PROG; | |
d4e9c4c3 | 859 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
860 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
861 | } | |
862 | } | |
863 | ||
864 | brelse(dibh); | |
865 | ||
866 | out: | |
867 | gfs2_trans_end(sdp); | |
868 | ||
869 | return error; | |
870 | } | |
871 | ||
872 | static int trunc_dealloc(struct gfs2_inode *ip, uint64_t size) | |
873 | { | |
874 | unsigned int height = ip->i_di.di_height; | |
875 | uint64_t lblock; | |
876 | struct metapath mp; | |
877 | int error; | |
878 | ||
879 | if (!size) | |
880 | lblock = 0; | |
18ec7d5c | 881 | else |
b3b94faa DT |
882 | lblock = (size - 1) >> ip->i_sbd->sd_sb.sb_bsize_shift; |
883 | ||
884 | find_metapath(ip, lblock, &mp); | |
885 | gfs2_alloc_get(ip); | |
886 | ||
887 | error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
888 | if (error) | |
889 | goto out; | |
890 | ||
891 | while (height--) { | |
892 | struct strip_mine sm; | |
893 | sm.sm_first = !!size; | |
894 | sm.sm_height = height; | |
895 | ||
896 | error = recursive_scan(ip, NULL, &mp, 0, 0, 1, do_strip, &sm); | |
897 | if (error) | |
898 | break; | |
899 | } | |
900 | ||
901 | gfs2_quota_unhold(ip); | |
902 | ||
903 | out: | |
904 | gfs2_alloc_put(ip); | |
905 | return error; | |
906 | } | |
907 | ||
908 | static int trunc_end(struct gfs2_inode *ip) | |
909 | { | |
910 | struct gfs2_sbd *sdp = ip->i_sbd; | |
911 | struct buffer_head *dibh; | |
912 | int error; | |
913 | ||
914 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
915 | if (error) | |
916 | return error; | |
917 | ||
918 | down_write(&ip->i_rw_mutex); | |
919 | ||
920 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
921 | if (error) | |
922 | goto out; | |
923 | ||
924 | if (!ip->i_di.di_size) { | |
925 | ip->i_di.di_height = 0; | |
926 | ip->i_di.di_goal_meta = | |
927 | ip->i_di.di_goal_data = | |
928 | ip->i_num.no_addr; | |
929 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); | |
930 | } | |
931 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
932 | ip->i_di.di_flags &= ~GFS2_DIF_TRUNC_IN_PROG; | |
933 | ||
d4e9c4c3 | 934 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
b3b94faa DT |
935 | gfs2_dinode_out(&ip->i_di, dibh->b_data); |
936 | brelse(dibh); | |
937 | ||
938 | out: | |
939 | up_write(&ip->i_rw_mutex); | |
940 | ||
941 | gfs2_trans_end(sdp); | |
942 | ||
943 | return error; | |
944 | } | |
945 | ||
946 | /** | |
947 | * do_shrink - make a file smaller | |
948 | * @ip: the inode | |
949 | * @size: the size to make the file | |
950 | * @truncator: function to truncate the last partial block | |
951 | * | |
952 | * Called with an exclusive lock on @ip. | |
953 | * | |
954 | * Returns: errno | |
955 | */ | |
956 | ||
aa6a85a9 | 957 | static int do_shrink(struct gfs2_inode *ip, uint64_t size) |
b3b94faa DT |
958 | { |
959 | int error; | |
960 | ||
aa6a85a9 | 961 | error = trunc_start(ip, size); |
b3b94faa DT |
962 | if (error < 0) |
963 | return error; | |
964 | if (error > 0) | |
965 | return 0; | |
966 | ||
967 | error = trunc_dealloc(ip, size); | |
968 | if (!error) | |
969 | error = trunc_end(ip); | |
970 | ||
971 | return error; | |
972 | } | |
973 | ||
974 | /** | |
666a2c53 | 975 | * gfs2_truncatei - make a file a given size |
b3b94faa DT |
976 | * @ip: the inode |
977 | * @size: the size to make the file | |
978 | * @truncator: function to truncate the last partial block | |
979 | * | |
980 | * The file size can grow, shrink, or stay the same size. | |
981 | * | |
982 | * Returns: errno | |
983 | */ | |
984 | ||
aa6a85a9 | 985 | int gfs2_truncatei(struct gfs2_inode *ip, uint64_t size) |
b3b94faa DT |
986 | { |
987 | int error; | |
988 | ||
