Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
b3b94faa DT |
10 | #include <linux/spinlock.h> |
11 | #include <linux/completion.h> | |
12 | #include <linux/buffer_head.h> | |
64dd153c | 13 | #include <linux/blkdev.h> |
5c676f6d | 14 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 15 | #include <linux/crc32.h> |
3974320c | 16 | #include <linux/iomap.h> |
b3b94faa DT |
17 | |
18 | #include "gfs2.h" | |
5c676f6d | 19 | #include "incore.h" |
b3b94faa DT |
20 | #include "bmap.h" |
21 | #include "glock.h" | |
22 | #include "inode.h" | |
b3b94faa | 23 | #include "meta_io.h" |
b3b94faa DT |
24 | #include "quota.h" |
25 | #include "rgrp.h" | |
45138990 | 26 | #include "log.h" |
4c16c36a | 27 | #include "super.h" |
b3b94faa | 28 | #include "trans.h" |
18ec7d5c | 29 | #include "dir.h" |
5c676f6d | 30 | #include "util.h" |
63997775 | 31 | #include "trace_gfs2.h" |
b3b94faa DT |
32 | |
33 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
34 | * block is 512, so __u16 is fine for that. It saves stack space to | |
35 | * keep it small. | |
36 | */ | |
37 | struct metapath { | |
dbac6710 | 38 | struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT]; |
b3b94faa | 39 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; |
5f8bd444 BP |
40 | int mp_fheight; /* find_metapath height */ |
41 | int mp_aheight; /* actual height (lookup height) */ | |
b3b94faa DT |
42 | }; |
43 | ||
f25ef0c1 SW |
44 | /** |
45 | * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page | |
46 | * @ip: the inode | |
47 | * @dibh: the dinode buffer | |
48 | * @block: the block number that was allocated | |
ff8f33c8 | 49 | * @page: The (optional) page. This is looked up if @page is NULL |
f25ef0c1 SW |
50 | * |
51 | * Returns: errno | |
52 | */ | |
53 | ||
54 | static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, | |
cd915493 | 55 | u64 block, struct page *page) |
f25ef0c1 | 56 | { |
f25ef0c1 SW |
57 | struct inode *inode = &ip->i_inode; |
58 | struct buffer_head *bh; | |
59 | int release = 0; | |
60 | ||
61 | if (!page || page->index) { | |
220cca2a | 62 | page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS); |
f25ef0c1 SW |
63 | if (!page) |
64 | return -ENOMEM; | |
65 | release = 1; | |
66 | } | |
67 | ||
68 | if (!PageUptodate(page)) { | |
69 | void *kaddr = kmap(page); | |
602c89d2 SW |
70 | u64 dsize = i_size_read(inode); |
71 | ||
235628c5 AG |
72 | if (dsize > gfs2_max_stuffed_size(ip)) |
73 | dsize = gfs2_max_stuffed_size(ip); | |
f25ef0c1 | 74 | |
602c89d2 | 75 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 76 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
f25ef0c1 SW |
77 | kunmap(page); |
78 | ||
79 | SetPageUptodate(page); | |
80 | } | |
81 | ||
82 | if (!page_has_buffers(page)) | |
47a9a527 FF |
83 | create_empty_buffers(page, BIT(inode->i_blkbits), |
84 | BIT(BH_Uptodate)); | |
f25ef0c1 SW |
85 | |
86 | bh = page_buffers(page); | |
87 | ||
88 | if (!buffer_mapped(bh)) | |
89 | map_bh(bh, inode->i_sb, block); | |
90 | ||
91 | set_buffer_uptodate(bh); | |
845802b1 | 92 | if (gfs2_is_jdata(ip)) |
350a9b0a | 93 | gfs2_trans_add_data(ip->i_gl, bh); |
845802b1 AG |
94 | else { |
95 | mark_buffer_dirty(bh); | |
96 | gfs2_ordered_add_inode(ip); | |
97 | } | |
f25ef0c1 SW |
98 | |
99 | if (release) { | |
100 | unlock_page(page); | |
09cbfeaf | 101 | put_page(page); |
f25ef0c1 SW |
102 | } |
103 | ||
104 | return 0; | |
105 | } | |
106 | ||
b3b94faa DT |
107 | /** |
108 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
109 | * @ip: The GFS2 inode to unstuff | |
ff8f33c8 | 110 | * @page: The (optional) page. This is looked up if the @page is NULL |
b3b94faa DT |
111 | * |
112 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
113 | * that the height can be grown in the traditional way. | |
114 | * | |
115 | * Returns: errno | |
116 | */ | |
117 | ||
f25ef0c1 | 118 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
119 | { |
120 | struct buffer_head *bh, *dibh; | |
48516ced | 121 | struct gfs2_dinode *di; |
cd915493 | 122 | u64 block = 0; |
18ec7d5c | 123 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
124 | int error; |
125 | ||
126 | down_write(&ip->i_rw_mutex); | |
127 | ||
128 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
129 | if (error) | |
130 | goto out; | |
907b9bce | 131 | |
a2e0f799 | 132 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
133 | /* Get a free block, fill it with the stuffed data, |
134 | and write it out to disk */ | |
135 | ||
b45e41d7 | 136 | unsigned int n = 1; |
6e87ed0f | 137 | error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL); |
09010978 SW |
138 | if (error) |
139 | goto out_brelse; | |
18ec7d5c | 140 | if (isdir) { |
5731be53 | 141 | gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 142 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
143 | if (error) |
144 | goto out_brelse; | |
48516ced | 145 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
146 | dibh, sizeof(struct gfs2_dinode)); |
147 | brelse(bh); | |
148 | } else { | |
f25ef0c1 | 149 | error = gfs2_unstuffer_page(ip, dibh, block, page); |
b3b94faa DT |
150 | if (error) |
151 | goto out_brelse; | |
152 | } | |
153 | } | |
154 | ||
155 | /* Set up the pointer to the new block */ | |
156 | ||
350a9b0a | 157 | gfs2_trans_add_meta(ip->i_gl, dibh); |
48516ced | 158 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
159 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
160 | ||
a2e0f799 | 161 | if (i_size_read(&ip->i_inode)) { |
48516ced | 162 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
163 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
164 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
165 | } |
166 | ||
ecc30c79 | 167 | ip->i_height = 1; |
48516ced | 168 | di->di_height = cpu_to_be16(1); |
b3b94faa | 169 | |
a91ea69f | 170 | out_brelse: |
b3b94faa | 171 | brelse(dibh); |
a91ea69f | 172 | out: |
b3b94faa | 173 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
174 | return error; |
175 | } | |
176 | ||
b3b94faa DT |
177 | |
178 | /** | |
179 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 180 | * @sdp: The superblock |
b3b94faa | 181 | * @block: The disk block to look up |
07e23d68 | 182 | * @mp: The metapath to return the result in |
9b8c81d1 | 183 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
184 | * |
185 | * This routine returns a struct metapath structure that defines a path | |
186 | * through the metadata of inode "ip" to get to block "block". | |
187 | * | |
188 | * Example: | |
189 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
190 | * filesystem with a blocksize of 4096. | |
191 | * | |
192 | * find_metapath() would return a struct metapath structure set to: | |
07e23d68 | 193 | * mp_fheight = 3, mp_list[0] = 0, mp_list[1] = 48, and mp_list[2] = 165. |
b3b94faa DT |
194 | * |
195 | * That means that in order to get to the block containing the byte at | |
196 | * offset 101342453, we would load the indirect block pointed to by pointer | |
197 | * 0 in the dinode. We would then load the indirect block pointed to by | |
198 | * pointer 48 in that indirect block. We would then load the data block | |
199 | * pointed to by pointer 165 in that indirect block. | |
200 | * | |
201 | * ---------------------------------------- | |
202 | * | Dinode | | | |
203 | * | | 4| | |
204 | * | |0 1 2 3 4 5 9| | |
205 | * | | 6| | |
206 | * ---------------------------------------- | |
207 | * | | |
208 | * | | |
209 | * V | |
210 | * ---------------------------------------- | |
211 | * | Indirect Block | | |
212 | * | 5| | |
213 | * | 4 4 4 4 4 5 5 1| | |
214 | * |0 5 6 7 8 9 0 1 2| | |
215 | * ---------------------------------------- | |
216 | * | | |
217 | * | | |
218 | * V | |
219 | * ---------------------------------------- | |
220 | * | Indirect Block | | |
221 | * | 1 1 1 1 1 5| | |
222 | * | 6 6 6 6 6 1| | |
223 | * |0 3 4 5 6 7 2| | |
224 | * ---------------------------------------- | |
225 | * | | |
226 | * | | |
227 | * V | |
228 | * ---------------------------------------- | |
229 | * | Data block containing offset | | |
230 | * | 101342453 | | |
231 | * | | | |
232 | * | | | |
233 | * ---------------------------------------- | |
234 | * | |
235 | */ | |
236 | ||
9b8c81d1 SW |
237 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
238 | struct metapath *mp, unsigned int height) | |
b3b94faa | 239 | { |
b3b94faa DT |
240 | unsigned int i; |
241 | ||
5f8bd444 | 242 | mp->mp_fheight = height; |
9b8c81d1 | 243 | for (i = height; i--;) |
7eabb77e | 244 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
245 | } |
246 | ||
5af4e7a0 | 247 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 248 | { |
5af4e7a0 BM |
249 | if (mp->mp_list[0] == 0) |
250 | return 2; | |
251 | return 1; | |
9b8c81d1 SW |
252 | } |
253 | ||
d552a2b9 | 254 | /** |
20cdc193 | 255 | * metaptr1 - Return the first possible metadata pointer in a metapath buffer |
d552a2b9 BP |
256 | * @height: The metadata height (0 = dinode) |
257 | * @mp: The metapath | |
258 | */ | |
259 | static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp) | |
260 | { | |
261 | struct buffer_head *bh = mp->mp_bh[height]; | |
262 | if (height == 0) | |
263 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode))); | |
264 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header))); | |
265 | } | |
266 | ||
b3b94faa DT |
267 | /** |
268 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
269 | * @height: The metadata height (0 = dinode) |
270 | * @mp: The metapath | |
271 | * | |
272 | * Return a pointer to the block number of the next height of the metadata | |
273 | * tree given a buffer containing the pointer to the current height of the | |
274 | * metadata tree. | |
275 | */ | |
276 | ||
9b8c81d1 | 277 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 278 | { |
d552a2b9 BP |
279 | __be64 *p = metaptr1(height, mp); |
280 | return p + mp->mp_list[height]; | |
b3b94faa DT |
281 | } |
282 | ||
7841b9f0 AG |
283 | static inline const __be64 *metaend(unsigned int height, const struct metapath *mp) |
284 | { | |
285 | const struct buffer_head *bh = mp->mp_bh[height]; | |
286 | return (const __be64 *)(bh->b_data + bh->b_size); | |
287 | } | |
