Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 4 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
b3b94faa DT |
7 | #include <linux/spinlock.h> |
8 | #include <linux/completion.h> | |
9 | #include <linux/buffer_head.h> | |
64dd153c | 10 | #include <linux/blkdev.h> |
5c676f6d | 11 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 12 | #include <linux/crc32.h> |
3974320c | 13 | #include <linux/iomap.h> |
98583b3e | 14 | #include <linux/ktime.h> |
b3b94faa DT |
15 | |
16 | #include "gfs2.h" | |
5c676f6d | 17 | #include "incore.h" |
b3b94faa DT |
18 | #include "bmap.h" |
19 | #include "glock.h" | |
20 | #include "inode.h" | |
b3b94faa | 21 | #include "meta_io.h" |
b3b94faa DT |
22 | #include "quota.h" |
23 | #include "rgrp.h" | |
45138990 | 24 | #include "log.h" |
4c16c36a | 25 | #include "super.h" |
b3b94faa | 26 | #include "trans.h" |
18ec7d5c | 27 | #include "dir.h" |
5c676f6d | 28 | #include "util.h" |
64bc06bb | 29 | #include "aops.h" |
63997775 | 30 | #include "trace_gfs2.h" |
b3b94faa DT |
31 | |
32 | /* This doesn't need to be that large as max 64 bit pointers in a 4k | |
33 | * block is 512, so __u16 is fine for that. It saves stack space to | |
34 | * keep it small. | |
35 | */ | |
36 | struct metapath { | |
dbac6710 | 37 | struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT]; |
b3b94faa | 38 | __u16 mp_list[GFS2_MAX_META_HEIGHT]; |
5f8bd444 BP |
39 | int mp_fheight; /* find_metapath height */ |
40 | int mp_aheight; /* actual height (lookup height) */ | |
b3b94faa DT |
41 | }; |
42 | ||
64bc06bb AG |
43 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length); |
44 | ||
f25ef0c1 SW |
45 | /** |
46 | * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page | |
47 | * @ip: the inode | |
48 | * @dibh: the dinode buffer | |
49 | * @block: the block number that was allocated | |
ff8f33c8 | 50 | * @page: The (optional) page. This is looked up if @page is NULL |
f25ef0c1 SW |
51 | * |
52 | * Returns: errno | |
53 | */ | |
54 | ||
55 | static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh, | |
cd915493 | 56 | u64 block, struct page *page) |
f25ef0c1 | 57 | { |
f25ef0c1 | 58 | struct inode *inode = &ip->i_inode; |
f25ef0c1 SW |
59 | |
60 | if (!PageUptodate(page)) { | |
61 | void *kaddr = kmap(page); | |
602c89d2 SW |
62 | u64 dsize = i_size_read(inode); |
63 | ||
235628c5 AG |
64 | if (dsize > gfs2_max_stuffed_size(ip)) |
65 | dsize = gfs2_max_stuffed_size(ip); | |
f25ef0c1 | 66 | |
602c89d2 | 67 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 68 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
f25ef0c1 SW |
69 | kunmap(page); |
70 | ||
71 | SetPageUptodate(page); | |
72 | } | |
73 | ||
2164f9b9 CH |
74 | if (gfs2_is_jdata(ip)) { |
75 | struct buffer_head *bh; | |
f25ef0c1 | 76 | |
2164f9b9 CH |
77 | if (!page_has_buffers(page)) |
78 | create_empty_buffers(page, BIT(inode->i_blkbits), | |
79 | BIT(BH_Uptodate)); | |
f25ef0c1 | 80 | |
2164f9b9 CH |
81 | bh = page_buffers(page); |
82 | if (!buffer_mapped(bh)) | |
83 | map_bh(bh, inode->i_sb, block); | |
f25ef0c1 | 84 | |
2164f9b9 | 85 | set_buffer_uptodate(bh); |
350a9b0a | 86 | gfs2_trans_add_data(ip->i_gl, bh); |
2164f9b9 CH |
87 | } else { |
88 | set_page_dirty(page); | |
845802b1 AG |
89 | gfs2_ordered_add_inode(ip); |
90 | } | |
f25ef0c1 | 91 | |
f25ef0c1 SW |
92 | return 0; |
93 | } | |
94 | ||
7a607a41 | 95 | static int __gfs2_unstuff_inode(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
96 | { |
97 | struct buffer_head *bh, *dibh; | |
48516ced | 98 | struct gfs2_dinode *di; |
cd915493 | 99 | u64 block = 0; |
18ec7d5c | 100 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
101 | int error; |
102 | ||
b3b94faa DT |
103 | error = gfs2_meta_inode_buffer(ip, &dibh); |
104 | if (error) | |
7a607a41 | 105 | return error; |
907b9bce | 106 | |
a2e0f799 | 107 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
108 | /* Get a free block, fill it with the stuffed data, |
109 | and write it out to disk */ | |
110 | ||
b45e41d7 | 111 | unsigned int n = 1; |
6e87ed0f | 112 | error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL); |
09010978 SW |
113 | if (error) |
114 | goto out_brelse; | |
18ec7d5c | 115 | if (isdir) { |
fbb27873 | 116 | gfs2_trans_remove_revoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 117 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
118 | if (error) |
119 | goto out_brelse; | |
48516ced | 120 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
121 | dibh, sizeof(struct gfs2_dinode)); |
122 | brelse(bh); | |
123 | } else { | |
f25ef0c1 | 124 | error = gfs2_unstuffer_page(ip, dibh, block, page); |
b3b94faa DT |
125 | if (error) |
126 | goto out_brelse; | |
127 | } | |
128 | } | |
129 | ||
130 | /* Set up the pointer to the new block */ | |
131 | ||
350a9b0a | 132 | gfs2_trans_add_meta(ip->i_gl, dibh); |
48516ced | 133 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
134 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
135 | ||
a2e0f799 | 136 | if (i_size_read(&ip->i_inode)) { |
48516ced | 137 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
138 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
139 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
140 | } |
141 | ||
ecc30c79 | 142 | ip->i_height = 1; |
48516ced | 143 | di->di_height = cpu_to_be16(1); |
b3b94faa | 144 | |
a91ea69f | 145 | out_brelse: |
b3b94faa | 146 | brelse(dibh); |
7a607a41 AG |
147 | return error; |
148 | } | |
149 | ||
150 | /** | |
151 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
152 | * @ip: The GFS2 inode to unstuff | |
153 | * | |
