]>
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) { | |
63 | page = grab_cache_page(inode->i_mapping, 0); | |
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; |
09010978 SW |
136 | error = gfs2_alloc_block(ip, &block, &n); |
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 | ||
272 | /** | |
9b8c81d1 SW |
273 | * lookup_metapath - Walk the metadata tree to a specific point |
274 | * @ip: The inode | |
b3b94faa | 275 | * @mp: The metapath |
b3b94faa | 276 | * |
9b8c81d1 SW |
277 | * Assumes that the inode's buffer has already been looked up and |
278 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
279 | * by find_metapath(). | |
280 | * | |
281 | * If this function encounters part of the tree which has not been | |
282 | * allocated, it returns the current height of the tree at the point | |
283 | * at which it found the unallocated block. Blocks which are found are | |
284 | * added to the mp->mp_bh[] list. | |
b3b94faa | 285 | * |
9b8c81d1 | 286 | * Returns: error or height of metadata tree |
b3b94faa DT |
287 | */ |
288 | ||
9b8c81d1 | 289 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 290 | { |
11707ea0 SW |
291 | unsigned int end_of_metadata = ip->i_height - 1; |
292 | unsigned int x; | |
9b8c81d1 SW |
293 | __be64 *ptr; |
294 | u64 dblock; | |
e23159d2 | 295 | int ret; |
11707ea0 SW |
296 | |
297 | for (x = 0; x < end_of_metadata; x++) { | |
9b8c81d1 SW |
298 | ptr = metapointer(x, mp); |
299 | dblock = be64_to_cpu(*ptr); | |
300 | if (!dblock) | |
301 | return x + 1; | |
11707ea0 | 302 | |
9b8c81d1 | 303 | ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, 0, &mp->mp_bh[x+1]); |
11707ea0 SW |
304 | if (ret) |
305 | return ret; | |
306 | } | |
307 | ||
9b8c81d1 | 308 | return ip->i_height; |
dbac6710 SW |
309 | } |
310 | ||
9b8c81d1 | 311 | static inline void release_metapath(struct metapath *mp) |
dbac6710 SW |
312 | { |
313 | int i; | |
314 | ||
9b8c81d1 SW |
315 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
316 | if (mp->mp_bh[i] == NULL) | |
317 | break; | |
318 | brelse(mp->mp_bh[i]); | |
319 | } | |
11707ea0 SW |
320 | } |
321 | ||
30cbf189 SW |
322 | /** |
323 | * gfs2_extent_length - Returns length of an extent of blocks | |
324 | * @start: Start of the buffer | |
325 | * @len: Length of the buffer in bytes | |
326 | * @ptr: Current position in the buffer | |
327 | * @limit: Max extent length to return (0 = unlimited) | |
328 | * @eob: Set to 1 if we hit "end of block" | |
329 | * | |
330 | * If the first block is zero (unallocated) it will return the number of | |
331 | * unallocated blocks in the extent, otherwise it will return the number | |
332 | * of contiguous blocks in the extent. | |
333 | * | |
334 | * Returns: The length of the extent (minimum of one block) | |
335 | */ | |
336 | ||
337 | static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, unsigned limit, int *eob) | |
338 | { | |
339 | const __be64 *end = (start + len); | |
340 | const __be64 *first = ptr; | |
341 | u64 d = be64_to_cpu(*ptr); | |
342 | ||
343 | *eob = 0; | |
344 | do { | |
345 | ptr++; | |
346 | if (ptr >= end) | |
347 | break; | |
348 | if (limit && --limit == 0) | |
349 | break; | |
350 | if (d) | |
351 | d++; | |
352 | } while(be64_to_cpu(*ptr) == d); | |
353 | if (ptr >= end) | |
354 | *eob = 1; | |
355 | return (ptr - first); | |
356 | } | |
357 | ||
9b8c81d1 | 358 | static inline void bmap_lock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 359 | { |
4cf1ed81 SW |
360 | if (create) |
361 | down_write(&ip->i_rw_mutex); | |
362 | else | |
363 | down_read(&ip->i_rw_mutex); | |
364 | } | |
365 | ||
9b8c81d1 | 366 | static inline void bmap_unlock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 367 | { |
4cf1ed81 SW |
368 | if (create) |
369 | up_write(&ip->i_rw_mutex); | |
370 | else | |
371 | up_read(&ip->i_rw_mutex); | |
372 | } | |
373 | ||
9b8c81d1 SW |
374 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
375 | struct gfs2_glock *gl, unsigned int i, | |
376 | unsigned offset, u64 bn) | |
377 | { | |
378 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
379 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
380 | sizeof(struct gfs2_dinode))); | |
381 | BUG_ON(i < 1); | |
382 | BUG_ON(mp->mp_bh[i] != NULL); | |
383 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
384 | gfs2_trans_add_bh(gl, mp->mp_bh[i], 1); | |
385 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); | |
386 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
387 | ptr += offset; | |
388 | *ptr = cpu_to_be64(bn); | |
389 | return ptr; | |
390 | } | |
391 | ||