288 | ||
289 | static void clone_metapath(struct metapath *clone, struct metapath *mp) | |
290 | { | |
291 | unsigned int hgt; | |
292 | ||
293 | *clone = *mp; | |
294 | for (hgt = 0; hgt < mp->mp_aheight; hgt++) | |
295 | get_bh(clone->mp_bh[hgt]); | |
296 | } | |
297 | ||
5cf26b1e | 298 | static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end) |
b99b98dc | 299 | { |
b99b98dc SW |
300 | const __be64 *t; |
301 | ||
5cf26b1e | 302 | for (t = start; t < end; t++) { |
c3ce5aa9 AG |
303 | struct buffer_head *rabh; |
304 | ||
b99b98dc SW |
305 | if (!*t) |
306 | continue; | |
307 | ||
308 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
309 | if (trylock_buffer(rabh)) { | |
310 | if (!buffer_uptodate(rabh)) { | |
311 | rabh->b_end_io = end_buffer_read_sync; | |
e477b24b CL |
312 | submit_bh(REQ_OP_READ, |
313 | REQ_RAHEAD | REQ_META | REQ_PRIO, | |
314 | rabh); | |
b99b98dc SW |
315 | continue; |
316 | } | |
317 | unlock_buffer(rabh); | |
318 | } | |
319 | brelse(rabh); | |
320 | } | |
321 | } | |
322 | ||
e8b43fe0 AG |
323 | static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, |
324 | unsigned int x, unsigned int h) | |
d552a2b9 | 325 | { |
e8b43fe0 AG |
326 | for (; x < h; x++) { |
327 | __be64 *ptr = metapointer(x, mp); | |
328 | u64 dblock = be64_to_cpu(*ptr); | |
329 | int ret; | |
d552a2b9 | 330 | |
e8b43fe0 AG |
331 | if (!dblock) |
332 | break; | |
333 | ret = gfs2_meta_indirect_buffer(ip, x + 1, dblock, &mp->mp_bh[x + 1]); | |
334 | if (ret) | |
335 | return ret; | |
336 | } | |
337 | mp->mp_aheight = x + 1; | |
338 | return 0; | |
d552a2b9 BP |
339 | } |
340 | ||
b3b94faa | 341 | /** |
9b8c81d1 SW |
342 | * lookup_metapath - Walk the metadata tree to a specific point |
343 | * @ip: The inode | |
b3b94faa | 344 | * @mp: The metapath |
b3b94faa | 345 | * |
9b8c81d1 SW |
346 | * Assumes that the inode's buffer has already been looked up and |
347 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
348 | * by find_metapath(). | |
349 | * | |
350 | * If this function encounters part of the tree which has not been | |
351 | * allocated, it returns the current height of the tree at the point | |
352 | * at which it found the unallocated block. Blocks which are found are | |
353 | * added to the mp->mp_bh[] list. | |
b3b94faa | 354 | * |
e8b43fe0 | 355 | * Returns: error |
b3b94faa DT |
356 | */ |
357 | ||
9b8c81d1 | 358 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 359 | { |
e8b43fe0 | 360 | return __fillup_metapath(ip, mp, 0, ip->i_height - 1); |
dbac6710 SW |
361 | } |
362 | ||
d552a2b9 BP |
363 | /** |
364 | * fillup_metapath - fill up buffers for the metadata path to a specific height | |
365 | * @ip: The inode | |
366 | * @mp: The metapath | |
367 | * @h: The height to which it should be mapped | |
368 | * | |
369 | * Similar to lookup_metapath, but does lookups for a range of heights | |
370 | * | |
c3ce5aa9 | 371 | * Returns: error or the number of buffers filled |
d552a2b9 BP |
372 | */ |
373 | ||
374 | static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h) | |
375 | { | |
e8b43fe0 | 376 | unsigned int x = 0; |
c3ce5aa9 | 377 | int ret; |
d552a2b9 BP |
378 | |
379 | if (h) { | |
380 | /* find the first buffer we need to look up. */ | |
e8b43fe0 AG |
381 | for (x = h - 1; x > 0; x--) { |
382 | if (mp->mp_bh[x]) | |
383 | break; | |
d552a2b9 BP |
384 | } |
385 | } | |
c3ce5aa9 AG |
386 | ret = __fillup_metapath(ip, mp, x, h); |
387 | if (ret) | |
388 | return ret; | |
389 | return mp->mp_aheight - x - 1; | |
d552a2b9 BP |
390 | } |
391 | ||
9b8c81d1 | 392 | static inline void release_metapath(struct metapath *mp) |
dbac6710 SW |
393 | { |
394 | int i; | |
395 | ||
9b8c81d1 SW |
396 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
397 | if (mp->mp_bh[i] == NULL) | |
398 | break; | |
399 | brelse(mp->mp_bh[i]); | |
400 | } | |
11707ea0 SW |
401 | } |
402 | ||
30cbf189 SW |
403 | /** |
404 | * gfs2_extent_length - Returns length of an extent of blocks | |
405 | * @start: Start of the buffer | |
406 | * @len: Length of the buffer in bytes | |
407 | * @ptr: Current position in the buffer | |
408 | * @limit: Max extent length to return (0 = unlimited) | |
409 | * @eob: Set to 1 if we hit "end of block" | |
410 | * | |
411 | * If the first block is zero (unallocated) it will return the number of | |
412 | * unallocated blocks in the extent, otherwise it will return the number | |
413 | * of contiguous blocks in the extent. | |
414 | * | |
415 | * Returns: The length of the extent (minimum of one block) | |
416 | */ | |
417 | ||
b650738c | 418 | static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob) |
30cbf189 SW |
419 | { |
420 | const __be64 *end = (start + len); | |
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; | |
429 | if (limit && --limit == 0) | |
430 | break; | |
431 | if (d) | |
432 | d++; | |
433 | } while(be64_to_cpu(*ptr) == d); | |
434 | if (ptr >= end) | |
435 | *eob = 1; | |
436 | return (ptr - first); | |
437 | } | |
438 | ||
7841b9f0 AG |
439 | typedef const __be64 *(*gfs2_metadata_walker)( |
440 | struct metapath *mp, | |
441 | const __be64 *start, const __be64 *end, | |
442 | u64 factor, void *data); | |
443 | ||
444 | #define WALK_STOP ((__be64 *)0) | |
445 | #define WALK_NEXT ((__be64 *)1) | |
446 | ||
447 | static int gfs2_walk_metadata(struct inode *inode, sector_t lblock, | |
448 | u64 len, struct metapath *mp, gfs2_metadata_walker walker, | |
449 | void *data) | |
450 | { | |
451 | struct metapath clone; | |
452 | struct gfs2_inode *ip = GFS2_I(inode); | |
453 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
454 | const __be64 *start, *end, *ptr; | |
455 | u64 factor = 1; | |
456 | unsigned int hgt; | |
457 | int ret = 0; | |
458 | ||
459 | for (hgt = ip->i_height - 1; hgt >= mp->mp_aheight; hgt--) | |
460 | factor *= sdp->sd_inptrs; | |
461 | ||
462 | for (;;) { | |
463 | u64 step; | |
464 | ||
465 | /* Walk indirect block. */ | |
466 | start = metapointer(hgt, mp); | |
467 | end = metaend(hgt, mp); | |
468 | ||
469 | step = (end - start) * factor; | |
470 | if (step > len) | |
471 | end = start + DIV_ROUND_UP_ULL(len, factor); | |
472 | ||
473 | ptr = walker(mp, start, end, factor, data); | |
474 | if (ptr == WALK_STOP) | |
475 | break; | |
476 | if (step >= len) | |
477 | break; | |
478 | len -= step; | |
479 | if (ptr != WALK_NEXT) { | |
480 | BUG_ON(!*ptr); | |
481 | mp->mp_list[hgt] += ptr - start; | |
482 | goto fill_up_metapath; | |
483 | } | |
484 | ||
485 | lower_metapath: | |
486 | /* Decrease height of metapath. */ | |
487 | if (mp != &clone) { | |
488 | clone_metapath(&clone, mp); | |
489 | mp = &clone; | |
490 | } | |
491 | brelse(mp->mp_bh[hgt]); | |
492 | mp->mp_bh[hgt] = NULL; | |
493 | if (!hgt) | |
494 | break; | |
495 | hgt--; | |
496 | factor *= sdp->sd_inptrs; | |
497 | ||
498 | /* Advance in metadata tree. */ | |
499 | (mp->mp_list[hgt])++; | |
500 | start = metapointer(hgt, mp); | |
501 | end = metaend(hgt, mp); | |
502 | if (start >= end) { | |
503 | mp->mp_list[hgt] = 0; | |
504 | if (!hgt) | |
505 | break; | |
506 | goto lower_metapath; | |
507 | } | |
508 | ||
509 | fill_up_metapath: | |
510 | /* Increase height of metapath. */ | |
511 | if (mp != &clone) { | |
512 | clone_metapath(&clone, mp); | |
513 | mp = &clone; | |
514 | } | |
515 | ret = fillup_metapath(ip, mp, ip->i_height - 1); | |
516 | if (ret < 0) | |
517 | break; | |
518 | hgt += ret; | |
519 | for (; ret; ret--) | |
520 | do_div(factor, sdp->sd_inptrs); | |
521 | mp->mp_aheight = hgt + 1; | |
522 | } | |
523 | if (mp == &clone) | |
524 | release_metapath(mp); | |
525 | return ret; | |
526 | } | |
527 | ||
528 | struct gfs2_hole_walker_args { | |
529 | u64 blocks; | |
530 | }; | |
531 | ||
532 | static const __be64 *gfs2_hole_walker(struct metapath *mp, | |
533 | const __be64 *start, const __be64 *end, | |
534 | u64 factor, void *data) | |
535 | { | |
536 | struct gfs2_hole_walker_args *args = data; | |
537 | const __be64 *ptr; | |
538 | ||
539 | for (ptr = start; ptr < end; ptr++) { | |
540 | if (*ptr) { | |
541 | args->blocks += (ptr - start) * factor; | |
542 | if (mp->mp_aheight == mp->mp_fheight) | |
543 | return WALK_STOP; | |
544 | return ptr; /* increase height */ | |
545 | } | |
546 | } | |
547 | args->blocks += (end - start) * factor; | |
548 | return WALK_NEXT; | |
549 | } | |
550 | ||
551 | /** | |
552 | * gfs2_hole_size - figure out the size of a hole | |
553 | * @inode: The inode | |
554 | * @lblock: The logical starting block number | |
555 | * @len: How far to look (in blocks) | |
556 | * @mp: The metapath at lblock | |
557 | * @iomap: The iomap to store the hole size in | |
558 | * | |
559 | * This function modifies @mp. | |
560 | * | |
561 | * Returns: errno on error | |
562 | */ | |
563 | static int gfs2_hole_size(struct inode *inode, sector_t lblock, u64 len, | |
564 | struct metapath *mp, struct iomap *iomap) | |
565 | { | |
566 | struct gfs2_hole_walker_args args = { }; | |
567 | int ret = 0; | |
568 | ||
569 | ret = gfs2_walk_metadata(inode, lblock, len, mp, gfs2_hole_walker, &args); | |
570 | if (!ret) | |
571 | iomap->length = args.blocks << inode->i_blkbits; | |
572 | return ret; | |
573 | } | |
574 | ||
9b8c81d1 SW |
575 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
576 | struct gfs2_glock *gl, unsigned int i, | |
577 | unsigned offset, u64 bn) | |
578 | { | |
579 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
580 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
581 | sizeof(struct gfs2_dinode))); | |
582 | BUG_ON(i < 1); | |
583 | BUG_ON(mp->mp_bh[i] != NULL); | |
584 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
350a9b0a | 585 | gfs2_trans_add_meta(gl, mp->mp_bh[i]); |
9b8c81d1 SW |
586 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
587 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
588 | ptr += offset; | |
589 | *ptr = cpu_to_be64(bn); | |
590 | return ptr; | |
591 | } | |
592 | ||
593 | enum alloc_state { | |
594 | ALLOC_DATA = 0, | |
595 | ALLOC_GROW_DEPTH = 1, | |
596 | ALLOC_GROW_HEIGHT = 2, | |
597 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
598 | }; | |
599 | ||
600 | /** | |
628e366d | 601 | * gfs2_iomap_alloc - Build a metadata tree of the requested height |
