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> | |
64dd153c | 13 | #include <linux/blkdev.h> |
5c676f6d | 14 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 15 | #include <linux/crc32.h> |
b3b94faa DT |
16 | |
17 | #include "gfs2.h" | |
5c676f6d | 18 | #include "incore.h" |
b3b94faa DT |
19 | #include "bmap.h" |
20 | #include "glock.h" | |
21 | #include "inode.h" | |
b3b94faa | 22 | #include "meta_io.h" |
b3b94faa DT |
23 | #include "quota.h" |
24 | #include "rgrp.h" | |
4c16c36a | 25 | #include "super.h" |
b3b94faa | 26 | #include "trans.h" |
18ec7d5c | 27 | #include "dir.h" |
5c676f6d | 28 | #include "util.h" |
63997775 | 29 | #include "trace_gfs2.h" |
b3b94faa DT |
30 | |
31 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
32 | * block is 512, so __u16 is fine for that. It saves stack space to | |
33 | * keep it small. | |
34 | */ | |
35 | struct metapath { | |
dbac6710 | 36 | struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT]; |
b3b94faa DT |
37 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; |
38 | }; | |
39 | ||
b3b94faa DT |
40 | struct strip_mine { |
41 | int sm_first; | |
42 | unsigned int sm_height; | |
43 | }; | |
44 | ||
f25ef0c1 SW |
45 | /** |
46 | * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page | |
47 | * @ip: the inode | |
48 | * @dibh: the dinode buffer | |
49 | * @block: the block number that was allocated | |
ff8f33c8 | 50 | * @page: The (optional) page. This is looked up if @page is NULL |
f25ef0c1 SW |
51 | * |
52 | * Returns: errno | |
53 | */ | |
54 | ||
55 | static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, | |
cd915493 | 56 | u64 block, struct page *page) |
f25ef0c1 | 57 | { |
f25ef0c1 SW |
58 | struct inode *inode = &ip->i_inode; |
59 | struct buffer_head *bh; | |
60 | int release = 0; | |
61 | ||
62 | if (!page || page->index) { | |
220cca2a | 63 | page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS); |
f25ef0c1 SW |
64 | if (!page) |
65 | return -ENOMEM; | |
66 | release = 1; | |
67 | } | |
68 | ||
69 | if (!PageUptodate(page)) { | |
70 | void *kaddr = kmap(page); | |
602c89d2 SW |
71 | u64 dsize = i_size_read(inode); |
72 | ||
73 | if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode))) | |
74 | dsize = dibh->b_size - sizeof(struct gfs2_dinode); | |
f25ef0c1 | 75 | |
602c89d2 SW |
76 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
77 | memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize); | |
f25ef0c1 SW |
78 | kunmap(page); |
79 | ||
80 | SetPageUptodate(page); | |
81 | } | |
82 | ||
83 | if (!page_has_buffers(page)) | |
84 | create_empty_buffers(page, 1 << inode->i_blkbits, | |
85 | (1 << BH_Uptodate)); | |
86 | ||
87 | bh = page_buffers(page); | |
88 | ||
89 | if (!buffer_mapped(bh)) | |
90 | map_bh(bh, inode->i_sb, block); | |
91 | ||
92 | set_buffer_uptodate(bh); | |
eaf96527 SW |
93 | if (!gfs2_is_jdata(ip)) |
94 | mark_buffer_dirty(bh); | |
bf36a713 | 95 | if (!gfs2_is_writeback(ip)) |
8475487b | 96 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
f25ef0c1 SW |
97 | |
98 | if (release) { | |
99 | unlock_page(page); | |
100 | page_cache_release(page); | |
101 | } | |
102 | ||
103 | return 0; | |
104 | } | |
105 | ||
b3b94faa DT |
106 | /** |
107 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
108 | * @ip: The GFS2 inode to unstuff | |
ff8f33c8 | 109 | * @page: The (optional) page. This is looked up if the @page is NULL |
b3b94faa DT |
110 | * |
111 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
112 | * that the height can be grown in the traditional way. | |
113 | * | |
114 | * Returns: errno | |
115 | */ | |
116 | ||
f25ef0c1 | 117 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
118 | { |
119 | struct buffer_head *bh, *dibh; | |
48516ced | 120 | struct gfs2_dinode *di; |
cd915493 | 121 | u64 block = 0; |
18ec7d5c | 122 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
123 | int error; |
124 | ||
125 | down_write(&ip->i_rw_mutex); | |
126 | ||
127 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
128 | if (error) | |
129 | goto out; | |
907b9bce | 130 | |
a2e0f799 | 131 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
132 | /* Get a free block, fill it with the stuffed data, |
133 | and write it out to disk */ | |
134 | ||
b45e41d7 | 135 | unsigned int n = 1; |
6e87ed0f | 136 | error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL); |
09010978 SW |
137 | if (error) |
138 | goto out_brelse; | |
18ec7d5c | 139 | if (isdir) { |
5731be53 | 140 | gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 141 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
142 | if (error) |
143 | goto out_brelse; | |
48516ced | 144 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
145 | dibh, sizeof(struct gfs2_dinode)); |
146 | brelse(bh); | |
147 | } else { | |
f25ef0c1 | 148 | error = gfs2_unstuffer_page(ip, dibh, block, page); |
b3b94faa DT |
149 | if (error) |
150 | goto out_brelse; | |
151 | } | |
152 | } | |
153 | ||
154 | /* Set up the pointer to the new block */ | |
155 | ||
d4e9c4c3 | 156 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
48516ced | 157 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
158 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
159 | ||
a2e0f799 | 160 | if (i_size_read(&ip->i_inode)) { |
48516ced | 161 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
162 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
163 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
164 | } |
165 | ||
ecc30c79 | 166 | ip->i_height = 1; |
48516ced | 167 | di->di_height = cpu_to_be16(1); |
b3b94faa | 168 | |
a91ea69f | 169 | out_brelse: |
b3b94faa | 170 | brelse(dibh); |
a91ea69f | 171 | out: |
b3b94faa | 172 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
173 | return error; |
