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> |
3974320c | 16 | #include <linux/iomap.h> |
b3b94faa DT |
17 | |
18 | #include "gfs2.h" | |
5c676f6d | 19 | #include "incore.h" |
b3b94faa DT |
20 | #include "bmap.h" |
21 | #include "glock.h" | |
22 | #include "inode.h" | |
b3b94faa | 23 | #include "meta_io.h" |
b3b94faa DT |
24 | #include "quota.h" |
25 | #include "rgrp.h" | |
45138990 | 26 | #include "log.h" |
4c16c36a | 27 | #include "super.h" |
b3b94faa | 28 | #include "trans.h" |
18ec7d5c | 29 | #include "dir.h" |
5c676f6d | 30 | #include "util.h" |
63997775 | 31 | #include "trace_gfs2.h" |
b3b94faa DT |
32 | |
33 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
34 | * block is 512, so __u16 is fine for that. It saves stack space to | |
35 | * keep it small. | |
36 | */ | |
37 | struct metapath { | |
dbac6710 | 38 | struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT]; |
b3b94faa | 39 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; |
5f8bd444 BP |
40 | int mp_fheight; /* find_metapath height */ |
41 | int mp_aheight; /* actual height (lookup height) */ | |
b3b94faa DT |
42 | }; |
43 | ||
f25ef0c1 SW |
44 | /** |
45 | * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page | |
46 | * @ip: the inode | |
47 | * @dibh: the dinode buffer | |
48 | * @block: the block number that was allocated | |
ff8f33c8 | 49 | * @page: The (optional) page. This is looked up if @page is NULL |
f25ef0c1 SW |
50 | * |
51 | * Returns: errno | |
52 | */ | |
53 | ||
54 | static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, | |
cd915493 | 55 | u64 block, struct page *page) |
f25ef0c1 | 56 | { |
f25ef0c1 SW |
57 | struct inode *inode = &ip->i_inode; |
58 | struct buffer_head *bh; | |
59 | int release = 0; | |
60 | ||
61 | if (!page || page->index) { | |
220cca2a | 62 | page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS); |
f25ef0c1 SW |
63 | if (!page) |
64 | return -ENOMEM; | |
65 | release = 1; | |
66 | } | |
67 | ||
68 | if (!PageUptodate(page)) { | |
69 | void *kaddr = kmap(page); | |
602c89d2 SW |
70 | u64 dsize = i_size_read(inode); |
71 | ||
72 | if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode))) | |
73 | dsize = dibh->b_size - sizeof(struct gfs2_dinode); | |
f25ef0c1 | 74 | |
602c89d2 | 75 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 76 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
f25ef0c1 SW |
77 | kunmap(page); |
78 | ||
79 | SetPageUptodate(page); | |
80 | } | |
81 | ||
82 | if (!page_has_buffers(page)) | |
47a9a527 FF |
83 | create_empty_buffers(page, BIT(inode->i_blkbits), |
84 | BIT(BH_Uptodate)); | |
f25ef0c1 SW |
85 | |
86 | bh = page_buffers(page); | |
87 | ||
88 | if (!buffer_mapped(bh)) | |
89 | map_bh(bh, inode->i_sb, block); | |
90 | ||
91 | set_buffer_uptodate(bh); | |
eaf96527 SW |
92 | if (!gfs2_is_jdata(ip)) |
93 | mark_buffer_dirty(bh); | |
bf36a713 | 94 | if (!gfs2_is_writeback(ip)) |
350a9b0a | 95 | gfs2_trans_add_data(ip->i_gl, bh); |
f25ef0c1 SW |
96 | |
97 | if (release) { | |
98 | unlock_page(page); | |
09cbfeaf | 99 | put_page(page); |
f25ef0c1 SW |
100 | } |
101 | ||
102 | return 0; | |
103 | } | |
104 | ||
b3b94faa DT |
105 | /** |
106 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
107 | * @ip: The GFS2 inode to unstuff | |
ff8f33c8 | 108 | * @page: The (optional) page. This is looked up if the @page is NULL |
b3b94faa DT |
109 | * |
110 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
111 | * that the height can be grown in the traditional way. | |
112 | * | |
113 | * Returns: errno | |
114 | */ | |
115 | ||
f25ef0c1 | 116 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
117 | { |
118 | struct buffer_head *bh, *dibh; | |
48516ced | 119 | struct gfs2_dinode *di; |
cd915493 | 120 | u64 block = 0; |
18ec7d5c | 121 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
122 | int error; |
123 | ||
124 | down_write(&ip->i_rw_mutex); | |
125 | ||
126 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
127 | if (error) | |
128 | goto out; | |
907b9bce | 129 | |
a2e0f799 | 130 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
131 | /* Get a free block, fill it with the stuffed data, |
132 | and write it out to disk */ | |
133 | ||
b45e41d7 | 134 | unsigned int n = 1; |
6e87ed0f | 135 | error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL); |
09010978 SW |
136 | if (error) |
137 | goto out_brelse; | |
18ec7d5c | 138 | if (isdir) { |
5731be53 | 139 | gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 140 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
141 | if (error) |
142 | goto out_brelse; | |
48516ced | 143 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
144 | dibh, sizeof(struct gfs2_dinode)); |
145 | brelse(bh); | |
146 | } else { | |
f25ef0c1 | 147 | error = gfs2_unstuffer_page(ip, dibh, block, page); |
b3b94faa DT |
148 | if (error) |
149 | goto out_brelse; | |
150 | } | |
151 | } | |
152 | ||
153 | /* Set up the pointer to the new block */ | |
154 | ||
350a9b0a | 155 | gfs2_trans_add_meta(ip->i_gl, dibh); |
48516ced | 156 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
157 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
158 | ||
a2e0f799 | 159 | if (i_size_read(&ip->i_inode)) { |
48516ced | 160 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
161 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
162 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
163 | } |
164 | ||
ecc30c79 | 165 | ip->i_height = 1; |
48516ced | 166 | di->di_height = cpu_to_be16(1); |
b3b94faa | 167 | |
a91ea69f | 168 | out_brelse: |
b3b94faa | 169 | brelse(dibh); |
a91ea69f | 170 | out: |
b3b94faa | 171 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
172 | return error; |
173 | } | |
174 | ||
b3b94faa DT |
175 | |
176 | /** | |
177 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 178 | * @sdp: The superblock |
b3b94faa DT |
179 | * @mp: The metapath to return the result in |
180 | * @block: The disk block to look up | |
9b8c81d1 | 181 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
182 | * |
183 | * This routine returns a struct metapath structure that defines a path | |
184 | * through the metadata of inode "ip" to get to block "block". | |
185 | * | |
186 | * Example: | |
187 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
188 | * filesystem with a blocksize of 4096. | |
189 | * | |
190 | * find_metapath() would return a struct metapath structure set to: | |
191 | * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48, | |
192 | * and mp_list[2] = 165. | |
193 | * | |
194 | * That means that in order to get to the block containing the byte at | |
195 | * offset 101342453, we would load the indirect block pointed to by pointer | |
196 | * 0 in the dinode. We would then load the indirect block pointed to by | |
197 | * pointer 48 in that indirect block. We would then load the data block | |
198 | * pointed to by pointer 165 in that indirect block. | |
199 | * | |
200 | * ---------------------------------------- | |
201 | * | Dinode | | | |
202 | * | | 4| | |
203 | * | |0 1 2 3 4 5 9| | |
204 | * | | 6| | |
205 | * ---------------------------------------- | |
206 | * | | |
207 | * | | |
208 | * V | |
209 | * ---------------------------------------- | |
210 | * | Indirect Block | | |
211 | * | 5| | |
212 | * | 4 4 4 4 4 5 5 1| | |
213 | * |0 5 6 7 8 9 0 1 2| | |
214 | * ---------------------------------------- | |
215 | * | | |
216 | * | | |
217 | * V | |
218 | * ---------------------------------------- | |
219 | * | Indirect Block | | |
220 | * | 1 1 1 1 1 5| | |
221 | * | 6 6 6 6 6 1| | |
222 | * |0 3 4 5 6 7 2| | |
223 | * ---------------------------------------- | |
224 | * | | |
225 | * | | |
226 | * V | |
227 | * ---------------------------------------- | |
228 | * | Data block containing offset | | |
229 | * | 101342453 | | |
230 | * | | | |
231 | * | | | |
232 | * ---------------------------------------- | |
233 | * | |
234 | */ | |
235 | ||
9b8c81d1 SW |
236 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
237 | struct metapath *mp, unsigned int height) | |
b3b94faa | 238 | { |
b3b94faa DT |
239 | unsigned int i; |
240 | ||
5f8bd444 | 241 | mp->mp_fheight = height; |
9b8c81d1 | 242 | for (i = height; i--;) |
7eabb77e | 243 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
244 | } |
245 | ||
5af4e7a0 | 246 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 247 | { |
5af4e7a0 BM |
248 | if (mp->mp_list[0] == 0) |
249 | return 2; | |
250 | return 1; | |
9b8c81d1 SW |
251 | } |
252 | ||
d552a2b9 | 253 | /** |
20cdc193 | 254 | * metaptr1 - Return the first possible metadata pointer in a metapath buffer |
