]>
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 | ||
235628c5 AG |
72 | if (dsize > gfs2_max_stuffed_size(ip)) |
73 | dsize = gfs2_max_stuffed_size(ip); | |
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 | ||
5cf26b1e | 282 | static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end) |
b99b98dc | 283 | { |
b99b98dc SW |
284 | const __be64 *t; |
285 | ||
5cf26b1e | 286 | for (t = start; t < end; t++) { |
c3ce5aa9 AG |
287 | struct buffer_head *rabh; |
288 | ||
b99b98dc SW |
289 | if (!*t) |
290 | continue; | |
291 | ||
292 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
293 | if (trylock_buffer(rabh)) { | |
294 | if (!buffer_uptodate(rabh)) { | |
295 | rabh->b_end_io = end_buffer_read_sync; | |
e477b24b CL |
296 | submit_bh(REQ_OP_READ, |
297 | REQ_RAHEAD | REQ_META | REQ_PRIO, | |
298 | rabh); | |
b99b98dc SW |
299 | continue; |
300 | } | |
301 | unlock_buffer(rabh); | |
302 | } | |
303 | brelse(rabh); | |
304 | } | |
305 | } | |
306 | ||
e8b43fe0 AG |
307 | static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, |
308 | unsigned int x, unsigned int h) | |
d552a2b9 | 309 | { |
e8b43fe0 AG |
310 | for (; x < h; x++) { |
311 | __be64 *ptr = metapointer(x, mp); | |
312 | u64 dblock = be64_to_cpu(*ptr); | |
313 | int ret; | |
d552a2b9 | 314 | |
e8b43fe0 AG |
315 | if (!dblock) |
316 | break; | |
317 | ret = gfs2_meta_indirect_buffer(ip, x + 1, dblock, &mp->mp_bh[x + 1]); | |
318 | if (ret) | |
319 | return ret; | |
320 | } | |
321 | mp->mp_aheight = x + 1; | |
322 | return 0; | |
d552a2b9 BP |
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 | * |
e8b43fe0 | 339 | * Returns: error |
b3b94faa DT |
340 | */ |
341 | ||
9b8c81d1 | 342 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 343 | { |
e8b43fe0 | 344 | return __fillup_metapath(ip, mp, 0, ip->i_height - 1); |
dbac6710 SW |
345 | } |
346 | ||
d552a2b9 BP |
347 | /** |
348 | * fillup_metapath - fill up buffers for the metadata path to a specific height | |
349 | * @ip: The inode | |
350 | * @mp: The metapath | |
351 | * @h: The height to which it should be mapped | |
352 | * | |
353 | * Similar to lookup_metapath, but does lookups for a range of heights | |
354 | * | |
c3ce5aa9 | 355 | * Returns: error or the number of buffers filled |
d552a2b9 BP |
356 | */ |
357 | ||
358 | static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h) | |
359 | { | |
e8b43fe0 | 360 | unsigned int x = 0; |
c3ce5aa9 | 361 | int ret; |
d552a2b9 BP |
362 | |
363 | if (h) { | |
364 | /* find the first buffer we need to look up. */ | |
e8b43fe0 AG |
365 | for (x = h - 1; x > 0; x--) { |
366 | if (mp->mp_bh[x]) | |
367 | break; | |
d552a2b9 BP |
368 | } |
369 | } | |
c3ce5aa9 AG |
370 | ret = __fillup_metapath(ip, mp, x, h); |
371 | if (ret) | |
372 | return ret; | |
373 | return mp->mp_aheight - x - 1; | |
d552a2b9 BP |
374 | } |
375 | ||
9b8c81d1 | 376 | static inline void release_metapath(struct metapath *mp) |
dbac6710 SW |
377 | { |
378 | int i; | |
379 | ||
9b8c81d1 SW |
380 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
381 | if (mp->mp_bh[i] == NULL) | |
382 | break; | |
383 | brelse(mp->mp_bh[i]); | |
384 | } | |
11707ea0 SW |
385 | } |
386 | ||
30cbf189 SW |
387 | /** |
388 | * gfs2_extent_length - Returns length of an extent of blocks | |
389 | * @start: Start of the buffer | |
390 | * @len: Length of the buffer in bytes | |
391 | * @ptr: Current position in the buffer | |
392 | * @limit: Max extent length to return (0 = unlimited) | |
393 | * @eob: Set to 1 if we hit "end of block" | |
394 | * | |
395 | * If the first block is zero (unallocated) it will return the number of | |
396 | * unallocated blocks in the extent, otherwise it will return the number | |
397 | * of contiguous blocks in the extent. | |
398 | * | |
399 | * Returns: The length of the extent (minimum of one block) | |
400 | */ | |
401 | ||
b650738c | 402 | static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob) |
30cbf189 SW |
403 | { |
404 | const __be64 *end = (start + len); | |
405 | const __be64 *first = ptr; | |
406 | u64 d = be64_to_cpu(*ptr); | |
407 | ||
408 | *eob = 0; | |
409 | do { | |
410 | ptr++; | |
411 | if (ptr >= end) | |
412 | break; | |
413 | if (limit && --limit == 0) | |
414 | break; | |
415 | if (d) | |
416 | d++; | |
417 | } while(be64_to_cpu(*ptr) == d); | |
418 | if (ptr >= end) | |
419 | *eob = 1; | |
420 | return (ptr - first); | |
421 | } | |
422 | ||
9b8c81d1 | 423 | static inline void bmap_lock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 424 | { |
4cf1ed81 SW |
425 | if (create) |
426 | down_write(&ip->i_rw_mutex); | |
427 | else | |
428 | down_read(&ip->i_rw_mutex); | |
429 | } | |
430 | ||
9b8c81d1 | 431 | static inline void bmap_unlock(struct gfs2_inode *ip, int create) |
4cf1ed81 | 432 | { |
4cf1ed81 SW |
433 | if (create) |
434 | up_write(&ip->i_rw_mutex); | |
435 | else | |
436 | up_read(&ip->i_rw_mutex); | |
437 | } | |
438 | ||
9b8c81d1 SW |
439 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
440 | struct gfs2_glock *gl, unsigned int i, | |
441 | unsigned offset, u64 bn) | |
442 | { | |
443 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
444 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
445 | sizeof(struct gfs2_dinode))); | |
446 | BUG_ON(i < 1); | |
447 | BUG_ON(mp->mp_bh[i] != NULL); | |
448 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
350a9b0a | 449 | gfs2_trans_add_meta(gl, mp->mp_bh[i]); |
9b8c81d1 SW |
450 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
451 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
452 | ptr += offset; | |
453 | *ptr = cpu_to_be64(bn); | |
454 | return ptr; | |
455 | } | |
456 | ||
457 | enum alloc_state { | |
458 | ALLOC_DATA = 0, | |
459 | ALLOC_GROW_DEPTH = 1, | |
460 | ALLOC_GROW_HEIGHT = 2, | |
461 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
462 | }; | |
463 | ||
464 | /** | |
465 | * gfs2_bmap_alloc - Build a metadata tree of the requested height | |
466 | * @inode: The GFS2 inode | |
467 | * @lblock: The logical starting block of the extent | |
468 | * @bh_map: This is used to return the mapping details | |
5f8bd444 BP |
469 | * @zero_new: True if newly allocated blocks should be zeroed |
470 | * @mp: The metapath, with proper height information calculated | |
9b8c81d1 | 471 | * @maxlen: The max number of data blocks to alloc |
5f8bd444 BP |
472 | * @dblock: Pointer to return the resulting new block |
473 | * @dblks: Pointer to return the number of blocks allocated | |
9b8c81d1 SW |
474 | * |
475 | * In this routine we may have to alloc: | |
476 | * i) Indirect blocks to grow the metadata tree height | |
477 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
478 | * iii) Data blocks | |
479 | * | |
480 | * The function is in two parts. The first part works out the total | |
481 | * number of blocks which we need. The second part does the actual | |
482 | * allocation asking for an extent at a time (if enough contiguous free | |
483 | * blocks are available, there will only be one request per bmap call) | |
484 | * and uses the state machine to initialise the blocks in order. | |
485 | * | |
486 | * Returns: errno on error | |
487 | */ | |
488 | ||
3974320c BP |
489 | static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap, |
490 | unsigned flags, struct metapath *mp) | |
9b8c81d1 SW |
491 | { |
492 | struct gfs2_inode *ip = GFS2_I(inode); | |
493 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
64dd153c | 494 | struct super_block *sb = sdp->sd_vfs; |
9b8c81d1 | 495 | struct buffer_head *dibh = mp->mp_bh[0]; |
5f8bd444 | 496 | u64 bn; |
5af4e7a0 | 497 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
3974320c | 498 | unsigned dblks = 0; |
9b8c81d1 | 499 | unsigned ptrs_per_blk; |
5f8bd444 | 500 | const unsigned end_of_metadata = mp->mp_fheight - 1; |
64dd153c | 501 | int ret; |
9b8c81d1 SW |
502 | enum alloc_state state; |
503 | __be64 *ptr; | |
504 | __be64 zero_bn = 0; | |
3974320c | 505 | size_t maxlen = iomap->length >> inode->i_blkbits; |
9b8c81d1 | 506 | |
5f8bd444 | 507 | BUG_ON(mp->mp_aheight < 1); |
9b8c81d1 SW |
508 | BUG_ON(dibh == NULL); |
509 | ||
350a9b0a | 510 | gfs2_trans_add_meta(ip->i_gl, dibh); |
9b8c81d1 | 511 | |
5f8bd444 | 512 | if (mp->mp_fheight == mp->mp_aheight) { |
9b8c81d1 | 513 | struct buffer_head *bh; |
3974320c BP |
514 | int eob; |
515 | ||
9b8c81d1 SW |
516 | /* Bottom indirect block exists, find unalloced extent size */ |
