]>
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) { |
5731be53 | 145 | gfs2_trans_add_unrevoke(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)) | |
679 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
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; | |
713 | /* Branching from existing tree */ | |
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; | |
724 | /* Tree complete, adding data blocks */ | |
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 | ||
64bc06bb AG |
928 | static int gfs2_write_lock(struct inode *inode) |
929 | { | |
930 | struct gfs2_inode *ip = GFS2_I(inode); | |
931 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
932 | int error; | |
933 | ||
934 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); | |
935 | error = gfs2_glock_nq(&ip->i_gh); | |
936 | if (error) | |
937 | goto out_uninit; | |
938 | if (&ip->i_inode == sdp->sd_rindex) { | |
939 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
940 | ||
941 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
942 | GL_NOCACHE, &m_ip->i_gh); | |
943 | if (error) | |
944 | goto out_unlock; | |
945 | } | |
946 | return 0; | |
947 | ||
948 | out_unlock: | |
949 | gfs2_glock_dq(&ip->i_gh); | |
950 | out_uninit: | |
951 | gfs2_holder_uninit(&ip->i_gh); | |
952 | return error; | |
953 | } | |
954 | ||
955 | static void gfs2_write_unlock(struct inode *inode) | |
956 | { | |
957 | struct gfs2_inode *ip = GFS2_I(inode); | |
958 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
959 | ||
960 | if (&ip->i_inode == sdp->sd_rindex) { | |
961 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
962 | ||
963 | gfs2_glock_dq_uninit(&m_ip->i_gh); | |
964 | } | |
965 | gfs2_glock_dq_uninit(&ip->i_gh); | |
966 | } | |
967 | ||
968 | static void gfs2_iomap_journaled_page_done(struct inode *inode, loff_t pos, | |
969 | unsigned copied, struct page *page, | |
970 | struct iomap *iomap) | |
971 | { | |
972 | struct gfs2_inode *ip = GFS2_I(inode); | |
973 | ||
974 | gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied); | |
975 | } | |
976 | ||
977 | static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, | |
978 | loff_t length, unsigned flags, | |
c26b5aa8 AG |
979 | struct iomap *iomap, |
980 | struct metapath *mp) | |
64bc06bb | 981 | { |
64bc06bb AG |
982 | struct gfs2_inode *ip = GFS2_I(inode); |
983 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
984 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; | |
985 | bool unstuff, alloc_required; | |
986 | int ret; | |
987 | ||
988 | ret = gfs2_write_lock(inode); | |
989 | if (ret) | |
990 | return ret; | |
991 | ||
992 | unstuff = gfs2_is_stuffed(ip) && | |
993 | pos + length > gfs2_max_stuffed_size(ip); | |
994 | ||
c26b5aa8 | 995 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, mp); |
64bc06bb | 996 | if (ret) |
c26b5aa8 | 997 | goto out_unlock; |
64bc06bb AG |
998 | |
999 | alloc_required = unstuff || iomap->type == IOMAP_HOLE; | |
1000 | ||
1001 | if (alloc_required || gfs2_is_jdata(ip)) | |
1002 | gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, | |
1003 | &ind_blocks); | |
1004 | ||
1005 | if (alloc_required) { | |
1006 | struct gfs2_alloc_parms ap = { | |
1007 | .target = data_blocks + ind_blocks | |
1008 | }; | |
1009 | ||
1010 | ret = gfs2_quota_lock_check(ip, &ap); | |
1011 | if (ret) | |
c26b5aa8 | 1012 | goto out_unlock; |
64bc06bb AG |
1013 | |
1014 | ret = gfs2_inplace_reserve(ip, &ap); | |
1015 | if (ret) | |
1016 | goto out_qunlock; | |
1017 | } | |
1018 | ||
1019 | rblocks = RES_DINODE + ind_blocks; | |
1020 | if (gfs2_is_jdata(ip)) | |
1021 | rblocks += data_blocks; | |
1022 | if (ind_blocks || data_blocks) | |
1023 | rblocks += RES_STATFS + RES_QUOTA; | |
1024 | if (inode == sdp->sd_rindex) | |
1025 | rblocks += 2 * RES_STATFS; | |
1026 | if (alloc_required) | |
1027 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); | |
1028 | ||
1029 | ret = gfs2_trans_begin(sdp, rblocks, iomap->length >> inode->i_blkbits); | |
1030 | if (ret) | |
1031 | goto out_trans_fail; | |
1032 | ||
1033 | if (unstuff) { | |
1034 | ret = gfs2_unstuff_dinode(ip, NULL); | |
1035 | if (ret) | |
1036 | goto out_trans_end; | |
c26b5aa8 | 1037 | release_metapath(mp); |
64bc06bb | 1038 | ret = gfs2_iomap_get(inode, iomap->offset, iomap->length, |
c26b5aa8 | 1039 | flags, iomap, mp); |
64bc06bb AG |
1040 | if (ret) |
1041 | goto out_trans_end; | |
1042 | } | |
1043 | ||
1044 | if (iomap->type == IOMAP_HOLE) { | |
c26b5aa8 | 1045 | ret = gfs2_iomap_alloc(inode, iomap, flags, mp); |
64bc06bb AG |
1046 | if (ret) { |
1047 | gfs2_trans_end(sdp); | |
1048 | gfs2_inplace_release(ip); | |
1049 | punch_hole(ip, iomap->offset, iomap->length); | |
1050 | goto out_qunlock; | |
1051 | } | |
1052 | } | |
fee5150c | 1053 | if (!gfs2_is_stuffed(ip) && gfs2_is_jdata(ip)) |
64bc06bb AG |
1054 | iomap->page_done = gfs2_iomap_journaled_page_done; |
1055 | return 0; | |
1056 | ||
1057 | out_trans_end: | |
1058 | gfs2_trans_end(sdp); | |
1059 | out_trans_fail: | |
1060 | if (alloc_required) | |
1061 | gfs2_inplace_release(ip); | |
1062 | out_qunlock: | |
1063 | if (alloc_required) | |
1064 | gfs2_quota_unlock(ip); | |
c26b5aa8 | 1065 | out_unlock: |
64bc06bb AG |
1066 | gfs2_write_unlock(inode); |
1067 | return ret; | |
1068 | } | |
1069 | ||
628e366d AG |
1070 | static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, |
1071 | unsigned flags, struct iomap *iomap) | |
1072 | { | |
1073 | struct gfs2_inode *ip = GFS2_I(inode); | |
1074 | struct metapath mp = { .mp_aheight = 1, }; | |
1075 | int ret; | |
1076 | ||
0ed91eca AG |
1077 | iomap->flags |= IOMAP_F_BUFFER_HEAD; |
1078 | ||
628e366d | 1079 | trace_gfs2_iomap_start(ip, pos, length, flags); |
967bcc91 | 1080 | if ((flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT)) { |
c26b5aa8 | 1081 | ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap, &mp); |
628e366d AG |
1082 | } else { |
1083 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); | |
c26b5aa8 | 1084 | |
967bcc91 AG |
1085 | /* |
1086 | * Silently fall back to buffered I/O for stuffed files or if | |
1087 | * we've hot a hole (see gfs2_file_direct_write). | |
1088 | */ | |
1089 | if ((flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT) && | |
1090 | iomap->type != IOMAP_MAPPED) | |
1091 | ret = -ENOTBLK; | |
628e366d | 1092 | } |
c26b5aa8 AG |
1093 | if (!ret) { |
1094 | get_bh(mp.mp_bh[0]); | |
1095 | iomap->private = mp.mp_bh[0]; | |
1096 | } | |
1097 | release_metapath(&mp); | |
628e366d AG |
1098 | trace_gfs2_iomap_end(ip, iomap, ret); |
1099 | return ret; | |
1100 | } | |
1101 | ||
64bc06bb AG |
1102 | static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length, |
1103 | ssize_t written, unsigned flags, struct iomap *iomap) | |
1104 | { | |
1105 | struct gfs2_inode *ip = GFS2_I(inode); | |
1106 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1107 | struct gfs2_trans *tr = current->journal_info; | |
1108 | struct buffer_head *dibh = iomap->private; | |
1109 | ||
967bcc91 | 1110 | if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) != IOMAP_WRITE) |
64bc06bb AG |
1111 | goto out; |
1112 | ||
1113 | if (iomap->type != IOMAP_INLINE) { | |
1114 | gfs2_ordered_add_inode(ip); | |
1115 | ||
1116 | if (tr->tr_num_buf_new) | |