d552a2b9 BP |
255 | * @height: The metadata height (0 = dinode) |
256 | * @mp: The metapath | |
257 | */ | |
258 | static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp) | |
259 | { | |
260 | struct buffer_head *bh = mp->mp_bh[height]; | |
261 | if (height == 0) | |
262 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode))); | |
263 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header))); | |
264 | } | |
265 | ||
b3b94faa DT |
266 | /** |
267 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
268 | * @height: The metadata height (0 = dinode) |
269 | * @mp: The metapath | |
270 | * | |
271 | * Return a pointer to the block number of the next height of the metadata | |
272 | * tree given a buffer containing the pointer to the current height of the | |
273 | * metadata tree. | |
274 | */ | |
275 | ||
9b8c81d1 | 276 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 277 | { |
d552a2b9 BP |
278 | __be64 *p = metaptr1(height, mp); |
279 | return p + mp->mp_list[height]; | |
b3b94faa DT |
280 | } |
281 | ||
b99b98dc SW |
282 | static void gfs2_metapath_ra(struct gfs2_glock *gl, |
283 | const struct buffer_head *bh, const __be64 *pos) | |
284 | { | |
285 | struct buffer_head *rabh; | |
286 | const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size); | |
287 | const __be64 *t; | |
288 | ||
289 | for (t = pos; t < endp; t++) { | |
290 | if (!*t) | |
291 | continue; | |
292 | ||
293 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
294 | if (trylock_buffer(rabh)) { | |
295 | if (!buffer_uptodate(rabh)) { | |
296 | rabh->b_end_io = end_buffer_read_sync; | |
e477b24b CL |
297 | submit_bh(REQ_OP_READ, |
298 | REQ_RAHEAD | REQ_META | REQ_PRIO, | |
299 | rabh); | |
b99b98dc SW |
300 | continue; |
301 | } | |
302 | unlock_buffer(rabh); | |
303 | } | |
304 | brelse(rabh); | |
305 | } | |
306 | } | |
307 | ||
d552a2b9 BP |
308 | /** |
309 | * lookup_mp_height - helper function for lookup_metapath | |
310 | * @ip: the inode | |
311 | * @mp: the metapath | |
312 | * @h: the height which needs looking up | |
313 | */ | |
314 | static int lookup_mp_height(struct gfs2_inode *ip, struct metapath *mp, int h) | |
315 | { | |
316 | __be64 *ptr = metapointer(h, mp); | |
317 | u64 dblock = be64_to_cpu(*ptr); | |
318 | ||
319 | if (!dblock) | |
320 | return h + 1; | |
321 | ||
322 | return gfs2_meta_indirect_buffer(ip, h + 1, dblock, &mp->mp_bh[h + 1]); | |
323 | } | |
324 | ||
b3b94faa | 325 | /** |
9b8c81d1 SW |
326 | * lookup_metapath - Walk the metadata tree to a specific point |
327 | * @ip: The inode | |
b3b94faa | 328 | * @mp: The metapath |
b3b94faa | 329 | * |
9b8c81d1 SW |
330 | * Assumes that the inode's buffer has already been looked up and |
331 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
332 | * by find_metapath(). | |
333 | * | |
334 | * If this function encounters part of the tree which has not been | |
335 | * allocated, it returns the current height of the tree at the point | |
336 | * at which it found the unallocated block. Blocks which are found are | |
337 | * added to the mp->mp_bh[] list. | |
b3b94faa | 338 | * |
9b8c81d1 | 339 | * Returns: error or height of metadata tree |
b3b94faa DT |
340 | */ |
341 | ||
9b8c81d1 | 342 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 343 | { |
11707ea0 SW |
344 | unsigned int end_of_metadata = ip->i_height - 1; |
345 | unsigned int x; | |
e23159d2 | 346 | int ret; |
11707ea0 SW |
347 | |
348 | for (x = 0; x < end_of_metadata; x++) { | |
d552a2b9 | 349 | ret = lookup_mp_height(ip, mp, x); |
11707ea0 | 350 | if (ret) |
5f8bd444 | 351 | goto out; |
11707ea0 SW |
352 | } |
353 | ||
5f8bd444 BP |
354 | ret = ip->i_height; |
355 | out: | |
356 | mp->mp_aheight = ret; | |
357 | return ret; | |
dbac6710 SW |
358 | } |
359 | ||
d552a2b9 BP |
360 | /** |
361 | * fillup_metapath - fill up buffers for the metadata path to a specific height | |
362 | * @ip: The inode | |
363 | * @mp: The metapath | |
364 | * @h: The height to which it should be mapped | |
365 | * | |
366 | * Similar to lookup_metapath, but does lookups for a range of heights | |
367 | * | |
368 | * Returns: error or height of metadata tree | |
369 | */ | |
370 | ||
371 | static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h) | |
372 | { | |
373 | unsigned int start_h = h - 1; | |
374 | int ret; | |
375 | ||
376 | if (h) { | |
377 | /* find the first buffer we need to look up. */ | |
378 | while (start_h > 0 && mp->mp_bh[start_h] == NULL) | |
379 | start_h--; | |
380 | for (; start_h < h; start_h++) { | |
381 | ret = lookup_mp_height(ip, mp, start_h); | |
382 | if (ret) | |
383 | return ret; | |
384 | } | |
385 | } | |
386 | return ip->i_height; | |
387 | } | |
388 | ||
9b8c81d1 | 389 | static inline void release_metapath(struct metapath *mp) |
dbac6710 SW |
390 | { |
391 | int i; | |
392 | ||
9b8c81d1 SW |
393 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
394 | if (mp->mp_bh[i] == NULL) | |
395 | break; | |
396 | brelse(mp->mp_bh[i]); | |
397 | } | |
11707ea0 SW |
398 | } |
399 | ||
30cbf189 SW |
400 | /** |
401 | * gfs2_extent_length - Returns length of an extent of blocks | |
402 | * @start: Start of the buffer | |
403 | * @len: Length of the buffer in bytes | |
404 | * @ptr: Current position in the buffer | |
405 | * @limit: Max extent length to return (0 = unlimited) | |
406 | * @eob: Set to 1 if we hit "end of block" | |
407 | * | |
408 | * If the first block is zero (unallocated) it will return the number of | |
409 | * unallocated blocks in the extent, otherwise it will return the number | |
410 | * of contiguous blocks in the extent. | |
411 | * | |
412 | * Returns: The length of the extent (minimum of one block) | |
413 | */ | |
414 | ||
b650738c | 415 | static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob) |
30cbf189 SW |
416 | { |
417 | const __be64 *end = (start + len); | |
418 | const __be64 *first = ptr; | |
419 | u64 d = be64_to_cpu(*ptr); | |
420 | ||
421 | *eob = 0; | |
422 | do { | |
423 | ptr++; | |
424 | if (ptr >= end) | |
425 | break; | |
426 | if (limit && --limit == 0) | |
427 | break; | |
428 | if (d) | |
429 | d++; | |
430 | } while(be64_to_cpu(*ptr) == d); | |
431 | if (ptr >= end) | |
432 | *eob = 1; | |
433 | return (ptr - first); | |
434 | } | |
435 | ||
9b8c81d1 | 436 | static inline void bmap_lock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 437 | { |
4cf1ed81 SW |
438 | if (create) |
439 | down_write(&ip->i_rw_mutex); | |
440 | else | |
441 | down_read(&ip->i_rw_mutex); | |
442 | } | |
443 | ||
9b8c81d1 | 444 | static inline void bmap_unlock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 445 | { |
4cf1ed81 SW |
446 | if (create) |
447 | up_write(&ip->i_rw_mutex); | |
448 | else | |
449 | up_read(&ip->i_rw_mutex); | |
450 | } | |
451 | ||
9b8c81d1 SW |
452 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
453 | struct gfs2_glock *gl, unsigned int i, | |
454 | unsigned offset, u64 bn) | |
455 | { | |
456 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
457 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
458 | sizeof(struct gfs2_dinode))); | |
459 | BUG_ON(i < 1); | |
460 | BUG_ON(mp->mp_bh[i] != NULL); | |
461 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
350a9b0a | 462 | gfs2_trans_add_meta(gl, mp->mp_bh[i]); |
9b8c81d1 SW |
463 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
464 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
465 | ptr += offset; | |
466 | *ptr = cpu_to_be64(bn); | |
467 | return ptr; | |
468 | } | |
469 | ||
470 | enum alloc_state { | |
471 | ALLOC_DATA = 0, | |
472 | ALLOC_GROW_DEPTH = 1, | |
473 | ALLOC_GROW_HEIGHT = 2, | |
474 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
475 | }; | |
476 | ||
d552a2b9 BP |
477 | static inline unsigned int hptrs(struct gfs2_sbd *sdp, const unsigned int hgt) |
478 | { | |
479 | if (hgt) | |
480 | return sdp->sd_inptrs; | |
481 | return sdp->sd_diptrs; | |
482 | } | |
483 | ||
9b8c81d1 SW |
484 | /** |
485 | * gfs2_bmap_alloc - Build a metadata tree of the requested height | |
486 | * @inode: The GFS2 inode | |
487 | * @lblock: The logical starting block of the extent | |
488 | * @bh_map: This is used to return the mapping details | |
5f8bd444 BP |
489 | * @zero_new: True if newly allocated blocks should be zeroed |
490 | * @mp: The metapath, with proper height information calculated | |
9b8c81d1 | 491 | * @maxlen: The max number of data blocks to alloc |
5f8bd444 BP |
492 | * @dblock: Pointer to return the resulting new block |
493 | * @dblks: Pointer to return the number of blocks allocated | |
9b8c81d1 SW |
