Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
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 | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
b3b94faa DT |
10 | #include <linux/spinlock.h> |
11 | #include <linux/completion.h> | |
12 | #include <linux/buffer_head.h> | |
5c676f6d | 13 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 14 | #include <linux/crc32.h> |
b3b94faa DT |
15 | |
16 | #include "gfs2.h" | |
5c676f6d | 17 | #include "incore.h" |
b3b94faa DT |
18 | #include "bmap.h" |
19 | #include "glock.h" | |
20 | #include "inode.h" | |
b3b94faa | 21 | #include "meta_io.h" |
b3b94faa DT |
22 | #include "quota.h" |
23 | #include "rgrp.h" | |
4c16c36a | 24 | #include "super.h" |
b3b94faa | 25 | #include "trans.h" |
18ec7d5c | 26 | #include "dir.h" |
5c676f6d | 27 | #include "util.h" |
63997775 | 28 | #include "trace_gfs2.h" |
b3b94faa DT |
29 | |
30 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
31 | * block is 512, so __u16 is fine for that. It saves stack space to | |
32 | * keep it small. | |
33 | */ | |
34 | struct metapath { | |
dbac6710 | 35 | struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT]; |
b3b94faa DT |
36 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; |
37 | }; | |
38 | ||
b3b94faa DT |
39 | struct strip_mine { |
40 | int sm_first; | |
41 | unsigned int sm_height; | |
42 | }; | |
43 | ||
f25ef0c1 SW |
44 | /** |
45 | * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page | |
46 | * @ip: the inode | |
47 | * @dibh: the dinode buffer | |
48 | * @block: the block number that was allocated | |
ff8f33c8 | 49 | * @page: The (optional) page. This is looked up if @page is NULL |
f25ef0c1 SW |
50 | * |
51 | * Returns: errno | |
52 | */ | |
53 | ||
54 | static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, | |
cd915493 | 55 | u64 block, struct page *page) |
f25ef0c1 | 56 | { |
f25ef0c1 SW |
57 | struct inode *inode = &ip->i_inode; |
58 | struct buffer_head *bh; | |
59 | int release = 0; | |
60 | ||
61 | if (!page || page->index) { | |
62 | page = grab_cache_page(inode->i_mapping, 0); | |
63 | if (!page) | |
64 | return -ENOMEM; | |
65 | release = 1; | |
66 | } | |
67 | ||
68 | if (!PageUptodate(page)) { | |
69 | void *kaddr = kmap(page); | |
602c89d2 SW |
70 | u64 dsize = i_size_read(inode); |
71 | ||
72 | if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode))) | |
73 | dsize = dibh->b_size - sizeof(struct gfs2_dinode); | |
f25ef0c1 | 74 | |
602c89d2 SW |
75 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
76 | memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize); | |
f25ef0c1 SW |
77 | kunmap(page); |
78 | ||
79 | SetPageUptodate(page); | |
80 | } | |
81 | ||
82 | if (!page_has_buffers(page)) | |
83 | create_empty_buffers(page, 1 << inode->i_blkbits, | |
84 | (1 << BH_Uptodate)); | |
85 | ||
86 | bh = page_buffers(page); | |
87 | ||
88 | if (!buffer_mapped(bh)) | |
89 | map_bh(bh, inode->i_sb, block); | |
90 | ||
91 | set_buffer_uptodate(bh); | |
eaf96527 SW |
92 | if (!gfs2_is_jdata(ip)) |
93 | mark_buffer_dirty(bh); | |
bf36a713 | 94 | if (!gfs2_is_writeback(ip)) |
8475487b | 95 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
f25ef0c1 SW |
96 | |
97 | if (release) { | |
98 | unlock_page(page); | |
99 | page_cache_release(page); | |
100 | } | |
101 | ||
102 | return 0; | |
103 | } | |
104 | ||
b3b94faa DT |
105 | /** |
106 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
107 | * @ip: The GFS2 inode to unstuff | |
ff8f33c8 | 108 | * @page: The (optional) page. This is looked up if the @page is NULL |
b3b94faa DT |
109 | * |
110 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
111 | * that the height can be grown in the traditional way. | |
112 | * | |
113 | * Returns: errno | |
114 | */ | |
115 | ||
f25ef0c1 | 116 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
117 | { |
118 | struct buffer_head *bh, *dibh; | |
48516ced | 119 | struct gfs2_dinode *di; |
cd915493 | 120 | u64 block = 0; |
18ec7d5c | 121 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
122 | int error; |
123 | ||
124 | down_write(&ip->i_rw_mutex); | |
125 | ||
126 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
127 | if (error) | |
128 | goto out; | |
907b9bce | 129 | |
a2e0f799 | 130 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
131 | /* Get a free block, fill it with the stuffed data, |
132 | and write it out to disk */ | |
133 | ||
b45e41d7 | 134 | unsigned int n = 1; |
09010978 SW |
135 | error = gfs2_alloc_block(ip, &block, &n); |
136 | if (error) | |
137 | goto out_brelse; | |
18ec7d5c | 138 | if (isdir) { |
5731be53 | 139 | gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 140 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
141 | if (error) |
142 | goto out_brelse; | |
48516ced | 143 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
144 | dibh, sizeof(struct gfs2_dinode)); |
145 | brelse(bh); | |
146 | } else { | |
f25ef0c1 | 147 | error = gfs2_unstuffer_page(ip, dibh, block, page); |
b3b94faa DT |
148 | if (error) |
149 | goto out_brelse; | |
150 | } | |
151 | } | |
152 | ||
153 | /* Set up the pointer to the new block */ | |
154 | ||
d4e9c4c3 | 155 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
48516ced | 156 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
157 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
158 | ||
a2e0f799 | 159 | if (i_size_read(&ip->i_inode)) { |
48516ced | 160 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
161 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
162 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