154 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
155 | * that the height can be grown in the traditional way. | |
156 | * | |
157 | * Returns: errno | |
158 | */ | |
159 | ||
160 | int gfs2_unstuff_dinode(struct gfs2_inode *ip) | |
161 | { | |
162 | struct inode *inode = &ip->i_inode; | |
163 | struct page *page; | |
164 | int error; | |
165 | ||
166 | down_write(&ip->i_rw_mutex); | |
167 | page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS); | |
168 | error = -ENOMEM; | |
169 | if (!page) | |
170 | goto out; | |
171 | error = __gfs2_unstuff_inode(ip, page); | |
172 | unlock_page(page); | |
173 | put_page(page); | |
a91ea69f | 174 | out: |
b3b94faa | 175 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
176 | return error; |
177 | } | |
178 | ||
b3b94faa DT |
179 | /** |
180 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 181 | * @sdp: The superblock |
b3b94faa | 182 | * @block: The disk block to look up |
07e23d68 | 183 | * @mp: The metapath to return the result in |
9b8c81d1 | 184 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
185 | * |
186 | * This routine returns a struct metapath structure that defines a path | |
187 | * through the metadata of inode "ip" to get to block "block". | |
188 | * | |
189 | * Example: | |
190 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
191 | * filesystem with a blocksize of 4096. | |
192 | * | |
193 | * find_metapath() would return a struct metapath structure set to: | |
07e23d68 | 194 | * mp_fheight = 3, mp_list[0] = 0, mp_list[1] = 48, and mp_list[2] = 165. |
b3b94faa DT |
195 | * |
196 | * That means that in order to get to the block containing the byte at | |
197 | * offset 101342453, we would load the indirect block pointed to by pointer | |
198 | * 0 in the dinode. We would then load the indirect block pointed to by | |
199 | * pointer 48 in that indirect block. We would then load the data block | |
200 | * pointed to by pointer 165 in that indirect block. | |
201 | * | |
202 | * ---------------------------------------- | |
203 | * | Dinode | | | |
204 | * | | 4| | |
205 | * | |0 1 2 3 4 5 9| | |
206 | * | | 6| | |
207 | * ---------------------------------------- | |
208 | * | | |
209 | * | | |
210 | * V | |
211 | * ---------------------------------------- | |
212 | * | Indirect Block | | |
213 | * | 5| | |
214 | * | 4 4 4 4 4 5 5 1| | |
215 | * |0 5 6 7 8 9 0 1 2| | |
216 | * ---------------------------------------- | |
217 | * | | |
218 | * | | |
219 | * V | |
220 | * ---------------------------------------- | |
221 | * | Indirect Block | | |
222 | * | 1 1 1 1 1 5| | |
223 | * | 6 6 6 6 6 1| | |
224 | * |0 3 4 5 6 7 2| | |
225 | * ---------------------------------------- | |
226 | * | | |
227 | * | | |
228 | * V | |
229 | * ---------------------------------------- | |
230 | * | Data block containing offset | | |
231 | * | 101342453 | | |
232 | * | | | |
233 | * | | | |
234 | * ---------------------------------------- | |
235 | * | |
236 | */ | |
237 | ||
9b8c81d1 SW |
238 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
239 | struct metapath *mp, unsigned int height) | |
b3b94faa | 240 | { |
b3b94faa DT |
241 | unsigned int i; |
242 | ||
5f8bd444 | 243 | mp->mp_fheight = height; |
9b8c81d1 | 244 | for (i = height; i--;) |
7eabb77e | 245 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
246 | } |
247 | ||
5af4e7a0 | 248 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 249 | { |
5af4e7a0 BM |
250 | if (mp->mp_list[0] == 0) |
251 | return 2; | |
252 | return 1; | |
9b8c81d1 SW |
253 | } |
254 | ||
d552a2b9 | 255 | /** |
20cdc193 | 256 | * metaptr1 - Return the first possible metadata pointer in a metapath buffer |
d552a2b9 BP |
257 | * @height: The metadata height (0 = dinode) |
258 | * @mp: The metapath | |
259 | */ | |
260 | static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp) | |
261 | { | |
262 | struct buffer_head *bh = mp->mp_bh[height]; | |
263 | if (height == 0) | |
264 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode))); | |
265 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header))); | |
266 | } | |
267 | ||
b3b94faa DT |
268 | /** |
269 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
270 | * @height: The metadata height (0 = dinode) |
271 | * @mp: The metapath | |
272 | * | |
273 | * Return a pointer to the block number of the next height of the metadata | |
274 | * tree given a buffer containing the pointer to the current height of the | |
275 | * metadata tree. | |
276 | */ | |
277 | ||
9b8c81d1 | 278 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 279 | { |
d552a2b9 BP |
280 | __be64 *p = metaptr1(height, mp); |
281 | return p + mp->mp_list[height]; | |
b3b94faa DT |
282 | } |
283 | ||
7841b9f0 AG |
284 | static inline const __be64 *metaend(unsigned int height, const struct metapath *mp) |
285 | { | |
286 | const struct buffer_head *bh = mp->mp_bh[height]; | |
287 | return (const __be64 *)(bh->b_data + bh->b_size); | |
288 | } | |
289 | ||
290 | static void clone_metapath(struct metapath *clone, struct metapath *mp) | |
291 | { | |
292 | unsigned int hgt; | |
293 | ||
294 | *clone = *mp; | |
295 | for (hgt = 0; hgt < mp->mp_aheight; hgt++) | |
296 | get_bh(clone->mp_bh[hgt]); | |
297 | } | |
298 | ||
5cf26b1e | 299 | static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end) |
b99b98dc | 300 | { |
b99b98dc SW |
301 | const __be64 *t; |
302 | ||
5cf26b1e | 303 | for (t = start; t < end; t++) { |
c3ce5aa9 AG |
304 | struct buffer_head *rabh; |
305 | ||
b99b98dc SW |
306 | if (!*t) |
307 | continue; | |
308 | ||
309 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