494 | * |
495 | * In this routine we may have to alloc: | |
496 | * i) Indirect blocks to grow the metadata tree height | |
497 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
498 | * iii) Data blocks | |
499 | * | |
500 | * The function is in two parts. The first part works out the total | |
501 | * number of blocks which we need. The second part does the actual | |
502 | * allocation asking for an extent at a time (if enough contiguous free | |
503 | * blocks are available, there will only be one request per bmap call) | |
504 | * and uses the state machine to initialise the blocks in order. | |
505 | * | |
506 | * Returns: errno on error | |
507 | */ | |
508 | ||
3974320c BP |
509 | static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap, |
510 | unsigned flags, struct metapath *mp) | |
9b8c81d1 SW |
511 | { |
512 | struct gfs2_inode *ip = GFS2_I(inode); | |
513 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
64dd153c | 514 | struct super_block *sb = sdp->sd_vfs; |
9b8c81d1 | 515 | struct buffer_head *dibh = mp->mp_bh[0]; |
5f8bd444 | 516 | u64 bn; |
5af4e7a0 | 517 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
3974320c | 518 | unsigned dblks = 0; |
9b8c81d1 | 519 | unsigned ptrs_per_blk; |
5f8bd444 | 520 | const unsigned end_of_metadata = mp->mp_fheight - 1; |
64dd153c | 521 | int ret; |
9b8c81d1 SW |
522 | enum alloc_state state; |
523 | __be64 *ptr; | |
524 | __be64 zero_bn = 0; | |
3974320c | 525 | size_t maxlen = iomap->length >> inode->i_blkbits; |
9b8c81d1 | 526 | |
5f8bd444 | 527 | BUG_ON(mp->mp_aheight < 1); |
9b8c81d1 SW |
528 | BUG_ON(dibh == NULL); |
529 | ||
350a9b0a | 530 | gfs2_trans_add_meta(ip->i_gl, dibh); |
9b8c81d1 | 531 | |
5f8bd444 | 532 | if (mp->mp_fheight == mp->mp_aheight) { |
9b8c81d1 | 533 | struct buffer_head *bh; |
3974320c BP |
534 | int eob; |
535 | ||
9b8c81d1 SW |
536 | /* Bottom indirect block exists, find unalloced extent size */ |
537 | ptr = metapointer(end_of_metadata, mp); | |
538 | bh = mp->mp_bh[end_of_metadata]; | |
3974320c BP |
539 | dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, |
540 | maxlen, &eob); | |
541 | BUG_ON(dblks < 1); | |
9b8c81d1 SW |
542 | state = ALLOC_DATA; |
543 | } else { | |
544 | /* Need to allocate indirect blocks */ | |
5f8bd444 BP |
545 | ptrs_per_blk = mp->mp_fheight > 1 ? sdp->sd_inptrs : |
546 | sdp->sd_diptrs; | |
3974320c BP |
547 | dblks = min(maxlen, (size_t)(ptrs_per_blk - |
548 | mp->mp_list[end_of_metadata])); | |
5f8bd444 | 549 | if (mp->mp_fheight == ip->i_height) { |
9b8c81d1 | 550 | /* Writing into existing tree, extend tree down */ |
5f8bd444 | 551 | iblks = mp->mp_fheight - mp->mp_aheight; |
9b8c81d1 SW |
552 | state = ALLOC_GROW_DEPTH; |
553 | } else { | |
554 | /* Building up tree height */ | |
555 | state = ALLOC_GROW_HEIGHT; | |
5f8bd444 | 556 | iblks = mp->mp_fheight - ip->i_height; |
5af4e7a0 | 557 | branch_start = metapath_branch_start(mp); |
5f8bd444 | 558 | iblks += (mp->mp_fheight - branch_start); |
9b8c81d1 SW |
559 | } |
560 | } | |
561 | ||
562 | /* start of the second part of the function (state machine) */ | |
563 | ||
3974320c | 564 | blks = dblks + iblks; |
5f8bd444 | 565 | i = mp->mp_aheight; |
9b8c81d1 | 566 | do { |
09010978 | 567 | int error; |
9b8c81d1 | 568 | n = blks - alloced; |
6e87ed0f | 569 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
570 | if (error) |
571 | return error; | |
9b8c81d1 SW |
572 | alloced += n; |
573 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
574 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
575 | switch (state) { | |
576 | /* Growing height of tree */ | |
577 | case ALLOC_GROW_HEIGHT: | |
578 | if (i == 1) { | |
579 | ptr = (__be64 *)(dibh->b_data + | |
580 | sizeof(struct gfs2_dinode)); | |
581 | zero_bn = *ptr; | |
582 | } | |
5f8bd444 BP |
583 | for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0; |
584 | i++, n--) | |
9b8c81d1 | 585 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); |
5f8bd444 | 586 | if (i - 1 == mp->mp_fheight - ip->i_height) { |
9b8c81d1 SW |
587 | i--; |
588 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
589 | sizeof(struct gfs2_meta_header), | |
590 | dibh, sizeof(struct gfs2_dinode)); | |
591 | gfs2_buffer_clear_tail(dibh, | |
592 | sizeof(struct gfs2_dinode) + | |
593 | sizeof(__be64)); | |
594 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
595 | sizeof(struct gfs2_meta_header)); | |
596 | *ptr = zero_bn; | |
597 | state = ALLOC_GROW_DEPTH; | |
5f8bd444 | 598 | for(i = branch_start; i < mp->mp_fheight; i++) { |
9b8c81d1 SW |
599 | if (mp->mp_bh[i] == NULL) |
600 | break; | |
601 | brelse(mp->mp_bh[i]); | |
602 | mp->mp_bh[i] = NULL; | |
603 | } | |
5af4e7a0 | 604 | i = branch_start; |
9b8c81d1 SW |
605 | } |
606 | if (n == 0) | |
607 | break; | |
608 | /* Branching from existing tree */ | |
609 | case ALLOC_GROW_DEPTH: | |
5f8bd444 | 610 | if (i > 1 && i < mp->mp_fheight) |
350a9b0a | 611 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]); |
5f8bd444 | 612 | for (; i < mp->mp_fheight && n > 0; i++, n--) |
9b8c81d1 SW |
613 | gfs2_indirect_init(mp, ip->i_gl, i, |
614 | mp->mp_list[i-1], bn++); | |
5f8bd444 | 615 | if (i == mp->mp_fheight) |
9b8c81d1 SW |
616 | state = ALLOC_DATA; |
617 | if (n == 0) | |
618 | break; | |
619 | /* Tree complete, adding data blocks */ | |
620 | case ALLOC_DATA: | |
3974320c | 621 | BUG_ON(n > dblks); |
9b8c81d1 | 622 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); |
350a9b0a | 623 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]); |
3974320c | 624 | dblks = n; |
9b8c81d1 | 625 | ptr = metapointer(end_of_metadata, mp); |
3974320c BP |
626 | iomap->addr = bn << inode->i_blkbits; |
627 | iomap->flags |= IOMAP_F_NEW; | |
9b8c81d1 SW |
628 | while (n-- > 0) |
629 | *ptr++ = cpu_to_be64(bn++); | |
3974320c BP |
630 | if (flags & IOMAP_ZERO) { |
631 | ret = sb_issue_zeroout(sb, iomap->addr >> inode->i_blkbits, | |
632 | dblks, GFP_NOFS); | |
64dd153c BM |
633 | if (ret) { |
634 | fs_err(sdp, | |
635 | "Failed to zero data buffers\n"); | |
3974320c | 636 | flags &= ~IOMAP_ZERO; |
64dd153c BM |
637 | } |
638 | } | |
9b8c81d1 SW |
639 | break; |
640 | } | |
3974320c | 641 | } while (iomap->addr == IOMAP_NULL_ADDR); |
9b8c81d1 | 642 | |
3974320c | 643 | iomap->length = (u64)dblks << inode->i_blkbits; |
5f8bd444 | 644 | ip->i_height = mp->mp_fheight; |
9b8c81d1 SW |
645 | gfs2_add_inode_blocks(&ip->i_inode, alloced); |
646 | gfs2_dinode_out(ip, mp->mp_bh[0]->b_data); | |
9b8c81d1 SW |
647 | return 0; |
648 | } | |
649 | ||
b3b94faa | 650 | /** |
3974320c | 651 | * hole_size - figure out the size of a hole |
fd88de56 | 652 | * @inode: The inode |
3974320c BP |
653 | * @lblock: The logical starting block number |
654 | * @mp: The metapath | |
b3b94faa | 655 | * |
3974320c | 656 | * Returns: The hole size in bytes |
b3b94faa | 657 | * |
b3b94faa | 658 | */ |
3974320c BP |
659 | static u64 hole_size(struct inode *inode, sector_t lblock, struct metapath *mp) |
660 | { | |
661 | struct gfs2_inode *ip = GFS2_I(inode); | |
662 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
663 | struct metapath mp_eof; | |
664 | u64 factor = 1; | |
665 | int hgt; | |
666 | u64 holesz = 0; | |
667 | const __be64 *first, *end, *ptr; | |
668 | const struct buffer_head *bh; | |
669 | u64 lblock_stop = (i_size_read(inode) - 1) >> inode->i_blkbits; | |
670 | int zeroptrs; | |
671 | bool done = false; | |
672 | ||
673 | /* Get another metapath, to the very last byte */ | |
674 | find_metapath(sdp, lblock_stop, &mp_eof, ip->i_height); | |
675 | for (hgt = ip->i_height - 1; hgt >= 0 && !done; hgt--) { | |
676 | bh = mp->mp_bh[hgt]; | |
677 | if (bh) { | |
678 | zeroptrs = 0; | |
679 | first = metapointer(hgt, mp); | |
680 | end = (const __be64 *)(bh->b_data + bh->b_size); | |
681 | ||
682 | for (ptr = first; ptr < end; ptr++) { | |
683 | if (*ptr) { | |
684 | done = true; | |
685 | break; | |
686 | } else { | |
687 | zeroptrs++; | |
688 | } | |
689 | } | |
690 | } else { | |
691 | zeroptrs = sdp->sd_inptrs; | |
692 | } | |
693 | if (factor * zeroptrs >= lblock_stop - lblock + 1) { | |
694 | holesz = lblock_stop - lblock + 1; | |
695 | break; | |
696 | } | |
697 | holesz += factor * zeroptrs; | |
b3b94faa | 698 | |
3974320c BP |
699 | factor *= sdp->sd_inptrs; |
700 | if (hgt && (mp->mp_list[hgt - 1] < mp_eof.mp_list[hgt - 1])) | |
701 | (mp->mp_list[hgt - 1])++; | |
702 | } | |
703 | return holesz << inode->i_blkbits; | |
704 | } | |
705 | ||
706 | static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap) | |
707 | { | |
708 | struct gfs2_inode *ip = GFS2_I(inode); | |
709 | ||