392 | enum alloc_state { | |
393 | ALLOC_DATA = 0, | |
394 | ALLOC_GROW_DEPTH = 1, | |
395 | ALLOC_GROW_HEIGHT = 2, | |
396 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
397 | }; | |
398 | ||
399 | /** | |
400 | * gfs2_bmap_alloc - Build a metadata tree of the requested height | |
401 | * @inode: The GFS2 inode | |
402 | * @lblock: The logical starting block of the extent | |
403 | * @bh_map: This is used to return the mapping details | |
404 | * @mp: The metapath | |
405 | * @sheight: The starting height (i.e. whats already mapped) | |
406 | * @height: The height to build to | |
407 | * @maxlen: The max number of data blocks to alloc | |
408 | * | |
409 | * In this routine we may have to alloc: | |
410 | * i) Indirect blocks to grow the metadata tree height | |
411 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
412 | * iii) Data blocks | |
413 | * | |
414 | * The function is in two parts. The first part works out the total | |
415 | * number of blocks which we need. The second part does the actual | |
416 | * allocation asking for an extent at a time (if enough contiguous free | |
417 | * blocks are available, there will only be one request per bmap call) | |
418 | * and uses the state machine to initialise the blocks in order. | |
419 | * | |
420 | * Returns: errno on error | |
421 | */ | |
422 | ||
423 | static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock, | |
424 | struct buffer_head *bh_map, struct metapath *mp, | |
425 | const unsigned int sheight, | |
426 | const unsigned int height, | |
427 | const unsigned int maxlen) | |
428 | { | |
429 | struct gfs2_inode *ip = GFS2_I(inode); | |
430 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
64dd153c | 431 | struct super_block *sb = sdp->sd_vfs; |
9b8c81d1 SW |
432 | struct buffer_head *dibh = mp->mp_bh[0]; |
433 | u64 bn, dblock = 0; | |
5af4e7a0 | 434 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
9b8c81d1 SW |
435 | unsigned dblks = 0; |
436 | unsigned ptrs_per_blk; | |
437 | const unsigned end_of_metadata = height - 1; | |
64dd153c | 438 | int ret; |
9b8c81d1 SW |
439 | int eob = 0; |
440 | enum alloc_state state; | |
441 | __be64 *ptr; | |
442 | __be64 zero_bn = 0; | |
443 | ||
444 | BUG_ON(sheight < 1); | |
445 | BUG_ON(dibh == NULL); | |
446 | ||
447 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
448 | ||
449 | if (height == sheight) { | |
450 | struct buffer_head *bh; | |
451 | /* Bottom indirect block exists, find unalloced extent size */ | |
452 | ptr = metapointer(end_of_metadata, mp); | |
453 | bh = mp->mp_bh[end_of_metadata]; | |
454 | dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, | |
455 | &eob); | |
456 | BUG_ON(dblks < 1); | |
457 | state = ALLOC_DATA; | |
458 | } else { | |
459 | /* Need to allocate indirect blocks */ | |
460 | ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs; | |
461 | dblks = min(maxlen, ptrs_per_blk - mp->mp_list[end_of_metadata]); | |
462 | if (height == ip->i_height) { | |
463 | /* Writing into existing tree, extend tree down */ | |
464 | iblks = height - sheight; | |
465 | state = ALLOC_GROW_DEPTH; | |
466 | } else { | |
467 | /* Building up tree height */ | |
468 | state = ALLOC_GROW_HEIGHT; | |
469 | iblks = height - ip->i_height; | |
5af4e7a0 BM |
470 | branch_start = metapath_branch_start(mp); |
471 | iblks += (height - branch_start); | |
9b8c81d1 SW |
472 | } |
473 | } | |
474 | ||
475 | /* start of the second part of the function (state machine) */ | |
476 | ||
477 | blks = dblks + iblks; | |
478 | i = sheight; | |
479 | do { | |
09010978 | 480 | int error; |
9b8c81d1 | 481 | n = blks - alloced; |
09010978 SW |
482 | error = gfs2_alloc_block(ip, &bn, &n); |
483 | if (error) | |
484 | return error; | |
9b8c81d1 SW |
485 | alloced += n; |
486 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
487 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
488 | switch (state) { | |
489 | /* Growing height of tree */ | |
490 | case ALLOC_GROW_HEIGHT: | |
491 | if (i == 1) { | |
492 | ptr = (__be64 *)(dibh->b_data + | |
493 | sizeof(struct gfs2_dinode)); | |
494 | zero_bn = *ptr; | |
495 | } | |
496 | for (; i - 1 < height - ip->i_height && n > 0; i++, n--) | |
497 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); | |
498 | if (i - 1 == height - ip->i_height) { | |
499 | i--; | |
500 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
501 | sizeof(struct gfs2_meta_header), | |
502 | dibh, sizeof(struct gfs2_dinode)); | |
503 | gfs2_buffer_clear_tail(dibh, | |
504 | sizeof(struct gfs2_dinode) + | |
505 | sizeof(__be64)); | |