1117 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1118 | else | |
1119 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
1120 | } | |
1121 | ||
1122 | if (inode == sdp->sd_rindex) { | |
1123 | adjust_fs_space(inode); | |
1124 | sdp->sd_rindex_uptodate = 0; | |
1125 | } | |
1126 | ||
1127 | gfs2_trans_end(sdp); | |
1128 | gfs2_inplace_release(ip); | |
1129 | ||
1130 | if (length != written && (iomap->flags & IOMAP_F_NEW)) { | |
1131 | /* Deallocate blocks that were just allocated. */ | |
1132 | loff_t blockmask = i_blocksize(inode) - 1; | |
1133 | loff_t end = (pos + length) & ~blockmask; | |
1134 | ||
1135 | pos = (pos + written + blockmask) & ~blockmask; | |
1136 | if (pos < end) { | |
1137 | truncate_pagecache_range(inode, pos, end - 1); | |
1138 | punch_hole(ip, pos, end - pos); | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | if (ip->i_qadata && ip->i_qadata->qa_qd_num) | |
1143 | gfs2_quota_unlock(ip); | |
1144 | gfs2_write_unlock(inode); | |
1145 | ||
1146 | out: | |
1147 | if (dibh) | |
1148 | brelse(dibh); | |
1149 | return 0; | |
1150 | } | |
1151 | ||
628e366d AG |
1152 | const struct iomap_ops gfs2_iomap_ops = { |
1153 | .iomap_begin = gfs2_iomap_begin, | |
64bc06bb | 1154 | .iomap_end = gfs2_iomap_end, |
628e366d AG |
1155 | }; |
1156 | ||
3974320c | 1157 | /** |
d39d18e0 | 1158 | * gfs2_block_map - Map one or more blocks of an inode to a disk block |
3974320c BP |
1159 | * @inode: The inode |
1160 | * @lblock: The logical block number | |
1161 | * @bh_map: The bh to be mapped | |
1162 | * @create: True if its ok to alloc blocks to satify the request | |
1163 | * | |
d39d18e0 AG |
1164 | * The size of the requested mapping is defined in bh_map->b_size. |
1165 | * | |
1166 | * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged | |
1167 | * when @lblock is not mapped. Sets buffer_mapped(bh_map) and | |
1168 | * bh_map->b_size to indicate the size of the mapping when @lblock and | |
1169 | * successive blocks are mapped, up to the requested size. | |
1170 | * | |
1171 | * Sets buffer_boundary() if a read of metadata will be required | |
1172 | * before the next block can be mapped. Sets buffer_new() if new | |
1173 | * blocks were allocated. | |
3974320c BP |
1174 | * |
1175 | * Returns: errno | |
1176 | */ | |
1177 | ||
1178 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
1179 | struct buffer_head *bh_map, int create) | |
1180 | { | |
1181 | struct gfs2_inode *ip = GFS2_I(inode); | |
628e366d AG |
1182 | loff_t pos = (loff_t)lblock << inode->i_blkbits; |
1183 | loff_t length = bh_map->b_size; | |
1184 | struct metapath mp = { .mp_aheight = 1, }; | |
1185 | struct iomap iomap = { }; | |
1186 | int ret; | |
3974320c BP |
1187 | |
1188 | clear_buffer_mapped(bh_map); | |
1189 | clear_buffer_new(bh_map); | |
1190 | clear_buffer_boundary(bh_map); | |
1191 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
1192 | ||
628e366d AG |
1193 | if (create) { |
1194 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, &iomap, &mp); | |
1195 | if (!ret && iomap.type == IOMAP_HOLE) | |
1196 | ret = gfs2_iomap_alloc(inode, &iomap, IOMAP_WRITE, &mp); | |
1197 | release_metapath(&mp); | |
1198 | } else { | |
1199 | ret = gfs2_iomap_get(inode, pos, length, 0, &iomap, &mp); | |
1200 | release_metapath(&mp); | |
3974320c | 1201 | } |
628e366d AG |
1202 | if (ret) |
1203 | goto out; | |
3974320c BP |
1204 | |
1205 | if (iomap.length > bh_map->b_size) { | |
1206 | iomap.length = bh_map->b_size; | |
7ee66c03 | 1207 | iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY; |
5f8bd444 | 1208 | } |
3974320c BP |
1209 | if (iomap.addr != IOMAP_NULL_ADDR) |
1210 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
1211 | bh_map->b_size = iomap.length; | |
7ee66c03 | 1212 | if (iomap.flags & IOMAP_F_GFS2_BOUNDARY) |
3974320c BP |
1213 | set_buffer_boundary(bh_map); |
1214 | if (iomap.flags & IOMAP_F_NEW) | |
1215 | set_buffer_new(bh_map); | |
1216 | ||
1217 | out: | |
1218 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
1219 | return ret; | |
fd88de56 SW |
1220 | } |
1221 | ||
941e6d7d SW |
1222 | /* |
1223 | * Deprecated: do not use in new code | |
1224 | */ | |
fd88de56 SW |
1225 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
1226 | { | |
23591256 | 1227 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 1228 | int ret; |
fd88de56 SW |
1229 | int create = *new; |
1230 | ||
1231 | BUG_ON(!extlen); | |
1232 | BUG_ON(!dblock); | |
1233 | BUG_ON(!new); | |
1234 | ||
47a9a527 | 1235 | bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 1236 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
1237 | *extlen = bh.b_size >> inode->i_blkbits; |
1238 | *dblock = bh.b_blocknr; | |
1239 | if (buffer_new(&bh)) | |
1240 | *new = 1; | |
1241 | else | |
1242 | *new = 0; | |
1243 | return ret; | |
b3b94faa DT |
1244 | } |
1245 | ||
ba7f7290 | 1246 | /** |
bdba0d5e | 1247 | * gfs2_block_zero_range - Deal with zeroing out data |
ba7f7290 SW |
1248 | * |
1249 | * This is partly borrowed from ext3. | |
1250 | */ | |
bdba0d5e AG |
1251 | static int gfs2_block_zero_range(struct inode *inode, loff_t from, |
1252 | unsigned int length) | |
ba7f7290 | 1253 | { |
bdba0d5e | 1254 | struct address_space *mapping = inode->i_mapping; |
ba7f7290 | 1255 | struct gfs2_inode *ip = GFS2_I(inode); |
09cbfeaf KS |
1256 | unsigned long index = from >> PAGE_SHIFT; |
1257 | unsigned offset = from & (PAGE_SIZE-1); | |
bdba0d5e | 1258 | unsigned blocksize, iblock, pos; |
ba7f7290 SW |
1259 | struct buffer_head *bh; |
1260 | struct page *page; | |
ba7f7290 SW |
1261 | int err; |
1262 | ||
220cca2a | 1263 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ba7f7290 SW |
1264 | if (!page) |
1265 | return 0; | |
1266 | ||
1267 | blocksize = inode->i_sb->s_blocksize; | |
09cbfeaf | 1268 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
ba7f7290 SW |
1269 | |
1270 | if (!page_has_buffers(page)) | |
1271 | create_empty_buffers(page, blocksize, 0); | |
1272 | ||
1273 | /* Find the buffer that contains "offset" */ | |
1274 | bh = page_buffers(page); | |
1275 | pos = blocksize; | |
1276 | while (offset >= pos) { | |
1277 | bh = bh->b_this_page; | |
1278 | iblock++; | |
1279 | pos += blocksize; | |
1280 | } | |
1281 | ||
1282 | err = 0; | |
1283 | ||
1284 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 1285 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
1286 | /* unmapped? It's a hole - nothing to do */ |
1287 | if (!buffer_mapped(bh)) | |
1288 | goto unlock; | |
1289 | } | |
1290 | ||
1291 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1292 | if (PageUptodate(page)) | |
1293 | set_buffer_uptodate(bh); | |
1294 | ||
1295 | if (!buffer_uptodate(bh)) { | |
1296 | err = -EIO; | |
dfec8a14 | 1297 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
ba7f7290 SW |
1298 | wait_on_buffer(bh); |
1299 | /* Uhhuh. Read error. Complain and punt. */ | |
1300 | if (!buffer_uptodate(bh)) | |
1301 | goto unlock; | |
1875f2f3 | 1302 | err = 0; |
ba7f7290 SW |
1303 | } |
1304 | ||
845802b1 | 1305 | if (gfs2_is_jdata(ip)) |
350a9b0a | 1306 | gfs2_trans_add_data(ip->i_gl, bh); |
845802b1 AG |
1307 | else |
1308 | gfs2_ordered_add_inode(ip); | |
ba7f7290 | 1309 | |
eebd2aa3 | 1310 | zero_user(page, offset, length); |
40bc9a27 | 1311 | mark_buffer_dirty(bh); |