310 | if (trylock_buffer(rabh)) { | |
311 | if (!buffer_uptodate(rabh)) { | |
312 | rabh->b_end_io = end_buffer_read_sync; | |
e477b24b CL |
313 | submit_bh(REQ_OP_READ, |
314 | REQ_RAHEAD | REQ_META | REQ_PRIO, | |
315 | rabh); | |
b99b98dc SW |
316 | continue; |
317 | } | |
318 | unlock_buffer(rabh); | |
319 | } | |
320 | brelse(rabh); | |
321 | } | |
322 | } | |
323 | ||
e8b43fe0 AG |
324 | static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, |
325 | unsigned int x, unsigned int h) | |
d552a2b9 | 326 | { |
e8b43fe0 AG |
327 | for (; x < h; x++) { |
328 | __be64 *ptr = metapointer(x, mp); | |
329 | u64 dblock = be64_to_cpu(*ptr); | |
330 | int ret; | |
d552a2b9 | 331 | |
e8b43fe0 AG |
332 | if (!dblock) |
333 | break; | |
6d8da302 | 334 | ret = gfs2_meta_buffer(ip, GFS2_METATYPE_IN, dblock, &mp->mp_bh[x + 1]); |
e8b43fe0 AG |
335 | if (ret) |
336 | return ret; | |
337 | } | |
338 | mp->mp_aheight = x + 1; | |
339 | return 0; | |
d552a2b9 BP |
340 | } |
341 | ||
b3b94faa | 342 | /** |
9b8c81d1 SW |
343 | * lookup_metapath - Walk the metadata tree to a specific point |
344 | * @ip: The inode | |
b3b94faa | 345 | * @mp: The metapath |
b3b94faa | 346 | * |
9b8c81d1 SW |
347 | * Assumes that the inode's buffer has already been looked up and |
348 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
349 | * by find_metapath(). | |
350 | * | |
351 | * If this function encounters part of the tree which has not been | |
352 | * allocated, it returns the current height of the tree at the point | |
353 | * at which it found the unallocated block. Blocks which are found are | |
354 | * added to the mp->mp_bh[] list. | |
b3b94faa | 355 | * |
e8b43fe0 | 356 | * Returns: error |
b3b94faa DT |
357 | */ |
358 | ||
9b8c81d1 | 359 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 360 | { |
e8b43fe0 | 361 | return __fillup_metapath(ip, mp, 0, ip->i_height - 1); |
dbac6710 SW |
362 | } |
363 | ||
d552a2b9 BP |
364 | /** |
365 | * fillup_metapath - fill up buffers for the metadata path to a specific height | |
366 | * @ip: The inode | |
367 | * @mp: The metapath | |
368 | * @h: The height to which it should be mapped | |
369 | * | |
370 | * Similar to lookup_metapath, but does lookups for a range of heights | |
371 | * | |
c3ce5aa9 | 372 | * Returns: error or the number of buffers filled |
d552a2b9 BP |
373 | */ |
374 | ||
375 | static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h) | |
376 | { | |
e8b43fe0 | 377 | unsigned int x = 0; |
c3ce5aa9 | 378 | int ret; |
d552a2b9 BP |
379 | |
380 | if (h) { | |
381 | /* find the first buffer we need to look up. */ | |
e8b43fe0 AG |
382 | for (x = h - 1; x > 0; x--) { |
383 | if (mp->mp_bh[x]) | |
384 | break; | |
d552a2b9 BP |
385 | } |
386 | } | |
c3ce5aa9 AG |
387 | ret = __fillup_metapath(ip, mp, x, h); |
388 | if (ret) | |
389 | return ret; | |
390 | return mp->mp_aheight - x - 1; | |
d552a2b9 BP |
391 | } |
392 | ||
a27a0c9b AG |
393 | static sector_t metapath_to_block(struct gfs2_sbd *sdp, struct metapath *mp) |
394 | { | |
395 | sector_t factor = 1, block = 0; | |
396 | int hgt; | |
397 | ||
398 | for (hgt = mp->mp_fheight - 1; hgt >= 0; hgt--) { | |
399 | if (hgt < mp->mp_aheight) | |
400 | block += mp->mp_list[hgt] * factor; | |
401 | factor *= sdp->sd_inptrs; | |
402 | } | |
403 | return block; | |
404 | } | |
405 | ||
64bc06bb | 406 | static void release_metapath(struct metapath *mp) |
dbac6710 SW |
407 | { |
408 | int i; | |
409 | ||
9b8c81d1 SW |
410 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
411 | if (mp->mp_bh[i] == NULL) | |
412 | break; | |
413 | brelse(mp->mp_bh[i]); | |
64bc06bb | 414 | mp->mp_bh[i] = NULL; |
9b8c81d1 | 415 | } |
11707ea0 SW |
416 | } |
417 | ||
30cbf189 SW |
418 | /** |
419 | * gfs2_extent_length - Returns length of an extent of blocks | |
bcfe9413 AG |
420 | * @bh: The metadata block |
421 | * @ptr: Current position in @bh | |
422 | * @limit: Max extent length to return | |
30cbf189 SW |
423 | * @eob: Set to 1 if we hit "end of block" |
424 | * | |
30cbf189 SW |
425 | * Returns: The length of the extent (minimum of one block) |
426 | */ | |
427 | ||
bcfe9413 | 428 | static inline unsigned int gfs2_extent_length(struct buffer_head *bh, __be64 *ptr, size_t limit, int *eob) |
30cbf189 | 429 | { |
bcfe9413 | 430 | const __be64 *end = (__be64 *)(bh->b_data + bh->b_size); |
30cbf189 SW |
431 | const __be64 *first = ptr; |
432 | u64 d = be64_to_cpu(*ptr); | |
433 | ||
434 | *eob = 0; | |
435 | do { | |
436 | ptr++; | |
437 | if (ptr >= end) | |
438 | break; | |
bcfe9413 | 439 | d++; |
30cbf189 SW |
440 | } while(be64_to_cpu(*ptr) == d); |
441 | if (ptr >= end) | |
442 | *eob = 1; | |
bcfe9413 | 443 | return ptr - first; |
30cbf189 SW |
444 | } |
445 | ||
a27a0c9b AG |
446 | enum walker_status { WALK_STOP, WALK_FOLLOW, WALK_CONTINUE }; |
447 | ||
448 | /* | |
449 | * gfs2_metadata_walker - walk an indirect block | |
450 | * @mp: Metapath to indirect block | |
451 | * @ptrs: Number of pointers to look at | |
452 | * | |
453 | * When returning WALK_FOLLOW, the walker must update @mp to point at the right | |
454 | * indirect block to follow. | |
455 | */ | |
456 | typedef enum walker_status (*gfs2_metadata_walker)(struct metapath *mp, | |
457 | unsigned int ptrs); | |
7841b9f0 | 458 | |
a27a0c9b AG |
459 | /* |
460 | * gfs2_walk_metadata - walk a tree of indirect blocks | |
461 | * @inode: The inode | |
462 | * @mp: Starting point of walk | |
463 | * @max_len: Maximum number of blocks to walk | |
464 | * @walker: Called during the walk | |
465 | * | |
