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> |
98583b3e | 17 | #include <linux/ktime.h> |
b3b94faa DT |
18 | |
19 | #include "gfs2.h" | |
5c676f6d | 20 | #include "incore.h" |
b3b94faa DT |
21 | #include "bmap.h" |
22 | #include "glock.h" | |
23 | #include "inode.h" | |
b3b94faa | 24 | #include "meta_io.h" |
b3b94faa DT |
25 | #include "quota.h" |
26 | #include "rgrp.h" | |
45138990 | 27 | #include "log.h" |
4c16c36a | 28 | #include "super.h" |
b3b94faa | 29 | #include "trans.h" |
18ec7d5c | 30 | #include "dir.h" |
5c676f6d | 31 | #include "util.h" |
64bc06bb | 32 | #include "aops.h" |
63997775 | 33 | #include "trace_gfs2.h" |
b3b94faa DT |
34 | |
35 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
36 | * block is 512, so __u16 is fine for that. It saves stack space to | |
37 | * keep it small. | |
38 | */ | |
39 | struct metapath { | |
dbac6710 | 40 | struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT]; |
b3b94faa | 41 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; |
5f8bd444 BP |
42 | int mp_fheight; /* find_metapath height */ |
43 | int mp_aheight; /* actual height (lookup height) */ | |
b3b94faa DT |
44 | }; |
45 | ||
64bc06bb AG |
46 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length); |
47 | ||
f25ef0c1 SW |
48 | /** |
49 | * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page | |
50 | * @ip: the inode | |
51 | * @dibh: the dinode buffer | |
52 | * @block: the block number that was allocated | |
ff8f33c8 | 53 | * @page: The (optional) page. This is looked up if @page is NULL |
f25ef0c1 SW |
54 | * |
55 | * Returns: errno | |
56 | */ | |
57 | ||
58 | static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, | |
cd915493 | 59 | u64 block, struct page *page) |
f25ef0c1 | 60 | { |
f25ef0c1 SW |
61 | struct inode *inode = &ip->i_inode; |
62 | struct buffer_head *bh; | |
63 | int release = 0; | |
64 | ||
65 | if (!page || page->index) { | |
220cca2a | 66 | page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS); |
f25ef0c1 SW |
67 | if (!page) |
68 | return -ENOMEM; | |
69 | release = 1; | |
70 | } | |
71 | ||
72 | if (!PageUptodate(page)) { | |
73 | void *kaddr = kmap(page); | |
602c89d2 SW |
74 | u64 dsize = i_size_read(inode); |
75 | ||
235628c5 AG |
76 | if (dsize > gfs2_max_stuffed_size(ip)) |
77 | dsize = gfs2_max_stuffed_size(ip); | |
f25ef0c1 | 78 | |
602c89d2 | 79 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 80 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
f25ef0c1 SW |
81 | kunmap(page); |
82 | ||
83 | SetPageUptodate(page); | |
84 | } | |
85 | ||
86 | if (!page_has_buffers(page)) | |
47a9a527 FF |
87 | create_empty_buffers(page, BIT(inode->i_blkbits), |
88 | BIT(BH_Uptodate)); | |
f25ef0c1 SW |
89 | |
90 | bh = page_buffers(page); | |
91 | ||
92 | if (!buffer_mapped(bh)) | |
93 | map_bh(bh, inode->i_sb, block); | |
94 | ||
95 | set_buffer_uptodate(bh); | |
845802b1 | 96 | if (gfs2_is_jdata(ip)) |
350a9b0a | 97 | gfs2_trans_add_data(ip->i_gl, bh); |
845802b1 AG |
98 | else { |
99 | mark_buffer_dirty(bh); | |
100 | gfs2_ordered_add_inode(ip); | |
101 | } | |
f25ef0c1 SW |
102 | |
103 | if (release) { | |
104 | unlock_page(page); | |
09cbfeaf | 105 | put_page(page); |
f25ef0c1 SW |
106 | } |
107 | ||
108 | return 0; | |
109 | } | |
110 | ||
b3b94faa DT |
111 | /** |
112 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
113 | * @ip: The GFS2 inode to unstuff | |
ff8f33c8 | 114 | * @page: The (optional) page. This is looked up if the @page is NULL |
b3b94faa DT |
115 | * |
116 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
117 | * that the height can be grown in the traditional way. | |
118 | * | |
119 | * Returns: errno | |
120 | */ | |
121 | ||
f25ef0c1 | 122 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
123 | { |
124 | struct buffer_head *bh, *dibh; | |
48516ced | 125 | struct gfs2_dinode *di; |
cd915493 | 126 | u64 block = 0; |
18ec7d5c | 127 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
128 | int error; |
129 | ||
130 | down_write(&ip->i_rw_mutex); | |
131 | ||
132 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
133 | if (error) | |
134 | goto out; | |
907b9bce | 135 | |
a2e0f799 | 136 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
137 | /* Get a free block, fill it with the stuffed data, |
138 | and write it out to disk */ | |
139 | ||
b45e41d7 | 140 | unsigned int n = 1; |
6e87ed0f | 141 | error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL); |
09010978 SW |
142 | if (error) |
143 | goto out_brelse; | |
18ec7d5c | 144 | if (isdir) { |
fbb27873 | 145 | gfs2_trans_remove_revoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 146 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
147 | if (error) |
148 | goto out_brelse; | |
48516ced | 149 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
150 | dibh, sizeof(struct gfs2_dinode)); |
151 | brelse(bh); | |
152 | } else { | |
f25ef0c1 | 153 | error = gfs2_unstuffer_page(ip, dibh, block, page); |
b3b94faa DT |
154 | if (error) |
155 | goto out_brelse; | |
156 | } | |
157 | } | |
158 | ||
159 | /* Set up the pointer to the new block */ | |
160 | ||
350a9b0a | 161 | gfs2_trans_add_meta(ip->i_gl, dibh); |
48516ced | 162 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
163 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
164 | ||
a2e0f799 | 165 | if (i_size_read(&ip->i_inode)) { |
48516ced | 166 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
167 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
168 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
169 | } |
170 | ||
ecc30c79 | 171 | ip->i_height = 1; |
48516ced | 172 | di->di_height = cpu_to_be16(1); |
b3b94faa | 173 | |
a91ea69f | 174 | out_brelse: |
b3b94faa | 175 | brelse(dibh); |
a91ea69f | 176 | out: |
b3b94faa | 177 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
178 | return error; |
179 | } | |
180 | ||
b3b94faa DT |
181 | |
182 | /** | |
183 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 184 | * @sdp: The superblock |
b3b94faa | 185 | * @block: The disk block to look up |
07e23d68 | 186 | * @mp: The metapath to return the result in |
9b8c81d1 | 187 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
188 | * |
189 | * This routine returns a struct metapath structure that defines a path | |
190 | * through the metadata of inode "ip" to get to block "block". | |
191 | * | |
192 | * Example: | |
193 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
194 | * filesystem with a blocksize of 4096. | |
195 | * | |
196 | * find_metapath() would return a struct metapath structure set to: | |
07e23d68 | 197 | * mp_fheight = 3, mp_list[0] = 0, mp_list[1] = 48, and mp_list[2] = 165. |
b3b94faa DT |
198 | * |
199 | * That means that in order to get to the block containing the byte at | |
200 | * offset 101342453, we would load the indirect block pointed to by pointer | |
201 | * 0 in the dinode. We would then load the indirect block pointed to by | |
202 | * pointer 48 in that indirect block. We would then load the data block | |
203 | * pointed to by pointer 165 in that indirect block. | |
204 | * | |
205 | * ---------------------------------------- | |
206 | * | Dinode | | | |
207 | * | | 4| | |
208 | * | |0 1 2 3 4 5 9| | |
209 | * | | 6| | |
210 | * ---------------------------------------- | |
211 | * | | |
212 | * | | |
213 | * V | |
214 | * ---------------------------------------- | |
215 | * | Indirect Block | | |
216 | * | 5| | |
217 | * | 4 4 4 4 4 5 5 1| | |
218 | * |0 5 6 7 8 9 0 1 2| | |
219 | * ---------------------------------------- | |
220 | * | | |
221 | * | | |
222 | * V | |
223 | * ---------------------------------------- | |
224 | * | Indirect Block | | |
225 | * | 1 1 1 1 1 5| | |
226 | * | 6 6 6 6 6 1| | |
227 | * |0 3 4 5 6 7 2| | |
228 | * ---------------------------------------- | |
229 | * | | |
230 | * | | |
231 | * V | |
232 | * ---------------------------------------- | |
233 | * | Data block containing offset | | |
234 | * | 101342453 | | |
235 | * | | | |
236 | * | | | |
237 | * ---------------------------------------- | |
238 | * | |
239 | */ | |
240 | ||
9b8c81d1 SW |
241 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
242 | struct metapath *mp, unsigned int height) | |
b3b94faa | 243 | { |
b3b94faa DT |
244 | unsigned int i; |
245 | ||
5f8bd444 | 246 | mp->mp_fheight = height; |
9b8c81d1 | 247 | for (i = height; i--;) |
7eabb77e | 248 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
249 | } |
250 | ||
5af4e7a0 | 251 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 252 | { |
5af4e7a0 BM |
253 | if (mp->mp_list[0] == 0) |
254 | return 2; | |
255 | return 1; | |
9b8c81d1 SW |
256 | } |
257 | ||
d552a2b9 | 258 | /** |
20cdc193 | 259 | * metaptr1 - Return the first possible metadata pointer in a metapath buffer |
d552a2b9 BP |
260 | * @height: The metadata height (0 = dinode) |
261 | * @mp: The metapath | |
262 | */ | |
263 | static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp) | |
264 | { | |
265 | struct buffer_head *bh = mp->mp_bh[height]; | |
266 | if (height == 0) | |
267 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode))); | |
268 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header))); | |
269 | } | |
270 | ||
b3b94faa DT |
271 | /** |
272 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
273 | * @height: The metadata height (0 = dinode) |
274 | * @mp: The metapath | |
275 | * | |
276 | * Return a pointer to the block number of the next height of the metadata | |
277 | * tree given a buffer containing the pointer to the current height of the | |
278 | * metadata tree. | |
279 | */ | |
280 | ||
9b8c81d1 | 281 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 282 | { |
d552a2b9 BP |
283 | __be64 *p = metaptr1(height, mp); |
284 | return p + mp->mp_list[height]; | |
b3b94faa DT |
285 | } |
286 | ||
7841b9f0 AG |
287 | static inline const __be64 *metaend(unsigned int height, const struct metapath *mp) |
288 | { | |
289 | const struct buffer_head *bh = mp->mp_bh[height]; | |
290 | return (const __be64 *)(bh->b_data + bh->b_size); | |
291 | } | |
292 | ||
293 | static void clone_metapath(struct metapath *clone, struct metapath *mp) | |
294 | { | |
295 | unsigned int hgt; | |
296 | ||
297 | *clone = *mp; | |
298 | for (hgt = 0; hgt < mp->mp_aheight; hgt++) | |
299 | get_bh(clone->mp_bh[hgt]); | |
300 | } | |
301 | ||
5cf26b1e | 302 | static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end) |
b99b98dc | 303 | { |
b99b98dc SW |
304 | const __be64 *t; |
305 | ||
5cf26b1e | 306 | for (t = start; t < end; t++) { |
c3ce5aa9 AG |
307 | struct buffer_head *rabh; |
308 | ||
b99b98dc SW |
309 | if (!*t) |
310 | continue; | |
311 | ||
312 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
313 | if (trylock_buffer(rabh)) { | |
314 | if (!buffer_uptodate(rabh)) { | |
315 | rabh->b_end_io = end_buffer_read_sync; | |
e477b24b CL |
316 | submit_bh(REQ_OP_READ, |
317 | REQ_RAHEAD | REQ_META | REQ_PRIO, | |
318 | rabh); | |
b99b98dc SW |
319 | continue; |
320 | } | |
321 | unlock_buffer(rabh); | |
322 | } | |
323 | brelse(rabh); | |
324 | } | |
325 | } | |
326 | ||
e8b43fe0 AG |
327 | static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, |
328 | unsigned int x, unsigned int h) | |
d552a2b9 | 329 | { |
e8b43fe0 AG |
330 | for (; x < h; x++) { |
331 | __be64 *ptr = metapointer(x, mp); | |
332 | u64 dblock = be64_to_cpu(*ptr); | |
333 | int ret; | |
d552a2b9 | 334 | |
e8b43fe0 AG |
335 | if (!dblock) |
336 | break; | |
337 | ret = gfs2_meta_indirect_buffer(ip, x + 1, dblock, &mp->mp_bh[x + 1]); | |
338 | if (ret) | |
339 | return ret; | |
340 | } | |
341 | mp->mp_aheight = x + 1; | |