163 | } |
164 | ||
ecc30c79 | 165 | ip->i_height = 1; |
48516ced | 166 | di->di_height = cpu_to_be16(1); |
b3b94faa | 167 | |
a91ea69f | 168 | out_brelse: |
b3b94faa | 169 | brelse(dibh); |
a91ea69f | 170 | out: |
b3b94faa | 171 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
172 | return error; |
173 | } | |
174 | ||
b3b94faa DT |
175 | |
176 | /** | |
177 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 178 | * @sdp: The superblock |
b3b94faa DT |
179 | * @mp: The metapath to return the result in |
180 | * @block: The disk block to look up | |
9b8c81d1 | 181 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
182 | * |
183 | * This routine returns a struct metapath structure that defines a path | |
184 | * through the metadata of inode "ip" to get to block "block". | |
185 | * | |
186 | * Example: | |
187 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
188 | * filesystem with a blocksize of 4096. | |
189 | * | |
190 | * find_metapath() would return a struct metapath structure set to: | |
191 | * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48, | |
192 | * and mp_list[2] = 165. | |
193 | * | |
194 | * That means that in order to get to the block containing the byte at | |
195 | * offset 101342453, we would load the indirect block pointed to by pointer | |
196 | * 0 in the dinode. We would then load the indirect block pointed to by | |
197 | * pointer 48 in that indirect block. We would then load the data block | |
198 | * pointed to by pointer 165 in that indirect block. | |
199 | * | |
200 | * ---------------------------------------- | |
201 | * | Dinode | | | |
202 | * | | 4| | |
203 | * | |0 1 2 3 4 5 9| | |
204 | * | | 6| | |
205 | * ---------------------------------------- | |
206 | * | | |
207 | * | | |
208 | * V | |
209 | * ---------------------------------------- | |
210 | * | Indirect Block | | |
211 | * | 5| | |
212 | * | 4 4 4 4 4 5 5 1| | |
213 | * |0 5 6 7 8 9 0 1 2| | |
214 | * ---------------------------------------- | |
215 | * | | |
216 | * | | |
217 | * V | |
218 | * ---------------------------------------- | |
219 | * | Indirect Block | | |
220 | * | 1 1 1 1 1 5| | |
221 | * | 6 6 6 6 6 1| | |
222 | * |0 3 4 5 6 7 2| | |
223 | * ---------------------------------------- | |
224 | * | | |
225 | * | | |
226 | * V | |
227 | * ---------------------------------------- | |
228 | * | Data block containing offset | | |
229 | * | 101342453 | | |
230 | * | | | |
231 | * | | | |
232 | * ---------------------------------------- | |
233 | * | |
234 | */ | |
235 | ||
9b8c81d1 SW |
236 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
237 | struct metapath *mp, unsigned int height) | |
b3b94faa | 238 | { |
b3b94faa DT |
239 | unsigned int i; |
240 | ||
9b8c81d1 | 241 | for (i = height; i--;) |
7eabb77e | 242 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
243 | |
244 | } | |
245 | ||
5af4e7a0 | 246 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 247 | { |
5af4e7a0 BM |
248 | if (mp->mp_list[0] == 0) |
249 | return 2; | |
250 | return 1; | |
9b8c81d1 SW |
251 | } |
252 | ||
b3b94faa DT |
253 | /** |
254 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
255 | * @height: The metadata height (0 = dinode) |
256 | * @mp: The metapath | |
257 | * | |
258 | * Return a pointer to the block number of the next height of the metadata | |
259 | * tree given a buffer containing the pointer to the current height of the | |
260 | * metadata tree. | |
261 | */ | |
262 | ||
9b8c81d1 | 263 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 264 | { |
dbac6710 | 265 | struct buffer_head *bh = mp->mp_bh[height]; |
b3b94faa DT |
266 | unsigned int head_size = (height > 0) ? |
267 | sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode); | |
9b8c81d1 | 268 | return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height]; |
b3b94faa DT |
269 | } |
270 | ||
271 | /** | |
9b8c81d1 SW |
272 | * lookup_metapath - Walk the metadata tree to a specific point |
273 | * @ip: The inode | |
b3b94faa | 274 | * @mp: The metapath |
b3b94faa | 275 | * |
9b8c81d1 SW |
276 | * Assumes that the inode's buffer has already been looked up and |
277 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
278 | * by find_metapath(). | |
279 | * | |
280 | * If this function encounters part of the tree which has not been | |
281 | * allocated, it returns the current height of the tree at the point | |
282 | * at which it found the unallocated block. Blocks which are found are | |
283 | * added to the mp->mp_bh[] list. | |
b3b94faa | 284 | * |
9b8c81d1 | 285 | * Returns: error or height of metadata tree |
b3b94faa DT |
286 | */ |
287 | ||
9b8c81d1 | 288 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 289 | { |
11707ea0 SW |
290 | unsigned int end_of_metadata = ip->i_height - 1; |
291 | unsigned int x; | |
9b8c81d1 SW |
292 | __be64 *ptr; |
293 | u64 dblock; | |
e23159d2 | 294 | int ret; |
11707ea0 SW |
295 | |
296 | for (x = 0; x < end_of_metadata; x++) { | |
9b8c81d1 SW |
297 | ptr = metapointer(x, mp); |
298 | dblock = be64_to_cpu(*ptr); | |
299 | if (!dblock) | |
300 | return x + 1; | |
11707ea0 | 301 | |
9b8c81d1 | 302 | ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, 0, &mp->mp_bh[x+1]); |
11707ea0 SW |
303 | if (ret) |
304 | return ret; | |
305 | } | |
306 | ||
9b8c81d1 | 307 | return ip->i_height; |
dbac6710 SW |
308 | } |
309 | ||
9b8c81d1 | 310 | static inline void release_metapath(struct metapath *mp) |
dbac6710 SW |
311 | { |
312 | int i; | |
313 | ||
9b8c81d1 SW |
314 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
315 | if (mp->mp_bh[i] == NULL) | |
