]>
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" |
64bc06bb | 31 | #include "aops.h" |
63997775 | 32 | #include "trace_gfs2.h" |
b3b94faa DT |
33 | |
34 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
35 | * block is 512, so __u16 is fine for that. It saves stack space to | |
36 | * keep it small. | |
37 | */ | |
38 | struct metapath { | |
dbac6710 | 39 | struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT]; |
b3b94faa | 40 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; |
5f8bd444 BP |
41 | int mp_fheight; /* find_metapath height */ |
42 | int mp_aheight; /* actual height (lookup height) */ | |
b3b94faa DT |
43 | }; |
44 | ||
64bc06bb AG |
45 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length); |
46 | ||
f25ef0c1 SW |
47 | /** |
48 | * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page | |
49 | * @ip: the inode | |
50 | * @dibh: the dinode buffer | |
51 | * @block: the block number that was allocated | |
ff8f33c8 | 52 | * @page: The (optional) page. This is looked up if @page is NULL |
f25ef0c1 SW |
53 | * |
54 | * Returns: errno | |
55 | */ | |
56 | ||
57 | static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, | |
cd915493 | 58 | u64 block, struct page *page) |
f25ef0c1 | 59 | { |
f25ef0c1 SW |
60 | struct inode *inode = &ip->i_inode; |
61 | struct buffer_head *bh; | |
62 | int release = 0; | |
63 | ||
64 | if (!page || page->index) { | |
220cca2a | 65 | page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS); |
f25ef0c1 SW |
66 | if (!page) |
67 | return -ENOMEM; | |
68 | release = 1; | |
69 | } | |
70 | ||
71 | if (!PageUptodate(page)) { | |
72 | void *kaddr = kmap(page); | |
602c89d2 SW |
73 | u64 dsize = i_size_read(inode); |
74 | ||
235628c5 AG |
75 | if (dsize > gfs2_max_stuffed_size(ip)) |
76 | dsize = gfs2_max_stuffed_size(ip); | |
f25ef0c1 | 77 | |
602c89d2 | 78 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 79 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
f25ef0c1 SW |
80 | kunmap(page); |
81 | ||
82 | SetPageUptodate(page); | |
83 | } | |
84 | ||
85 | if (!page_has_buffers(page)) | |
47a9a527 FF |
86 | create_empty_buffers(page, BIT(inode->i_blkbits), |
87 | BIT(BH_Uptodate)); | |
f25ef0c1 SW |
88 | |
89 | bh = page_buffers(page); | |
90 | ||
91 | if (!buffer_mapped(bh)) | |
92 | map_bh(bh, inode->i_sb, block); | |
93 | ||
94 | set_buffer_uptodate(bh); | |
845802b1 | 95 | if (gfs2_is_jdata(ip)) |
350a9b0a | 96 | gfs2_trans_add_data(ip->i_gl, bh); |
845802b1 AG |
97 | else { |
98 | mark_buffer_dirty(bh); | |
99 | gfs2_ordered_add_inode(ip); | |
100 | } | |
f25ef0c1 SW |
101 | |
102 | if (release) { | |
103 | unlock_page(page); | |
09cbfeaf | 104 | put_page(page); |
f25ef0c1 SW |
105 | } |
106 | ||
107 | return 0; | |
108 | } | |
109 | ||
b3b94faa DT |
110 | /** |
111 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
112 | * @ip: The GFS2 inode to unstuff | |
ff8f33c8 | 113 | * @page: The (optional) page. This is looked up if the @page is NULL |
b3b94faa DT |
114 | * |
115 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
116 | * that the height can be grown in the traditional way. | |
117 | * | |
118 | * Returns: errno | |
119 | */ | |
120 | ||
f25ef0c1 | 121 | int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
122 | { |
123 | struct buffer_head *bh, *dibh; | |
48516ced | 124 | struct gfs2_dinode *di; |
cd915493 | 125 | u64 block = 0; |
18ec7d5c | 126 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
127 | int error; |
128 | ||
129 | down_write(&ip->i_rw_mutex); | |
130 | ||
131 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
132 | if (error) | |
133 | goto out; | |
907b9bce | 134 | |
a2e0f799 | 135 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
136 | /* Get a free block, fill it with the stuffed data, |
137 | and write it out to disk */ | |
138 | ||
b45e41d7 | 139 | unsigned int n = 1; |
6e87ed0f | 140 | error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL); |
09010978 SW |
141 | if (error) |
142 | goto out_brelse; | |
18ec7d5c | 143 | if (isdir) { |
5731be53 | 144 | gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 145 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
146 | if (error) |
147 | goto out_brelse; | |
48516ced | 148 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
149 | dibh, sizeof(struct gfs2_dinode)); |
150 | brelse(bh); | |
151 | } else { | |
f25ef0c1 | 152 | error = gfs2_unstuffer_page(ip, dibh, block, page); |
b3b94faa DT |
153 | if (error) |
154 | goto out_brelse; | |
155 | } | |
156 | } | |
157 | ||
158 | /* Set up the pointer to the new block */ | |
159 | ||
350a9b0a | 160 | gfs2_trans_add_meta(ip->i_gl, dibh); |
48516ced | 161 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
162 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
163 | ||
a2e0f799 | 164 | if (i_size_read(&ip->i_inode)) { |
48516ced | 165 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
166 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
167 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
168 | } |
169 | ||
ecc30c79 | 170 | ip->i_height = 1; |
48516ced | 171 | di->di_height = cpu_to_be16(1); |
b3b94faa | 172 | |
a91ea69f | 173 | out_brelse: |
b3b94faa | 174 | brelse(dibh); |
a91ea69f | 175 | out: |
b3b94faa | 176 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
177 | return error; |
178 | } | |
179 | ||
b3b94faa DT |
180 | |
181 | /** | |
182 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 183 | * @sdp: The superblock |
b3b94faa | 184 | * @block: The disk block to look up |
07e23d68 | 185 | * @mp: The metapath to return the result in |
9b8c81d1 | 186 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
187 | * |
188 | * This routine returns a struct metapath structure that defines a path | |
189 | * through the metadata of inode "ip" to get to block "block". | |
190 | * | |
191 | * Example: | |
192 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
193 | * filesystem with a blocksize of 4096. | |
194 | * | |
195 | * find_metapath() would return a struct metapath structure set to: | |
07e23d68 | 196 | * mp_fheight = 3, mp_list[0] = 0, mp_list[1] = 48, and mp_list[2] = 165. |
b3b94faa DT |
197 | * |
198 | * That means that in order to get to the block containing the byte at | |
199 | * offset 101342453, we would load the indirect block pointed to by pointer | |
200 | * 0 in the dinode. We would then load the indirect block pointed to by | |
201 | * pointer 48 in that indirect block. We would then load the data block | |
202 | * pointed to by pointer 165 in that indirect block. | |
203 | * | |
204 | * ---------------------------------------- | |
205 | * | Dinode | | | |
206 | * | | 4| | |
207 | * | |0 1 2 3 4 5 9| | |
208 | * | | 6| | |
209 | * ---------------------------------------- | |
210 | * | | |
211 | * | | |
212 | * V | |
213 | * ---------------------------------------- | |
214 | * | Indirect Block | | |
215 | * | 5| | |
216 | * | 4 4 4 4 4 5 5 1| | |
217 | * |0 5 6 7 8 9 0 1 2| | |
218 | * ---------------------------------------- | |
219 | * | | |
220 | * | | |
221 | * V | |
222 | * ---------------------------------------- | |
223 | * | Indirect Block | | |
224 | * | 1 1 1 1 1 5| | |
225 | * | 6 6 6 6 6 1| | |
226 | * |0 3 4 5 6 7 2| | |
227 | * ---------------------------------------- | |
228 | * | | |
229 | * | | |
230 | * V | |
231 | * ---------------------------------------- | |
232 | * | Data block containing offset | | |
233 | * | 101342453 | | |
234 | * | | | |
235 | * | | | |
236 | * ---------------------------------------- | |
237 | * | |
238 | */ | |
239 | ||
9b8c81d1 SW |
240 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
241 | struct metapath *mp, unsigned int height) | |
b3b94faa | 242 | { |
b3b94faa DT |
243 | unsigned int i; |
244 | ||
5f8bd444 | 245 | mp->mp_fheight = height; |
9b8c81d1 | 246 | for (i = height; i--;) |
7eabb77e | 247 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
248 | } |
249 | ||
5af4e7a0 | 250 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 251 | { |
5af4e7a0 BM |
252 | if (mp->mp_list[0] == 0) |
253 | return 2; | |
254 | return 1; | |
9b8c81d1 SW |
255 | } |
256 | ||
d552a2b9 | 257 | /** |
20cdc193 | 258 | * metaptr1 - Return the first possible metadata pointer in a metapath buffer |
d552a2b9 BP |
259 | * @height: The metadata height (0 = dinode) |
260 | * @mp: The metapath | |
261 | */ | |
262 | static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp) | |
263 | { | |
264 | struct buffer_head *bh = mp->mp_bh[height]; | |
265 | if (height == 0) | |
266 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode))); | |
267 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header))); | |
268 | } | |
269 | ||
b3b94faa DT |
270 | /** |
271 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
272 | * @height: The metadata height (0 = dinode) |
273 | * @mp: The metapath | |
274 | * | |
275 | * Return a pointer to the block number of the next height of the metadata | |
276 | * tree given a buffer containing the pointer to the current height of the | |
277 | * metadata tree. | |
278 | */ | |
279 | ||
9b8c81d1 | 280 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 281 | { |
d552a2b9 BP |
282 | __be64 *p = metaptr1(height, mp); |
283 | return p + mp->mp_list[height]; | |
b3b94faa DT |
284 | } |
285 | ||
7841b9f0 AG |
286 | static inline const __be64 *metaend(unsigned int height, const struct metapath *mp) |
287 | { | |
288 | const struct buffer_head *bh = mp->mp_bh[height]; | |
289 | return (const __be64 *)(bh->b_data + bh->b_size); | |
290 | } | |
291 | ||
292 | static void clone_metapath(struct metapath *clone, struct metapath *mp) | |
293 | { | |
294 | unsigned int hgt; | |
295 | ||
296 | *clone = *mp; | |
297 | for (hgt = 0; hgt < mp->mp_aheight; hgt++) | |
298 | get_bh(clone->mp_bh[hgt]); | |
299 | } | |
300 | ||
5cf26b1e | 301 | static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end) |
b99b98dc | 302 | { |
b99b98dc SW |
303 | const __be64 *t; |
304 | ||
5cf26b1e | 305 | for (t = start; t < end; t++) { |
c3ce5aa9 AG |
306 | struct buffer_head *rabh; |
307 | ||
b99b98dc SW |
308 | if (!*t) |
309 | continue; | |
310 | ||
311 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
312 | if (trylock_buffer(rabh)) { | |
313 | if (!buffer_uptodate(rabh)) { | |
314 | rabh->b_end_io = end_buffer_read_sync; | |
e477b24b CL |
315 | submit_bh(REQ_OP_READ, |
316 | REQ_RAHEAD | REQ_META | REQ_PRIO, | |
317 | rabh); | |
b99b98dc SW |
318 | continue; |
319 | } | |
320 | unlock_buffer(rabh); | |
321 | } | |
322 | brelse(rabh); | |
323 | } | |
324 | } | |
325 | ||
e8b43fe0 AG |
326 | static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, |
327 | unsigned int x, unsigned int h) | |
d552a2b9 | 328 | { |
e8b43fe0 AG |
329 | for (; x < h; x++) { |
330 | __be64 *ptr = metapointer(x, mp); | |
331 | u64 dblock = be64_to_cpu(*ptr); | |
332 | int ret; | |
d552a2b9 | 333 | |
e8b43fe0 AG |
334 | if (!dblock) |
335 | break; | |
336 | ret = gfs2_meta_indirect_buffer(ip, x + 1, dblock, &mp->mp_bh[x + 1]); | |
337 | if (ret) | |
338 | return ret; | |
339 | } | |
340 | mp->mp_aheight = x + 1; | |
341 | return 0; | |
d552a2b9 BP |
342 | } |
343 | ||
b3b94faa | 344 | /** |