342 | return 0; | |
d552a2b9 BP |
343 | } |
344 | ||
b3b94faa | 345 | /** |
9b8c81d1 SW |
346 | * lookup_metapath - Walk the metadata tree to a specific point |
347 | * @ip: The inode | |
b3b94faa | 348 | * @mp: The metapath |
b3b94faa | 349 | * |
9b8c81d1 SW |
350 | * Assumes that the inode's buffer has already been looked up and |
351 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
352 | * by find_metapath(). | |
353 | * | |
354 | * If this function encounters part of the tree which has not been | |
355 | * allocated, it returns the current height of the tree at the point | |
356 | * at which it found the unallocated block. Blocks which are found are | |
357 | * added to the mp->mp_bh[] list. | |
b3b94faa | 358 | * |
e8b43fe0 | 359 | * Returns: error |
b3b94faa DT |
360 | */ |
361 | ||
9b8c81d1 | 362 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 363 | { |
e8b43fe0 | 364 | return __fillup_metapath(ip, mp, 0, ip->i_height - 1); |
dbac6710 SW |
365 | } |
366 | ||
d552a2b9 BP |
367 | /** |
368 | * fillup_metapath - fill up buffers for the metadata path to a specific height | |
369 | * @ip: The inode | |
370 | * @mp: The metapath | |
371 | * @h: The height to which it should be mapped | |
372 | * | |
373 | * Similar to lookup_metapath, but does lookups for a range of heights | |
374 | * | |
c3ce5aa9 | 375 | * Returns: error or the number of buffers filled |
d552a2b9 BP |
376 | */ |
377 | ||
378 | static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h) | |
379 | { | |
e8b43fe0 | 380 | unsigned int x = 0; |
c3ce5aa9 | 381 | int ret; |
d552a2b9 BP |
382 | |
383 | if (h) { | |
384 | /* find the first buffer we need to look up. */ | |
e8b43fe0 AG |
385 | for (x = h - 1; x > 0; x--) { |
386 | if (mp->mp_bh[x]) | |
387 | break; | |
d552a2b9 BP |
388 | } |
389 | } | |
c3ce5aa9 AG |
390 | ret = __fillup_metapath(ip, mp, x, h); |
391 | if (ret) | |
392 | return ret; | |
393 | return mp->mp_aheight - x - 1; | |
d552a2b9 BP |
394 | } |
395 | ||
64bc06bb | 396 | static void release_metapath(struct metapath *mp) |
dbac6710 SW |
397 | { |
398 | int i; | |
399 | ||
9b8c81d1 SW |
400 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
401 | if (mp->mp_bh[i] == NULL) | |
402 | break; | |
403 | brelse(mp->mp_bh[i]); | |
64bc06bb | 404 | mp->mp_bh[i] = NULL; |
9b8c81d1 | 405 | } |
11707ea0 SW |
406 | } |
407 | ||
30cbf189 SW |
408 | /** |
409 | * gfs2_extent_length - Returns length of an extent of blocks | |
bcfe9413 AG |
410 | * @bh: The metadata block |
411 | * @ptr: Current position in @bh | |
412 | * @limit: Max extent length to return | |
30cbf189 SW |
413 | * @eob: Set to 1 if we hit "end of block" |
414 | * | |
30cbf189 SW |
415 | * Returns: The length of the extent (minimum of one block) |
416 | */ | |
417 | ||
bcfe9413 | 418 | static inline unsigned int gfs2_extent_length(struct buffer_head *bh, __be64 *ptr, size_t limit, int *eob) |
30cbf189 | 419 | { |
bcfe9413 | 420 | const __be64 *end = (__be64 *)(bh->b_data + bh->b_size); |
30cbf189 SW |
421 | const __be64 *first = ptr; |
422 | u64 d = be64_to_cpu(*ptr); | |
423 | ||
424 | *eob = 0; | |
425 | do { | |
426 | ptr++; | |
427 | if (ptr >= end) | |
428 | break; | |
bcfe9413 | 429 | d++; |
30cbf189 SW |
430 | } while(be64_to_cpu(*ptr) == d); |
431 | if (ptr >= end) | |
432 | *eob = 1; | |
bcfe9413 | 433 | return ptr - first; |
30cbf189 SW |
434 | } |
435 | ||
7841b9f0 AG |
436 | typedef const __be64 *(*gfs2_metadata_walker)( |
437 | struct metapath *mp, | |
438 | const __be64 *start, const __be64 *end, | |
439 | u64 factor, void *data); | |
440 | ||
441 | #define WALK_STOP ((__be64 *)0) | |
442 | #define WALK_NEXT ((__be64 *)1) | |
443 | ||
444 | static int gfs2_walk_metadata(struct inode *inode, sector_t lblock, | |
445 | u64 len, struct metapath *mp, gfs2_metadata_walker walker, | |
446 | void *data) | |
447 | { | |
448 | struct metapath clone; | |
449 | struct gfs2_inode *ip = GFS2_I(inode); | |
450 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
451 | const __be64 *start, *end, *ptr; | |
452 | u64 factor = 1; | |
453 | unsigned int hgt; | |
454 | int ret = 0; | |
455 | ||
456 | for (hgt = ip->i_height - 1; hgt >= mp->mp_aheight; hgt--) | |
457 | factor *= sdp->sd_inptrs; | |
458 | ||
459 | for (;;) { | |
460 | u64 step; | |
461 | ||
462 | /* Walk indirect block. */ | |
463 | start = metapointer(hgt, mp); | |
464 | end = metaend(hgt, mp); | |
465 | ||
466 | step = (end - start) * factor; | |
467 | if (step > len) | |
468 | end = start + DIV_ROUND_UP_ULL(len, factor); | |
469 | ||
470 | ptr = walker(mp, start, end, factor, data); | |
471 | if (ptr == WALK_STOP) | |
472 | break; | |
473 | if (step >= len) | |
474 | break; | |
475 | len -= step; | |
476 | if (ptr != WALK_NEXT) { | |
477 | BUG_ON(!*ptr); | |
478 | mp->mp_list[hgt] += ptr - start; | |
479 | goto fill_up_metapath; | |
480 | } | |
481 | ||
482 | lower_metapath: | |
483 | /* Decrease height of metapath. */ | |
484 | if (mp != &clone) { | |
485 | clone_metapath(&clone, mp); | |
486 | mp = &clone; | |
487 | } | |
488 | brelse(mp->mp_bh[hgt]); | |
489 | mp->mp_bh[hgt] = NULL; | |
490 | if (!hgt) | |
491 | break; | |
492 | hgt--; | |
493 | factor *= sdp->sd_inptrs; | |
494 | ||
495 | /* Advance in metadata tree. */ | |
496 | (mp->mp_list[hgt])++; | |
497 | start = metapointer(hgt, mp); | |
498 | end = metaend(hgt, mp); | |
499 | if (start >= end) { | |
500 | mp->mp_list[hgt] = 0; | |
501 | if (!hgt) | |
502 | break; | |
503 | goto lower_metapath; | |
504 | } | |
505 | ||
506 | fill_up_metapath: | |
507 | /* Increase height of metapath. */ | |
508 | if (mp != &clone) { | |
509 | clone_metapath(&clone, mp); | |
510 | mp = &clone; | |
511 | } | |
512 | ret = fillup_metapath(ip, mp, ip->i_height - 1); | |
513 | if (ret < 0) | |
514 | break; | |
515 | hgt += ret; | |
516 | for (; ret; ret--) | |
517 | do_div(factor, sdp->sd_inptrs); | |
518 | mp->mp_aheight = hgt + 1; | |
519 | } | |
520 | if (mp == &clone) | |
521 | release_metapath(mp); | |
522 | return ret; | |
523 | } | |
524 | ||
525 | struct gfs2_hole_walker_args { | |
526 | u64 blocks; | |
527 | }; | |
528 | ||
529 | static const __be64 *gfs2_hole_walker(struct metapath *mp, | |
530 | const __be64 *start, const __be64 *end, | |
531 | u64 factor, void *data) | |
532 | { | |
533 | struct gfs2_hole_walker_args *args = data; | |
534 | const __be64 *ptr; | |
535 | ||
536 | for (ptr = start; ptr < end; ptr++) { | |
537 | if (*ptr) { | |
538 | args->blocks += (ptr - start) * factor; | |
539 | if (mp->mp_aheight == mp->mp_fheight) | |
540 | return WALK_STOP; | |
541 | return ptr; /* increase height */ | |
542 | } | |
543 | } | |
544 | args->blocks += (end - start) * factor; | |
545 | return WALK_NEXT; | |
546 | } | |
547 | ||
548 | /** | |
549 | * gfs2_hole_size - figure out the size of a hole | |
550 | * @inode: The inode | |
551 | * @lblock: The logical starting block number | |
552 | * @len: How far to look (in blocks) | |
553 | * @mp: The metapath at lblock | |
554 | * @iomap: The iomap to store the hole size in | |
555 | * | |
556 | * This function modifies @mp. | |
557 | * | |
558 | * Returns: errno on error | |
559 | */ | |
560 | static int gfs2_hole_size(struct inode *inode, sector_t lblock, u64 len, | |
561 | struct metapath *mp, struct iomap *iomap) | |
562 | { | |
563 | struct gfs2_hole_walker_args args = { }; | |
564 | int ret = 0; | |
565 | ||
566 | ret = gfs2_walk_metadata(inode, lblock, len, mp, gfs2_hole_walker, &args); | |
567 | if (!ret) | |
568 | iomap->length = args.blocks << inode->i_blkbits; | |
569 | return ret; | |
570 | } | |
571 | ||
9b8c81d1 SW |
572 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
573 | struct gfs2_glock *gl, unsigned int i, | |
574 | unsigned offset, u64 bn) | |
575 | { | |
576 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
577 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
578 | sizeof(struct gfs2_dinode))); | |
579 | BUG_ON(i < 1); | |
580 | BUG_ON(mp->mp_bh[i] != NULL); | |
581 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
350a9b0a | 582 | gfs2_trans_add_meta(gl, mp->mp_bh[i]); |
9b8c81d1 SW |
583 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
584 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
585 | ptr += offset; | |
586 | *ptr = cpu_to_be64(bn); | |
587 | return ptr; | |
588 | } | |
589 | ||
590 | enum alloc_state { | |
591 | ALLOC_DATA = 0, | |
592 | ALLOC_GROW_DEPTH = 1, | |
593 | ALLOC_GROW_HEIGHT = 2, | |
594 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
595 | }; | |
596 | ||
597 | /** | |
628e366d | 598 | * gfs2_iomap_alloc - Build a metadata tree of the requested height |
9b8c81d1 | 599 | * @inode: The GFS2 inode |
628e366d AG |
600 | * @iomap: The iomap structure |
601 | * @flags: iomap flags | |
5f8bd444 | 602 | * @mp: The metapath, with proper height information calculated |
9b8c81d1 SW |
603 | * |
604 | * In this routine we may have to alloc: | |
605 | * i) Indirect blocks to grow the metadata tree height | |
606 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
607 | * iii) Data blocks | |
608 | * | |
64bc06bb AG |
609 | * This function is called after gfs2_iomap_get, which works out the |
610 | * total number of blocks which we need via gfs2_alloc_size. | |
611 | * | |
612 | * We then do the actual allocation asking for an extent at a time (if | |
613 | * enough contiguous free blocks are available, there will only be one | |
614 | * allocation request per call) and uses the state machine to initialise | |
615 | * the blocks in order. | |
9b8c81d1 | 616 | * |
628e366d AG |
617 | * Right now, this function will allocate at most one indirect block |
618 | * worth of data -- with a default block size of 4K, that's slightly | |
619 | * less than 2M. If this limitation is ever removed to allow huge | |
620 | * allocations, we would probably still want to limit the iomap size we | |
621 | * return to avoid stalling other tasks during huge writes; the next | |
622 | * iomap iteration would then find the blocks already allocated. | |
623 | * | |
9b8c81d1 SW |
624 | * Returns: errno on error |
625 | */ | |
626 | ||
3974320c BP |
627 | static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap, |
628 | unsigned flags, struct metapath *mp) | |
9b8c81d1 SW |
629 | { |
630 | struct gfs2_inode *ip = GFS2_I(inode); | |
631 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
632 | struct buffer_head *dibh = mp->mp_bh[0]; | |
5f8bd444 | 633 | u64 bn; |
5af4e7a0 | 634 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
64bc06bb | 635 | size_t dblks = iomap->length >> inode->i_blkbits; |
5f8bd444 | 636 | const unsigned end_of_metadata = mp->mp_fheight - 1; |
628e366d | 637 | int ret; |
9b8c81d1 SW |
638 | enum alloc_state state; |
639 | __be64 *ptr; | |
640 | __be64 zero_bn = 0; | |
641 | ||
5f8bd444 | 642 | BUG_ON(mp->mp_aheight < 1); |
9b8c81d1 | 643 | BUG_ON(dibh == NULL); |
64bc06bb | 644 | BUG_ON(dblks < 1); |
9b8c81d1 | 645 | |
350a9b0a | 646 | gfs2_trans_add_meta(ip->i_gl, dibh); |
9b8c81d1 | 647 | |
628e366d AG |
648 | down_write(&ip->i_rw_mutex); |
649 | ||
5f8bd444 | 650 | if (mp->mp_fheight == mp->mp_aheight) { |
64bc06bb | 651 | /* Bottom indirect block exists */ |
9b8c81d1 SW |
652 | state = ALLOC_DATA; |
653 | } else { | |
654 | /* Need to allocate indirect blocks */ | |
5f8bd444 | 655 | if (mp->mp_fheight == ip->i_height) { |
9b8c81d1 | 656 | /* Writing into existing tree, extend tree down */ |
5f8bd444 | 657 | iblks = mp->mp_fheight - mp->mp_aheight; |
9b8c81d1 SW |
658 | state = ALLOC_GROW_DEPTH; |
659 | } else { | |
660 | /* Building up tree height */ | |
661 | state = ALLOC_GROW_HEIGHT; | |