466 | * Returns 1 if the walk was stopped by @walker, 0 if we went past @max_len or | |
467 | * past the end of metadata, and a negative error code otherwise. | |
468 | */ | |
7841b9f0 | 469 | |
a27a0c9b AG |
470 | static int gfs2_walk_metadata(struct inode *inode, struct metapath *mp, |
471 | u64 max_len, gfs2_metadata_walker walker) | |
7841b9f0 | 472 | { |
7841b9f0 AG |
473 | struct gfs2_inode *ip = GFS2_I(inode); |
474 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7841b9f0 AG |
475 | u64 factor = 1; |
476 | unsigned int hgt; | |
a27a0c9b | 477 | int ret; |
7841b9f0 | 478 | |
a27a0c9b AG |
479 | /* |
480 | * The walk starts in the lowest allocated indirect block, which may be | |
481 | * before the position indicated by @mp. Adjust @max_len accordingly | |
482 | * to avoid a short walk. | |
483 | */ | |
484 | for (hgt = mp->mp_fheight - 1; hgt >= mp->mp_aheight; hgt--) { | |
485 | max_len += mp->mp_list[hgt] * factor; | |
486 | mp->mp_list[hgt] = 0; | |
7841b9f0 | 487 | factor *= sdp->sd_inptrs; |
a27a0c9b | 488 | } |
7841b9f0 AG |
489 | |
490 | for (;;) { | |
a27a0c9b AG |
491 | u16 start = mp->mp_list[hgt]; |
492 | enum walker_status status; | |
493 | unsigned int ptrs; | |
494 | u64 len; | |
7841b9f0 AG |
495 | |
496 | /* Walk indirect block. */ | |
a27a0c9b AG |
497 | ptrs = (hgt >= 1 ? sdp->sd_inptrs : sdp->sd_diptrs) - start; |
498 | len = ptrs * factor; | |
499 | if (len > max_len) | |
500 | ptrs = DIV_ROUND_UP_ULL(max_len, factor); | |
501 | status = walker(mp, ptrs); | |
502 | switch (status) { | |
503 | case WALK_STOP: | |
504 | return 1; | |
505 | case WALK_FOLLOW: | |
506 | BUG_ON(mp->mp_aheight == mp->mp_fheight); | |
507 | ptrs = mp->mp_list[hgt] - start; | |
508 | len = ptrs * factor; | |
7841b9f0 | 509 | break; |
a27a0c9b | 510 | case WALK_CONTINUE: |
7841b9f0 | 511 | break; |
7841b9f0 | 512 | } |
a27a0c9b AG |
513 | if (len >= max_len) |
514 | break; | |
515 | max_len -= len; | |
516 | if (status == WALK_FOLLOW) | |
517 | goto fill_up_metapath; | |
7841b9f0 AG |
518 | |
519 | lower_metapath: | |
520 | /* Decrease height of metapath. */ | |
7841b9f0 AG |
521 | brelse(mp->mp_bh[hgt]); |
522 | mp->mp_bh[hgt] = NULL; | |
a27a0c9b | 523 | mp->mp_list[hgt] = 0; |
7841b9f0 AG |
524 | if (!hgt) |
525 | break; | |
526 | hgt--; | |
527 | factor *= sdp->sd_inptrs; | |
528 | ||
529 | /* Advance in metadata tree. */ | |
530 | (mp->mp_list[hgt])++; | |
566a2ab3 AG |
531 | if (hgt) { |
532 | if (mp->mp_list[hgt] >= sdp->sd_inptrs) | |
533 | goto lower_metapath; | |
534 | } else { | |
535 | if (mp->mp_list[hgt] >= sdp->sd_diptrs) | |
7841b9f0 | 536 | break; |
7841b9f0 AG |
537 | } |
538 | ||
539 | fill_up_metapath: | |
540 | /* Increase height of metapath. */ | |
7841b9f0 AG |
541 | ret = fillup_metapath(ip, mp, ip->i_height - 1); |
542 | if (ret < 0) | |
a27a0c9b | 543 | return ret; |
7841b9f0 AG |
544 | hgt += ret; |
545 | for (; ret; ret--) | |
546 | do_div(factor, sdp->sd_inptrs); | |
547 | mp->mp_aheight = hgt + 1; | |
548 | } | |
a27a0c9b | 549 | return 0; |
7841b9f0 AG |
550 | } |
551 | ||
a27a0c9b AG |
552 | static enum walker_status gfs2_hole_walker(struct metapath *mp, |
553 | unsigned int ptrs) | |
7841b9f0 | 554 | { |
a27a0c9b AG |
555 | const __be64 *start, *ptr, *end; |
556 | unsigned int hgt; | |
557 | ||
558 | hgt = mp->mp_aheight - 1; | |
559 | start = metapointer(hgt, mp); | |
560 | end = start + ptrs; | |
7841b9f0 AG |
561 | |
562 | for (ptr = start; ptr < end; ptr++) { | |
563 | if (*ptr) { | |
a27a0c9b | 564 | mp->mp_list[hgt] += ptr - start; |
7841b9f0 AG |
565 | if (mp->mp_aheight == mp->mp_fheight) |
566 | return WALK_STOP; | |
a27a0c9b | 567 | return WALK_FOLLOW; |
7841b9f0 AG |
568 | } |
569 | } | |
a27a0c9b | 570 | return WALK_CONTINUE; |
7841b9f0 AG |
571 | } |
572 | ||
573 | /** | |
574 | * gfs2_hole_size - figure out the size of a hole | |
575 | * @inode: The inode | |
576 | * @lblock: The logical starting block number | |
577 | * @len: How far to look (in blocks) | |
578 | * @mp: The metapath at lblock | |
579 | * @iomap: The iomap to store the hole size in | |
580 | * | |
581 | * This function modifies @mp. | |
582 | * | |
583 | * Returns: errno on error | |
584 | */ | |
585 | static int gfs2_hole_size(struct inode *inode, sector_t lblock, u64 len, | |
586 | struct metapath *mp, struct iomap *iomap) | |
587 | { | |
a27a0c9b AG |
588 | struct metapath clone; |
589 | u64 hole_size; | |
590 | int ret; | |
591 | ||
592 | clone_metapath(&clone, mp); | |
593 | ret = gfs2_walk_metadata(inode, &clone, len, gfs2_hole_walker); | |
594 | if (ret < 0) | |
595 | goto out; | |
7841b9f0 | 596 | |
a27a0c9b AG |
597 | if (ret == 1) |
598 | hole_size = metapath_to_block(GFS2_SB(inode), &clone) - lblock; | |
599 | else | |
600 | hole_size = len; | |
601 | iomap->length = hole_size << inode->i_blkbits; | |
602 | ret = 0; | |
603 | ||
604 | out: | |
605 | release_metapath(&clone); | |
7841b9f0 AG |
606 | return ret; |
607 | } | |
608 | ||
9b8c81d1 SW |
609 | static inline __be64 *gfs2_indirect_init(struct metapath *mp, |
610 | struct gfs2_glock *gl, unsigned int i, | |
611 | unsigned offset, u64 bn) | |
612 | { | |
613 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
614 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
615 | sizeof(struct gfs2_dinode))); | |
616 | BUG_ON(i < 1); | |
617 | BUG_ON(mp->mp_bh[i] != NULL); | |
618 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
350a9b0a | 619 | gfs2_trans_add_meta(gl, mp->mp_bh[i]); |
9b8c81d1 SW |
620 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
621 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
622 | ptr += offset; | |