517 | ptr = metapointer(end_of_metadata, mp); | |
518 | bh = mp->mp_bh[end_of_metadata]; | |
3974320c BP |
519 | dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, |
520 | maxlen, &eob); | |
521 | BUG_ON(dblks < 1); | |
9b8c81d1 SW |
522 | state = ALLOC_DATA; |
523 | } else { | |
524 | /* Need to allocate indirect blocks */ | |
5f8bd444 BP |
525 | ptrs_per_blk = mp->mp_fheight > 1 ? sdp->sd_inptrs : |
526 | sdp->sd_diptrs; | |
3974320c BP |
527 | dblks = min(maxlen, (size_t)(ptrs_per_blk - |
528 | mp->mp_list[end_of_metadata])); | |
5f8bd444 | 529 | if (mp->mp_fheight == ip->i_height) { |
9b8c81d1 | 530 | /* Writing into existing tree, extend tree down */ |
5f8bd444 | 531 | iblks = mp->mp_fheight - mp->mp_aheight; |
9b8c81d1 SW |
532 | state = ALLOC_GROW_DEPTH; |
533 | } else { | |
534 | /* Building up tree height */ | |
535 | state = ALLOC_GROW_HEIGHT; | |
5f8bd444 | 536 | iblks = mp->mp_fheight - ip->i_height; |
5af4e7a0 | 537 | branch_start = metapath_branch_start(mp); |
5f8bd444 | 538 | iblks += (mp->mp_fheight - branch_start); |
9b8c81d1 SW |
539 | } |
540 | } | |
541 | ||
542 | /* start of the second part of the function (state machine) */ | |
543 | ||
3974320c | 544 | blks = dblks + iblks; |
5f8bd444 | 545 | i = mp->mp_aheight; |
9b8c81d1 | 546 | do { |
09010978 | 547 | int error; |
9b8c81d1 | 548 | n = blks - alloced; |
6e87ed0f | 549 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
550 | if (error) |
551 | return error; | |
9b8c81d1 SW |
552 | alloced += n; |
553 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
554 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
555 | switch (state) { | |
556 | /* Growing height of tree */ | |
557 | case ALLOC_GROW_HEIGHT: | |
558 | if (i == 1) { | |
559 | ptr = (__be64 *)(dibh->b_data + | |
560 | sizeof(struct gfs2_dinode)); | |
561 | zero_bn = *ptr; | |
562 | } | |
5f8bd444 BP |
563 | for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0; |
564 | i++, n--) | |
9b8c81d1 | 565 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); |
5f8bd444 | 566 | if (i - 1 == mp->mp_fheight - ip->i_height) { |
9b8c81d1 SW |
567 | i--; |
568 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
569 | sizeof(struct gfs2_meta_header), | |
570 | dibh, sizeof(struct gfs2_dinode)); | |
571 | gfs2_buffer_clear_tail(dibh, | |
572 | sizeof(struct gfs2_dinode) + | |
573 | sizeof(__be64)); | |
574 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
575 | sizeof(struct gfs2_meta_header)); | |
576 | *ptr = zero_bn; | |
577 | state = ALLOC_GROW_DEPTH; | |
5f8bd444 | 578 | for(i = branch_start; i < mp->mp_fheight; i++) { |
9b8c81d1 SW |
579 | if (mp->mp_bh[i] == NULL) |
580 | break; | |
581 | brelse(mp->mp_bh[i]); | |
582 | mp->mp_bh[i] = NULL; | |
583 | } | |
5af4e7a0 | 584 | i = branch_start; |
9b8c81d1 SW |
585 | } |
586 | if (n == 0) | |
587 | break; | |
588 | /* Branching from existing tree */ | |
589 | case ALLOC_GROW_DEPTH: | |
5f8bd444 | 590 | if (i > 1 && i < mp->mp_fheight) |
350a9b0a | 591 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]); |
5f8bd444 | 592 | for (; i < mp->mp_fheight && n > 0; i++, n--) |
9b8c81d1 SW |
593 | gfs2_indirect_init(mp, ip->i_gl, i, |
594 | mp->mp_list[i-1], bn++); | |
5f8bd444 | 595 | if (i == mp->mp_fheight) |
9b8c81d1 SW |
596 | state = ALLOC_DATA; |
597 | if (n == 0) | |
598 | break; | |
599 | /* Tree complete, adding data blocks */ | |
600 | case ALLOC_DATA: | |
3974320c | 601 | BUG_ON(n > dblks); |
9b8c81d1 | 602 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); |
350a9b0a | 603 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]); |
3974320c | 604 | dblks = n; |
9b8c81d1 | 605 | ptr = metapointer(end_of_metadata, mp); |
3974320c BP |
606 | iomap->addr = bn << inode->i_blkbits; |
607 | iomap->flags |= IOMAP_F_NEW; | |
9b8c81d1 SW |
608 | while (n-- > 0) |
609 | *ptr++ = cpu_to_be64(bn++); | |
3974320c BP |
610 | if (flags & IOMAP_ZERO) { |
611 | ret = sb_issue_zeroout(sb, iomap->addr >> inode->i_blkbits, | |
612 | dblks, GFP_NOFS); | |
64dd153c BM |
613 | if (ret) { |
614 | fs_err(sdp, | |
615 | "Failed to zero data buffers\n"); | |
3974320c | 616 | flags &= ~IOMAP_ZERO; |
64dd153c BM |
617 | } |
618 | } | |
9b8c81d1 SW |
619 | break; |
620 | } | |
3974320c | 621 | } while (iomap->addr == IOMAP_NULL_ADDR); |
9b8c81d1 | 622 | |
3974320c | 623 | iomap->length = (u64)dblks << inode->i_blkbits; |
5f8bd444 | 624 | ip->i_height = mp->mp_fheight; |
9b8c81d1 SW |
625 | gfs2_add_inode_blocks(&ip->i_inode, alloced); |
626 | gfs2_dinode_out(ip, mp->mp_bh[0]->b_data); | |
9b8c81d1 SW |
627 | return 0; |
628 | } | |
629 | ||
b3b94faa | 630 | /** |
3974320c | 631 | * hole_size - figure out the size of a hole |
fd88de56 | 632 | * @inode: The inode |
3974320c BP |
633 | * @lblock: The logical starting block number |
634 | * @mp: The metapath | |
b3b94faa | 635 | * |
3974320c | 636 | * Returns: The hole size in bytes |
b3b94faa | 637 | * |
b3b94faa | 638 | */ |
3974320c BP |
639 | static u64 hole_size(struct inode *inode, sector_t lblock, struct metapath *mp) |
640 | { | |
641 | struct gfs2_inode *ip = GFS2_I(inode); | |
642 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
643 | struct metapath mp_eof; | |
644 | u64 factor = 1; | |
645 | int hgt; | |
646 | u64 holesz = 0; | |
647 | const __be64 *first, *end, *ptr; | |
648 | const struct buffer_head *bh; | |
649 | u64 lblock_stop = (i_size_read(inode) - 1) >> inode->i_blkbits; | |
650 | int zeroptrs; | |
651 | bool done = false; | |
652 | ||
653 | /* Get another metapath, to the very last byte */ | |
654 | find_metapath(sdp, lblock_stop, &mp_eof, ip->i_height); | |
655 | for (hgt = ip->i_height - 1; hgt >= 0 && !done; hgt--) { | |
656 | bh = mp->mp_bh[hgt]; | |
657 | if (bh) { | |
658 | zeroptrs = 0; | |
659 | first = metapointer(hgt, mp); | |
660 | end = (const __be64 *)(bh->b_data + bh->b_size); | |
661 | ||
662 | for (ptr = first; ptr < end; ptr++) { | |
663 | if (*ptr) { | |
664 | done = true; | |
665 | break; | |
666 | } else { | |
667 | zeroptrs++; | |
668 | } | |
669 | } | |
670 | } else { | |
671 | zeroptrs = sdp->sd_inptrs; | |
672 | } | |
673 | if (factor * zeroptrs >= lblock_stop - lblock + 1) { | |
674 | holesz = lblock_stop - lblock + 1; | |
675 | break; | |
676 | } | |
677 | holesz += factor * zeroptrs; | |
b3b94faa | 678 | |
3974320c BP |
679 | factor *= sdp->sd_inptrs; |
680 | if (hgt && (mp->mp_list[hgt - 1] < mp_eof.mp_list[hgt - 1])) | |
681 | (mp->mp_list[hgt - 1])++; | |
682 | } | |
683 | return holesz << inode->i_blkbits; | |
684 | } | |
685 | ||
686 | static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap) | |
687 | { | |
688 | struct gfs2_inode *ip = GFS2_I(inode); | |
689 | ||
690 | iomap->addr = (ip->i_no_addr << inode->i_blkbits) + | |
691 | sizeof(struct gfs2_dinode); | |
692 | iomap->offset = 0; | |
693 | iomap->length = i_size_read(inode); | |
694 | iomap->type = IOMAP_MAPPED; | |
695 | iomap->flags = IOMAP_F_DATA_INLINE; | |
696 | } | |
697 | ||
698 | /** | |
699 | * gfs2_iomap_begin - Map blocks from an inode to disk blocks | |
700 | * @inode: The inode | |
701 | * @pos: Starting position in bytes | |
702 | * @length: Length to map, in bytes | |
703 | * @flags: iomap flags | |
704 | * @iomap: The iomap structure | |
705 | * | |
706 | * Returns: errno | |
707 | */ | |
708 | int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, | |
709 | unsigned flags, struct iomap *iomap) | |
b3b94faa | 710 | { |
feaa7bba SW |
711 | struct gfs2_inode *ip = GFS2_I(inode); |
712 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
3974320c | 713 | struct metapath mp = { .mp_aheight = 1, }; |
20cdc193 | 714 | unsigned int factor = sdp->sd_sb.sb_bsize; |
ecc30c79 | 715 | const u64 *arr = sdp->sd_heightsize; |
9b8c81d1 | 716 | __be64 *ptr; |
3974320c BP |
717 | sector_t lblock; |
718 | sector_t lend; | |
49edd5bf | 719 | int ret = 0; |
9b8c81d1 SW |
720 | int eob; |
721 | unsigned int len; | |
722 | struct buffer_head *bh; | |
723 | u8 height; | |
7276b3b0 | 724 | |
3974320c BP |
725 | trace_gfs2_iomap_start(ip, pos, length, flags); |
726 | if (!length) { | |
727 | ret = -EINVAL; | |
728 | goto out; | |
729 | } | |
b3b94faa | 730 | |
49edd5bf AG |
731 | if (gfs2_is_stuffed(ip)) { |
732 | if (flags & IOMAP_REPORT) { | |
733 | gfs2_stuffed_iomap(inode, iomap); | |
734 | if (pos >= iomap->length) | |
735 | ret = -ENOENT; | |
736 | goto out; | |
737 | } | |
738 | BUG_ON(!(flags & IOMAP_WRITE)); | |
3974320c BP |
739 | } |