5f8bd444 | 662 | iblks = mp->mp_fheight - ip->i_height; |
5af4e7a0 | 663 | branch_start = metapath_branch_start(mp); |
5f8bd444 | 664 | iblks += (mp->mp_fheight - branch_start); |
9b8c81d1 SW |
665 | } |
666 | } | |
667 | ||
668 | /* start of the second part of the function (state machine) */ | |
669 | ||
3974320c | 670 | blks = dblks + iblks; |
5f8bd444 | 671 | i = mp->mp_aheight; |
9b8c81d1 SW |
672 | do { |
673 | n = blks - alloced; | |
628e366d AG |
674 | ret = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
675 | if (ret) | |
676 | goto out; | |
9b8c81d1 SW |
677 | alloced += n; |
678 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
fbb27873 | 679 | gfs2_trans_remove_revoke(sdp, bn, n); |
9b8c81d1 SW |
680 | switch (state) { |
681 | /* Growing height of tree */ | |
682 | case ALLOC_GROW_HEIGHT: | |
683 | if (i == 1) { | |
684 | ptr = (__be64 *)(dibh->b_data + | |
685 | sizeof(struct gfs2_dinode)); | |
686 | zero_bn = *ptr; | |
687 | } | |
5f8bd444 BP |
688 | for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0; |
689 | i++, n--) | |
9b8c81d1 | 690 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); |
5f8bd444 | 691 | if (i - 1 == mp->mp_fheight - ip->i_height) { |
9b8c81d1 SW |
692 | i--; |
693 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
694 | sizeof(struct gfs2_meta_header), | |
695 | dibh, sizeof(struct gfs2_dinode)); | |
696 | gfs2_buffer_clear_tail(dibh, | |
697 | sizeof(struct gfs2_dinode) + | |
698 | sizeof(__be64)); | |
699 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
700 | sizeof(struct gfs2_meta_header)); | |
701 | *ptr = zero_bn; | |
702 | state = ALLOC_GROW_DEPTH; | |
5f8bd444 | 703 | for(i = branch_start; i < mp->mp_fheight; i++) { |
9b8c81d1 SW |
704 | if (mp->mp_bh[i] == NULL) |
705 | break; | |
706 | brelse(mp->mp_bh[i]); | |
707 | mp->mp_bh[i] = NULL; | |
708 | } | |
5af4e7a0 | 709 | i = branch_start; |
9b8c81d1 SW |
710 | } |
711 | if (n == 0) | |
712 | break; | |
0a4c9265 | 713 | /* fall through - To branching from existing tree */ |
9b8c81d1 | 714 | case ALLOC_GROW_DEPTH: |
5f8bd444 | 715 | if (i > 1 && i < mp->mp_fheight) |
350a9b0a | 716 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]); |
5f8bd444 | 717 | for (; i < mp->mp_fheight && n > 0; i++, n--) |
9b8c81d1 SW |
718 | gfs2_indirect_init(mp, ip->i_gl, i, |
719 | mp->mp_list[i-1], bn++); | |
5f8bd444 | 720 | if (i == mp->mp_fheight) |
9b8c81d1 SW |
721 | state = ALLOC_DATA; |
722 | if (n == 0) | |
723 | break; | |
0a4c9265 | 724 | /* fall through - To tree complete, adding data blocks */ |
9b8c81d1 | 725 | case ALLOC_DATA: |
3974320c | 726 | BUG_ON(n > dblks); |
9b8c81d1 | 727 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); |
350a9b0a | 728 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]); |
3974320c | 729 | dblks = n; |
9b8c81d1 | 730 | ptr = metapointer(end_of_metadata, mp); |
3974320c | 731 | iomap->addr = bn << inode->i_blkbits; |
628e366d | 732 | iomap->flags |= IOMAP_F_MERGED | IOMAP_F_NEW; |
9b8c81d1 SW |
733 | while (n-- > 0) |
734 | *ptr++ = cpu_to_be64(bn++); | |
735 | break; | |
736 | } | |
3974320c | 737 | } while (iomap->addr == IOMAP_NULL_ADDR); |
9b8c81d1 | 738 | |
d505a96a | 739 | iomap->type = IOMAP_MAPPED; |
3974320c | 740 | iomap->length = (u64)dblks << inode->i_blkbits; |
5f8bd444 | 741 | ip->i_height = mp->mp_fheight; |
9b8c81d1 | 742 | gfs2_add_inode_blocks(&ip->i_inode, alloced); |
628e366d AG |
743 | gfs2_dinode_out(ip, dibh->b_data); |
744 | out: | |
745 | up_write(&ip->i_rw_mutex); | |
746 | return ret; | |
9b8c81d1 SW |
747 | } |
748 | ||
7ee66c03 CH |
749 | #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE |
750 | ||
64bc06bb AG |
751 | /** |
752 | * gfs2_alloc_size - Compute the maximum allocation size | |
753 | * @inode: The inode | |
754 | * @mp: The metapath | |
755 | * @size: Requested size in blocks | |
756 | * | |
757 | * Compute the maximum size of the next allocation at @mp. | |
758 | * | |
759 | * Returns: size in blocks | |
760 | */ | |
761 | static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size) | |
3974320c BP |
762 | { |
763 | struct gfs2_inode *ip = GFS2_I(inode); | |
64bc06bb AG |
764 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
765 | const __be64 *first, *ptr, *end; | |
766 | ||
767 | /* | |
768 | * For writes to stuffed files, this function is called twice via | |
769 | * gfs2_iomap_get, before and after unstuffing. The size we return the | |
770 | * first time needs to be large enough to get the reservation and | |
771 | * allocation sizes right. The size we return the second time must | |
772 | * be exact or else gfs2_iomap_alloc won't do the right thing. | |
773 | */ | |
774 | ||
775 | if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) { | |
776 | unsigned int maxsize = mp->mp_fheight > 1 ? | |
777 | sdp->sd_inptrs : sdp->sd_diptrs; | |
778 | maxsize -= mp->mp_list[mp->mp_fheight - 1]; | |
779 | if (size > maxsize) | |
780 | size = maxsize; | |
781 | return size; | |
782 | } | |
3974320c | 783 | |
64bc06bb AG |
784 | first = metapointer(ip->i_height - 1, mp); |
785 | end = metaend(ip->i_height - 1, mp); | |
786 | if (end - first > size) | |
787 | end = first + size; | |
788 | for (ptr = first; ptr < end; ptr++) { | |
789 | if (*ptr) | |
790 | break; | |
791 | } | |
792 | return ptr - first; | |
3974320c BP |
793 | } |
794 | ||
795 | /** | |
628e366d | 796 | * gfs2_iomap_get - Map blocks from an inode to disk blocks |
3974320c BP |
797 | * @inode: The inode |
798 | * @pos: Starting position in bytes | |
799 | * @length: Length to map, in bytes | |
800 | * @flags: iomap flags | |
801 | * @iomap: The iomap structure | |
628e366d | 802 | * @mp: The metapath |
3974320c BP |
803 | * |
804 | * Returns: errno | |
805 | */ | |
628e366d AG |
806 | static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length, |
807 | unsigned flags, struct iomap *iomap, | |
808 | struct metapath *mp) | |
b3b94faa | 809 | { |
feaa7bba SW |
810 | struct gfs2_inode *ip = GFS2_I(inode); |
811 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
d505a96a | 812 | loff_t size = i_size_read(inode); |
9b8c81d1 | 813 | __be64 *ptr; |
3974320c | 814 | sector_t lblock; |
628e366d AG |
815 | sector_t lblock_stop; |
816 | int ret; | |
9b8c81d1 | 817 | int eob; |
628e366d | 818 | u64 len; |
d505a96a | 819 | struct buffer_head *dibh = NULL, *bh; |
9b8c81d1 | 820 | u8 height; |
7276b3b0 | 821 | |
628e366d AG |
822 | if (!length) |
823 | return -EINVAL; | |
b3b94faa | 824 | |
d505a96a AG |
825 | down_read(&ip->i_rw_mutex); |
826 | ||
827 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
828 | if (ret) | |
829 | goto unlock; | |
c26b5aa8 | 830 | mp->mp_bh[0] = dibh; |
d505a96a | 831 | |
49edd5bf | 832 | if (gfs2_is_stuffed(ip)) { |
d505a96a AG |
833 | if (flags & IOMAP_WRITE) { |
834 | loff_t max_size = gfs2_max_stuffed_size(ip); | |
835 | ||
836 | if (pos + length > max_size) | |
837 | goto unstuff; | |
838 | iomap->length = max_size; | |
839 | } else { | |
840 | if (pos >= size) { | |
841 | if (flags & IOMAP_REPORT) { | |
842 | ret = -ENOENT; | |
843 | goto unlock; | |
844 | } else { | |
845 | /* report a hole */ | |
846 | iomap->offset = pos; | |
847 | iomap->length = length; | |
848 | goto do_alloc; | |
849 | } | |
850 | } | |
851 | iomap->length = size; | |
49edd5bf | 852 | } |
d505a96a AG |
853 | iomap->addr = (ip->i_no_addr << inode->i_blkbits) + |
854 | sizeof(struct gfs2_dinode); | |
855 | iomap->type = IOMAP_INLINE; | |
64bc06bb | 856 | iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode); |
d505a96a | 857 | goto out; |
3974320c | 858 | } |
d505a96a AG |
859 | |
860 | unstuff: | |
3974320c | 861 | lblock = pos >> inode->i_blkbits; |
3974320c | 862 | iomap->offset = lblock << inode->i_blkbits; |
628e366d AG |
863 | lblock_stop = (pos + length - 1) >> inode->i_blkbits; |
864 | len = lblock_stop - lblock + 1; | |
d505a96a | 865 | iomap->length = len << inode->i_blkbits; |
628e366d | 866 | |
9b8c81d1 | 867 | height = ip->i_height; |
9a38662b | 868 | while ((lblock + 1) * sdp->sd_sb.sb_bsize > sdp->sd_heightsize[height]) |
9b8c81d1 | 869 | height++; |
628e366d | 870 | find_metapath(sdp, lblock, mp, height); |
9b8c81d1 SW |
871 | if (height > ip->i_height || gfs2_is_stuffed(ip)) |
872 | goto do_alloc; | |
3974320c | 873 | |
628e366d | 874 | ret = lookup_metapath(ip, mp); |
e8b43fe0 | 875 | if (ret) |
628e366d | 876 | goto unlock; |
3974320c | 877 | |
628e366d | 878 | if (mp->mp_aheight != ip->i_height) |
9b8c81d1 | 879 | goto do_alloc; |
3974320c | 880 | |
628e366d | 881 | ptr = metapointer(ip->i_height - 1, mp); |
9b8c81d1 SW |
882 | if (*ptr == 0) |
883 | goto do_alloc; | |
3974320c | 884 | |
628e366d | 885 | bh = mp->mp_bh[ip->i_height - 1]; |
bcfe9413 | 886 | len = gfs2_extent_length(bh, ptr, len, &eob); |
3974320c | 887 | |
628e366d AG |
888 | iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits; |
889 | iomap->length = len << inode->i_blkbits; | |
890 | iomap->type = IOMAP_MAPPED; | |
0ed91eca | 891 | iomap->flags |= IOMAP_F_MERGED; |
9b8c81d1 | 892 | if (eob) |
7ee66c03 | 893 | iomap->flags |= IOMAP_F_GFS2_BOUNDARY; |
3974320c | 894 | |
3974320c | 895 | out: |
628e366d AG |
896 | iomap->bdev = inode->i_sb->s_bdev; |
897 | unlock: | |
898 | up_read(&ip->i_rw_mutex); | |
9b8c81d1 | 899 | return ret; |
30cbf189 | 900 | |
9b8c81d1 | 901 | do_alloc: |
628e366d | 902 | iomap->addr = IOMAP_NULL_ADDR; |
628e366d | 903 | iomap->type = IOMAP_HOLE; |
628e366d | 904 | if (flags & IOMAP_REPORT) { |
49edd5bf | 905 | if (pos >= size) |
3974320c | 906 | ret = -ENOENT; |
628e366d AG |
907 | else if (height == ip->i_height) |
908 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
49edd5bf AG |
909 | else |
910 | iomap->length = size - pos; | |
64bc06bb AG |
911 | } else if (flags & IOMAP_WRITE) { |
912 | u64 alloc_size; | |
913 | ||
967bcc91 AG |
914 | if (flags & IOMAP_DIRECT) |
915 | goto out; /* (see gfs2_file_direct_write) */ | |
916 | ||
64bc06bb AG |
917 | len = gfs2_alloc_size(inode, mp, len); |
918 | alloc_size = len << inode->i_blkbits; | |
919 | if (alloc_size < iomap->length) | |
920 | iomap->length = alloc_size; | |
921 | } else { | |
d505a96a AG |
922 | if (pos < size && height == ip->i_height) |
923 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
b3b94faa | 924 | } |
628e366d | 925 | goto out; |
3974320c BP |
926 | } |
927 | ||
7c70b896 BP |
928 | /** |
929 | * gfs2_lblk_to_dblk - convert logical block to disk block | |
930 | * @inode: the inode of the file we're mapping | |
931 | * @lblock: the block relative to the start of the file | |
932 | * @dblock: the returned dblock, if no error | |
933 | * | |
934 | * This function maps a single block from a file logical block (relative to | |
935 | * the start of the file) to a file system absolute block using iomap. | |
936 | * | |
937 | * Returns: the absolute file system block, or an error | |
938 | */ | |
939 | int gfs2_lblk_to_dblk(struct inode *inode, u32 lblock, u64 *dblock) | |
940 | { | |
941 | struct iomap iomap = { }; | |
942 | struct metapath mp = { .mp_aheight = 1, }; | |
943 | loff_t pos = (loff_t)lblock << inode->i_blkbits; | |
944 | int ret; | |
945 | ||
946 | ret = gfs2_iomap_get(inode, pos, i_blocksize(inode), 0, &iomap, &mp); | |
947 | release_metapath(&mp); | |
948 | if (ret == 0) | |
949 | *dblock = iomap.addr >> inode->i_blkbits; | |
950 | ||
951 | return ret; | |
952 | } | |
953 | ||
64bc06bb AG |
954 | static int gfs2_write_lock(struct inode *inode) |
955 | { | |
956 | struct gfs2_inode *ip = GFS2_I(inode); | |
957 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
958 | int error; | |
959 | ||
960 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); | |