9b8c81d1 SW |
345 | * lookup_metapath - Walk the metadata tree to a specific point |
346 | * @ip: The inode | |
b3b94faa | 347 | * @mp: The metapath |
b3b94faa | 348 | * |
9b8c81d1 SW |
349 | * Assumes that the inode's buffer has already been looked up and |
350 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
351 | * by find_metapath(). | |
352 | * | |
353 | * If this function encounters part of the tree which has not been | |
354 | * allocated, it returns the current height of the tree at the point | |
355 | * at which it found the unallocated block. Blocks which are found are | |
356 | * added to the mp->mp_bh[] list. | |
b3b94faa | 357 | * |
e8b43fe0 | 358 | * Returns: error |
b3b94faa DT |
359 | */ |
360 | ||
9b8c81d1 | 361 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 362 | { |
e8b43fe0 | 363 | return __fillup_metapath(ip, mp, 0, ip->i_height - 1); |
dbac6710 SW |
364 | } |
365 | ||
d552a2b9 BP |
366 | /** |
367 | * fillup_metapath - fill up buffers for the metadata path to a specific height | |
368 | * @ip: The inode | |
369 | * @mp: The metapath | |
370 | * @h: The height to which it should be mapped | |
371 | * | |
372 | * Similar to lookup_metapath, but does lookups for a range of heights | |
373 | * | |
c3ce5aa9 | 374 | * Returns: error or the number of buffers filled |
d552a2b9 BP |
375 | */ |
376 | ||
377 | static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h) | |
378 | { | |
e8b43fe0 | 379 | unsigned int x = 0; |
c3ce5aa9 | 380 | int ret; |
d552a2b9 BP |
381 | |
382 | if (h) { | |
383 | /* find the first buffer we need to look up. */ | |
e8b43fe0 AG |
384 | for (x = h - 1; x > 0; x--) { |
385 | if (mp->mp_bh[x]) | |
386 | break; | |
d552a2b9 BP |
387 | } |
388 | } | |
c3ce5aa9 AG |
389 | ret = __fillup_metapath(ip, mp, x, h); |
390 | if (ret) | |
391 | return ret; | |
392 | return mp->mp_aheight - x - 1; | |
d552a2b9 BP |
393 | } |
394 | ||
64bc06bb | 395 | static void release_metapath(struct metapath *mp) |
dbac6710 SW |
396 | { |
397 | int i; | |
398 | ||
9b8c81d1 SW |
399 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
400 | if (mp->mp_bh[i] == NULL) | |
401 | break; | |
402 | brelse(mp->mp_bh[i]); | |
64bc06bb | 403 | mp->mp_bh[i] = NULL; |
9b8c81d1 | 404 | } |
11707ea0 SW |
405 | } |
406 | ||
30cbf189 SW |
407 | /** |
408 | * gfs2_extent_length - Returns length of an extent of blocks | |
409 | * @start: Start of the buffer | |
410 | * @len: Length of the buffer in bytes | |
411 | * @ptr: Current position in the buffer | |
412 | * @limit: Max extent length to return (0 = unlimited) | |
413 | * @eob: Set to 1 if we hit "end of block" | |
414 | * | |
415 | * If the first block is zero (unallocated) it will return the number of | |
416 | * unallocated blocks in the extent, otherwise it will return the number | |
417 | * of contiguous blocks in the extent. | |
418 | * | |
419 | * Returns: The length of the extent (minimum of one block) | |
420 | */ | |
421 | ||
b650738c | 422 | static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob) |
30cbf189 SW |
423 | { |
424 | const __be64 *end = (start + len); | |
425 | const __be64 *first = ptr; | |
426 | u64 d = be64_to_cpu(*ptr); | |
427 | ||
428 | *eob = 0; | |
429 | do { | |
430 | ptr++; | |
431 | if (ptr >= end) | |
432 | break; | |
433 | if (limit && --limit == 0) | |
434 | break; | |
435 | if (d) | |
436 | d++; | |
437 | } while(be64_to_cpu(*ptr) == d); | |
438 | if (ptr >= end) | |
439 | *eob = 1; | |
440 | return (ptr - first); | |
441 | } | |
442 | ||
7841b9f0 AG |
443 | typedef const __be64 *(*gfs2_metadata_walker)( |
444 | struct metapath *mp, | |
445 | const __be64 *start, const __be64 *end, | |
446 | u64 factor, void *data); | |
447 | ||
448 | #define WALK_STOP ((__be64 *)0) | |
449 | #define WALK_NEXT ((__be64 *)1) | |
450 | ||
451 | static int gfs2_walk_metadata(struct inode *inode, sector_t lblock, | |
452 | u64 len, struct metapath *mp, gfs2_metadata_walker walker, | |
453 | void *data) | |
454 | { | |
455 | struct metapath clone; | |
456 | struct gfs2_inode *ip = GFS2_I(inode); | |
457 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
458 | const __be64 *start, *end, *ptr; | |
459 | u64 factor = 1; | |
460 | unsigned int hgt; | |
461 | int ret = 0; | |
462 | ||
463 | for (hgt = ip->i_height - 1; hgt >= mp->mp_aheight; hgt--) | |
464 | factor *= sdp->sd_inptrs; | |
465 | ||
466 | for (;;) { | |
467 | u64 step; | |
468 | ||
469 | /* Walk indirect block. */ | |
470 | start = metapointer(hgt, mp); | |
471 | end = metaend(hgt, mp); | |
472 | ||
473 | step = (end - start) * factor; | |
474 | if (step > len) | |
475 | end = start + DIV_ROUND_UP_ULL(len, factor); | |
476 | ||
477 | ptr = walker(mp, start, end, factor, data); | |
478 | if (ptr == WALK_STOP) | |
479 | break; | |
480 | if (step >= len) | |
481 | break; | |
482 | len -= step; | |
483 | if (ptr != WALK_NEXT) { | |
484 | BUG_ON(!*ptr); | |
485 | mp->mp_list[hgt] += ptr - start; | |
486 | goto fill_up_metapath; | |
487 | } | |
488 | ||
489 | lower_metapath: | |
490 | /* Decrease height of metapath. */ | |
491 | if (mp != &clone) { | |
492 | clone_metapath(&clone, mp); | |
493 | mp = &clone; | |
494 | } | |
495 | brelse(mp->mp_bh[hgt]); | |
496 | mp->mp_bh[hgt] = NULL; | |
497 | if (!hgt) | |
498 | break; | |
499 | hgt--; | |
500 | factor *= sdp->sd_inptrs; | |
501 | ||
502 | /* Advance in metadata tree. */ | |
503 | (mp->mp_list[hgt])++; | |
504 | start = metapointer(hgt, mp); | |
505 | end = metaend(hgt, mp); | |
506 | if (start >= end) { | |
507 | mp->mp_list[hgt] = 0; | |
508 | if (!hgt) | |
509 | break; | |
510 | goto lower_metapath; | |
511 | } | |
512 | ||
513 | fill_up_metapath: | |
514 | /* Increase height of metapath. */ | |
515 | if (mp != &clone) { | |
516 | clone_metapath(&clone, mp); | |
517 | mp = &clone; | |
518 | } | |
519 | ret = fillup_metapath(ip, mp, ip->i_height - 1); | |
520 | if (ret < 0) | |
521 | break; | |
522 | hgt += ret; | |
523 | for (; ret; ret--) | |
524 | do_div(factor, sdp->sd_inptrs); | |
525 | mp->mp_aheight = hgt + 1; | |
526 | } | |
527 | if (mp == &clone) | |
528 | release_metapath(mp); | |
529 | return ret; | |
530 | } | |
531 | ||
532 | struct gfs2_hole_walker_args { | |
533 | u64 blocks; | |
534 | }; | |
535 | ||
536 | static const __be64 *gfs2_hole_walker(struct metapath *mp, | |
537 | const __be64 *start, const __be64 *end, | |
538 | u64 factor, void *data) | |
539 | { | |
540 | struct gfs2_hole_walker_args *args = data; | |
541 | const __be64 *ptr; | |
542 | ||
543 | for (ptr = start; ptr < end; ptr++) { | |
544 | if (*ptr) { | |
545 | args->blocks += (ptr - start) * factor; | |
546 | if (mp->mp_aheight == mp->mp_fheight) | |
547 | return WALK_STOP; | |
548 | return ptr; /* increase height */ | |
549 | } | |
550 | } | |
551 | args->blocks += (end - start) * factor; | |
552 | return WALK_NEXT; | |
553 | } | |
554 | ||
555 | /** | |
556 | * gfs2_hole_size - figure out the size of a hole | |
557 | * @inode: The inode | |
558 | * @lblock: The logical starting block number | |
559 | * @len: How far to look (in blocks) | |
560 | * @mp: The metapath at lblock | |
561 | * @iomap: The iomap to store the hole size in | |
562 | * | |
563 | * This function modifies @mp. | |
564 | * | |
565 | * Returns: errno on error | |
566 | */ | |
567 | static int gfs2_hole_size(struct inode *inode, sector_t lblock, u64 len, | |
568 | struct metapath *mp, struct iomap *iomap) | |
569 | { | |
570 | struct gfs2_hole_walker_args args = { }; | |
571 | int ret = 0; | |
572 | ||
573 | ret = gfs2_walk_metadata(inode, lblock, len, mp, gfs2_hole_walker, &args); | |
574 | if (!ret) | |
575 | iomap->length = args.blocks << inode->i_blkbits; | |
576 | return ret; | |
577 | } | |
578 | ||
9b8c81d1 SW |
579 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
580 | struct gfs2_glock *gl, unsigned int i, | |
581 | unsigned offset, u64 bn) | |
582 | { | |
583 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
584 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
585 | sizeof(struct gfs2_dinode))); | |
586 | BUG_ON(i < 1); | |
587 | BUG_ON(mp->mp_bh[i] != NULL); | |
588 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
350a9b0a | 589 | gfs2_trans_add_meta(gl, mp->mp_bh[i]); |
9b8c81d1 SW |
590 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
591 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
592 | ptr += offset; | |
593 | *ptr = cpu_to_be64(bn); | |
594 | return ptr; | |
595 | } | |
596 | ||
597 | enum alloc_state { | |
598 | ALLOC_DATA = 0, | |
599 | ALLOC_GROW_DEPTH = 1, | |
600 | ALLOC_GROW_HEIGHT = 2, | |
601 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
602 | }; | |
603 | ||
604 | /** | |
628e366d | 605 | * gfs2_iomap_alloc - Build a metadata tree of the requested height |
9b8c81d1 | 606 | * @inode: The GFS2 inode |
628e366d AG |
607 | * @iomap: The iomap structure |
608 | * @flags: iomap flags | |
5f8bd444 | 609 | * @mp: The metapath, with proper height information calculated |
9b8c81d1 SW |
610 | * |
611 | * In this routine we may have to alloc: | |
612 | * i) Indirect blocks to grow the metadata tree height | |
613 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
614 | * iii) Data blocks | |
615 | * | |
64bc06bb AG |
616 | * This function is called after gfs2_iomap_get, which works out the |
617 | * total number of blocks which we need via gfs2_alloc_size. | |
618 | * | |
619 | * We then do the actual allocation asking for an extent at a time (if | |
620 | * enough contiguous free blocks are available, there will only be one | |
621 | * allocation request per call) and uses the state machine to initialise | |
622 | * the blocks in order. | |
9b8c81d1 | 623 | * |
628e366d AG |
624 | * Right now, this function will allocate at most one indirect block |
625 | * worth of data -- with a default block size of 4K, that's slightly | |
626 | * less than 2M. If this limitation is ever removed to allow huge | |
627 | * allocations, we would probably still want to limit the iomap size we | |
628 | * return to avoid stalling other tasks during huge writes; the next | |
629 | * iomap iteration would then find the blocks already allocated. | |
630 | * | |
9b8c81d1 SW |
631 | * Returns: errno on error |
632 | */ | |
633 | ||
3974320c BP |
634 | static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap, |
635 | unsigned flags, struct metapath *mp) | |
9b8c81d1 SW |
636 | { |
637 | struct gfs2_inode *ip = GFS2_I(inode); | |
638 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
639 | struct buffer_head *dibh = mp->mp_bh[0]; | |
5f8bd444 | 640 | u64 bn; |
5af4e7a0 | 641 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
64bc06bb | 642 | size_t dblks = iomap->length >> inode->i_blkbits; |
5f8bd444 | 643 | const unsigned end_of_metadata = mp->mp_fheight - 1; |
628e366d | 644 | int ret; |
9b8c81d1 SW |
645 | enum alloc_state state; |
646 | __be64 *ptr; | |
647 | __be64 zero_bn = 0; | |
648 | ||
5f8bd444 | 649 | BUG_ON(mp->mp_aheight < 1); |
9b8c81d1 | 650 | BUG_ON(dibh == NULL); |
64bc06bb | 651 | BUG_ON(dblks < 1); |
9b8c81d1 | 652 | |
350a9b0a | 653 | gfs2_trans_add_meta(ip->i_gl, dibh); |
9b8c81d1 | 654 | |
628e366d AG |
655 | down_write(&ip->i_rw_mutex); |
656 | ||