623 | *ptr = cpu_to_be64(bn); | |
624 | return ptr; | |
625 | } | |
626 | ||
627 | enum alloc_state { | |
628 | ALLOC_DATA = 0, | |
629 | ALLOC_GROW_DEPTH = 1, | |
630 | ALLOC_GROW_HEIGHT = 2, | |
631 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
632 | }; | |
633 | ||
634 | /** | |
54992257 | 635 | * __gfs2_iomap_alloc - Build a metadata tree of the requested height |
9b8c81d1 | 636 | * @inode: The GFS2 inode |
628e366d | 637 | * @iomap: The iomap structure |
5f8bd444 | 638 | * @mp: The metapath, with proper height information calculated |
9b8c81d1 SW |
639 | * |
640 | * In this routine we may have to alloc: | |
641 | * i) Indirect blocks to grow the metadata tree height | |
642 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
643 | * iii) Data blocks | |
644 | * | |
54992257 | 645 | * This function is called after __gfs2_iomap_get, which works out the |
64bc06bb AG |
646 | * total number of blocks which we need via gfs2_alloc_size. |
647 | * | |
648 | * We then do the actual allocation asking for an extent at a time (if | |
649 | * enough contiguous free blocks are available, there will only be one | |
650 | * allocation request per call) and uses the state machine to initialise | |
651 | * the blocks in order. | |
9b8c81d1 | 652 | * |
628e366d AG |
653 | * Right now, this function will allocate at most one indirect block |
654 | * worth of data -- with a default block size of 4K, that's slightly | |
655 | * less than 2M. If this limitation is ever removed to allow huge | |
656 | * allocations, we would probably still want to limit the iomap size we | |
657 | * return to avoid stalling other tasks during huge writes; the next | |
658 | * iomap iteration would then find the blocks already allocated. | |
659 | * | |
9b8c81d1 SW |
660 | * Returns: errno on error |
661 | */ | |
662 | ||
54992257 AG |
663 | static int __gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap, |
664 | struct metapath *mp) | |
9b8c81d1 SW |
665 | { |
666 | struct gfs2_inode *ip = GFS2_I(inode); | |
667 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
668 | struct buffer_head *dibh = mp->mp_bh[0]; | |
5f8bd444 | 669 | u64 bn; |
5af4e7a0 | 670 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
64bc06bb | 671 | size_t dblks = iomap->length >> inode->i_blkbits; |
5f8bd444 | 672 | const unsigned end_of_metadata = mp->mp_fheight - 1; |
628e366d | 673 | int ret; |
9b8c81d1 SW |
674 | enum alloc_state state; |
675 | __be64 *ptr; | |
676 | __be64 zero_bn = 0; | |
677 | ||
5f8bd444 | 678 | BUG_ON(mp->mp_aheight < 1); |
9b8c81d1 | 679 | BUG_ON(dibh == NULL); |
64bc06bb | 680 | BUG_ON(dblks < 1); |
9b8c81d1 | 681 | |
350a9b0a | 682 | gfs2_trans_add_meta(ip->i_gl, dibh); |
9b8c81d1 | 683 | |
628e366d AG |
684 | down_write(&ip->i_rw_mutex); |
685 | ||
5f8bd444 | 686 | if (mp->mp_fheight == mp->mp_aheight) { |
64bc06bb | 687 | /* Bottom indirect block exists */ |
9b8c81d1 SW |
688 | state = ALLOC_DATA; |
689 | } else { | |
690 | /* Need to allocate indirect blocks */ | |
5f8bd444 | 691 | if (mp->mp_fheight == ip->i_height) { |
9b8c81d1 | 692 | /* Writing into existing tree, extend tree down */ |
5f8bd444 | 693 | iblks = mp->mp_fheight - mp->mp_aheight; |
9b8c81d1 SW |
694 | state = ALLOC_GROW_DEPTH; |
695 | } else { | |
696 | /* Building up tree height */ | |
697 | state = ALLOC_GROW_HEIGHT; | |
5f8bd444 | 698 | iblks = mp->mp_fheight - ip->i_height; |
5af4e7a0 | 699 | branch_start = metapath_branch_start(mp); |
5f8bd444 | 700 | iblks += (mp->mp_fheight - branch_start); |
9b8c81d1 SW |
701 | } |
702 | } | |
703 | ||
704 | /* start of the second part of the function (state machine) */ | |
705 | ||
3974320c | 706 | blks = dblks + iblks; |
5f8bd444 | 707 | i = mp->mp_aheight; |
9b8c81d1 SW |
708 | do { |
709 | n = blks - alloced; | |
628e366d AG |
710 | ret = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
711 | if (ret) | |
712 | goto out; | |
9b8c81d1 SW |
713 | alloced += n; |
714 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
fbb27873 | 715 | gfs2_trans_remove_revoke(sdp, bn, n); |
9b8c81d1 SW |
716 | switch (state) { |
717 | /* Growing height of tree */ | |
718 | case ALLOC_GROW_HEIGHT: | |
719 | if (i == 1) { | |
720 | ptr = (__be64 *)(dibh->b_data + | |
721 | sizeof(struct gfs2_dinode)); | |
722 | zero_bn = *ptr; | |
723 | } | |
5f8bd444 BP |
724 | for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0; |
725 | i++, n--) | |
9b8c81d1 | 726 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); |
5f8bd444 | 727 | if (i - 1 == mp->mp_fheight - ip->i_height) { |
9b8c81d1 SW |
728 | i--; |
729 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
730 | sizeof(struct gfs2_meta_header), | |
731 | dibh, sizeof(struct gfs2_dinode)); | |
732 | gfs2_buffer_clear_tail(dibh, | |
733 | sizeof(struct gfs2_dinode) + | |
734 | sizeof(__be64)); | |
735 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
736 | sizeof(struct gfs2_meta_header)); | |
737 | *ptr = zero_bn; | |
738 | state = ALLOC_GROW_DEPTH; | |
5f8bd444 | 739 | for(i = branch_start; i < mp->mp_fheight; i++) { |
9b8c81d1 SW |
740 | if (mp->mp_bh[i] == NULL) |
741 | break; | |
742 | brelse(mp->mp_bh[i]); | |
743 | mp->mp_bh[i] = NULL; | |
744 | } | |
5af4e7a0 | 745 | i = branch_start; |
9b8c81d1 SW |
746 | } |
747 | if (n == 0) | |
748 | break; | |
df561f66 | 749 | fallthrough; /* To branching from existing tree */ |
9b8c81d1 | 750 | case ALLOC_GROW_DEPTH: |
5f8bd444 | 751 | if (i > 1 && i < mp->mp_fheight) |
350a9b0a | 752 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]); |
5f8bd444 | 753 | for (; i < mp->mp_fheight && n > 0; i++, n--) |