961 | error = gfs2_glock_nq(&ip->i_gh); | |
962 | if (error) | |
963 | goto out_uninit; | |
964 | if (&ip->i_inode == sdp->sd_rindex) { | |
965 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
966 | ||
967 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
968 | GL_NOCACHE, &m_ip->i_gh); | |
969 | if (error) | |
970 | goto out_unlock; | |
971 | } | |
972 | return 0; | |
973 | ||
974 | out_unlock: | |
975 | gfs2_glock_dq(&ip->i_gh); | |
976 | out_uninit: | |
977 | gfs2_holder_uninit(&ip->i_gh); | |
978 | return error; | |
979 | } | |
980 | ||
981 | static void gfs2_write_unlock(struct inode *inode) | |
982 | { | |
983 | struct gfs2_inode *ip = GFS2_I(inode); | |
984 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
985 | ||
986 | if (&ip->i_inode == sdp->sd_rindex) { | |
987 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
988 | ||
989 | gfs2_glock_dq_uninit(&m_ip->i_gh); | |
990 | } | |
991 | gfs2_glock_dq_uninit(&ip->i_gh); | |
992 | } | |
993 | ||
d0a22a4b AG |
994 | static int gfs2_iomap_page_prepare(struct inode *inode, loff_t pos, |
995 | unsigned len, struct iomap *iomap) | |
996 | { | |
997 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
998 | ||
999 | return gfs2_trans_begin(sdp, RES_DINODE + (len >> inode->i_blkbits), 0); | |
1000 | } | |
1001 | ||
df0db3ec AG |
1002 | static void gfs2_iomap_page_done(struct inode *inode, loff_t pos, |
1003 | unsigned copied, struct page *page, | |
1004 | struct iomap *iomap) | |
64bc06bb AG |
1005 | { |
1006 | struct gfs2_inode *ip = GFS2_I(inode); | |
d0a22a4b | 1007 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
64bc06bb | 1008 | |
d0a22a4b | 1009 | if (page && !gfs2_is_stuffed(ip)) |
df0db3ec | 1010 | gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied); |
d0a22a4b | 1011 | gfs2_trans_end(sdp); |
64bc06bb AG |
1012 | } |
1013 | ||
df0db3ec | 1014 | static const struct iomap_page_ops gfs2_iomap_page_ops = { |
d0a22a4b | 1015 | .page_prepare = gfs2_iomap_page_prepare, |
df0db3ec AG |
1016 | .page_done = gfs2_iomap_page_done, |
1017 | }; | |
1018 | ||
64bc06bb AG |
1019 | static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, |
1020 | loff_t length, unsigned flags, | |
c26b5aa8 AG |
1021 | struct iomap *iomap, |
1022 | struct metapath *mp) | |
64bc06bb | 1023 | { |
64bc06bb AG |
1024 | struct gfs2_inode *ip = GFS2_I(inode); |
1025 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1026 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; | |
1027 | bool unstuff, alloc_required; | |
1028 | int ret; | |
1029 | ||
1030 | ret = gfs2_write_lock(inode); | |
1031 | if (ret) | |
1032 | return ret; | |
1033 | ||
1034 | unstuff = gfs2_is_stuffed(ip) && | |
1035 | pos + length > gfs2_max_stuffed_size(ip); | |
1036 | ||
c26b5aa8 | 1037 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, mp); |
64bc06bb | 1038 | if (ret) |
c26b5aa8 | 1039 | goto out_unlock; |
64bc06bb AG |
1040 | |
1041 | alloc_required = unstuff || iomap->type == IOMAP_HOLE; | |
1042 | ||
1043 | if (alloc_required || gfs2_is_jdata(ip)) | |
1044 | gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, | |
1045 | &ind_blocks); | |
1046 | ||
1047 | if (alloc_required) { | |
1048 | struct gfs2_alloc_parms ap = { | |
1049 | .target = data_blocks + ind_blocks | |
1050 | }; | |
1051 | ||
1052 | ret = gfs2_quota_lock_check(ip, &ap); | |
1053 | if (ret) | |
c26b5aa8 | 1054 | goto out_unlock; |
64bc06bb AG |
1055 | |
1056 | ret = gfs2_inplace_reserve(ip, &ap); | |
1057 | if (ret) | |
1058 | goto out_qunlock; | |
1059 | } | |
1060 | ||
1061 | rblocks = RES_DINODE + ind_blocks; | |
1062 | if (gfs2_is_jdata(ip)) | |
1063 | rblocks += data_blocks; | |
1064 | if (ind_blocks || data_blocks) | |
1065 | rblocks += RES_STATFS + RES_QUOTA; | |
1066 | if (inode == sdp->sd_rindex) | |
1067 | rblocks += 2 * RES_STATFS; | |
1068 | if (alloc_required) | |
1069 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); | |
1070 | ||
d0a22a4b AG |
1071 | if (unstuff || iomap->type == IOMAP_HOLE) { |
1072 | struct gfs2_trans *tr; | |
64bc06bb | 1073 | |
d0a22a4b AG |
1074 | ret = gfs2_trans_begin(sdp, rblocks, |
1075 | iomap->length >> inode->i_blkbits); | |
64bc06bb | 1076 | if (ret) |
d0a22a4b AG |
1077 | goto out_trans_fail; |
1078 | ||
1079 | if (unstuff) { | |
1080 | ret = gfs2_unstuff_dinode(ip, NULL); | |
1081 | if (ret) | |
1082 | goto out_trans_end; | |
1083 | release_metapath(mp); | |
1084 | ret = gfs2_iomap_get(inode, iomap->offset, | |
1085 | iomap->length, flags, iomap, mp); | |
1086 | if (ret) | |
1087 | goto out_trans_end; | |
1088 | } | |
64bc06bb | 1089 | |
d0a22a4b AG |
1090 | if (iomap->type == IOMAP_HOLE) { |
1091 | ret = gfs2_iomap_alloc(inode, iomap, flags, mp); | |
1092 | if (ret) { | |
1093 | gfs2_trans_end(sdp); | |
1094 | gfs2_inplace_release(ip); | |
1095 | punch_hole(ip, iomap->offset, iomap->length); | |
1096 | goto out_qunlock; | |
1097 | } | |
64bc06bb | 1098 | } |
d0a22a4b AG |
1099 | |
1100 | tr = current->journal_info; | |
1101 | if (tr->tr_num_buf_new) | |
1102 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1103 | else | |
1104 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[0]); | |
1105 | ||
1106 | gfs2_trans_end(sdp); | |
64bc06bb | 1107 | } |
d0a22a4b AG |
1108 | |
1109 | if (gfs2_is_stuffed(ip) || gfs2_is_jdata(ip)) | |
df0db3ec | 1110 | iomap->page_ops = &gfs2_iomap_page_ops; |
64bc06bb AG |
1111 | return 0; |
1112 | ||
1113 | out_trans_end: | |
1114 | gfs2_trans_end(sdp); | |
1115 | out_trans_fail: | |
1116 | if (alloc_required) | |
1117 | gfs2_inplace_release(ip); | |
1118 | out_qunlock: | |
1119 | if (alloc_required) | |
1120 | gfs2_quota_unlock(ip); | |
c26b5aa8 | 1121 | out_unlock: |
64bc06bb AG |
1122 | gfs2_write_unlock(inode); |
1123 | return ret; | |
1124 | } | |
1125 | ||
628e366d AG |
1126 | static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, |
1127 | unsigned flags, struct iomap *iomap) | |
1128 | { | |
1129 | struct gfs2_inode *ip = GFS2_I(inode); | |
1130 | struct metapath mp = { .mp_aheight = 1, }; | |
1131 | int ret; | |
1132 | ||
0ed91eca AG |
1133 | iomap->flags |= IOMAP_F_BUFFER_HEAD; |
1134 | ||
628e366d | 1135 | trace_gfs2_iomap_start(ip, pos, length, flags); |
967bcc91 | 1136 | if ((flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT)) { |
c26b5aa8 | 1137 | ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap, &mp); |
628e366d AG |
1138 | } else { |
1139 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); | |
c26b5aa8 | 1140 | |
967bcc91 AG |
1141 | /* |
1142 | * Silently fall back to buffered I/O for stuffed files or if | |
1143 | * we've hot a hole (see gfs2_file_direct_write). | |
1144 | */ | |
1145 | if ((flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT) && | |
1146 | iomap->type != IOMAP_MAPPED) | |
1147 | ret = -ENOTBLK; | |
628e366d | 1148 | } |
c26b5aa8 | 1149 | release_metapath(&mp); |
628e366d AG |
1150 | trace_gfs2_iomap_end(ip, iomap, ret); |
1151 | return ret; | |
1152 | } | |
1153 | ||
64bc06bb AG |
1154 | static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length, |
1155 | ssize_t written, unsigned flags, struct iomap *iomap) | |
1156 | { | |
1157 | struct gfs2_inode *ip = GFS2_I(inode); | |
1158 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
64bc06bb | 1159 | |
967bcc91 | 1160 | if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) != IOMAP_WRITE) |
64bc06bb AG |
1161 | goto out; |
1162 | ||
d0a22a4b | 1163 | if (!gfs2_is_stuffed(ip)) |
64bc06bb AG |
1164 | gfs2_ordered_add_inode(ip); |
1165 | ||
d0a22a4b | 1166 | if (inode == sdp->sd_rindex) |
64bc06bb | 1167 | adjust_fs_space(inode); |
64bc06bb | 1168 | |
64bc06bb AG |
1169 | gfs2_inplace_release(ip); |
1170 | ||
1171 | if (length != written && (iomap->flags & IOMAP_F_NEW)) { | |
1172 | /* Deallocate blocks that were just allocated. */ | |
1173 | loff_t blockmask = i_blocksize(inode) - 1; | |
1174 | loff_t end = (pos + length) & ~blockmask; | |
1175 | ||
1176 | pos = (pos + written + blockmask) & ~blockmask; | |
1177 | if (pos < end) { | |
1178 | truncate_pagecache_range(inode, pos, end - 1); | |
1179 | punch_hole(ip, pos, end - pos); | |
1180 | } | |
1181 | } | |
1182 | ||
1183 | if (ip->i_qadata && ip->i_qadata->qa_qd_num) | |
1184 | gfs2_quota_unlock(ip); | |
1185 | gfs2_write_unlock(inode); | |
1186 | ||
1187 | out: | |
64bc06bb AG |
1188 | return 0; |
1189 | } | |
1190 | ||
628e366d AG |
1191 | const struct iomap_ops gfs2_iomap_ops = { |
1192 | .iomap_begin = gfs2_iomap_begin, | |
64bc06bb | 1193 | .iomap_end = gfs2_iomap_end, |
628e366d AG |
1194 | }; |
1195 | ||
3974320c | 1196 | /** |
d39d18e0 | 1197 | * gfs2_block_map - Map one or more blocks of an inode to a disk block |
3974320c BP |
1198 | * @inode: The inode |
1199 | * @lblock: The logical block number | |
1200 | * @bh_map: The bh to be mapped | |
1201 | * @create: True if its ok to alloc blocks to satify the request | |
1202 | * | |
d39d18e0 AG |
1203 | * The size of the requested mapping is defined in bh_map->b_size. |
1204 | * | |
1205 | * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged | |
1206 | * when @lblock is not mapped. Sets buffer_mapped(bh_map) and | |
1207 | * bh_map->b_size to indicate the size of the mapping when @lblock and | |
1208 | * successive blocks are mapped, up to the requested size. | |
1209 | * | |
1210 | * Sets buffer_boundary() if a read of metadata will be required | |
1211 | * before the next block can be mapped. Sets buffer_new() if new | |
1212 | * blocks were allocated. | |
3974320c BP |
1213 | * |
1214 | * Returns: errno | |
1215 | */ | |
1216 | ||
1217 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
1218 | struct buffer_head *bh_map, int create) | |
1219 | { | |
1220 | struct gfs2_inode *ip = GFS2_I(inode); | |
628e366d AG |
1221 | loff_t pos = (loff_t)lblock << inode->i_blkbits; |
1222 | loff_t length = bh_map->b_size; | |
1223 | struct metapath mp = { .mp_aheight = 1, }; | |
1224 | struct iomap iomap = { }; | |
1225 | int ret; | |
3974320c BP |
1226 | |
1227 | clear_buffer_mapped(bh_map); | |
1228 | clear_buffer_new(bh_map); | |
1229 | clear_buffer_boundary(bh_map); | |
1230 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
1231 | ||
628e366d AG |
1232 | if (create) { |
1233 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, &iomap, &mp); | |
1234 | if (!ret && iomap.type == IOMAP_HOLE) | |
1235 | ret = gfs2_iomap_alloc(inode, &iomap, IOMAP_WRITE, &mp); | |
1236 | release_metapath(&mp); | |
1237 | } else { | |
1238 | ret = gfs2_iomap_get(inode, pos, length, 0, &iomap, &mp); | |
1239 | release_metapath(&mp); | |
3974320c | 1240 | } |
628e366d AG |
1241 | if (ret) |
1242 | goto out; | |
3974320c BP |
1243 | |
1244 | if (iomap.length > bh_map->b_size) { | |
1245 | iomap.length = bh_map->b_size; | |
7ee66c03 | 1246 | iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY; |
5f8bd444 | 1247 | } |
3974320c BP |
1248 | if (iomap.addr != IOMAP_NULL_ADDR) |
1249 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
1250 | bh_map->b_size = iomap.length; | |
7ee66c03 | 1251 | if (iomap.flags & IOMAP_F_GFS2_BOUNDARY) |
3974320c BP |
1252 | set_buffer_boundary(bh_map); |
1253 | if (iomap.flags & IOMAP_F_NEW) | |
1254 | set_buffer_new(bh_map); | |
1255 | ||
1256 | out: | |
1257 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
1258 | return ret; | |
fd88de56 SW |
1259 | } |
1260 | ||
941e6d7d SW |
1261 | /* |
1262 | * Deprecated: do not use in new code | |
1263 | */ | |
fd88de56 SW |
1264 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
1265 | { | |
23591256 | 1266 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 1267 | int ret; |
fd88de56 SW |
1268 | int create = *new; |