5f8bd444 | 657 | if (mp->mp_fheight == mp->mp_aheight) { |
64bc06bb | 658 | /* Bottom indirect block exists */ |
9b8c81d1 SW |
659 | state = ALLOC_DATA; |
660 | } else { | |
661 | /* Need to allocate indirect blocks */ | |
5f8bd444 | 662 | if (mp->mp_fheight == ip->i_height) { |
9b8c81d1 | 663 | /* Writing into existing tree, extend tree down */ |
5f8bd444 | 664 | iblks = mp->mp_fheight - mp->mp_aheight; |
9b8c81d1 SW |
665 | state = ALLOC_GROW_DEPTH; |
666 | } else { | |
667 | /* Building up tree height */ | |
668 | state = ALLOC_GROW_HEIGHT; | |
5f8bd444 | 669 | iblks = mp->mp_fheight - ip->i_height; |
5af4e7a0 | 670 | branch_start = metapath_branch_start(mp); |
5f8bd444 | 671 | iblks += (mp->mp_fheight - branch_start); |
9b8c81d1 SW |
672 | } |
673 | } | |
674 | ||
675 | /* start of the second part of the function (state machine) */ | |
676 | ||
3974320c | 677 | blks = dblks + iblks; |
5f8bd444 | 678 | i = mp->mp_aheight; |
9b8c81d1 SW |
679 | do { |
680 | n = blks - alloced; | |
628e366d AG |
681 | ret = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
682 | if (ret) | |
683 | goto out; | |
9b8c81d1 SW |
684 | alloced += n; |
685 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
686 | gfs2_trans_add_unrevoke(sdp, bn, n); | |
687 | switch (state) { | |
688 | /* Growing height of tree */ | |
689 | case ALLOC_GROW_HEIGHT: | |
690 | if (i == 1) { | |
691 | ptr = (__be64 *)(dibh->b_data + | |
692 | sizeof(struct gfs2_dinode)); | |
693 | zero_bn = *ptr; | |
694 | } | |
5f8bd444 BP |
695 | for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0; |
696 | i++, n--) | |
9b8c81d1 | 697 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); |
5f8bd444 | 698 | if (i - 1 == mp->mp_fheight - ip->i_height) { |
9b8c81d1 SW |
699 | i--; |
700 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
701 | sizeof(struct gfs2_meta_header), | |
702 | dibh, sizeof(struct gfs2_dinode)); | |
703 | gfs2_buffer_clear_tail(dibh, | |
704 | sizeof(struct gfs2_dinode) + | |
705 | sizeof(__be64)); | |
706 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
707 | sizeof(struct gfs2_meta_header)); | |
708 | *ptr = zero_bn; | |
709 | state = ALLOC_GROW_DEPTH; | |
5f8bd444 | 710 | for(i = branch_start; i < mp->mp_fheight; i++) { |
9b8c81d1 SW |
711 | if (mp->mp_bh[i] == NULL) |
712 | break; | |
713 | brelse(mp->mp_bh[i]); | |
714 | mp->mp_bh[i] = NULL; | |
715 | } | |
5af4e7a0 | 716 | i = branch_start; |
9b8c81d1 SW |
717 | } |
718 | if (n == 0) | |
719 | break; | |
720 | /* Branching from existing tree */ | |
721 | case ALLOC_GROW_DEPTH: | |
5f8bd444 | 722 | if (i > 1 && i < mp->mp_fheight) |
350a9b0a | 723 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]); |
5f8bd444 | 724 | for (; i < mp->mp_fheight && n > 0; i++, n--) |
9b8c81d1 SW |
725 | gfs2_indirect_init(mp, ip->i_gl, i, |
726 | mp->mp_list[i-1], bn++); | |
5f8bd444 | 727 | if (i == mp->mp_fheight) |
9b8c81d1 SW |
728 | state = ALLOC_DATA; |
729 | if (n == 0) | |
730 | break; | |
731 | /* Tree complete, adding data blocks */ | |
732 | case ALLOC_DATA: | |
3974320c | 733 | BUG_ON(n > dblks); |
9b8c81d1 | 734 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); |
350a9b0a | 735 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]); |
3974320c | 736 | dblks = n; |
9b8c81d1 | 737 | ptr = metapointer(end_of_metadata, mp); |
3974320c | 738 | iomap->addr = bn << inode->i_blkbits; |
628e366d | 739 | iomap->flags |= IOMAP_F_MERGED | IOMAP_F_NEW; |
9b8c81d1 SW |
740 | while (n-- > 0) |
741 | *ptr++ = cpu_to_be64(bn++); | |
742 | break; | |
743 | } | |
3974320c | 744 | } while (iomap->addr == IOMAP_NULL_ADDR); |
9b8c81d1 | 745 | |
d505a96a | 746 | iomap->type = IOMAP_MAPPED; |
3974320c | 747 | iomap->length = (u64)dblks << inode->i_blkbits; |
5f8bd444 | 748 | ip->i_height = mp->mp_fheight; |
9b8c81d1 | 749 | gfs2_add_inode_blocks(&ip->i_inode, alloced); |
628e366d AG |
750 | gfs2_dinode_out(ip, dibh->b_data); |
751 | out: | |
752 | up_write(&ip->i_rw_mutex); | |
753 | return ret; | |
9b8c81d1 SW |
754 | } |
755 | ||
7ee66c03 CH |
756 | #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE |
757 | ||
64bc06bb AG |
758 | /** |
759 | * gfs2_alloc_size - Compute the maximum allocation size | |
760 | * @inode: The inode | |
761 | * @mp: The metapath | |
762 | * @size: Requested size in blocks | |
763 | * | |
764 | * Compute the maximum size of the next allocation at @mp. | |
765 | * | |
766 | * Returns: size in blocks | |
767 | */ | |
768 | static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size) | |
769 | { | |
770 | struct gfs2_inode *ip = GFS2_I(inode); | |
771 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
772 | const __be64 *first, *ptr, *end; | |
773 | ||
774 | /* | |
775 | * For writes to stuffed files, this function is called twice via | |
776 | * gfs2_iomap_get, before and after unstuffing. The size we return the | |
777 | * first time needs to be large enough to get the reservation and | |
778 | * allocation sizes right. The size we return the second time must | |
779 | * be exact or else gfs2_iomap_alloc won't do the right thing. | |
780 | */ | |
781 | ||
782 | if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) { | |
783 | unsigned int maxsize = mp->mp_fheight > 1 ? | |
784 | sdp->sd_inptrs : sdp->sd_diptrs; | |
785 | maxsize -= mp->mp_list[mp->mp_fheight - 1]; | |
786 | if (size > maxsize) | |
787 | size = maxsize; | |
788 | return size; | |
789 | } | |
790 | ||
791 | first = metapointer(ip->i_height - 1, mp); | |
792 | end = metaend(ip->i_height - 1, mp); | |
793 | if (end - first > size) | |
794 | end = first + size; | |
795 | for (ptr = first; ptr < end; ptr++) { | |
796 | if (*ptr) | |
797 | break; | |
798 | } | |
799 | return ptr - first; | |
800 | } | |
801 | ||
3974320c | 802 | /** |
628e366d | 803 | * gfs2_iomap_get - Map blocks from an inode to disk blocks |
3974320c BP |
804 | * @inode: The inode |
805 | * @pos: Starting position in bytes | |
806 | * @length: Length to map, in bytes | |
807 | * @flags: iomap flags | |
808 | * @iomap: The iomap structure | |
628e366d | 809 | * @mp: The metapath |
3974320c BP |
810 | * |
811 | * Returns: errno | |
812 | */ | |
628e366d AG |
813 | static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length, |
814 | unsigned flags, struct iomap *iomap, | |
815 | struct metapath *mp) | |
b3b94faa | 816 | { |
feaa7bba SW |
817 | struct gfs2_inode *ip = GFS2_I(inode); |
818 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
d505a96a | 819 | loff_t size = i_size_read(inode); |
9b8c81d1 | 820 | __be64 *ptr; |
3974320c | 821 | sector_t lblock; |
628e366d AG |
822 | sector_t lblock_stop; |
823 | int ret; | |
9b8c81d1 | 824 | int eob; |
628e366d | 825 | u64 len; |
d505a96a | 826 | struct buffer_head *dibh = NULL, *bh; |
9b8c81d1 | 827 | u8 height; |
7276b3b0 | 828 | |
628e366d AG |
829 | if (!length) |
830 | return -EINVAL; | |
b3b94faa | 831 | |
d505a96a AG |
832 | down_read(&ip->i_rw_mutex); |
833 | ||
834 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
835 | if (ret) | |
836 | goto unlock; | |
64bc06bb | 837 | iomap->private = dibh; |
d505a96a | 838 | |
49edd5bf | 839 | if (gfs2_is_stuffed(ip)) { |
d505a96a AG |
840 | if (flags & IOMAP_WRITE) { |
841 | loff_t max_size = gfs2_max_stuffed_size(ip); | |
842 | ||
843 | if (pos + length > max_size) | |
844 | goto unstuff; | |
845 | iomap->length = max_size; | |
846 | } else { | |
847 | if (pos >= size) { | |
848 | if (flags & IOMAP_REPORT) { | |
849 | ret = -ENOENT; | |
850 | goto unlock; | |
851 | } else { | |
852 | /* report a hole */ | |
853 | iomap->offset = pos; | |
854 | iomap->length = length; | |
855 | goto do_alloc; | |
856 | } | |
857 | } | |
858 | iomap->length = size; | |
49edd5bf | 859 | } |
d505a96a AG |
860 | iomap->addr = (ip->i_no_addr << inode->i_blkbits) + |
861 | sizeof(struct gfs2_dinode); | |
862 | iomap->type = IOMAP_INLINE; | |
64bc06bb | 863 | iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode); |
d505a96a | 864 | goto out; |
3974320c | 865 | } |
d505a96a AG |
866 | |
867 | unstuff: | |
3974320c | 868 | lblock = pos >> inode->i_blkbits; |
3974320c | 869 | iomap->offset = lblock << inode->i_blkbits; |
628e366d AG |
870 | lblock_stop = (pos + length - 1) >> inode->i_blkbits; |
871 | len = lblock_stop - lblock + 1; | |
d505a96a | 872 | iomap->length = len << inode->i_blkbits; |
628e366d | 873 | |
d505a96a AG |
874 | get_bh(dibh); |
875 | mp->mp_bh[0] = dibh; | |
b3b94faa | 876 | |
9b8c81d1 | 877 | height = ip->i_height; |
9a38662b | 878 | while ((lblock + 1) * sdp->sd_sb.sb_bsize > sdp->sd_heightsize[height]) |
9b8c81d1 | 879 | height++; |
628e366d | 880 | find_metapath(sdp, lblock, mp, height); |
9b8c81d1 SW |
881 | if (height > ip->i_height || gfs2_is_stuffed(ip)) |
882 | goto do_alloc; | |
3974320c | 883 | |
628e366d | 884 | ret = lookup_metapath(ip, mp); |
e8b43fe0 | 885 | if (ret) |
628e366d | 886 | goto unlock; |
3974320c | 887 | |
628e366d | 888 | if (mp->mp_aheight != ip->i_height) |
9b8c81d1 | 889 | goto do_alloc; |
3974320c | 890 | |
628e366d | 891 | ptr = metapointer(ip->i_height - 1, mp); |
9b8c81d1 SW |
892 | if (*ptr == 0) |
893 | goto do_alloc; | |
3974320c | 894 | |
628e366d AG |
895 | bh = mp->mp_bh[ip->i_height - 1]; |
896 | len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, len, &eob); | |
3974320c | 897 | |
628e366d AG |
898 | iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits; |
899 | iomap->length = len << inode->i_blkbits; | |
900 | iomap->type = IOMAP_MAPPED; | |
901 | iomap->flags = IOMAP_F_MERGED; | |
9b8c81d1 | 902 | if (eob) |
7ee66c03 | 903 | iomap->flags |= IOMAP_F_GFS2_BOUNDARY; |
3974320c | 904 | |
3974320c | 905 | out: |
628e366d AG |
906 | iomap->bdev = inode->i_sb->s_bdev; |
907 | unlock: | |
908 | up_read(&ip->i_rw_mutex); | |
64bc06bb | 909 | if (ret && dibh) |
d505a96a | 910 | brelse(dibh); |
9b8c81d1 | 911 | return ret; |
30cbf189 | 912 | |
9b8c81d1 | 913 | do_alloc: |
628e366d | 914 | iomap->addr = IOMAP_NULL_ADDR; |
628e366d | 915 | iomap->type = IOMAP_HOLE; |
628e366d | 916 | if (flags & IOMAP_REPORT) { |
49edd5bf | 917 | if (pos >= size) |
3974320c | 918 | ret = -ENOENT; |
628e366d AG |
919 | else if (height == ip->i_height) |
920 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
49edd5bf AG |
921 | else |
922 | iomap->length = size - pos; | |
64bc06bb AG |
923 | } else if (flags & IOMAP_WRITE) { |
924 | u64 alloc_size; | |
925 | ||
926 | len = gfs2_alloc_size(inode, mp, len); | |
927 | alloc_size = len << inode->i_blkbits; | |
928 | if (alloc_size < iomap->length) | |
929 | iomap->length = alloc_size; | |
930 | } else { | |
d505a96a AG |
931 | if (pos < size && height == ip->i_height) |
932 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
b3b94faa | 933 | } |
628e366d | 934 | goto out; |
3974320c BP |
935 | } |
936 | ||
64bc06bb AG |
937 | static int gfs2_write_lock(struct inode *inode) |
938 | { | |
939 | struct gfs2_inode *ip = GFS2_I(inode); | |
940 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
941 | int error; | |
942 | ||
943 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); | |
944 | error = gfs2_glock_nq(&ip->i_gh); | |
945 | if (error) | |
946 | goto out_uninit; | |
947 | if (&ip->i_inode == sdp->sd_rindex) { | |