9b8c81d1 SW |
754 | gfs2_indirect_init(mp, ip->i_gl, i, |
755 | mp->mp_list[i-1], bn++); | |
5f8bd444 | 756 | if (i == mp->mp_fheight) |
9b8c81d1 SW |
757 | state = ALLOC_DATA; |
758 | if (n == 0) | |
759 | break; | |
df561f66 | 760 | fallthrough; /* To tree complete, adding data blocks */ |
9b8c81d1 | 761 | case ALLOC_DATA: |
3974320c | 762 | BUG_ON(n > dblks); |
9b8c81d1 | 763 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); |
350a9b0a | 764 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]); |
3974320c | 765 | dblks = n; |
9b8c81d1 | 766 | ptr = metapointer(end_of_metadata, mp); |
3974320c | 767 | iomap->addr = bn << inode->i_blkbits; |
628e366d | 768 | iomap->flags |= IOMAP_F_MERGED | IOMAP_F_NEW; |
9b8c81d1 SW |
769 | while (n-- > 0) |
770 | *ptr++ = cpu_to_be64(bn++); | |
771 | break; | |
772 | } | |
3974320c | 773 | } while (iomap->addr == IOMAP_NULL_ADDR); |
9b8c81d1 | 774 | |
d505a96a | 775 | iomap->type = IOMAP_MAPPED; |
3974320c | 776 | iomap->length = (u64)dblks << inode->i_blkbits; |
5f8bd444 | 777 | ip->i_height = mp->mp_fheight; |
9b8c81d1 | 778 | gfs2_add_inode_blocks(&ip->i_inode, alloced); |
628e366d AG |
779 | gfs2_dinode_out(ip, dibh->b_data); |
780 | out: | |
781 | up_write(&ip->i_rw_mutex); | |
782 | return ret; | |
9b8c81d1 SW |
783 | } |
784 | ||
7ee66c03 CH |
785 | #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE |
786 | ||
64bc06bb AG |
787 | /** |
788 | * gfs2_alloc_size - Compute the maximum allocation size | |
789 | * @inode: The inode | |
790 | * @mp: The metapath | |
791 | * @size: Requested size in blocks | |
792 | * | |
793 | * Compute the maximum size of the next allocation at @mp. | |
794 | * | |
795 | * Returns: size in blocks | |
796 | */ | |
797 | static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size) | |
3974320c BP |
798 | { |
799 | struct gfs2_inode *ip = GFS2_I(inode); | |
64bc06bb AG |
800 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
801 | const __be64 *first, *ptr, *end; | |
802 | ||
803 | /* | |
804 | * For writes to stuffed files, this function is called twice via | |
54992257 | 805 | * __gfs2_iomap_get, before and after unstuffing. The size we return the |
64bc06bb AG |
806 | * first time needs to be large enough to get the reservation and |
807 | * allocation sizes right. The size we return the second time must | |
54992257 | 808 | * be exact or else __gfs2_iomap_alloc won't do the right thing. |
64bc06bb AG |
809 | */ |
810 | ||
811 | if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) { | |
812 | unsigned int maxsize = mp->mp_fheight > 1 ? | |
813 | sdp->sd_inptrs : sdp->sd_diptrs; | |
814 | maxsize -= mp->mp_list[mp->mp_fheight - 1]; | |
815 | if (size > maxsize) | |
816 | size = maxsize; | |
817 | return size; | |
818 | } | |
3974320c | 819 | |
64bc06bb AG |
820 | first = metapointer(ip->i_height - 1, mp); |
821 | end = metaend(ip->i_height - 1, mp); | |
822 | if (end - first > size) | |
823 | end = first + size; | |
824 | for (ptr = first; ptr < end; ptr++) { | |
825 | if (*ptr) | |
826 | break; | |
827 | } | |
828 | return ptr - first; | |
3974320c BP |
829 | } |
830 | ||
831 | /** | |
54992257 | 832 | * __gfs2_iomap_get - Map blocks from an inode to disk blocks |
3974320c BP |
833 | * @inode: The inode |
834 | * @pos: Starting position in bytes | |
835 | * @length: Length to map, in bytes | |
836 | * @flags: iomap flags | |
837 | * @iomap: The iomap structure | |
628e366d | 838 | * @mp: The metapath |
3974320c BP |
839 | * |
840 | * Returns: errno | |
841 | */ | |
54992257 AG |
842 | static int __gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length, |
843 | unsigned flags, struct iomap *iomap, | |
844 | struct metapath *mp) | |
b3b94faa | 845 | { |
feaa7bba SW |
846 | struct gfs2_inode *ip = GFS2_I(inode); |
847 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
d505a96a | 848 | loff_t size = i_size_read(inode); |
9b8c81d1 | 849 | __be64 *ptr; |
3974320c | 850 | sector_t lblock; |
628e366d AG |
851 | sector_t lblock_stop; |
852 | int ret; | |
9b8c81d1 | 853 | int eob; |
628e366d | 854 | u64 len; |
d505a96a | 855 | struct buffer_head *dibh = NULL, *bh; |
9b8c81d1 | 856 | u8 height; |
7276b3b0 | 857 | |
628e366d AG |
858 | if (!length) |
859 | return -EINVAL; | |
b3b94faa | 860 | |
d505a96a AG |
861 | down_read(&ip->i_rw_mutex); |
862 | ||
863 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
864 | if (ret) | |
865 | goto unlock; | |
c26b5aa8 | 866 | mp->mp_bh[0] = dibh; |
d505a96a | 867 | |
49edd5bf | 868 | if (gfs2_is_stuffed(ip)) { |
d505a96a AG |
869 | if (flags & IOMAP_WRITE) { |
870 | loff_t max_size = gfs2_max_stuffed_size(ip); | |
871 | ||
872 | if (pos + length > max_size) | |
873 | goto unstuff; | |
874 | iomap->length = max_size; | |
875 | } else { | |
876 | if (pos >= size) { | |
877 | if (flags & IOMAP_REPORT) { | |
878 | ret = -ENOENT; | |
879 | goto unlock; | |
880 | } else { | |
d505a96a AG |
881 | iomap->offset = pos; |
882 | iomap->length = length; | |
566a2ab3 | 883 | goto hole_found; |
d505a96a AG |
884 | } |
885 | } | |
886 | iomap->length = size; | |
49edd5bf | 887 | } |
d505a96a AG |
888 | iomap->addr = (ip->i_no_addr << inode->i_blkbits) + |
889 | sizeof(struct gfs2_dinode); | |
890 | iomap->type = IOMAP_INLINE; | |
64bc06bb | 891 | iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode); |
d505a96a | 892 | goto out; |
3974320c | 893 | } |
d505a96a AG |
894 | |
895 | unstuff: | |
3974320c | 896 | lblock = pos >> inode->i_blkbits; |
3974320c | 897 | iomap->offset = lblock << inode->i_blkbits; |
628e366d AG |
898 | lblock_stop = (pos + length - 1) >> inode->i_blkbits; |