740 | ||
741 | lblock = pos >> inode->i_blkbits; | |
742 | lend = (pos + length + sdp->sd_sb.sb_bsize - 1) >> inode->i_blkbits; | |
743 | ||
744 | iomap->offset = lblock << inode->i_blkbits; | |
745 | iomap->addr = IOMAP_NULL_ADDR; | |
746 | iomap->type = IOMAP_HOLE; | |
747 | iomap->length = (u64)(lend - lblock) << inode->i_blkbits; | |
748 | iomap->flags = IOMAP_F_MERGED; | |
49edd5bf | 749 | bmap_lock(ip, flags & IOMAP_WRITE); |
20cdc193 AG |
750 | |
751 | /* | |
752 | * Directory data blocks have a struct gfs2_meta_header header, so the | |
753 | * remaining size is smaller than the filesystem block size. Logical | |
754 | * block numbers for directories are in units of this remaining size! | |
755 | */ | |
ecc30c79 | 756 | if (gfs2_is_dir(ip)) { |
20cdc193 | 757 | factor = sdp->sd_jbsize; |
ecc30c79 SW |
758 | arr = sdp->sd_jheightsize; |
759 | } | |
4cf1ed81 | 760 | |
9b8c81d1 SW |
761 | ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]); |
762 | if (ret) | |
3974320c | 763 | goto out_release; |
b3b94faa | 764 | |
9b8c81d1 | 765 | height = ip->i_height; |
3974320c | 766 | while ((lblock + 1) * factor > arr[height]) |
9b8c81d1 SW |
767 | height++; |
768 | find_metapath(sdp, lblock, &mp, height); | |
9b8c81d1 SW |
769 | if (height > ip->i_height || gfs2_is_stuffed(ip)) |
770 | goto do_alloc; | |
3974320c | 771 | |
9b8c81d1 | 772 | ret = lookup_metapath(ip, &mp); |
e8b43fe0 | 773 | if (ret) |
3974320c BP |
774 | goto out_release; |
775 | ||
5f8bd444 | 776 | if (mp.mp_aheight != ip->i_height) |
9b8c81d1 | 777 | goto do_alloc; |
3974320c | 778 | |
9b8c81d1 SW |
779 | ptr = metapointer(ip->i_height - 1, &mp); |
780 | if (*ptr == 0) | |
781 | goto do_alloc; | |
3974320c BP |
782 | |
783 | iomap->type = IOMAP_MAPPED; | |
784 | iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits; | |
785 | ||
9b8c81d1 | 786 | bh = mp.mp_bh[ip->i_height - 1]; |
3974320c | 787 | len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, lend - lblock, &eob); |
9b8c81d1 | 788 | if (eob) |
3974320c BP |
789 | iomap->flags |= IOMAP_F_BOUNDARY; |
790 | iomap->length = (u64)len << inode->i_blkbits; | |
791 | ||
3974320c | 792 | out_release: |
9b8c81d1 | 793 | release_metapath(&mp); |
49edd5bf | 794 | bmap_unlock(ip, flags & IOMAP_WRITE); |
3974320c BP |
795 | out: |
796 | trace_gfs2_iomap_end(ip, iomap, ret); | |
9b8c81d1 | 797 | return ret; |
30cbf189 | 798 | |
9b8c81d1 | 799 | do_alloc: |
49edd5bf AG |
800 | if (flags & IOMAP_WRITE) { |
801 | ret = gfs2_iomap_alloc(inode, iomap, flags, &mp); | |
802 | } else if (flags & IOMAP_REPORT) { | |
803 | loff_t size = i_size_read(inode); | |
804 | if (pos >= size) | |
3974320c | 805 | ret = -ENOENT; |
49edd5bf AG |
806 | else if (height <= ip->i_height) |
807 | iomap->length = hole_size(inode, lblock, &mp); | |
808 | else | |
809 | iomap->length = size - pos; | |
b3b94faa | 810 | } |
3974320c BP |
811 | goto out_release; |
812 | } | |
813 | ||
814 | /** | |
d39d18e0 | 815 | * gfs2_block_map - Map one or more blocks of an inode to a disk block |
3974320c BP |
816 | * @inode: The inode |
817 | * @lblock: The logical block number | |
818 | * @bh_map: The bh to be mapped | |
819 | * @create: True if its ok to alloc blocks to satify the request | |
820 | * | |
d39d18e0 AG |
821 | * The size of the requested mapping is defined in bh_map->b_size. |
822 | * | |
823 | * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged | |
824 | * when @lblock is not mapped. Sets buffer_mapped(bh_map) and | |
825 | * bh_map->b_size to indicate the size of the mapping when @lblock and | |
826 | * successive blocks are mapped, up to the requested size. | |
827 | * | |
828 | * Sets buffer_boundary() if a read of metadata will be required | |
829 | * before the next block can be mapped. Sets buffer_new() if new | |
830 | * blocks were allocated. | |
3974320c BP |
831 | * |
832 | * Returns: errno | |
833 | */ | |
834 | ||
835 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
836 | struct buffer_head *bh_map, int create) | |
837 | { | |
838 | struct gfs2_inode *ip = GFS2_I(inode); | |
839 | struct iomap iomap; | |
840 | int ret, flags = 0; | |
841 | ||
842 | clear_buffer_mapped(bh_map); | |
843 | clear_buffer_new(bh_map); | |
844 | clear_buffer_boundary(bh_map); | |
845 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
846 | ||
847 | if (create) | |
848 | flags |= IOMAP_WRITE; | |
5f8bd444 | 849 | if (buffer_zeronew(bh_map)) |
3974320c BP |
850 | flags |= IOMAP_ZERO; |
851 | ret = gfs2_iomap_begin(inode, (loff_t)lblock << inode->i_blkbits, | |
852 | bh_map->b_size, flags, &iomap); | |
853 | if (ret) { | |
854 | if (!create && ret == -ENOENT) { | |
855 | /* Return unmapped buffer beyond the end of file. */ | |
856 | ret = 0; | |
857 | } | |
858 | goto out; | |
859 | } | |
860 | ||
861 | if (iomap.length > bh_map->b_size) { | |
862 | iomap.length = bh_map->b_size; | |
863 | iomap.flags &= ~IOMAP_F_BOUNDARY; | |
5f8bd444 | 864 | } |
3974320c BP |
865 | if (iomap.addr != IOMAP_NULL_ADDR) |
866 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
867 | bh_map->b_size = iomap.length; | |
868 | if (iomap.flags & IOMAP_F_BOUNDARY) | |
869 | set_buffer_boundary(bh_map); | |
870 | if (iomap.flags & IOMAP_F_NEW) | |
871 | set_buffer_new(bh_map); | |
872 | ||
873 | out: | |
874 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
875 | return ret; | |
fd88de56 SW |
876 | } |
877 | ||
941e6d7d SW |
878 | /* |
879 | * Deprecated: do not use in new code | |
880 | */ | |
fd88de56 SW |
881 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
882 | { | |
23591256 | 883 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 884 | int ret; |
fd88de56 SW |
885 | int create = *new; |
886 | ||
887 | BUG_ON(!extlen); | |
888 | BUG_ON(!dblock); | |
889 | BUG_ON(!new); | |
890 | ||
47a9a527 | 891 | bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 892 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
893 | *extlen = bh.b_size >> inode->i_blkbits; |
894 | *dblock = bh.b_blocknr; | |
895 | if (buffer_new(&bh)) | |
896 | *new = 1; | |
897 | else | |
898 | *new = 0; | |
899 | return ret; | |
b3b94faa DT |
900 | } |
901 | ||
ba7f7290 | 902 | /** |
bdba0d5e | 903 | * gfs2_block_zero_range - Deal with zeroing out data |
ba7f7290 SW |
904 | * |
905 | * This is partly borrowed from ext3. | |
906 | */ | |
bdba0d5e AG |
907 | static int gfs2_block_zero_range(struct inode *inode, loff_t from, |
908 | unsigned int length) | |
ba7f7290 | 909 | { |
bdba0d5e | 910 | struct address_space *mapping = inode->i_mapping; |
ba7f7290 | 911 | struct gfs2_inode *ip = GFS2_I(inode); |
09cbfeaf KS |
912 | unsigned long index = from >> PAGE_SHIFT; |
913 | unsigned offset = from & (PAGE_SIZE-1); | |
bdba0d5e | 914 | unsigned blocksize, iblock, pos; |
ba7f7290 SW |
915 | struct buffer_head *bh; |
916 | struct page *page; | |
ba7f7290 SW |
917 | int err; |
918 | ||
220cca2a | 919 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ba7f7290 SW |
920 | if (!page) |
921 | return 0; | |
922 | ||
923 | blocksize = inode->i_sb->s_blocksize; | |
09cbfeaf | 924 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
ba7f7290 SW |
925 | |
926 | if (!page_has_buffers(page)) | |
927 | create_empty_buffers(page, blocksize, 0); | |
928 | ||
929 | /* Find the buffer that contains "offset" */ | |
930 | bh = page_buffers(page); | |
931 | pos = blocksize; | |
932 | while (offset >= pos) { | |
933 | bh = bh->b_this_page; | |
934 | iblock++; | |
935 | pos += blocksize; | |
936 | } | |
937 | ||
938 | err = 0; | |
939 | ||
940 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 941 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
942 | /* unmapped? It's a hole - nothing to do */ |
943 | if (!buffer_mapped(bh)) | |
944 | goto unlock; | |
945 | } | |
946 | ||
947 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
948 | if (PageUptodate(page)) | |
949 | set_buffer_uptodate(bh); | |
950 | ||
951 | if (!buffer_uptodate(bh)) { | |
952 | err = -EIO; | |
dfec8a14 | 953 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
ba7f7290 SW |
954 | wait_on_buffer(bh); |
955 | /* Uhhuh. Read error. Complain and punt. */ | |
956 | if (!buffer_uptodate(bh)) | |
957 | goto unlock; | |
1875f2f3 | 958 | err = 0; |
ba7f7290 SW |
959 | } |
960 | ||
bf36a713 | 961 | if (!gfs2_is_writeback(ip)) |
350a9b0a | 962 | gfs2_trans_add_data(ip->i_gl, bh); |
ba7f7290 | 963 | |
eebd2aa3 | 964 | zero_user(page, offset, length); |