1269 | ||
1270 | BUG_ON(!extlen); | |
1271 | BUG_ON(!dblock); | |
1272 | BUG_ON(!new); | |
1273 | ||
47a9a527 | 1274 | bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 1275 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
1276 | *extlen = bh.b_size >> inode->i_blkbits; |
1277 | *dblock = bh.b_blocknr; | |
1278 | if (buffer_new(&bh)) | |
1279 | *new = 1; | |
1280 | else | |
1281 | *new = 0; | |
1282 | return ret; | |
b3b94faa DT |
1283 | } |
1284 | ||
ba7f7290 | 1285 | /** |
bdba0d5e | 1286 | * gfs2_block_zero_range - Deal with zeroing out data |
ba7f7290 SW |
1287 | * |
1288 | * This is partly borrowed from ext3. | |
1289 | */ | |
bdba0d5e AG |
1290 | static int gfs2_block_zero_range(struct inode *inode, loff_t from, |
1291 | unsigned int length) | |
ba7f7290 | 1292 | { |
bdba0d5e | 1293 | struct address_space *mapping = inode->i_mapping; |
ba7f7290 | 1294 | struct gfs2_inode *ip = GFS2_I(inode); |
09cbfeaf KS |
1295 | unsigned long index = from >> PAGE_SHIFT; |
1296 | unsigned offset = from & (PAGE_SIZE-1); | |
bdba0d5e | 1297 | unsigned blocksize, iblock, pos; |
ba7f7290 SW |
1298 | struct buffer_head *bh; |
1299 | struct page *page; | |
ba7f7290 SW |
1300 | int err; |
1301 | ||
220cca2a | 1302 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ba7f7290 SW |
1303 | if (!page) |
1304 | return 0; | |
1305 | ||
1306 | blocksize = inode->i_sb->s_blocksize; | |
09cbfeaf | 1307 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
ba7f7290 SW |
1308 | |
1309 | if (!page_has_buffers(page)) | |
1310 | create_empty_buffers(page, blocksize, 0); | |
1311 | ||
1312 | /* Find the buffer that contains "offset" */ | |
1313 | bh = page_buffers(page); | |
1314 | pos = blocksize; | |
1315 | while (offset >= pos) { | |
1316 | bh = bh->b_this_page; | |
1317 | iblock++; | |
1318 | pos += blocksize; | |
1319 | } | |
1320 | ||
1321 | err = 0; | |
1322 | ||
1323 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 1324 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
1325 | /* unmapped? It's a hole - nothing to do */ |
1326 | if (!buffer_mapped(bh)) | |
1327 | goto unlock; | |
1328 | } | |
1329 | ||
1330 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1331 | if (PageUptodate(page)) | |
1332 | set_buffer_uptodate(bh); | |
1333 | ||
1334 | if (!buffer_uptodate(bh)) { | |
1335 | err = -EIO; | |
dfec8a14 | 1336 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
ba7f7290 SW |
1337 | wait_on_buffer(bh); |
1338 | /* Uhhuh. Read error. Complain and punt. */ | |
1339 | if (!buffer_uptodate(bh)) | |
1340 | goto unlock; | |
1875f2f3 | 1341 | err = 0; |
ba7f7290 SW |
1342 | } |
1343 | ||
845802b1 | 1344 | if (gfs2_is_jdata(ip)) |
350a9b0a | 1345 | gfs2_trans_add_data(ip->i_gl, bh); |
845802b1 AG |
1346 | else |
1347 | gfs2_ordered_add_inode(ip); | |
ba7f7290 | 1348 | |
eebd2aa3 | 1349 | zero_user(page, offset, length); |
40bc9a27 | 1350 | mark_buffer_dirty(bh); |
ba7f7290 SW |
1351 | unlock: |
1352 | unlock_page(page); | |
09cbfeaf | 1353 | put_page(page); |
ba7f7290 SW |
1354 | return err; |
1355 | } | |
1356 | ||
c62baf65 FF |
1357 | #define GFS2_JTRUNC_REVOKES 8192 |
1358 | ||
fa731fc4 SW |
1359 | /** |
1360 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
1361 | * @inode: The inode being truncated | |
1362 | * @oldsize: The original (larger) size | |
1363 | * @newsize: The new smaller size | |
1364 | * | |
1365 | * With jdata files, we have to journal a revoke for each block which is | |
1366 | * truncated. As a result, we need to split this into separate transactions | |
1367 | * if the number of pages being truncated gets too large. | |
1368 | */ | |
1369 | ||
fa731fc4 SW |
1370 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
1371 | { | |
1372 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1373 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1374 | u64 chunk; | |
1375 | int error; | |
1376 | ||
1377 | while (oldsize != newsize) { | |
e7fdf004 AG |
1378 | struct gfs2_trans *tr; |
1379 | unsigned int offs; | |
1380 | ||
fa731fc4 SW |
1381 | chunk = oldsize - newsize; |
1382 | if (chunk > max_chunk) | |
1383 | chunk = max_chunk; | |
e7fdf004 AG |
1384 | |
1385 | offs = oldsize & ~PAGE_MASK; | |
1386 | if (offs && chunk > PAGE_SIZE) | |
1387 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1388 | ||
7caef267 | 1389 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1390 | oldsize -= chunk; |
e7fdf004 AG |
1391 | |
1392 | tr = current->journal_info; | |
1393 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1394 | continue; | |
1395 | ||
fa731fc4 SW |
1396 | gfs2_trans_end(sdp); |
1397 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1398 | if (error) | |
1399 | return error; | |
1400 | } | |
1401 | ||
1402 | return 0; | |
1403 | } | |
1404 | ||
8b5860a3 | 1405 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1406 | { |
ff8f33c8 SW |
1407 | struct gfs2_inode *ip = GFS2_I(inode); |
1408 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
80990f40 | 1409 | struct buffer_head *dibh = NULL; |
b3b94faa | 1410 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1411 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1412 | int error; |
1413 | ||
fa731fc4 SW |
1414 | if (journaled) |
1415 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1416 | else | |
1417 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1418 | if (error) |
1419 | return error; | |
1420 | ||
1421 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1422 | if (error) | |
1423 | goto out; | |
1424 | ||
350a9b0a | 1425 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1426 | |
b3b94faa | 1427 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1428 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1429 | } else { |
bdba0d5e AG |
1430 | unsigned int blocksize = i_blocksize(inode); |
1431 | unsigned int offs = newsize & (blocksize - 1); | |
1432 | if (offs) { | |
1433 | error = gfs2_block_zero_range(inode, newsize, | |
1434 | blocksize - offs); | |
ff8f33c8 | 1435 | if (error) |
80990f40 | 1436 | goto out; |
b3b94faa | 1437 | } |
ff8f33c8 | 1438 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1439 | } |
1440 | ||
ff8f33c8 | 1441 | i_size_write(inode, newsize); |
078cd827 | 1442 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
ff8f33c8 | 1443 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1444 | |
fa731fc4 SW |
1445 | if (journaled) |
1446 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1447 | else | |
7caef267 | 1448 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1449 | |
a91ea69f | 1450 | out: |
80990f40 AG |
1451 | brelse(dibh); |
1452 | if (current->journal_info) | |
1453 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1454 | return error; |
1455 | } | |
1456 | ||
628e366d AG |
1457 | int gfs2_iomap_get_alloc(struct inode *inode, loff_t pos, loff_t length, |
1458 | struct iomap *iomap) | |
1459 | { | |
1460 | struct metapath mp = { .mp_aheight = 1, }; | |
1461 | int ret; | |
1462 | ||
1463 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp); | |
1464 | if (!ret && iomap->type == IOMAP_HOLE) | |
1465 | ret = gfs2_iomap_alloc(inode, iomap, IOMAP_WRITE, &mp); | |
1466 | release_metapath(&mp); | |
1467 | return ret; | |
1468 | } | |
1469 | ||
d552a2b9 BP |
1470 | /** |
1471 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1472 | * @ip: inode | |
1473 | * @rg_gh: holder of resource group glock | |
5cf26b1e AG |
1474 | * @bh: buffer head to sweep |
1475 | * @start: starting point in bh | |
1476 | * @end: end point in bh | |
1477 | * @meta: true if bh points to metadata (rather than data) | |
d552a2b9 | 1478 | * @btotal: place to keep count of total blocks freed |
d552a2b9 BP |
1479 | * |
1480 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1481 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1482 | * block has references to multiple rgrps, we break it into individual | |
1483 | * transactions. This allows other processes to use the rgrps while we're | |
1484 | * focused on a single one, for better concurrency / performance. | |
1485 | * At every transaction boundary, we rewrite the inode into the journal. | |
1486 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1487 | * if we're interrupted by power-outages. | |
1488 | * | |
1489 | * Returns: 0, or return code if an error occurred. | |
1490 | * *btotal has the total number of blocks freed | |
1491 | */ | |
1492 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
5cf26b1e AG |
1493 | struct buffer_head *bh, __be64 *start, __be64 *end, |
1494 | bool meta, u32 *btotal) | |
b3b94faa | 1495 | { |
9b8c81d1 | 1496 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1497 | struct gfs2_rgrpd *rgd; |
1498 | struct gfs2_trans *tr; | |
5cf26b1e | 1499 | __be64 *p; |
d552a2b9 BP |
1500 | int blks_outside_rgrp; |
1501 | u64 bn, bstart, isize_blks; | |
1502 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
d552a2b9 BP |
1503 | int ret = 0; |
1504 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1505 | ||
d552a2b9 | 1506 | more_rgrps: |
5cf26b1e AG |
1507 | rgd = NULL; |
1508 | if (gfs2_holder_initialized(rd_gh)) { | |
1509 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); | |
1510 | gfs2_assert_withdraw(sdp, | |
1511 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1512 | } | |
d552a2b9 BP |
1513 | blks_outside_rgrp = 0; |
1514 | bstart = 0; | |
1515 | blen = 0; | |
d552a2b9 | 1516 | |
5cf26b1e | 1517 | for (p = start; p < end; p++) { |
d552a2b9 BP |
1518 | if (!*p) |
1519 | continue; | |
1520 | bn = be64_to_cpu(*p); | |
5cf26b1e AG |
1521 | |
1522 | if (rgd) { | |
1523 | if (!rgrp_contains_block(rgd, bn)) { | |
1524 | blks_outside_rgrp++; | |
1525 | continue; | |
1526 | } | |
d552a2b9 | 1527 | } else { |
90bcab99 | 1528 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
5cf26b1e AG |
1529 | if (unlikely(!rgd)) { |
1530 | ret = -EIO; | |
1531 | goto out; | |
1532 | } | |
d552a2b9 BP |
1533 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1534 | 0, rd_gh); | |
1535 | if (ret) | |
1536 | goto out; | |
1537 | ||
1538 | /* Must be done with the rgrp glock held: */ | |
1539 | if (gfs2_rs_active(&ip->i_res) && | |
1540 | rgd == ip->i_res.rs_rbm.rgd) | |
1541 | gfs2_rs_deltree(&ip->i_res); | |
1542 | } | |
1543 | ||
d552a2b9 BP |
1544 | /* The size of our transactions will be unknown until we |
1545 | actually process all the metadata blocks that relate to | |
1546 | the rgrp. So we estimate. We know it can't be more than | |
1547 | the dinode's i_blocks and we don't want to exceed the | |
1548 | journal flush threshold, sd_log_thresh2. */ | |
1549 | if (current->journal_info == NULL) { | |
1550 | unsigned int jblocks_rqsted, revokes; | |
1551 | ||
1552 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1553 | RES_INDIRECT; | |
1554 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1555 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1556 | jblocks_rqsted += | |
1557 | atomic_read(&sdp->sd_log_thresh2); | |
1558 | else | |
1559 | jblocks_rqsted += isize_blks; | |
1560 | revokes = jblocks_rqsted; | |
1561 | if (meta) | |
5cf26b1e | 1562 | revokes += end - start; |
d552a2b9 BP |
1563 | else if (ip->i_depth) |
1564 | revokes += sdp->sd_inptrs; | |
1565 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1566 | if (ret) | |
1567 | goto out_unlock; | |
1568 | down_write(&ip->i_rw_mutex); | |
1569 | } | |
1570 | /* check if we will exceed the transaction blocks requested */ | |
1571 | tr = current->journal_info; | |
1572 | if (tr->tr_num_buf_new + RES_STATFS + | |
1573 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1574 | /* We set blks_outside_rgrp to ensure the loop will | |