506 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
507 | sizeof(struct gfs2_meta_header)); | |
508 | *ptr = zero_bn; | |
509 | state = ALLOC_GROW_DEPTH; | |
5af4e7a0 | 510 | for(i = branch_start; i < height; i++) { |
9b8c81d1 SW |
511 | if (mp->mp_bh[i] == NULL) |
512 | break; | |
513 | brelse(mp->mp_bh[i]); | |
514 | mp->mp_bh[i] = NULL; | |
515 | } | |
5af4e7a0 | 516 | i = branch_start; |
9b8c81d1 SW |
517 | } |
518 | if (n == 0) | |
519 | break; | |
520 | /* Branching from existing tree */ | |
521 | case ALLOC_GROW_DEPTH: | |
522 | if (i > 1 && i < height) | |
523 | gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[i-1], 1); | |
524 | for (; i < height && n > 0; i++, n--) | |
525 | gfs2_indirect_init(mp, ip->i_gl, i, | |
526 | mp->mp_list[i-1], bn++); | |
527 | if (i == height) | |
528 | state = ALLOC_DATA; | |
529 | if (n == 0) | |
530 | break; | |
531 | /* Tree complete, adding data blocks */ | |
532 | case ALLOC_DATA: | |
533 | BUG_ON(n > dblks); | |
534 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); | |
535 | gfs2_trans_add_bh(ip->i_gl, mp->mp_bh[end_of_metadata], 1); | |
536 | dblks = n; | |
537 | ptr = metapointer(end_of_metadata, mp); | |
538 | dblock = bn; | |
539 | while (n-- > 0) | |
540 | *ptr++ = cpu_to_be64(bn++); | |
64dd153c BM |
541 | if (buffer_zeronew(bh_map)) { |
542 | ret = sb_issue_zeroout(sb, dblock, dblks, | |
543 | GFP_NOFS); | |
544 | if (ret) { | |
545 | fs_err(sdp, | |
546 | "Failed to zero data buffers\n"); | |
547 | clear_buffer_zeronew(bh_map); | |
548 | } | |
549 | } | |
9b8c81d1 SW |
550 | break; |
551 | } | |
07ccb7bf | 552 | } while ((state != ALLOC_DATA) || !dblock); |
9b8c81d1 SW |
553 | |
554 | ip->i_height = height; | |
555 | gfs2_add_inode_blocks(&ip->i_inode, alloced); | |
556 | gfs2_dinode_out(ip, mp->mp_bh[0]->b_data); | |
557 | map_bh(bh_map, inode->i_sb, dblock); | |
558 | bh_map->b_size = dblks << inode->i_blkbits; | |
559 | set_buffer_new(bh_map); | |
560 | return 0; | |
561 | } | |
562 | ||
b3b94faa | 563 | /** |
4cf1ed81 | 564 | * gfs2_block_map - Map a block from an inode to a disk block |
fd88de56 | 565 | * @inode: The inode |
b3b94faa | 566 | * @lblock: The logical block number |
4cf1ed81 | 567 | * @bh_map: The bh to be mapped |
9b8c81d1 | 568 | * @create: True if its ok to alloc blocks to satify the request |
b3b94faa | 569 | * |
9b8c81d1 SW |
570 | * Sets buffer_mapped() if successful, sets buffer_boundary() if a |
571 | * read of metadata will be required before the next block can be | |
572 | * mapped. Sets buffer_new() if new blocks were allocated. | |
b3b94faa DT |
573 | * |
574 | * Returns: errno | |
575 | */ | |
576 | ||
e9e1ef2b BP |
577 | int gfs2_block_map(struct inode *inode, sector_t lblock, |
578 | struct buffer_head *bh_map, int create) | |
b3b94faa | 579 | { |
feaa7bba SW |
580 | struct gfs2_inode *ip = GFS2_I(inode); |
581 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
ecc30c79 | 582 | unsigned int bsize = sdp->sd_sb.sb_bsize; |
9b8c81d1 | 583 | const unsigned int maxlen = bh_map->b_size >> inode->i_blkbits; |
ecc30c79 | 584 | const u64 *arr = sdp->sd_heightsize; |
9b8c81d1 SW |
585 | __be64 *ptr; |
586 | u64 size; | |
587 | struct metapath mp; | |
588 | int ret; | |
589 | int eob; | |
590 | unsigned int len; | |
591 | struct buffer_head *bh; | |
592 | u8 height; | |
7276b3b0 | 593 | |
9b8c81d1 | 594 | BUG_ON(maxlen == 0); |
b3b94faa | 595 | |
dbac6710 | 596 | memset(mp.mp_bh, 0, sizeof(mp.mp_bh)); |
9b8c81d1 | 597 | bmap_lock(ip, create); |
4cf1ed81 SW |
598 | clear_buffer_mapped(bh_map); |
599 | clear_buffer_new(bh_map); | |
600 | clear_buffer_boundary(bh_map); | |
63997775 | 601 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); |
ecc30c79 SW |
602 | if (gfs2_is_dir(ip)) { |
603 | bsize = sdp->sd_jbsize; | |
604 | arr = sdp->sd_jheightsize; | |
605 | } | |
4cf1ed81 | 606 | |
9b8c81d1 SW |
607 | ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]); |
608 | if (ret) | |
609 | goto out; | |
b3b94faa | 610 | |
9b8c81d1 SW |
611 | height = ip->i_height; |
612 | size = (lblock + 1) * bsize; | |
613 | while (size > arr[height]) | |
614 | height++; | |
615 | find_metapath(sdp, lblock, &mp, height); | |
616 | ret = 1; | |
617 | if (height > ip->i_height || gfs2_is_stuffed(ip)) | |
618 | goto do_alloc; | |
619 | ret = lookup_metapath(ip, &mp); | |
620 | if (ret < 0) | |
621 | goto out; | |
622 | if (ret != ip->i_height) | |
623 | goto do_alloc; | |
624 | ptr = metapointer(ip->i_height - 1, &mp); | |
625 | if (*ptr == 0) | |
626 | goto do_alloc; | |
627 | map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr)); | |