316 | break; | |
317 | brelse(mp->mp_bh[i]); | |
318 | } | |
11707ea0 SW |
319 | } |
320 | ||
30cbf189 SW |
321 | /** |
322 | * gfs2_extent_length - Returns length of an extent of blocks | |
323 | * @start: Start of the buffer | |
324 | * @len: Length of the buffer in bytes | |
325 | * @ptr: Current position in the buffer | |
326 | * @limit: Max extent length to return (0 = unlimited) | |
327 | * @eob: Set to 1 if we hit "end of block" | |
328 | * | |
329 | * If the first block is zero (unallocated) it will return the number of | |
330 | * unallocated blocks in the extent, otherwise it will return the number | |
331 | * of contiguous blocks in the extent. | |
332 | * | |
333 | * Returns: The length of the extent (minimum of one block) | |
334 | */ | |
335 | ||
336 | static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob) | |
337 | { | |
338 | const __be64 *end = (start + len); | |
339 | const __be64 *first = ptr; | |
340 | u64 d = be64_to_cpu(*ptr); | |
341 | ||
342 | *eob = 0; | |
343 | do { | |
344 | ptr++; | |
345 | if (ptr >= end) | |
346 | break; | |
347 | if (limit && --limit == 0) | |
348 | break; | |
349 | if (d) | |
350 | d++; | |
351 | } while(be64_to_cpu(*ptr) == d); | |
352 | if (ptr >= end) | |
353 | *eob = 1; | |
354 | return (ptr - first); | |
355 | } | |
356 | ||
9b8c81d1 | 357 | static inline void bmap_lock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 358 | { |
4cf1ed81 SW |
359 | if (create) |
360 | down_write(&ip->i_rw_mutex); | |
361 | else | |
362 | down_read(&ip->i_rw_mutex); | |
363 | } | |
364 | ||
9b8c81d1 | 365 | static inline void bmap_unlock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 366 | { |
4cf1ed81 SW |
367 | if (create) |
368 | up_write(&ip->i_rw_mutex); | |
369 | else | |
370 | up_read(&ip->i_rw_mutex); | |
371 | } | |
372 | ||
9b8c81d1 SW |
373 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
374 | struct gfs2_glock *gl, unsigned int i, | |
375 | unsigned offset, u64 bn) | |
376 | { | |
377 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
378 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
379 | sizeof(struct gfs2_dinode))); | |
380 | BUG_ON(i < 1); | |
381 | BUG_ON(mp->mp_bh[i] != NULL); | |
382 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
383 | gfs2_trans_add_bh(gl, mp->mp_bh[i], 1); | |
384 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); | |
385 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
386 | ptr += offset; | |
387 | *ptr = cpu_to_be64(bn); | |
388 | return ptr; | |
389 | } | |
390 | ||
391 | enum alloc_state { | |
392 | ALLOC_DATA = 0, | |
393 | ALLOC_GROW_DEPTH = 1, | |
394 | ALLOC_GROW_HEIGHT = 2, | |
395 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
396 | }; | |
397 | ||
398 | /** | |
399 | * gfs2_bmap_alloc - Build a metadata tree of the requested height | |
400 | * @inode: The GFS2 inode | |
401 | * @lblock: The logical starting block of the extent | |
402 | * @bh_map: This is used to return the mapping details | |
403 | * @mp: The metapath | |
404 | * @sheight: The starting height (i.e. whats already mapped) | |
405 | * @height: The height to build to | |
406 | * @maxlen: The max number of data blocks to alloc | |
407 | * | |
408 | * In this routine we may have to alloc: | |
409 | * i) Indirect blocks to grow the metadata tree height | |
410 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
411 | * iii) Data blocks | |
412 | * | |
413 | * The function is in two parts. The first part works out the total | |
414 | * number of blocks which we need. The second part does the actual | |
415 | * allocation asking for an extent at a time (if enough contiguous free | |
416 | * blocks are available, there will only be one request per bmap call) | |
417 | * and uses the state machine to initialise the blocks in order. | |
418 | * | |
419 | * Returns: errno on error | |
420 | */ | |
421 | ||
422 | static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock, | |
423 | struct buffer_head *bh_map, struct metapath *mp, | |
424 | const unsigned int sheight, | |
425 | const unsigned int height, | |
426 | const unsigned int maxlen) | |
427 | { | |
428 | struct gfs2_inode *ip = GFS2_I(inode); | |
429 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
430 | struct buffer_head *dibh = mp->mp_bh[0]; | |
431 | u64 bn, dblock = 0; | |
5af4e7a0 | 432 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
9b8c81d1 SW |
433 | unsigned dblks = 0; |
434 | unsigned ptrs_per_blk; | |
435 | const unsigned end_of_metadata = height - 1; | |
436 | int eob = 0; | |
437 | enum alloc_state state; | |
438 | __be64 *ptr; | |
439 | __be64 zero_bn = 0; | |
440 | ||
441 | BUG_ON(sheight < 1); | |
442 | BUG_ON(dibh == NULL); | |
443 | ||
444 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
445 | ||
446 | if (height == sheight) { | |
447 | struct buffer_head *bh; | |
448 | /* Bottom indirect block exists, find unalloced extent size */ | |
449 | ptr = metapointer(end_of_metadata, mp); | |
450 | bh = mp->mp_bh[end_of_metadata]; | |
451 | dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, | |
452 | &eob); | |
453 | BUG_ON(dblks < 1); | |
454 | state = ALLOC_DATA; | |
455 | } else { | |
456 | /* Need to allocate indirect blocks */ | |
457 | ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs; | |
458 | dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]); | |
459 | if (height == ip->i_height) { | |
460 | /* Writing into existing tree, extend tree down */ | |
461 | iblks = height - sheight; | |
462 | state = ALLOC_GROW_DEPTH; | |
463 | } else { | |
464 | /* Building up tree height */ | |