948 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
949 | ||
950 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
951 | GL_NOCACHE, &m_ip->i_gh); | |
952 | if (error) | |
953 | goto out_unlock; | |
954 | } | |
955 | return 0; | |
956 | ||
957 | out_unlock: | |
958 | gfs2_glock_dq(&ip->i_gh); | |
959 | out_uninit: | |
960 | gfs2_holder_uninit(&ip->i_gh); | |
961 | return error; | |
962 | } | |
963 | ||
964 | static void gfs2_write_unlock(struct inode *inode) | |
965 | { | |
966 | struct gfs2_inode *ip = GFS2_I(inode); | |
967 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
968 | ||
969 | if (&ip->i_inode == sdp->sd_rindex) { | |
970 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
971 | ||
972 | gfs2_glock_dq_uninit(&m_ip->i_gh); | |
973 | } | |
974 | gfs2_glock_dq_uninit(&ip->i_gh); | |
975 | } | |
976 | ||
977 | static void gfs2_iomap_journaled_page_done(struct inode *inode, loff_t pos, | |
978 | unsigned copied, struct page *page, | |
979 | struct iomap *iomap) | |
980 | { | |
981 | struct gfs2_inode *ip = GFS2_I(inode); | |
982 | ||
983 | gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied); | |
984 | } | |
985 | ||
986 | static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, | |
987 | loff_t length, unsigned flags, | |
988 | struct iomap *iomap) | |
989 | { | |
990 | struct metapath mp = { .mp_aheight = 1, }; | |
991 | struct gfs2_inode *ip = GFS2_I(inode); | |
992 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
993 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; | |
994 | bool unstuff, alloc_required; | |
995 | int ret; | |
996 | ||
997 | ret = gfs2_write_lock(inode); | |
998 | if (ret) | |
999 | return ret; | |
1000 | ||
1001 | unstuff = gfs2_is_stuffed(ip) && | |
1002 | pos + length > gfs2_max_stuffed_size(ip); | |
1003 | ||
1004 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); | |
1005 | if (ret) | |
1006 | goto out_release; | |
1007 | ||
1008 | alloc_required = unstuff || iomap->type == IOMAP_HOLE; | |
1009 | ||
1010 | if (alloc_required || gfs2_is_jdata(ip)) | |
1011 | gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, | |
1012 | &ind_blocks); | |
1013 | ||
1014 | if (alloc_required) { | |
1015 | struct gfs2_alloc_parms ap = { | |
1016 | .target = data_blocks + ind_blocks | |
1017 | }; | |
1018 | ||
1019 | ret = gfs2_quota_lock_check(ip, &ap); | |
1020 | if (ret) | |
1021 | goto out_release; | |
1022 | ||
1023 | ret = gfs2_inplace_reserve(ip, &ap); | |
1024 | if (ret) | |
1025 | goto out_qunlock; | |
1026 | } | |
1027 | ||
1028 | rblocks = RES_DINODE + ind_blocks; | |
1029 | if (gfs2_is_jdata(ip)) | |
1030 | rblocks += data_blocks; | |
1031 | if (ind_blocks || data_blocks) | |
1032 | rblocks += RES_STATFS + RES_QUOTA; | |
1033 | if (inode == sdp->sd_rindex) | |
1034 | rblocks += 2 * RES_STATFS; | |
1035 | if (alloc_required) | |
1036 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); | |
1037 | ||
1038 | ret = gfs2_trans_begin(sdp, rblocks, iomap->length >> inode->i_blkbits); | |
1039 | if (ret) | |
1040 | goto out_trans_fail; | |
1041 | ||
1042 | if (unstuff) { | |
1043 | ret = gfs2_unstuff_dinode(ip, NULL); | |
1044 | if (ret) | |
1045 | goto out_trans_end; | |
1046 | release_metapath(&mp); | |
1047 | brelse(iomap->private); | |
1048 | iomap->private = NULL; | |
1049 | ret = gfs2_iomap_get(inode, iomap->offset, iomap->length, | |
1050 | flags, iomap, &mp); | |
1051 | if (ret) | |
1052 | goto out_trans_end; | |
1053 | } | |
1054 | ||
1055 | if (iomap->type == IOMAP_HOLE) { | |
1056 | ret = gfs2_iomap_alloc(inode, iomap, flags, &mp); | |
1057 | if (ret) { | |
1058 | gfs2_trans_end(sdp); | |
1059 | gfs2_inplace_release(ip); | |
1060 | punch_hole(ip, iomap->offset, iomap->length); | |
1061 | goto out_qunlock; | |
1062 | } | |
1063 | } | |
1064 | release_metapath(&mp); | |
1065 | if (gfs2_is_jdata(ip)) | |
1066 | iomap->page_done = gfs2_iomap_journaled_page_done; | |
1067 | return 0; | |
1068 | ||
1069 | out_trans_end: | |
1070 | gfs2_trans_end(sdp); | |
1071 | out_trans_fail: | |
1072 | if (alloc_required) | |
1073 | gfs2_inplace_release(ip); | |
1074 | out_qunlock: | |
1075 | if (alloc_required) | |
1076 | gfs2_quota_unlock(ip); | |
1077 | out_release: | |
1078 | if (iomap->private) | |
1079 | brelse(iomap->private); | |
1080 | release_metapath(&mp); | |
1081 | gfs2_write_unlock(inode); | |
1082 | return ret; | |
1083 | } | |
1084 | ||
628e366d AG |
1085 | static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, |
1086 | unsigned flags, struct iomap *iomap) | |
1087 | { | |
1088 | struct gfs2_inode *ip = GFS2_I(inode); | |
1089 | struct metapath mp = { .mp_aheight = 1, }; | |
1090 | int ret; | |
1091 | ||
1092 | trace_gfs2_iomap_start(ip, pos, length, flags); | |
1093 | if (flags & IOMAP_WRITE) { | |
64bc06bb | 1094 | ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap); |
628e366d AG |
1095 | } else { |
1096 | ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); | |
1097 | release_metapath(&mp); | |
1098 | } | |
1099 | trace_gfs2_iomap_end(ip, iomap, ret); | |
1100 | return ret; | |
1101 | } | |
1102 | ||
64bc06bb AG |
1103 | static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length, |
1104 | ssize_t written, unsigned flags, struct iomap *iomap) | |
1105 | { | |
1106 | struct gfs2_inode *ip = GFS2_I(inode); | |
1107 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1108 | struct gfs2_trans *tr = current->journal_info; | |
1109 | struct buffer_head *dibh = iomap->private; | |
1110 | ||
1111 | if (!(flags & IOMAP_WRITE)) | |
1112 | goto out; | |
1113 | ||
1114 | if (iomap->type != IOMAP_INLINE) { | |
1115 | gfs2_ordered_add_inode(ip); | |
1116 | ||
1117 | if (tr->tr_num_buf_new) | |
1118 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1119 | else | |
1120 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
1121 | } | |
1122 | ||
1123 | if (inode == sdp->sd_rindex) { | |
1124 | adjust_fs_space(inode); | |
1125 | sdp->sd_rindex_uptodate = 0; | |
1126 | } | |
1127 | ||
1128 | gfs2_trans_end(sdp); | |
1129 | gfs2_inplace_release(ip); | |
1130 | ||
1131 | if (length != written && (iomap->flags & IOMAP_F_NEW)) { | |
1132 | /* Deallocate blocks that were just allocated. */ | |
1133 | loff_t blockmask = i_blocksize(inode) - 1; | |
1134 | loff_t end = (pos + length) & ~blockmask; | |
1135 | ||
1136 | pos = (pos + written + blockmask) & ~blockmask; | |
1137 | if (pos < end) { | |
1138 | truncate_pagecache_range(inode, pos, end - 1); | |
1139 | punch_hole(ip, pos, end - pos); | |
1140 | } | |
1141 | } | |
1142 | ||
1143 | if (ip->i_qadata && ip->i_qadata->qa_qd_num) | |
1144 | gfs2_quota_unlock(ip); | |
1145 | gfs2_write_unlock(inode); | |
1146 | ||
1147 | out: | |
1148 | if (dibh) | |
1149 | brelse(dibh); | |
1150 | return 0; | |
1151 | } | |
1152 | ||
628e366d AG |
1153 | const struct iomap_ops gfs2_iomap_ops = { |
1154 | .iomap_begin = gfs2_iomap_begin, | |
64bc06bb | 1155 | .iomap_end = gfs2_iomap_end, |
628e366d AG |
1156 | }; |
1157 | ||
3974320c | 1158 | /** |
d39d18e0 | 1159 | * gfs2_block_map - Map one or more blocks of an inode to a disk block |
3974320c BP |
1160 | * @inode: The inode |
1161 | * @lblock: The logical block number | |
1162 | * @bh_map: The bh to be mapped | |
1163 | * @create: True if its ok to alloc blocks to satify the request | |
1164 | * | |
d39d18e0 AG |
1165 | * The size of the requested mapping is defined in bh_map->b_size. |
1166 | * | |
1167 | * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged | |
1168 | * when @lblock is not mapped. Sets buffer_mapped(bh_map) and | |
1169 | * bh_map->b_size to indicate the size of the mapping when @lblock and | |
1170 | * successive blocks are mapped, up to the requested size. | |
1171 | * | |
1172 | * Sets buffer_boundary() if a read of metadata will be required | |
1173 | * before the next block can be mapped. Sets buffer_new() if new | |
1174 | * blocks were allocated. | |
3974320c BP |
1175 | * |
1176 | * Returns: errno | |
1177 | */ | |
1178 | ||
1179 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
1180 | struct buffer_head *bh_map, int create) | |
1181 | { | |
1182 | struct gfs2_inode *ip = GFS2_I(inode); | |
628e366d AG |
1183 | loff_t pos = (loff_t)lblock << inode->i_blkbits; |
1184 | loff_t length = bh_map->b_size; | |
1185 | struct metapath mp = { .mp_aheight = 1, }; | |
1186 | struct iomap iomap = { }; | |
1187 | int ret; | |
3974320c BP |
1188 | |
1189 | clear_buffer_mapped(bh_map); | |
1190 | clear_buffer_new(bh_map); | |
1191 | clear_buffer_boundary(bh_map); | |
1192 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
1193 | ||
628e366d AG |
1194 | if (create) { |
1195 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, &iomap, &mp); | |
1196 | if (!ret && iomap.type == IOMAP_HOLE) | |
1197 | ret = gfs2_iomap_alloc(inode, &iomap, IOMAP_WRITE, &mp); | |
1198 | release_metapath(&mp); | |
1199 | } else { | |
1200 | ret = gfs2_iomap_get(inode, pos, length, 0, &iomap, &mp); | |
1201 | release_metapath(&mp); | |
3974320c | 1202 | } |
628e366d AG |
1203 | if (ret) |
1204 | goto out; | |
3974320c BP |
1205 | |
1206 | if (iomap.length > bh_map->b_size) { | |
1207 | iomap.length = bh_map->b_size; | |
7ee66c03 | 1208 | iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY; |
5f8bd444 | 1209 | } |
3974320c BP |
1210 | if (iomap.addr != IOMAP_NULL_ADDR) |
1211 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
1212 | bh_map->b_size = iomap.length; | |
7ee66c03 | 1213 | if (iomap.flags & IOMAP_F_GFS2_BOUNDARY) |
3974320c BP |
1214 | set_buffer_boundary(bh_map); |
1215 | if (iomap.flags & IOMAP_F_NEW) | |
1216 | set_buffer_new(bh_map); | |
1217 | ||
1218 | out: | |
1219 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
1220 | return ret; | |
fd88de56 SW |
1221 | } |
1222 | ||
941e6d7d SW |
1223 | /* |
1224 | * Deprecated: do not use in new code | |
1225 | */ | |
fd88de56 SW |
1226 | int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen) |
1227 | { | |
23591256 | 1228 | struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 }; |
7a6bbacb | 1229 | int ret; |
fd88de56 SW |
1230 | int create = *new; |
1231 | ||
1232 | BUG_ON(!extlen); | |
1233 | BUG_ON(!dblock); | |
1234 | BUG_ON(!new); | |
1235 | ||
47a9a527 | 1236 | bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5)); |
e9e1ef2b | 1237 | ret = gfs2_block_map(inode, lblock, &bh, create); |
7a6bbacb SW |
1238 | *extlen = bh.b_size >> inode->i_blkbits; |
1239 | *dblock = bh.b_blocknr; | |
1240 | if (buffer_new(&bh)) | |
1241 | *new = 1; | |
1242 | else | |
1243 | *new = 0; | |
1244 | return ret; | |
b3b94faa DT |
1245 | } |
1246 | ||
ba7f7290 | 1247 | /** |
bdba0d5e | 1248 | * gfs2_block_zero_range - Deal with zeroing out data |
ba7f7290 SW |
1249 | * |
1250 | * This is partly borrowed from ext3. | |
1251 | */ | |
bdba0d5e AG |
1252 | static int gfs2_block_zero_range(struct inode *inode, loff_t from, |
1253 | unsigned int length) | |
ba7f7290 | 1254 | { |
bdba0d5e | 1255 | struct address_space *mapping = inode->i_mapping; |
ba7f7290 | 1256 | struct gfs2_inode *ip = GFS2_I(inode); |
09cbfeaf KS |
1257 | unsigned long index = from >> PAGE_SHIFT; |
1258 | unsigned offset = from & (PAGE_SIZE-1); | |