40bc9a27 | 965 | mark_buffer_dirty(bh); |
ba7f7290 SW |
966 | unlock: |
967 | unlock_page(page); | |
09cbfeaf | 968 | put_page(page); |
ba7f7290 SW |
969 | return err; |
970 | } | |
971 | ||
c62baf65 FF |
972 | #define GFS2_JTRUNC_REVOKES 8192 |
973 | ||
fa731fc4 SW |
974 | /** |
975 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
976 | * @inode: The inode being truncated | |
977 | * @oldsize: The original (larger) size | |
978 | * @newsize: The new smaller size | |
979 | * | |
980 | * With jdata files, we have to journal a revoke for each block which is | |
981 | * truncated. As a result, we need to split this into separate transactions | |
982 | * if the number of pages being truncated gets too large. | |
983 | */ | |
984 | ||
fa731fc4 SW |
985 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
986 | { | |
987 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
988 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
989 | u64 chunk; | |
990 | int error; | |
991 | ||
992 | while (oldsize != newsize) { | |
e7fdf004 AG |
993 | struct gfs2_trans *tr; |
994 | unsigned int offs; | |
995 | ||
fa731fc4 SW |
996 | chunk = oldsize - newsize; |
997 | if (chunk > max_chunk) | |
998 | chunk = max_chunk; | |
e7fdf004 AG |
999 | |
1000 | offs = oldsize & ~PAGE_MASK; | |
1001 | if (offs && chunk > PAGE_SIZE) | |
1002 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1003 | ||
7caef267 | 1004 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1005 | oldsize -= chunk; |
e7fdf004 AG |
1006 | |
1007 | tr = current->journal_info; | |
1008 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1009 | continue; | |
1010 | ||
fa731fc4 SW |
1011 | gfs2_trans_end(sdp); |
1012 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1013 | if (error) | |
1014 | return error; | |
1015 | } | |
1016 | ||
1017 | return 0; | |
1018 | } | |
1019 | ||
8b5860a3 | 1020 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1021 | { |
ff8f33c8 SW |
1022 | struct gfs2_inode *ip = GFS2_I(inode); |
1023 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
80990f40 | 1024 | struct buffer_head *dibh = NULL; |
b3b94faa | 1025 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1026 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1027 | int error; |
1028 | ||
fa731fc4 SW |
1029 | if (journaled) |
1030 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1031 | else | |
1032 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1033 | if (error) |
1034 | return error; | |
1035 | ||
1036 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1037 | if (error) | |
1038 | goto out; | |
1039 | ||
350a9b0a | 1040 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1041 | |
b3b94faa | 1042 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1043 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1044 | } else { |
bdba0d5e AG |
1045 | unsigned int blocksize = i_blocksize(inode); |
1046 | unsigned int offs = newsize & (blocksize - 1); | |
1047 | if (offs) { | |
1048 | error = gfs2_block_zero_range(inode, newsize, | |
1049 | blocksize - offs); | |
ff8f33c8 | 1050 | if (error) |
80990f40 | 1051 | goto out; |
b3b94faa | 1052 | } |
ff8f33c8 | 1053 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1054 | } |
1055 | ||
ff8f33c8 | 1056 | i_size_write(inode, newsize); |
078cd827 | 1057 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
ff8f33c8 | 1058 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1059 | |
fa731fc4 SW |
1060 | if (journaled) |
1061 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1062 | else | |
7caef267 | 1063 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1064 | |
a91ea69f | 1065 | out: |
80990f40 AG |
1066 | brelse(dibh); |
1067 | if (current->journal_info) | |
1068 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1069 | return error; |
1070 | } | |
1071 | ||
d552a2b9 BP |
1072 | /** |
1073 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1074 | * @ip: inode | |
1075 | * @rg_gh: holder of resource group glock | |
5cf26b1e AG |
1076 | * @bh: buffer head to sweep |
1077 | * @start: starting point in bh | |
1078 | * @end: end point in bh | |
1079 | * @meta: true if bh points to metadata (rather than data) | |
d552a2b9 | 1080 | * @btotal: place to keep count of total blocks freed |
d552a2b9 BP |
1081 | * |
1082 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1083 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1084 | * block has references to multiple rgrps, we break it into individual | |
1085 | * transactions. This allows other processes to use the rgrps while we're | |
1086 | * focused on a single one, for better concurrency / performance. | |
1087 | * At every transaction boundary, we rewrite the inode into the journal. | |
1088 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1089 | * if we're interrupted by power-outages. | |
1090 | * | |
1091 | * Returns: 0, or return code if an error occurred. | |
1092 | * *btotal has the total number of blocks freed | |
1093 | */ | |
1094 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
5cf26b1e AG |
1095 | struct buffer_head *bh, __be64 *start, __be64 *end, |
1096 | bool meta, u32 *btotal) | |
b3b94faa | 1097 | { |
9b8c81d1 | 1098 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1099 | struct gfs2_rgrpd *rgd; |
1100 | struct gfs2_trans *tr; | |
5cf26b1e | 1101 | __be64 *p; |
d552a2b9 BP |
1102 | int blks_outside_rgrp; |
1103 | u64 bn, bstart, isize_blks; | |
1104 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
d552a2b9 BP |
1105 | int ret = 0; |
1106 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1107 | ||
d552a2b9 | 1108 | more_rgrps: |
5cf26b1e AG |
1109 | rgd = NULL; |
1110 | if (gfs2_holder_initialized(rd_gh)) { | |
1111 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); | |
1112 | gfs2_assert_withdraw(sdp, | |
1113 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1114 | } | |
d552a2b9 BP |
1115 | blks_outside_rgrp = 0; |
1116 | bstart = 0; | |
1117 | blen = 0; | |
d552a2b9 | 1118 | |
5cf26b1e | 1119 | for (p = start; p < end; p++) { |
d552a2b9 BP |
1120 | if (!*p) |
1121 | continue; | |
1122 | bn = be64_to_cpu(*p); | |
5cf26b1e AG |
1123 | |
1124 | if (rgd) { | |
1125 | if (!rgrp_contains_block(rgd, bn)) { | |
1126 | blks_outside_rgrp++; | |
1127 | continue; | |
1128 | } | |
d552a2b9 | 1129 | } else { |
90bcab99 | 1130 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
5cf26b1e AG |
1131 | if (unlikely(!rgd)) { |
1132 | ret = -EIO; | |
1133 | goto out; | |
1134 | } | |
d552a2b9 BP |
1135 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1136 | 0, rd_gh); | |
1137 | if (ret) | |
1138 | goto out; | |
1139 | ||
1140 | /* Must be done with the rgrp glock held: */ | |
1141 | if (gfs2_rs_active(&ip->i_res) && | |
1142 | rgd == ip->i_res.rs_rbm.rgd) | |
1143 | gfs2_rs_deltree(&ip->i_res); | |
1144 | } | |
1145 | ||
d552a2b9 BP |
1146 | /* The size of our transactions will be unknown until we |
1147 | actually process all the metadata blocks that relate to | |
1148 | the rgrp. So we estimate. We know it can't be more than | |
1149 | the dinode's i_blocks and we don't want to exceed the | |
1150 | journal flush threshold, sd_log_thresh2. */ | |
1151 | if (current->journal_info == NULL) { | |
1152 | unsigned int jblocks_rqsted, revokes; | |
1153 | ||
1154 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1155 | RES_INDIRECT; | |
1156 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1157 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1158 | jblocks_rqsted += | |
1159 | atomic_read(&sdp->sd_log_thresh2); | |
1160 | else | |
1161 | jblocks_rqsted += isize_blks; | |
1162 | revokes = jblocks_rqsted; | |
1163 | if (meta) | |
5cf26b1e | 1164 | revokes += end - start; |
d552a2b9 BP |
1165 | else if (ip->i_depth) |
1166 | revokes += sdp->sd_inptrs; | |
1167 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1168 | if (ret) | |
1169 | goto out_unlock; | |
1170 | down_write(&ip->i_rw_mutex); | |
1171 | } | |
1172 | /* check if we will exceed the transaction blocks requested */ | |
1173 | tr = current->journal_info; | |
1174 | if (tr->tr_num_buf_new + RES_STATFS + | |
1175 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1176 | /* We set blks_outside_rgrp to ensure the loop will | |
1177 | be repeated for the same rgrp, but with a new | |
1178 | transaction. */ | |
1179 | blks_outside_rgrp++; | |
1180 | /* This next part is tricky. If the buffer was added | |