465 | state = ALLOC_GROW_HEIGHT; | |
466 | iblks = height - ip->i_height; | |
5af4e7a0 BM |
467 | branch_start = metapath_branch_start(mp); |
468 | iblks += (height - branch_start); | |
9b8c81d1 SW |
469 | } |
470 | } | |
471 | ||
472 | /* start of the second part of the function (state machine) */ | |
473 | ||
474 | blks = dblks + iblks; | |
475 | i = sheight; | |
476 | do { | |
09010978 | 477 | int error; |
9b8c81d1 | 478 | n = blks - alloced; |
09010978 SW |
479 | error = gfs2_alloc_block(ip, &bn, &n); |
480 | if (error) | |
481 | return error; | |
9b8c81d1 SW |
482 | alloced += n; |
483 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
484 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
485 | switch (state) { | |
486 | /* Growing height of tree */ | |
487 | case ALLOC_GROW_HEIGHT: | |
488 | if (i == 1) { | |
489 | ptr = (__be64 *)(dibh->b_data + | |
490 | sizeof(struct gfs2_dinode)); | |
491 | zero_bn = *ptr; | |
492 | } | |
493 | for (; i - 1 < height - ip->i_height && n > 0; i++, n--) | |
494 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); | |
495 | if (i - 1 == height - ip->i_height) { | |
496 | i--; | |
497 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
498 | sizeof(struct gfs2_meta_header), | |
499 | dibh, sizeof(struct gfs2_dinode)); | |
500 | gfs2_buffer_clear_tail(dibh, | |
501 | sizeof(struct gfs2_dinode) + | |
502 | sizeof(__be64)); | |
503 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
504 | sizeof(struct gfs2_meta_header)); | |
505 | *ptr = zero_bn; | |
506 | state = ALLOC_GROW_DEPTH; | |
5af4e7a0 | 507 | for(i = branch_start; i < height; i++) { |
9b8c81d1 SW |
508 | if (mp->mp_bh[i] == NULL) |
509 | break; | |
510 | brelse(mp->mp_bh[i]); | |
511 | mp->mp_bh[i] = NULL; | |
512 | } | |
5af4e7a0 | 513 | i = branch_start; |
9b8c81d1 SW |
514 | } |
515 | if (n == 0) | |
516 | break; | |
517 | /* Branching from existing tree */ | |
518 | case ALLOC_GROW_DEPTH: | |
519 | if (i > 1 && i < height) | |
520 | gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1); | |
521 | for (; i < height && n > 0; i++, n--) | |
522 | gfs2_indirect_init(mp, ip->i_gl, i, | |
523 | mp->mp_list[i-1], bn++); | |
524 | if (i == height) | |
525 | state = ALLOC_DATA; | |
526 | if (n == 0) | |
527 | break; | |
528 | /* Tree complete, adding data blocks */ | |
529 | case ALLOC_DATA: | |
530 | BUG_ON(n > dblks); | |
531 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); | |
532 | gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1); | |
533 | dblks = n; | |
534 | ptr = metapointer(end_of_metadata, mp); | |
535 | dblock = bn; | |
536 | while (n-- > 0) | |
537 | *ptr++ = cpu_to_be64(bn++); | |
538 | break; | |
539 | } | |
07ccb7bf | 540 | } while ((state != ALLOC_DATA) || !dblock); |
9b8c81d1 SW |
541 | |
542 | ip->i_height = height; | |
543 | gfs2_add_inode_blocks(&ip->i_inode, alloced); | |
544 | gfs2_dinode_out(ip, mp->mp_bh[0]->b_data); | |
545 | map_bh(bh_map, inode->i_sb, dblock); | |
546 | bh_map->b_size = dblks << inode->i_blkbits; | |
547 | set_buffer_new(bh_map); | |
548 | return 0; | |
549 | } | |
550 | ||
b3b94faa | 551 | /** |
4cf1ed81 | 552 | * gfs2_block_map - Map a block from an inode to a disk block |
fd88de56 | 553 | * @inode: The inode |
b3b94faa | 554 | * @lblock: The logical block number |
4cf1ed81 | 555 | * @bh_map: The bh to be mapped |
9b8c81d1 | 556 | * @create: True if its ok to alloc blocks to satify the request |
b3b94faa | 557 | * |
9b8c81d1 SW |
558 | * Sets buffer_mapped() if successful, sets buffer_boundary() if a |
559 | * read of metadata will be required before the next block can be | |
560 | * mapped. Sets buffer_new() if new blocks were allocated. | |
b3b94faa DT |
561 | * |
562 | * Returns: errno | |
563 | */ | |
564 | ||
e9e1ef2b BP |
565 | int gfs2_block_map(struct inode *inode, sector_t lblock, |
566 | struct buffer_head *bh_map, int create) | |
b3b94faa | 567 | { |
feaa7bba SW |
568 | struct gfs2_inode *ip = GFS2_I(inode); |
569 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
ecc30c79 | 570 | unsigned int bsize = sdp->sd_sb.sb_bsize; |
9b8c81d1 | 571 | const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits; |
ecc30c79 | 572 | const u64 *arr = sdp->sd_heightsize; |
9b8c81d1 SW |
573 | __be64 *ptr; |
574 | u64 size; | |
575 | struct metapath mp; | |
576 | int ret; | |
577 | int eob; | |
578 | unsigned int len; | |
579 | struct buffer_head *bh; | |
580 | u8 height; | |
7276b3b0 | 581 | |
9b8c81d1 | 582 | BUG_ON(maxlen == 0); |
b3b94faa | 583 | |
dbac6710 | 584 | memset(mp.mp_bh, 0, sizeof(mp.mp_bh)); |
9b8c81d1 | 585 | bmap_lock(ip, create); |
4cf1ed81 SW |
586 | clear_buffer_mapped(bh_map); |
587 | clear_buffer_new(bh_map); | |
588 | clear_buffer_boundary(bh_map); | |
63997775 | 589 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); |
ecc30c79 SW |
590 | if (gfs2_is_dir(ip)) { |
591 | bsize = sdp->sd_jbsize; | |
592 | arr = sdp->sd_jheightsize; | |
593 | } | |
4cf1ed81 | 594 | |
9b8c81d1 SW |
595 | ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]); |
596 | if (ret) | |
597 | goto out; | |
b3b94faa | 598 | |
9b8c81d1 SW |
599 | height = ip->i_height; |
600 | size = (lblock + 1) * bsize; | |
601 | while (size > arr[height]) | |
602 | height++; | |
603 | find_metapath(sdp, lblock, &mp, height); | |
604 | ret = 1; | |
605 | if (height > ip->i_height || gfs2_is_stuffed(ip)) | |
606 | goto do_alloc; | |
607 | ret = lookup_metapath(ip, &mp); | |
608 | if (ret < 0) | |
609 | goto out; | |
610 | if (ret != ip->i_height) | |
611 | goto do_alloc; | |
612 | ptr = metapointer(ip->i_height - 1, &mp); | |