ba7f7290 SW |
1312 | unlock: |
1313 | unlock_page(page); | |
09cbfeaf | 1314 | put_page(page); |
ba7f7290 SW |
1315 | return err; |
1316 | } | |
1317 | ||
c62baf65 FF |
1318 | #define GFS2_JTRUNC_REVOKES 8192 |
1319 | ||
fa731fc4 SW |
1320 | /** |
1321 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
1322 | * @inode: The inode being truncated | |
1323 | * @oldsize: The original (larger) size | |
1324 | * @newsize: The new smaller size | |
1325 | * | |
1326 | * With jdata files, we have to journal a revoke for each block which is | |
1327 | * truncated. As a result, we need to split this into separate transactions | |
1328 | * if the number of pages being truncated gets too large. | |
1329 | */ | |
1330 | ||
fa731fc4 SW |
1331 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
1332 | { | |
1333 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1334 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1335 | u64 chunk; | |
1336 | int error; | |
1337 | ||
1338 | while (oldsize != newsize) { | |
e7fdf004 AG |
1339 | struct gfs2_trans *tr; |
1340 | unsigned int offs; | |
1341 | ||
fa731fc4 SW |
1342 | chunk = oldsize - newsize; |
1343 | if (chunk > max_chunk) | |
1344 | chunk = max_chunk; | |
e7fdf004 AG |
1345 | |
1346 | offs = oldsize & ~PAGE_MASK; | |
1347 | if (offs && chunk > PAGE_SIZE) | |
1348 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1349 | ||
7caef267 | 1350 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1351 | oldsize -= chunk; |
e7fdf004 AG |
1352 | |
1353 | tr = current->journal_info; | |
1354 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1355 | continue; | |
1356 | ||
fa731fc4 SW |
1357 | gfs2_trans_end(sdp); |
1358 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1359 | if (error) | |
1360 | return error; | |
1361 | } | |
1362 | ||
1363 | return 0; | |
1364 | } | |
1365 | ||
8b5860a3 | 1366 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1367 | { |
ff8f33c8 SW |
1368 | struct gfs2_inode *ip = GFS2_I(inode); |
1369 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
80990f40 | 1370 | struct buffer_head *dibh = NULL; |
b3b94faa | 1371 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1372 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1373 | int error; |
1374 | ||
fa731fc4 SW |
1375 | if (journaled) |
1376 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1377 | else | |
1378 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1379 | if (error) |
1380 | return error; | |
1381 | ||
1382 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1383 | if (error) | |
1384 | goto out; | |
1385 | ||
350a9b0a | 1386 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1387 | |
b3b94faa | 1388 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1389 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1390 | } else { |
bdba0d5e AG |
1391 | unsigned int blocksize = i_blocksize(inode); |
1392 | unsigned int offs = newsize & (blocksize - 1); | |
1393 | if (offs) { | |
1394 | error = gfs2_block_zero_range(inode, newsize, | |
1395 | blocksize - offs); | |
ff8f33c8 | 1396 | if (error) |
80990f40 | 1397 | goto out; |
b3b94faa | 1398 | } |
ff8f33c8 | 1399 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1400 | } |
1401 | ||
ff8f33c8 | 1402 | i_size_write(inode, newsize); |
078cd827 | 1403 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
ff8f33c8 | 1404 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1405 | |
fa731fc4 SW |
1406 | if (journaled) |
1407 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1408 | else | |
7caef267 | 1409 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1410 | |
a91ea69f | 1411 | out: |
80990f40 AG |
1412 | brelse(dibh); |
1413 | if (current->journal_info) | |
1414 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1415 | return error; |
1416 | } | |
1417 | ||
628e366d AG |
1418 | int gfs2_iomap_get_alloc(struct inode *inode, loff_t pos, loff_t length, |
1419 | struct iomap *iomap) | |
1420 | { | |
1421 | struct metapath mp = { .mp_aheight = 1, }; | |
1422 | int ret; | |
1423 | ||
1424 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp); | |
1425 | if (!ret && iomap->type == IOMAP_HOLE) | |
1426 | ret = gfs2_iomap_alloc(inode, iomap, IOMAP_WRITE, &mp); | |
1427 | release_metapath(&mp); | |
1428 | return ret; | |
1429 | } | |
1430 | ||
d552a2b9 BP |
1431 | /** |
1432 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1433 | * @ip: inode | |
1434 | * @rg_gh: holder of resource group glock | |
5cf26b1e AG |
1435 | * @bh: buffer head to sweep |
1436 | * @start: starting point in bh | |
1437 | * @end: end point in bh | |
1438 | * @meta: true if bh points to metadata (rather than data) | |
d552a2b9 | 1439 | * @btotal: place to keep count of total blocks freed |
d552a2b9 BP |
1440 | * |
1441 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1442 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1443 | * block has references to multiple rgrps, we break it into individual | |
1444 | * transactions. This allows other processes to use the rgrps while we're | |
1445 | * focused on a single one, for better concurrency / performance. | |
1446 | * At every transaction boundary, we rewrite the inode into the journal. | |
1447 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1448 | * if we're interrupted by power-outages. | |
1449 | * | |
1450 | * Returns: 0, or return code if an error occurred. | |
1451 | * *btotal has the total number of blocks freed | |
1452 | */ | |
1453 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
5cf26b1e AG |
1454 | struct buffer_head *bh, __be64 *start, __be64 *end, |
1455 | bool meta, u32 *btotal) | |
b3b94faa | 1456 | { |
9b8c81d1 | 1457 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1458 | struct gfs2_rgrpd *rgd; |
1459 | struct gfs2_trans *tr; | |
5cf26b1e | 1460 | __be64 *p; |
d552a2b9 BP |
1461 | int blks_outside_rgrp; |
1462 | u64 bn, bstart, isize_blks; | |
1463 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
d552a2b9 BP |
1464 | int ret = 0; |
1465 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1466 | ||
d552a2b9 | 1467 | more_rgrps: |
5cf26b1e AG |
1468 | rgd = NULL; |
1469 | if (gfs2_holder_initialized(rd_gh)) { | |
1470 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); | |
1471 | gfs2_assert_withdraw(sdp, | |
1472 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1473 | } | |
d552a2b9 BP |
1474 | blks_outside_rgrp = 0; |
1475 | bstart = 0; | |
1476 | blen = 0; | |
d552a2b9 | 1477 | |
5cf26b1e | 1478 | for (p = start; p < end; p++) { |
d552a2b9 BP |
1479 | if (!*p) |
1480 | continue; | |
1481 | bn = be64_to_cpu(*p); | |
5cf26b1e AG |
1482 | |
1483 | if (rgd) { | |
1484 | if (!rgrp_contains_block(rgd, bn)) { | |
1485 | blks_outside_rgrp++; | |
1486 | continue; | |
1487 | } | |
d552a2b9 | 1488 | } else { |
90bcab99 | 1489 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
5cf26b1e AG |
1490 | if (unlikely(!rgd)) { |
1491 | ret = -EIO; | |
1492 | goto out; | |
1493 | } | |
d552a2b9 BP |
1494 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1495 | 0, rd_gh); | |
1496 | if (ret) | |
1497 | goto out; | |
1498 | ||
1499 | /* Must be done with the rgrp glock held: */ | |
1500 | if (gfs2_rs_active(&ip->i_res) && | |
1501 | rgd == ip->i_res.rs_rbm.rgd) | |