710 | iomap->addr = (ip->i_no_addr << inode->i_blkbits) + | |
711 | sizeof(struct gfs2_dinode); | |
712 | iomap->offset = 0; | |
713 | iomap->length = i_size_read(inode); | |
714 | iomap->type = IOMAP_MAPPED; | |
715 | iomap->flags = IOMAP_F_DATA_INLINE; | |
716 | } | |
717 | ||
718 | /** | |
719 | * gfs2_iomap_begin - Map blocks from an inode to disk blocks | |
720 | * @inode: The inode | |
721 | * @pos: Starting position in bytes | |
722 | * @length: Length to map, in bytes | |
723 | * @flags: iomap flags | |
724 | * @iomap: The iomap structure | |
725 | * | |
726 | * Returns: errno | |
727 | */ | |
728 | int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, | |
729 | unsigned flags, struct iomap *iomap) | |
b3b94faa | 730 | { |
feaa7bba SW |
731 | struct gfs2_inode *ip = GFS2_I(inode); |
732 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
3974320c | 733 | struct metapath mp = { .mp_aheight = 1, }; |
20cdc193 | 734 | unsigned int factor = sdp->sd_sb.sb_bsize; |
ecc30c79 | 735 | const u64 *arr = sdp->sd_heightsize; |
9b8c81d1 | 736 | __be64 *ptr; |
3974320c BP |
737 | sector_t lblock; |
738 | sector_t lend; | |
9b8c81d1 SW |
739 | int ret; |
740 | int eob; | |
741 | unsigned int len; | |
742 | struct buffer_head *bh; | |
743 | u8 height; | |
7276b3b0 | 744 | |
3974320c BP |
745 | trace_gfs2_iomap_start(ip, pos, length, flags); |
746 | if (!length) { | |
747 | ret = -EINVAL; | |
748 | goto out; | |
749 | } | |
b3b94faa | 750 | |
3974320c BP |
751 | if ((flags & IOMAP_REPORT) && gfs2_is_stuffed(ip)) { |
752 | gfs2_stuffed_iomap(inode, iomap); | |
753 | if (pos >= iomap->length) | |
754 | return -ENOENT; | |
755 | ret = 0; | |
756 | goto out; | |
757 | } | |
758 | ||
759 | lblock = pos >> inode->i_blkbits; | |
760 | lend = (pos + length + sdp->sd_sb.sb_bsize - 1) >> inode->i_blkbits; | |
761 | ||
762 | iomap->offset = lblock << inode->i_blkbits; | |
763 | iomap->addr = IOMAP_NULL_ADDR; | |
764 | iomap->type = IOMAP_HOLE; | |
765 | iomap->length = (u64)(lend - lblock) << inode->i_blkbits; | |
766 | iomap->flags = IOMAP_F_MERGED; | |
767 | bmap_lock(ip, 0); | |
20cdc193 AG |
768 | |
769 | /* | |
770 | * Directory data blocks have a struct gfs2_meta_header header, so the | |
771 | * remaining size is smaller than the filesystem block size. Logical | |
772 | * block numbers for directories are in units of this remaining size! | |
773 | */ | |
ecc30c79 | 774 | if (gfs2_is_dir(ip)) { |
20cdc193 | 775 | factor = sdp->sd_jbsize; |
ecc30c79 SW |
776 | arr = sdp->sd_jheightsize; |
777 | } | |
4cf1ed81 | 778 | |
9b8c81d1 SW |
779 | ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]); |
780 | if (ret) | |
3974320c | 781 | goto out_release; |
b3b94faa | 782 | |
9b8c81d1 | 783 | height = ip->i_height; |
3974320c | 784 | while ((lblock + 1) * factor > arr[height]) |
9b8c81d1 SW |
785 | height++; |
786 | find_metapath(sdp, lblock, &mp, height); | |
9b8c81d1 SW |
787 | if (height > ip->i_height || gfs2_is_stuffed(ip)) |
788 | goto do_alloc; | |
3974320c | 789 | |
9b8c81d1 SW |
790 | ret = lookup_metapath(ip, &mp); |
791 | if (ret < 0) | |
3974320c BP |
792 | goto out_release; |
793 | ||
5f8bd444 | 794 | if (mp.mp_aheight != ip->i_height) |
9b8c81d1 | 795 | goto do_alloc; |
3974320c | 796 | |
9b8c81d1 SW |
797 | ptr = metapointer(ip->i_height - 1, &mp); |
798 | if (*ptr == 0) | |
799 | goto do_alloc; | |
3974320c BP |
800 | |
801 | iomap->type = IOMAP_MAPPED; | |
802 | iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits; | |
803 | ||
9b8c81d1 | 804 | bh = mp.mp_bh[ip->i_height - 1]; |
3974320c | 805 | len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, lend - lblock, &eob); |
9b8c81d1 | 806 | if (eob) |
3974320c BP |
807 | iomap->flags |= IOMAP_F_BOUNDARY; |
808 | iomap->length = (u64)len << inode->i_blkbits; | |
809 | ||
9b8c81d1 | 810 | ret = 0; |
3974320c BP |
811 | |
812 | out_release: | |
9b8c81d1 | 813 | release_metapath(&mp); |
3974320c BP |
814 | bmap_unlock(ip, 0); |
815 | out: | |
816 | trace_gfs2_iomap_end(ip, iomap, ret); | |
9b8c81d1 | 817 | return ret; |
30cbf189 | 818 | |
9b8c81d1 | 819 | do_alloc: |
3974320c BP |
820 | if (!(flags & IOMAP_WRITE)) { |
821 | if (pos >= i_size_read(inode)) { | |
822 | ret = -ENOENT; | |
823 | goto out_release; | |
824 | } | |
9b8c81d1 | 825 | ret = 0; |
3974320c BP |
826 | iomap->length = hole_size(inode, lblock, &mp); |
827 | goto out_release; | |
b3b94faa | 828 | } |
9b8c81d1 | 829 | |
3974320c BP |
830 | ret = gfs2_iomap_alloc(inode, iomap, flags, &mp); |
831 | goto out_release; | |
832 | } | |
833 | ||
834 | /** | |
835 | * gfs2_block_map - Map a block from an inode to a disk block | |
836 | * @inode: The inode | |
837 | * @lblock: The logical block number | |
838 | * @bh_map: The bh to be mapped | |
839 | * @create: True if its ok to alloc blocks to satify the request | |
840 | * | |
841 | * Sets buffer_mapped() if successful, sets buffer_boundary() if a | |
842 | * read of metadata will be required before the next block can be | |
843 | * mapped. Sets buffer_new() if new blocks were allocated. | |
844 | * | |
845 | * Returns: errno | |
846 | */ | |
847 | ||
848 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
849 | struct buffer_head *bh_map, int create) | |
850 | { | |
851 | struct gfs2_inode *ip = GFS2_I(inode); | |
852 | struct iomap iomap; | |
853 | int ret, flags = 0; | |
854 | ||
855 | clear_buffer_mapped(bh_map); | |
856 | clear_buffer_new(bh_map); | |
857 | clear_buffer_boundary(bh_map); | |
858 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
859 | ||
860 | if (create) | |
861 | flags |= IOMAP_WRITE; | |
5f8bd444 | 862 | if (buffer_zeronew(bh_map)) |
3974320c BP |
863 | flags |= IOMAP_ZERO; |
864 | ret = gfs2_iomap_begin(inode, (loff_t)lblock << inode->i_blkbits, | |
865 | bh_map->b_size, flags, &iomap); | |
866 | if (ret) { | |
867 | if (!create && ret == -ENOENT) { | |
868 | /* Return unmapped buffer beyond the end of file. */ | |
869 | ret = 0; | |
870 | } | |
871 | goto out; | |
872 | } | |
873 | ||
874 | if (iomap.length > bh_map->b_size) { | |
875 | iomap.length = bh_map->b_size; | |
876 | iomap.flags &= ~IOMAP_F_BOUNDARY; | |
5f8bd444 | 877 | } |
3974320c BP |
878 | if (iomap.addr != IOMAP_NULL_ADDR) |
879 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
880 | bh_map->b_size = iomap.length; | |
881 | if (iomap.flags & IOMAP_F_BOUNDARY) | |
882 | set_buffer_boundary(bh_map); | |
883 | if (iomap.flags & IOMAP_F_NEW) | |
884 | set_buffer_new(bh_map); | |
885 | ||
886 | out: | |
887 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
888 | return ret; | |
fd88de56 SW |
889 | } |
890 | ||
941e6d7d SW |
891 | /* |
892 | * Deprecated: do not use in new code | |
893 | */ | |
fd88de56 SW |
894 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
895 | { | |
23591256 | 896 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 897 | int ret; |
fd88de56 SW |
898 | int create = *new; |
899 | ||
900 | BUG_ON(!extlen); | |
901 | BUG_ON(!dblock); | |
902 | BUG_ON(!new); | |
903 | ||
47a9a527 | 904 | bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 905 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
906 | *extlen = bh.b_size >> inode->i_blkbits; |
907 | *dblock = bh.b_blocknr; | |
908 | if (buffer_new(&bh)) | |
909 | *new = 1; | |
910 | else | |
911 | *new = 0; | |
912 | return ret; | |
b3b94faa DT |
913 | } |
914 | ||
ba7f7290 SW |
915 | /** |
916 | * gfs2_block_truncate_page - Deal with zeroing out data for truncate | |
917 | * | |
918 | * This is partly borrowed from ext3. | |
919 | */ | |
ff8f33c8 | 920 | static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from) |
ba7f7290 SW |
921 | { |
922 | struct inode *inode = mapping->host; | |
923 | struct gfs2_inode *ip = GFS2_I(inode); | |
09cbfeaf KS |
924 | unsigned long index = from >> PAGE_SHIFT; |
925 | unsigned offset = from & (PAGE_SIZE-1); | |
ba7f7290 SW |
926 | unsigned blocksize, iblock, length, pos; |
927 | struct buffer_head *bh; | |
928 | struct page *page; | |
ba7f7290 SW |
929 | int err; |
930 | ||
220cca2a | 931 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ba7f7290 SW |
932 | if (!page) |
933 | return 0; | |
934 | ||
935 | blocksize = inode->i_sb->s_blocksize; | |
936 | length = blocksize - (offset & (blocksize - 1)); | |
09cbfeaf | 937 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
ba7f7290 SW |
938 | |
939 | if (!page_has_buffers(page)) | |
940 | create_empty_buffers(page, blocksize, 0); | |
941 | ||
942 | /* Find the buffer that contains "offset" */ | |
943 | bh = page_buffers(page); | |
944 | pos = blocksize; | |
945 | while (offset >= pos) { | |