174 | } | |
175 | ||
b3b94faa DT |
176 | |
177 | /** | |
178 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 179 | * @sdp: The superblock |
b3b94faa DT |
180 | * @mp: The metapath to return the result in |
181 | * @block: The disk block to look up | |
9b8c81d1 | 182 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
183 | * |
184 | * This routine returns a struct metapath structure that defines a path | |
185 | * through the metadata of inode "ip" to get to block "block". | |
186 | * | |
187 | * Example: | |
188 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
189 | * filesystem with a blocksize of 4096. | |
190 | * | |
191 | * find_metapath() would return a struct metapath structure set to: | |
192 | * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48, | |
193 | * and mp_list[2] = 165. | |
194 | * | |
195 | * That means that in order to get to the block containing the byte at | |
196 | * offset 101342453, we would load the indirect block pointed to by pointer | |
197 | * 0 in the dinode. We would then load the indirect block pointed to by | |
198 | * pointer 48 in that indirect block. We would then load the data block | |
199 | * pointed to by pointer 165 in that indirect block. | |
200 | * | |
201 | * ---------------------------------------- | |
202 | * | Dinode | | | |
203 | * | | 4| | |
204 | * | |0 1 2 3 4 5 9| | |
205 | * | | 6| | |
206 | * ---------------------------------------- | |
207 | * | | |
208 | * | | |
209 | * V | |
210 | * ---------------------------------------- | |
211 | * | Indirect Block | | |
212 | * | 5| | |
213 | * | 4 4 4 4 4 5 5 1| | |
214 | * |0 5 6 7 8 9 0 1 2| | |
215 | * ---------------------------------------- | |
216 | * | | |
217 | * | | |
218 | * V | |
219 | * ---------------------------------------- | |
220 | * | Indirect Block | | |
221 | * | 1 1 1 1 1 5| | |
222 | * | 6 6 6 6 6 1| | |
223 | * |0 3 4 5 6 7 2| | |
224 | * ---------------------------------------- | |
225 | * | | |
226 | * | | |
227 | * V | |
228 | * ---------------------------------------- | |
229 | * | Data block containing offset | | |
230 | * | 101342453 | | |
231 | * | | | |
232 | * | | | |
233 | * ---------------------------------------- | |
234 | * | |
235 | */ | |
236 | ||
9b8c81d1 SW |
237 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
238 | struct metapath *mp, unsigned int height) | |
b3b94faa | 239 | { |
b3b94faa DT |
240 | unsigned int i; |
241 | ||
9b8c81d1 | 242 | for (i = height; i--;) |
7eabb77e | 243 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
244 | |
245 | } | |
246 | ||
5af4e7a0 | 247 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 248 | { |
5af4e7a0 BM |
249 | if (mp->mp_list[0] == 0) |
250 | return 2; | |
251 | return 1; | |
9b8c81d1 SW |
252 | } |
253 | ||
b3b94faa DT |
254 | /** |
255 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
256 | * @height: The metadata height (0 = dinode) |
257 | * @mp: The metapath | |
258 | * | |
259 | * Return a pointer to the block number of the next height of the metadata | |
260 | * tree given a buffer containing the pointer to the current height of the | |
261 | * metadata tree. | |
262 | */ | |
263 | ||
9b8c81d1 | 264 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 265 | { |
dbac6710 | 266 | struct buffer_head *bh = mp->mp_bh[height]; |
b3b94faa DT |
267 | unsigned int head_size = (height > 0) ? |
268 | sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode); | |
9b8c81d1 | 269 | return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height]; |
b3b94faa DT |
270 | } |
271 | ||
b99b98dc SW |
272 | static void gfs2_metapath_ra(struct gfs2_glock *gl, |
273 | const struct buffer_head *bh, const __be64 *pos) | |
274 | { | |
275 | struct buffer_head *rabh; | |
276 | const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size); | |
277 | const __be64 *t; | |
278 | ||
279 | for (t = pos; t < endp; t++) { | |
280 | if (!*t) | |
281 | continue; | |
282 | ||
283 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
284 | if (trylock_buffer(rabh)) { | |
285 | if (!buffer_uptodate(rabh)) { | |
286 | rabh->b_end_io = end_buffer_read_sync; | |
287 | submit_bh(READA | REQ_META, rabh); | |
288 | continue; | |
289 | } | |
290 | unlock_buffer(rabh); | |
291 | } | |
292 | brelse(rabh); | |
293 | } | |
294 | } | |
295 | ||
b3b94faa | 296 | /** |
9b8c81d1 SW |
297 | * lookup_metapath - Walk the metadata tree to a specific point |
298 | * @ip: The inode | |
b3b94faa | 299 | * @mp: The metapath |
b3b94faa | 300 | * |
9b8c81d1 SW |
301 | * Assumes that the inode's buffer has already been looked up and |
302 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
303 | * by find_metapath(). | |
304 | * | |
305 | * If this function encounters part of the tree which has not been | |
306 | * allocated, it returns the current height of the tree at the point | |
307 | * at which it found the unallocated block. Blocks which are found are | |
308 | * added to the mp->mp_bh[] list. | |
b3b94faa | 309 | * |
9b8c81d1 | 310 | * Returns: error or height of metadata tree |
b3b94faa DT |
311 | */ |
312 | ||
9b8c81d1 | 313 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 314 | { |
11707ea0 SW |
315 | unsigned int end_of_metadata = ip->i_height - 1; |
316 | unsigned int x; | |
9b8c81d1 SW |
317 | __be64 *ptr; |
318 | u64 dblock; | |
e23159d2 | 319 | int ret; |
11707ea0 SW |
320 | |
321 | for (x = 0; x < end_of_metadata; x++) { | |
9b8c81d1 SW |
322 | ptr = metapointer(x, mp); |
323 | dblock = be64_to_cpu(*ptr); | |
324 | if (!dblock) | |
325 | return x + 1; | |
11707ea0 | 326 | |
f2f9c812 | 327 | ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, &mp->mp_bh[x+1]); |
11707ea0 SW |
328 | if (ret) |
329 | return ret; | |
330 | } | |
331 | ||
9b8c81d1 | 332 | return ip->i_height; |