1502 | gfs2_rs_deltree(&ip->i_res); | |
1503 | } | |
1504 | ||
d552a2b9 BP |
1505 | /* The size of our transactions will be unknown until we |
1506 | actually process all the metadata blocks that relate to | |
1507 | the rgrp. So we estimate. We know it can't be more than | |
1508 | the dinode's i_blocks and we don't want to exceed the | |
1509 | journal flush threshold, sd_log_thresh2. */ | |
1510 | if (current->journal_info == NULL) { | |
1511 | unsigned int jblocks_rqsted, revokes; | |
1512 | ||
1513 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1514 | RES_INDIRECT; | |
1515 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1516 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1517 | jblocks_rqsted += | |
1518 | atomic_read(&sdp->sd_log_thresh2); | |
1519 | else | |
1520 | jblocks_rqsted += isize_blks; | |
1521 | revokes = jblocks_rqsted; | |
1522 | if (meta) | |
5cf26b1e | 1523 | revokes += end - start; |
d552a2b9 BP |
1524 | else if (ip->i_depth) |
1525 | revokes += sdp->sd_inptrs; | |
1526 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1527 | if (ret) | |
1528 | goto out_unlock; | |
1529 | down_write(&ip->i_rw_mutex); | |
1530 | } | |
1531 | /* check if we will exceed the transaction blocks requested */ | |
1532 | tr = current->journal_info; | |
1533 | if (tr->tr_num_buf_new + RES_STATFS + | |
1534 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1535 | /* We set blks_outside_rgrp to ensure the loop will | |
1536 | be repeated for the same rgrp, but with a new | |
1537 | transaction. */ | |
1538 | blks_outside_rgrp++; | |
1539 | /* This next part is tricky. If the buffer was added | |
1540 | to the transaction, we've already set some block | |
1541 | pointers to 0, so we better follow through and free | |
1542 | them, or we will introduce corruption (so break). | |
1543 | This may be impossible, or at least rare, but I | |
1544 | decided to cover the case regardless. | |
1545 | ||
1546 | If the buffer was not added to the transaction | |
1547 | (this call), doing so would exceed our transaction | |
1548 | size, so we need to end the transaction and start a | |
1549 | new one (so goto). */ | |
1550 | ||
1551 | if (buf_in_tr) | |
1552 | break; | |
1553 | goto out_unlock; | |
1554 | } | |
1555 | ||
1556 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1557 | buf_in_tr = true; | |
1558 | *p = 0; | |
1559 | if (bstart + blen == bn) { | |
1560 | blen++; | |
1561 | continue; | |
1562 | } | |
1563 | if (bstart) { | |
0ddeded4 | 1564 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1565 | (*btotal) += blen; |
1566 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1567 | } | |
1568 | bstart = bn; | |
1569 | blen = 1; | |
1570 | } | |
1571 | if (bstart) { | |
0ddeded4 | 1572 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1573 | (*btotal) += blen; |
1574 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1575 | } | |
1576 | out_unlock: | |
1577 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1578 | outside the rgrp we just processed, | |
1579 | do it all over again. */ | |
1580 | if (current->journal_info) { | |
5cf26b1e AG |
1581 | struct buffer_head *dibh; |
1582 | ||
1583 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1584 | if (ret) | |
1585 | goto out; | |
d552a2b9 BP |
1586 | |
1587 | /* Every transaction boundary, we rewrite the dinode | |
1588 | to keep its di_blocks current in case of failure. */ | |
1589 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = | |
b32c8c76 | 1590 | current_time(&ip->i_inode); |
d552a2b9 BP |
1591 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1592 | gfs2_dinode_out(ip, dibh->b_data); | |
5cf26b1e | 1593 | brelse(dibh); |
d552a2b9 BP |
1594 | up_write(&ip->i_rw_mutex); |
1595 | gfs2_trans_end(sdp); | |
1596 | } | |
1597 | gfs2_glock_dq_uninit(rd_gh); | |
1598 | cond_resched(); | |
1599 | goto more_rgrps; | |
1600 | } | |
1601 | out: | |
1602 | return ret; | |
1603 | } | |
1604 | ||
10d2cf94 AG |
1605 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h) |
1606 | { | |
1607 | if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0]))) | |
1608 | return false; | |
1609 | return true; | |
1610 | } | |
1611 | ||
d552a2b9 BP |
1612 | /** |
1613 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
d552a2b9 BP |
1614 | * @mp: starting metapath |
1615 | * @h: desired height to search | |
1616 | * | |
10d2cf94 | 1617 | * Assumes the metapath is valid (with buffers) out to height h. |
d552a2b9 BP |
1618 | * Returns: true if a non-null pointer was found in the metapath buffer |
1619 | * false if all remaining pointers are NULL in the buffer | |
1620 | */ | |
1621 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
10d2cf94 AG |
1622 | unsigned int h, |
1623 | __u16 *end_list, unsigned int end_aligned) | |
d552a2b9 | 1624 | { |
10d2cf94 AG |
1625 | struct buffer_head *bh = mp->mp_bh[h]; |
1626 | __be64 *first, *ptr, *end; | |
1627 | ||
1628 | first = metaptr1(h, mp); | |
1629 | ptr = first + mp->mp_list[h]; | |
1630 | end = (__be64 *)(bh->b_data + bh->b_size); | |
1631 | if (end_list && mp_eq_to_hgt(mp, end_list, h)) { | |
1632 | bool keep_end = h < end_aligned; | |
1633 | end = first + end_list[h] + keep_end; | |
1634 | } | |
d552a2b9 | 1635 | |
10d2cf94 | 1636 | while (ptr < end) { |
c4a9d189 | 1637 | if (*ptr) { /* if we have a non-null pointer */ |
10d2cf94 | 1638 | mp->mp_list[h] = ptr - first; |
c4a9d189 BP |
1639 | h++; |
1640 | if (h < GFS2_MAX_META_HEIGHT) | |
10d2cf94 | 1641 | mp->mp_list[h] = 0; |
d552a2b9 | 1642 | return true; |
c4a9d189 | 1643 | } |
10d2cf94 | 1644 | ptr++; |
d552a2b9 | 1645 | } |
10d2cf94 | 1646 | return false; |
d552a2b9 BP |
1647 | } |
1648 | ||
1649 | enum dealloc_states { | |
1650 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1651 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1652 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1653 | DEALLOC_DONE = 3, /* process complete */ | |
1654 | }; | |
b3b94faa | 1655 | |
5cf26b1e AG |
1656 | static inline void |
1657 | metapointer_range(struct metapath *mp, int height, | |
1658 | __u16 *start_list, unsigned int start_aligned, | |
10d2cf94 | 1659 | __u16 *end_list, unsigned int end_aligned, |
5cf26b1e AG |
1660 | __be64 **start, __be64 **end) |
1661 | { | |
1662 | struct buffer_head *bh = mp->mp_bh[height]; | |
1663 | __be64 *first; | |
1664 | ||
1665 | first = metaptr1(height, mp); | |
1666 | *start = first; | |
1667 | if (mp_eq_to_hgt(mp, start_list, height)) { | |
1668 | bool keep_start = height < start_aligned; | |
1669 | *start = first + start_list[height] + keep_start; | |
1670 | } | |
1671 | *end = (__be64 *)(bh->b_data + bh->b_size); | |
10d2cf94 AG |
1672 | if (end_list && mp_eq_to_hgt(mp, end_list, height)) { |
1673 | bool keep_end = height < end_aligned; | |
1674 | *end = first + end_list[height] + keep_end; | |
1675 | } | |
1676 | } | |
1677 | ||
1678 | static inline bool walk_done(struct gfs2_sbd *sdp, | |
1679 | struct metapath *mp, int height, | |
1680 | __u16 *end_list, unsigned int end_aligned) | |
1681 | { | |
1682 | __u16 end; | |
1683 | ||
1684 | if (end_list) { | |
1685 | bool keep_end = height < end_aligned; | |
1686 | if (!mp_eq_to_hgt(mp, end_list, height)) | |
1687 | return false; | |
1688 | end = end_list[height] + keep_end; | |