613 | if (*ptr == 0) | |
614 | goto do_alloc; | |
615 | map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr)); | |
616 | bh = mp.mp_bh[ip->i_height - 1]; | |
617 | len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob); | |
618 | bh_map->b_size = (len << inode->i_blkbits); | |
619 | if (eob) | |
620 | set_buffer_boundary(bh_map); | |
621 | ret = 0; | |
622 | out: | |
623 | release_metapath(&mp); | |
63997775 | 624 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); |
9b8c81d1 SW |
625 | bmap_unlock(ip, create); |
626 | return ret; | |
30cbf189 | 627 | |
9b8c81d1 SW |
628 | do_alloc: |
629 | /* All allocations are done here, firstly check create flag */ | |
630 | if (!create) { | |
631 | BUG_ON(gfs2_is_stuffed(ip)); | |
632 | ret = 0; | |
633 | goto out; | |
b3b94faa | 634 | } |
9b8c81d1 SW |
635 | |
636 | /* At this point ret is the tree depth of already allocated blocks */ | |
637 | ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen); | |
638 | goto out; | |
fd88de56 SW |
639 | } |
640 | ||
941e6d7d SW |
641 | /* |
642 | * Deprecated: do not use in new code | |
643 | */ | |
fd88de56 SW |
644 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
645 | { | |
23591256 | 646 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 647 | int ret; |
fd88de56 SW |
648 | int create = *new; |
649 | ||
650 | BUG_ON(!extlen); | |
651 | BUG_ON(!dblock); | |
652 | BUG_ON(!new); | |
653 | ||
9b8c81d1 | 654 | bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 655 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
656 | *extlen = bh.b_size >> inode->i_blkbits; |
657 | *dblock = bh.b_blocknr; | |
658 | if (buffer_new(&bh)) | |
659 | *new = 1; | |
660 | else | |
661 | *new = 0; | |
662 | return ret; | |
b3b94faa DT |
663 | } |
664 | ||
b3b94faa DT |
665 | /** |
666 | * do_strip - Look for a layer a particular layer of the file and strip it off | |
667 | * @ip: the inode | |
668 | * @dibh: the dinode buffer | |
669 | * @bh: A buffer of pointers | |
670 | * @top: The first pointer in the buffer | |
671 | * @bottom: One more than the last pointer | |
672 | * @height: the height this buffer is at | |
673 | * @data: a pointer to a struct strip_mine | |
674 | * | |
675 | * Returns: errno | |
676 | */ | |
677 | ||
678 | static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh, | |
b44b84d7 | 679 | struct buffer_head *bh, __be64 *top, __be64 *bottom, |
d56fa8a1 | 680 | unsigned int height, struct strip_mine *sm) |
b3b94faa | 681 | { |
feaa7bba | 682 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa | 683 | struct gfs2_rgrp_list rlist; |
cd915493 | 684 | u64 bn, bstart; |
4c16c36a | 685 | u32 blen, btotal; |
b44b84d7 | 686 | __be64 *p; |
b3b94faa DT |
687 | unsigned int rg_blocks = 0; |
688 | int metadata; | |
689 | unsigned int revokes = 0; | |
690 | int x; | |
e06dfc49 | 691 | int error = 0; |
b3b94faa DT |
692 | |
693 | if (!*top) | |
694 | sm->sm_first = 0; | |
695 | ||
696 | if (height != sm->sm_height) | |
697 | return 0; | |
698 | ||
699 | if (sm->sm_first) { | |
700 | top++; | |
701 | sm->sm_first = 0; | |
702 | } | |
703 | ||
ecc30c79 | 704 | metadata = (height != ip->i_height - 1); |
b3b94faa DT |
705 | if (metadata) |
706 | revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
6d3117b4 SW |
707 | else if (ip->i_depth) |
708 | revokes = sdp->sd_inptrs; | |
b3b94faa | 709 | |
b3b94faa DT |
710 | if (error) |
711 | return error; | |
712 | ||
713 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); | |
714 | bstart = 0; | |
715 | blen = 0; | |
716 | ||
717 | for (p = top; p < bottom; p++) { | |
718 | if (!*p) | |
719 | continue; | |
720 | ||
721 | bn = be64_to_cpu(*p); | |
722 | ||
723 | if (bstart + blen == bn) | |
724 | blen++; | |
725 | else { | |
726 | if (bstart) | |
727 | gfs2_rlist_add(sdp, &rlist, bstart); | |
728 | ||
729 | bstart = bn; | |
730 | blen = 1; | |
731 | } | |
732 | } | |
733 | ||
734 | if (bstart) | |
735 | gfs2_rlist_add(sdp, &rlist, bstart); | |
736 | else | |
737 | goto out; /* Nothing to do */ | |
738 | ||
fe6c991c | 739 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
740 | |
741 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
742 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 743 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 744 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
745 | } |
746 | ||
747 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
748 | if (error) | |
749 | goto out_rlist; | |
750 | ||
751 | error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE + | |
752 | RES_INDIRECT + RES_STATFS + RES_QUOTA, | |
753 | revokes); | |
754 | if (error) | |
755 | goto out_rg_gunlock; | |
756 | ||
757 | down_write(&ip->i_rw_mutex); | |
758 | ||
d4e9c4c3 SW |
759 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
760 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
b3b94faa DT |
761 | |
762 | bstart = 0; | |
763 | blen = 0; | |
4c16c36a | 764 | btotal = 0; |
b3b94faa DT |
765 | |
766 | for (p = top; p < bottom; p++) { | |
767 | if (!*p) | |
768 | continue; | |
769 | ||
770 | bn = be64_to_cpu(*p); | |
771 | ||
772 | if (bstart + blen == bn) | |
773 | blen++; | |
774 | else { | |
775 | if (bstart) { | |
46fcb2ed | 776 | __gfs2_free_blocks(ip, bstart, blen, metadata); |
4c16c36a | 777 | btotal += blen; |
b3b94faa DT |
778 | } |
779 | ||
780 | bstart = bn; | |
781 | blen = 1; | |
782 | } | |
783 | ||
784 | *p = 0; | |
77658aad | 785 | gfs2_add_inode_blocks(&ip->i_inode, -1); |
b3b94faa DT |
786 | } |
787 | if (bstart) { | |
46fcb2ed | 788 | __gfs2_free_blocks(ip, bstart, blen, metadata); |