dbac6710 SW |
333 | } |
334 | ||
9b8c81d1 | 335 | static inline void release_metapath(struct metapath *mp) |
dbac6710 SW |
336 | { |
337 | int i; | |
338 | ||
9b8c81d1 SW |
339 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
340 | if (mp->mp_bh[i] == NULL) | |
341 | break; | |
342 | brelse(mp->mp_bh[i]); | |
343 | } | |
11707ea0 SW |
344 | } |
345 | ||
30cbf189 SW |
346 | /** |
347 | * gfs2_extent_length - Returns length of an extent of blocks | |
348 | * @start: Start of the buffer | |
349 | * @len: Length of the buffer in bytes | |
350 | * @ptr: Current position in the buffer | |
351 | * @limit: Max extent length to return (0 = unlimited) | |
352 | * @eob: Set to 1 if we hit "end of block" | |
353 | * | |
354 | * If the first block is zero (unallocated) it will return the number of | |
355 | * unallocated blocks in the extent, otherwise it will return the number | |
356 | * of contiguous blocks in the extent. | |
357 | * | |
358 | * Returns: The length of the extent (minimum of one block) | |
359 | */ | |
360 | ||
361 | static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob) | |
362 | { | |
363 | const __be64 *end = (start + len); | |
364 | const __be64 *first = ptr; | |
365 | u64 d = be64_to_cpu(*ptr); | |
366 | ||
367 | *eob = 0; | |
368 | do { | |
369 | ptr++; | |
370 | if (ptr >= end) | |
371 | break; | |
372 | if (limit && --limit == 0) | |
373 | break; | |
374 | if (d) | |
375 | d++; | |
376 | } while(be64_to_cpu(*ptr) == d); | |
377 | if (ptr >= end) | |
378 | *eob = 1; | |
379 | return (ptr - first); | |
380 | } | |
381 | ||
9b8c81d1 | 382 | static inline void bmap_lock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 383 | { |
4cf1ed81 SW |
384 | if (create) |
385 | down_write(&ip->i_rw_mutex); | |
386 | else | |
387 | down_read(&ip->i_rw_mutex); | |
388 | } | |
389 | ||
9b8c81d1 | 390 | static inline void bmap_unlock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 391 | { |
4cf1ed81 SW |
392 | if (create) |
393 | up_write(&ip->i_rw_mutex); | |
394 | else | |
395 | up_read(&ip->i_rw_mutex); | |
396 | } | |
397 | ||
9b8c81d1 SW |
398 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
399 | struct gfs2_glock *gl, unsigned int i, | |
400 | unsigned offset, u64 bn) | |
401 | { | |
402 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
403 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
404 | sizeof(struct gfs2_dinode))); | |
405 | BUG_ON(i < 1); | |
406 | BUG_ON(mp->mp_bh[i] != NULL); | |
407 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
408 | gfs2_trans_add_bh(gl, mp->mp_bh[i], 1); | |
409 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); | |
410 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
411 | ptr += offset; | |
412 | *ptr = cpu_to_be64(bn); | |
413 | return ptr; | |
414 | } | |
415 | ||
416 | enum alloc_state { | |
417 | ALLOC_DATA = 0, | |
418 | ALLOC_GROW_DEPTH = 1, | |
419 | ALLOC_GROW_HEIGHT = 2, | |
420 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
421 | }; | |
422 | ||
423 | /** | |
424 | * gfs2_bmap_alloc - Build a metadata tree of the requested height | |
425 | * @inode: The GFS2 inode | |
426 | * @lblock: The logical starting block of the extent | |
427 | * @bh_map: This is used to return the mapping details | |
428 | * @mp: The metapath | |
429 | * @sheight: The starting height (i.e. whats already mapped) | |
430 | * @height: The height to build to | |
431 | * @maxlen: The max number of data blocks to alloc | |
432 | * | |
433 | * In this routine we may have to alloc: | |
434 | * i) Indirect blocks to grow the metadata tree height | |
435 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
436 | * iii) Data blocks | |
437 | * | |
438 | * The function is in two parts. The first part works out the total | |
439 | * number of blocks which we need. The second part does the actual | |
440 | * allocation asking for an extent at a time (if enough contiguous free | |
441 | * blocks are available, there will only be one request per bmap call) | |
442 | * and uses the state machine to initialise the blocks in order. | |
443 | * | |
444 | * Returns: errno on error | |
445 | */ | |
446 | ||
447 | static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock, | |
448 | struct buffer_head *bh_map, struct metapath *mp, | |
449 | const unsigned int sheight, | |
450 | const unsigned int height, | |
451 | const unsigned int maxlen) | |
452 | { | |
453 | struct gfs2_inode *ip = GFS2_I(inode); | |
454 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
64dd153c | 455 | struct super_block *sb = sdp->sd_vfs; |
9b8c81d1 SW |
456 | struct buffer_head *dibh = mp->mp_bh[0]; |
457 | u64 bn, dblock = 0; | |
5af4e7a0 | 458 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
9b8c81d1 SW |
459 | unsigned dblks = 0; |
460 | unsigned ptrs_per_blk; | |
461 | const unsigned end_of_metadata = height - 1; | |
64dd153c | 462 | int ret; |
9b8c81d1 SW |
463 | int eob = 0; |
464 | enum alloc_state state; | |
465 | __be64 *ptr; | |
466 | __be64 zero_bn = 0; | |
467 | ||
468 | BUG_ON(sheight < 1); | |
469 | BUG_ON(dibh == NULL); | |
470 | ||
471 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
472 | ||
473 | if (height == sheight) { | |
474 | struct buffer_head *bh; | |
475 | /* Bottom indirect block exists, find unalloced extent size */ | |
476 | ptr = metapointer(end_of_metadata, mp); | |
477 | bh = mp->mp_bh[end_of_metadata]; | |
478 | dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, | |
479 | &eob); | |
480 | BUG_ON(dblks < 1); | |
481 | state = ALLOC_DATA; | |
482 | } else { | |
483 | /* Need to allocate indirect blocks */ | |
484 | ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs; | |
485 | dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]); | |
486 | if (height == ip->i_height) { | |