628 | bh = mp.mp_bh[ip->i_height - 1]; | |
629 | len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob); | |
630 | bh_map->b_size = (len << inode->i_blkbits); | |
631 | if (eob) | |
632 | set_buffer_boundary(bh_map); | |
633 | ret = 0; | |
634 | out: | |
635 | release_metapath(&mp); | |
63997775 | 636 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); |
9b8c81d1 SW |
637 | bmap_unlock(ip, create); |
638 | return ret; | |
30cbf189 | 639 | |
9b8c81d1 SW |
640 | do_alloc: |
641 | /* All allocations are done here, firstly check create flag */ | |
642 | if (!create) { | |
643 | BUG_ON(gfs2_is_stuffed(ip)); | |
644 | ret = 0; | |
645 | goto out; | |
b3b94faa | 646 | } |
9b8c81d1 SW |
647 | |
648 | /* At this point ret is the tree depth of already allocated blocks */ | |
649 | ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen); | |
650 | goto out; | |
fd88de56 SW |
651 | } |
652 | ||
941e6d7d SW |
653 | /* |
654 | * Deprecated: do not use in new code | |
655 | */ | |
fd88de56 SW |
656 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
657 | { | |
23591256 | 658 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 659 | int ret; |
fd88de56 SW |
660 | int create = *new; |
661 | ||
662 | BUG_ON(!extlen); | |
663 | BUG_ON(!dblock); | |
664 | BUG_ON(!new); | |
665 | ||
9b8c81d1 | 666 | bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 667 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
668 | *extlen = bh.b_size >> inode->i_blkbits; |
669 | *dblock = bh.b_blocknr; | |
670 | if (buffer_new(&bh)) | |
671 | *new = 1; | |
672 | else | |
673 | *new = 0; | |
674 | return ret; | |
b3b94faa DT |
675 | } |
676 | ||
b3b94faa DT |
677 | /** |
678 | * do_strip - Look for a layer a particular layer of the file and strip it off | |
679 | * @ip: the inode | |
680 | * @dibh: the dinode buffer | |
681 | * @bh: A buffer of pointers | |
682 | * @top: The first pointer in the buffer | |
683 | * @bottom: One more than the last pointer | |
684 | * @height: the height this buffer is at | |
685 | * @data: a pointer to a struct strip_mine | |
686 | * | |
687 | * Returns: errno | |
688 | */ | |
689 | ||
690 | static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh, | |
b44b84d7 | 691 | struct buffer_head *bh, __be64 *top, __be64 *bottom, |
d56fa8a1 | 692 | unsigned int height, struct strip_mine *sm) |
b3b94faa | 693 | { |
feaa7bba | 694 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa | 695 | struct gfs2_rgrp_list rlist; |
cd915493 | 696 | u64 bn, bstart; |
4c16c36a | 697 | u32 blen, btotal; |
b44b84d7 | 698 | __be64 *p; |
b3b94faa DT |
699 | unsigned int rg_blocks = 0; |
700 | int metadata; | |
701 | unsigned int revokes = 0; | |
702 | int x; | |
e06dfc49 | 703 | int error = 0; |
b3b94faa DT |
704 | |
705 | if (!*top) | |
706 | sm->sm_first = 0; | |
707 | ||
708 | if (height != sm->sm_height) | |
709 | return 0; | |
710 | ||
711 | if (sm->sm_first) { | |
712 | top++; | |
713 | sm->sm_first = 0; | |
714 | } | |
715 | ||
ecc30c79 | 716 | metadata = (height != ip->i_height - 1); |
b3b94faa DT |
717 | if (metadata) |
718 | revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
6d3117b4 SW |
719 | else if (ip->i_depth) |
720 | revokes = sdp->sd_inptrs; | |
b3b94faa | 721 | |
b3b94faa DT |
722 | if (error) |
723 | return error; | |
724 | ||
725 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); | |
726 | bstart = 0; | |
727 | blen = 0; | |
728 | ||
729 | for (p = top; p < bottom; p++) { | |
730 | if (!*p) | |
731 | continue; | |
732 | ||
733 | bn = be64_to_cpu(*p); | |
734 | ||
735 | if (bstart + blen == bn) | |
736 | blen++; | |
737 | else { | |
738 | if (bstart) | |
70b0c365 | 739 | gfs2_rlist_add(ip, &rlist, bstart); |
b3b94faa DT |
740 | |
741 | bstart = bn; | |
742 | blen = 1; | |
743 | } | |
744 | } | |
745 | ||
746 | if (bstart) | |
70b0c365 | 747 | gfs2_rlist_add(ip, &rlist, bstart); |
b3b94faa DT |
748 | else |
749 | goto out; /* Nothing to do */ | |
750 | ||
fe6c991c | 751 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
752 | |
753 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
754 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 755 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
bb8d8a6f | 756 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
757 | } |
758 | ||
759 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
760 | if (error) | |
761 | goto out_rlist; | |
762 | ||
763 | error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE + | |
764 | RES_INDIRECT + RES_STATFS + RES_QUOTA, | |
765 | revokes); | |
766 | if (error) | |
767 | goto out_rg_gunlock; | |
768 | ||