946 | bh = bh->b_this_page; | |
947 | iblock++; | |
948 | pos += blocksize; | |
949 | } | |
950 | ||
951 | err = 0; | |
952 | ||
953 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 954 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
955 | /* unmapped? It's a hole - nothing to do */ |
956 | if (!buffer_mapped(bh)) | |
957 | goto unlock; | |
958 | } | |
959 | ||
960 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
961 | if (PageUptodate(page)) | |
962 | set_buffer_uptodate(bh); | |
963 | ||
964 | if (!buffer_uptodate(bh)) { | |
965 | err = -EIO; | |
dfec8a14 | 966 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
ba7f7290 SW |
967 | wait_on_buffer(bh); |
968 | /* Uhhuh. Read error. Complain and punt. */ | |
969 | if (!buffer_uptodate(bh)) | |
970 | goto unlock; | |
1875f2f3 | 971 | err = 0; |
ba7f7290 SW |
972 | } |
973 | ||
bf36a713 | 974 | if (!gfs2_is_writeback(ip)) |
350a9b0a | 975 | gfs2_trans_add_data(ip->i_gl, bh); |
ba7f7290 | 976 | |
eebd2aa3 | 977 | zero_user(page, offset, length); |
40bc9a27 | 978 | mark_buffer_dirty(bh); |
ba7f7290 SW |
979 | unlock: |
980 | unlock_page(page); | |
09cbfeaf | 981 | put_page(page); |
ba7f7290 SW |
982 | return err; |
983 | } | |
984 | ||
c62baf65 FF |
985 | #define GFS2_JTRUNC_REVOKES 8192 |
986 | ||
fa731fc4 SW |
987 | /** |
988 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
989 | * @inode: The inode being truncated | |
990 | * @oldsize: The original (larger) size | |
991 | * @newsize: The new smaller size | |
992 | * | |
993 | * With jdata files, we have to journal a revoke for each block which is | |
994 | * truncated. As a result, we need to split this into separate transactions | |
995 | * if the number of pages being truncated gets too large. | |
996 | */ | |
997 | ||
fa731fc4 SW |
998 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
999 | { | |
1000 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1001 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1002 | u64 chunk; | |
1003 | int error; | |
1004 | ||
1005 | while (oldsize != newsize) { | |
e7fdf004 AG |
1006 | struct gfs2_trans *tr; |
1007 | unsigned int offs; | |
1008 | ||
fa731fc4 SW |
1009 | chunk = oldsize - newsize; |
1010 | if (chunk > max_chunk) | |
1011 | chunk = max_chunk; | |
e7fdf004 AG |
1012 | |
1013 | offs = oldsize & ~PAGE_MASK; | |
1014 | if (offs && chunk > PAGE_SIZE) | |
1015 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1016 | ||
7caef267 | 1017 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1018 | oldsize -= chunk; |
e7fdf004 AG |
1019 | |
1020 | tr = current->journal_info; | |
1021 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1022 | continue; | |
1023 | ||
fa731fc4 SW |
1024 | gfs2_trans_end(sdp); |
1025 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1026 | if (error) | |
1027 | return error; | |
1028 | } | |
1029 | ||
1030 | return 0; | |
1031 | } | |
1032 | ||
8b5860a3 | 1033 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1034 | { |
ff8f33c8 SW |
1035 | struct gfs2_inode *ip = GFS2_I(inode); |
1036 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1037 | struct address_space *mapping = inode->i_mapping; | |
80990f40 | 1038 | struct buffer_head *dibh = NULL; |
b3b94faa | 1039 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1040 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1041 | int error; |
1042 | ||
fa731fc4 SW |
1043 | if (journaled) |
1044 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1045 | else | |
1046 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1047 | if (error) |
1048 | return error; | |
1049 | ||
1050 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1051 | if (error) | |
1052 | goto out; | |
1053 | ||
350a9b0a | 1054 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1055 | |
b3b94faa | 1056 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1057 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1058 | } else { |
ff8f33c8 SW |
1059 | if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) { |
1060 | error = gfs2_block_truncate_page(mapping, newsize); | |
1061 | if (error) | |
80990f40 | 1062 | goto out; |
b3b94faa | 1063 | } |
ff8f33c8 | 1064 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1065 | } |
1066 | ||
ff8f33c8 | 1067 | i_size_write(inode, newsize); |
078cd827 | 1068 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
ff8f33c8 | 1069 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1070 | |
fa731fc4 SW |
1071 | if (journaled) |
1072 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1073 | else | |
7caef267 | 1074 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1075 | |
a91ea69f | 1076 | out: |
80990f40 AG |
1077 | brelse(dibh); |
1078 | if (current->journal_info) | |
1079 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1080 | return error; |
1081 | } | |
1082 | ||
d552a2b9 BP |
1083 | /** |
1084 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1085 | * @ip: inode | |
1086 | * @rg_gh: holder of resource group glock | |
1087 | * @mp: current metapath fully populated with buffers | |
1088 | * @btotal: place to keep count of total blocks freed | |
1089 | * @hgt: height we're processing | |
1090 | * @first: true if this is the first call to this function for this height | |
1091 | * | |
1092 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1093 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1094 | * block has references to multiple rgrps, we break it into individual | |
1095 | * transactions. This allows other processes to use the rgrps while we're | |
1096 | * focused on a single one, for better concurrency / performance. | |
1097 | * At every transaction boundary, we rewrite the inode into the journal. | |
1098 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1099 | * if we're interrupted by power-outages. | |
1100 | * | |
1101 | * Returns: 0, or return code if an error occurred. | |
1102 | * *btotal has the total number of blocks freed | |
1103 | */ | |
1104 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
1105 | const struct metapath *mp, u32 *btotal, int hgt, | |
1106 | bool preserve1) | |
b3b94faa | 1107 | { |
9b8c81d1 | 1108 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1109 | struct gfs2_rgrpd *rgd; |
1110 | struct gfs2_trans *tr; | |
1111 | struct buffer_head *bh = mp->mp_bh[hgt]; | |
1112 | __be64 *top, *bottom, *p; | |
1113 | int blks_outside_rgrp; | |
1114 | u64 bn, bstart, isize_blks; | |
1115 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
1116 | int meta = ((hgt != ip->i_height - 1) ? 1 : 0); | |
1117 | int ret = 0; | |
1118 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1119 | ||
1120 | if (gfs2_metatype_check(sdp, bh, | |
1121 | (hgt ? GFS2_METATYPE_IN : GFS2_METATYPE_DI))) | |
1122 | return -EIO; | |
1123 | ||
1124 | more_rgrps: | |
1125 | blks_outside_rgrp = 0; | |
1126 | bstart = 0; | |
1127 | blen = 0; | |
1128 | top = metapointer(hgt, mp); /* first ptr from metapath */ | |
1129 | /* If we're keeping some data at the truncation point, we've got to | |
1130 | preserve the metadata tree by adding 1 to the starting metapath. */ | |
1131 | if (preserve1) | |
1132 | top++; | |
1133 | ||
1134 | bottom = (__be64 *)(bh->b_data + bh->b_size); | |
1135 | ||
1136 | for (p = top; p < bottom; p++) { | |
1137 | if (!*p) | |
1138 | continue; | |
1139 | bn = be64_to_cpu(*p); | |
1140 | if (gfs2_holder_initialized(rd_gh)) { | |
6f6597ba | 1141 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); |
d552a2b9 BP |
1142 | gfs2_assert_withdraw(sdp, |
1143 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1144 | } else { | |
90bcab99 | 1145 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
d552a2b9 BP |
1146 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1147 | 0, rd_gh); | |
1148 | if (ret) | |
1149 | goto out; | |
1150 | ||
1151 | /* Must be done with the rgrp glock held: */ | |
1152 | if (gfs2_rs_active(&ip->i_res) && | |
1153 | rgd == ip->i_res.rs_rbm.rgd) | |
1154 | gfs2_rs_deltree(&ip->i_res); | |
1155 | } | |
1156 | ||
1157 | if (!rgrp_contains_block(rgd, bn)) { | |
1158 | blks_outside_rgrp++; | |
1159 | continue; | |
1160 | } | |
1161 | ||
1162 | /* The size of our transactions will be unknown until we | |
1163 | actually process all the metadata blocks that relate to | |
1164 | the rgrp. So we estimate. We know it can't be more than | |
1165 | the dinode's i_blocks and we don't want to exceed the | |
1166 | journal flush threshold, sd_log_thresh2. */ | |
1167 | if (current->journal_info == NULL) { | |
1168 | unsigned int jblocks_rqsted, revokes; | |
1169 | ||
1170 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1171 | RES_INDIRECT; | |