bdba0d5e | 1259 | unsigned blocksize, iblock, pos; |
ba7f7290 SW |
1260 | struct buffer_head *bh; |
1261 | struct page *page; | |
ba7f7290 SW |
1262 | int err; |
1263 | ||
220cca2a | 1264 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ba7f7290 SW |
1265 | if (!page) |
1266 | return 0; | |
1267 | ||
1268 | blocksize = inode->i_sb->s_blocksize; | |
09cbfeaf | 1269 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
ba7f7290 SW |
1270 | |
1271 | if (!page_has_buffers(page)) | |
1272 | create_empty_buffers(page, blocksize, 0); | |
1273 | ||
1274 | /* Find the buffer that contains "offset" */ | |
1275 | bh = page_buffers(page); | |
1276 | pos = blocksize; | |
1277 | while (offset >= pos) { | |
1278 | bh = bh->b_this_page; | |
1279 | iblock++; | |
1280 | pos += blocksize; | |
1281 | } | |
1282 | ||
1283 | err = 0; | |
1284 | ||
1285 | if (!buffer_mapped(bh)) { | |
e9e1ef2b | 1286 | gfs2_block_map(inode, iblock, bh, 0); |
ba7f7290 SW |
1287 | /* unmapped? It's a hole - nothing to do */ |
1288 | if (!buffer_mapped(bh)) | |
1289 | goto unlock; | |
1290 | } | |
1291 | ||
1292 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1293 | if (PageUptodate(page)) | |
1294 | set_buffer_uptodate(bh); | |
1295 | ||
1296 | if (!buffer_uptodate(bh)) { | |
1297 | err = -EIO; | |
dfec8a14 | 1298 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
ba7f7290 SW |
1299 | wait_on_buffer(bh); |
1300 | /* Uhhuh. Read error. Complain and punt. */ | |
1301 | if (!buffer_uptodate(bh)) | |
1302 | goto unlock; | |
1875f2f3 | 1303 | err = 0; |
ba7f7290 SW |
1304 | } |
1305 | ||
845802b1 | 1306 | if (gfs2_is_jdata(ip)) |
350a9b0a | 1307 | gfs2_trans_add_data(ip->i_gl, bh); |
845802b1 AG |
1308 | else |
1309 | gfs2_ordered_add_inode(ip); | |
ba7f7290 | 1310 | |
eebd2aa3 | 1311 | zero_user(page, offset, length); |
40bc9a27 | 1312 | mark_buffer_dirty(bh); |
ba7f7290 SW |
1313 | unlock: |
1314 | unlock_page(page); | |
09cbfeaf | 1315 | put_page(page); |
ba7f7290 SW |
1316 | return err; |
1317 | } | |
1318 | ||
c62baf65 FF |
1319 | #define GFS2_JTRUNC_REVOKES 8192 |
1320 | ||
fa731fc4 SW |
1321 | /** |
1322 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
1323 | * @inode: The inode being truncated | |
1324 | * @oldsize: The original (larger) size | |
1325 | * @newsize: The new smaller size | |
1326 | * | |
1327 | * With jdata files, we have to journal a revoke for each block which is | |
1328 | * truncated. As a result, we need to split this into separate transactions | |
1329 | * if the number of pages being truncated gets too large. | |
1330 | */ | |
1331 | ||
fa731fc4 SW |
1332 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
1333 | { | |
1334 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1335 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1336 | u64 chunk; | |
1337 | int error; | |
1338 | ||
1339 | while (oldsize != newsize) { | |
e7fdf004 AG |
1340 | struct gfs2_trans *tr; |
1341 | unsigned int offs; | |
1342 | ||
fa731fc4 SW |
1343 | chunk = oldsize - newsize; |
1344 | if (chunk > max_chunk) | |
1345 | chunk = max_chunk; | |
e7fdf004 AG |
1346 | |
1347 | offs = oldsize & ~PAGE_MASK; | |
1348 | if (offs && chunk > PAGE_SIZE) | |
1349 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1350 | ||
7caef267 | 1351 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1352 | oldsize -= chunk; |
e7fdf004 AG |
1353 | |
1354 | tr = current->journal_info; | |
1355 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1356 | continue; | |
1357 | ||
fa731fc4 SW |
1358 | gfs2_trans_end(sdp); |
1359 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1360 | if (error) | |
1361 | return error; | |
1362 | } | |
1363 | ||
1364 | return 0; | |
1365 | } | |
1366 | ||
8b5860a3 | 1367 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1368 | { |
ff8f33c8 SW |
1369 | struct gfs2_inode *ip = GFS2_I(inode); |
1370 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
80990f40 | 1371 | struct buffer_head *dibh = NULL; |
b3b94faa | 1372 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1373 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1374 | int error; |
1375 | ||
fa731fc4 SW |
1376 | if (journaled) |
1377 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1378 | else | |
1379 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1380 | if (error) |
1381 | return error; | |
1382 | ||
1383 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1384 | if (error) | |
1385 | goto out; | |
1386 | ||
350a9b0a | 1387 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1388 | |
b3b94faa | 1389 | if (gfs2_is_stuffed(ip)) { |
ff8f33c8 | 1390 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
b3b94faa | 1391 | } else { |
bdba0d5e AG |
1392 | unsigned int blocksize = i_blocksize(inode); |
1393 | unsigned int offs = newsize & (blocksize - 1); | |
1394 | if (offs) { | |
1395 | error = gfs2_block_zero_range(inode, newsize, | |
1396 | blocksize - offs); | |
ff8f33c8 | 1397 | if (error) |
80990f40 | 1398 | goto out; |
b3b94faa | 1399 | } |
ff8f33c8 | 1400 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa DT |
1401 | } |
1402 | ||
ff8f33c8 | 1403 | i_size_write(inode, newsize); |
078cd827 | 1404 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
ff8f33c8 | 1405 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1406 | |
fa731fc4 SW |
1407 | if (journaled) |
1408 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1409 | else | |
7caef267 | 1410 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1411 | |
a91ea69f | 1412 | out: |
80990f40 AG |
1413 | brelse(dibh); |
1414 | if (current->journal_info) | |
1415 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1416 | return error; |
1417 | } | |
1418 | ||
628e366d AG |
1419 | int gfs2_iomap_get_alloc(struct inode *inode, loff_t pos, loff_t length, |
1420 | struct iomap *iomap) | |
1421 | { | |
1422 | struct metapath mp = { .mp_aheight = 1, }; | |
1423 | int ret; | |
1424 | ||
1425 | ret = gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp); | |
1426 | if (!ret && iomap->type == IOMAP_HOLE) | |
1427 | ret = gfs2_iomap_alloc(inode, iomap, IOMAP_WRITE, &mp); | |
1428 | release_metapath(&mp); | |
1429 | return ret; | |
1430 | } | |
1431 | ||
d552a2b9 BP |
1432 | /** |
1433 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1434 | * @ip: inode | |
1435 | * @rg_gh: holder of resource group glock | |
5cf26b1e AG |
1436 | * @bh: buffer head to sweep |
1437 | * @start: starting point in bh | |
1438 | * @end: end point in bh | |
1439 | * @meta: true if bh points to metadata (rather than data) | |
d552a2b9 | 1440 | * @btotal: place to keep count of total blocks freed |
d552a2b9 BP |
1441 | * |
1442 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1443 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1444 | * block has references to multiple rgrps, we break it into individual | |
1445 | * transactions. This allows other processes to use the rgrps while we're | |
1446 | * focused on a single one, for better concurrency / performance. | |
1447 | * At every transaction boundary, we rewrite the inode into the journal. | |
1448 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1449 | * if we're interrupted by power-outages. | |
1450 | * | |
1451 | * Returns: 0, or return code if an error occurred. | |
1452 | * *btotal has the total number of blocks freed | |
1453 | */ | |
1454 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
5cf26b1e AG |
1455 | struct buffer_head *bh, __be64 *start, __be64 *end, |
1456 | bool meta, u32 *btotal) | |
b3b94faa | 1457 | { |
9b8c81d1 | 1458 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1459 | struct gfs2_rgrpd *rgd; |
1460 | struct gfs2_trans *tr; | |
5cf26b1e | 1461 | __be64 *p; |
d552a2b9 BP |
1462 | int blks_outside_rgrp; |
1463 | u64 bn, bstart, isize_blks; | |
1464 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
d552a2b9 BP |
1465 | int ret = 0; |
1466 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1467 | ||
d552a2b9 | 1468 | more_rgrps: |
5cf26b1e AG |
1469 | rgd = NULL; |
1470 | if (gfs2_holder_initialized(rd_gh)) { | |
1471 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); | |
1472 | gfs2_assert_withdraw(sdp, | |
1473 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1474 | } | |
d552a2b9 BP |
1475 | blks_outside_rgrp = 0; |
1476 | bstart = 0; | |
1477 | blen = 0; | |
d552a2b9 | 1478 | |
5cf26b1e | 1479 | for (p = start; p < end; p++) { |
d552a2b9 BP |
1480 | if (!*p) |
1481 | continue; | |
1482 | bn = be64_to_cpu(*p); | |
5cf26b1e AG |
1483 | |
1484 | if (rgd) { | |
1485 | if (!rgrp_contains_block(rgd, bn)) { | |
1486 | blks_outside_rgrp++; | |
1487 | continue; | |
1488 | } | |
d552a2b9 | 1489 | } else { |
90bcab99 | 1490 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
5cf26b1e AG |
1491 | if (unlikely(!rgd)) { |
1492 | ret = -EIO; | |
1493 | goto out; | |
1494 | } | |
d552a2b9 BP |
1495 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1496 | 0, rd_gh); | |
1497 | if (ret) | |
1498 | goto out; | |
1499 | ||
1500 | /* Must be done with the rgrp glock held: */ | |
1501 | if (gfs2_rs_active(&ip->i_res) && | |
1502 | rgd == ip->i_res.rs_rbm.rgd) | |
1503 | gfs2_rs_deltree(&ip->i_res); | |
1504 | } | |
1505 | ||
d552a2b9 BP |
1506 | /* The size of our transactions will be unknown until we |
1507 | actually process all the metadata blocks that relate to | |
1508 | the rgrp. So we estimate. We know it can't be more than | |
1509 | the dinode's i_blocks and we don't want to exceed the | |
1510 | journal flush threshold, sd_log_thresh2. */ | |
1511 | if (current->journal_info == NULL) { | |
1512 | unsigned int jblocks_rqsted, revokes; | |
1513 | ||
1514 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1515 | RES_INDIRECT; | |
1516 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1517 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1518 | jblocks_rqsted += | |
1519 | atomic_read(&sdp->sd_log_thresh2); | |
1520 | else | |
1521 | jblocks_rqsted += isize_blks; | |
1522 | revokes = jblocks_rqsted; | |
1523 | if (meta) | |
5cf26b1e | 1524 | revokes += end - start; |
d552a2b9 BP |
1525 | else if (ip->i_depth) |
1526 | revokes += sdp->sd_inptrs; | |
1527 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1528 | if (ret) | |
1529 | goto out_unlock; | |
1530 | down_write(&ip->i_rw_mutex); | |
1531 | } | |
1532 | /* check if we will exceed the transaction blocks requested */ | |
1533 | tr = current->journal_info; | |
1534 | if (tr->tr_num_buf_new + RES_STATFS + | |
1535 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1536 | /* We set blks_outside_rgrp to ensure the loop will | |
1537 | be repeated for the same rgrp, but with a new | |
1538 | transaction. */ | |
1539 | blks_outside_rgrp++; | |
1540 | /* This next part is tricky. If the buffer was added | |
1541 | to the transaction, we've already set some block | |
1542 | pointers to 0, so we better follow through and free | |
1543 | them, or we will introduce corruption (so break). | |
1544 | This may be impossible, or at least rare, but I | |
1545 | decided to cover the case regardless. | |
1546 | ||
1547 | If the buffer was not added to the transaction | |
1548 | (this call), doing so would exceed our transaction | |