9b8c81d1 | 602 | * @inode: The GFS2 inode |
628e366d AG |
603 | * @iomap: The iomap structure |
604 | * @flags: iomap flags | |
5f8bd444 | 605 | * @mp: The metapath, with proper height information calculated |
9b8c81d1 SW |
606 | * |
607 | * In this routine we may have to alloc: | |
608 | * i) Indirect blocks to grow the metadata tree height | |
609 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
610 | * iii) Data blocks | |
611 | * | |
612 | * The function is in two parts. The first part works out the total | |
613 | * number of blocks which we need. The second part does the actual | |
614 | * allocation asking for an extent at a time (if enough contiguous free | |
615 | * blocks are available, there will only be one request per bmap call) | |
616 | * and uses the state machine to initialise the blocks in order. | |
617 | * | |
628e366d AG |
618 | * Right now, this function will allocate at most one indirect block |
619 | * worth of data -- with a default block size of 4K, that's slightly | |
620 | * less than 2M. If this limitation is ever removed to allow huge | |
621 | * allocations, we would probably still want to limit the iomap size we | |
622 | * return to avoid stalling other tasks during huge writes; the next | |
623 | * iomap iteration would then find the blocks already allocated. | |
624 | * | |
9b8c81d1 SW |
625 | * Returns: errno on error |
626 | */ | |
627 | ||
3974320c BP |
628 | static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap, |
629 | unsigned flags, struct metapath *mp) | |
9b8c81d1 SW |
630 | { |
631 | struct gfs2_inode *ip = GFS2_I(inode); | |
632 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
633 | struct buffer_head *dibh = mp->mp_bh[0]; | |
5f8bd444 | 634 | u64 bn; |
5af4e7a0 | 635 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
3974320c | 636 | unsigned dblks = 0; |
9b8c81d1 | 637 | unsigned ptrs_per_blk; |
5f8bd444 | 638 | const unsigned end_of_metadata = mp->mp_fheight - 1; |
628e366d | 639 | int ret; |
9b8c81d1 SW |
640 | enum alloc_state state; |
641 | __be64 *ptr; | |
642 | __be64 zero_bn = 0; | |
3974320c | 643 | size_t maxlen = iomap->length >> inode->i_blkbits; |
9b8c81d1 | 644 | |
5f8bd444 | 645 | BUG_ON(mp->mp_aheight < 1); |
9b8c81d1 SW |
646 | BUG_ON(dibh == NULL); |
647 | ||
350a9b0a | 648 | gfs2_trans_add_meta(ip->i_gl, dibh); |
9b8c81d1 | 649 | |
628e366d AG |
650 | down_write(&ip->i_rw_mutex); |
651 | ||
5f8bd444 | 652 | if (mp->mp_fheight == mp->mp_aheight) { |
9b8c81d1 | 653 | struct buffer_head *bh; |
3974320c BP |
654 | int eob; |
655 | ||
9b8c81d1 SW |
656 | /* Bottom indirect block exists, find unalloced extent size */ |
657 | ptr = metapointer(end_of_metadata, mp); | |
658 | bh = mp->mp_bh[end_of_metadata]; | |
3974320c BP |
659 | dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, |
660 | maxlen, &eob); | |
661 | BUG_ON(dblks < 1); | |
9b8c81d1 SW |
662 | state = ALLOC_DATA; |
663 | } else { | |
664 | /* Need to allocate indirect blocks */ | |
5f8bd444 BP |
665 | ptrs_per_blk = mp->mp_fheight > 1 ? sdp->sd_inptrs : |
666 | sdp->sd_diptrs; | |
3974320c BP |
667 | dblks = min(maxlen, (size_t)(ptrs_per_blk - |
668 | mp->mp_list[end_of_metadata])); | |
5f8bd444 | 669 | if (mp->mp_fheight == ip->i_height) { |
9b8c81d1 | 670 | /* Writing into existing tree, extend tree down */ |
5f8bd444 | 671 | iblks = mp->mp_fheight - mp->mp_aheight; |
9b8c81d1 SW |
672 | state = ALLOC_GROW_DEPTH; |
673 | } else { | |
674 | /* Building up tree height */ | |
675 | state = ALLOC_GROW_HEIGHT; | |
5f8bd444 | 676 | iblks = mp->mp_fheight - ip->i_height; |
5af4e7a0 | 677 | branch_start = metapath_branch_start(mp); |
5f8bd444 | 678 | iblks += (mp->mp_fheight - branch_start); |
9b8c81d1 SW |
679 | } |
680 | } | |
681 | ||
682 | /* start of the second part of the function (state machine) */ | |
683 | ||
3974320c | 684 | blks = dblks + iblks; |
5f8bd444 | 685 | i = mp->mp_aheight; |
9b8c81d1 SW |
686 | do { |
687 | n = blks - alloced; | |
628e366d AG |
688 | ret = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
689 | if (ret) | |
690 | goto out; | |
9b8c81d1 SW |
691 | alloced += n; |
692 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
693 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
694 | switch (state) { | |
695 | /* Growing height of tree */ | |
696 | case ALLOC_GROW_HEIGHT: | |
697 | if (i == 1) { | |
698 | ptr = (__be64 *)(dibh->b_data + | |
699 | sizeof(struct gfs2_dinode)); | |
700 | zero_bn = *ptr; | |
701 | } | |
5f8bd444 BP |
702 | for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0; |
703 | i++, n--) | |
9b8c81d1 | 704 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); |
5f8bd444 | 705 | if (i - 1 == mp->mp_fheight - ip->i_height) { |
9b8c81d1 SW |
706 | i--; |
707 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
708 | sizeof(struct gfs2_meta_header), | |
709 | dibh, sizeof(struct gfs2_dinode)); | |
710 | gfs2_buffer_clear_tail(dibh, | |
711 | sizeof(struct gfs2_dinode) + | |
712 | sizeof(__be64)); | |
713 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
714 | sizeof(struct gfs2_meta_header)); | |
715 | *ptr = zero_bn; | |
716 | state = ALLOC_GROW_DEPTH; | |
5f8bd444 | 717 | for(i = branch_start; i < mp->mp_fheight; i++) { |
9b8c81d1 SW |
718 | if (mp->mp_bh[i] == NULL) |
719 | break; | |
720 | brelse(mp->mp_bh[i]); | |
721 | mp->mp_bh[i] = NULL; | |
722 | } | |
5af4e7a0 | 723 | i = branch_start; |
9b8c81d1 SW |
724 | } |
725 | if (n == 0) | |
726 | break; | |
727 | /* Branching from existing tree */ | |
728 | case ALLOC_GROW_DEPTH: | |
5f8bd444 | 729 | if (i > 1 && i < mp->mp_fheight) |
350a9b0a | 730 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]); |
5f8bd444 | 731 | for (; i < mp->mp_fheight && n > 0; i++, n--) |
9b8c81d1 SW |
732 | gfs2_indirect_init(mp, ip->i_gl, i, |
733 | mp->mp_list[i-1], bn++); | |
5f8bd444 | 734 | if (i == mp->mp_fheight) |
9b8c81d1 SW |
735 | state = ALLOC_DATA; |
736 | if (n == 0) | |
737 | break; | |
738 | /* Tree complete, adding data blocks */ | |
739 | case ALLOC_DATA: | |
3974320c | 740 | BUG_ON(n > dblks); |
9b8c81d1 | 741 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); |
350a9b0a | 742 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]); |
3974320c | 743 | dblks = n; |
9b8c81d1 | 744 | ptr = metapointer(end_of_metadata, mp); |
3974320c | 745 | iomap->addr = bn << inode->i_blkbits; |
628e366d | 746 | iomap->flags |= IOMAP_F_MERGED | IOMAP_F_NEW; |
9b8c81d1 SW |
747 | while (n-- > 0) |
748 | *ptr++ = cpu_to_be64(bn++); | |
749 | break; | |
750 | } | |
3974320c | 751 | } while (iomap->addr == IOMAP_NULL_ADDR); |
9b8c81d1 | 752 | |
3974320c | 753 | iomap->length = (u64)dblks << inode->i_blkbits; |
5f8bd444 | 754 | ip->i_height = mp->mp_fheight; |
9b8c81d1 | 755 | gfs2_add_inode_blocks(&ip->i_inode, alloced); |
628e366d AG |
756 | gfs2_dinode_out(ip, dibh->b_data); |
757 | out: | |
758 | up_write(&ip->i_rw_mutex); | |
759 | return ret; | |
9b8c81d1 SW |
760 | } |
761 | ||
3974320c BP |
762 | static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap) |
763 | { | |
764 | struct gfs2_inode *ip = GFS2_I(inode); | |
765 | ||
766 | iomap->addr = (ip->i_no_addr << inode->i_blkbits) + | |
767 | sizeof(struct gfs2_dinode); | |
768 | iomap->offset = 0; | |
769 | iomap->length = i_size_read(inode); | |
19319b53 | 770 | iomap->type = IOMAP_INLINE; |
3974320c BP |
771 | } |
772 | ||
7ee66c03 CH |
773 | #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE |
774 | ||
3974320c | 775 | /** |
628e366d | 776 | * gfs2_iomap_get - Map blocks from an inode to disk blocks |
3974320c BP |
777 | * @inode: The inode |
778 | * @pos: Starting position in bytes | |
779 | * @length: Length to map, in bytes | |
780 | * @flags: iomap flags | |
781 | * @iomap: The iomap structure | |
628e366d | 782 | * @mp: The metapath |
3974320c BP |
783 | * |
784 | * Returns: errno | |
785 | */ | |
628e366d AG |
786 | static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length, |
787 | unsigned flags, struct iomap *iomap, | |
788 | struct metapath *mp) | |
b3b94faa | 789 | { |
feaa7bba SW |
790 | struct gfs2_inode *ip = GFS2_I(inode); |
791 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
9b8c81d1 | 792 | __be64 *ptr; |
3974320c | 793 | sector_t lblock; |
628e366d AG |
794 | sector_t lblock_stop; |
795 | int ret; | |
9b8c81d1 | 796 | int eob; |
628e366d | 797 | u64 len; |
9b8c81d1 SW |
798 | struct buffer_head *bh; |
799 | u8 height; | |
7276b3b0 | 800 | |
628e366d AG |
801 | if (!length) |
802 | return -EINVAL; | |
b3b94faa | 803 | |
49edd5bf AG |
804 | if (gfs2_is_stuffed(ip)) { |
805 | if (flags & IOMAP_REPORT) { | |
628e366d AG |
806 | if (pos >= i_size_read(inode)) |
807 | return -ENOENT; | |
49edd5bf | 808 | gfs2_stuffed_iomap(inode, iomap); |
628e366d | 809 | return 0; |
49edd5bf AG |
810 | } |
811 | BUG_ON(!(flags & IOMAP_WRITE)); | |
3974320c | 812 | } |
3974320c | 813 | lblock = pos >> inode->i_blkbits; |
3974320c | 814 | iomap->offset = lblock << inode->i_blkbits; |
628e366d AG |
815 | lblock_stop = (pos + length - 1) >> inode->i_blkbits; |
816 | len = lblock_stop - lblock + 1; | |
817 | ||
818 | down_read(&ip->i_rw_mutex); | |
20cdc193 | 819 | |
628e366d | 820 | ret = gfs2_meta_inode_buffer(ip, &mp->mp_bh[0]); |
9b8c81d1 | 821 | if (ret) |
628e366d | 822 | goto unlock; |
b3b94faa | 823 | |
9b8c81d1 | 824 | height = ip->i_height; |
9a38662b | 825 | while ((lblock + 1) * sdp->sd_sb.sb_bsize > sdp->sd_heightsize[height]) |
9b8c81d1 | 826 | height++; |
628e366d | 827 | find_metapath(sdp, lblock, mp, height); |
9b8c81d1 SW |
828 | if (height > ip->i_height || gfs2_is_stuffed(ip)) |
829 | goto do_alloc; | |
3974320c | 830 | |
628e366d | 831 | ret = lookup_metapath(ip, mp); |
e8b43fe0 | 832 | if (ret) |
628e366d | 833 | goto unlock; |
3974320c | 834 | |
628e366d | 835 | if (mp->mp_aheight != ip->i_height) |
9b8c81d1 | 836 | goto do_alloc; |
3974320c | 837 | |
628e366d | 838 | ptr = metapointer(ip->i_height - 1, mp); |
9b8c81d1 SW |
839 | if (*ptr == 0) |
840 | goto do_alloc; | |
3974320c | 841 | |
628e366d AG |
842 | bh = mp->mp_bh[ip->i_height - 1]; |
843 | len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, len, &eob); | |
3974320c | 844 | |
628e366d AG |