1172 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1173 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1174 | jblocks_rqsted += | |
1175 | atomic_read(&sdp->sd_log_thresh2); | |
1176 | else | |
1177 | jblocks_rqsted += isize_blks; | |
1178 | revokes = jblocks_rqsted; | |
1179 | if (meta) | |
1180 | revokes += hptrs(sdp, hgt); | |
1181 | else if (ip->i_depth) | |
1182 | revokes += sdp->sd_inptrs; | |
1183 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1184 | if (ret) | |
1185 | goto out_unlock; | |
1186 | down_write(&ip->i_rw_mutex); | |
1187 | } | |
1188 | /* check if we will exceed the transaction blocks requested */ | |
1189 | tr = current->journal_info; | |
1190 | if (tr->tr_num_buf_new + RES_STATFS + | |
1191 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1192 | /* We set blks_outside_rgrp to ensure the loop will | |
1193 | be repeated for the same rgrp, but with a new | |
1194 | transaction. */ | |
1195 | blks_outside_rgrp++; | |
1196 | /* This next part is tricky. If the buffer was added | |
1197 | to the transaction, we've already set some block | |
1198 | pointers to 0, so we better follow through and free | |
1199 | them, or we will introduce corruption (so break). | |
1200 | This may be impossible, or at least rare, but I | |
1201 | decided to cover the case regardless. | |
1202 | ||
1203 | If the buffer was not added to the transaction | |
1204 | (this call), doing so would exceed our transaction | |
1205 | size, so we need to end the transaction and start a | |
1206 | new one (so goto). */ | |
1207 | ||
1208 | if (buf_in_tr) | |
1209 | break; | |
1210 | goto out_unlock; | |
1211 | } | |
1212 | ||
1213 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1214 | buf_in_tr = true; | |
1215 | *p = 0; | |
1216 | if (bstart + blen == bn) { | |
1217 | blen++; | |
1218 | continue; | |
1219 | } | |
1220 | if (bstart) { | |
1221 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1222 | (*btotal) += blen; | |
1223 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1224 | } | |
1225 | bstart = bn; | |
1226 | blen = 1; | |
1227 | } | |
1228 | if (bstart) { | |
1229 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1230 | (*btotal) += blen; | |
1231 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1232 | } | |
1233 | out_unlock: | |
1234 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1235 | outside the rgrp we just processed, | |
1236 | do it all over again. */ | |
1237 | if (current->journal_info) { | |
1238 | struct buffer_head *dibh = mp->mp_bh[0]; | |
1239 | ||
1240 | /* Every transaction boundary, we rewrite the dinode | |
1241 | to keep its di_blocks current in case of failure. */ | |
1242 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = | |
b32c8c76 | 1243 | current_time(&ip->i_inode); |
d552a2b9 BP |
1244 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1245 | gfs2_dinode_out(ip, dibh->b_data); | |
1246 | up_write(&ip->i_rw_mutex); | |
1247 | gfs2_trans_end(sdp); | |
1248 | } | |
1249 | gfs2_glock_dq_uninit(rd_gh); | |
1250 | cond_resched(); | |
1251 | goto more_rgrps; | |
1252 | } | |
1253 | out: | |
1254 | return ret; | |
1255 | } | |
1256 | ||
1257 | /** | |
1258 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
1259 | * assumes the metapath is valid (with buffers) out to height h | |
1260 | * @mp: starting metapath | |
1261 | * @h: desired height to search | |
1262 | * | |
1263 | * Returns: true if a non-null pointer was found in the metapath buffer | |
1264 | * false if all remaining pointers are NULL in the buffer | |
1265 | */ | |
1266 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
1267 | unsigned int h) | |
1268 | { | |
1269 | __be64 *ptr; | |
1270 | unsigned int ptrs = hptrs(sdp, h) - 1; | |
1271 | ||
1272 | while (true) { | |
1273 | ptr = metapointer(h, mp); | |
c4a9d189 BP |
1274 | if (*ptr) { /* if we have a non-null pointer */ |
1275 | /* Now zero the metapath after the current height. */ | |
1276 | h++; | |
1277 | if (h < GFS2_MAX_META_HEIGHT) | |
1278 | memset(&mp->mp_list[h], 0, | |
1279 | (GFS2_MAX_META_HEIGHT - h) * | |
1280 | sizeof(mp->mp_list[0])); | |
d552a2b9 | 1281 | return true; |
c4a9d189 | 1282 | } |
d552a2b9 BP |
1283 | |
1284 | if (mp->mp_list[h] < ptrs) | |
1285 | mp->mp_list[h]++; | |
1286 | else | |
1287 | return false; /* no more pointers in this buffer */ | |
1288 | } | |
1289 | } | |
1290 | ||
1291 | enum dealloc_states { | |
1292 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1293 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1294 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1295 | DEALLOC_DONE = 3, /* process complete */ | |
1296 | }; | |
b3b94faa | 1297 | |
c4a9d189 BP |
1298 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *nbof, unsigned int h) |
1299 | { | |
1300 | if (memcmp(mp->mp_list, nbof, h * sizeof(mp->mp_list[0]))) | |
1301 | return false; | |
1302 | return true; | |
1303 | } | |
1304 | ||
d552a2b9 BP |
1305 | /** |
1306 | * trunc_dealloc - truncate a file down to a desired size | |
1307 | * @ip: inode to truncate | |
1308 | * @newsize: The desired size of the file | |
1309 | * | |
1310 | * This function truncates a file to newsize. It works from the | |
1311 | * bottom up, and from the right to the left. In other words, it strips off | |
1312 | * the highest layer (data) before stripping any of the metadata. Doing it | |
1313 | * this way is best in case the operation is interrupted by power failure, etc. | |
1314 | * The dinode is rewritten in every transaction to guarantee integrity. | |
1315 | */ | |
1316 | static int trunc_dealloc(struct gfs2_inode *ip, u64 newsize) | |
1317 | { | |
1318 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1319 | struct metapath mp; | |
1320 | struct buffer_head *dibh, *bh; | |
1321 | struct gfs2_holder rd_gh; | |
1322 | u64 lblock; | |
1323 | __u16 nbof[GFS2_MAX_META_HEIGHT]; /* new beginning of truncation */ | |
1324 | unsigned int strip_h = ip->i_height - 1; | |
1325 | u32 btotal = 0; | |
1326 | int ret, state; | |
1327 | int mp_h; /* metapath buffers are read in to this height */ | |
1328 | sector_t last_ra = 0; | |
1329 | u64 prev_bnr = 0; | |
1330 | bool preserve1; /* need to preserve the first meta pointer? */ | |
1331 | ||
1332 | if (!newsize) | |
b3b94faa | 1333 | lblock = 0; |
18ec7d5c | 1334 | else |
d552a2b9 | 1335 | lblock = (newsize - 1) >> sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 1336 | |
d552a2b9 | 1337 | memset(&mp, 0, sizeof(mp)); |
9b8c81d1 | 1338 | find_metapath(sdp, lblock, &mp, ip->i_height); |
b3b94faa | 1339 | |
d552a2b9 BP |
1340 | memcpy(&nbof, &mp.mp_list, sizeof(nbof)); |
1341 | ||
1342 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1343 | if (ret) | |
1344 | return ret; | |
b3b94faa | 1345 | |
d552a2b9 BP |
1346 | mp.mp_bh[0] = dibh; |
1347 | ret = lookup_metapath(ip, &mp); | |
1348 | if (ret == ip->i_height) | |
1349 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ | |
1350 | else | |
1351 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1352 | |
d552a2b9 BP |
1353 | ret = gfs2_rindex_update(sdp); |
1354 | if (ret) | |
1355 | goto out_metapath; | |
1356 | ||
1357 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1358 | if (ret) | |
1359 | goto out_metapath; | |
1360 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1361 | ||
1362 | mp_h = strip_h; | |
1363 | ||
1364 | while (state != DEALLOC_DONE) { | |
1365 | switch (state) { | |
1366 | /* Truncate a full metapath at the given strip height. | |
1367 | * Note that strip_h == mp_h in order to be in this state. */ | |
1368 | case DEALLOC_MP_FULL: | |
1369 | if (mp_h > 0) { /* issue read-ahead on metadata */ | |
1370 | __be64 *top; | |
1371 | ||
1372 | bh = mp.mp_bh[mp_h - 1]; | |
1373 | if (bh->b_blocknr != last_ra) { | |
1374 | last_ra = bh->b_blocknr; | |
1375 | top = metaptr1(mp_h - 1, &mp); | |
1376 | gfs2_metapath_ra(ip->i_gl, bh, top); | |
1377 | } | |
1378 | } | |
1379 | /* If we're truncating to a non-zero size and the mp is | |
1380 | at the beginning of file for the strip height, we | |
1381 | need to preserve the first metadata pointer. */ | |
c4a9d189 | 1382 | preserve1 = (newsize && mp_eq_to_hgt(&mp, nbof, mp_h)); |
d552a2b9 BP |
1383 | bh = mp.mp_bh[mp_h]; |
1384 | gfs2_assert_withdraw(sdp, bh); | |
1385 | if (gfs2_assert_withdraw(sdp, | |
1386 | prev_bnr != bh->b_blocknr)) { | |
1387 | printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, " | |
1388 | "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n", | |
1389 | sdp->sd_fsname, | |
1390 | (unsigned long long)ip->i_no_addr, | |