1549 | size, so we need to end the transaction and start a | |
1550 | new one (so goto). */ | |
1551 | ||
1552 | if (buf_in_tr) | |
1553 | break; | |
1554 | goto out_unlock; | |
1555 | } | |
1556 | ||
1557 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1558 | buf_in_tr = true; | |
1559 | *p = 0; | |
1560 | if (bstart + blen == bn) { | |
1561 | blen++; | |
1562 | continue; | |
1563 | } | |
1564 | if (bstart) { | |
1565 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1566 | (*btotal) += blen; | |
1567 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1568 | } | |
1569 | bstart = bn; | |
1570 | blen = 1; | |
1571 | } | |
1572 | if (bstart) { | |
1573 | __gfs2_free_blocks(ip, bstart, (u32)blen, meta); | |
1574 | (*btotal) += blen; | |
1575 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1576 | } | |
1577 | out_unlock: | |
1578 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1579 | outside the rgrp we just processed, | |
1580 | do it all over again. */ | |
1581 | if (current->journal_info) { | |
5cf26b1e AG |
1582 | struct buffer_head *dibh; |
1583 | ||
1584 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1585 | if (ret) | |
1586 | goto out; | |
d552a2b9 BP |
1587 | |
1588 | /* Every transaction boundary, we rewrite the dinode | |
1589 | to keep its di_blocks current in case of failure. */ | |
1590 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = | |
b32c8c76 | 1591 | current_time(&ip->i_inode); |
d552a2b9 BP |
1592 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1593 | gfs2_dinode_out(ip, dibh->b_data); | |
5cf26b1e | 1594 | brelse(dibh); |
d552a2b9 BP |
1595 | up_write(&ip->i_rw_mutex); |
1596 | gfs2_trans_end(sdp); | |
1597 | } | |
1598 | gfs2_glock_dq_uninit(rd_gh); | |
1599 | cond_resched(); | |
1600 | goto more_rgrps; | |
1601 | } | |
1602 | out: | |
1603 | return ret; | |
1604 | } | |
1605 | ||
10d2cf94 AG |
1606 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h) |
1607 | { | |
1608 | if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0]))) | |
1609 | return false; | |
1610 | return true; | |
1611 | } | |
1612 | ||
d552a2b9 BP |
1613 | /** |
1614 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
d552a2b9 BP |
1615 | * @mp: starting metapath |
1616 | * @h: desired height to search | |
1617 | * | |
10d2cf94 | 1618 | * Assumes the metapath is valid (with buffers) out to height h. |
d552a2b9 BP |
1619 | * Returns: true if a non-null pointer was found in the metapath buffer |
1620 | * false if all remaining pointers are NULL in the buffer | |
1621 | */ | |
1622 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
10d2cf94 AG |
1623 | unsigned int h, |
1624 | __u16 *end_list, unsigned int end_aligned) | |
d552a2b9 | 1625 | { |
10d2cf94 AG |
1626 | struct buffer_head *bh = mp->mp_bh[h]; |
1627 | __be64 *first, *ptr, *end; | |
1628 | ||
1629 | first = metaptr1(h, mp); | |
1630 | ptr = first + mp->mp_list[h]; | |
1631 | end = (__be64 *)(bh->b_data + bh->b_size); | |
1632 | if (end_list && mp_eq_to_hgt(mp, end_list, h)) { | |
1633 | bool keep_end = h < end_aligned; | |
1634 | end = first + end_list[h] + keep_end; | |
1635 | } | |
d552a2b9 | 1636 | |
10d2cf94 | 1637 | while (ptr < end) { |
c4a9d189 | 1638 | if (*ptr) { /* if we have a non-null pointer */ |
10d2cf94 | 1639 | mp->mp_list[h] = ptr - first; |
c4a9d189 BP |
1640 | h++; |
1641 | if (h < GFS2_MAX_META_HEIGHT) | |
10d2cf94 | 1642 | mp->mp_list[h] = 0; |
d552a2b9 | 1643 | return true; |
c4a9d189 | 1644 | } |
10d2cf94 | 1645 | ptr++; |
d552a2b9 | 1646 | } |
10d2cf94 | 1647 | return false; |
d552a2b9 BP |
1648 | } |
1649 | ||
1650 | enum dealloc_states { | |
1651 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1652 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1653 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1654 | DEALLOC_DONE = 3, /* process complete */ | |
1655 | }; | |
b3b94faa | 1656 | |
5cf26b1e AG |
1657 | static inline void |
1658 | metapointer_range(struct metapath *mp, int height, | |
1659 | __u16 *start_list, unsigned int start_aligned, | |
10d2cf94 | 1660 | __u16 *end_list, unsigned int end_aligned, |
5cf26b1e AG |
1661 | __be64 **start, __be64 **end) |
1662 | { | |
1663 | struct buffer_head *bh = mp->mp_bh[height]; | |
1664 | __be64 *first; | |
1665 | ||
1666 | first = metaptr1(height, mp); | |
1667 | *start = first; | |
1668 | if (mp_eq_to_hgt(mp, start_list, height)) { | |
1669 | bool keep_start = height < start_aligned; | |
1670 | *start = first + start_list[height] + keep_start; | |
1671 | } | |
1672 | *end = (__be64 *)(bh->b_data + bh->b_size); | |
10d2cf94 AG |
1673 | if (end_list && mp_eq_to_hgt(mp, end_list, height)) { |
1674 | bool keep_end = height < end_aligned; | |
1675 | *end = first + end_list[height] + keep_end; | |
1676 | } | |
1677 | } | |
1678 | ||
1679 | static inline bool walk_done(struct gfs2_sbd *sdp, | |
1680 | struct metapath *mp, int height, | |
1681 | __u16 *end_list, unsigned int end_aligned) | |
1682 | { | |
1683 | __u16 end; | |
1684 | ||
1685 | if (end_list) { | |
1686 | bool keep_end = height < end_aligned; | |
1687 | if (!mp_eq_to_hgt(mp, end_list, height)) | |
1688 | return false; | |
1689 | end = end_list[height] + keep_end; | |
1690 | } else | |
1691 | end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
1692 | return mp->mp_list[height] >= end; | |
5cf26b1e AG |
1693 | } |
1694 | ||
d552a2b9 | 1695 | /** |
10d2cf94 | 1696 | * punch_hole - deallocate blocks in a file |
d552a2b9 | 1697 | * @ip: inode to truncate |
10d2cf94 AG |
1698 | * @offset: the start of the hole |
1699 | * @length: the size of the hole (or 0 for truncate) | |
1700 | * | |
1701 | * Punch a hole into a file or truncate a file at a given position. This | |
1702 | * function operates in whole blocks (@offset and @length are rounded | |
1703 | * accordingly); partially filled blocks must be cleared otherwise. | |
d552a2b9 | 1704 | * |
10d2cf94 AG |
1705 | * This function works from the bottom up, and from the right to the left. In |
1706 | * other words, it strips off the highest layer (data) before stripping any of | |
1707 | * the metadata. Doing it this way is best in case the operation is interrupted | |
1708 | * by power failure, etc. The dinode is rewritten in every transaction to | |
1709 | * guarantee integrity. | |
d552a2b9 | 1710 | */ |
10d2cf94 | 1711 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length) |
d552a2b9 BP |
1712 | { |
1713 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
bb491ce6 | 1714 | u64 maxsize = sdp->sd_heightsize[ip->i_height]; |
10d2cf94 | 1715 | struct metapath mp = {}; |
d552a2b9 BP |
1716 | struct buffer_head *dibh, *bh; |
1717 | struct gfs2_holder rd_gh; | |
cb7f0903 | 1718 | unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; |
10d2cf94 AG |
1719 | u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift; |
1720 | __u16 start_list[GFS2_MAX_META_HEIGHT]; | |
1721 | __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL; | |
4e56a641 | 1722 | unsigned int start_aligned, uninitialized_var(end_aligned); |
d552a2b9 BP |
1723 | unsigned int strip_h = ip->i_height - 1; |
1724 | u32 btotal = 0; | |
1725 | int ret, state; | |
1726 | int mp_h; /* metapath buffers are read in to this height */ | |
d552a2b9 | 1727 | u64 prev_bnr = 0; |
5cf26b1e | 1728 | __be64 *start, *end; |
b3b94faa | 1729 | |
bb491ce6 AG |
1730 | if (offset >= maxsize) { |
1731 | /* | |
1732 | * The starting point lies beyond the allocated meta-data; | |
1733 | * there are no blocks do deallocate. | |
1734 | */ | |
1735 | return 0; | |
1736 | } | |
1737 | ||
10d2cf94 AG |
1738 | /* |
1739 | * The start position of the hole is defined by lblock, start_list, and | |
1740 | * start_aligned. The end position of the hole is defined by lend, | |
1741 | * end_list, and end_aligned. | |
1742 | * | |
1743 | * start_aligned and end_aligned define down to which height the start | |
1744 | * and end positions are aligned to the metadata tree (i.e., the | |
1745 | * position is a multiple of the metadata granularity at the height | |
1746 | * above). This determines at which heights additional meta pointers | |
1747 | * needs to be preserved for the remaining data. | |
1748 | */ | |
b3b94faa | 1749 | |
10d2cf94 | 1750 | if (length) { |
10d2cf94 AG |
1751 | u64 end_offset = offset + length; |
1752 | u64 lend; | |
1753 | ||
1754 | /* | |
1755 | * Clip the end at the maximum file size for the given height: | |
1756 | * that's how far the metadata goes; files bigger than that | |
1757 | * will have additional layers of indirection. | |
1758 | */ | |
1759 | if (end_offset > maxsize) | |
1760 | end_offset = maxsize; | |
1761 | lend = end_offset >> bsize_shift; | |
1762 | ||
1763 | if (lblock >= lend) | |
1764 | return 0; | |
1765 | ||
1766 | find_metapath(sdp, lend, &mp, ip->i_height); | |
1767 | end_list = __end_list; | |
1768 | memcpy(end_list, mp.mp_list, sizeof(mp.mp_list)); | |
1769 | ||
1770 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { | |
1771 | if (end_list[mp_h]) | |
1772 | break; | |
1773 | } | |
1774 | end_aligned = mp_h; | |
1775 | } | |
1776 | ||
1777 | find_metapath(sdp, lblock, &mp, ip->i_height); | |
cb7f0903 AG |
1778 | memcpy(start_list, mp.mp_list, sizeof(start_list)); |
1779 | ||
cb7f0903 AG |
1780 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { |
1781 | if (start_list[mp_h]) | |
1782 | break; | |
1783 | } | |
1784 | start_aligned = mp_h; | |
d552a2b9 BP |
1785 | |
1786 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1787 | if (ret) | |
1788 | return ret; | |
b3b94faa | 1789 | |
d552a2b9 BP |
1790 | mp.mp_bh[0] = dibh; |
1791 | ret = lookup_metapath(ip, &mp); | |
e8b43fe0 AG |
1792 | if (ret) |
1793 | goto out_metapath; | |
c3ce5aa9 AG |
1794 | |
1795 | /* issue read-ahead on metadata */ | |
5cf26b1e AG |
1796 | for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) { |
1797 | metapointer_range(&mp, mp_h, start_list, start_aligned, | |
10d2cf94 | 1798 | end_list, end_aligned, &start, &end); |
5cf26b1e AG |
1799 | gfs2_metapath_ra(ip->i_gl, start, end); |
1800 | } | |
c3ce5aa9 | 1801 | |
e8b43fe0 | 1802 | if (mp.mp_aheight == ip->i_height) |
d552a2b9 BP |
1803 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ |
1804 | else | |
1805 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1806 | |
d552a2b9 BP |
1807 | ret = gfs2_rindex_update(sdp); |
1808 | if (ret) | |
1809 | goto out_metapath; | |
1810 | ||
1811 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1812 | if (ret) | |
1813 | goto out_metapath; | |
1814 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1815 | ||
1816 | mp_h = strip_h; | |
1817 | ||
1818 | while (state != DEALLOC_DONE) { | |
1819 | switch (state) { | |
1820 | /* Truncate a full metapath at the given strip height. | |
1821 | * Note that strip_h == mp_h in order to be in this state. */ | |
1822 | case DEALLOC_MP_FULL: | |
d552a2b9 BP |
1823 | bh = mp.mp_bh[mp_h]; |
1824 | gfs2_assert_withdraw(sdp, bh); | |
1825 | if (gfs2_assert_withdraw(sdp, | |
1826 | prev_bnr != bh->b_blocknr)) { | |
1827 | printk(KERN_EMERG "GFS2: fsid=%s:inode %llu, " | |
1828 | "block:%llu, i_h:%u, s_h:%u, mp_h:%u\n", | |
1829 | sdp->sd_fsname, | |
1830 | (unsigned long long)ip->i_no_addr, | |