4c16c36a | 789 | btotal += blen; |
b3b94faa DT |
790 | } |
791 | ||
4c16c36a BP |
792 | gfs2_statfs_change(sdp, 0, +btotal, 0); |
793 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
794 | ip->i_inode.i_gid); | |
795 | ||
4bd91ba1 | 796 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
b3b94faa | 797 | |
539e5d6b | 798 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
799 | |
800 | up_write(&ip->i_rw_mutex); | |
801 | ||
802 | gfs2_trans_end(sdp); | |
803 | ||
a91ea69f | 804 | out_rg_gunlock: |
b3b94faa | 805 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 806 | out_rlist: |
b3b94faa | 807 | gfs2_rlist_free(&rlist); |
a91ea69f | 808 | out: |
b3b94faa DT |
809 | return error; |
810 | } | |
811 | ||
d56fa8a1 SW |
812 | /** |
813 | * recursive_scan - recursively scan through the end of a file | |
814 | * @ip: the inode | |
815 | * @dibh: the dinode buffer | |
816 | * @mp: the path through the metadata to the point to start | |
817 | * @height: the height the recursion is at | |
818 | * @block: the indirect block to look at | |
819 | * @first: 1 if this is the first block | |
820 | * @sm: data opaque to this function to pass to @bc | |
821 | * | |
822 | * When this is first called @height and @block should be zero and | |
823 | * @first should be 1. | |
824 | * | |
825 | * Returns: errno | |
826 | */ | |
827 | ||
828 | static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh, | |
829 | struct metapath *mp, unsigned int height, | |
830 | u64 block, int first, struct strip_mine *sm) | |
831 | { | |
832 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
833 | struct buffer_head *bh = NULL; | |
834 | __be64 *top, *bottom; | |
835 | u64 bn; | |
836 | int error; | |
837 | int mh_size = sizeof(struct gfs2_meta_header); | |
838 | ||
839 | if (!height) { | |
840 | error = gfs2_meta_inode_buffer(ip, &bh); | |
841 | if (error) | |
842 | return error; | |
843 | dibh = bh; | |
844 | ||
845 | top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0]; | |
846 | bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs; | |
847 | } else { | |
848 | error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh); | |
849 | if (error) | |
850 | return error; | |
851 | ||
852 | top = (__be64 *)(bh->b_data + mh_size) + | |
853 | (first ? mp->mp_list[height] : 0); | |
854 | ||
855 | bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs; | |
856 | } | |
857 | ||
858 | error = do_strip(ip, dibh, bh, top, bottom, height, sm); | |
859 | if (error) | |
860 | goto out; | |
861 | ||
862 | if (height < ip->i_height - 1) | |
863 | for (; top < bottom; top++, first = 0) { | |
864 | if (!*top) | |
865 | continue; | |
866 | ||
867 | bn = be64_to_cpu(*top); | |
868 | ||
869 | error = recursive_scan(ip, dibh, mp, height + 1, bn, | |
870 | first, sm); | |
871 | if (error) | |
872 | break; | |
873 | } | |
874 | ||
875 | out: | |
876 | brelse(bh); | |
877 | return error; | |
878 | } | |
879 | ||
880 | ||
ba7f7290 SW |
881 | /** |
882 | * gfs2_block_truncate_page - Deal with zeroing out data for truncate | |
883 | * | |
884 | * This is partly borrowed from ext3. | |
885 | */ | |
ff8f33c8 | 886 | static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from) |
ba7f7290 SW |
887 | { |
888 | struct inode *inode = mapping->host; | |
889 | struct gfs2_inode *ip = GFS2_I(inode); | |
ba7f7290 SW |
890 | unsigned long index = from >> PAGE_CACHE_SHIFT; |
891 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
892 | unsigned blocksize, iblock, length, pos; | |
893 | struct buffer_head *bh; | |
894 | struct page *page; | |
ba7f7290 SW |
895 | int err; |
896 | ||
897 | page = grab_cache_page(mapping, index); | |
898 | if (!page) | |
899 | return 0; | |
900 | ||
901 | blocksize = inode->i_sb->s_blocksize; | |
902 | length = blocksize - (offset & (blocksize - 1)); | |
903 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
904 | ||
905 | if (!page_has_buffers(page)) | |
906 | create_empty_buffers(page, blocksize, 0); | |
907 | ||
908 | /* Find the buffer that contains "offset" */ | |
909 | bh = page_buffers(page); | |
910 | pos = blocksize; | |
911 | while (offset >= pos) { | |
912 | bh = bh->b_this_page; | |
913 | iblock++; | |
914 | pos += blocksize; | |
915 | } | |
916 | ||
917 | err = 0; | |
918 | ||
919 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 920 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
921 | /* unmapped? It's a hole - nothing to do */ |
922 | if (!buffer_mapped(bh)) | |
923 | goto unlock; | |
924 | } | |
925 | ||
926 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
927 | if (PageUptodate(page)) | |
928 | set_buffer_uptodate(bh); | |
929 | ||
930 | if (!buffer_uptodate(bh)) { | |
931 | err = -EIO; | |
932 | ll_rw_block(READ, 1, &bh); | |
933 | wait_on_buffer(bh); | |
934 | /* Uhhuh. Read error. Complain and punt. */ | |
935 | if (!buffer_uptodate(bh)) | |
936 | goto unlock; | |
1875f2f3 | 937 | err = 0; |
ba7f7290 SW |
938 | } |
939 | ||
bf36a713 | 940 | if (!gfs2_is_writeback(ip)) |
ba7f7290 SW |
941 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
942 | ||
eebd2aa3 | 943 | zero_user(page, offset, length); |
40bc9a27 | 944 | mark_buffer_dirty(bh); |
ba7f7290 SW |
945 | unlock: |
946 | unlock_page(page); | |
947 | page_cache_release(page); | |
948 | return err; | |
949 | } | |
950 | ||
ff8f33c8 | 951 | static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize) |
b3b94faa | 952 | { |
ff8f33c8 SW |
953 | struct gfs2_inode *ip = GFS2_I(inode); |
954 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
955 | struct address_space *mapping = inode->i_mapping; | |