1689 | } else | |
1690 | end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
1691 | return mp->mp_list[height] >= end; | |
5cf26b1e AG |
1692 | } |
1693 | ||
d552a2b9 | 1694 | /** |
10d2cf94 | 1695 | * punch_hole - deallocate blocks in a file |
d552a2b9 | 1696 | * @ip: inode to truncate |
10d2cf94 AG |
1697 | * @offset: the start of the hole |
1698 | * @length: the size of the hole (or 0 for truncate) | |
1699 | * | |
1700 | * Punch a hole into a file or truncate a file at a given position. This | |
1701 | * function operates in whole blocks (@offset and @length are rounded | |
1702 | * accordingly); partially filled blocks must be cleared otherwise. | |
d552a2b9 | 1703 | * |
10d2cf94 AG |
1704 | * This function works from the bottom up, and from the right to the left. In |
1705 | * other words, it strips off the highest layer (data) before stripping any of | |
1706 | * the metadata. Doing it this way is best in case the operation is interrupted | |
1707 | * by power failure, etc. The dinode is rewritten in every transaction to | |
1708 | * guarantee integrity. | |
d552a2b9 | 1709 | */ |
10d2cf94 | 1710 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length) |
d552a2b9 BP |
1711 | { |
1712 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
bb491ce6 | 1713 | u64 maxsize = sdp->sd_heightsize[ip->i_height]; |
10d2cf94 | 1714 | struct metapath mp = {}; |
d552a2b9 BP |
1715 | struct buffer_head *dibh, *bh; |
1716 | struct gfs2_holder rd_gh; | |
cb7f0903 | 1717 | unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; |
10d2cf94 AG |
1718 | u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift; |
1719 | __u16 start_list[GFS2_MAX_META_HEIGHT]; | |
1720 | __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL; | |
4e56a641 | 1721 | unsigned int start_aligned, uninitialized_var(end_aligned); |
d552a2b9 BP |
1722 | unsigned int strip_h = ip->i_height - 1; |
1723 | u32 btotal = 0; | |
1724 | int ret, state; | |
1725 | int mp_h; /* metapath buffers are read in to this height */ | |
d552a2b9 | 1726 | u64 prev_bnr = 0; |
5cf26b1e | 1727 | __be64 *start, *end; |
b3b94faa | 1728 | |
bb491ce6 AG |
1729 | if (offset >= maxsize) { |
1730 | /* | |
1731 | * The starting point lies beyond the allocated meta-data; | |
1732 | * there are no blocks do deallocate. | |
1733 | */ | |
1734 | return 0; | |
1735 | } | |
1736 | ||
10d2cf94 AG |
1737 | /* |
1738 | * The start position of the hole is defined by lblock, start_list, and | |
1739 | * start_aligned. The end position of the hole is defined by lend, | |
1740 | * end_list, and end_aligned. | |
1741 | * | |
1742 | * start_aligned and end_aligned define down to which height the start | |
1743 | * and end positions are aligned to the metadata tree (i.e., the | |
1744 | * position is a multiple of the metadata granularity at the height | |
1745 | * above). This determines at which heights additional meta pointers | |
1746 | * needs to be preserved for the remaining data. | |
1747 | */ | |
b3b94faa | 1748 | |
10d2cf94 | 1749 | if (length) { |
10d2cf94 AG |
1750 | u64 end_offset = offset + length; |
1751 | u64 lend; | |
1752 | ||
1753 | /* | |
1754 | * Clip the end at the maximum file size for the given height: | |
1755 | * that's how far the metadata goes; files bigger than that | |
1756 | * will have additional layers of indirection. | |
1757 | */ | |
1758 | if (end_offset > maxsize) | |
1759 | end_offset = maxsize; | |
1760 | lend = end_offset >> bsize_shift; | |
1761 | ||
1762 | if (lblock >= lend) | |
1763 | return 0; | |
1764 | ||
1765 | find_metapath(sdp, lend, &mp, ip->i_height); | |
1766 | end_list = __end_list; | |
1767 | memcpy(end_list, mp.mp_list, sizeof(mp.mp_list)); | |
1768 | ||
1769 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { | |
1770 | if (end_list[mp_h]) | |
1771 | break; | |
1772 | } | |
1773 | end_aligned = mp_h; | |
1774 | } | |
1775 | ||
1776 | find_metapath(sdp, lblock, &mp, ip->i_height); | |
cb7f0903 AG |
1777 | memcpy(start_list, mp.mp_list, sizeof(start_list)); |
1778 | ||
cb7f0903 AG |
1779 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { |
1780 | if (start_list[mp_h]) | |
1781 | break; | |
1782 | } | |
1783 | start_aligned = mp_h; | |
d552a2b9 BP |
1784 | |
1785 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1786 | if (ret) | |
1787 | return ret; | |
b3b94faa | 1788 | |
d552a2b9 BP |
1789 | mp.mp_bh[0] = dibh; |
1790 | ret = lookup_metapath(ip, &mp); | |
e8b43fe0 AG |
1791 | if (ret) |
1792 | goto out_metapath; | |
c3ce5aa9 AG |
1793 | |
1794 | /* issue read-ahead on metadata */ | |
5cf26b1e AG |
1795 | for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) { |
1796 | metapointer_range(&mp, mp_h, start_list, start_aligned, | |
10d2cf94 | 1797 | end_list, end_aligned, &start, &end); |
5cf26b1e AG |
1798 | gfs2_metapath_ra(ip->i_gl, start, end); |
1799 | } | |
c3ce5aa9 | 1800 | |
e8b43fe0 | 1801 | if (mp.mp_aheight == ip->i_height) |
d552a2b9 BP |
1802 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ |
1803 | else | |
1804 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1805 | |
d552a2b9 BP |
1806 | ret = gfs2_rindex_update(sdp); |
1807 | if (ret) | |
1808 | goto out_metapath; | |
1809 | ||
1810 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1811 | if (ret) | |
1812 | goto out_metapath; | |
1813 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1814 | ||
1815 | mp_h = strip_h; | |
1816 | ||
1817 | while (state != DEALLOC_DONE) { | |
1818 | switch (state) { | |
1819 | /* Truncate a full metapath at the given strip height. | |
1820 | * Note that strip_h == mp_h in order to be in this state. */ | |
1821 | case DEALLOC_MP_FULL: | |
d552a2b9 BP |
1822 | bh = mp.mp_bh[mp_h]; |
1823 | gfs2_assert_withdraw(sdp, bh); | |
1824 | if (gfs2_assert_withdraw(sdp, | |
1825 | prev_bnr != bh->b_blocknr)) { | |
1826 | printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, " | |
1827 | "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n", | |
1828 | sdp->sd_fsname, | |
1829 | (unsigned long long)ip->i_no_addr, | |
1830 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1831 | } | |
1832 | prev_bnr = bh->b_blocknr; | |
cb7f0903 | 1833 | |
5cf26b1e AG |
1834 | if (gfs2_metatype_check(sdp, bh, |
1835 | (mp_h ? GFS2_METATYPE_IN : | |
1836 | GFS2_METATYPE_DI))) { | |
1837 | ret = -EIO; | |
1838 | goto out; | |
1839 | } | |
1840 | ||
10d2cf94 AG |
1841 | /* |
1842 | * Below, passing end_aligned as 0 gives us the | |
1843 | * metapointer range excluding the end point: the end | |
1844 | * point is the first metapath we must not deallocate! | |
1845 | */ | |
1846 | ||
5cf26b1e | 1847 | metapointer_range(&mp, mp_h, start_list, start_aligned, |
10d2cf94 | 1848 | end_list, 0 /* end_aligned */, |
5cf26b1e AG |
1849 | &start, &end); |
1850 | ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h], | |
1851 | start, end, | |
1852 | mp_h != ip->i_height - 1, | |
1853 | &btotal); | |
cb7f0903 | 1854 | |
d552a2b9 BP |
1855 | /* If we hit an error or just swept dinode buffer, |
1856 | just exit. */ | |
1857 | if (ret || !mp_h) { | |
1858 | state = DEALLOC_DONE; | |
1859 | break; | |
1860 | } | |
1861 | state = DEALLOC_MP_LOWER; | |
1862 | break; | |
1863 | ||
1864 | /* lower the metapath strip height */ | |
1865 | case DEALLOC_MP_LOWER: | |