1831 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1832 | } | |
1833 | prev_bnr = bh->b_blocknr; | |
cb7f0903 | 1834 | |
5cf26b1e AG |
1835 | if (gfs2_metatype_check(sdp, bh, |
1836 | (mp_h ? GFS2_METATYPE_IN : | |
1837 | GFS2_METATYPE_DI))) { | |
1838 | ret = -EIO; | |
1839 | goto out; | |
1840 | } | |
1841 | ||
10d2cf94 AG |
1842 | /* |
1843 | * Below, passing end_aligned as 0 gives us the | |
1844 | * metapointer range excluding the end point: the end | |
1845 | * point is the first metapath we must not deallocate! | |
1846 | */ | |
1847 | ||
5cf26b1e | 1848 | metapointer_range(&mp, mp_h, start_list, start_aligned, |
10d2cf94 | 1849 | end_list, 0 /* end_aligned */, |
5cf26b1e AG |
1850 | &start, &end); |
1851 | ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h], | |
1852 | start, end, | |
1853 | mp_h != ip->i_height - 1, | |
1854 | &btotal); | |
cb7f0903 | 1855 | |
d552a2b9 BP |
1856 | /* If we hit an error or just swept dinode buffer, |
1857 | just exit. */ | |
1858 | if (ret || !mp_h) { | |
1859 | state = DEALLOC_DONE; | |
1860 | break; | |
1861 | } | |
1862 | state = DEALLOC_MP_LOWER; | |
1863 | break; | |
1864 | ||
1865 | /* lower the metapath strip height */ | |
1866 | case DEALLOC_MP_LOWER: | |
1867 | /* We're done with the current buffer, so release it, | |
1868 | unless it's the dinode buffer. Then back up to the | |
1869 | previous pointer. */ | |
1870 | if (mp_h) { | |
1871 | brelse(mp.mp_bh[mp_h]); | |
1872 | mp.mp_bh[mp_h] = NULL; | |
1873 | } | |
1874 | /* If we can't get any lower in height, we've stripped | |
1875 | off all we can. Next step is to back up and start | |
1876 | stripping the previous level of metadata. */ | |
1877 | if (mp_h == 0) { | |
1878 | strip_h--; | |
cb7f0903 | 1879 | memcpy(mp.mp_list, start_list, sizeof(start_list)); |
d552a2b9 BP |
1880 | mp_h = strip_h; |
1881 | state = DEALLOC_FILL_MP; | |
1882 | break; | |
1883 | } | |
1884 | mp.mp_list[mp_h] = 0; | |
1885 | mp_h--; /* search one metadata height down */ | |
d552a2b9 | 1886 | mp.mp_list[mp_h]++; |
10d2cf94 AG |
1887 | if (walk_done(sdp, &mp, mp_h, end_list, end_aligned)) |
1888 | break; | |
d552a2b9 BP |
1889 | /* Here we've found a part of the metapath that is not |
1890 | * allocated. We need to search at that height for the | |
1891 | * next non-null pointer. */ | |
10d2cf94 | 1892 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) { |
d552a2b9 BP |
1893 | state = DEALLOC_FILL_MP; |
1894 | mp_h++; | |
1895 | } | |
1896 | /* No more non-null pointers at this height. Back up | |
1897 | to the previous height and try again. */ | |
1898 | break; /* loop around in the same state */ | |
1899 | ||
1900 | /* Fill the metapath with buffers to the given height. */ | |
1901 | case DEALLOC_FILL_MP: | |
1902 | /* Fill the buffers out to the current height. */ | |
1903 | ret = fillup_metapath(ip, &mp, mp_h); | |
c3ce5aa9 | 1904 | if (ret < 0) |
d552a2b9 | 1905 | goto out; |
c3ce5aa9 AG |
1906 | |
1907 | /* issue read-ahead on metadata */ | |
1908 | if (mp.mp_aheight > 1) { | |
5cf26b1e AG |
1909 | for (; ret > 1; ret--) { |
1910 | metapointer_range(&mp, mp.mp_aheight - ret, | |
1911 | start_list, start_aligned, | |
10d2cf94 | 1912 | end_list, end_aligned, |
5cf26b1e AG |
1913 | &start, &end); |
1914 | gfs2_metapath_ra(ip->i_gl, start, end); | |
1915 | } | |
c3ce5aa9 | 1916 | } |
d552a2b9 BP |
1917 | |
1918 | /* If buffers found for the entire strip height */ | |
e8b43fe0 | 1919 | if (mp.mp_aheight - 1 == strip_h) { |
d552a2b9 BP |
1920 | state = DEALLOC_MP_FULL; |
1921 | break; | |
1922 | } | |
e8b43fe0 AG |
1923 | if (mp.mp_aheight < ip->i_height) /* We have a partial height */ |
1924 | mp_h = mp.mp_aheight - 1; | |
d552a2b9 BP |
1925 | |
1926 | /* If we find a non-null block pointer, crawl a bit | |
1927 | higher up in the metapath and try again, otherwise | |
1928 | we need to look lower for a new starting point. */ | |
10d2cf94 | 1929 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) |
d552a2b9 BP |
1930 | mp_h++; |
1931 | else | |
1932 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1933 | break; |
d552a2b9 | 1934 | } |
b3b94faa DT |
1935 | } |
1936 | ||
d552a2b9 BP |
1937 | if (btotal) { |
1938 | if (current->journal_info == NULL) { | |
1939 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1940 | RES_QUOTA, 0); | |
1941 | if (ret) | |
1942 | goto out; | |
1943 | down_write(&ip->i_rw_mutex); | |
1944 | } | |
1945 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
1946 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
1947 | ip->i_inode.i_gid); | |
b32c8c76 | 1948 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
d552a2b9 BP |
1949 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1950 | gfs2_dinode_out(ip, dibh->b_data); | |
1951 | up_write(&ip->i_rw_mutex); | |
1952 | gfs2_trans_end(sdp); | |
1953 | } | |
b3b94faa | 1954 | |
d552a2b9 BP |
1955 | out: |
1956 | if (gfs2_holder_initialized(&rd_gh)) | |
1957 | gfs2_glock_dq_uninit(&rd_gh); | |
1958 | if (current->journal_info) { | |
1959 | up_write(&ip->i_rw_mutex); | |
1960 | gfs2_trans_end(sdp); | |
1961 | cond_resched(); | |
1962 | } | |
1963 | gfs2_quota_unhold(ip); | |
1964 | out_metapath: | |
1965 | release_metapath(&mp); | |
1966 | return ret; | |
b3b94faa DT |
1967 | } |
1968 | ||
1969 | static int trunc_end(struct gfs2_inode *ip) | |
1970 | { | |
feaa7bba | 1971 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1972 | struct buffer_head *dibh; |
1973 | int error; | |
1974 | ||
1975 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1976 | if (error) | |
1977 | return error; | |
1978 | ||
1979 | down_write(&ip->i_rw_mutex); | |
1980 | ||
1981 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1982 | if (error) | |
1983 | goto out; | |
1984 | ||
a2e0f799 | 1985 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1986 | ip->i_height = 0; |
ce276b06 | 1987 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 1988 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 1989 | gfs2_ordered_del_inode(ip); |
b3b94faa | 1990 | } |
078cd827 | 1991 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
383f01fb | 1992 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1993 | |
350a9b0a | 1994 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 1995 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1996 | brelse(dibh); |
1997 | ||
a91ea69f | 1998 | out: |
b3b94faa | 1999 | up_write(&ip->i_rw_mutex); |
b3b94faa | 2000 | gfs2_trans_end(sdp); |
b3b94faa DT |
2001 | return error; |
2002 | } | |
2003 | ||
2004 | /** | |
2005 | * do_shrink - make a file smaller | |
ff8f33c8 | 2006 | * @inode: the inode |
ff8f33c8 | 2007 | * @newsize: the size to make the file |
b3b94faa | 2008 | * |
ff8f33c8 SW |
2009 | * Called with an exclusive lock on @inode. The @size must |
2010 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
2011 | * |
2012 | * Returns: errno | |
2013 | */ | |
2014 | ||
8b5860a3 | 2015 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 2016 | { |
ff8f33c8 | 2017 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
2018 | int error; |
2019 | ||
8b5860a3 | 2020 | error = trunc_start(inode, newsize); |
b3b94faa DT |
2021 | if (error < 0) |
2022 | return error; | |
ff8f33c8 | 2023 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
2024 | return 0; |
2025 | ||
10d2cf94 | 2026 | error = punch_hole(ip, newsize, 0); |
ff8f33c8 | 2027 | if (error == 0) |
b3b94faa DT |
2028 | error = trunc_end(ip); |
2029 | ||
2030 | return error; | |
2031 | } | |
2032 | ||
ff8f33c8 | 2033 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 2034 | { |
ff8f33c8 SW |
2035 | int ret; |
2036 | ||
8b5860a3 | 2037 | ret = do_shrink(inode, inode->i_size); |
ff8f33c8 SW |
2038 | WARN_ON(ret != 0); |
2039 | } | |
2040 | ||
2041 | /** | |
2042 | * do_grow - Touch and update inode size | |
2043 | * @inode: The inode | |
2044 | * @size: The new size | |
2045 | * | |
2046 | * This function updates the timestamps on the inode and | |
2047 | * may also increase the size of the inode. This function | |
2048 | * must not be called with @size any smaller than the current | |
2049 | * inode size. | |
2050 | * | |
2051 | * Although it is not strictly required to unstuff files here, | |
2052 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
2053 | * code which will result in a buffer overrun if the size is larger | |
2054 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 2055 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
2056 | * just update the inode size directly. |
2057 | * | |
2058 | * Returns: 0 on success, or -ve on error | |
2059 | */ | |
2060 | ||
2061 | static int do_grow(struct inode *inode, u64 size) | |
2062 | { | |
2063 | struct gfs2_inode *ip = GFS2_I(inode); | |
2064 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 2065 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
2066 | struct buffer_head *dibh; |
2067 | int error; | |
2f7ee358 | 2068 | int unstuff = 0; |
a13b8c5f | 2069 | |
235628c5 | 2070 | if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) { |
b8fbf471 | 2071 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 2072 | if (error) |
5407e242 | 2073 | return error; |
ff8f33c8 | 2074 | |
7b9cff46 | 2075 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
2076 | if (error) |
2077 | goto do_grow_qunlock; | |
2f7ee358 | 2078 | unstuff = 1; |
ff8f33c8 SW |
2079 | } |
2080 | ||
a01aedfe BP |
2081 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
2082 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? | |
2083 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 2084 | if (error) |
ff8f33c8 | 2085 | goto do_grow_release; |
a13b8c5f | 2086 | |
2f7ee358 | 2087 | if (unstuff) { |
ff8f33c8 SW |
2088 | error = gfs2_unstuff_dinode(ip, NULL); |
2089 | if (error) | |
2090 | goto do_end_trans; | |
2091 | } | |
a13b8c5f WC |
2092 | |
2093 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2094 | if (error) | |
ff8f33c8 | 2095 | goto do_end_trans; |
a13b8c5f | 2096 | |
ff8f33c8 | 2097 | i_size_write(inode, size); |
078cd827 | 2098 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
350a9b0a | 2099 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
2100 | gfs2_dinode_out(ip, dibh->b_data); |
2101 | brelse(dibh); | |
2102 | ||
ff8f33c8 | 2103 | do_end_trans: |
a13b8c5f | 2104 | gfs2_trans_end(sdp); |
ff8f33c8 | 2105 | do_grow_release: |
2f7ee358 | 2106 | if (unstuff) { |
ff8f33c8 SW |
2107 | gfs2_inplace_release(ip); |
2108 | do_grow_qunlock: | |
2109 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 2110 | } |
a13b8c5f WC |
2111 | return error; |
2112 | } | |
2113 | ||
b3b94faa | 2114 | /** |
ff8f33c8 SW |
2115 | * gfs2_setattr_size - make a file a given size |
2116 | * @inode: the inode | |
2117 | * @newsize: the size to make the file | |
b3b94faa | 2118 | * |
ff8f33c8 | 2119 | * The file size can grow, shrink, or stay the same size. This |
3e7aafc3 | 2120 | * is called holding i_rwsem and an exclusive glock on the inode |
ff8f33c8 | 2121 | * in question. |
b3b94faa DT |
2122 | * |
2123 | * Returns: errno | |
2124 | */ | |
2125 | ||
ff8f33c8 | 2126 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 2127 | { |