845 | iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits; |
846 | iomap->length = len << inode->i_blkbits; | |
847 | iomap->type = IOMAP_MAPPED; | |
848 | iomap->flags = IOMAP_F_MERGED; | |
9b8c81d1 | 849 | if (eob) |
7ee66c03 | 850 | iomap->flags |= IOMAP_F_GFS2_BOUNDARY; |
3974320c | 851 | |
3974320c | 852 | out: |
628e366d AG |
853 | iomap->bdev = inode->i_sb->s_bdev; |
854 | unlock: | |
855 | up_read(&ip->i_rw_mutex); | |
9b8c81d1 | 856 | return ret; |
30cbf189 | 857 | |
9b8c81d1 | 858 | do_alloc: |
628e366d AG |
859 | iomap->addr = IOMAP_NULL_ADDR; |
860 | iomap->length = len << inode->i_blkbits; | |
861 | iomap->type = IOMAP_HOLE; | |
862 | iomap->flags = 0; | |
863 | if (flags & IOMAP_REPORT) { | |
49edd5bf AG |
864 | loff_t size = i_size_read(inode); |
865 | if (pos >= size) | |
3974320c | 866 | ret = -ENOENT; |
628e366d AG |
867 | else if (height == ip->i_height) |
868 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
49edd5bf AG |
869 | else |
870 | iomap->length = size - pos; | |
b3b94faa | 871 | } |
628e366d | 872 | goto out; |
3974320c BP |
873 | } |
874 | ||
628e366d AG |
875 | static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, |
876 | unsigned flags, struct iomap *iomap) | |
877 | { | |
878 | struct gfs2_inode *ip = GFS2_I(inode); | |
879 | struct metapath mp = { .mp_aheight = 1, }; | |
880 | int ret; | |
881 | ||
882 | trace_gfs2_iomap_start(ip, pos, length, flags); | |
883 | if (flags & IOMAP_WRITE) { | |
884 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); | |
885 | if (!ret && iomap->type == IOMAP_HOLE) | |
886 | ret = gfs2_iomap_alloc(inode, iomap, flags, &mp); | |
887 | release_metapath(&mp); | |
888 | } else { | |
889 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); | |
890 | release_metapath(&mp); | |
891 | } | |
892 | trace_gfs2_iomap_end(ip, iomap, ret); | |
893 | return ret; | |
894 | } | |
895 | ||
896 | const struct iomap_ops gfs2_iomap_ops = { | |
897 | .iomap_begin = gfs2_iomap_begin, | |
898 | }; | |
899 | ||
3974320c | 900 | /** |
d39d18e0 | 901 | * gfs2_block_map - Map one or more blocks of an inode to a disk block |
3974320c BP |
902 | * @inode: The inode |
903 | * @lblock: The logical block number | |
904 | * @bh_map: The bh to be mapped | |
905 | * @create: True if its ok to alloc blocks to satify the request | |
906 | * | |
d39d18e0 AG |
907 | * The size of the requested mapping is defined in bh_map->b_size. |
908 | * | |
909 | * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged | |
910 | * when @lblock is not mapped. Sets buffer_mapped(bh_map) and | |
911 | * bh_map->b_size to indicate the size of the mapping when @lblock and | |
912 | * successive blocks are mapped, up to the requested size. | |
913 | * | |
914 | * Sets buffer_boundary() if a read of metadata will be required | |
915 | * before the next block can be mapped. Sets buffer_new() if new | |
916 | * blocks were allocated. | |
3974320c BP |
917 | * |
918 | * Returns: errno | |
919 | */ | |
920 | ||
921 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
922 | struct buffer_head *bh_map, int create) | |
923 | { | |
924 | struct gfs2_inode *ip = GFS2_I(inode); | |
628e366d AG |
925 | loff_t pos = (loff_t)lblock << inode->i_blkbits; |
926 | loff_t length = bh_map->b_size; | |
927 | struct metapath mp = { .mp_aheight = 1, }; | |
928 | struct iomap iomap = { }; | |
929 | int ret; | |
3974320c BP |
930 | |
931 | clear_buffer_mapped(bh_map); | |
932 | clear_buffer_new(bh_map); | |
933 | clear_buffer_boundary(bh_map); | |
934 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
935 | ||
628e366d AG |
936 | if (create) { |
937 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, &iomap, &mp); | |
938 | if (!ret && iomap.type == IOMAP_HOLE) | |
939 | ret = gfs2_iomap_alloc(inode, &iomap, IOMAP_WRITE, &mp); | |
940 | release_metapath(&mp); | |
941 | } else { | |
942 | ret = gfs2_iomap_get(inode, pos, length, 0, &iomap, &mp); | |
943 | release_metapath(&mp); | |
944 | ||
945 | /* Return unmapped buffer beyond the end of file. */ | |
946 | if (ret == -ENOENT) { | |
3974320c | 947 | ret = 0; |
628e366d | 948 | goto out; |
3974320c | 949 | } |
3974320c | 950 | } |
628e366d AG |
951 | if (ret) |
952 | goto out; | |
3974320c BP |
953 | |
954 | if (iomap.length > bh_map->b_size) { | |
955 | iomap.length = bh_map->b_size; | |
7ee66c03 | 956 | iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY; |
5f8bd444 | 957 | } |
3974320c BP |
958 | if (iomap.addr != IOMAP_NULL_ADDR) |
959 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
960 | bh_map->b_size = iomap.length; | |
7ee66c03 | 961 | if (iomap.flags & IOMAP_F_GFS2_BOUNDARY) |
3974320c BP |
962 | set_buffer_boundary(bh_map); |
963 | if (iomap.flags & IOMAP_F_NEW) | |
964 | set_buffer_new(bh_map); | |
965 | ||
966 | out: | |
967 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
968 | return ret; | |
fd88de56 SW |
969 | } |
970 | ||
941e6d7d SW |
971 | /* |
972 | * Deprecated: do not use in new code | |
973 | */ | |
fd88de56 SW |
974 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
975 | { | |
23591256 | 976 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 977 | int ret; |
fd88de56 SW |
978 | int create = *new; |
979 | ||
980 | BUG_ON(!extlen); | |
981 | BUG_ON(!dblock); | |
982 | BUG_ON(!new); | |
983 | ||
47a9a527 | 984 | bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 985 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
986 | *extlen = bh.b_size >> inode->i_blkbits; |
987 | *dblock = bh.b_blocknr; | |
988 | if (buffer_new(&bh)) | |
989 | *new = 1; | |
990 | else | |
991 | *new = 0; | |
992 | return ret; | |
b3b94faa DT |
993 | } |
994 | ||
ba7f7290 | 995 | /** |
bdba0d5e | 996 | * gfs2_block_zero_range - Deal with zeroing out data |
ba7f7290 SW |
997 | * |
998 | * This is partly borrowed from ext3. | |
999 | */ | |
bdba0d5e AG |
1000 | static int gfs2_block_zero_range(struct inode *inode, loff_t from, |
1001 | unsigned int length) | |
ba7f7290 | 1002 | { |
bdba0d5e | 1003 | struct address_space *mapping = inode->i_mapping; |
ba7f7290 | 1004 | struct gfs2_inode *ip = GFS2_I(inode); |
09cbfeaf KS |
1005 | unsigned long index = from >> PAGE_SHIFT; |
1006 | unsigned offset = from & (PAGE_SIZE-1); | |
bdba0d5e | 1007 | unsigned blocksize, iblock, pos; |
ba7f7290 SW |
1008 | struct buffer_head *bh; |
1009 | struct page *page; | |
ba7f7290 SW |
1010 | int err; |
1011 | ||
220cca2a | 1012 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ba7f7290 SW |
1013 | if (!page) |
1014 | return 0; | |
1015 | ||
1016 | blocksize = inode->i_sb->s_blocksize; | |
09cbfeaf | 1017 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
ba7f7290 SW |
1018 | |
1019 | if (!page_has_buffers(page)) | |
1020 | create_empty_buffers(page, blocksize, 0); | |
1021 | ||
1022 | /* Find the buffer that contains "offset" */ | |
1023 | bh = page_buffers(page); | |
1024 | pos = blocksize; | |
1025 | while (offset >= pos) { | |
1026 | bh = bh->b_this_page; | |
1027 | iblock++; | |
1028 | pos += blocksize; | |
1029 | } | |
1030 | ||
1031 | err = 0; | |
1032 | ||
1033 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 1034 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
1035 | /* unmapped? It's a hole - nothing to do */ |
1036 | if (!buffer_mapped(bh)) | |
1037 | goto unlock; | |
1038 | } | |
1039 | ||
1040 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1041 | if (PageUptodate(page)) | |
1042 | set_buffer_uptodate(bh); | |
1043 | ||
1044 | if (!buffer_uptodate(bh)) { | |
1045 | err = -EIO; | |
dfec8a14 | 1046 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
ba7f7290 SW |
1047 | wait_on_buffer(bh); |
1048 | /* Uhhuh. Read error. Complain and punt. */ | |
1049 | if (!buffer_uptodate(bh)) | |
1050 | goto unlock; | |
1875f2f3 | 1051 | err = 0; |
ba7f7290 SW |
1052 | } |
1053 | ||
845802b1 | 1054 | if (gfs2_is_jdata(ip)) |
350a9b0a | 1055 | gfs2_trans_add_data(ip->i_gl, bh); |
845802b1 AG |
1056 | else |
1057 | gfs2_ordered_add_inode(ip); | |
ba7f7290 | 1058 | |
eebd2aa3 | 1059 | zero_user(page, offset, length); |
40bc9a27 | 1060 | mark_buffer_dirty(bh); |
ba7f7290 SW |
1061 | unlock: |
1062 | unlock_page(page); | |
09cbfeaf | 1063 | put_page(page); |
ba7f7290 SW |
1064 | return err; |
1065 | } | |
1066 | ||
c62baf65 FF |
1067 | #define GFS2_JTRUNC_REVOKES 8192 |
1068 | ||
fa731fc4 SW |
1069 | /** |
1070 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
1071 | * @inode: The inode being truncated | |
1072 | * @oldsize: The original (larger) size | |
1073 | * @newsize: The new smaller size | |
1074 | * | |
1075 | * With jdata files, we have to journal a revoke for each block which is | |
1076 | * truncated. As a result, we need to split this into separate transactions | |
1077 | * if the number of pages being truncated gets too large. | |
1078 | */ | |
1079 | ||
fa731fc4 SW |
1080 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
1081 | { | |
1082 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1083 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1084 | u64 chunk; | |
1085 | int error; | |
1086 | ||
1087 | while (oldsize != newsize) { | |
e7fdf004 AG |
1088 | struct gfs2_trans *tr; |
1089 | unsigned int offs; | |
1090 | ||
fa731fc4 SW |
1091 | chunk = oldsize - newsize; |
1092 | if (chunk > max_chunk) | |
1093 | chunk = max_chunk; | |
e7fdf004 AG |
1094 | |
1095 | offs = oldsize & ~PAGE_MASK; | |
1096 | if (offs && chunk > PAGE_SIZE) | |
1097 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1098 | ||
7caef267 | 1099 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1100 | oldsize -= chunk; |
e7fdf004 AG |
1101 | |
1102 | tr = current->journal_info; | |
1103 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1104 | continue; | |
1105 | ||
fa731fc4 SW |
1106 | gfs2_trans_end(sdp); |
1107 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1108 | if (error) | |
1109 | return error; | |
1110 | } | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
8b5860a3 | 1115 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1116 | { |
ff8f33c8 SW |
1117 | struct gfs2_inode *ip = GFS2_I(inode); |
1118 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
80990f40 | 1119 | struct buffer_head *dibh = NULL; |