1181 | to the transaction, we've already set some block | |
1182 | pointers to 0, so we better follow through and free | |
1183 | them, or we will introduce corruption (so break). | |
1184 | This may be impossible, or at least rare, but I | |
1185 | decided to cover the case regardless. | |
1186 | ||
1187 | If the buffer was not added to the transaction | |
1188 | (this call), doing so would exceed our transaction | |
1189 | size, so we need to end the transaction and start a | |
1190 | new one (so goto). */ | |
1191 | ||
1192 | if (buf_in_tr) | |
1193 | break; | |
1194 | goto out_unlock; | |
1195 | } | |
1196 | ||
1197 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1198 | buf_in_tr = true; | |
1199 | *p = 0; | |
1200 | if (bstart + blen == bn) { | |
1201 | blen++; | |
1202 | continue; | |
1203 | } | |
1204 | if (bstart) { | |
1205 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1206 | (*btotal) += blen; | |
1207 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1208 | } | |
1209 | bstart = bn; | |
1210 | blen = 1; | |
1211 | } | |
1212 | if (bstart) { | |
1213 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1214 | (*btotal) += blen; | |
1215 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1216 | } | |
1217 | out_unlock: | |
1218 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1219 | outside the rgrp we just processed, | |
1220 | do it all over again. */ | |
1221 | if (current->journal_info) { | |
5cf26b1e AG |
1222 | struct buffer_head *dibh; |
1223 | ||
1224 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1225 | if (ret) | |
1226 | goto out; | |
d552a2b9 BP |
1227 | |
1228 | /* Every transaction boundary, we rewrite the dinode | |
1229 | to keep its di_blocks current in case of failure. */ | |
1230 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = | |
b32c8c76 | 1231 | current_time(&ip->i_inode); |
d552a2b9 BP |
1232 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1233 | gfs2_dinode_out(ip, dibh->b_data); | |
5cf26b1e | 1234 | brelse(dibh); |
d552a2b9 BP |
1235 | up_write(&ip->i_rw_mutex); |
1236 | gfs2_trans_end(sdp); | |
1237 | } | |
1238 | gfs2_glock_dq_uninit(rd_gh); | |
1239 | cond_resched(); | |
1240 | goto more_rgrps; | |
1241 | } | |
1242 | out: | |
1243 | return ret; | |
1244 | } | |
1245 | ||
10d2cf94 AG |
1246 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h) |
1247 | { | |
1248 | if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0]))) | |
1249 | return false; | |
1250 | return true; | |
1251 | } | |
1252 | ||
d552a2b9 BP |
1253 | /** |
1254 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
d552a2b9 BP |
1255 | * @mp: starting metapath |
1256 | * @h: desired height to search | |
1257 | * | |
10d2cf94 | 1258 | * Assumes the metapath is valid (with buffers) out to height h. |
d552a2b9 BP |
1259 | * Returns: true if a non-null pointer was found in the metapath buffer |
1260 | * false if all remaining pointers are NULL in the buffer | |
1261 | */ | |
1262 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
10d2cf94 AG |
1263 | unsigned int h, |
1264 | __u16 *end_list, unsigned int end_aligned) | |
d552a2b9 | 1265 | { |
10d2cf94 AG |
1266 | struct buffer_head *bh = mp->mp_bh[h]; |
1267 | __be64 *first, *ptr, *end; | |
1268 | ||
1269 | first = metaptr1(h, mp); | |
1270 | ptr = first + mp->mp_list[h]; | |
1271 | end = (__be64 *)(bh->b_data + bh->b_size); | |
1272 | if (end_list && mp_eq_to_hgt(mp, end_list, h)) { | |
1273 | bool keep_end = h < end_aligned; | |
1274 | end = first + end_list[h] + keep_end; | |
1275 | } | |
d552a2b9 | 1276 | |
10d2cf94 | 1277 | while (ptr < end) { |
c4a9d189 | 1278 | if (*ptr) { /* if we have a non-null pointer */ |
10d2cf94 | 1279 | mp->mp_list[h] = ptr - first; |
c4a9d189 BP |
1280 | h++; |
1281 | if (h < GFS2_MAX_META_HEIGHT) | |
10d2cf94 | 1282 | mp->mp_list[h] = 0; |
d552a2b9 | 1283 | return true; |
c4a9d189 | 1284 | } |
10d2cf94 | 1285 | ptr++; |
d552a2b9 | 1286 | } |
10d2cf94 | 1287 | return false; |
d552a2b9 BP |
1288 | } |
1289 | ||
1290 | enum dealloc_states { | |
1291 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1292 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1293 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1294 | DEALLOC_DONE = 3, /* process complete */ | |
1295 | }; | |
b3b94faa | 1296 | |
5cf26b1e AG |
1297 | static inline void |
1298 | metapointer_range(struct metapath *mp, int height, | |
1299 | __u16 *start_list, unsigned int start_aligned, | |
10d2cf94 | 1300 | __u16 *end_list, unsigned int end_aligned, |
5cf26b1e AG |
1301 | __be64 **start, __be64 **end) |
1302 | { | |
1303 | struct buffer_head *bh = mp->mp_bh[height]; | |
1304 | __be64 *first; | |
1305 | ||
1306 | first = metaptr1(height, mp); | |
1307 | *start = first; | |
1308 | if (mp_eq_to_hgt(mp, start_list, height)) { | |
1309 | bool keep_start = height < start_aligned; | |
1310 | *start = first + start_list[height] + keep_start; | |
1311 | } | |
1312 | *end = (__be64 *)(bh->b_data + bh->b_size); | |
10d2cf94 AG |
1313 | if (end_list && mp_eq_to_hgt(mp, end_list, height)) { |
1314 | bool keep_end = height < end_aligned; | |
1315 | *end = first + end_list[height] + keep_end; | |
1316 | } | |
1317 | } | |
1318 | ||
1319 | static inline bool walk_done(struct gfs2_sbd *sdp, | |
1320 | struct metapath *mp, int height, | |
1321 | __u16 *end_list, unsigned int end_aligned) | |
1322 | { | |
1323 | __u16 end; | |
1324 | ||
1325 | if (end_list) { | |
1326 | bool keep_end = height < end_aligned; | |
1327 | if (!mp_eq_to_hgt(mp, end_list, height)) | |
1328 | return false; | |
1329 | end = end_list[height] + keep_end; | |
1330 | } else | |
1331 | end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
1332 | return mp->mp_list[height] >= end; | |
5cf26b1e AG |
1333 | } |
1334 | ||
d552a2b9 | 1335 | /** |
10d2cf94 | 1336 | * punch_hole - deallocate blocks in a file |
d552a2b9 | 1337 | * @ip: inode to truncate |
10d2cf94 AG |
1338 | * @offset: the start of the hole |
1339 | * @length: the size of the hole (or 0 for truncate) | |
1340 | * | |
1341 | * Punch a hole into a file or truncate a file at a given position. This | |
1342 | * function operates in whole blocks (@offset and @length are rounded | |
1343 | * accordingly); partially filled blocks must be cleared otherwise. | |
d552a2b9 | 1344 | * |
10d2cf94 AG |
1345 | * This function works from the bottom up, and from the right to the left. In |
1346 | * other words, it strips off the highest layer (data) before stripping any of | |
1347 | * the metadata. Doing it this way is best in case the operation is interrupted | |
1348 | * by power failure, etc. The dinode is rewritten in every transaction to | |
1349 | * guarantee integrity. | |
d552a2b9 | 1350 | */ |
10d2cf94 | 1351 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length) |
d552a2b9 BP |
1352 | { |
1353 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
10d2cf94 | 1354 | struct metapath mp = {}; |
d552a2b9 BP |
1355 | struct buffer_head *dibh, *bh; |
1356 | struct gfs2_holder rd_gh; | |
cb7f0903 | 1357 | unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; |
10d2cf94 AG |
1358 | u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift; |
1359 | __u16 start_list[GFS2_MAX_META_HEIGHT]; | |
1360 | __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL; | |
4e56a641 | 1361 | unsigned int start_aligned, uninitialized_var(end_aligned); |
d552a2b9 BP |
1362 | unsigned int strip_h = ip->i_height - 1; |
1363 | u32 btotal = 0; | |
1364 | int ret, state; | |
1365 | int mp_h; /* metapath buffers are read in to this height */ | |
d552a2b9 | 1366 | u64 prev_bnr = 0; |
5cf26b1e | 1367 | __be64 *start, *end; |
b3b94faa | 1368 | |
10d2cf94 AG |
1369 | /* |
1370 | * The start position of the hole is defined by lblock, start_list, and | |
1371 | * start_aligned. The end position of the hole is defined by lend, | |
1372 | * end_list, and end_aligned. | |
1373 | * | |
1374 | * start_aligned and end_aligned define down to which height the start | |
1375 | * and end positions are aligned to the metadata tree (i.e., the | |
1376 | * position is a multiple of the metadata granularity at the height | |
1377 | * above). This determines at which heights additional meta pointers | |
1378 | * needs to be preserved for the remaining data. | |
1379 | */ | |
b3b94faa | 1380 | |
10d2cf94 AG |
1381 | if (length) { |
1382 | u64 maxsize = sdp->sd_heightsize[ip->i_height]; | |
1383 | u64 end_offset = offset + length; | |