1575 | be repeated for the same rgrp, but with a new | |
1576 | transaction. */ | |
1577 | blks_outside_rgrp++; | |
1578 | /* This next part is tricky. If the buffer was added | |
1579 | to the transaction, we've already set some block | |
1580 | pointers to 0, so we better follow through and free | |
1581 | them, or we will introduce corruption (so break). | |
1582 | This may be impossible, or at least rare, but I | |
1583 | decided to cover the case regardless. | |
1584 | ||
1585 | If the buffer was not added to the transaction | |
1586 | (this call), doing so would exceed our transaction | |
1587 | size, so we need to end the transaction and start a | |
1588 | new one (so goto). */ | |
1589 | ||
1590 | if (buf_in_tr) | |
1591 | break; | |
1592 | goto out_unlock; | |
1593 | } | |
1594 | ||
1595 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1596 | buf_in_tr = true; | |
1597 | *p = 0; | |
1598 | if (bstart + blen == bn) { | |
1599 | blen++; | |
1600 | continue; | |
1601 | } | |
1602 | if (bstart) { | |
0ddeded4 | 1603 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1604 | (*btotal) += blen; |
1605 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1606 | } | |
1607 | bstart = bn; | |
1608 | blen = 1; | |
1609 | } | |
1610 | if (bstart) { | |
0ddeded4 | 1611 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1612 | (*btotal) += blen; |
1613 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1614 | } | |
1615 | out_unlock: | |
1616 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1617 | outside the rgrp we just processed, | |
1618 | do it all over again. */ | |
1619 | if (current->journal_info) { | |
5cf26b1e AG |
1620 | struct buffer_head *dibh; |
1621 | ||
1622 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1623 | if (ret) | |
1624 | goto out; | |
d552a2b9 BP |
1625 | |
1626 | /* Every transaction boundary, we rewrite the dinode | |
1627 | to keep its di_blocks current in case of failure. */ | |
1628 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = | |
b32c8c76 | 1629 | current_time(&ip->i_inode); |
d552a2b9 BP |
1630 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1631 | gfs2_dinode_out(ip, dibh->b_data); | |
5cf26b1e | 1632 | brelse(dibh); |
d552a2b9 BP |
1633 | up_write(&ip->i_rw_mutex); |
1634 | gfs2_trans_end(sdp); | |
1635 | } | |
1636 | gfs2_glock_dq_uninit(rd_gh); | |
1637 | cond_resched(); | |
1638 | goto more_rgrps; | |
1639 | } | |
1640 | out: | |
1641 | return ret; | |
1642 | } | |
1643 | ||
10d2cf94 AG |
1644 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h) |
1645 | { | |
1646 | if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0]))) | |
1647 | return false; | |
1648 | return true; | |
1649 | } | |
1650 | ||
d552a2b9 BP |
1651 | /** |
1652 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
d552a2b9 BP |
1653 | * @mp: starting metapath |
1654 | * @h: desired height to search | |
1655 | * | |
10d2cf94 | 1656 | * Assumes the metapath is valid (with buffers) out to height h. |
d552a2b9 BP |
1657 | * Returns: true if a non-null pointer was found in the metapath buffer |
1658 | * false if all remaining pointers are NULL in the buffer | |
1659 | */ | |
1660 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
10d2cf94 AG |
1661 | unsigned int h, |
1662 | __u16 *end_list, unsigned int end_aligned) | |
d552a2b9 | 1663 | { |
10d2cf94 AG |
1664 | struct buffer_head *bh = mp->mp_bh[h]; |
1665 | __be64 *first, *ptr, *end; | |
1666 | ||
1667 | first = metaptr1(h, mp); | |
1668 | ptr = first + mp->mp_list[h]; | |
1669 | end = (__be64 *)(bh->b_data + bh->b_size); | |
1670 | if (end_list && mp_eq_to_hgt(mp, end_list, h)) { | |
1671 | bool keep_end = h < end_aligned; | |
1672 | end = first + end_list[h] + keep_end; | |
1673 | } | |
d552a2b9 | 1674 | |
10d2cf94 | 1675 | while (ptr < end) { |
c4a9d189 | 1676 | if (*ptr) { /* if we have a non-null pointer */ |
10d2cf94 | 1677 | mp->mp_list[h] = ptr - first; |
c4a9d189 BP |
1678 | h++; |
1679 | if (h < GFS2_MAX_META_HEIGHT) | |
10d2cf94 | 1680 | mp->mp_list[h] = 0; |
d552a2b9 | 1681 | return true; |
c4a9d189 | 1682 | } |
10d2cf94 | 1683 | ptr++; |
d552a2b9 | 1684 | } |
10d2cf94 | 1685 | return false; |
d552a2b9 BP |
1686 | } |
1687 | ||
1688 | enum dealloc_states { | |
1689 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1690 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1691 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1692 | DEALLOC_DONE = 3, /* process complete */ | |
1693 | }; | |
b3b94faa | 1694 | |
5cf26b1e AG |
1695 | static inline void |
1696 | metapointer_range(struct metapath *mp, int height, | |
1697 | __u16 *start_list, unsigned int start_aligned, | |
10d2cf94 | 1698 | __u16 *end_list, unsigned int end_aligned, |
5cf26b1e AG |
1699 | __be64 **start, __be64 **end) |
1700 | { | |
1701 | struct buffer_head *bh = mp->mp_bh[height]; | |
1702 | __be64 *first; | |
1703 | ||
1704 | first = metaptr1(height, mp); | |
1705 | *start = first; | |
1706 | if (mp_eq_to_hgt(mp, start_list, height)) { | |
1707 | bool keep_start = height < start_aligned; | |
1708 | *start = first + start_list[height] + keep_start; | |
1709 | } | |
1710 | *end = (__be64 *)(bh->b_data + bh->b_size); | |
10d2cf94 AG |
1711 | if (end_list && mp_eq_to_hgt(mp, end_list, height)) { |
1712 | bool keep_end = height < end_aligned; | |
1713 | *end = first + end_list[height] + keep_end; | |
1714 | } | |
1715 | } | |
1716 | ||
1717 | static inline bool walk_done(struct gfs2_sbd *sdp, | |
1718 | struct metapath *mp, int height, | |
1719 | __u16 *end_list, unsigned int end_aligned) | |
1720 | { | |
1721 | __u16 end; | |
1722 | ||
1723 | if (end_list) { | |
1724 | bool keep_end = height < end_aligned; | |
1725 | if (!mp_eq_to_hgt(mp, end_list, height)) | |
1726 | return false; | |
1727 | end = end_list[height] + keep_end; | |
1728 | } else | |
1729 | end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
1730 | return mp->mp_list[height] >= end; | |
5cf26b1e AG |
1731 | } |
1732 | ||
d552a2b9 | 1733 | /** |
10d2cf94 | 1734 | * punch_hole - deallocate blocks in a file |
d552a2b9 | 1735 | * @ip: inode to truncate |
10d2cf94 AG |
1736 | * @offset: the start of the hole |
1737 | * @length: the size of the hole (or 0 for truncate) | |
1738 | * | |
1739 | * Punch a hole into a file or truncate a file at a given position. This | |
1740 | * function operates in whole blocks (@offset and @length are rounded | |
1741 | * accordingly); partially filled blocks must be cleared otherwise. | |
d552a2b9 | 1742 | * |
10d2cf94 AG |
1743 | * This function works from the bottom up, and from the right to the left. In |
1744 | * other words, it strips off the highest layer (data) before stripping any of | |
1745 | * the metadata. Doing it this way is best in case the operation is interrupted | |
1746 | * by power failure, etc. The dinode is rewritten in every transaction to | |
1747 | * guarantee integrity. | |
d552a2b9 | 1748 | */ |
10d2cf94 | 1749 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length) |
d552a2b9 BP |
1750 | { |
1751 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
bb491ce6 | 1752 | u64 maxsize = sdp->sd_heightsize[ip->i_height]; |
10d2cf94 | 1753 | struct metapath mp = {}; |
d552a2b9 BP |
1754 | struct buffer_head *dibh, *bh; |
1755 | struct gfs2_holder rd_gh; | |
cb7f0903 | 1756 | unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; |
10d2cf94 AG |
1757 | u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift; |
1758 | __u16 start_list[GFS2_MAX_META_HEIGHT]; | |
1759 | __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL; | |
4e56a641 | 1760 | unsigned int start_aligned, uninitialized_var(end_aligned); |
d552a2b9 BP |
1761 | unsigned int strip_h = ip->i_height - 1; |
1762 | u32 btotal = 0; | |
1763 | int ret, state; | |
1764 | int mp_h; /* metapath buffers are read in to this height */ | |
d552a2b9 | 1765 | u64 prev_bnr = 0; |
5cf26b1e | 1766 | __be64 *start, *end; |
b3b94faa | 1767 | |
bb491ce6 AG |
1768 | if (offset >= maxsize) { |
1769 | /* | |
1770 | * The starting point lies beyond the allocated meta-data; | |
1771 | * there are no blocks do deallocate. | |
1772 | */ | |
1773 | return 0; | |
1774 | } | |
1775 | ||
10d2cf94 AG |
1776 | /* |
1777 | * The start position of the hole is defined by lblock, start_list, and | |
1778 | * start_aligned. The end position of the hole is defined by lend, | |
1779 | * end_list, and end_aligned. | |
1780 | * | |
1781 | * start_aligned and end_aligned define down to which height the start | |
1782 | * and end positions are aligned to the metadata tree (i.e., the | |
1783 | * position is a multiple of the metadata granularity at the height | |
1784 | * above). This determines at which heights additional meta pointers | |
1785 | * needs to be preserved for the remaining data. | |
1786 | */ | |
b3b94faa | 1787 | |
10d2cf94 | 1788 | if (length) { |
10d2cf94 AG |
1789 | u64 end_offset = offset + length; |
1790 | u64 lend; | |
1791 | ||
1792 | /* | |
1793 | * Clip the end at the maximum file size for the given height: | |
1794 | * that's how far the metadata goes; files bigger than that | |
1795 | * will have additional layers of indirection. | |
1796 | */ | |
1797 | if (end_offset > maxsize) | |
1798 | end_offset = maxsize; | |
1799 | lend = end_offset >> bsize_shift; | |
1800 | ||
1801 | if (lblock >= lend) | |
1802 | return 0; | |
1803 | ||
1804 | find_metapath(sdp, lend, &mp, ip->i_height); | |
1805 | end_list = __end_list; | |
1806 | memcpy(end_list, mp.mp_list, sizeof(mp.mp_list)); | |
1807 | ||
1808 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { | |
1809 | if (end_list[mp_h]) | |
1810 | break; | |
1811 | } | |
1812 | end_aligned = mp_h; | |
1813 | } | |
1814 | ||
1815 | find_metapath(sdp, lblock, &mp, ip->i_height); | |
cb7f0903 AG |
1816 | memcpy(start_list, mp.mp_list, sizeof(start_list)); |
1817 | ||
cb7f0903 AG |
1818 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { |
1819 | if (start_list[mp_h]) | |
1820 | break; | |
1821 | } | |
1822 | start_aligned = mp_h; | |
d552a2b9 BP |
1823 | |
1824 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1825 | if (ret) | |
1826 | return ret; | |
b3b94faa | 1827 | |
d552a2b9 BP |
1828 | mp.mp_bh[0] = dibh; |
1829 | ret = lookup_metapath(ip, &mp); | |
e8b43fe0 AG |
1830 | if (ret) |
1831 | goto out_metapath; | |
c3ce5aa9 AG |
1832 | |
1833 | /* issue read-ahead on metadata */ | |
5cf26b1e AG |
1834 | for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) { |
1835 | metapointer_range(&mp, mp_h, start_list, start_aligned, | |
10d2cf94 | 1836 | end_list, end_aligned, &start, &end); |
5cf26b1e AG |
1837 | gfs2_metapath_ra(ip->i_gl, start, end); |
1838 | } | |
c3ce5aa9 | 1839 | |
e8b43fe0 | 1840 | if (mp.mp_aheight == ip->i_height) |
d552a2b9 BP |
1841 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ |
1842 | else | |
1843 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1844 | |
d552a2b9 BP |
1845 | ret = gfs2_rindex_update(sdp); |
1846 | if (ret) | |
1847 | goto out_metapath; | |
1848 | ||
1849 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1850 | if (ret) | |
1851 | goto out_metapath; | |
1852 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1853 | ||
1854 | mp_h = strip_h; | |
1855 | ||
1856 | while (state != DEALLOC_DONE) { | |
1857 | switch (state) { | |
1858 | /* Truncate a full metapath at the given strip height. | |
1859 | * Note that strip_h == mp_h in order to be in this state. */ | |
1860 | case DEALLOC_MP_FULL: | |
d552a2b9 BP |
1861 | bh = mp.mp_bh[mp_h]; |
1862 | gfs2_assert_withdraw(sdp, bh); | |
1863 | if (gfs2_assert_withdraw(sdp, | |
1864 | prev_bnr != bh->b_blocknr)) { | |