769 | down_write(&ip->i_rw_mutex); | |
770 | ||
d4e9c4c3 SW |
771 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
772 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
b3b94faa DT |
773 | |
774 | bstart = 0; | |
775 | blen = 0; | |
4c16c36a | 776 | btotal = 0; |
b3b94faa DT |
777 | |
778 | for (p = top; p < bottom; p++) { | |
779 | if (!*p) | |
780 | continue; | |
781 | ||
782 | bn = be64_to_cpu(*p); | |
783 | ||
784 | if (bstart + blen == bn) | |
785 | blen++; | |
786 | else { | |
787 | if (bstart) { | |
46fcb2ed | 788 | __gfs2_free_blocks(ip, bstart, blen, metadata); |
4c16c36a | 789 | btotal += blen; |
b3b94faa DT |
790 | } |
791 | ||
792 | bstart = bn; | |
793 | blen = 1; | |
794 | } | |
795 | ||
796 | *p = 0; | |
77658aad | 797 | gfs2_add_inode_blocks(&ip->i_inode, -1); |
b3b94faa DT |
798 | } |
799 | if (bstart) { | |
46fcb2ed | 800 | __gfs2_free_blocks(ip, bstart, blen, metadata); |
4c16c36a | 801 | btotal += blen; |
b3b94faa DT |
802 | } |
803 | ||
4c16c36a BP |
804 | gfs2_statfs_change(sdp, 0, +btotal, 0); |
805 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
806 | ip->i_inode.i_gid); | |
807 | ||
4bd91ba1 | 808 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
b3b94faa | 809 | |
539e5d6b | 810 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
811 | |
812 | up_write(&ip->i_rw_mutex); | |
813 | ||
814 | gfs2_trans_end(sdp); | |
815 | ||
a91ea69f | 816 | out_rg_gunlock: |
b3b94faa | 817 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 818 | out_rlist: |
b3b94faa | 819 | gfs2_rlist_free(&rlist); |
a91ea69f | 820 | out: |
b3b94faa DT |
821 | return error; |
822 | } | |
823 | ||
d56fa8a1 SW |
824 | /** |
825 | * recursive_scan - recursively scan through the end of a file | |
826 | * @ip: the inode | |
827 | * @dibh: the dinode buffer | |
828 | * @mp: the path through the metadata to the point to start | |
829 | * @height: the height the recursion is at | |
830 | * @block: the indirect block to look at | |
831 | * @first: 1 if this is the first block | |
832 | * @sm: data opaque to this function to pass to @bc | |
833 | * | |
834 | * When this is first called @height and @block should be zero and | |
835 | * @first should be 1. | |
836 | * | |
837 | * Returns: errno | |
838 | */ | |
839 | ||
840 | static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh, | |
841 | struct metapath *mp, unsigned int height, | |
842 | u64 block, int first, struct strip_mine *sm) | |
843 | { | |
844 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
845 | struct buffer_head *bh = NULL; | |
bd5437a7 | 846 | __be64 *top, *bottom, *t2; |
d56fa8a1 SW |
847 | u64 bn; |
848 | int error; | |
849 | int mh_size = sizeof(struct gfs2_meta_header); | |
850 | ||
851 | if (!height) { | |
852 | error = gfs2_meta_inode_buffer(ip, &bh); | |
853 | if (error) | |
854 | return error; | |
855 | dibh = bh; | |
856 | ||
857 | top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0]; | |
858 | bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs; | |
859 | } else { | |
860 | error = gfs2_meta_indirect_buffer(ip, height, block, 0, &bh); | |
861 | if (error) | |
862 | return error; | |
863 | ||
864 | top = (__be64 *)(bh->b_data + mh_size) + | |
865 | (first ? mp->mp_list[height] : 0); | |
866 | ||
867 | bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs; | |
868 | } | |
869 | ||
870 | error = do_strip(ip, dibh, bh, top, bottom, height, sm); | |
871 | if (error) | |
872 | goto out; | |
873 | ||
bd5437a7 BP |
874 | if (height < ip->i_height - 1) { |
875 | struct buffer_head *rabh; | |
876 | ||
877 | for (t2 = top; t2 < bottom; t2++, first = 0) { | |
878 | if (!*t2) | |
879 | continue; | |
880 | ||
881 | bn = be64_to_cpu(*t2); | |
882 | rabh = gfs2_getbuf(ip->i_gl, bn, CREATE); | |
883 | if (trylock_buffer(rabh)) { | |
884 | if (buffer_uptodate(rabh)) { | |
885 | unlock_buffer(rabh); | |
886 | brelse(rabh); | |
887 | continue; | |
888 | } | |
889 | rabh->b_end_io = end_buffer_read_sync; | |
890 | submit_bh(READA | REQ_META, rabh); | |
891 | continue; | |
892 | } | |
893 | brelse(rabh); | |
894 | } | |
d56fa8a1 SW |
895 | for (; top < bottom; top++, first = 0) { |
896 | if (!*top) | |
897 | continue; | |
898 | ||
899 | bn = be64_to_cpu(*top); | |
900 | ||
901 | error = recursive_scan(ip, dibh, mp, height + 1, bn, | |
902 | first, sm); | |
903 | if (error) | |
904 | break; | |
905 | } | |
bd5437a7 | 906 | } |
d56fa8a1 SW |
907 | out: |
908 | brelse(bh); | |
909 | return error; | |
910 | } | |
911 | ||
912 | ||
ba7f7290 SW |
913 | /** |
914 | * gfs2_block_truncate_page - Deal with zeroing out data for truncate | |