1866 | /* We're done with the current buffer, so release it, | |
1867 | unless it's the dinode buffer. Then back up to the | |
1868 | previous pointer. */ | |
1869 | if (mp_h) { | |
1870 | brelse(mp.mp_bh[mp_h]); | |
1871 | mp.mp_bh[mp_h] = NULL; | |
1872 | } | |
1873 | /* If we can't get any lower in height, we've stripped | |
1874 | off all we can. Next step is to back up and start | |
1875 | stripping the previous level of metadata. */ | |
1876 | if (mp_h == 0) { | |
1877 | strip_h--; | |
cb7f0903 | 1878 | memcpy(mp.mp_list, start_list, sizeof(start_list)); |
d552a2b9 BP |
1879 | mp_h = strip_h; |
1880 | state = DEALLOC_FILL_MP; | |
1881 | break; | |
1882 | } | |
1883 | mp.mp_list[mp_h] = 0; | |
1884 | mp_h--; /* search one metadata height down */ | |
d552a2b9 | 1885 | mp.mp_list[mp_h]++; |
10d2cf94 AG |
1886 | if (walk_done(sdp, &mp, mp_h, end_list, end_aligned)) |
1887 | break; | |
d552a2b9 BP |
1888 | /* Here we've found a part of the metapath that is not |
1889 | * allocated. We need to search at that height for the | |
1890 | * next non-null pointer. */ | |
10d2cf94 | 1891 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) { |
d552a2b9 BP |
1892 | state = DEALLOC_FILL_MP; |
1893 | mp_h++; | |
1894 | } | |
1895 | /* No more non-null pointers at this height. Back up | |
1896 | to the previous height and try again. */ | |
1897 | break; /* loop around in the same state */ | |
1898 | ||
1899 | /* Fill the metapath with buffers to the given height. */ | |
1900 | case DEALLOC_FILL_MP: | |
1901 | /* Fill the buffers out to the current height. */ | |
1902 | ret = fillup_metapath(ip, &mp, mp_h); | |
c3ce5aa9 | 1903 | if (ret < 0) |
d552a2b9 | 1904 | goto out; |
c3ce5aa9 | 1905 | |
e7445ced AG |
1906 | /* On the first pass, issue read-ahead on metadata. */ |
1907 | if (mp.mp_aheight > 1 && strip_h == ip->i_height - 1) { | |
1908 | unsigned int height = mp.mp_aheight - 1; | |
1909 | ||
1910 | /* No read-ahead for data blocks. */ | |
1911 | if (mp.mp_aheight - 1 == strip_h) | |
1912 | height--; | |
1913 | ||
1914 | for (; height >= mp.mp_aheight - ret; height--) { | |
1915 | metapointer_range(&mp, height, | |
5cf26b1e | 1916 | start_list, start_aligned, |
10d2cf94 | 1917 | end_list, end_aligned, |
5cf26b1e AG |
1918 | &start, &end); |
1919 | gfs2_metapath_ra(ip->i_gl, start, end); | |
1920 | } | |
c3ce5aa9 | 1921 | } |
d552a2b9 BP |
1922 | |
1923 | /* If buffers found for the entire strip height */ | |
e8b43fe0 | 1924 | if (mp.mp_aheight - 1 == strip_h) { |
d552a2b9 BP |
1925 | state = DEALLOC_MP_FULL; |
1926 | break; | |
1927 | } | |
e8b43fe0 AG |
1928 | if (mp.mp_aheight < ip->i_height) /* We have a partial height */ |
1929 | mp_h = mp.mp_aheight - 1; | |
d552a2b9 BP |
1930 | |
1931 | /* If we find a non-null block pointer, crawl a bit | |
1932 | higher up in the metapath and try again, otherwise | |
1933 | we need to look lower for a new starting point. */ | |
10d2cf94 | 1934 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) |
d552a2b9 BP |
1935 | mp_h++; |
1936 | else | |
1937 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1938 | break; |
d552a2b9 | 1939 | } |
b3b94faa DT |
1940 | } |
1941 | ||
d552a2b9 BP |
1942 | if (btotal) { |
1943 | if (current->journal_info == NULL) { | |
1944 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1945 | RES_QUOTA, 0); | |
1946 | if (ret) | |
1947 | goto out; | |
1948 | down_write(&ip->i_rw_mutex); | |
1949 | } | |
1950 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
1951 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
1952 | ip->i_inode.i_gid); | |
b32c8c76 | 1953 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
d552a2b9 BP |
1954 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1955 | gfs2_dinode_out(ip, dibh->b_data); | |
1956 | up_write(&ip->i_rw_mutex); | |
1957 | gfs2_trans_end(sdp); | |
1958 | } | |
b3b94faa | 1959 | |
d552a2b9 BP |
1960 | out: |
1961 | if (gfs2_holder_initialized(&rd_gh)) | |
1962 | gfs2_glock_dq_uninit(&rd_gh); | |
1963 | if (current->journal_info) { | |
1964 | up_write(&ip->i_rw_mutex); | |
1965 | gfs2_trans_end(sdp); | |
1966 | cond_resched(); | |
1967 | } | |
1968 | gfs2_quota_unhold(ip); | |
1969 | out_metapath: | |
1970 | release_metapath(&mp); | |
1971 | return ret; | |
b3b94faa DT |
1972 | } |
1973 | ||
1974 | static int trunc_end(struct gfs2_inode *ip) | |
1975 | { | |
feaa7bba | 1976 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1977 | struct buffer_head *dibh; |
1978 | int error; | |
1979 | ||
1980 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1981 | if (error) | |
1982 | return error; | |
1983 | ||
1984 | down_write(&ip->i_rw_mutex); | |
1985 | ||
1986 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1987 | if (error) | |
1988 | goto out; | |
1989 | ||
a2e0f799 | 1990 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1991 | ip->i_height = 0; |
ce276b06 | 1992 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 1993 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 1994 | gfs2_ordered_del_inode(ip); |
b3b94faa | 1995 | } |
078cd827 | 1996 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
383f01fb | 1997 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1998 | |
350a9b0a | 1999 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 2000 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
2001 | brelse(dibh); |
2002 | ||
a91ea69f | 2003 | out: |
b3b94faa | 2004 | up_write(&ip->i_rw_mutex); |
b3b94faa | 2005 | gfs2_trans_end(sdp); |
b3b94faa DT |
2006 | return error; |
2007 | } | |
2008 | ||
2009 | /** | |
2010 | * do_shrink - make a file smaller | |
ff8f33c8 | 2011 | * @inode: the inode |
ff8f33c8 | 2012 | * @newsize: the size to make the file |
b3b94faa | 2013 | * |
ff8f33c8 SW |
2014 | * Called with an exclusive lock on @inode. The @size must |
2015 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
2016 | * |
2017 | * Returns: errno | |
2018 | */ | |
2019 | ||
8b5860a3 | 2020 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 2021 | { |
ff8f33c8 | 2022 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
2023 | int error; |
2024 | ||
8b5860a3 | 2025 | error = trunc_start(inode, newsize); |
b3b94faa DT |
2026 | if (error < 0) |
2027 | return error; | |
ff8f33c8 | 2028 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
2029 | return 0; |
2030 | ||
10d2cf94 | 2031 | error = punch_hole(ip, newsize, 0); |
ff8f33c8 | 2032 | if (error == 0) |
b3b94faa DT |
2033 | error = trunc_end(ip); |
2034 | ||
2035 | return error; | |
2036 | } | |
2037 | ||
ff8f33c8 | 2038 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 2039 | { |
ff8f33c8 SW |
2040 | int ret; |
2041 | ||
8b5860a3 | 2042 | ret = do_shrink(inode, inode->i_size); |
ff8f33c8 SW |
2043 | WARN_ON(ret != 0); |
2044 | } | |
2045 | ||
2046 | /** | |
2047 | * do_grow - Touch and update inode size | |
2048 | * @inode: The inode | |
2049 | * @size: The new size | |
2050 | * | |
2051 | * This function updates the timestamps on the inode and | |
2052 | * may also increase the size of the inode. This function | |
2053 | * must not be called with @size any smaller than the current | |
2054 | * inode size. | |
2055 | * | |
2056 | * Although it is not strictly required to unstuff files here, | |
2057 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