b3b94faa | 1120 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1121 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1122 | int error; |
1123 | ||
fa731fc4 SW |
1124 | if (journaled) |
1125 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1126 | else | |
1127 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1128 | if (error) |
1129 | return error; | |
1130 | ||
1131 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1132 | if (error) | |
1133 | goto out; | |
1134 | ||
350a9b0a | 1135 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1136 | |
b3b94faa | 1137 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1138 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1139 | } else { |
bdba0d5e AG |
1140 | unsigned int blocksize = i_blocksize(inode); |
1141 | unsigned int offs = newsize & (blocksize - 1); | |
1142 | if (offs) { | |
1143 | error = gfs2_block_zero_range(inode, newsize, | |
1144 | blocksize - offs); | |
ff8f33c8 | 1145 | if (error) |
80990f40 | 1146 | goto out; |
b3b94faa | 1147 | } |
ff8f33c8 | 1148 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1149 | } |
1150 | ||
ff8f33c8 | 1151 | i_size_write(inode, newsize); |
078cd827 | 1152 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
ff8f33c8 | 1153 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1154 | |
fa731fc4 SW |
1155 | if (journaled) |
1156 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1157 | else | |
7caef267 | 1158 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1159 | |
a91ea69f | 1160 | out: |
80990f40 AG |
1161 | brelse(dibh); |
1162 | if (current->journal_info) | |
1163 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1164 | return error; |
1165 | } | |
1166 | ||
628e366d AG |
1167 | int gfs2_iomap_get_alloc(struct inode *inode, loff_t pos, loff_t length, |
1168 | struct iomap *iomap) | |
1169 | { | |
1170 | struct metapath mp = { .mp_aheight = 1, }; | |
1171 | int ret; | |
1172 | ||
1173 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp); | |
1174 | if (!ret && iomap->type == IOMAP_HOLE) | |
1175 | ret = gfs2_iomap_alloc(inode, iomap, IOMAP_WRITE, &mp); | |
1176 | release_metapath(&mp); | |
1177 | return ret; | |
1178 | } | |
1179 | ||
d552a2b9 BP |
1180 | /** |
1181 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1182 | * @ip: inode | |
1183 | * @rg_gh: holder of resource group glock | |
5cf26b1e AG |
1184 | * @bh: buffer head to sweep |
1185 | * @start: starting point in bh | |
1186 | * @end: end point in bh | |
1187 | * @meta: true if bh points to metadata (rather than data) | |
d552a2b9 | 1188 | * @btotal: place to keep count of total blocks freed |
d552a2b9 BP |
1189 | * |
1190 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1191 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1192 | * block has references to multiple rgrps, we break it into individual | |
1193 | * transactions. This allows other processes to use the rgrps while we're | |
1194 | * focused on a single one, for better concurrency / performance. | |
1195 | * At every transaction boundary, we rewrite the inode into the journal. | |
1196 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1197 | * if we're interrupted by power-outages. | |
1198 | * | |
1199 | * Returns: 0, or return code if an error occurred. | |
1200 | * *btotal has the total number of blocks freed | |
1201 | */ | |
1202 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
5cf26b1e AG |
1203 | struct buffer_head *bh, __be64 *start, __be64 *end, |
1204 | bool meta, u32 *btotal) | |
b3b94faa | 1205 | { |
9b8c81d1 | 1206 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1207 | struct gfs2_rgrpd *rgd; |
1208 | struct gfs2_trans *tr; | |
5cf26b1e | 1209 | __be64 *p; |
d552a2b9 BP |
1210 | int blks_outside_rgrp; |
1211 | u64 bn, bstart, isize_blks; | |
1212 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
d552a2b9 BP |
1213 | int ret = 0; |
1214 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1215 | ||
d552a2b9 | 1216 | more_rgrps: |
5cf26b1e AG |
1217 | rgd = NULL; |
1218 | if (gfs2_holder_initialized(rd_gh)) { | |
1219 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); | |
1220 | gfs2_assert_withdraw(sdp, | |
1221 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1222 | } | |
d552a2b9 BP |
1223 | blks_outside_rgrp = 0; |
1224 | bstart = 0; | |
1225 | blen = 0; | |
d552a2b9 | 1226 | |
5cf26b1e | 1227 | for (p = start; p < end; p++) { |
d552a2b9 BP |
1228 | if (!*p) |
1229 | continue; | |
1230 | bn = be64_to_cpu(*p); | |
5cf26b1e AG |
1231 | |
1232 | if (rgd) { | |
1233 | if (!rgrp_contains_block(rgd, bn)) { | |
1234 | blks_outside_rgrp++; | |
1235 | continue; | |
1236 | } | |
d552a2b9 | 1237 | } else { |
90bcab99 | 1238 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
5cf26b1e AG |
1239 | if (unlikely(!rgd)) { |
1240 | ret = -EIO; | |
1241 | goto out; | |
1242 | } | |
d552a2b9 BP |
1243 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1244 | 0, rd_gh); | |
1245 | if (ret) | |
1246 | goto out; | |
1247 | ||
1248 | /* Must be done with the rgrp glock held: */ | |
1249 | if (gfs2_rs_active(&ip->i_res) && | |
1250 | rgd == ip->i_res.rs_rbm.rgd) | |
1251 | gfs2_rs_deltree(&ip->i_res); | |
1252 | } | |
1253 | ||
d552a2b9 BP |
1254 | /* The size of our transactions will be unknown until we |
1255 | actually process all the metadata blocks that relate to | |
1256 | the rgrp. So we estimate. We know it can't be more than | |
1257 | the dinode's i_blocks and we don't want to exceed the | |
1258 | journal flush threshold, sd_log_thresh2. */ | |
1259 | if (current->journal_info == NULL) { | |
1260 | unsigned int jblocks_rqsted, revokes; | |
1261 | ||
1262 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1263 | RES_INDIRECT; | |
1264 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1265 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1266 | jblocks_rqsted += | |
1267 | atomic_read(&sdp->sd_log_thresh2); | |
1268 | else | |
1269 | jblocks_rqsted += isize_blks; | |
1270 | revokes = jblocks_rqsted; | |
1271 | if (meta) | |
5cf26b1e | 1272 | revokes += end - start; |
d552a2b9 BP |
1273 | else if (ip->i_depth) |
1274 | revokes += sdp->sd_inptrs; | |
1275 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1276 | if (ret) | |
1277 | goto out_unlock; | |
1278 | down_write(&ip->i_rw_mutex); | |
1279 | } | |
1280 | /* check if we will exceed the transaction blocks requested */ | |
1281 | tr = current->journal_info; | |
1282 | if (tr->tr_num_buf_new + RES_STATFS + | |
1283 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1284 | /* We set blks_outside_rgrp to ensure the loop will | |
1285 | be repeated for the same rgrp, but with a new | |
1286 | transaction. */ | |
1287 | blks_outside_rgrp++; | |
1288 | /* This next part is tricky. If the buffer was added | |
1289 | to the transaction, we've already set some block | |
1290 | pointers to 0, so we better follow through and free | |
1291 | them, or we will introduce corruption (so break). | |
1292 | This may be impossible, or at least rare, but I | |
1293 | decided to cover the case regardless. | |
1294 | ||
1295 | If the buffer was not added to the transaction | |
1296 | (this call), doing so would exceed our transaction | |
1297 | size, so we need to end the transaction and start a | |
1298 | new one (so goto). */ | |
1299 | ||
1300 | if (buf_in_tr) | |
1301 | break; | |
1302 | goto out_unlock; | |
1303 | } | |
1304 | ||
1305 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1306 | buf_in_tr = true; | |
1307 | *p = 0; | |
1308 | if (bstart + blen == bn) { | |
1309 | blen++; | |
1310 | continue; | |
1311 | } | |
1312 | if (bstart) { | |
1313 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1314 | (*btotal) += blen; | |
1315 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1316 | } | |
1317 | bstart = bn; | |
1318 | blen = 1; | |
1319 | } | |
1320 | if (bstart) { | |
1321 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1322 | (*btotal) += blen; | |
1323 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1324 | } | |
1325 | out_unlock: | |
1326 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1327 | outside the rgrp we just processed, | |
1328 | do it all over again. */ | |
1329 | if (current->journal_info) { | |
5cf26b1e AG |
1330 | struct buffer_head *dibh; |
1331 | ||
1332 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1333 | if (ret) | |
1334 | goto out; | |
d552a2b9 BP |
1335 | |
1336 | /* Every transaction boundary, we rewrite the dinode | |
1337 | to keep its di_blocks current in case of failure. */ | |
1338 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = | |
b32c8c76 | 1339 | current_time(&ip->i_inode); |
d552a2b9 BP |
1340 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1341 | gfs2_dinode_out(ip, dibh->b_data); | |
5cf26b1e | 1342 | brelse(dibh); |
d552a2b9 BP |
1343 | up_write(&ip->i_rw_mutex); |
1344 | gfs2_trans_end(sdp); | |
1345 | } | |
1346 | gfs2_glock_dq_uninit(rd_gh); | |
1347 | cond_resched(); | |
1348 | goto more_rgrps; | |
1349 | } | |
1350 | out: | |
1351 | return ret; | |
1352 | } | |
1353 | ||
10d2cf94 AG |
1354 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h) |
1355 | { | |
1356 | if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0]))) | |
1357 | return false; | |
1358 | return true; | |
1359 | } | |
1360 | ||
d552a2b9 BP |
1361 | /** |
1362 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
d552a2b9 BP |
1363 | * @mp: starting metapath |
1364 | * @h: desired height to search | |
1365 | * | |
10d2cf94 | 1366 | * Assumes the metapath is valid (with buffers) out to height h. |
d552a2b9 BP |
1367 | * Returns: true if a non-null pointer was found in the metapath buffer |
1368 | * false if all remaining pointers are NULL in the buffer | |
1369 | */ | |
1370 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
10d2cf94 AG |
1371 | unsigned int h, |
1372 | __u16 *end_list, unsigned int end_aligned) | |
d552a2b9 | 1373 | { |
10d2cf94 AG |
1374 | struct buffer_head *bh = mp->mp_bh[h]; |
1375 | __be64 *first, *ptr, *end; | |
1376 | ||
1377 | first = metaptr1(h, mp); | |
1378 | ptr = first + mp->mp_list[h]; | |
1379 | end = (__be64 *)(bh->b_data + bh->b_size); | |