915 | * | |
916 | * This is partly borrowed from ext3. | |
917 | */ | |
ff8f33c8 | 918 | static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from) |
ba7f7290 SW |
919 | { |
920 | struct inode *inode = mapping->host; | |
921 | struct gfs2_inode *ip = GFS2_I(inode); | |
ba7f7290 SW |
922 | unsigned long index = from >> PAGE_CACHE_SHIFT; |
923 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
924 | unsigned blocksize, iblock, length, pos; | |
925 | struct buffer_head *bh; | |
926 | struct page *page; | |
ba7f7290 SW |
927 | int err; |
928 | ||
929 | page = grab_cache_page(mapping, index); | |
930 | if (!page) | |
931 | return 0; | |
932 | ||
933 | blocksize = inode->i_sb->s_blocksize; | |
934 | length = blocksize - (offset & (blocksize - 1)); | |
935 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
936 | ||
937 | if (!page_has_buffers(page)) | |
938 | create_empty_buffers(page, blocksize, 0); | |
939 | ||
940 | /* Find the buffer that contains "offset" */ | |
941 | bh = page_buffers(page); | |
942 | pos = blocksize; | |
943 | while (offset >= pos) { | |
944 | bh = bh->b_this_page; | |
945 | iblock++; | |
946 | pos += blocksize; | |
947 | } | |
948 | ||
949 | err = 0; | |
950 | ||
951 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 952 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
953 | /* unmapped? It's a hole - nothing to do */ |
954 | if (!buffer_mapped(bh)) | |
955 | goto unlock; | |
956 | } | |
957 | ||
958 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
959 | if (PageUptodate(page)) | |
960 | set_buffer_uptodate(bh); | |
961 | ||
962 | if (!buffer_uptodate(bh)) { | |
963 | err = -EIO; | |
964 | ll_rw_block(READ, 1, &bh); | |
965 | wait_on_buffer(bh); | |
966 | /* Uhhuh. Read error. Complain and punt. */ | |
967 | if (!buffer_uptodate(bh)) | |
968 | goto unlock; | |
1875f2f3 | 969 | err = 0; |
ba7f7290 SW |
970 | } |
971 | ||
bf36a713 | 972 | if (!gfs2_is_writeback(ip)) |
ba7f7290 SW |
973 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
974 | ||
eebd2aa3 | 975 | zero_user(page, offset, length); |
40bc9a27 | 976 | mark_buffer_dirty(bh); |
ba7f7290 SW |
977 | unlock: |
978 | unlock_page(page); | |
979 | page_cache_release(page); | |
980 | return err; | |
981 | } | |
982 | ||
ff8f33c8 | 983 | static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize) |
b3b94faa | 984 | { |
ff8f33c8 SW |
985 | struct gfs2_inode *ip = GFS2_I(inode); |
986 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
987 | struct address_space *mapping = inode->i_mapping; | |
b3b94faa DT |
988 | struct buffer_head *dibh; |
989 | int journaled = gfs2_is_jdata(ip); | |
990 | int error; | |
991 | ||
992 | error = gfs2_trans_begin(sdp, | |
c5392124 | 993 | RES_DINODE + (journaled ? RES_JDATA : 0), 0); |
b3b94faa DT |
994 | if (error) |
995 | return error; | |
996 | ||
997 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
998 | if (error) | |
999 | goto out; | |
1000 | ||
ff8f33c8 SW |
1001 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
1002 | ||
b3b94faa | 1003 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1004 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1005 | } else { |
ff8f33c8 SW |
1006 | if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) { |
1007 | error = gfs2_block_truncate_page(mapping, newsize); | |
1008 | if (error) | |
1009 | goto out_brelse; | |
b3b94faa | 1010 | } |
ff8f33c8 | 1011 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1012 | } |
1013 | ||
ff8f33c8 | 1014 | i_size_write(inode, newsize); |
ff8f33c8 SW |
1015 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
1016 | gfs2_dinode_out(ip, dibh->b_data); | |
b3b94faa | 1017 | |
ff8f33c8 SW |
1018 | truncate_pagecache(inode, oldsize, newsize); |
1019 | out_brelse: | |
1020 | brelse(dibh); | |
a91ea69f | 1021 | out: |
b3b94faa | 1022 | gfs2_trans_end(sdp); |
b3b94faa DT |
1023 | return error; |
1024 | } | |
1025 | ||
cd915493 | 1026 | static int trunc_dealloc(struct gfs2_inode *ip, u64 size) |
b3b94faa | 1027 | { |
9b8c81d1 | 1028 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
ecc30c79 | 1029 | unsigned int height = ip->i_height; |
cd915493 | 1030 | u64 lblock; |
b3b94faa DT |
1031 | struct metapath mp; |
1032 | int error; | |
1033 | ||
1034 | if (!size) | |
1035 | lblock = 0; | |
18ec7d5c | 1036 | else |
9b8c81d1 | 1037 | lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 1038 | |
9b8c81d1 | 1039 | find_metapath(sdp, lblock, &mp, ip->i_height); |
182fe5ab CG |
1040 | if (!gfs2_alloc_get(ip)) |
1041 | return -ENOMEM; | |
b3b94faa DT |
1042 | |