af5c2697 | 2128 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 2129 | int ret; |
b3b94faa | 2130 | |
ff8f33c8 | 2131 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 2132 | |
ff8f33c8 SW |
2133 | ret = inode_newsize_ok(inode, newsize); |
2134 | if (ret) | |
2135 | return ret; | |
b3b94faa | 2136 | |
562c72aa CH |
2137 | inode_dio_wait(inode); |
2138 | ||
b54e9a0b | 2139 | ret = gfs2_rsqa_alloc(ip); |
d2b47cfb | 2140 | if (ret) |
2b3dcf35 | 2141 | goto out; |
d2b47cfb | 2142 | |
8b5860a3 | 2143 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
2144 | ret = do_grow(inode, newsize); |
2145 | goto out; | |
2146 | } | |
ff8f33c8 | 2147 | |
8b5860a3 | 2148 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 2149 | out: |
a097dc7e | 2150 | gfs2_rsqa_delete(ip, NULL); |
2b3dcf35 | 2151 | return ret; |
b3b94faa DT |
2152 | } |
2153 | ||
2154 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
2155 | { | |
2156 | int error; | |
10d2cf94 | 2157 | error = punch_hole(ip, i_size_read(&ip->i_inode), 0); |
b3b94faa DT |
2158 | if (!error) |
2159 | error = trunc_end(ip); | |
2160 | return error; | |
2161 | } | |
2162 | ||
2163 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
2164 | { | |
10d2cf94 | 2165 | return punch_hole(ip, 0, 0); |
b3b94faa DT |
2166 | } |
2167 | ||
b50f227b SW |
2168 | /** |
2169 | * gfs2_free_journal_extents - Free cached journal bmap info | |
2170 | * @jd: The journal | |
2171 | * | |
2172 | */ | |
2173 | ||
2174 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
2175 | { | |
2176 | struct gfs2_journal_extent *jext; | |
2177 | ||
2178 | while(!list_empty(&jd->extent_list)) { | |
2179 | jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list); | |
2180 | list_del(&jext->list); | |
2181 | kfree(jext); | |
2182 | } | |
2183 | } | |
2184 | ||
2185 | /** | |
2186 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
2187 | * @jd: The journal descriptor | |
2188 | * @lblock: The logical block at start of new extent | |
c62baf65 | 2189 | * @dblock: The physical block at start of new extent |
b50f227b SW |
2190 | * @blocks: Size of extent in fs blocks |
2191 | * | |
2192 | * Returns: 0 on success or -ENOMEM | |
2193 | */ | |
2194 | ||
2195 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
2196 | { | |
2197 | struct gfs2_journal_extent *jext; | |
2198 | ||
2199 | if (!list_empty(&jd->extent_list)) { | |
2200 | jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list); | |
2201 | if ((jext->dblock + jext->blocks) == dblock) { | |
2202 | jext->blocks += blocks; | |
2203 | return 0; | |
2204 | } | |
2205 | } | |
2206 | ||
2207 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
2208 | if (jext == NULL) | |
2209 | return -ENOMEM; | |
2210 | jext->dblock = dblock; | |
2211 | jext->lblock = lblock; | |
2212 | jext->blocks = blocks; | |
2213 | list_add_tail(&jext->list, &jd->extent_list); | |
2214 | jd->nr_extents++; | |
2215 | return 0; | |
2216 | } | |
2217 | ||
2218 | /** | |
2219 | * gfs2_map_journal_extents - Cache journal bmap info | |
2220 | * @sdp: The super block | |
2221 | * @jd: The journal to map | |
2222 | * | |
2223 | * Create a reusable "extent" mapping from all logical | |
2224 | * blocks to all physical blocks for the given journal. This will save | |
2225 | * us time when writing journal blocks. Most journals will have only one | |
2226 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
2227 | * arranges the journal blocks sequentially to maximize performance. | |
2228 | * So the extent would map the first block for the entire file length. | |
2229 | * However, gfs2_jadd can happen while file activity is happening, so | |
2230 | * those journals may not be sequential. Less likely is the case where | |
2231 | * the users created their own journals by mounting the metafs and | |
2232 | * laying it out. But it's still possible. These journals might have | |
2233 | * several extents. | |
2234 | * | |
2235 | * Returns: 0 on success, or error on failure | |
2236 | */ | |
2237 | ||
2238 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
2239 | { | |
2240 | u64 lblock = 0; | |
2241 | u64 lblock_stop; | |
2242 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
2243 | struct buffer_head bh; | |
2244 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
2245 | u64 size; | |
2246 | int rc; | |
2247 | ||
2248 | lblock_stop = i_size_read(jd->jd_inode) >> shift; | |
2249 | size = (lblock_stop - lblock) << shift; | |
2250 | jd->nr_extents = 0; | |
2251 | WARN_ON(!list_empty(&jd->extent_list)); | |
2252 | ||
2253 | do { | |
2254 | bh.b_state = 0; | |
2255 | bh.b_blocknr = 0; | |
2256 | bh.b_size = size; | |
2257 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
2258 | if (rc || !buffer_mapped(&bh)) | |
2259 | goto fail; | |
2260 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
2261 | if (rc) | |
2262 | goto fail; | |
2263 | size -= bh.b_size; | |
2264 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2265 | } while(size > 0); | |
2266 | ||
2267 | fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid, | |
2268 | jd->nr_extents); | |
2269 | return 0; | |
2270 | ||
2271 | fail: | |
2272 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
2273 | rc, jd->jd_jid, | |
2274 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
2275 | jd->nr_extents); | |
2276 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
2277 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
2278 | bh.b_state, (unsigned long long)bh.b_size); | |
2279 | gfs2_free_journal_extents(jd); | |
2280 | return rc; | |
2281 | } | |
2282 | ||
b3b94faa DT |
2283 | /** |
2284 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
2285 | * @ip: the file being written to | |
2286 | * @offset: the offset to write to | |
2287 | * @len: the number of bytes being written | |
b3b94faa | 2288 | * |
461cb419 | 2289 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
2290 | */ |
2291 | ||
cd915493 | 2292 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 2293 | unsigned int len) |
b3b94faa | 2294 | { |
feaa7bba | 2295 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
2296 | struct buffer_head bh; |
2297 | unsigned int shift; | |
2298 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 2299 | u64 end_of_file; |
b3b94faa | 2300 | |
b3b94faa DT |
2301 | if (!len) |
2302 | return 0; | |
2303 | ||
2304 | if (gfs2_is_stuffed(ip)) { | |
235628c5 | 2305 | if (offset + len > gfs2_max_stuffed_size(ip)) |
461cb419 | 2306 | return 1; |
b3b94faa DT |
2307 | return 0; |
2308 | } | |
2309 | ||
941e6d7d | 2310 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 2311 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 2312 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
2313 | lblock = offset >> shift; |
2314 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
2315 | if (lblock_stop > end_of_file) | |
461cb419 | 2316 | return 1; |
b3b94faa | 2317 | |
941e6d7d SW |
2318 | size = (lblock_stop - lblock) << shift; |
2319 | do { | |
2320 | bh.b_state = 0; | |
2321 | bh.b_size = size; | |
2322 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
2323 | if (!buffer_mapped(&bh)) | |
461cb419 | 2324 | return 1; |
941e6d7d SW |
2325 | size -= bh.b_size; |
2326 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2327 | } while(size > 0); | |
b3b94faa DT |
2328 | |
2329 | return 0; | |
2330 | } | |
2331 | ||
4e56a641 AG |
2332 | static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length) |
2333 | { | |
2334 | struct gfs2_inode *ip = GFS2_I(inode); | |
2335 | struct buffer_head *dibh; | |
2336 | int error; | |
2337 | ||
2338 | if (offset >= inode->i_size) | |
2339 | return 0; | |
2340 | if (offset + length > inode->i_size) | |
2341 | length = inode->i_size - offset; | |
2342 | ||
2343 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2344 | if (error) | |
2345 | return error; | |
2346 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
2347 | memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0, | |
2348 | length); | |
2349 | brelse(dibh); | |
2350 | return 0; | |
2351 | } | |
2352 | ||
2353 | static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset, | |
2354 | loff_t length) | |
2355 | { | |
2356 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2357 | loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
2358 | int error; | |
2359 | ||
2360 | while (length) { | |
2361 | struct gfs2_trans *tr; | |
2362 | loff_t chunk; | |
2363 | unsigned int offs; | |
2364 | ||
2365 | chunk = length; | |
2366 | if (chunk > max_chunk) | |
2367 | chunk = max_chunk; | |
2368 | ||
2369 | offs = offset & ~PAGE_MASK; | |
2370 | if (offs && chunk > PAGE_SIZE) | |
2371 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
2372 | ||
2373 | truncate_pagecache_range(inode, offset, chunk); | |
2374 | offset += chunk; | |
2375 | length -= chunk; | |
2376 | ||
2377 | tr = current->journal_info; | |
2378 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
2379 | continue; | |
2380 | ||
2381 | gfs2_trans_end(sdp); | |
2382 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
2383 | if (error) | |
2384 | return error; | |
2385 | } | |
2386 | return 0; | |
2387 | } | |
2388 | ||
2389 | int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length) | |
2390 | { | |
2391 | struct inode *inode = file_inode(file); | |
2392 | struct gfs2_inode *ip = GFS2_I(inode); | |
2393 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2394 | int error; | |
2395 | ||
2396 | if (gfs2_is_jdata(ip)) | |
2397 | error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA, | |
2398 | GFS2_JTRUNC_REVOKES); | |
2399 | else | |
2400 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2401 | if (error) | |
2402 | return error; | |
2403 | ||
2404 | if (gfs2_is_stuffed(ip)) { | |
2405 | error = stuffed_zero_range(inode, offset, length); | |
2406 | if (error) | |
2407 | goto out; | |
2408 | } else { | |
2409 | unsigned int start_off, end_off, blocksize; | |
2410 | ||
2411 | blocksize = i_blocksize(inode); | |
2412 | start_off = offset & (blocksize - 1); | |
2413 | end_off = (offset + length) & (blocksize - 1); | |
2414 | if (start_off) { | |
2415 | unsigned int len = length; | |
2416 | if (length > blocksize - start_off) | |
2417 | len = blocksize - start_off; | |
2418 | error = gfs2_block_zero_range(inode, offset, len); | |
2419 | if (error) | |
2420 | goto out; | |
2421 | if (start_off + length < blocksize) | |
2422 | end_off = 0; | |
2423 | } | |
2424 | if (end_off) { | |
2425 | error = gfs2_block_zero_range(inode, | |
2426 | offset + length - end_off, end_off); | |
2427 | if (error) | |
2428 | goto out; | |
2429 | } | |
2430 | } | |
2431 | ||
2432 | if (gfs2_is_jdata(ip)) { | |
2433 | BUG_ON(!current->journal_info); | |
2434 | gfs2_journaled_truncate_range(inode, offset, length); | |
2435 | } else | |
2436 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2437 | ||
2438 | file_update_time(file); | |
2439 | mark_inode_dirty(inode); | |
2440 | ||
2441 | if (current->journal_info) | |
2442 | gfs2_trans_end(sdp); | |
2443 | ||
2444 | if (!gfs2_is_stuffed(ip)) | |
2445 | error = punch_hole(ip, offset, length); | |
2446 | ||
2447 | out: | |
2448 | if (current->journal_info) | |
2449 | gfs2_trans_end(sdp); | |
2450 | return error; | |
2451 | } |