1380 | if (end_list && mp_eq_to_hgt(mp, end_list, h)) { | |
1381 | bool keep_end = h < end_aligned; | |
1382 | end = first + end_list[h] + keep_end; | |
1383 | } | |
d552a2b9 | 1384 | |
10d2cf94 | 1385 | while (ptr < end) { |
c4a9d189 | 1386 | if (*ptr) { /* if we have a non-null pointer */ |
10d2cf94 | 1387 | mp->mp_list[h] = ptr - first; |
c4a9d189 BP |
1388 | h++; |
1389 | if (h < GFS2_MAX_META_HEIGHT) | |
10d2cf94 | 1390 | mp->mp_list[h] = 0; |
d552a2b9 | 1391 | return true; |
c4a9d189 | 1392 | } |
10d2cf94 | 1393 | ptr++; |
d552a2b9 | 1394 | } |
10d2cf94 | 1395 | return false; |
d552a2b9 BP |
1396 | } |
1397 | ||
1398 | enum dealloc_states { | |
1399 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1400 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1401 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1402 | DEALLOC_DONE = 3, /* process complete */ | |
1403 | }; | |
b3b94faa | 1404 | |
5cf26b1e AG |
1405 | static inline void |
1406 | metapointer_range(struct metapath *mp, int height, | |
1407 | __u16 *start_list, unsigned int start_aligned, | |
10d2cf94 | 1408 | __u16 *end_list, unsigned int end_aligned, |
5cf26b1e AG |
1409 | __be64 **start, __be64 **end) |
1410 | { | |
1411 | struct buffer_head *bh = mp->mp_bh[height]; | |
1412 | __be64 *first; | |
1413 | ||
1414 | first = metaptr1(height, mp); | |
1415 | *start = first; | |
1416 | if (mp_eq_to_hgt(mp, start_list, height)) { | |
1417 | bool keep_start = height < start_aligned; | |
1418 | *start = first + start_list[height] + keep_start; | |
1419 | } | |
1420 | *end = (__be64 *)(bh->b_data + bh->b_size); | |
10d2cf94 AG |
1421 | if (end_list && mp_eq_to_hgt(mp, end_list, height)) { |
1422 | bool keep_end = height < end_aligned; | |
1423 | *end = first + end_list[height] + keep_end; | |
1424 | } | |
1425 | } | |
1426 | ||
1427 | static inline bool walk_done(struct gfs2_sbd *sdp, | |
1428 | struct metapath *mp, int height, | |
1429 | __u16 *end_list, unsigned int end_aligned) | |
1430 | { | |
1431 | __u16 end; | |
1432 | ||
1433 | if (end_list) { | |
1434 | bool keep_end = height < end_aligned; | |
1435 | if (!mp_eq_to_hgt(mp, end_list, height)) | |
1436 | return false; | |
1437 | end = end_list[height] + keep_end; | |
1438 | } else | |
1439 | end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
1440 | return mp->mp_list[height] >= end; | |
5cf26b1e AG |
1441 | } |
1442 | ||
d552a2b9 | 1443 | /** |
10d2cf94 | 1444 | * punch_hole - deallocate blocks in a file |
d552a2b9 | 1445 | * @ip: inode to truncate |
10d2cf94 AG |
1446 | * @offset: the start of the hole |
1447 | * @length: the size of the hole (or 0 for truncate) | |
1448 | * | |
1449 | * Punch a hole into a file or truncate a file at a given position. This | |
1450 | * function operates in whole blocks (@offset and @length are rounded | |
1451 | * accordingly); partially filled blocks must be cleared otherwise. | |
d552a2b9 | 1452 | * |
10d2cf94 AG |
1453 | * This function works from the bottom up, and from the right to the left. In |
1454 | * other words, it strips off the highest layer (data) before stripping any of | |
1455 | * the metadata. Doing it this way is best in case the operation is interrupted | |
1456 | * by power failure, etc. The dinode is rewritten in every transaction to | |
1457 | * guarantee integrity. | |
d552a2b9 | 1458 | */ |
10d2cf94 | 1459 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length) |
d552a2b9 BP |
1460 | { |
1461 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
bb491ce6 | 1462 | u64 maxsize = sdp->sd_heightsize[ip->i_height]; |
10d2cf94 | 1463 | struct metapath mp = {}; |
d552a2b9 BP |
1464 | struct buffer_head *dibh, *bh; |
1465 | struct gfs2_holder rd_gh; | |
cb7f0903 | 1466 | unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; |
10d2cf94 AG |
1467 | u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift; |
1468 | __u16 start_list[GFS2_MAX_META_HEIGHT]; | |
1469 | __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL; | |
4e56a641 | 1470 | unsigned int start_aligned, uninitialized_var(end_aligned); |
d552a2b9 BP |
1471 | unsigned int strip_h = ip->i_height - 1; |
1472 | u32 btotal = 0; | |
1473 | int ret, state; | |
1474 | int mp_h; /* metapath buffers are read in to this height */ | |
d552a2b9 | 1475 | u64 prev_bnr = 0; |
5cf26b1e | 1476 | __be64 *start, *end; |
b3b94faa | 1477 | |
bb491ce6 AG |
1478 | if (offset >= maxsize) { |
1479 | /* | |
1480 | * The starting point lies beyond the allocated meta-data; | |
1481 | * there are no blocks do deallocate. | |
1482 | */ | |
1483 | return 0; | |
1484 | } | |
1485 | ||
10d2cf94 AG |
1486 | /* |
1487 | * The start position of the hole is defined by lblock, start_list, and | |
1488 | * start_aligned. The end position of the hole is defined by lend, | |
1489 | * end_list, and end_aligned. | |
1490 | * | |
1491 | * start_aligned and end_aligned define down to which height the start | |
1492 | * and end positions are aligned to the metadata tree (i.e., the | |
1493 | * position is a multiple of the metadata granularity at the height | |
1494 | * above). This determines at which heights additional meta pointers | |
1495 | * needs to be preserved for the remaining data. | |
1496 | */ | |
b3b94faa | 1497 | |
10d2cf94 | 1498 | if (length) { |
10d2cf94 AG |
1499 | u64 end_offset = offset + length; |
1500 | u64 lend; | |
1501 | ||
1502 | /* | |
1503 | * Clip the end at the maximum file size for the given height: | |
1504 | * that's how far the metadata goes; files bigger than that | |
1505 | * will have additional layers of indirection. | |
1506 | */ | |
1507 | if (end_offset > maxsize) | |
1508 | end_offset = maxsize; | |
1509 | lend = end_offset >> bsize_shift; | |
1510 | ||
1511 | if (lblock >= lend) | |
1512 | return 0; | |
1513 | ||
1514 | find_metapath(sdp, lend, &mp, ip->i_height); | |
1515 | end_list = __end_list; | |
1516 | memcpy(end_list, mp.mp_list, sizeof(mp.mp_list)); | |
1517 | ||
1518 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { | |
1519 | if (end_list[mp_h]) | |
1520 | break; | |
1521 | } | |
1522 | end_aligned = mp_h; | |
1523 | } | |
1524 | ||
1525 | find_metapath(sdp, lblock, &mp, ip->i_height); | |
cb7f0903 AG |
1526 | memcpy(start_list, mp.mp_list, sizeof(start_list)); |
1527 | ||
cb7f0903 AG |
1528 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { |
1529 | if (start_list[mp_h]) | |
1530 | break; | |
1531 | } | |
1532 | start_aligned = mp_h; | |
d552a2b9 BP |
1533 | |
1534 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1535 | if (ret) | |
1536 | return ret; | |
b3b94faa | 1537 | |
d552a2b9 BP |
1538 | mp.mp_bh[0] = dibh; |
1539 | ret = lookup_metapath(ip, &mp); | |
e8b43fe0 AG |
1540 | if (ret) |
1541 | goto out_metapath; | |
c3ce5aa9 AG |
1542 | |
1543 | /* issue read-ahead on metadata */ | |
5cf26b1e AG |
1544 | for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) { |
1545 | metapointer_range(&mp, mp_h, start_list, start_aligned, | |
10d2cf94 | 1546 | end_list, end_aligned, &start, &end); |
5cf26b1e AG |
1547 | gfs2_metapath_ra(ip->i_gl, start, end); |
1548 | } | |
c3ce5aa9 | 1549 | |
e8b43fe0 | 1550 | if (mp.mp_aheight == ip->i_height) |
d552a2b9 BP |
1551 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ |
1552 | else | |
1553 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1554 | |
d552a2b9 BP |
1555 | ret = gfs2_rindex_update(sdp); |
1556 | if (ret) | |
1557 | goto out_metapath; | |
1558 | ||
1559 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1560 | if (ret) | |
1561 | goto out_metapath; | |
1562 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1563 | ||
1564 | mp_h = strip_h; | |
1565 | ||
1566 | while (state != DEALLOC_DONE) { | |
1567 | switch (state) { | |
1568 | /* Truncate a full metapath at the given strip height. | |
1569 | * Note that strip_h == mp_h in order to be in this state. */ | |
1570 | case DEALLOC_MP_FULL: | |
d552a2b9 BP |
1571 | bh = mp.mp_bh[mp_h]; |
1572 | gfs2_assert_withdraw(sdp, bh); | |
1573 | if (gfs2_assert_withdraw(sdp, | |
1574 | prev_bnr != bh->b_blocknr)) { | |
1575 | printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, " | |
1576 | "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n", | |
1577 | sdp->sd_fsname, | |
1578 | (unsigned long long)ip->i_no_addr, | |
1579 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1580 | } | |
1581 | prev_bnr = bh->b_blocknr; | |
cb7f0903 | 1582 | |
5cf26b1e AG |
1583 | if (gfs2_metatype_check(sdp, bh, |
1584 | (mp_h ? GFS2_METATYPE_IN : | |
1585 | GFS2_METATYPE_DI))) { | |
1586 | ret = -EIO; | |
1587 | goto out; | |
1588 | } | |
1589 | ||
10d2cf94 AG |
1590 | /* |
1591 | * Below, passing end_aligned as 0 gives us the | |
1592 | * metapointer range excluding the end point: the end | |
1593 | * point is the first metapath we must not deallocate! | |
1594 | */ | |
1595 | ||
5cf26b1e | 1596 | metapointer_range(&mp, mp_h, start_list, start_aligned, |
10d2cf94 | 1597 | end_list, 0 /* end_aligned */, |
5cf26b1e AG |
1598 | &start, &end); |
1599 | ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h], | |
1600 | start, end, | |
1601 | mp_h != ip->i_height - 1, | |
1602 | &btotal); | |
cb7f0903 | 1603 | |
d552a2b9 BP |
1604 | /* If we hit an error or just swept dinode buffer, |
1605 | just exit. */ | |
1606 | if (ret || !mp_h) { | |
1607 | state = DEALLOC_DONE; | |
1608 | break; | |
1609 | } | |
1610 | state = DEALLOC_MP_LOWER; | |
1611 | break; | |
1612 | ||
1613 | /* lower the metapath strip height */ | |
1614 | case DEALLOC_MP_LOWER: | |
1615 | /* We're done with the current buffer, so release it, | |
1616 | unless it's the dinode buffer. Then back up to the | |
1617 | previous pointer. */ | |
1618 | if (mp_h) { | |
1619 | brelse(mp.mp_bh[mp_h]); | |
1620 | mp.mp_bh[mp_h] = NULL; | |
1621 | } | |
1622 | /* If we can't get any lower in height, we've stripped | |
1623 | off all we can. Next step is to back up and start | |
1624 | stripping the previous level of metadata. */ | |
1625 | if (mp_h == 0) { | |
1626 | strip_h--; | |
cb7f0903 | 1627 | memcpy(mp.mp_list, start_list, sizeof(start_list)); |
d552a2b9 BP |
1628 | mp_h = strip_h; |
1629 | state = DEALLOC_FILL_MP; | |
1630 | break; | |
1631 | } | |
1632 | mp.mp_list[mp_h] = 0; | |