989 | if (gfs2_assert_warn(ip->i_sbd, S_ISREG(ip->i_di.di_mode))) | |
990 | return -EINVAL; | |
991 | ||
992 | if (size > ip->i_di.di_size) | |
993 | error = do_grow(ip, size); | |
994 | else | |
aa6a85a9 | 995 | error = do_shrink(ip, size); |
b3b94faa DT |
996 | |
997 | return error; | |
998 | } | |
999 | ||
1000 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1001 | { | |
1002 | int error; | |
1003 | error = trunc_dealloc(ip, ip->i_di.di_size); | |
1004 | if (!error) | |
1005 | error = trunc_end(ip); | |
1006 | return error; | |
1007 | } | |
1008 | ||
1009 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1010 | { | |
1011 | return trunc_dealloc(ip, 0); | |
1012 | } | |
1013 | ||
1014 | /** | |
1015 | * gfs2_write_calc_reserv - calculate number of blocks needed to write to a file | |
1016 | * @ip: the file | |
1017 | * @len: the number of bytes to be written to the file | |
1018 | * @data_blocks: returns the number of data blocks required | |
1019 | * @ind_blocks: returns the number of indirect blocks required | |
1020 | * | |
1021 | */ | |
1022 | ||
1023 | void gfs2_write_calc_reserv(struct gfs2_inode *ip, unsigned int len, | |
1024 | unsigned int *data_blocks, unsigned int *ind_blocks) | |
1025 | { | |
1026 | struct gfs2_sbd *sdp = ip->i_sbd; | |
1027 | unsigned int tmp; | |
1028 | ||
18ec7d5c | 1029 | if (gfs2_is_dir(ip)) { |
5c676f6d | 1030 | *data_blocks = DIV_ROUND_UP(len, sdp->sd_jbsize) + 2; |
b3b94faa DT |
1031 | *ind_blocks = 3 * (sdp->sd_max_jheight - 1); |
1032 | } else { | |
1033 | *data_blocks = (len >> sdp->sd_sb.sb_bsize_shift) + 3; | |
1034 | *ind_blocks = 3 * (sdp->sd_max_height - 1); | |
1035 | } | |
1036 | ||
1037 | for (tmp = *data_blocks; tmp > sdp->sd_diptrs;) { | |
5c676f6d | 1038 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); |
b3b94faa DT |
1039 | *ind_blocks += tmp; |
1040 | } | |
1041 | } | |
1042 | ||
1043 | /** | |
1044 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
1045 | * @ip: the file being written to | |
1046 | * @offset: the offset to write to | |
1047 | * @len: the number of bytes being written | |
1048 | * @alloc_required: set to 1 if an alloc is required, 0 otherwise | |
1049 | * | |
1050 | * Returns: errno | |
1051 | */ | |
1052 | ||
1053 | int gfs2_write_alloc_required(struct gfs2_inode *ip, uint64_t offset, | |
1054 | unsigned int len, int *alloc_required) | |
1055 | { | |
1056 | struct gfs2_sbd *sdp = ip->i_sbd; | |
1057 | uint64_t lblock, lblock_stop, dblock; | |
1058 | uint32_t extlen; | |
1059 | int new = 0; | |
1060 | int error = 0; | |
1061 | ||
1062 | *alloc_required = 0; | |
1063 | ||
1064 | if (!len) | |
1065 | return 0; | |
1066 | ||
1067 | if (gfs2_is_stuffed(ip)) { | |
1068 | if (offset + len > | |
1069 | sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
1070 | *alloc_required = 1; | |
1071 | return 0; | |
1072 | } | |
1073 | ||
18ec7d5c | 1074 | if (gfs2_is_dir(ip)) { |
b3b94faa DT |
1075 | unsigned int bsize = sdp->sd_jbsize; |
1076 | lblock = offset; | |
1077 | do_div(lblock, bsize); | |
1078 | lblock_stop = offset + len + bsize - 1; | |
1079 | do_div(lblock_stop, bsize); | |
1080 | } else { | |
1081 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
1082 | lblock = offset >> shift; | |
1083 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
1084 | } | |
1085 | ||
1086 | for (; lblock < lblock_stop; lblock += extlen) { | |
1087 | error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen); | |
1088 | if (error) | |
1089 | return error; | |
1090 | ||
1091 | if (!dblock) { | |
1092 | *alloc_required = 1; | |
1093 | return 0; | |
1094 | } | |
1095 | } | |
1096 | ||
1097 | return 0; | |
1098 | } | |
1099 |