487 | /* Writing into existing tree, extend tree down */ | |
488 | iblks = height - sheight; | |
489 | state = ALLOC_GROW_DEPTH; | |
490 | } else { | |
491 | /* Building up tree height */ | |
492 | state = ALLOC_GROW_HEIGHT; | |
493 | iblks = height - ip->i_height; | |
5af4e7a0 BM |
494 | branch_start = metapath_branch_start(mp); |
495 | iblks += (height - branch_start); | |
9b8c81d1 SW |
496 | } |
497 | } | |
498 | ||
499 | /* start of the second part of the function (state machine) */ | |
500 | ||
501 | blks = dblks + iblks; | |
502 | i = sheight; | |
503 | do { | |
09010978 | 504 | int error; |
9b8c81d1 | 505 | n = blks - alloced; |
6e87ed0f | 506 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
507 | if (error) |
508 | return error; | |
9b8c81d1 SW |
509 | alloced += n; |
510 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
511 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
512 | switch (state) { | |
513 | /* Growing height of tree */ | |
514 | case ALLOC_GROW_HEIGHT: | |
515 | if (i == 1) { | |
516 | ptr = (__be64 *)(dibh->b_data + | |
517 | sizeof(struct gfs2_dinode)); | |
518 | zero_bn = *ptr; | |
519 | } | |
520 | for (; i - 1 < height - ip->i_height && n > 0; i++, n--) | |
521 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); | |
522 | if (i - 1 == height - ip->i_height) { | |
523 | i--; | |
524 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
525 | sizeof(struct gfs2_meta_header), | |
526 | dibh, sizeof(struct gfs2_dinode)); | |
527 | gfs2_buffer_clear_tail(dibh, | |
528 | sizeof(struct gfs2_dinode) + | |
529 | sizeof(__be64)); | |
530 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
531 | sizeof(struct gfs2_meta_header)); | |
532 | *ptr = zero_bn; | |
533 | state = ALLOC_GROW_DEPTH; | |
5af4e7a0 | 534 | for(i = branch_start; i < height; i++) { |
9b8c81d1 SW |
535 | if (mp->mp_bh[i] == NULL) |
536 | break; | |
537 | brelse(mp->mp_bh[i]); | |
538 | mp->mp_bh[i] = NULL; | |
539 | } | |
5af4e7a0 | 540 | i = branch_start; |
9b8c81d1 SW |
541 | } |
542 | if (n == 0) | |
543 | break; | |
544 | /* Branching from existing tree */ | |
545 | case ALLOC_GROW_DEPTH: | |
546 | if (i > 1 && i < height) | |
547 | gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1); | |
548 | for (; i < height && n > 0; i++, n--) | |
549 | gfs2_indirect_init(mp, ip->i_gl, i, | |
550 | mp->mp_list[i-1], bn++); | |
551 | if (i == height) | |
552 | state = ALLOC_DATA; | |
553 | if (n == 0) | |
554 | break; | |
555 | /* Tree complete, adding data blocks */ | |
556 | case ALLOC_DATA: | |
557 | BUG_ON(n > dblks); | |
558 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); | |
559 | gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1); | |
560 | dblks = n; | |
561 | ptr = metapointer(end_of_metadata, mp); | |
562 | dblock = bn; | |
563 | while (n-- > 0) | |
564 | *ptr++ = cpu_to_be64(bn++); | |
64dd153c BM |
565 | if (buffer_zeronew(bh_map)) { |
566 | ret = sb_issue_zeroout(sb, dblock, dblks, | |
567 | GFP_NOFS); | |
568 | if (ret) { | |
569 | fs_err(sdp, | |
570 | "Failed to zero data buffers\n"); | |
571 | clear_buffer_zeronew(bh_map); | |
572 | } | |
573 | } | |
9b8c81d1 SW |
574 | break; |
575 | } | |
07ccb7bf | 576 | } while ((state != ALLOC_DATA) || !dblock); |
9b8c81d1 SW |
577 | |
578 | ip->i_height = height; | |
579 | gfs2_add_inode_blocks(&ip->i_inode, alloced); | |
580 | gfs2_dinode_out(ip, mp->mp_bh[0]->b_data); | |
581 | map_bh(bh_map, inode->i_sb, dblock); | |
582 | bh_map->b_size = dblks << inode->i_blkbits; | |
583 | set_buffer_new(bh_map); | |
584 | return 0; | |
585 | } | |
586 | ||
b3b94faa | 587 | /** |
4cf1ed81 | 588 | * gfs2_block_map - Map a block from an inode to a disk block |
fd88de56 | 589 | * @inode: The inode |
b3b94faa | 590 | * @lblock: The logical block number |
4cf1ed81 | 591 | * @bh_map: The bh to be mapped |
9b8c81d1 | 592 | * @create: True if its ok to alloc blocks to satify the request |
b3b94faa | 593 | * |
9b8c81d1 SW |
594 | * Sets buffer_mapped() if successful, sets buffer_boundary() if a |
595 | * read of metadata will be required before the next block can be | |
596 | * mapped. Sets buffer_new() if new blocks were allocated. | |
b3b94faa DT |
597 | * |
598 | * Returns: errno | |
599 | */ | |
600 | ||
e9e1ef2b BP |
601 | int gfs2_block_map(struct inode *inode, sector_t lblock, |
602 | struct buffer_head *bh_map, int create) | |
b3b94faa | 603 | { |
feaa7bba SW |
604 | struct gfs2_inode *ip = GFS2_I(inode); |
605 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
ecc30c79 | 606 | unsigned int bsize = sdp->sd_sb.sb_bsize; |
9b8c81d1 | 607 | const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits; |
ecc30c79 | 608 | const u64 *arr = sdp->sd_heightsize; |
9b8c81d1 SW |
609 | __be64 *ptr; |
610 | u64 size; | |
611 | struct metapath mp; | |
612 | int ret; | |
613 | int eob; | |
614 | unsigned int len; | |
615 | struct buffer_head *bh; | |
616 | u8 height; | |
7276b3b0 | 617 | |
9b8c81d1 | 618 | BUG_ON(maxlen == 0); |
b3b94faa | 619 | |
dbac6710 | 620 | memset(mp.mp_bh, 0, sizeof(mp.mp_bh)); |
9b8c81d1 | 621 | bmap_lock(ip, create); |
4cf1ed81 SW |
622 | clear_buffer_mapped(bh_map); |
623 | clear_buffer_new(bh_map); | |
624 | clear_buffer_boundary(bh_map); | |
63997775 | 625 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); |
ecc30c79 SW |
626 | if (gfs2_is_dir(ip)) { |
627 | bsize = sdp->sd_jbsize; | |
628 | arr = sdp->sd_jheightsize; | |
629 | } | |
4cf1ed81 | 630 | |
9b8c81d1 SW |
631 | ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]); |
632 | if (ret) | |
633 | goto out; | |
b3b94faa | 634 | |
9b8c81d1 SW |
635 | height = ip->i_height; |
636 | size = (lblock + 1) * bsize; | |
637 | while (size > arr[height]) | |
638 | height++; | |
639 | find_metapath(sdp, lblock, &mp, height); | |