1043 | error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
1044 | if (error) | |
1045 | goto out; | |
1046 | ||
1047 | while (height--) { | |
1048 | struct strip_mine sm; | |
1049 | sm.sm_first = !!size; | |
1050 | sm.sm_height = height; | |
1051 | ||
d56fa8a1 | 1052 | error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm); |
b3b94faa DT |
1053 | if (error) |
1054 | break; | |
1055 | } | |
1056 | ||
1057 | gfs2_quota_unhold(ip); | |
1058 | ||
a91ea69f | 1059 | out: |
b3b94faa DT |
1060 | gfs2_alloc_put(ip); |
1061 | return error; | |
1062 | } | |
1063 | ||
1064 | static int trunc_end(struct gfs2_inode *ip) | |
1065 | { | |
feaa7bba | 1066 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1067 | struct buffer_head *dibh; |
1068 | int error; | |
1069 | ||
1070 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1071 | if (error) | |
1072 | return error; | |
1073 | ||
1074 | down_write(&ip->i_rw_mutex); | |
1075 | ||
1076 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1077 | if (error) | |
1078 | goto out; | |
1079 | ||
a2e0f799 | 1080 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1081 | ip->i_height = 0; |
ce276b06 | 1082 | ip->i_goal = ip->i_no_addr; |
b3b94faa DT |
1083 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
1084 | } | |
4bd91ba1 | 1085 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
383f01fb | 1086 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1087 | |
d4e9c4c3 | 1088 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); |
539e5d6b | 1089 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1090 | brelse(dibh); |
1091 | ||
a91ea69f | 1092 | out: |
b3b94faa | 1093 | up_write(&ip->i_rw_mutex); |
b3b94faa | 1094 | gfs2_trans_end(sdp); |
b3b94faa DT |
1095 | return error; |
1096 | } | |
1097 | ||
1098 | /** | |
1099 | * do_shrink - make a file smaller | |
ff8f33c8 SW |
1100 | * @inode: the inode |
1101 | * @oldsize: the current inode size | |
1102 | * @newsize: the size to make the file | |
b3b94faa | 1103 | * |
ff8f33c8 SW |
1104 | * Called with an exclusive lock on @inode. The @size must |
1105 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
1106 | * |
1107 | * Returns: errno | |
1108 | */ | |
1109 | ||
ff8f33c8 | 1110 | static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize) |
b3b94faa | 1111 | { |
ff8f33c8 | 1112 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
1113 | int error; |
1114 | ||
ff8f33c8 | 1115 | error = trunc_start(inode, oldsize, newsize); |
b3b94faa DT |
1116 | if (error < 0) |
1117 | return error; | |
ff8f33c8 | 1118 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
1119 | return 0; |
1120 | ||
ff8f33c8 SW |
1121 | error = trunc_dealloc(ip, newsize); |
1122 | if (error == 0) | |
b3b94faa DT |
1123 | error = trunc_end(ip); |
1124 | ||
1125 | return error; | |
1126 | } | |
1127 | ||
ff8f33c8 | 1128 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 1129 | { |
ff8f33c8 SW |
1130 | u64 size = inode->i_size; |
1131 | int ret; | |
1132 | ||
1133 | ret = do_shrink(inode, size, size); | |
1134 | WARN_ON(ret != 0); | |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * do_grow - Touch and update inode size | |
1139 | * @inode: The inode | |
1140 | * @size: The new size | |
1141 | * | |
1142 | * This function updates the timestamps on the inode and | |
1143 | * may also increase the size of the inode. This function | |
1144 | * must not be called with @size any smaller than the current | |
1145 | * inode size. | |
1146 | * | |
1147 | * Although it is not strictly required to unstuff files here, | |
1148 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
1149 | * code which will result in a buffer overrun if the size is larger | |
1150 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 1151 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
1152 | * just update the inode size directly. |
1153 | * | |
1154 | * Returns: 0 on success, or -ve on error | |
1155 | */ | |
1156 | ||
1157 | static int do_grow(struct inode *inode, u64 size) | |
1158 | { | |
1159 | struct gfs2_inode *ip = GFS2_I(inode); | |
1160 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
a13b8c5f | 1161 | struct buffer_head *dibh; |
ff8f33c8 | 1162 | struct gfs2_alloc *al = NULL; |
a13b8c5f WC |
1163 | int error; |
1164 | ||
ff8f33c8 SW |
1165 | if (gfs2_is_stuffed(ip) && |
1166 | (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) { | |
1167 | al = gfs2_alloc_get(ip); | |
1168 | if (al == NULL) | |
1169 | return -ENOMEM; | |
1170 | ||
1171 | error = gfs2_quota_lock_check(ip); | |
1172 | if (error) | |
1173 | goto do_grow_alloc_put; | |
1174 | ||