1633 | mp_h--; /* search one metadata height down */ | |
d552a2b9 | 1634 | mp.mp_list[mp_h]++; |
10d2cf94 AG |
1635 | if (walk_done(sdp, &mp, mp_h, end_list, end_aligned)) |
1636 | break; | |
d552a2b9 BP |
1637 | /* Here we've found a part of the metapath that is not |
1638 | * allocated. We need to search at that height for the | |
1639 | * next non-null pointer. */ | |
10d2cf94 | 1640 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) { |
d552a2b9 BP |
1641 | state = DEALLOC_FILL_MP; |
1642 | mp_h++; | |
1643 | } | |
1644 | /* No more non-null pointers at this height. Back up | |
1645 | to the previous height and try again. */ | |
1646 | break; /* loop around in the same state */ | |
1647 | ||
1648 | /* Fill the metapath with buffers to the given height. */ | |
1649 | case DEALLOC_FILL_MP: | |
1650 | /* Fill the buffers out to the current height. */ | |
1651 | ret = fillup_metapath(ip, &mp, mp_h); | |
c3ce5aa9 | 1652 | if (ret < 0) |
d552a2b9 | 1653 | goto out; |
c3ce5aa9 AG |
1654 | |
1655 | /* issue read-ahead on metadata */ | |
1656 | if (mp.mp_aheight > 1) { | |
5cf26b1e AG |
1657 | for (; ret > 1; ret--) { |
1658 | metapointer_range(&mp, mp.mp_aheight - ret, | |
1659 | start_list, start_aligned, | |
10d2cf94 | 1660 | end_list, end_aligned, |
5cf26b1e AG |
1661 | &start, &end); |
1662 | gfs2_metapath_ra(ip->i_gl, start, end); | |
1663 | } | |
c3ce5aa9 | 1664 | } |
d552a2b9 BP |
1665 | |
1666 | /* If buffers found for the entire strip height */ | |
e8b43fe0 | 1667 | if (mp.mp_aheight - 1 == strip_h) { |
d552a2b9 BP |
1668 | state = DEALLOC_MP_FULL; |
1669 | break; | |
1670 | } | |
e8b43fe0 AG |
1671 | if (mp.mp_aheight < ip->i_height) /* We have a partial height */ |
1672 | mp_h = mp.mp_aheight - 1; | |
d552a2b9 BP |
1673 | |
1674 | /* If we find a non-null block pointer, crawl a bit | |
1675 | higher up in the metapath and try again, otherwise | |
1676 | we need to look lower for a new starting point. */ | |
10d2cf94 | 1677 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) |
d552a2b9 BP |
1678 | mp_h++; |
1679 | else | |
1680 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1681 | break; |
d552a2b9 | 1682 | } |
b3b94faa DT |
1683 | } |
1684 | ||
d552a2b9 BP |
1685 | if (btotal) { |
1686 | if (current->journal_info == NULL) { | |
1687 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1688 | RES_QUOTA, 0); | |
1689 | if (ret) | |
1690 | goto out; | |
1691 | down_write(&ip->i_rw_mutex); | |
1692 | } | |
1693 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
1694 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
1695 | ip->i_inode.i_gid); | |
b32c8c76 | 1696 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
d552a2b9 BP |
1697 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1698 | gfs2_dinode_out(ip, dibh->b_data); | |
1699 | up_write(&ip->i_rw_mutex); | |
1700 | gfs2_trans_end(sdp); | |
1701 | } | |
b3b94faa | 1702 | |
d552a2b9 BP |
1703 | out: |
1704 | if (gfs2_holder_initialized(&rd_gh)) | |
1705 | gfs2_glock_dq_uninit(&rd_gh); | |
1706 | if (current->journal_info) { | |
1707 | up_write(&ip->i_rw_mutex); | |
1708 | gfs2_trans_end(sdp); | |
1709 | cond_resched(); | |
1710 | } | |
1711 | gfs2_quota_unhold(ip); | |
1712 | out_metapath: | |
1713 | release_metapath(&mp); | |
1714 | return ret; | |
b3b94faa DT |
1715 | } |
1716 | ||
1717 | static int trunc_end(struct gfs2_inode *ip) | |
1718 | { | |
feaa7bba | 1719 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1720 | struct buffer_head *dibh; |
1721 | int error; | |
1722 | ||
1723 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1724 | if (error) | |
1725 | return error; | |
1726 | ||
1727 | down_write(&ip->i_rw_mutex); | |
1728 | ||
1729 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1730 | if (error) | |
1731 | goto out; | |
1732 | ||
a2e0f799 | 1733 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1734 | ip->i_height = 0; |
ce276b06 | 1735 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 1736 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 1737 | gfs2_ordered_del_inode(ip); |
b3b94faa | 1738 | } |
078cd827 | 1739 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
383f01fb | 1740 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1741 | |
350a9b0a | 1742 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 1743 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1744 | brelse(dibh); |
1745 | ||
a91ea69f | 1746 | out: |
b3b94faa | 1747 | up_write(&ip->i_rw_mutex); |
b3b94faa | 1748 | gfs2_trans_end(sdp); |
b3b94faa DT |
1749 | return error; |
1750 | } | |
1751 | ||
1752 | /** | |
1753 | * do_shrink - make a file smaller | |
ff8f33c8 | 1754 | * @inode: the inode |
ff8f33c8 | 1755 | * @newsize: the size to make the file |
b3b94faa | 1756 | * |
ff8f33c8 SW |
1757 | * Called with an exclusive lock on @inode. The @size must |
1758 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
1759 | * |
1760 | * Returns: errno | |
1761 | */ | |
1762 | ||
8b5860a3 | 1763 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 1764 | { |
ff8f33c8 | 1765 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
1766 | int error; |
1767 | ||
8b5860a3 | 1768 | error = trunc_start(inode, newsize); |
b3b94faa DT |
1769 | if (error < 0) |
1770 | return error; | |
ff8f33c8 | 1771 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
1772 | return 0; |
1773 | ||
10d2cf94 | 1774 | error = punch_hole(ip, newsize, 0); |
ff8f33c8 | 1775 | if (error == 0) |
b3b94faa DT |
1776 | error = trunc_end(ip); |
1777 | ||
1778 | return error; | |
1779 | } | |
1780 | ||
ff8f33c8 | 1781 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 1782 | { |
ff8f33c8 SW |
1783 | int ret; |
1784 | ||
8b5860a3 | 1785 | ret = do_shrink(inode, inode->i_size); |
ff8f33c8 SW |
1786 | WARN_ON(ret != 0); |
1787 | } | |
1788 | ||
1789 | /** | |
1790 | * do_grow - Touch and update inode size | |
1791 | * @inode: The inode | |
1792 | * @size: The new size | |
1793 | * | |
1794 | * This function updates the timestamps on the inode and | |
1795 | * may also increase the size of the inode. This function | |
1796 | * must not be called with @size any smaller than the current | |
1797 | * inode size. | |
1798 | * | |
1799 | * Although it is not strictly required to unstuff files here, | |
1800 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
1801 | * code which will result in a buffer overrun if the size is larger | |
1802 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 1803 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
1804 | * just update the inode size directly. |
1805 | * | |
1806 | * Returns: 0 on success, or -ve on error | |
1807 | */ | |
1808 | ||
1809 | static int do_grow(struct inode *inode, u64 size) | |
1810 | { | |
1811 | struct gfs2_inode *ip = GFS2_I(inode); | |
1812 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 1813 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
1814 | struct buffer_head *dibh; |
1815 | int error; | |
2f7ee358 | 1816 | int unstuff = 0; |
a13b8c5f | 1817 | |
235628c5 | 1818 | if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) { |
b8fbf471 | 1819 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 1820 | if (error) |
5407e242 | 1821 | return error; |
ff8f33c8 | 1822 | |
7b9cff46 | 1823 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
1824 | if (error) |
1825 | goto do_grow_qunlock; | |
2f7ee358 | 1826 | unstuff = 1; |
ff8f33c8 SW |
1827 | } |
1828 | ||
a01aedfe BP |
1829 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
1830 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? | |
1831 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 1832 | if (error) |
ff8f33c8 | 1833 | goto do_grow_release; |
a13b8c5f | 1834 | |
2f7ee358 | 1835 | if (unstuff) { |
ff8f33c8 SW |
1836 | error = gfs2_unstuff_dinode(ip, NULL); |
1837 | if (error) | |
1838 | goto do_end_trans; | |
1839 | } | |
a13b8c5f WC |
1840 | |
1841 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1842 | if (error) | |
ff8f33c8 | 1843 | goto do_end_trans; |
a13b8c5f | 1844 | |
ff8f33c8 | 1845 | i_size_write(inode, size); |
078cd827 | 1846 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
350a9b0a | 1847 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
1848 | gfs2_dinode_out(ip, dibh->b_data); |
1849 | brelse(dibh); | |
1850 | ||
ff8f33c8 | 1851 | do_end_trans: |
a13b8c5f | 1852 | gfs2_trans_end(sdp); |
ff8f33c8 | 1853 | do_grow_release: |
2f7ee358 | 1854 | if (unstuff) { |
ff8f33c8 SW |
1855 | gfs2_inplace_release(ip); |
1856 | do_grow_qunlock: | |
1857 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 1858 | } |
a13b8c5f WC |
1859 | return error; |
1860 | } | |
1861 | ||
b3b94faa | 1862 | /** |
ff8f33c8 SW |
1863 | * gfs2_setattr_size - make a file a given size |
1864 | * @inode: the inode | |
1865 | * @newsize: the size to make the file | |
b3b94faa | 1866 | * |
ff8f33c8 | 1867 | * The file size can grow, shrink, or stay the same size. This |
3e7aafc3 | 1868 | * is called holding i_rwsem and an exclusive glock on the inode |
ff8f33c8 | 1869 | * in question. |
b3b94faa DT |
1870 | * |
1871 | * Returns: errno | |
1872 | */ | |
1873 | ||
ff8f33c8 | 1874 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 1875 | { |
af5c2697 | 1876 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 1877 | int ret; |
b3b94faa | 1878 | |
ff8f33c8 | 1879 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 1880 | |
ff8f33c8 SW |
1881 | ret = inode_newsize_ok(inode, newsize); |
1882 | if (ret) | |
1883 | return ret; | |
b3b94faa | 1884 | |
562c72aa CH |
1885 | inode_dio_wait(inode); |
1886 | ||
b54e9a0b | 1887 | ret = gfs2_rsqa_alloc(ip); |
d2b47cfb | 1888 | if (ret) |
2b3dcf35 | 1889 | goto out; |
d2b47cfb | 1890 | |
8b5860a3 | 1891 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
1892 | ret = do_grow(inode, newsize); |
1893 | goto out; | |
1894 | } | |
ff8f33c8 | 1895 | |
8b5860a3 | 1896 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 1897 | out: |
a097dc7e | 1898 | gfs2_rsqa_delete(ip, NULL); |
2b3dcf35 | 1899 | return ret; |
b3b94faa DT |
1900 | } |
1901 | ||
1902 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
1903 | { | |
1904 | int error; | |
10d2cf94 | 1905 | error = punch_hole(ip, i_size_read(&ip->i_inode), 0); |