2058 | * code which will result in a buffer overrun if the size is larger | |
2059 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 2060 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
2061 | * just update the inode size directly. |
2062 | * | |
2063 | * Returns: 0 on success, or -ve on error | |
2064 | */ | |
2065 | ||
2066 | static int do_grow(struct inode *inode, u64 size) | |
2067 | { | |
2068 | struct gfs2_inode *ip = GFS2_I(inode); | |
2069 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 2070 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
2071 | struct buffer_head *dibh; |
2072 | int error; | |
2f7ee358 | 2073 | int unstuff = 0; |
a13b8c5f | 2074 | |
235628c5 | 2075 | if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) { |
b8fbf471 | 2076 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 2077 | if (error) |
5407e242 | 2078 | return error; |
ff8f33c8 | 2079 | |
7b9cff46 | 2080 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
2081 | if (error) |
2082 | goto do_grow_qunlock; | |
2f7ee358 | 2083 | unstuff = 1; |
ff8f33c8 SW |
2084 | } |
2085 | ||
a01aedfe | 2086 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
bc020561 BP |
2087 | (unstuff && |
2088 | gfs2_is_jdata(ip) ? RES_JDATA : 0) + | |
a01aedfe BP |
2089 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? |
2090 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 2091 | if (error) |
ff8f33c8 | 2092 | goto do_grow_release; |
a13b8c5f | 2093 | |
2f7ee358 | 2094 | if (unstuff) { |
ff8f33c8 SW |
2095 | error = gfs2_unstuff_dinode(ip, NULL); |
2096 | if (error) | |
2097 | goto do_end_trans; | |
2098 | } | |
a13b8c5f WC |
2099 | |
2100 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2101 | if (error) | |
ff8f33c8 | 2102 | goto do_end_trans; |
a13b8c5f | 2103 | |
ff8f33c8 | 2104 | i_size_write(inode, size); |
078cd827 | 2105 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
350a9b0a | 2106 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
2107 | gfs2_dinode_out(ip, dibh->b_data); |
2108 | brelse(dibh); | |
2109 | ||
ff8f33c8 | 2110 | do_end_trans: |
a13b8c5f | 2111 | gfs2_trans_end(sdp); |
ff8f33c8 | 2112 | do_grow_release: |
2f7ee358 | 2113 | if (unstuff) { |
ff8f33c8 SW |
2114 | gfs2_inplace_release(ip); |
2115 | do_grow_qunlock: | |
2116 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 2117 | } |
a13b8c5f WC |
2118 | return error; |
2119 | } | |
2120 | ||
b3b94faa | 2121 | /** |
ff8f33c8 SW |
2122 | * gfs2_setattr_size - make a file a given size |
2123 | * @inode: the inode | |
2124 | * @newsize: the size to make the file | |
b3b94faa | 2125 | * |
ff8f33c8 | 2126 | * The file size can grow, shrink, or stay the same size. This |
3e7aafc3 | 2127 | * is called holding i_rwsem and an exclusive glock on the inode |
ff8f33c8 | 2128 | * in question. |
b3b94faa DT |
2129 | * |
2130 | * Returns: errno | |
2131 | */ | |
2132 | ||
ff8f33c8 | 2133 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 2134 | { |
af5c2697 | 2135 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 2136 | int ret; |
b3b94faa | 2137 | |
ff8f33c8 | 2138 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 2139 | |
ff8f33c8 SW |
2140 | ret = inode_newsize_ok(inode, newsize); |
2141 | if (ret) | |
2142 | return ret; | |
b3b94faa | 2143 | |
562c72aa CH |
2144 | inode_dio_wait(inode); |
2145 | ||
b54e9a0b | 2146 | ret = gfs2_rsqa_alloc(ip); |
d2b47cfb | 2147 | if (ret) |
2b3dcf35 | 2148 | goto out; |
d2b47cfb | 2149 | |
8b5860a3 | 2150 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
2151 | ret = do_grow(inode, newsize); |
2152 | goto out; | |
2153 | } | |
ff8f33c8 | 2154 | |
8b5860a3 | 2155 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 2156 | out: |
a097dc7e | 2157 | gfs2_rsqa_delete(ip, NULL); |
2b3dcf35 | 2158 | return ret; |
b3b94faa DT |
2159 | } |
2160 | ||
2161 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
2162 | { | |
2163 | int error; | |
10d2cf94 | 2164 | error = punch_hole(ip, i_size_read(&ip->i_inode), 0); |
b3b94faa DT |
2165 | if (!error) |
2166 | error = trunc_end(ip); | |
2167 | return error; | |
2168 | } | |
2169 | ||
2170 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
2171 | { | |
10d2cf94 | 2172 | return punch_hole(ip, 0, 0); |
b3b94faa DT |
2173 | } |
2174 | ||
b50f227b SW |
2175 | /** |
2176 | * gfs2_free_journal_extents - Free cached journal bmap info | |
2177 | * @jd: The journal | |
2178 | * | |
2179 | */ | |
2180 | ||
2181 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
2182 | { | |
2183 | struct gfs2_journal_extent *jext; | |
2184 | ||
2185 | while(!list_empty(&jd->extent_list)) { | |
2186 | jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list); | |
2187 | list_del(&jext->list); | |
2188 | kfree(jext); | |
2189 | } | |
2190 | } | |
2191 | ||
2192 | /** | |
2193 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
2194 | * @jd: The journal descriptor | |
2195 | * @lblock: The logical block at start of new extent | |
c62baf65 | 2196 | * @dblock: The physical block at start of new extent |
b50f227b SW |
2197 | * @blocks: Size of extent in fs blocks |
2198 | * | |
2199 | * Returns: 0 on success or -ENOMEM | |
2200 | */ | |
2201 | ||
2202 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
2203 | { | |
2204 | struct gfs2_journal_extent *jext; | |
2205 | ||
2206 | if (!list_empty(&jd->extent_list)) { | |
2207 | jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list); | |
2208 | if ((jext->dblock + jext->blocks) == dblock) { | |
2209 | jext->blocks += blocks; | |
2210 | return 0; | |
2211 | } | |
2212 | } | |
2213 | ||
2214 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
2215 | if (jext == NULL) | |
2216 | return -ENOMEM; | |
2217 | jext->dblock = dblock; | |
2218 | jext->lblock = lblock; | |
2219 | jext->blocks = blocks; | |
2220 | list_add_tail(&jext->list, &jd->extent_list); | |
2221 | jd->nr_extents++; | |
2222 | return 0; | |
2223 | } | |
2224 | ||
2225 | /** | |
2226 | * gfs2_map_journal_extents - Cache journal bmap info | |
2227 | * @sdp: The super block | |
2228 | * @jd: The journal to map | |
2229 | * | |
2230 | * Create a reusable "extent" mapping from all logical | |
2231 | * blocks to all physical blocks for the given journal. This will save | |
2232 | * us time when writing journal blocks. Most journals will have only one | |
2233 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
2234 | * arranges the journal blocks sequentially to maximize performance. | |
2235 | * So the extent would map the first block for the entire file length. | |
2236 | * However, gfs2_jadd can happen while file activity is happening, so | |
2237 | * those journals may not be sequential. Less likely is the case where | |
2238 | * the users created their own journals by mounting the metafs and | |
2239 | * laying it out. But it's still possible. These journals might have | |
2240 | * several extents. | |
2241 | * | |
2242 | * Returns: 0 on success, or error on failure | |
2243 | */ | |
2244 | ||
2245 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
2246 | { | |
2247 | u64 lblock = 0; | |
2248 | u64 lblock_stop; | |
2249 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
2250 | struct buffer_head bh; | |
2251 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
2252 | u64 size; | |