899 | len = lblock_stop - lblock + 1; | |
d505a96a | 900 | iomap->length = len << inode->i_blkbits; |
628e366d | 901 | |
9b8c81d1 | 902 | height = ip->i_height; |
9a38662b | 903 | while ((lblock + 1) * sdp->sd_sb.sb_bsize > sdp->sd_heightsize[height]) |
9b8c81d1 | 904 | height++; |
628e366d | 905 | find_metapath(sdp, lblock, mp, height); |
9b8c81d1 SW |
906 | if (height > ip->i_height || gfs2_is_stuffed(ip)) |
907 | goto do_alloc; | |
3974320c | 908 | |
628e366d | 909 | ret = lookup_metapath(ip, mp); |
e8b43fe0 | 910 | if (ret) |
628e366d | 911 | goto unlock; |
3974320c | 912 | |
628e366d | 913 | if (mp->mp_aheight != ip->i_height) |
9b8c81d1 | 914 | goto do_alloc; |
3974320c | 915 | |
628e366d | 916 | ptr = metapointer(ip->i_height - 1, mp); |
9b8c81d1 SW |
917 | if (*ptr == 0) |
918 | goto do_alloc; | |
3974320c | 919 | |
628e366d | 920 | bh = mp->mp_bh[ip->i_height - 1]; |
bcfe9413 | 921 | len = gfs2_extent_length(bh, ptr, len, &eob); |
3974320c | 922 | |
628e366d AG |
923 | iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits; |
924 | iomap->length = len << inode->i_blkbits; | |
925 | iomap->type = IOMAP_MAPPED; | |
0ed91eca | 926 | iomap->flags |= IOMAP_F_MERGED; |
9b8c81d1 | 927 | if (eob) |
7ee66c03 | 928 | iomap->flags |= IOMAP_F_GFS2_BOUNDARY; |
3974320c | 929 | |
3974320c | 930 | out: |
628e366d AG |
931 | iomap->bdev = inode->i_sb->s_bdev; |
932 | unlock: | |
933 | up_read(&ip->i_rw_mutex); | |
9b8c81d1 | 934 | return ret; |
30cbf189 | 935 | |
9b8c81d1 | 936 | do_alloc: |
628e366d | 937 | if (flags & IOMAP_REPORT) { |
49edd5bf | 938 | if (pos >= size) |
3974320c | 939 | ret = -ENOENT; |
628e366d AG |
940 | else if (height == ip->i_height) |
941 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
49edd5bf AG |
942 | else |
943 | iomap->length = size - pos; | |
64bc06bb AG |
944 | } else if (flags & IOMAP_WRITE) { |
945 | u64 alloc_size; | |
946 | ||
967bcc91 AG |
947 | if (flags & IOMAP_DIRECT) |
948 | goto out; /* (see gfs2_file_direct_write) */ | |
949 | ||
64bc06bb AG |
950 | len = gfs2_alloc_size(inode, mp, len); |
951 | alloc_size = len << inode->i_blkbits; | |
952 | if (alloc_size < iomap->length) | |
953 | iomap->length = alloc_size; | |
954 | } else { | |
d505a96a AG |
955 | if (pos < size && height == ip->i_height) |
956 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
b3b94faa | 957 | } |
566a2ab3 AG |
958 | hole_found: |
959 | iomap->addr = IOMAP_NULL_ADDR; | |
960 | iomap->type = IOMAP_HOLE; | |
628e366d | 961 | goto out; |
3974320c BP |
962 | } |
963 | ||
64bc06bb AG |
964 | static int gfs2_write_lock(struct inode *inode) |
965 | { | |
966 | struct gfs2_inode *ip = GFS2_I(inode); | |
967 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
968 | int error; | |
969 | ||
970 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); | |
971 | error = gfs2_glock_nq(&ip->i_gh); | |
972 | if (error) | |
973 | goto out_uninit; | |
974 | if (&ip->i_inode == sdp->sd_rindex) { | |
975 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
976 | ||
977 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
978 | GL_NOCACHE, &m_ip->i_gh); | |
979 | if (error) | |
980 | goto out_unlock; | |
981 | } | |
982 | return 0; | |
983 | ||
984 | out_unlock: | |
985 | gfs2_glock_dq(&ip->i_gh); | |
986 | out_uninit: | |
987 | gfs2_holder_uninit(&ip->i_gh); | |
988 | return error; | |
989 | } | |
990 | ||
991 | static void gfs2_write_unlock(struct inode *inode) | |
992 | { | |
993 | struct gfs2_inode *ip = GFS2_I(inode); | |
994 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
995 | ||
996 | if (&ip->i_inode == sdp->sd_rindex) { | |
997 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
998 | ||
999 | gfs2_glock_dq_uninit(&m_ip->i_gh); | |
1000 | } | |
1001 | gfs2_glock_dq_uninit(&ip->i_gh); | |
1002 | } | |
1003 | ||
d0a22a4b AG |
1004 | static int gfs2_iomap_page_prepare(struct inode *inode, loff_t pos, |
1005 | unsigned len, struct iomap *iomap) | |
1006 | { | |
2741b672 | 1007 | unsigned int blockmask = i_blocksize(inode) - 1; |
d0a22a4b | 1008 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
2741b672 | 1009 | unsigned int blocks; |
d0a22a4b | 1010 | |
2741b672 AG |
1011 | blocks = ((pos & blockmask) + len + blockmask) >> inode->i_blkbits; |
1012 | return gfs2_trans_begin(sdp, RES_DINODE + blocks, 0); | |
d0a22a4b AG |
1013 | } |
1014 | ||
df0db3ec AG |
1015 | static void gfs2_iomap_page_done(struct inode *inode, loff_t pos, |
1016 | unsigned copied, struct page *page, | |
1017 | struct iomap *iomap) | |
64bc06bb | 1018 | { |
706cb549 | 1019 | struct gfs2_trans *tr = current->journal_info; |
64bc06bb | 1020 | struct gfs2_inode *ip = GFS2_I(inode); |
d0a22a4b | 1021 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
64bc06bb | 1022 | |
d0a22a4b | 1023 | if (page && !gfs2_is_stuffed(ip)) |
df0db3ec | 1024 | gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied); |
706cb549 AG |
1025 | |
1026 | if (tr->tr_num_buf_new) | |
1027 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1028 | ||
d0a22a4b | 1029 | gfs2_trans_end(sdp); |
64bc06bb AG |
1030 | } |
1031 | ||
df0db3ec | 1032 | static const struct iomap_page_ops gfs2_iomap_page_ops = { |
d0a22a4b | 1033 | .page_prepare = gfs2_iomap_page_prepare, |
df0db3ec AG |
1034 | .page_done = gfs2_iomap_page_done, |
1035 | }; | |
1036 | ||
64bc06bb AG |
1037 | static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, |
1038 | loff_t length, unsigned flags, | |
c26b5aa8 AG |
1039 | struct iomap *iomap, |
1040 | struct metapath *mp) | |
64bc06bb | 1041 | { |
64bc06bb AG |
1042 | struct gfs2_inode *ip = GFS2_I(inode); |