b3b94faa DT |
956 | struct buffer_head *dibh; |
957 | int journaled = gfs2_is_jdata(ip); | |
958 | int error; | |
959 | ||
960 | error = gfs2_trans_begin(sdp, | |
c5392124 | 961 | RES_DINODE + (journaled ? RES_JDATA : 0), 0); |
b3b94faa DT |
962 | if (error) |
963 | return error; | |
964 | ||
965 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
966 | if (error) | |
967 | goto out; | |
968 | ||
ff8f33c8 SW |
969 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
970 | ||
b3b94faa | 971 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 972 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 973 | } else { |
ff8f33c8 SW |
974 | if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) { |
975 | error = gfs2_block_truncate_page(mapping, newsize); | |
976 | if (error) | |
977 | goto out_brelse; | |
b3b94faa | 978 | } |
ff8f33c8 | 979 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
980 | } |
981 | ||
ff8f33c8 | 982 | i_size_write(inode, newsize); |
ff8f33c8 SW |
983 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
984 | gfs2_dinode_out(ip, dibh->b_data); | |
b3b94faa | 985 | |
ff8f33c8 SW |
986 | truncate_pagecache(inode, oldsize, newsize); |
987 | out_brelse: | |
988 | brelse(dibh); | |
a91ea69f | 989 | out: |
b3b94faa | 990 | gfs2_trans_end(sdp); |
b3b94faa DT |
991 | return error; |
992 | } | |
993 | ||
cd915493 | 994 | static int trunc_dealloc(struct gfs2_inode *ip, u64 size) |
b3b94faa | 995 | { |
9b8c81d1 | 996 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
ecc30c79 | 997 | unsigned int height = ip->i_height; |
cd915493 | 998 | u64 lblock; |
b3b94faa DT |
999 | struct metapath mp; |
1000 | int error; | |
1001 | ||
1002 | if (!size) | |
1003 | lblock = 0; | |
18ec7d5c | 1004 | else |
9b8c81d1 | 1005 | lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 1006 | |
9b8c81d1 | 1007 | find_metapath(sdp, lblock, &mp, ip->i_height); |
182fe5ab CG |
1008 | if (!gfs2_alloc_get(ip)) |
1009 | return -ENOMEM; | |
b3b94faa DT |
1010 | |
1011 | error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
1012 | if (error) | |
1013 | goto out; | |
1014 | ||
1015 | while (height--) { | |
1016 | struct strip_mine sm; | |
1017 | sm.sm_first = !!size; | |
1018 | sm.sm_height = height; | |
1019 | ||
d56fa8a1 | 1020 | error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm); |
b3b94faa DT |
1021 | if (error) |
1022 | break; | |
1023 | } | |
1024 | ||
1025 | gfs2_quota_unhold(ip); | |
1026 | ||
a91ea69f | 1027 | out: |
b3b94faa DT |
1028 | gfs2_alloc_put(ip); |
1029 | return error; | |
1030 | } | |
1031 | ||
1032 | static int trunc_end(struct gfs2_inode *ip) | |
1033 | { | |
feaa7bba | 1034 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1035 | struct buffer_head *dibh; |
1036 | int error; | |
1037 | ||
1038 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1039 | if (error) | |
1040 | return error; | |
1041 | ||
1042 | down_write(&ip->i_rw_mutex); | |
1043 | ||
1044 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1045 | if (error) | |
1046 | goto out; | |
1047 | ||
a2e0f799 | 1048 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1049 | ip->i_height = 0; |
ce276b06 | 1050 | ip->i_goal = ip->i_no_addr; |
b3b94faa DT |
1051 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
1052 | } | |
4bd91ba1 | 1053 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
383f01fb | 1054 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1055 | |
d4e9c4c3 | 1056 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
539e5d6b | 1057 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1058 | brelse(dibh); |
1059 | ||
a91ea69f | 1060 | out: |
b3b94faa | 1061 | up_write(&ip->i_rw_mutex); |
b3b94faa | 1062 | gfs2_trans_end(sdp); |
b3b94faa DT |
1063 | return error; |
1064 | } | |
1065 | ||
1066 | /** | |
1067 | * do_shrink - make a file smaller | |
ff8f33c8 SW |
1068 | * @inode: the inode |
1069 | * @oldsize: the current inode size | |
1070 | * @newsize: the size to make the file | |
b3b94faa | 1071 | * |
ff8f33c8 SW |
1072 | * Called with an exclusive lock on @inode. The @size must |
1073 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
1074 | * |
1075 | * Returns: errno | |
1076 | */ | |
1077 | ||
ff8f33c8 | 1078 | static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize) |
b3b94faa | 1079 | { |
ff8f33c8 | 1080 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
1081 | int error; |
1082 | ||
ff8f33c8 | 1083 | error = trunc_start(inode, oldsize, newsize); |
b3b94faa DT |
1084 | if (error < 0) |
1085 | return error; | |
ff8f33c8 | 1086 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
1087 | return 0; |
1088 | ||
ff8f33c8 SW |
1089 | error = trunc_dealloc(ip, newsize); |
1090 | if (error == 0) | |
b3b94faa DT |
1091 | error = trunc_end(ip); |
1092 | ||
1093 | return error; | |
1094 | } | |
1095 | ||
ff8f33c8 | 1096 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 1097 | { |
ff8f33c8 SW |
1098 | u64 size = inode->i_size; |
1099 | int ret; | |
1100 | ||
1101 | ret = do_shrink(inode, size, size); | |
1102 | WARN_ON(ret != 0); | |
1103 | } | |
1104 | ||
1105 | /** | |
1106 | * do_grow - Touch and update inode size | |
1107 | * @inode: The inode | |
1108 | * @size: The new size | |
1109 | * | |
1110 | * This function updates the timestamps on the inode and | |
1111 | * may also increase the size of the inode. This function | |
1112 | * must not be called with @size any smaller than the current | |
1113 | * inode size. | |
1114 | * | |
1115 | * Although it is not strictly required to unstuff files here, | |
1116 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