640 | ret = 1; | |
641 | if (height > ip->i_height || gfs2_is_stuffed(ip)) | |
642 | goto do_alloc; | |
643 | ret = lookup_metapath(ip, &mp); | |
644 | if (ret < 0) | |
645 | goto out; | |
646 | if (ret != ip->i_height) | |
647 | goto do_alloc; | |
648 | ptr = metapointer(ip->i_height - 1, &mp); | |
649 | if (*ptr == 0) | |
650 | goto do_alloc; | |
651 | map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr)); | |
652 | bh = mp.mp_bh[ip->i_height - 1]; | |
653 | len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob); | |
654 | bh_map->b_size = (len << inode->i_blkbits); | |
655 | if (eob) | |
656 | set_buffer_boundary(bh_map); | |
657 | ret = 0; | |
658 | out: | |
659 | release_metapath(&mp); | |
63997775 | 660 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); |
9b8c81d1 SW |
661 | bmap_unlock(ip, create); |
662 | return ret; | |
30cbf189 | 663 | |
9b8c81d1 SW |
664 | do_alloc: |
665 | /* All allocations are done here, firstly check create flag */ | |
666 | if (!create) { | |
667 | BUG_ON(gfs2_is_stuffed(ip)); | |
668 | ret = 0; | |
669 | goto out; | |
b3b94faa | 670 | } |
9b8c81d1 SW |
671 | |
672 | /* At this point ret is the tree depth of already allocated blocks */ | |
673 | ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen); | |
674 | goto out; | |
fd88de56 SW |
675 | } |
676 | ||
941e6d7d SW |
677 | /* |
678 | * Deprecated: do not use in new code | |
679 | */ | |
fd88de56 SW |
680 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
681 | { | |
23591256 | 682 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 683 | int ret; |
fd88de56 SW |
684 | int create = *new; |
685 | ||
686 | BUG_ON(!extlen); | |
687 | BUG_ON(!dblock); | |
688 | BUG_ON(!new); | |
689 | ||
9b8c81d1 | 690 | bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 691 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
692 | *extlen = bh.b_size >> inode->i_blkbits; |
693 | *dblock = bh.b_blocknr; | |
694 | if (buffer_new(&bh)) | |
695 | *new = 1; | |
696 | else | |
697 | *new = 0; | |
698 | return ret; | |
b3b94faa DT |
699 | } |
700 | ||
b3b94faa DT |
701 | /** |
702 | * do_strip - Look for a layer a particular layer of the file and strip it off | |
703 | * @ip: the inode | |
704 | * @dibh: the dinode buffer | |
705 | * @bh: A buffer of pointers | |
706 | * @top: The first pointer in the buffer | |
707 | * @bottom: One more than the last pointer | |
708 | * @height: the height this buffer is at | |
709 | * @data: a pointer to a struct strip_mine | |
710 | * | |
711 | * Returns: errno | |
712 | */ | |
713 | ||
714 | static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh, | |
b44b84d7 | 715 | struct buffer_head *bh, __be64 *top, __be64 *bottom, |
d56fa8a1 | 716 | unsigned int height, struct strip_mine *sm) |
b3b94faa | 717 | { |
feaa7bba | 718 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa | 719 | struct gfs2_rgrp_list rlist; |
cd915493 | 720 | u64 bn, bstart; |
4c16c36a | 721 | u32 blen, btotal; |
b44b84d7 | 722 | __be64 *p; |
b3b94faa DT |
723 | unsigned int rg_blocks = 0; |
724 | int metadata; | |
725 | unsigned int revokes = 0; | |
726 | int x; | |
5e2f7d61 BP |
727 | int error; |
728 | ||
729 | error = gfs2_rindex_update(sdp); | |
730 | if (error) | |
731 | return error; | |
b3b94faa DT |
732 | |
733 | if (!*top) | |
734 | sm->sm_first = 0; | |
735 | ||
736 | if (height != sm->sm_height) | |
737 | return 0; | |
738 | ||
739 | if (sm->sm_first) { | |
740 | top++; | |
741 | sm->sm_first = 0; | |
742 | } | |
743 | ||
ecc30c79 | 744 | metadata = (height != ip->i_height - 1); |
b3b94faa DT |
745 | if (metadata) |
746 | revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
6d3117b4 SW |
747 | else if (ip->i_depth) |
748 | revokes = sdp->sd_inptrs; | |
b3b94faa | 749 | |
b3b94faa DT |
750 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); |
751 | bstart = 0; | |
752 | blen = 0; | |
753 | ||
754 | for (p = top; p < bottom; p++) { | |
755 | if (!*p) | |
756 | continue; | |
757 | ||
758 | bn = be64_to_cpu(*p); | |
759 | ||
760 | if (bstart + blen == bn) | |
761 | blen++; | |
762 | else { | |
763 | if (bstart) | |
70b0c365 | 764 | gfs2_rlist_add(ip, &rlist, bstart); |
b3b94faa DT |
765 | |
766 | bstart = bn; | |
767 | blen = 1; | |
768 | } | |
769 | } | |
770 | ||
771 | if (bstart) | |
70b0c365 | 772 | gfs2_rlist_add(ip, &rlist, bstart); |
b3b94faa DT |
773 | else |
774 | goto out; /* Nothing to do */ | |
775 | ||
fe6c991c | 776 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
777 | |
778 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
779 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 780 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 781 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
782 | } |
783 | ||
784 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
785 | if (error) | |
786 | goto out_rlist; | |
787 | ||
8e2e0047 | 788 | if (gfs2_rs_active(ip->i_res)) /* needs to be done with the rgrp glock held */ |
4a993fb1 | 789 | gfs2_rs_deltree(ip, ip->i_res); |
8e2e0047 | 790 | |
b3b94faa DT |
791 | error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE + |
792 | RES_INDIRECT + RES_STATFS + RES_QUOTA, | |
793 | revokes); | |
794 | if (error) | |
795 | goto out_rg_gunlock; | |
796 | ||
797 | down_write(&ip->i_rw_mutex); | |
798 | ||
d4e9c4c3 SW |
799 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
800 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
b3b94faa DT |
801 | |
802 | bstart = 0; | |
803 | blen = 0; | |
4c16c36a | 804 | btotal = 0; |
b3b94faa DT |
805 | |
806 | for (p = top; p < bottom; p++) { | |
807 | if (!*p) | |
808 | continue; | |
809 | ||
810 | bn = be64_to_cpu(*p); | |
811 | ||
812 | if (bstart + blen == bn) | |
813 | blen++; | |
814 | else { | |
815 | if (bstart) { | |