1865 | printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, " | |
1866 | "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n", | |
1867 | sdp->sd_fsname, | |
1868 | (unsigned long long)ip->i_no_addr, | |
1869 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1870 | } | |
1871 | prev_bnr = bh->b_blocknr; | |
cb7f0903 | 1872 | |
5cf26b1e AG |
1873 | if (gfs2_metatype_check(sdp, bh, |
1874 | (mp_h ? GFS2_METATYPE_IN : | |
1875 | GFS2_METATYPE_DI))) { | |
1876 | ret = -EIO; | |
1877 | goto out; | |
1878 | } | |
1879 | ||
10d2cf94 AG |
1880 | /* |
1881 | * Below, passing end_aligned as 0 gives us the | |
1882 | * metapointer range excluding the end point: the end | |
1883 | * point is the first metapath we must not deallocate! | |
1884 | */ | |
1885 | ||
5cf26b1e | 1886 | metapointer_range(&mp, mp_h, start_list, start_aligned, |
10d2cf94 | 1887 | end_list, 0 /* end_aligned */, |
5cf26b1e AG |
1888 | &start, &end); |
1889 | ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h], | |
1890 | start, end, | |
1891 | mp_h != ip->i_height - 1, | |
1892 | &btotal); | |
cb7f0903 | 1893 | |
d552a2b9 BP |
1894 | /* If we hit an error or just swept dinode buffer, |
1895 | just exit. */ | |
1896 | if (ret || !mp_h) { | |
1897 | state = DEALLOC_DONE; | |
1898 | break; | |
1899 | } | |
1900 | state = DEALLOC_MP_LOWER; | |
1901 | break; | |
1902 | ||
1903 | /* lower the metapath strip height */ | |
1904 | case DEALLOC_MP_LOWER: | |
1905 | /* We're done with the current buffer, so release it, | |
1906 | unless it's the dinode buffer. Then back up to the | |
1907 | previous pointer. */ | |
1908 | if (mp_h) { | |
1909 | brelse(mp.mp_bh[mp_h]); | |
1910 | mp.mp_bh[mp_h] = NULL; | |
1911 | } | |
1912 | /* If we can't get any lower in height, we've stripped | |
1913 | off all we can. Next step is to back up and start | |
1914 | stripping the previous level of metadata. */ | |
1915 | if (mp_h == 0) { | |
1916 | strip_h--; | |
cb7f0903 | 1917 | memcpy(mp.mp_list, start_list, sizeof(start_list)); |
d552a2b9 BP |
1918 | mp_h = strip_h; |
1919 | state = DEALLOC_FILL_MP; | |
1920 | break; | |
1921 | } | |
1922 | mp.mp_list[mp_h] = 0; | |
1923 | mp_h--; /* search one metadata height down */ | |
d552a2b9 | 1924 | mp.mp_list[mp_h]++; |
10d2cf94 AG |
1925 | if (walk_done(sdp, &mp, mp_h, end_list, end_aligned)) |
1926 | break; | |
d552a2b9 BP |
1927 | /* Here we've found a part of the metapath that is not |
1928 | * allocated. We need to search at that height for the | |
1929 | * next non-null pointer. */ | |
10d2cf94 | 1930 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) { |
d552a2b9 BP |
1931 | state = DEALLOC_FILL_MP; |
1932 | mp_h++; | |
1933 | } | |
1934 | /* No more non-null pointers at this height. Back up | |
1935 | to the previous height and try again. */ | |
1936 | break; /* loop around in the same state */ | |
1937 | ||
1938 | /* Fill the metapath with buffers to the given height. */ | |
1939 | case DEALLOC_FILL_MP: | |
1940 | /* Fill the buffers out to the current height. */ | |
1941 | ret = fillup_metapath(ip, &mp, mp_h); | |
c3ce5aa9 | 1942 | if (ret < 0) |
d552a2b9 | 1943 | goto out; |
c3ce5aa9 | 1944 | |
e7445ced AG |
1945 | /* On the first pass, issue read-ahead on metadata. */ |
1946 | if (mp.mp_aheight > 1 && strip_h == ip->i_height - 1) { | |
1947 | unsigned int height = mp.mp_aheight - 1; | |
1948 | ||
1949 | /* No read-ahead for data blocks. */ | |
1950 | if (mp.mp_aheight - 1 == strip_h) | |
1951 | height--; | |
1952 | ||
1953 | for (; height >= mp.mp_aheight - ret; height--) { | |
1954 | metapointer_range(&mp, height, | |
5cf26b1e | 1955 | start_list, start_aligned, |
10d2cf94 | 1956 | end_list, end_aligned, |
5cf26b1e AG |
1957 | &start, &end); |
1958 | gfs2_metapath_ra(ip->i_gl, start, end); | |
1959 | } | |
c3ce5aa9 | 1960 | } |
d552a2b9 BP |
1961 | |
1962 | /* If buffers found for the entire strip height */ | |
e8b43fe0 | 1963 | if (mp.mp_aheight - 1 == strip_h) { |
d552a2b9 BP |
1964 | state = DEALLOC_MP_FULL; |
1965 | break; | |
1966 | } | |
e8b43fe0 AG |
1967 | if (mp.mp_aheight < ip->i_height) /* We have a partial height */ |
1968 | mp_h = mp.mp_aheight - 1; | |
d552a2b9 BP |
1969 | |
1970 | /* If we find a non-null block pointer, crawl a bit | |
1971 | higher up in the metapath and try again, otherwise | |
1972 | we need to look lower for a new starting point. */ | |
10d2cf94 | 1973 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) |
d552a2b9 BP |
1974 | mp_h++; |
1975 | else | |
1976 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1977 | break; |
d552a2b9 | 1978 | } |
b3b94faa DT |
1979 | } |
1980 | ||
d552a2b9 BP |
1981 | if (btotal) { |
1982 | if (current->journal_info == NULL) { | |
1983 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1984 | RES_QUOTA, 0); | |
1985 | if (ret) | |
1986 | goto out; | |
1987 | down_write(&ip->i_rw_mutex); | |
1988 | } | |
1989 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
1990 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
1991 | ip->i_inode.i_gid); | |
b32c8c76 | 1992 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
d552a2b9 BP |
1993 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1994 | gfs2_dinode_out(ip, dibh->b_data); | |
1995 | up_write(&ip->i_rw_mutex); | |
1996 | gfs2_trans_end(sdp); | |
1997 | } | |
b3b94faa | 1998 | |
d552a2b9 BP |
1999 | out: |
2000 | if (gfs2_holder_initialized(&rd_gh)) | |
2001 | gfs2_glock_dq_uninit(&rd_gh); | |
2002 | if (current->journal_info) { | |
2003 | up_write(&ip->i_rw_mutex); | |
2004 | gfs2_trans_end(sdp); | |
2005 | cond_resched(); | |
2006 | } | |
2007 | gfs2_quota_unhold(ip); | |
2008 | out_metapath: | |
2009 | release_metapath(&mp); | |
2010 | return ret; | |
b3b94faa DT |
2011 | } |
2012 | ||
2013 | static int trunc_end(struct gfs2_inode *ip) | |
2014 | { | |
feaa7bba | 2015 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2016 | struct buffer_head *dibh; |
2017 | int error; | |
2018 | ||
2019 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2020 | if (error) | |
2021 | return error; | |
2022 | ||
2023 | down_write(&ip->i_rw_mutex); | |
2024 | ||
2025 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2026 | if (error) | |
2027 | goto out; | |
2028 | ||
a2e0f799 | 2029 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 2030 | ip->i_height = 0; |
ce276b06 | 2031 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 2032 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 2033 | gfs2_ordered_del_inode(ip); |
b3b94faa | 2034 | } |
078cd827 | 2035 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
383f01fb | 2036 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 2037 | |
350a9b0a | 2038 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 2039 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
2040 | brelse(dibh); |
2041 | ||
a91ea69f | 2042 | out: |
b3b94faa | 2043 | up_write(&ip->i_rw_mutex); |
b3b94faa | 2044 | gfs2_trans_end(sdp); |
b3b94faa DT |
2045 | return error; |
2046 | } | |
2047 | ||
2048 | /** | |
2049 | * do_shrink - make a file smaller | |
ff8f33c8 | 2050 | * @inode: the inode |
ff8f33c8 | 2051 | * @newsize: the size to make the file |
b3b94faa | 2052 | * |
ff8f33c8 SW |
2053 | * Called with an exclusive lock on @inode. The @size must |
2054 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
2055 | * |
2056 | * Returns: errno | |
2057 | */ | |
2058 | ||
8b5860a3 | 2059 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 2060 | { |
ff8f33c8 | 2061 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
2062 | int error; |
2063 | ||
8b5860a3 | 2064 | error = trunc_start(inode, newsize); |
b3b94faa DT |
2065 | if (error < 0) |
2066 | return error; | |
ff8f33c8 | 2067 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
2068 | return 0; |
2069 | ||
10d2cf94 | 2070 | error = punch_hole(ip, newsize, 0); |
ff8f33c8 | 2071 | if (error == 0) |
b3b94faa DT |
2072 | error = trunc_end(ip); |
2073 | ||
2074 | return error; | |
2075 | } | |
2076 | ||
ff8f33c8 | 2077 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 2078 | { |
ff8f33c8 SW |
2079 | int ret; |
2080 | ||
8b5860a3 | 2081 | ret = do_shrink(inode, inode->i_size); |
ff8f33c8 SW |
2082 | WARN_ON(ret != 0); |
2083 | } | |
2084 | ||
2085 | /** | |
2086 | * do_grow - Touch and update inode size | |
2087 | * @inode: The inode | |
2088 | * @size: The new size | |
2089 | * | |
2090 | * This function updates the timestamps on the inode and | |
2091 | * may also increase the size of the inode. This function | |
2092 | * must not be called with @size any smaller than the current | |
2093 | * inode size. | |
2094 | * | |
2095 | * Although it is not strictly required to unstuff files here, | |
2096 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
2097 | * code which will result in a buffer overrun if the size is larger | |
2098 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 2099 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
2100 | * just update the inode size directly. |
2101 | * | |
2102 | * Returns: 0 on success, or -ve on error | |
2103 | */ | |
2104 | ||
2105 | static int do_grow(struct inode *inode, u64 size) | |
2106 | { | |
2107 | struct gfs2_inode *ip = GFS2_I(inode); | |
2108 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 2109 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
2110 | struct buffer_head *dibh; |
2111 | int error; | |
2f7ee358 | 2112 | int unstuff = 0; |
a13b8c5f | 2113 | |
235628c5 | 2114 | if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) { |
b8fbf471 | 2115 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 2116 | if (error) |
5407e242 | 2117 | return error; |
ff8f33c8 | 2118 | |
7b9cff46 | 2119 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
2120 | if (error) |
2121 | goto do_grow_qunlock; | |
2f7ee358 | 2122 | unstuff = 1; |
ff8f33c8 SW |
2123 | } |
2124 | ||
a01aedfe | 2125 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
bc020561 BP |
2126 | (unstuff && |
2127 | gfs2_is_jdata(ip) ? RES_JDATA : 0) + | |
a01aedfe BP |
2128 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? |
2129 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 2130 | if (error) |
ff8f33c8 | 2131 | goto do_grow_release; |
a13b8c5f | 2132 | |
2f7ee358 | 2133 | if (unstuff) { |
ff8f33c8 SW |
2134 | error = gfs2_unstuff_dinode(ip, NULL); |
2135 | if (error) | |
2136 | goto do_end_trans; | |
2137 | } | |
a13b8c5f WC |
2138 | |
2139 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2140 | if (error) | |
ff8f33c8 | 2141 | goto do_end_trans; |
a13b8c5f | 2142 | |
ff8f33c8 | 2143 | i_size_write(inode, size); |
078cd827 | 2144 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
350a9b0a | 2145 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
2146 | gfs2_dinode_out(ip, dibh->b_data); |
2147 | brelse(dibh); | |
2148 | ||
ff8f33c8 | 2149 | do_end_trans: |
a13b8c5f | 2150 | gfs2_trans_end(sdp); |
ff8f33c8 | 2151 | do_grow_release: |
2f7ee358 | 2152 | if (unstuff) { |
ff8f33c8 SW |
2153 | gfs2_inplace_release(ip); |
2154 | do_grow_qunlock: | |
2155 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 2156 | } |
a13b8c5f WC |
2157 | return error; |
2158 | } | |
2159 | ||
b3b94faa | 2160 | /** |
ff8f33c8 SW |
2161 | * gfs2_setattr_size - make a file a given size |
2162 | * @inode: the inode | |
2163 | * @newsize: the size to make the file | |
b3b94faa | 2164 | * |
ff8f33c8 | 2165 | * The file size can grow, shrink, or stay the same size. This |
3e7aafc3 | 2166 | * is called holding i_rwsem and an exclusive glock on the inode |
ff8f33c8 | 2167 | * in question. |
b3b94faa DT |
2168 | * |
2169 | * Returns: errno | |
2170 | */ | |
2171 | ||