2253 | int rc; | |
98583b3e | 2254 | ktime_t start, end; |
b50f227b | 2255 | |
98583b3e | 2256 | start = ktime_get(); |
b50f227b SW |
2257 | lblock_stop = i_size_read(jd->jd_inode) >> shift; |
2258 | size = (lblock_stop - lblock) << shift; | |
2259 | jd->nr_extents = 0; | |
2260 | WARN_ON(!list_empty(&jd->extent_list)); | |
2261 | ||
2262 | do { | |
2263 | bh.b_state = 0; | |
2264 | bh.b_blocknr = 0; | |
2265 | bh.b_size = size; | |
2266 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
2267 | if (rc || !buffer_mapped(&bh)) | |
2268 | goto fail; | |
2269 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
2270 | if (rc) | |
2271 | goto fail; | |
2272 | size -= bh.b_size; | |
2273 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2274 | } while(size > 0); | |
2275 | ||
98583b3e AD |
2276 | end = ktime_get(); |
2277 | fs_info(sdp, "journal %d mapped with %u extents in %lldms\n", jd->jd_jid, | |
2278 | jd->nr_extents, ktime_ms_delta(end, start)); | |
b50f227b SW |
2279 | return 0; |
2280 | ||
2281 | fail: | |
2282 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
2283 | rc, jd->jd_jid, | |
2284 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
2285 | jd->nr_extents); | |
2286 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
2287 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
2288 | bh.b_state, (unsigned long long)bh.b_size); | |
2289 | gfs2_free_journal_extents(jd); | |
2290 | return rc; | |
2291 | } | |
2292 | ||
b3b94faa DT |
2293 | /** |
2294 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
2295 | * @ip: the file being written to | |
2296 | * @offset: the offset to write to | |
2297 | * @len: the number of bytes being written | |
b3b94faa | 2298 | * |
461cb419 | 2299 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
2300 | */ |
2301 | ||
cd915493 | 2302 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 2303 | unsigned int len) |
b3b94faa | 2304 | { |
feaa7bba | 2305 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
2306 | struct buffer_head bh; |
2307 | unsigned int shift; | |
2308 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 2309 | u64 end_of_file; |
b3b94faa | 2310 | |
b3b94faa DT |
2311 | if (!len) |
2312 | return 0; | |
2313 | ||
2314 | if (gfs2_is_stuffed(ip)) { | |
235628c5 | 2315 | if (offset + len > gfs2_max_stuffed_size(ip)) |
461cb419 | 2316 | return 1; |
b3b94faa DT |
2317 | return 0; |
2318 | } | |
2319 | ||
941e6d7d | 2320 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 2321 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 2322 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
2323 | lblock = offset >> shift; |
2324 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
77612578 | 2325 | if (lblock_stop > end_of_file && ip != GFS2_I(sdp->sd_rindex)) |
461cb419 | 2326 | return 1; |
b3b94faa | 2327 | |
941e6d7d SW |
2328 | size = (lblock_stop - lblock) << shift; |
2329 | do { | |
2330 | bh.b_state = 0; | |
2331 | bh.b_size = size; | |
2332 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
2333 | if (!buffer_mapped(&bh)) | |
461cb419 | 2334 | return 1; |
941e6d7d SW |
2335 | size -= bh.b_size; |
2336 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2337 | } while(size > 0); | |
b3b94faa DT |
2338 | |
2339 | return 0; | |
2340 | } | |
2341 | ||
4e56a641 AG |
2342 | static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length) |
2343 | { | |
2344 | struct gfs2_inode *ip = GFS2_I(inode); | |
2345 | struct buffer_head *dibh; | |
2346 | int error; | |
2347 | ||
2348 | if (offset >= inode->i_size) | |
2349 | return 0; | |
2350 | if (offset + length > inode->i_size) | |
2351 | length = inode->i_size - offset; | |
2352 | ||
2353 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2354 | if (error) | |
2355 | return error; | |
2356 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
2357 | memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0, | |
2358 | length); | |
2359 | brelse(dibh); | |
2360 | return 0; | |
2361 | } | |
2362 | ||
2363 | static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset, | |
2364 | loff_t length) | |
2365 | { | |
2366 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2367 | loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
2368 | int error; | |
2369 | ||
2370 | while (length) { | |
2371 | struct gfs2_trans *tr; | |
2372 | loff_t chunk; | |
2373 | unsigned int offs; | |
2374 | ||
2375 | chunk = length; | |
2376 | if (chunk > max_chunk) | |
2377 | chunk = max_chunk; | |
2378 | ||
2379 | offs = offset & ~PAGE_MASK; | |
2380 | if (offs && chunk > PAGE_SIZE) | |
2381 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
2382 | ||
2383 | truncate_pagecache_range(inode, offset, chunk); | |
2384 | offset += chunk; | |
2385 | length -= chunk; | |
2386 | ||
2387 | tr = current->journal_info; | |
2388 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
2389 | continue; | |
2390 | ||
2391 | gfs2_trans_end(sdp); | |
2392 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
2393 | if (error) | |
2394 | return error; | |
2395 | } | |
2396 | return 0; | |
2397 | } | |
2398 | ||
2399 | int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length) | |
2400 | { | |
2401 | struct inode *inode = file_inode(file); | |
2402 | struct gfs2_inode *ip = GFS2_I(inode); | |
2403 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2404 | int error; | |
2405 | ||
2406 | if (gfs2_is_jdata(ip)) | |
2407 | error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA, | |
2408 | GFS2_JTRUNC_REVOKES); | |
2409 | else | |
2410 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2411 | if (error) | |
2412 | return error; | |
2413 | ||
2414 | if (gfs2_is_stuffed(ip)) { | |
2415 | error = stuffed_zero_range(inode, offset, length); | |
2416 | if (error) | |
2417 | goto out; | |
2418 | } else { | |
00251a16 | 2419 | unsigned int start_off, end_len, blocksize; |
4e56a641 AG |
2420 | |
2421 | blocksize = i_blocksize(inode); | |
2422 | start_off = offset & (blocksize - 1); | |
00251a16 | 2423 | end_len = (offset + length) & (blocksize - 1); |
4e56a641 AG |
2424 | if (start_off) { |
2425 | unsigned int len = length; | |
2426 | if (length > blocksize - start_off) | |
2427 | len = blocksize - start_off; | |
2428 | error = gfs2_block_zero_range(inode, offset, len); | |
2429 | if (error) | |
2430 | goto out; | |
2431 | if (start_off + length < blocksize) | |
00251a16 | 2432 | end_len = 0; |
4e56a641 | 2433 | } |
00251a16 | 2434 | if (end_len) { |
4e56a641 | 2435 | error = gfs2_block_zero_range(inode, |
00251a16 | 2436 | offset + length - end_len, end_len); |
4e56a641 AG |
2437 | if (error) |
2438 | goto out; | |
2439 | } | |
2440 | } | |
2441 | ||
2442 | if (gfs2_is_jdata(ip)) { | |
2443 | BUG_ON(!current->journal_info); | |
2444 | gfs2_journaled_truncate_range(inode, offset, length); | |
2445 | } else | |
2446 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2447 | ||
2448 | file_update_time(file); | |
2449 | mark_inode_dirty(inode); | |
2450 | ||
2451 | if (current->journal_info) | |
2452 | gfs2_trans_end(sdp); | |
2453 | ||
2454 | if (!gfs2_is_stuffed(ip)) | |
2455 | error = punch_hole(ip, offset, length); | |
2456 | ||
2457 | out: | |
2458 | if (current->journal_info) | |
2459 | gfs2_trans_end(sdp); | |
2460 | return error; | |
2461 | } |