46fcb2ed | 816 | __gfs2_free_blocks(ip, bstart, blen, metadata); |
4c16c36a | 817 | btotal += blen; |
b3b94faa DT |
818 | } |
819 | ||
820 | bstart = bn; | |
821 | blen = 1; | |
822 | } | |
823 | ||
824 | *p = 0; | |
77658aad | 825 | gfs2_add_inode_blocks(&ip->i_inode, -1); |
b3b94faa DT |
826 | } |
827 | if (bstart) { | |
46fcb2ed | 828 | __gfs2_free_blocks(ip, bstart, blen, metadata); |
4c16c36a | 829 | btotal += blen; |
b3b94faa DT |
830 | } |
831 | ||
4c16c36a BP |
832 | gfs2_statfs_change(sdp, 0, +btotal, 0); |
833 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
834 | ip->i_inode.i_gid); | |
835 | ||
4bd91ba1 | 836 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
b3b94faa | 837 | |
539e5d6b | 838 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
839 | |
840 | up_write(&ip->i_rw_mutex); | |
841 | ||
842 | gfs2_trans_end(sdp); | |
843 | ||
a91ea69f | 844 | out_rg_gunlock: |
b3b94faa | 845 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 846 | out_rlist: |
b3b94faa | 847 | gfs2_rlist_free(&rlist); |
a91ea69f | 848 | out: |
b3b94faa DT |
849 | return error; |
850 | } | |
851 | ||
d56fa8a1 SW |
852 | /** |
853 | * recursive_scan - recursively scan through the end of a file | |
854 | * @ip: the inode | |
855 | * @dibh: the dinode buffer | |
856 | * @mp: the path through the metadata to the point to start | |
857 | * @height: the height the recursion is at | |
858 | * @block: the indirect block to look at | |
859 | * @first: 1 if this is the first block | |
860 | * @sm: data opaque to this function to pass to @bc | |
861 | * | |
862 | * When this is first called @height and @block should be zero and | |
863 | * @first should be 1. | |
864 | * | |
865 | * Returns: errno | |
866 | */ | |
867 | ||
868 | static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh, | |
869 | struct metapath *mp, unsigned int height, | |
870 | u64 block, int first, struct strip_mine *sm) | |
871 | { | |
872 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
873 | struct buffer_head *bh = NULL; | |
b99b98dc | 874 | __be64 *top, *bottom; |
d56fa8a1 SW |
875 | u64 bn; |
876 | int error; | |
877 | int mh_size = sizeof(struct gfs2_meta_header); | |
878 | ||
879 | if (!height) { | |
880 | error = gfs2_meta_inode_buffer(ip, &bh); | |
881 | if (error) | |
882 | return error; | |
883 | dibh = bh; | |
884 | ||
885 | top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0]; | |
886 | bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs; | |
887 | } else { | |
f2f9c812 | 888 | error = gfs2_meta_indirect_buffer(ip, height, block, &bh); |
d56fa8a1 SW |
889 | if (error) |
890 | return error; | |
891 | ||
892 | top = (__be64 *)(bh->b_data + mh_size) + | |
893 | (first ? mp->mp_list[height] : 0); | |
894 | ||
895 | bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs; | |
896 | } | |
897 | ||
898 | error = do_strip(ip, dibh, bh, top, bottom, height, sm); | |
899 | if (error) | |
900 | goto out; | |
901 | ||
bd5437a7 | 902 | if (height < ip->i_height - 1) { |
bd5437a7 | 903 | |
b99b98dc | 904 | gfs2_metapath_ra(ip->i_gl, bh, top); |
bd5437a7 | 905 | |
d56fa8a1 SW |
906 | for (; top < bottom; top++, first = 0) { |
907 | if (!*top) | |
908 | continue; | |
909 | ||
910 | bn = be64_to_cpu(*top); | |
911 | ||
912 | error = recursive_scan(ip, dibh, mp, height + 1, bn, | |
913 | first, sm); | |
914 | if (error) | |
915 | break; | |
916 | } | |
bd5437a7 | 917 | } |
d56fa8a1 SW |
918 | out: |
919 | brelse(bh); | |
920 | return error; | |
921 | } | |
922 | ||
923 | ||
ba7f7290 SW |
924 | /** |
925 | * gfs2_block_truncate_page - Deal with zeroing out data for truncate | |
926 | * | |
927 | * This is partly borrowed from ext3. | |
928 | */ | |
ff8f33c8 | 929 | static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from) |
ba7f7290 SW |
930 | { |
931 | struct inode *inode = mapping->host; | |
932 | struct gfs2_inode *ip = GFS2_I(inode); | |
ba7f7290 SW |
933 | unsigned long index = from >> PAGE_CACHE_SHIFT; |
934 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
935 | unsigned blocksize, iblock, length, pos; | |
936 | struct buffer_head *bh; | |
937 | struct page *page; | |
ba7f7290 SW |
938 | int err; |
939 | ||
220cca2a | 940 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ba7f7290 SW |
941 | if (!page) |
942 | return 0; | |
943 | ||
944 | blocksize = inode->i_sb->s_blocksize; | |
945 | length = blocksize - (offset & (blocksize - 1)); | |
946 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
947 | ||
948 | if (!page_has_buffers(page)) | |
949 | create_empty_buffers(page, blocksize, 0); | |
950 | ||
951 | /* Find the buffer that contains "offset" */ | |
952 | bh = page_buffers(page); | |
953 | pos = blocksize; | |
954 | while (offset >= pos) { | |
955 | bh = bh->b_this_page; | |
956 | iblock++; | |
957 | pos += blocksize; | |
958 | } | |
959 | ||
960 | err = 0; | |
961 | ||
962 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 963 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
964 | /* unmapped? It's a hole - nothing to do */ |
965 | if (!buffer_mapped(bh)) | |
966 | goto unlock; | |
967 | } | |
968 | ||
969 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
970 | if (PageUptodate(page)) | |
971 | set_buffer_uptodate(bh); | |
972 | ||
973 | if (!buffer_uptodate(bh)) { | |
974 | err = -EIO; | |
975 | ll_rw_block(READ, 1, &bh); | |
976 | wait_on_buffer(bh); | |
977 | /* Uhhuh. Read error. Complain and punt. */ | |
978 | if (!buffer_uptodate(bh)) | |
979 | goto unlock; | |
1875f2f3 | 980 | err = 0; |
ba7f7290 SW |
981 | } |
982 | ||
bf36a713 | 983 | if (!gfs2_is_writeback(ip)) |
ba7f7290 SW |
984 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
985 | ||
eebd2aa3 | 986 | zero_user(page, offset, length); |
40bc9a27 | 987 | mark_buffer_dirty(bh); |
ba7f7290 SW |
988 | unlock: |