b3b94faa DT |
1906 | if (!error) |
1907 | error = trunc_end(ip); | |
1908 | return error; | |
1909 | } | |
1910 | ||
1911 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
1912 | { | |
10d2cf94 | 1913 | return punch_hole(ip, 0, 0); |
b3b94faa DT |
1914 | } |
1915 | ||
b50f227b SW |
1916 | /** |
1917 | * gfs2_free_journal_extents - Free cached journal bmap info | |
1918 | * @jd: The journal | |
1919 | * | |
1920 | */ | |
1921 | ||
1922 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
1923 | { | |
1924 | struct gfs2_journal_extent *jext; | |
1925 | ||
1926 | while(!list_empty(&jd->extent_list)) { | |
1927 | jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list); | |
1928 | list_del(&jext->list); | |
1929 | kfree(jext); | |
1930 | } | |
1931 | } | |
1932 | ||
1933 | /** | |
1934 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
1935 | * @jd: The journal descriptor | |
1936 | * @lblock: The logical block at start of new extent | |
c62baf65 | 1937 | * @dblock: The physical block at start of new extent |
b50f227b SW |
1938 | * @blocks: Size of extent in fs blocks |
1939 | * | |
1940 | * Returns: 0 on success or -ENOMEM | |
1941 | */ | |
1942 | ||
1943 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
1944 | { | |
1945 | struct gfs2_journal_extent *jext; | |
1946 | ||
1947 | if (!list_empty(&jd->extent_list)) { | |
1948 | jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list); | |
1949 | if ((jext->dblock + jext->blocks) == dblock) { | |
1950 | jext->blocks += blocks; | |
1951 | return 0; | |
1952 | } | |
1953 | } | |
1954 | ||
1955 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
1956 | if (jext == NULL) | |
1957 | return -ENOMEM; | |
1958 | jext->dblock = dblock; | |
1959 | jext->lblock = lblock; | |
1960 | jext->blocks = blocks; | |
1961 | list_add_tail(&jext->list, &jd->extent_list); | |
1962 | jd->nr_extents++; | |
1963 | return 0; | |
1964 | } | |
1965 | ||
1966 | /** | |
1967 | * gfs2_map_journal_extents - Cache journal bmap info | |
1968 | * @sdp: The super block | |
1969 | * @jd: The journal to map | |
1970 | * | |
1971 | * Create a reusable "extent" mapping from all logical | |
1972 | * blocks to all physical blocks for the given journal. This will save | |
1973 | * us time when writing journal blocks. Most journals will have only one | |
1974 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
1975 | * arranges the journal blocks sequentially to maximize performance. | |
1976 | * So the extent would map the first block for the entire file length. | |
1977 | * However, gfs2_jadd can happen while file activity is happening, so | |
1978 | * those journals may not be sequential. Less likely is the case where | |
1979 | * the users created their own journals by mounting the metafs and | |
1980 | * laying it out. But it's still possible. These journals might have | |
1981 | * several extents. | |
1982 | * | |
1983 | * Returns: 0 on success, or error on failure | |
1984 | */ | |
1985 | ||
1986 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
1987 | { | |
1988 | u64 lblock = 0; | |
1989 | u64 lblock_stop; | |
1990 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
1991 | struct buffer_head bh; | |
1992 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
1993 | u64 size; | |
1994 | int rc; | |
1995 | ||
1996 | lblock_stop = i_size_read(jd->jd_inode) >> shift; | |
1997 | size = (lblock_stop - lblock) << shift; | |
1998 | jd->nr_extents = 0; | |
1999 | WARN_ON(!list_empty(&jd->extent_list)); | |
2000 | ||
2001 | do { | |
2002 | bh.b_state = 0; | |
2003 | bh.b_blocknr = 0; | |
2004 | bh.b_size = size; | |
2005 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
2006 | if (rc || !buffer_mapped(&bh)) | |
2007 | goto fail; | |
2008 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
2009 | if (rc) | |
2010 | goto fail; | |
2011 | size -= bh.b_size; | |
2012 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2013 | } while(size > 0); | |
2014 | ||
2015 | fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid, | |
2016 | jd->nr_extents); | |
2017 | return 0; | |
2018 | ||
2019 | fail: | |
2020 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
2021 | rc, jd->jd_jid, | |
2022 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
2023 | jd->nr_extents); | |
2024 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
2025 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
2026 | bh.b_state, (unsigned long long)bh.b_size); | |
2027 | gfs2_free_journal_extents(jd); | |
2028 | return rc; | |
2029 | } | |
2030 | ||
b3b94faa DT |
2031 | /** |
2032 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
2033 | * @ip: the file being written to | |
2034 | * @offset: the offset to write to | |
2035 | * @len: the number of bytes being written | |
b3b94faa | 2036 | * |
461cb419 | 2037 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
2038 | */ |
2039 | ||
cd915493 | 2040 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 2041 | unsigned int len) |
b3b94faa | 2042 | { |
feaa7bba | 2043 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
2044 | struct buffer_head bh; |
2045 | unsigned int shift; | |
2046 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 2047 | u64 end_of_file; |
b3b94faa | 2048 | |
b3b94faa DT |
2049 | if (!len) |
2050 | return 0; | |
2051 | ||
2052 | if (gfs2_is_stuffed(ip)) { | |
235628c5 | 2053 | if (offset + len > gfs2_max_stuffed_size(ip)) |
461cb419 | 2054 | return 1; |
b3b94faa DT |
2055 | return 0; |
2056 | } | |
2057 | ||
941e6d7d | 2058 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 2059 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 2060 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
2061 | lblock = offset >> shift; |
2062 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
2063 | if (lblock_stop > end_of_file) | |
461cb419 | 2064 | return 1; |
b3b94faa | 2065 | |
941e6d7d SW |
2066 | size = (lblock_stop - lblock) << shift; |
2067 | do { | |
2068 | bh.b_state = 0; | |
2069 | bh.b_size = size; | |
2070 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
2071 | if (!buffer_mapped(&bh)) | |
461cb419 | 2072 | return 1; |
941e6d7d SW |
2073 | size -= bh.b_size; |
2074 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2075 | } while(size > 0); | |
b3b94faa DT |
2076 | |
2077 | return 0; | |
2078 | } | |
2079 | ||
4e56a641 AG |
2080 | static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length) |
2081 | { | |
2082 | struct gfs2_inode *ip = GFS2_I(inode); | |
2083 | struct buffer_head *dibh; | |
2084 | int error; | |
2085 | ||
2086 | if (offset >= inode->i_size) | |
2087 | return 0; | |
2088 | if (offset + length > inode->i_size) | |
2089 | length = inode->i_size - offset; | |
2090 | ||
2091 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2092 | if (error) | |
2093 | return error; | |
2094 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
2095 | memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0, | |
2096 | length); | |
2097 | brelse(dibh); | |
2098 | return 0; | |
2099 | } | |
2100 | ||
2101 | static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset, | |
2102 | loff_t length) | |
2103 | { | |
2104 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2105 | loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
2106 | int error; | |
2107 | ||
2108 | while (length) { | |
2109 | struct gfs2_trans *tr; | |
2110 | loff_t chunk; | |
2111 | unsigned int offs; | |
2112 | ||
2113 | chunk = length; | |
2114 | if (chunk > max_chunk) | |
2115 | chunk = max_chunk; | |
2116 | ||
2117 | offs = offset & ~PAGE_MASK; | |
2118 | if (offs && chunk > PAGE_SIZE) | |
2119 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
2120 | ||
2121 | truncate_pagecache_range(inode, offset, chunk); | |
2122 | offset += chunk; | |
2123 | length -= chunk; | |
2124 | ||
2125 | tr = current->journal_info; | |
2126 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
2127 | continue; | |
2128 | ||
2129 | gfs2_trans_end(sdp); | |
2130 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
2131 | if (error) | |
2132 | return error; | |
2133 | } | |
2134 | return 0; | |
2135 | } | |
2136 | ||
2137 | int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length) | |
2138 | { | |
2139 | struct inode *inode = file_inode(file); | |
2140 | struct gfs2_inode *ip = GFS2_I(inode); | |
2141 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2142 | int error; | |
2143 | ||
2144 | if (gfs2_is_jdata(ip)) | |
2145 | error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA, | |
2146 | GFS2_JTRUNC_REVOKES); | |
2147 | else | |
2148 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2149 | if (error) | |
2150 | return error; | |
2151 | ||
2152 | if (gfs2_is_stuffed(ip)) { | |
2153 | error = stuffed_zero_range(inode, offset, length); | |
2154 | if (error) | |
2155 | goto out; | |
2156 | } else { | |
2157 | unsigned int start_off, end_off, blocksize; | |
2158 | ||
2159 | blocksize = i_blocksize(inode); | |
2160 | start_off = offset & (blocksize - 1); | |
2161 | end_off = (offset + length) & (blocksize - 1); | |
2162 | if (start_off) { | |
2163 | unsigned int len = length; | |
2164 | if (length > blocksize - start_off) | |
2165 | len = blocksize - start_off; | |
2166 | error = gfs2_block_zero_range(inode, offset, len); | |
2167 | if (error) | |
2168 | goto out; | |
2169 | if (start_off + length < blocksize) | |
2170 | end_off = 0; | |
2171 | } | |
2172 | if (end_off) { | |
2173 | error = gfs2_block_zero_range(inode, | |
2174 | offset + length - end_off, end_off); | |
2175 | if (error) | |
2176 | goto out; | |
2177 | } | |
2178 | } | |
2179 | ||
2180 | if (gfs2_is_jdata(ip)) { | |
2181 | BUG_ON(!current->journal_info); | |
2182 | gfs2_journaled_truncate_range(inode, offset, length); | |
2183 | } else | |
2184 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2185 | ||
2186 | file_update_time(file); | |
2187 | mark_inode_dirty(inode); | |
2188 | ||
2189 | if (current->journal_info) | |
2190 | gfs2_trans_end(sdp); | |
2191 | ||
2192 | if (!gfs2_is_stuffed(ip)) | |
2193 | error = punch_hole(ip, offset, length); | |
2194 | ||
2195 | out: | |
2196 | if (current->journal_info) | |
2197 | gfs2_trans_end(sdp); | |
2198 | return error; | |
2199 | } |