1175 | al->al_requested = 1; | |
1176 | error = gfs2_inplace_reserve(ip); | |
1177 | if (error) | |
1178 | goto do_grow_qunlock; | |
1179 | } | |
1180 | ||
bf97b673 | 1181 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT, 0); |
a13b8c5f | 1182 | if (error) |
ff8f33c8 | 1183 | goto do_grow_release; |
a13b8c5f | 1184 | |
ff8f33c8 SW |
1185 | if (al) { |
1186 | error = gfs2_unstuff_dinode(ip, NULL); | |
1187 | if (error) | |
1188 | goto do_end_trans; | |
1189 | } | |
a13b8c5f WC |
1190 | |
1191 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1192 | if (error) | |
ff8f33c8 | 1193 | goto do_end_trans; |
a13b8c5f | 1194 | |
ff8f33c8 | 1195 | i_size_write(inode, size); |
a13b8c5f WC |
1196 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; |
1197 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
1198 | gfs2_dinode_out(ip, dibh->b_data); | |
1199 | brelse(dibh); | |
1200 | ||
ff8f33c8 | 1201 | do_end_trans: |
a13b8c5f | 1202 | gfs2_trans_end(sdp); |
ff8f33c8 SW |
1203 | do_grow_release: |
1204 | if (al) { | |
1205 | gfs2_inplace_release(ip); | |
1206 | do_grow_qunlock: | |
1207 | gfs2_quota_unlock(ip); | |
1208 | do_grow_alloc_put: | |
1209 | gfs2_alloc_put(ip); | |
1210 | } | |
a13b8c5f WC |
1211 | return error; |
1212 | } | |
1213 | ||
b3b94faa | 1214 | /** |
ff8f33c8 SW |
1215 | * gfs2_setattr_size - make a file a given size |
1216 | * @inode: the inode | |
1217 | * @newsize: the size to make the file | |
b3b94faa | 1218 | * |
ff8f33c8 SW |
1219 | * The file size can grow, shrink, or stay the same size. This |
1220 | * is called holding i_mutex and an exclusive glock on the inode | |
1221 | * in question. | |
b3b94faa DT |
1222 | * |
1223 | * Returns: errno | |
1224 | */ | |
1225 | ||
ff8f33c8 | 1226 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 1227 | { |
ff8f33c8 SW |
1228 | int ret; |
1229 | u64 oldsize; | |
b3b94faa | 1230 | |
ff8f33c8 | 1231 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 1232 | |
ff8f33c8 SW |
1233 | ret = inode_newsize_ok(inode, newsize); |
1234 | if (ret) | |
1235 | return ret; | |
b3b94faa | 1236 | |
562c72aa CH |
1237 | inode_dio_wait(inode); |
1238 | ||
ff8f33c8 SW |
1239 | oldsize = inode->i_size; |
1240 | if (newsize >= oldsize) | |
1241 | return do_grow(inode, newsize); | |
1242 | ||
1243 | return do_shrink(inode, oldsize, newsize); | |
b3b94faa DT |
1244 | } |
1245 | ||
1246 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1247 | { | |
1248 | int error; | |
a2e0f799 | 1249 | error = trunc_dealloc(ip, i_size_read(&ip->i_inode)); |
b3b94faa DT |
1250 | if (!error) |
1251 | error = trunc_end(ip); | |
1252 | return error; | |
1253 | } | |
1254 | ||
1255 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1256 | { | |
1257 | return trunc_dealloc(ip, 0); | |
1258 | } | |
1259 | ||
b3b94faa DT |
1260 | /** |
1261 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
1262 | * @ip: the file being written to | |
1263 | * @offset: the offset to write to | |
1264 | * @len: the number of bytes being written | |
b3b94faa | 1265 | * |
461cb419 | 1266 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
1267 | */ |
1268 | ||
cd915493 | 1269 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 1270 | unsigned int len) |
b3b94faa | 1271 | { |
feaa7bba | 1272 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
1273 | struct buffer_head bh; |
1274 | unsigned int shift; | |
1275 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 1276 | u64 end_of_file; |
b3b94faa | 1277 | |
b3b94faa DT |
1278 | if (!len) |
1279 | return 0; | |
1280 | ||
1281 | if (gfs2_is_stuffed(ip)) { | |
1282 | if (offset + len > | |
1283 | sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
461cb419 | 1284 | return 1; |
b3b94faa DT |
1285 | return 0; |
1286 | } | |
1287 | ||
941e6d7d | 1288 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 1289 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 1290 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
1291 | lblock = offset >> shift; |
1292 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
1293 | if (lblock_stop > end_of_file) | |
461cb419 | 1294 | return 1; |
b3b94faa | 1295 | |
941e6d7d SW |
1296 | size = (lblock_stop - lblock) << shift; |
1297 | do { | |
1298 | bh.b_state = 0; | |
1299 | bh.b_size = size; | |
1300 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
1301 | if (!buffer_mapped(&bh)) | |
461cb419 | 1302 | return 1; |
941e6d7d SW |
1303 | size -= bh.b_size; |
1304 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
1305 | } while(size > 0); | |
b3b94faa DT |
1306 | |
1307 | return 0; | |
1308 | } | |
1309 |