989 | unlock_page(page); | |
990 | page_cache_release(page); | |
991 | return err; | |
992 | } | |
993 | ||
ff8f33c8 | 994 | static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize) |
b3b94faa | 995 | { |
ff8f33c8 SW |
996 | struct gfs2_inode *ip = GFS2_I(inode); |
997 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
998 | struct address_space *mapping = inode->i_mapping; | |
b3b94faa DT |
999 | struct buffer_head *dibh; |
1000 | int journaled = gfs2_is_jdata(ip); | |
1001 | int error; | |
1002 | ||
1003 | error = gfs2_trans_begin(sdp, | |
c5392124 | 1004 | RES_DINODE + (journaled ? RES_JDATA : 0), 0); |
b3b94faa DT |
1005 | if (error) |
1006 | return error; | |
1007 | ||
1008 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1009 | if (error) | |
1010 | goto out; | |
1011 | ||
ff8f33c8 SW |
1012 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
1013 | ||
b3b94faa | 1014 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1015 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1016 | } else { |
ff8f33c8 SW |
1017 | if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) { |
1018 | error = gfs2_block_truncate_page(mapping, newsize); | |
1019 | if (error) | |
1020 | goto out_brelse; | |
b3b94faa | 1021 | } |
ff8f33c8 | 1022 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1023 | } |
1024 | ||
ff8f33c8 | 1025 | i_size_write(inode, newsize); |
ff8f33c8 SW |
1026 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
1027 | gfs2_dinode_out(ip, dibh->b_data); | |
b3b94faa | 1028 | |
ff8f33c8 SW |
1029 | truncate_pagecache(inode, oldsize, newsize); |
1030 | out_brelse: | |
1031 | brelse(dibh); | |
a91ea69f | 1032 | out: |
b3b94faa | 1033 | gfs2_trans_end(sdp); |
b3b94faa DT |
1034 | return error; |
1035 | } | |
1036 | ||
cd915493 | 1037 | static int trunc_dealloc(struct gfs2_inode *ip, u64 size) |
b3b94faa | 1038 | { |
9b8c81d1 | 1039 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
ecc30c79 | 1040 | unsigned int height = ip->i_height; |
cd915493 | 1041 | u64 lblock; |
b3b94faa DT |
1042 | struct metapath mp; |
1043 | int error; | |
1044 | ||
1045 | if (!size) | |
1046 | lblock = 0; | |
18ec7d5c | 1047 | else |
9b8c81d1 | 1048 | lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 1049 | |
9b8c81d1 | 1050 | find_metapath(sdp, lblock, &mp, ip->i_height); |
5407e242 BP |
1051 | error = gfs2_rindex_update(sdp); |
1052 | if (error) | |
1053 | return error; | |
b3b94faa DT |
1054 | |
1055 | error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
1056 | if (error) | |
5407e242 | 1057 | return error; |
b3b94faa DT |
1058 | |
1059 | while (height--) { | |
1060 | struct strip_mine sm; | |
1061 | sm.sm_first = !!size; | |
1062 | sm.sm_height = height; | |
1063 | ||
d56fa8a1 | 1064 | error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm); |
b3b94faa DT |
1065 | if (error) |
1066 | break; | |
1067 | } | |
1068 | ||
1069 | gfs2_quota_unhold(ip); | |
1070 | ||
b3b94faa DT |
1071 | return error; |
1072 | } | |
1073 | ||
1074 | static int trunc_end(struct gfs2_inode *ip) | |
1075 | { | |
feaa7bba | 1076 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1077 | struct buffer_head *dibh; |
1078 | int error; | |
1079 | ||
1080 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1081 | if (error) | |
1082 | return error; | |
1083 | ||
1084 | down_write(&ip->i_rw_mutex); | |
1085 | ||
1086 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1087 | if (error) | |
1088 | goto out; | |
1089 | ||
a2e0f799 | 1090 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1091 | ip->i_height = 0; |
ce276b06 | 1092 | ip->i_goal = ip->i_no_addr; |
b3b94faa DT |
1093 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
1094 | } | |
4bd91ba1 | 1095 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
383f01fb | 1096 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1097 | |
d4e9c4c3 | 1098 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
539e5d6b | 1099 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1100 | brelse(dibh); |
1101 | ||
a91ea69f | 1102 | out: |
b3b94faa | 1103 | up_write(&ip->i_rw_mutex); |
b3b94faa | 1104 | gfs2_trans_end(sdp); |
b3b94faa DT |
1105 | return error; |
1106 | } | |
1107 | ||
1108 | /** | |
1109 | * do_shrink - make a file smaller | |
ff8f33c8 SW |
1110 | * @inode: the inode |
1111 | * @oldsize: the current inode size | |
1112 | * @newsize: the size to make the file | |
b3b94faa | 1113 | * |
ff8f33c8 SW |
1114 | * Called with an exclusive lock on @inode. The @size must |
1115 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
1116 | * |
1117 | * Returns: errno | |
1118 | */ | |
1119 | ||
ff8f33c8 | 1120 | static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize) |
b3b94faa | 1121 | { |
ff8f33c8 | 1122 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
1123 | int error; |
1124 | ||
ff8f33c8 | 1125 | error = trunc_start(inode, oldsize, newsize); |
b3b94faa DT |
1126 | if (error < 0) |
1127 | return error; | |
ff8f33c8 | 1128 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
1129 | return 0; |
1130 | ||
ff8f33c8 SW |
1131 | error = trunc_dealloc(ip, newsize); |
1132 | if (error == 0) | |
b3b94faa DT |
1133 | error = trunc_end(ip); |
1134 | ||
1135 | return error; | |
1136 | } | |
1137 | ||
ff8f33c8 | 1138 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 1139 | { |
ff8f33c8 SW |
1140 | u64 size = inode->i_size; |
1141 | int ret; | |
1142 | ||
1143 | ret = do_shrink(inode, size, size); | |
1144 | WARN_ON(ret != 0); | |
1145 | } | |
1146 | ||
1147 | /** | |
1148 | * do_grow - Touch and update inode size | |
1149 | * @inode: The inode | |
1150 | * @size: The new size | |
1151 | * | |
1152 | * This function updates the timestamps on the inode and | |
1153 | * may also increase the size of the inode. This function | |