1117 | * code which will result in a buffer overrun if the size is larger | |
1118 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 1119 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
1120 | * just update the inode size directly. |
1121 | * | |
1122 | * Returns: 0 on success, or -ve on error | |
1123 | */ | |
1124 | ||
1125 | static int do_grow(struct inode *inode, u64 size) | |
1126 | { | |
1127 | struct gfs2_inode *ip = GFS2_I(inode); | |
1128 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
a13b8c5f | 1129 | struct buffer_head *dibh; |
ff8f33c8 | 1130 | struct gfs2_alloc *al = NULL; |
a13b8c5f WC |
1131 | int error; |
1132 | ||
ff8f33c8 SW |
1133 | if (gfs2_is_stuffed(ip) && |
1134 | (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) { | |
1135 | al = gfs2_alloc_get(ip); | |
1136 | if (al == NULL) | |
1137 | return -ENOMEM; | |
1138 | ||
1139 | error = gfs2_quota_lock_check(ip); | |
1140 | if (error) | |
1141 | goto do_grow_alloc_put; | |
1142 | ||
1143 | al->al_requested = 1; | |
1144 | error = gfs2_inplace_reserve(ip); | |
1145 | if (error) | |
1146 | goto do_grow_qunlock; | |
1147 | } | |
1148 | ||
bf97b673 | 1149 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0); |
a13b8c5f | 1150 | if (error) |
ff8f33c8 | 1151 | goto do_grow_release; |
a13b8c5f | 1152 | |
ff8f33c8 SW |
1153 | if (al) { |
1154 | error = gfs2_unstuff_dinode(ip, NULL); | |
1155 | if (error) | |
1156 | goto do_end_trans; | |
1157 | } | |
a13b8c5f WC |
1158 | |
1159 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1160 | if (error) | |
ff8f33c8 | 1161 | goto do_end_trans; |
a13b8c5f | 1162 | |
ff8f33c8 | 1163 | i_size_write(inode, size); |
a13b8c5f WC |
1164 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
1165 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
1166 | gfs2_dinode_out(ip, dibh->b_data); | |
1167 | brelse(dibh); | |
1168 | ||
ff8f33c8 | 1169 | do_end_trans: |
a13b8c5f | 1170 | gfs2_trans_end(sdp); |
ff8f33c8 SW |
1171 | do_grow_release: |
1172 | if (al) { | |
1173 | gfs2_inplace_release(ip); | |
1174 | do_grow_qunlock: | |
1175 | gfs2_quota_unlock(ip); | |
1176 | do_grow_alloc_put: | |
1177 | gfs2_alloc_put(ip); | |
1178 | } | |
a13b8c5f WC |
1179 | return error; |
1180 | } | |
1181 | ||
b3b94faa | 1182 | /** |
ff8f33c8 SW |
1183 | * gfs2_setattr_size - make a file a given size |
1184 | * @inode: the inode | |
1185 | * @newsize: the size to make the file | |
b3b94faa | 1186 | * |
ff8f33c8 SW |
1187 | * The file size can grow, shrink, or stay the same size. This |
1188 | * is called holding i_mutex and an exclusive glock on the inode | |
1189 | * in question. | |
b3b94faa DT |
1190 | * |
1191 | * Returns: errno | |
1192 | */ | |
1193 | ||
ff8f33c8 | 1194 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 1195 | { |
ff8f33c8 SW |
1196 | int ret; |
1197 | u64 oldsize; | |
b3b94faa | 1198 | |
ff8f33c8 | 1199 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 1200 | |
ff8f33c8 SW |
1201 | ret = inode_newsize_ok(inode, newsize); |
1202 | if (ret) | |
1203 | return ret; | |
b3b94faa | 1204 | |
562c72aa CH |
1205 | inode_dio_wait(inode); |
1206 | ||
ff8f33c8 SW |
1207 | oldsize = inode->i_size; |
1208 | if (newsize >= oldsize) | |
1209 | return do_grow(inode, newsize); | |
1210 | ||
1211 | return do_shrink(inode, oldsize, newsize); | |
b3b94faa DT |
1212 | } |
1213 | ||
1214 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1215 | { | |
1216 | int error; | |
a2e0f799 | 1217 | error = trunc_dealloc(ip, i_size_read(&ip->i_inode)); |
b3b94faa DT |
1218 | if (!error) |
1219 | error = trunc_end(ip); | |
1220 | return error; | |
1221 | } | |
1222 | ||
1223 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1224 | { | |
1225 | return trunc_dealloc(ip, 0); | |
1226 | } | |
1227 | ||
b3b94faa DT |
1228 | /** |
1229 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
1230 | * @ip: the file being written to | |
1231 | * @offset: the offset to write to | |
1232 | * @len: the number of bytes being written | |
b3b94faa | 1233 | * |
461cb419 | 1234 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
1235 | */ |
1236 | ||
cd915493 | 1237 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 1238 | unsigned int len) |
b3b94faa | 1239 | { |
feaa7bba | 1240 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
1241 | struct buffer_head bh; |
1242 | unsigned int shift; | |
1243 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 1244 | u64 end_of_file; |
b3b94faa | 1245 | |
b3b94faa DT |
1246 | if (!len) |
1247 | return 0; | |
1248 | ||
1249 | if (gfs2_is_stuffed(ip)) { | |
1250 | if (offset + len > | |
1251 | sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
461cb419 | 1252 | return 1; |
b3b94faa DT |
1253 | return 0; |
1254 | } | |
1255 | ||
941e6d7d | 1256 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 1257 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 1258 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
1259 | lblock = offset >> shift; |
1260 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
1261 | if (lblock_stop > end_of_file) | |
461cb419 | 1262 | return 1; |
b3b94faa | 1263 | |
941e6d7d SW |
1264 | size = (lblock_stop - lblock) << shift; |
1265 | do { | |
1266 | bh.b_state = 0; | |
1267 | bh.b_size = size; | |
1268 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
1269 | if (!buffer_mapped(&bh)) | |
461cb419 | 1270 | return 1; |
941e6d7d SW |
1271 | size -= bh.b_size; |
1272 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
1273 | } while(size > 0); | |
b3b94faa DT |
1274 | |
1275 | return 0; | |
1276 | } | |
1277 |