1391 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1392 | } | |
1393 | prev_bnr = bh->b_blocknr; | |
1394 | ret = sweep_bh_for_rgrps(ip, &rd_gh, &mp, &btotal, | |
1395 | mp_h, preserve1); | |
1396 | /* If we hit an error or just swept dinode buffer, | |
1397 | just exit. */ | |
1398 | if (ret || !mp_h) { | |
1399 | state = DEALLOC_DONE; | |
1400 | break; | |
1401 | } | |
1402 | state = DEALLOC_MP_LOWER; | |
1403 | break; | |
1404 | ||
1405 | /* lower the metapath strip height */ | |
1406 | case DEALLOC_MP_LOWER: | |
1407 | /* We're done with the current buffer, so release it, | |
1408 | unless it's the dinode buffer. Then back up to the | |
1409 | previous pointer. */ | |
1410 | if (mp_h) { | |
1411 | brelse(mp.mp_bh[mp_h]); | |
1412 | mp.mp_bh[mp_h] = NULL; | |
1413 | } | |
1414 | /* If we can't get any lower in height, we've stripped | |
1415 | off all we can. Next step is to back up and start | |
1416 | stripping the previous level of metadata. */ | |
1417 | if (mp_h == 0) { | |
1418 | strip_h--; | |
1419 | memcpy(&mp.mp_list, &nbof, sizeof(nbof)); | |
1420 | mp_h = strip_h; | |
1421 | state = DEALLOC_FILL_MP; | |
1422 | break; | |
1423 | } | |
1424 | mp.mp_list[mp_h] = 0; | |
1425 | mp_h--; /* search one metadata height down */ | |
1426 | if (mp.mp_list[mp_h] >= hptrs(sdp, mp_h) - 1) | |
1427 | break; /* loop around in the same state */ | |
1428 | mp.mp_list[mp_h]++; | |
1429 | /* Here we've found a part of the metapath that is not | |
1430 | * allocated. We need to search at that height for the | |
1431 | * next non-null pointer. */ | |
1432 | if (find_nonnull_ptr(sdp, &mp, mp_h)) { | |
1433 | state = DEALLOC_FILL_MP; | |
1434 | mp_h++; | |
1435 | } | |
1436 | /* No more non-null pointers at this height. Back up | |
1437 | to the previous height and try again. */ | |
1438 | break; /* loop around in the same state */ | |
1439 | ||
1440 | /* Fill the metapath with buffers to the given height. */ | |
1441 | case DEALLOC_FILL_MP: | |
1442 | /* Fill the buffers out to the current height. */ | |
1443 | ret = fillup_metapath(ip, &mp, mp_h); | |
1444 | if (ret < 0) | |
1445 | goto out; | |
1446 | ||
1447 | /* If buffers found for the entire strip height */ | |
1448 | if ((ret == ip->i_height) && (mp_h == strip_h)) { | |
1449 | state = DEALLOC_MP_FULL; | |
1450 | break; | |
1451 | } | |
1452 | if (ret < ip->i_height) /* We have a partial height */ | |
1453 | mp_h = ret - 1; | |
1454 | ||
1455 | /* If we find a non-null block pointer, crawl a bit | |
1456 | higher up in the metapath and try again, otherwise | |
1457 | we need to look lower for a new starting point. */ | |
1458 | if (find_nonnull_ptr(sdp, &mp, mp_h)) | |
1459 | mp_h++; | |
1460 | else | |
1461 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1462 | break; |
d552a2b9 | 1463 | } |
b3b94faa DT |
1464 | } |
1465 | ||
d552a2b9 BP |
1466 | if (btotal) { |
1467 | if (current->journal_info == NULL) { | |
1468 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1469 | RES_QUOTA, 0); | |
1470 | if (ret) | |
1471 | goto out; | |
1472 | down_write(&ip->i_rw_mutex); | |
1473 | } | |
1474 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
1475 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
1476 | ip->i_inode.i_gid); | |
b32c8c76 | 1477 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
d552a2b9 BP |
1478 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1479 | gfs2_dinode_out(ip, dibh->b_data); | |
1480 | up_write(&ip->i_rw_mutex); | |
1481 | gfs2_trans_end(sdp); | |
1482 | } | |
b3b94faa | 1483 | |
d552a2b9 BP |
1484 | out: |
1485 | if (gfs2_holder_initialized(&rd_gh)) | |
1486 | gfs2_glock_dq_uninit(&rd_gh); | |
1487 | if (current->journal_info) { | |
1488 | up_write(&ip->i_rw_mutex); | |
1489 | gfs2_trans_end(sdp); | |
1490 | cond_resched(); | |
1491 | } | |
1492 | gfs2_quota_unhold(ip); | |
1493 | out_metapath: | |
1494 | release_metapath(&mp); | |
1495 | return ret; | |
b3b94faa DT |
1496 | } |
1497 | ||
1498 | static int trunc_end(struct gfs2_inode *ip) | |
1499 | { | |
feaa7bba | 1500 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1501 | struct buffer_head *dibh; |
1502 | int error; | |
1503 | ||
1504 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1505 | if (error) | |
1506 | return error; | |
1507 | ||
1508 | down_write(&ip->i_rw_mutex); | |
1509 | ||
1510 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1511 | if (error) | |
1512 | goto out; | |
1513 | ||
a2e0f799 | 1514 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1515 | ip->i_height = 0; |
ce276b06 | 1516 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 1517 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 1518 | gfs2_ordered_del_inode(ip); |
b3b94faa | 1519 | } |
078cd827 | 1520 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
383f01fb | 1521 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1522 | |
350a9b0a | 1523 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 1524 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1525 | brelse(dibh); |
1526 | ||
a91ea69f | 1527 | out: |
b3b94faa | 1528 | up_write(&ip->i_rw_mutex); |
b3b94faa | 1529 | gfs2_trans_end(sdp); |
b3b94faa DT |
1530 | return error; |
1531 | } | |
1532 | ||
1533 | /** | |
1534 | * do_shrink - make a file smaller | |
ff8f33c8 | 1535 | * @inode: the inode |
ff8f33c8 | 1536 | * @newsize: the size to make the file |
b3b94faa | 1537 | * |
ff8f33c8 SW |
1538 | * Called with an exclusive lock on @inode. The @size must |
1539 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
1540 | * |
1541 | * Returns: errno | |
1542 | */ | |
1543 | ||
8b5860a3 | 1544 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 1545 | { |
ff8f33c8 | 1546 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
1547 | int error; |
1548 | ||
8b5860a3 | 1549 | error = trunc_start(inode, newsize); |
b3b94faa DT |
1550 | if (error < 0) |
1551 | return error; | |
ff8f33c8 | 1552 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
1553 | return 0; |
1554 | ||
ff8f33c8 SW |
1555 | error = trunc_dealloc(ip, newsize); |
1556 | if (error == 0) | |
b3b94faa DT |
1557 | error = trunc_end(ip); |
1558 | ||
1559 | return error; | |
1560 | } | |
1561 | ||
ff8f33c8 | 1562 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 1563 | { |
ff8f33c8 SW |
1564 | int ret; |
1565 | ||
8b5860a3 | 1566 | ret = do_shrink(inode, inode->i_size); |
ff8f33c8 SW |
1567 | WARN_ON(ret != 0); |
1568 | } | |
1569 | ||
1570 | /** | |
1571 | * do_grow - Touch and update inode size | |
1572 | * @inode: The inode | |
1573 | * @size: The new size | |
1574 | * | |
1575 | * This function updates the timestamps on the inode and | |
1576 | * may also increase the size of the inode. This function | |
1577 | * must not be called with @size any smaller than the current | |
1578 | * inode size. | |
1579 | * | |
1580 | * Although it is not strictly required to unstuff files here, | |
1581 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
1582 | * code which will result in a buffer overrun if the size is larger | |
1583 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 1584 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
1585 | * just update the inode size directly. |
1586 | * | |
1587 | * Returns: 0 on success, or -ve on error | |
1588 | */ | |
1589 | ||
1590 | static int do_grow(struct inode *inode, u64 size) | |
1591 | { | |
1592 | struct gfs2_inode *ip = GFS2_I(inode); | |
1593 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 1594 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
1595 | struct buffer_head *dibh; |
1596 | int error; | |
2f7ee358 | 1597 | int unstuff = 0; |
a13b8c5f | 1598 | |
ff8f33c8 SW |
1599 | if (gfs2_is_stuffed(ip) && |
1600 | (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) { | |
b8fbf471 | 1601 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 1602 | if (error) |
5407e242 | 1603 | return error; |
ff8f33c8 | 1604 | |
7b9cff46 | 1605 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
1606 | if (error) |
1607 | goto do_grow_qunlock; | |
2f7ee358 | 1608 | unstuff = 1; |
ff8f33c8 SW |
1609 | } |
1610 | ||
a01aedfe BP |
1611 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
1612 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? | |
1613 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 1614 | if (error) |
ff8f33c8 | 1615 | goto do_grow_release; |
a13b8c5f | 1616 | |
2f7ee358 | 1617 | if (unstuff) { |
ff8f33c8 SW |
1618 | error = gfs2_unstuff_dinode(ip, NULL); |
1619 | if (error) | |
1620 | goto do_end_trans; | |
1621 | } | |
a13b8c5f WC |
1622 | |
1623 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1624 | if (error) | |