1154 | * must not be called with @size any smaller than the current | |
1155 | * inode size. | |
1156 | * | |
1157 | * Although it is not strictly required to unstuff files here, | |
1158 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
1159 | * code which will result in a buffer overrun if the size is larger | |
1160 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 1161 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
1162 | * just update the inode size directly. |
1163 | * | |
1164 | * Returns: 0 on success, or -ve on error | |
1165 | */ | |
1166 | ||
1167 | static int do_grow(struct inode *inode, u64 size) | |
1168 | { | |
1169 | struct gfs2_inode *ip = GFS2_I(inode); | |
1170 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
a13b8c5f WC |
1171 | struct buffer_head *dibh; |
1172 | int error; | |
2f7ee358 | 1173 | int unstuff = 0; |
a13b8c5f | 1174 | |
ff8f33c8 SW |
1175 | if (gfs2_is_stuffed(ip) && |
1176 | (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) { | |
ff8f33c8 SW |
1177 | error = gfs2_quota_lock_check(ip); |
1178 | if (error) | |
5407e242 | 1179 | return error; |
ff8f33c8 | 1180 | |
564e12b1 | 1181 | error = gfs2_inplace_reserve(ip, 1); |
ff8f33c8 SW |
1182 | if (error) |
1183 | goto do_grow_qunlock; | |
2f7ee358 | 1184 | unstuff = 1; |
ff8f33c8 SW |
1185 | } |
1186 | ||
bf97b673 | 1187 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0); |
a13b8c5f | 1188 | if (error) |
ff8f33c8 | 1189 | goto do_grow_release; |
a13b8c5f | 1190 | |
2f7ee358 | 1191 | if (unstuff) { |
ff8f33c8 SW |
1192 | error = gfs2_unstuff_dinode(ip, NULL); |
1193 | if (error) | |
1194 | goto do_end_trans; | |
1195 | } | |
a13b8c5f WC |
1196 | |
1197 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1198 | if (error) | |
ff8f33c8 | 1199 | goto do_end_trans; |
a13b8c5f | 1200 | |
ff8f33c8 | 1201 | i_size_write(inode, size); |
a13b8c5f WC |
1202 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
1203 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
1204 | gfs2_dinode_out(ip, dibh->b_data); | |
1205 | brelse(dibh); | |
1206 | ||
ff8f33c8 | 1207 | do_end_trans: |
a13b8c5f | 1208 | gfs2_trans_end(sdp); |
ff8f33c8 | 1209 | do_grow_release: |
2f7ee358 | 1210 | if (unstuff) { |
ff8f33c8 SW |
1211 | gfs2_inplace_release(ip); |
1212 | do_grow_qunlock: | |
1213 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 1214 | } |
a13b8c5f WC |
1215 | return error; |
1216 | } | |
1217 | ||
b3b94faa | 1218 | /** |
ff8f33c8 SW |
1219 | * gfs2_setattr_size - make a file a given size |
1220 | * @inode: the inode | |
1221 | * @newsize: the size to make the file | |
b3b94faa | 1222 | * |
ff8f33c8 SW |
1223 | * The file size can grow, shrink, or stay the same size. This |
1224 | * is called holding i_mutex and an exclusive glock on the inode | |
1225 | * in question. | |
b3b94faa DT |
1226 | * |
1227 | * Returns: errno | |
1228 | */ | |
1229 | ||
ff8f33c8 | 1230 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 1231 | { |
ff8f33c8 SW |
1232 | int ret; |
1233 | u64 oldsize; | |
b3b94faa | 1234 | |
ff8f33c8 | 1235 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 1236 | |
ff8f33c8 SW |
1237 | ret = inode_newsize_ok(inode, newsize); |
1238 | if (ret) | |
1239 | return ret; | |
b3b94faa | 1240 | |
562c72aa CH |
1241 | inode_dio_wait(inode); |
1242 | ||
ff8f33c8 SW |
1243 | oldsize = inode->i_size; |
1244 | if (newsize >= oldsize) | |
1245 | return do_grow(inode, newsize); | |
1246 | ||
1247 | return do_shrink(inode, oldsize, newsize); | |
b3b94faa DT |
1248 | } |
1249 | ||
1250 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1251 | { | |
1252 | int error; | |
a2e0f799 | 1253 | error = trunc_dealloc(ip, i_size_read(&ip->i_inode)); |
b3b94faa DT |
1254 | if (!error) |
1255 | error = trunc_end(ip); | |
1256 | return error; | |
1257 | } | |
1258 | ||
1259 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1260 | { | |
1261 | return trunc_dealloc(ip, 0); | |
1262 | } | |
1263 | ||
b3b94faa DT |
1264 | /** |
1265 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
1266 | * @ip: the file being written to | |
1267 | * @offset: the offset to write to | |
1268 | * @len: the number of bytes being written | |
b3b94faa | 1269 | * |
461cb419 | 1270 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
1271 | */ |
1272 | ||
cd915493 | 1273 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 1274 | unsigned int len) |
b3b94faa | 1275 | { |
feaa7bba | 1276 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
1277 | struct buffer_head bh; |
1278 | unsigned int shift; | |
1279 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 1280 | u64 end_of_file; |
b3b94faa | 1281 | |
b3b94faa DT |
1282 | if (!len) |
1283 | return 0; | |
1284 | ||
1285 | if (gfs2_is_stuffed(ip)) { | |
1286 | if (offset + len > | |
1287 | sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
461cb419 | 1288 | return 1; |
b3b94faa DT |
1289 | return 0; |
1290 | } | |
1291 | ||
941e6d7d | 1292 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 1293 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 1294 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
1295 | lblock = offset >> shift; |
1296 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
1297 | if (lblock_stop > end_of_file) | |
461cb419 | 1298 | return 1; |
b3b94faa | 1299 | |
941e6d7d SW |
1300 | size = (lblock_stop - lblock) << shift; |
1301 | do { | |
1302 | bh.b_state = 0; | |
1303 | bh.b_size = size; | |
1304 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
1305 | if (!buffer_mapped(&bh)) | |
461cb419 | 1306 | return 1; |
941e6d7d SW |
1307 | size -= bh.b_size; |
1308 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
1309 | } while(size > 0); | |
b3b94faa DT |
1310 | |
1311 | return 0; | |
1312 | } | |
1313 |