ff8f33c8 | 2172 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 2173 | { |
af5c2697 | 2174 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 2175 | int ret; |
b3b94faa | 2176 | |
ff8f33c8 | 2177 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 2178 | |
ff8f33c8 SW |
2179 | ret = inode_newsize_ok(inode, newsize); |
2180 | if (ret) | |
2181 | return ret; | |
b3b94faa | 2182 | |
562c72aa CH |
2183 | inode_dio_wait(inode); |
2184 | ||
b54e9a0b | 2185 | ret = gfs2_rsqa_alloc(ip); |
d2b47cfb | 2186 | if (ret) |
2b3dcf35 | 2187 | goto out; |
d2b47cfb | 2188 | |
8b5860a3 | 2189 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
2190 | ret = do_grow(inode, newsize); |
2191 | goto out; | |
2192 | } | |
ff8f33c8 | 2193 | |
8b5860a3 | 2194 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 2195 | out: |
a097dc7e | 2196 | gfs2_rsqa_delete(ip, NULL); |
2b3dcf35 | 2197 | return ret; |
b3b94faa DT |
2198 | } |
2199 | ||
2200 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
2201 | { | |
2202 | int error; | |
10d2cf94 | 2203 | error = punch_hole(ip, i_size_read(&ip->i_inode), 0); |
b3b94faa DT |
2204 | if (!error) |
2205 | error = trunc_end(ip); | |
2206 | return error; | |
2207 | } | |
2208 | ||
2209 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
2210 | { | |
10d2cf94 | 2211 | return punch_hole(ip, 0, 0); |
b3b94faa DT |
2212 | } |
2213 | ||
b50f227b SW |
2214 | /** |
2215 | * gfs2_free_journal_extents - Free cached journal bmap info | |
2216 | * @jd: The journal | |
2217 | * | |
2218 | */ | |
2219 | ||
2220 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
2221 | { | |
2222 | struct gfs2_journal_extent *jext; | |
2223 | ||
2224 | while(!list_empty(&jd->extent_list)) { | |
2225 | jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list); | |
2226 | list_del(&jext->list); | |
2227 | kfree(jext); | |
2228 | } | |
2229 | } | |
2230 | ||
2231 | /** | |
2232 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
2233 | * @jd: The journal descriptor | |
2234 | * @lblock: The logical block at start of new extent | |
c62baf65 | 2235 | * @dblock: The physical block at start of new extent |
b50f227b SW |
2236 | * @blocks: Size of extent in fs blocks |
2237 | * | |
2238 | * Returns: 0 on success or -ENOMEM | |
2239 | */ | |
2240 | ||
2241 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
2242 | { | |
2243 | struct gfs2_journal_extent *jext; | |
2244 | ||
2245 | if (!list_empty(&jd->extent_list)) { | |
2246 | jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list); | |
2247 | if ((jext->dblock + jext->blocks) == dblock) { | |
2248 | jext->blocks += blocks; | |
2249 | return 0; | |
2250 | } | |
2251 | } | |
2252 | ||
2253 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
2254 | if (jext == NULL) | |
2255 | return -ENOMEM; | |
2256 | jext->dblock = dblock; | |
2257 | jext->lblock = lblock; | |
2258 | jext->blocks = blocks; | |
2259 | list_add_tail(&jext->list, &jd->extent_list); | |
2260 | jd->nr_extents++; | |
2261 | return 0; | |
2262 | } | |
2263 | ||
2264 | /** | |
2265 | * gfs2_map_journal_extents - Cache journal bmap info | |
2266 | * @sdp: The super block | |
2267 | * @jd: The journal to map | |
2268 | * | |
2269 | * Create a reusable "extent" mapping from all logical | |
2270 | * blocks to all physical blocks for the given journal. This will save | |
2271 | * us time when writing journal blocks. Most journals will have only one | |
2272 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
2273 | * arranges the journal blocks sequentially to maximize performance. | |
2274 | * So the extent would map the first block for the entire file length. | |
2275 | * However, gfs2_jadd can happen while file activity is happening, so | |
2276 | * those journals may not be sequential. Less likely is the case where | |
2277 | * the users created their own journals by mounting the metafs and | |
2278 | * laying it out. But it's still possible. These journals might have | |
2279 | * several extents. | |
2280 | * | |
2281 | * Returns: 0 on success, or error on failure | |
2282 | */ | |
2283 | ||
2284 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
2285 | { | |
2286 | u64 lblock = 0; | |
2287 | u64 lblock_stop; | |
2288 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
2289 | struct buffer_head bh; | |
2290 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
2291 | u64 size; | |
2292 | int rc; | |
98583b3e | 2293 | ktime_t start, end; |
b50f227b | 2294 | |
98583b3e | 2295 | start = ktime_get(); |
b50f227b SW |
2296 | lblock_stop = i_size_read(jd->jd_inode) >> shift; |
2297 | size = (lblock_stop - lblock) << shift; | |
2298 | jd->nr_extents = 0; | |
2299 | WARN_ON(!list_empty(&jd->extent_list)); | |
2300 | ||
2301 | do { | |
2302 | bh.b_state = 0; | |
2303 | bh.b_blocknr = 0; | |
2304 | bh.b_size = size; | |
2305 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
2306 | if (rc || !buffer_mapped(&bh)) | |
2307 | goto fail; | |
2308 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
2309 | if (rc) | |
2310 | goto fail; | |
2311 | size -= bh.b_size; | |
2312 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2313 | } while(size > 0); | |
2314 | ||
98583b3e AD |
2315 | end = ktime_get(); |
2316 | fs_info(sdp, "journal %d mapped with %u extents in %lldms\n", jd->jd_jid, | |
2317 | jd->nr_extents, ktime_ms_delta(end, start)); | |
b50f227b SW |
2318 | return 0; |
2319 | ||
2320 | fail: | |
2321 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
2322 | rc, jd->jd_jid, | |
2323 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
2324 | jd->nr_extents); | |
2325 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
2326 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
2327 | bh.b_state, (unsigned long long)bh.b_size); | |
2328 | gfs2_free_journal_extents(jd); | |
2329 | return rc; | |
2330 | } | |
2331 | ||
b3b94faa DT |
2332 | /** |
2333 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
2334 | * @ip: the file being written to | |
2335 | * @offset: the offset to write to | |
2336 | * @len: the number of bytes being written | |
b3b94faa | 2337 | * |
461cb419 | 2338 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
2339 | */ |
2340 | ||
cd915493 | 2341 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 2342 | unsigned int len) |
b3b94faa | 2343 | { |
feaa7bba | 2344 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
2345 | struct buffer_head bh; |
2346 | unsigned int shift; | |
2347 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 2348 | u64 end_of_file; |
b3b94faa | 2349 | |
b3b94faa DT |
2350 | if (!len) |
2351 | return 0; | |
2352 | ||
2353 | if (gfs2_is_stuffed(ip)) { | |
235628c5 | 2354 | if (offset + len > gfs2_max_stuffed_size(ip)) |
461cb419 | 2355 | return 1; |
b3b94faa DT |
2356 | return 0; |
2357 | } | |
2358 | ||
941e6d7d | 2359 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 2360 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 2361 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
2362 | lblock = offset >> shift; |
2363 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
77612578 | 2364 | if (lblock_stop > end_of_file && ip != GFS2_I(sdp->sd_rindex)) |
461cb419 | 2365 | return 1; |
b3b94faa | 2366 | |
941e6d7d SW |
2367 | size = (lblock_stop - lblock) << shift; |
2368 | do { | |
2369 | bh.b_state = 0; | |
2370 | bh.b_size = size; | |
2371 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
2372 | if (!buffer_mapped(&bh)) | |
461cb419 | 2373 | return 1; |
941e6d7d SW |
2374 | size -= bh.b_size; |
2375 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2376 | } while(size > 0); | |
b3b94faa DT |
2377 | |
2378 | return 0; | |
2379 | } | |
2380 | ||
4e56a641 AG |
2381 | static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length) |
2382 | { | |
2383 | struct gfs2_inode *ip = GFS2_I(inode); | |
2384 | struct buffer_head *dibh; | |
2385 | int error; | |
2386 | ||
2387 | if (offset >= inode->i_size) | |
2388 | return 0; | |
2389 | if (offset + length > inode->i_size) | |
2390 | length = inode->i_size - offset; | |
2391 | ||
2392 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2393 | if (error) | |
2394 | return error; | |
2395 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
2396 | memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0, | |
2397 | length); | |
2398 | brelse(dibh); | |
2399 | return 0; | |
2400 | } | |
2401 | ||
2402 | static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset, | |
2403 | loff_t length) | |
2404 | { | |
2405 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2406 | loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
2407 | int error; | |
2408 | ||
2409 | while (length) { | |
2410 | struct gfs2_trans *tr; | |
2411 | loff_t chunk; | |
2412 | unsigned int offs; | |
2413 | ||
2414 | chunk = length; | |
2415 | if (chunk > max_chunk) | |
2416 | chunk = max_chunk; | |
2417 | ||
2418 | offs = offset & ~PAGE_MASK; | |
2419 | if (offs && chunk > PAGE_SIZE) | |
2420 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
2421 | ||
2422 | truncate_pagecache_range(inode, offset, chunk); | |
2423 | offset += chunk; | |
2424 | length -= chunk; | |
2425 | ||
2426 | tr = current->journal_info; | |
2427 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
2428 | continue; | |
2429 | ||
2430 | gfs2_trans_end(sdp); | |
2431 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
2432 | if (error) | |
2433 | return error; | |
2434 | } | |
2435 | return 0; | |
2436 | } | |
2437 | ||
2438 | int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length) | |
2439 | { | |
2440 | struct inode *inode = file_inode(file); | |
2441 | struct gfs2_inode *ip = GFS2_I(inode); | |
2442 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2443 | int error; | |
2444 | ||
2445 | if (gfs2_is_jdata(ip)) | |
2446 | error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA, | |
2447 | GFS2_JTRUNC_REVOKES); | |
2448 | else | |
2449 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2450 | if (error) | |
2451 | return error; | |
2452 | ||
2453 | if (gfs2_is_stuffed(ip)) { | |
2454 | error = stuffed_zero_range(inode, offset, length); | |
2455 | if (error) | |
2456 | goto out; | |
2457 | } else { | |
00251a16 | 2458 | unsigned int start_off, end_len, blocksize; |
4e56a641 AG |
2459 | |
2460 | blocksize = i_blocksize(inode); | |
2461 | start_off = offset & (blocksize - 1); | |
00251a16 | 2462 | end_len = (offset + length) & (blocksize - 1); |
4e56a641 AG |
2463 | if (start_off) { |
2464 | unsigned int len = length; | |
2465 | if (length > blocksize - start_off) | |
2466 | len = blocksize - start_off; | |
2467 | error = gfs2_block_zero_range(inode, offset, len); | |
2468 | if (error) | |
2469 | goto out; | |
2470 | if (start_off + length < blocksize) | |
00251a16 | 2471 | end_len = 0; |
4e56a641 | 2472 | } |
00251a16 | 2473 | if (end_len) { |
4e56a641 | 2474 | error = gfs2_block_zero_range(inode, |
00251a16 | 2475 | offset + length - end_len, end_len); |
4e56a641 AG |
2476 | if (error) |
2477 | goto out; | |
2478 | } | |
2479 | } | |
2480 | ||
2481 | if (gfs2_is_jdata(ip)) { | |
2482 | BUG_ON(!current->journal_info); | |
2483 | gfs2_journaled_truncate_range(inode, offset, length); | |
2484 | } else | |
2485 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2486 | ||
2487 | file_update_time(file); | |
2488 | mark_inode_dirty(inode); | |
2489 | ||
2490 | if (current->journal_info) | |
2491 | gfs2_trans_end(sdp); | |
2492 | ||
2493 | if (!gfs2_is_stuffed(ip)) | |
2494 | error = punch_hole(ip, offset, length); | |
2495 | ||
2496 | out: | |
2497 | if (current->journal_info) | |
2498 | gfs2_trans_end(sdp); | |
2499 | return error; | |
2500 | } |