ff8f33c8 | 1625 | goto do_end_trans; |
a13b8c5f | 1626 | |
ff8f33c8 | 1627 | i_size_write(inode, size); |
078cd827 | 1628 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
350a9b0a | 1629 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
1630 | gfs2_dinode_out(ip, dibh->b_data); |
1631 | brelse(dibh); | |
1632 | ||
ff8f33c8 | 1633 | do_end_trans: |
a13b8c5f | 1634 | gfs2_trans_end(sdp); |
ff8f33c8 | 1635 | do_grow_release: |
2f7ee358 | 1636 | if (unstuff) { |
ff8f33c8 SW |
1637 | gfs2_inplace_release(ip); |
1638 | do_grow_qunlock: | |
1639 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 1640 | } |
a13b8c5f WC |
1641 | return error; |
1642 | } | |
1643 | ||
b3b94faa | 1644 | /** |
ff8f33c8 SW |
1645 | * gfs2_setattr_size - make a file a given size |
1646 | * @inode: the inode | |
1647 | * @newsize: the size to make the file | |
b3b94faa | 1648 | * |
ff8f33c8 SW |
1649 | * The file size can grow, shrink, or stay the same size. This |
1650 | * is called holding i_mutex and an exclusive glock on the inode | |
1651 | * in question. | |
b3b94faa DT |
1652 | * |
1653 | * Returns: errno | |
1654 | */ | |
1655 | ||
ff8f33c8 | 1656 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 1657 | { |
af5c2697 | 1658 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 1659 | int ret; |
b3b94faa | 1660 | |
ff8f33c8 | 1661 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 1662 | |
ff8f33c8 SW |
1663 | ret = inode_newsize_ok(inode, newsize); |
1664 | if (ret) | |
1665 | return ret; | |
b3b94faa | 1666 | |
562c72aa CH |
1667 | inode_dio_wait(inode); |
1668 | ||
b54e9a0b | 1669 | ret = gfs2_rsqa_alloc(ip); |
d2b47cfb | 1670 | if (ret) |
2b3dcf35 | 1671 | goto out; |
d2b47cfb | 1672 | |
8b5860a3 | 1673 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
1674 | ret = do_grow(inode, newsize); |
1675 | goto out; | |
1676 | } | |
ff8f33c8 | 1677 | |
8b5860a3 | 1678 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 1679 | out: |
a097dc7e | 1680 | gfs2_rsqa_delete(ip, NULL); |
2b3dcf35 | 1681 | return ret; |
b3b94faa DT |
1682 | } |
1683 | ||
1684 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1685 | { | |
1686 | int error; | |
a2e0f799 | 1687 | error = trunc_dealloc(ip, i_size_read(&ip->i_inode)); |
b3b94faa DT |
1688 | if (!error) |
1689 | error = trunc_end(ip); | |
1690 | return error; | |
1691 | } | |
1692 | ||
1693 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1694 | { | |
1695 | return trunc_dealloc(ip, 0); | |
1696 | } | |
1697 | ||
b50f227b SW |
1698 | /** |
1699 | * gfs2_free_journal_extents - Free cached journal bmap info | |
1700 | * @jd: The journal | |
1701 | * | |
1702 | */ | |
1703 | ||
1704 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
1705 | { | |
1706 | struct gfs2_journal_extent *jext; | |
1707 | ||
1708 | while(!list_empty(&jd->extent_list)) { | |
1709 | jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list); | |
1710 | list_del(&jext->list); | |
1711 | kfree(jext); | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | /** | |
1716 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
1717 | * @jd: The journal descriptor | |
1718 | * @lblock: The logical block at start of new extent | |
c62baf65 | 1719 | * @dblock: The physical block at start of new extent |
b50f227b SW |
1720 | * @blocks: Size of extent in fs blocks |
1721 | * | |
1722 | * Returns: 0 on success or -ENOMEM | |
1723 | */ | |
1724 | ||
1725 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
1726 | { | |
1727 | struct gfs2_journal_extent *jext; | |
1728 | ||
1729 | if (!list_empty(&jd->extent_list)) { | |
1730 | jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list); | |
1731 | if ((jext->dblock + jext->blocks) == dblock) { | |
1732 | jext->blocks += blocks; | |
1733 | return 0; | |
1734 | } | |
1735 | } | |
1736 | ||
1737 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
1738 | if (jext == NULL) | |
1739 | return -ENOMEM; | |
1740 | jext->dblock = dblock; | |
1741 | jext->lblock = lblock; | |
1742 | jext->blocks = blocks; | |
1743 | list_add_tail(&jext->list, &jd->extent_list); | |
1744 | jd->nr_extents++; | |
1745 | return 0; | |
1746 | } | |
1747 | ||
1748 | /** | |
1749 | * gfs2_map_journal_extents - Cache journal bmap info | |
1750 | * @sdp: The super block | |
1751 | * @jd: The journal to map | |
1752 | * | |
1753 | * Create a reusable "extent" mapping from all logical | |
1754 | * blocks to all physical blocks for the given journal. This will save | |
1755 | * us time when writing journal blocks. Most journals will have only one | |
1756 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
1757 | * arranges the journal blocks sequentially to maximize performance. | |
1758 | * So the extent would map the first block for the entire file length. | |
1759 | * However, gfs2_jadd can happen while file activity is happening, so | |
1760 | * those journals may not be sequential. Less likely is the case where | |
1761 | * the users created their own journals by mounting the metafs and | |
1762 | * laying it out. But it's still possible. These journals might have | |
1763 | * several extents. | |
1764 | * | |
1765 | * Returns: 0 on success, or error on failure | |
1766 | */ | |
1767 | ||
1768 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
1769 | { | |
1770 | u64 lblock = 0; | |
1771 | u64 lblock_stop; | |
1772 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
1773 | struct buffer_head bh; | |
1774 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
1775 | u64 size; | |
1776 | int rc; | |
1777 | ||
1778 | lblock_stop = i_size_read(jd->jd_inode) >> shift; | |
1779 | size = (lblock_stop - lblock) << shift; | |
1780 | jd->nr_extents = 0; | |
1781 | WARN_ON(!list_empty(&jd->extent_list)); | |
1782 | ||
1783 | do { | |
1784 | bh.b_state = 0; | |
1785 | bh.b_blocknr = 0; | |
1786 | bh.b_size = size; | |
1787 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
1788 | if (rc || !buffer_mapped(&bh)) | |
1789 | goto fail; | |
1790 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
1791 | if (rc) | |
1792 | goto fail; | |
1793 | size -= bh.b_size; | |
1794 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
1795 | } while(size > 0); | |
1796 | ||
1797 | fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid, | |
1798 | jd->nr_extents); | |
1799 | return 0; | |
1800 | ||
1801 | fail: | |
1802 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
1803 | rc, jd->jd_jid, | |
1804 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
1805 | jd->nr_extents); | |
1806 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
1807 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
1808 | bh.b_state, (unsigned long long)bh.b_size); | |
1809 | gfs2_free_journal_extents(jd); | |
1810 | return rc; | |
1811 | } | |
1812 | ||
b3b94faa DT |
1813 | /** |
1814 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
1815 | * @ip: the file being written to | |
1816 | * @offset: the offset to write to | |
1817 | * @len: the number of bytes being written | |
b3b94faa | 1818 | * |
461cb419 | 1819 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
1820 | */ |
1821 | ||
cd915493 | 1822 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 1823 | unsigned int len) |
b3b94faa | 1824 | { |
feaa7bba | 1825 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
1826 | struct buffer_head bh; |
1827 | unsigned int shift; | |
1828 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 1829 | u64 end_of_file; |
b3b94faa | 1830 | |
b3b94faa DT |
1831 | if (!len) |
1832 | return 0; | |
1833 | ||
1834 | if (gfs2_is_stuffed(ip)) { | |
1835 | if (offset + len > | |
1836 | sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
461cb419 | 1837 | return 1; |
b3b94faa DT |
1838 | return 0; |
1839 | } | |
1840 | ||
941e6d7d | 1841 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 1842 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 1843 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
1844 | lblock = offset >> shift; |
1845 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
1846 | if (lblock_stop > end_of_file) | |
461cb419 | 1847 | return 1; |
b3b94faa | 1848 | |
941e6d7d SW |
1849 | size = (lblock_stop - lblock) << shift; |
1850 | do { | |
1851 | bh.b_state = 0; | |
1852 | bh.b_size = size; | |
1853 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
1854 | if (!buffer_mapped(&bh)) | |
461cb419 | 1855 | return 1; |
941e6d7d SW |
1856 | size -= bh.b_size; |
1857 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
1858 | } while(size > 0); | |
b3b94faa DT |
1859 | |
1860 | return 0; | |
1861 | } | |
1862 |