1384 | u64 lend; | |
1385 | ||
1386 | /* | |
1387 | * Clip the end at the maximum file size for the given height: | |
1388 | * that's how far the metadata goes; files bigger than that | |
1389 | * will have additional layers of indirection. | |
1390 | */ | |
1391 | if (end_offset > maxsize) | |
1392 | end_offset = maxsize; | |
1393 | lend = end_offset >> bsize_shift; | |
1394 | ||
1395 | if (lblock >= lend) | |
1396 | return 0; | |
1397 | ||
1398 | find_metapath(sdp, lend, &mp, ip->i_height); | |
1399 | end_list = __end_list; | |
1400 | memcpy(end_list, mp.mp_list, sizeof(mp.mp_list)); | |
1401 | ||
1402 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { | |
1403 | if (end_list[mp_h]) | |
1404 | break; | |
1405 | } | |
1406 | end_aligned = mp_h; | |
1407 | } | |
1408 | ||
1409 | find_metapath(sdp, lblock, &mp, ip->i_height); | |
cb7f0903 AG |
1410 | memcpy(start_list, mp.mp_list, sizeof(start_list)); |
1411 | ||
cb7f0903 AG |
1412 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { |
1413 | if (start_list[mp_h]) | |
1414 | break; | |
1415 | } | |
1416 | start_aligned = mp_h; | |
d552a2b9 BP |
1417 | |
1418 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1419 | if (ret) | |
1420 | return ret; | |
b3b94faa | 1421 | |
d552a2b9 BP |
1422 | mp.mp_bh[0] = dibh; |
1423 | ret = lookup_metapath(ip, &mp); | |
e8b43fe0 AG |
1424 | if (ret) |
1425 | goto out_metapath; | |
c3ce5aa9 AG |
1426 | |
1427 | /* issue read-ahead on metadata */ | |
5cf26b1e AG |
1428 | for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) { |
1429 | metapointer_range(&mp, mp_h, start_list, start_aligned, | |
10d2cf94 | 1430 | end_list, end_aligned, &start, &end); |
5cf26b1e AG |
1431 | gfs2_metapath_ra(ip->i_gl, start, end); |
1432 | } | |
c3ce5aa9 | 1433 | |
e8b43fe0 | 1434 | if (mp.mp_aheight == ip->i_height) |
d552a2b9 BP |
1435 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ |
1436 | else | |
1437 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1438 | |
d552a2b9 BP |
1439 | ret = gfs2_rindex_update(sdp); |
1440 | if (ret) | |
1441 | goto out_metapath; | |
1442 | ||
1443 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1444 | if (ret) | |
1445 | goto out_metapath; | |
1446 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1447 | ||
1448 | mp_h = strip_h; | |
1449 | ||
1450 | while (state != DEALLOC_DONE) { | |
1451 | switch (state) { | |
1452 | /* Truncate a full metapath at the given strip height. | |
1453 | * Note that strip_h == mp_h in order to be in this state. */ | |
1454 | case DEALLOC_MP_FULL: | |
d552a2b9 BP |
1455 | bh = mp.mp_bh[mp_h]; |
1456 | gfs2_assert_withdraw(sdp, bh); | |
1457 | if (gfs2_assert_withdraw(sdp, | |
1458 | prev_bnr != bh->b_blocknr)) { | |
1459 | printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, " | |
1460 | "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n", | |
1461 | sdp->sd_fsname, | |
1462 | (unsigned long long)ip->i_no_addr, | |
1463 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1464 | } | |
1465 | prev_bnr = bh->b_blocknr; | |
cb7f0903 | 1466 | |
5cf26b1e AG |
1467 | if (gfs2_metatype_check(sdp, bh, |
1468 | (mp_h ? GFS2_METATYPE_IN : | |
1469 | GFS2_METATYPE_DI))) { | |
1470 | ret = -EIO; | |
1471 | goto out; | |
1472 | } | |
1473 | ||
10d2cf94 AG |
1474 | /* |
1475 | * Below, passing end_aligned as 0 gives us the | |
1476 | * metapointer range excluding the end point: the end | |
1477 | * point is the first metapath we must not deallocate! | |
1478 | */ | |
1479 | ||
5cf26b1e | 1480 | metapointer_range(&mp, mp_h, start_list, start_aligned, |
10d2cf94 | 1481 | end_list, 0 /* end_aligned */, |
5cf26b1e AG |
1482 | &start, &end); |
1483 | ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h], | |
1484 | start, end, | |
1485 | mp_h != ip->i_height - 1, | |
1486 | &btotal); | |
cb7f0903 | 1487 | |
d552a2b9 BP |
1488 | /* If we hit an error or just swept dinode buffer, |
1489 | just exit. */ | |
1490 | if (ret || !mp_h) { | |
1491 | state = DEALLOC_DONE; | |
1492 | break; | |
1493 | } | |
1494 | state = DEALLOC_MP_LOWER; | |
1495 | break; | |
1496 | ||
1497 | /* lower the metapath strip height */ | |
1498 | case DEALLOC_MP_LOWER: | |
1499 | /* We're done with the current buffer, so release it, | |
1500 | unless it's the dinode buffer. Then back up to the | |
1501 | previous pointer. */ | |
1502 | if (mp_h) { | |
1503 | brelse(mp.mp_bh[mp_h]); | |
1504 | mp.mp_bh[mp_h] = NULL; | |
1505 | } | |
1506 | /* If we can't get any lower in height, we've stripped | |
1507 | off all we can. Next step is to back up and start | |
1508 | stripping the previous level of metadata. */ | |
1509 | if (mp_h == 0) { | |
1510 | strip_h--; | |
cb7f0903 | 1511 | memcpy(mp.mp_list, start_list, sizeof(start_list)); |
d552a2b9 BP |
1512 | mp_h = strip_h; |
1513 | state = DEALLOC_FILL_MP; | |
1514 | break; | |
1515 | } | |
1516 | mp.mp_list[mp_h] = 0; | |
1517 | mp_h--; /* search one metadata height down */ | |
d552a2b9 | 1518 | mp.mp_list[mp_h]++; |
10d2cf94 AG |
1519 | if (walk_done(sdp, &mp, mp_h, end_list, end_aligned)) |
1520 | break; | |
d552a2b9 BP |
1521 | /* Here we've found a part of the metapath that is not |
1522 | * allocated. We need to search at that height for the | |
1523 | * next non-null pointer. */ | |
10d2cf94 | 1524 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) { |
d552a2b9 BP |
1525 | state = DEALLOC_FILL_MP; |
1526 | mp_h++; | |
1527 | } | |
1528 | /* No more non-null pointers at this height. Back up | |
1529 | to the previous height and try again. */ | |
1530 | break; /* loop around in the same state */ | |
1531 | ||
1532 | /* Fill the metapath with buffers to the given height. */ | |
1533 | case DEALLOC_FILL_MP: | |
1534 | /* Fill the buffers out to the current height. */ | |
1535 | ret = fillup_metapath(ip, &mp, mp_h); | |
c3ce5aa9 | 1536 | if (ret < 0) |
d552a2b9 | 1537 | goto out; |
c3ce5aa9 AG |
1538 | |
1539 | /* issue read-ahead on metadata */ | |
1540 | if (mp.mp_aheight > 1) { | |
5cf26b1e AG |
1541 | for (; ret > 1; ret--) { |
1542 | metapointer_range(&mp, mp.mp_aheight - ret, | |
1543 | start_list, start_aligned, | |
10d2cf94 | 1544 | end_list, end_aligned, |
5cf26b1e AG |
1545 | &start, &end); |
1546 | gfs2_metapath_ra(ip->i_gl, start, end); | |
1547 | } | |
c3ce5aa9 | 1548 | } |
d552a2b9 BP |
1549 | |
1550 | /* If buffers found for the entire strip height */ | |
e8b43fe0 | 1551 | if (mp.mp_aheight - 1 == strip_h) { |
d552a2b9 BP |
1552 | state = DEALLOC_MP_FULL; |
1553 | break; | |
1554 | } | |
e8b43fe0 AG |
1555 | if (mp.mp_aheight < ip->i_height) /* We have a partial height */ |
1556 | mp_h = mp.mp_aheight - 1; | |
d552a2b9 BP |
1557 | |
1558 | /* If we find a non-null block pointer, crawl a bit | |
1559 | higher up in the metapath and try again, otherwise | |
1560 | we need to look lower for a new starting point. */ | |
10d2cf94 | 1561 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) |
d552a2b9 BP |
1562 | mp_h++; |
1563 | else | |
1564 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1565 | break; |
d552a2b9 | 1566 | } |
b3b94faa DT |
1567 | } |
1568 | ||
d552a2b9 BP |
1569 | if (btotal) { |
1570 | if (current->journal_info == NULL) { | |
1571 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1572 | RES_QUOTA, 0); | |
1573 | if (ret) | |
1574 | goto out; | |
1575 | down_write(&ip->i_rw_mutex); | |
1576 | } | |
1577 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
1578 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
1579 | ip->i_inode.i_gid); | |
b32c8c76 | 1580 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
d552a2b9 BP |
1581 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1582 | gfs2_dinode_out(ip, dibh->b_data); | |
1583 | up_write(&ip->i_rw_mutex); | |
1584 | gfs2_trans_end(sdp); | |
1585 | } | |
b3b94faa | 1586 | |
d552a2b9 BP |
1587 | out: |
1588 | if (gfs2_holder_initialized(&rd_gh)) | |
1589 | gfs2_glock_dq_uninit(&rd_gh); | |
1590 | if (current->journal_info) { | |
1591 | up_write(&ip->i_rw_mutex); | |
1592 | gfs2_trans_end(sdp); | |
1593 | cond_resched(); | |
1594 | } | |
1595 | gfs2_quota_unhold(ip); | |
1596 | out_metapath: | |
1597 | release_metapath(&mp); | |
1598 | return ret; | |
b3b94faa DT |
1599 | } |
1600 | ||
1601 | static int trunc_end(struct gfs2_inode *ip) | |
1602 | { | |
feaa7bba | 1603 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1604 | struct buffer_head *dibh; |
1605 | int error; | |
1606 | ||
1607 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1608 | if (error) | |
1609 | return error; | |
1610 | ||
1611 | down_write(&ip->i_rw_mutex); | |
1612 | ||
1613 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1614 | if (error) | |
1615 | goto out; | |
1616 | ||
a2e0f799 | 1617 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1618 | ip->i_height = 0; |
ce276b06 | 1619 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 1620 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 1621 | gfs2_ordered_del_inode(ip); |
b3b94faa | 1622 | } |
078cd827 | 1623 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
383f01fb | 1624 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1625 | |
350a9b0a | 1626 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 1627 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1628 | brelse(dibh); |
1629 | ||
a91ea69f | 1630 | out: |
b3b94faa | 1631 | up_write(&ip->i_rw_mutex); |
b3b94faa | 1632 | gfs2_trans_end(sdp); |
b3b94faa DT |
1633 | return error; |
1634 | } | |
1635 | ||
1636 | /** | |
1637 | * do_shrink - make a file smaller | |
ff8f33c8 | 1638 | * @inode: the inode |
ff8f33c8 | 1639 | * @newsize: the size to make the file |
b3b94faa | 1640 | * |
ff8f33c8 SW |
1641 | * Called with an exclusive lock on @inode. The @size must |
1642 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
1643 | * |
1644 | * Returns: errno | |
1645 | */ | |
1646 | ||
8b5860a3 | 1647 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 1648 | { |
ff8f33c8 | 1649 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
1650 | int error; |
1651 | ||
8b5860a3 | 1652 | error = trunc_start(inode, newsize); |
b3b94faa DT |
1653 | if (error < 0) |
1654 | return error; | |
ff8f33c8 | 1655 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
1656 | return 0; |
1657 | ||
10d2cf94 | 1658 | error = punch_hole(ip, newsize, 0); |
ff8f33c8 | 1659 | if (error == 0) |
b3b94faa DT |
1660 | error = trunc_end(ip); |
1661 | ||
1662 | return error; | |
1663 | } | |
1664 | ||
ff8f33c8 | 1665 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 1666 | { |
ff8f33c8 SW |
1667 | int ret; |
1668 | ||
8b5860a3 | 1669 | ret = do_shrink(inode, inode->i_size); |
ff8f33c8 SW |
1670 | WARN_ON(ret != 0); |
1671 | } | |
1672 | ||
1673 | /** | |
1674 | * do_grow - Touch and update inode size | |
1675 | * @inode: The inode | |
1676 | * @size: The new size | |
1677 | * | |
1678 | * This function updates the timestamps on the inode and | |
1679 | * may also increase the size of the inode. This function | |
1680 | * must not be called with @size any smaller than the current | |
1681 | * inode size. | |
1682 | * | |
1683 | * Although it is not strictly required to unstuff files here, | |
1684 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
1685 | * code which will result in a buffer overrun if the size is larger | |
1686 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 1687 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
1688 | * just update the inode size directly. |
1689 | * | |
1690 | * Returns: 0 on success, or -ve on error | |
1691 | */ | |
1692 | ||
1693 | static int do_grow(struct inode *inode, u64 size) | |
1694 | { | |
1695 | struct gfs2_inode *ip = GFS2_I(inode); | |
1696 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 1697 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
1698 | struct buffer_head *dibh; |
1699 | int error; | |
2f7ee358 | 1700 | int unstuff = 0; |
a13b8c5f | 1701 | |
235628c5 | 1702 | if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) { |
b8fbf471 | 1703 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 1704 | if (error) |
5407e242 | 1705 | return error; |
ff8f33c8 | 1706 | |
7b9cff46 | 1707 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
1708 | if (error) |
1709 | goto do_grow_qunlock; | |
2f7ee358 | 1710 | unstuff = 1; |
ff8f33c8 SW |
1711 | } |
1712 | ||
a01aedfe BP |
1713 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
1714 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? | |
1715 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 1716 | if (error) |
ff8f33c8 | 1717 | goto do_grow_release; |
a13b8c5f | 1718 | |
2f7ee358 | 1719 | if (unstuff) { |
ff8f33c8 SW |
1720 | error = gfs2_unstuff_dinode(ip, NULL); |
1721 | if (error) | |
1722 | goto do_end_trans; | |
1723 | } | |
a13b8c5f WC |
1724 | |
1725 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1726 | if (error) | |
ff8f33c8 | 1727 | goto do_end_trans; |
a13b8c5f | 1728 | |
ff8f33c8 | 1729 | i_size_write(inode, size); |
078cd827 | 1730 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
350a9b0a | 1731 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
1732 | gfs2_dinode_out(ip, dibh->b_data); |
1733 | brelse(dibh); | |
1734 | ||
ff8f33c8 | 1735 | do_end_trans: |
a13b8c5f | 1736 | gfs2_trans_end(sdp); |
ff8f33c8 | 1737 | do_grow_release: |
2f7ee358 | 1738 | if (unstuff) { |
ff8f33c8 SW |
1739 | gfs2_inplace_release(ip); |
1740 | do_grow_qunlock: | |
1741 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 1742 | } |
a13b8c5f WC |
1743 | return error; |
1744 | } | |
1745 | ||
b3b94faa | 1746 | /** |
ff8f33c8 SW |
1747 | * gfs2_setattr_size - make a file a given size |
1748 | * @inode: the inode | |
1749 | * @newsize: the size to make the file | |
b3b94faa | 1750 | * |
ff8f33c8 SW |
1751 | * The file size can grow, shrink, or stay the same size. This |
1752 | * is called holding i_mutex and an exclusive glock on the inode | |
1753 | * in question. | |
b3b94faa DT |
1754 | * |
1755 | * Returns: errno | |
1756 | */ | |
1757 | ||
ff8f33c8 | 1758 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 1759 | { |
af5c2697 | 1760 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 1761 | int ret; |
b3b94faa | 1762 | |
ff8f33c8 | 1763 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 1764 | |
ff8f33c8 SW |
1765 | ret = inode_newsize_ok(inode, newsize); |
1766 | if (ret) | |
1767 | return ret; | |
b3b94faa | 1768 | |
562c72aa CH |
1769 | inode_dio_wait(inode); |
1770 | ||
b54e9a0b | 1771 | ret = gfs2_rsqa_alloc(ip); |
d2b47cfb | 1772 | if (ret) |
2b3dcf35 | 1773 | goto out; |
d2b47cfb | 1774 | |
8b5860a3 | 1775 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
1776 | ret = do_grow(inode, newsize); |
1777 | goto out; | |
1778 | } | |
ff8f33c8 | 1779 | |
8b5860a3 | 1780 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 1781 | out: |
a097dc7e | 1782 | gfs2_rsqa_delete(ip, NULL); |
2b3dcf35 | 1783 | return ret; |
b3b94faa DT |
1784 | } |
1785 | ||
1786 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1787 | { | |
1788 | int error; | |
10d2cf94 | 1789 | error = punch_hole(ip, i_size_read(&ip->i_inode), 0); |
b3b94faa DT |
1790 | if (!error) |
1791 | error = trunc_end(ip); | |
1792 | return error; | |
1793 | } | |
1794 | ||
1795 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1796 | { | |
10d2cf94 | 1797 | return punch_hole(ip, 0, 0); |
b3b94faa DT |
1798 | } |
1799 | ||
b50f227b SW |
1800 | /** |
1801 | * gfs2_free_journal_extents - Free cached journal bmap info | |
1802 | * @jd: The journal | |
1803 | * | |
1804 | */ | |
1805 | ||
1806 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
1807 | { | |
1808 | struct gfs2_journal_extent *jext; | |
1809 | ||
1810 | while(!list_empty(&jd->extent_list)) { | |
1811 | jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list); | |
1812 | list_del(&jext->list); | |
1813 | kfree(jext); | |
1814 | } | |
1815 | } | |
1816 | ||
1817 | /** | |
1818 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
1819 | * @jd: The journal descriptor | |
1820 | * @lblock: The logical block at start of new extent | |
c62baf65 | 1821 | * @dblock: The physical block at start of new extent |
b50f227b SW |
1822 | * @blocks: Size of extent in fs blocks |
1823 | * | |
1824 | * Returns: 0 on success or -ENOMEM | |
1825 | */ | |
1826 | ||
1827 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
1828 | { | |
1829 | struct gfs2_journal_extent *jext; | |
1830 | ||
1831 | if (!list_empty(&jd->extent_list)) { | |
1832 | jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list); | |
1833 | if ((jext->dblock + jext->blocks) == dblock) { | |
1834 | jext->blocks += blocks; | |
1835 | return 0; | |
1836 | } | |
1837 | } | |
1838 | ||
1839 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
1840 | if (jext == NULL) | |
1841 | return -ENOMEM; | |
1842 | jext->dblock = dblock; | |
1843 | jext->lblock = lblock; | |
1844 | jext->blocks = blocks; | |
1845 | list_add_tail(&jext->list, &jd->extent_list); | |
1846 | jd->nr_extents++; | |
1847 | return 0; | |
1848 | } | |
1849 | ||
1850 | /** | |
1851 | * gfs2_map_journal_extents - Cache journal bmap info | |
1852 | * @sdp: The super block | |
1853 | * @jd: The journal to map | |
1854 | * | |
1855 | * Create a reusable "extent" mapping from all logical | |
1856 | * blocks to all physical blocks for the given journal. This will save | |
1857 | * us time when writing journal blocks. Most journals will have only one | |
1858 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
1859 | * arranges the journal blocks sequentially to maximize performance. | |
1860 | * So the extent would map the first block for the entire file length. | |
1861 | * However, gfs2_jadd can happen while file activity is happening, so | |
1862 | * those journals may not be sequential. Less likely is the case where | |
1863 | * the users created their own journals by mounting the metafs and | |
1864 | * laying it out. But it's still possible. These journals might have | |
1865 | * several extents. | |
1866 | * | |
1867 | * Returns: 0 on success, or error on failure | |
1868 | */ | |
1869 | ||
1870 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
1871 | { | |
1872 | u64 lblock = 0; | |
1873 | u64 lblock_stop; | |
1874 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
1875 | struct buffer_head bh; | |
1876 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
1877 | u64 size; | |
1878 | int rc; | |
1879 | ||
1880 | lblock_stop = i_size_read(jd->jd_inode) >> shift; | |
1881 | size = (lblock_stop - lblock) << shift; | |
1882 | jd->nr_extents = 0; | |
1883 | WARN_ON(!list_empty(&jd->extent_list)); | |
1884 | ||
1885 | do { | |
1886 | bh.b_state = 0; | |
1887 | bh.b_blocknr = 0; | |
1888 | bh.b_size = size; | |
1889 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
1890 | if (rc || !buffer_mapped(&bh)) | |
1891 | goto fail; | |
1892 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
1893 | if (rc) | |
1894 | goto fail; | |
1895 | size -= bh.b_size; | |
1896 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
1897 | } while(size > 0); | |
1898 | ||
1899 | fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid, | |
1900 | jd->nr_extents); | |
1901 | return 0; | |
1902 | ||
1903 | fail: | |
1904 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
1905 | rc, jd->jd_jid, | |
1906 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
1907 | jd->nr_extents); | |
1908 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
1909 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
1910 | bh.b_state, (unsigned long long)bh.b_size); | |
1911 | gfs2_free_journal_extents(jd); | |
1912 | return rc; | |
1913 | } | |
1914 | ||
b3b94faa DT |
1915 | /** |
1916 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
1917 | * @ip: the file being written to | |
1918 | * @offset: the offset to write to | |
1919 | * @len: the number of bytes being written | |
b3b94faa | 1920 | * |
461cb419 | 1921 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
1922 | */ |
1923 | ||
cd915493 | 1924 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 1925 | unsigned int len) |
b3b94faa | 1926 | { |
feaa7bba | 1927 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
1928 | struct buffer_head bh; |
1929 | unsigned int shift; | |
1930 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 1931 | u64 end_of_file; |
b3b94faa | 1932 | |
b3b94faa DT |
1933 | if (!len) |
1934 | return 0; | |
1935 | ||
1936 | if (gfs2_is_stuffed(ip)) { | |
235628c5 | 1937 | if (offset + len > gfs2_max_stuffed_size(ip)) |
461cb419 | 1938 | return 1; |
b3b94faa DT |
1939 | return 0; |
1940 | } | |
1941 | ||
941e6d7d | 1942 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 1943 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 1944 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
1945 | lblock = offset >> shift; |
1946 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
1947 | if (lblock_stop > end_of_file) | |
461cb419 | 1948 | return 1; |
b3b94faa | 1949 | |
941e6d7d SW |
1950 | size = (lblock_stop - lblock) << shift; |
1951 | do { | |
1952 | bh.b_state = 0; | |
1953 | bh.b_size = size; | |
1954 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
1955 | if (!buffer_mapped(&bh)) | |
461cb419 | 1956 | return 1; |
941e6d7d SW |
1957 | size -= bh.b_size; |
1958 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
1959 | } while(size > 0); | |
b3b94faa DT |
1960 | |
1961 | return 0; | |
1962 | } | |
1963 | ||
4e56a641 AG |
1964 | static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length) |
1965 | { | |
1966 | struct gfs2_inode *ip = GFS2_I(inode); | |
1967 | struct buffer_head *dibh; | |
1968 | int error; | |
1969 | ||
1970 | if (offset >= inode->i_size) | |
1971 | return 0; | |
1972 | if (offset + length > inode->i_size) | |
1973 | length = inode->i_size - offset; | |
1974 | ||
1975 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1976 | if (error) | |
1977 | return error; | |
1978 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
1979 | memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0, | |
1980 | length); | |
1981 | brelse(dibh); | |
1982 | return 0; | |
1983 | } | |
1984 | ||
1985 | static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset, | |
1986 | loff_t length) | |
1987 | { | |
1988 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1989 | loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1990 | int error; | |
1991 | ||
1992 | while (length) { | |
1993 | struct gfs2_trans *tr; | |
1994 | loff_t chunk; | |
1995 | unsigned int offs; | |
1996 | ||
1997 | chunk = length; | |
1998 | if (chunk > max_chunk) | |
1999 | chunk = max_chunk; | |
2000 | ||
2001 | offs = offset & ~PAGE_MASK; | |
2002 | if (offs && chunk > PAGE_SIZE) | |
2003 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
2004 | ||
2005 | truncate_pagecache_range(inode, offset, chunk); | |
2006 | offset += chunk; | |
2007 | length -= chunk; | |
2008 | ||
2009 | tr = current->journal_info; | |
2010 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
2011 | continue; | |
2012 | ||
2013 | gfs2_trans_end(sdp); | |
2014 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
2015 | if (error) | |
2016 | return error; | |
2017 | } | |
2018 | return 0; | |
2019 | } | |
2020 | ||
2021 | int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length) | |
2022 | { | |
2023 | struct inode *inode = file_inode(file); | |
2024 | struct gfs2_inode *ip = GFS2_I(inode); | |
2025 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2026 | int error; | |
2027 | ||
2028 | if (gfs2_is_jdata(ip)) | |
2029 | error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA, | |
2030 | GFS2_JTRUNC_REVOKES); | |
2031 | else | |
2032 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2033 | if (error) | |
2034 | return error; | |
2035 | ||
2036 | if (gfs2_is_stuffed(ip)) { | |
2037 | error = stuffed_zero_range(inode, offset, length); | |
2038 | if (error) | |
2039 | goto out; | |
2040 | } else { | |
2041 | unsigned int start_off, end_off, blocksize; | |
2042 | ||
2043 | blocksize = i_blocksize(inode); | |
2044 | start_off = offset & (blocksize - 1); | |
2045 | end_off = (offset + length) & (blocksize - 1); | |
2046 | if (start_off) { | |
2047 | unsigned int len = length; | |
2048 | if (length > blocksize - start_off) | |
2049 | len = blocksize - start_off; | |
2050 | error = gfs2_block_zero_range(inode, offset, len); | |
2051 | if (error) | |
2052 | goto out; | |
2053 | if (start_off + length < blocksize) | |
2054 | end_off = 0; | |
2055 | } | |
2056 | if (end_off) { | |
2057 | error = gfs2_block_zero_range(inode, | |
2058 | offset + length - end_off, end_off); | |
2059 | if (error) | |
2060 | goto out; | |
2061 | } | |
2062 | } | |
2063 | ||
2064 | if (gfs2_is_jdata(ip)) { | |
2065 | BUG_ON(!current->journal_info); | |
2066 | gfs2_journaled_truncate_range(inode, offset, length); | |
2067 | } else | |
2068 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2069 | ||
2070 | file_update_time(file); | |
2071 | mark_inode_dirty(inode); | |
2072 | ||
2073 | if (current->journal_info) | |
2074 | gfs2_trans_end(sdp); | |
2075 | ||
2076 | if (!gfs2_is_stuffed(ip)) | |
2077 | error = punch_hole(ip, offset, length); | |
2078 | ||
2079 | out: | |
2080 | if (current->journal_info) | |
2081 | gfs2_trans_end(sdp); | |
2082 | return error; | |
2083 | } |