1043 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
34aad20b | 1044 | bool unstuff; |
64bc06bb AG |
1045 | int ret; |
1046 | ||
64bc06bb AG |
1047 | unstuff = gfs2_is_stuffed(ip) && |
1048 | pos + length > gfs2_max_stuffed_size(ip); | |
1049 | ||
34aad20b AG |
1050 | if (unstuff || iomap->type == IOMAP_HOLE) { |
1051 | unsigned int data_blocks, ind_blocks; | |
1052 | struct gfs2_alloc_parms ap = {}; | |
1053 | unsigned int rblocks; | |
1054 | struct gfs2_trans *tr; | |
64bc06bb | 1055 | |
64bc06bb AG |
1056 | gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, |
1057 | &ind_blocks); | |
34aad20b | 1058 | ap.target = data_blocks + ind_blocks; |
64bc06bb AG |
1059 | ret = gfs2_quota_lock_check(ip, &ap); |
1060 | if (ret) | |
34aad20b | 1061 | return ret; |
64bc06bb AG |
1062 | |
1063 | ret = gfs2_inplace_reserve(ip, &ap); | |
1064 | if (ret) | |
1065 | goto out_qunlock; | |
64bc06bb | 1066 | |
34aad20b AG |
1067 | rblocks = RES_DINODE + ind_blocks; |
1068 | if (gfs2_is_jdata(ip)) | |
1069 | rblocks += data_blocks; | |
1070 | if (ind_blocks || data_blocks) | |
1071 | rblocks += RES_STATFS + RES_QUOTA; | |
1072 | if (inode == sdp->sd_rindex) | |
1073 | rblocks += 2 * RES_STATFS; | |
64bc06bb AG |
1074 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
1075 | ||
d0a22a4b AG |
1076 | ret = gfs2_trans_begin(sdp, rblocks, |
1077 | iomap->length >> inode->i_blkbits); | |
64bc06bb | 1078 | if (ret) |
d0a22a4b AG |
1079 | goto out_trans_fail; |
1080 | ||
1081 | if (unstuff) { | |
7a607a41 | 1082 | ret = gfs2_unstuff_dinode(ip); |
d0a22a4b AG |
1083 | if (ret) |
1084 | goto out_trans_end; | |
1085 | release_metapath(mp); | |
54992257 AG |
1086 | ret = __gfs2_iomap_get(inode, iomap->offset, |
1087 | iomap->length, flags, iomap, mp); | |
d0a22a4b AG |
1088 | if (ret) |
1089 | goto out_trans_end; | |
1090 | } | |
64bc06bb | 1091 | |
d0a22a4b | 1092 | if (iomap->type == IOMAP_HOLE) { |
54992257 | 1093 | ret = __gfs2_iomap_alloc(inode, iomap, mp); |
d0a22a4b AG |
1094 | if (ret) { |
1095 | gfs2_trans_end(sdp); | |
1096 | gfs2_inplace_release(ip); | |
1097 | punch_hole(ip, iomap->offset, iomap->length); | |
1098 | goto out_qunlock; | |
1099 | } | |
64bc06bb | 1100 | } |
d0a22a4b AG |
1101 | |
1102 | tr = current->journal_info; | |
1103 | if (tr->tr_num_buf_new) | |
1104 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
d0a22a4b AG |
1105 | |
1106 | gfs2_trans_end(sdp); | |
64bc06bb | 1107 | } |
d0a22a4b AG |
1108 | |
1109 | if (gfs2_is_stuffed(ip) || gfs2_is_jdata(ip)) | |
df0db3ec | 1110 | iomap->page_ops = &gfs2_iomap_page_ops; |
64bc06bb AG |
1111 | return 0; |
1112 | ||
1113 | out_trans_end: | |
1114 | gfs2_trans_end(sdp); | |
1115 | out_trans_fail: | |
34aad20b | 1116 | gfs2_inplace_release(ip); |
64bc06bb | 1117 | out_qunlock: |
34aad20b | 1118 | gfs2_quota_unlock(ip); |
64bc06bb AG |
1119 | return ret; |
1120 | } | |
1121 | ||
34aad20b AG |
1122 | static inline bool gfs2_iomap_need_write_lock(unsigned flags) |
1123 | { | |
1124 | return (flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT); | |
1125 | } | |
1126 | ||
628e366d | 1127 | static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, |
c039b997 GR |
1128 | unsigned flags, struct iomap *iomap, |
1129 | struct iomap *srcmap) | |
628e366d AG |
1130 | { |
1131 | struct gfs2_inode *ip = GFS2_I(inode); | |
1132 | struct metapath mp = { .mp_aheight = 1, }; | |
1133 | int ret; | |
1134 | ||
2164f9b9 CH |
1135 | if (gfs2_is_jdata(ip)) |
1136 | iomap->flags |= IOMAP_F_BUFFER_HEAD; | |
0ed91eca | 1137 | |
628e366d | 1138 | trace_gfs2_iomap_start(ip, pos, length, flags); |
34aad20b AG |
1139 | if (gfs2_iomap_need_write_lock(flags)) { |
1140 | ret = gfs2_write_lock(inode); | |
1141 | if (ret) | |
1142 | goto out; | |
628e366d | 1143 | } |
34aad20b | 1144 | |
54992257 | 1145 | ret = __gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); |
34aad20b AG |
1146 | if (ret) |
1147 | goto out_unlock; | |
1148 | ||
72d36d05 | 1149 | switch(flags & (IOMAP_WRITE | IOMAP_ZERO)) { |
34aad20b AG |
1150 | case IOMAP_WRITE: |
1151 | if (flags & IOMAP_DIRECT) { | |
1152 | /* | |
1153 | * Silently fall back to buffered I/O for stuffed files | |
1154 | * or if we've got a hole (see gfs2_file_direct_write). | |
1155 | */ | |
1156 | if (iomap->type != IOMAP_MAPPED) | |
1157 | ret = -ENOTBLK; | |
1158 | goto out_unlock; | |
1159 | } | |
1160 | break; | |
72d36d05 AG |
1161 | case IOMAP_ZERO: |
1162 | if (iomap->type == IOMAP_HOLE) | |
1163 | goto out_unlock; | |
1164 | break; | |
34aad20b AG |
1165 | default: |
1166 | goto out_unlock; | |
1167 | } | |
1168 | ||
1169 | ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap, &mp); | |
1170 | ||
1171 | out_unlock: | |
1172 | if (ret && gfs2_iomap_need_write_lock(flags)) | |
1173 | gfs2_write_unlock(inode); | |
c26b5aa8 | 1174 | release_metapath(&mp); |
34aad20b | 1175 | out: |
628e366d AG |
1176 | trace_gfs2_iomap_end(ip, iomap, ret); |
1177 | return ret; | |
1178 | } | |
1179 | ||
64bc06bb AG |
1180 | static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length, |
1181 | ssize_t written, unsigned flags, struct iomap *iomap) | |
1182 | { | |
1183 | struct gfs2_inode *ip = GFS2_I(inode); | |
1184 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
64bc06bb | 1185 | |
72d36d05 | 1186 | switch (flags & (IOMAP_WRITE | IOMAP_ZERO)) { |
34aad20b AG |
1187 | case IOMAP_WRITE: |
1188 | if (flags & IOMAP_DIRECT) | |
1189 | return 0; | |
1190 | break; | |
72d36d05 AG |
1191 | case IOMAP_ZERO: |
1192 | if (iomap->type == IOMAP_HOLE) | |
1193 | return 0; | |
1194 | break; | |
34aad20b AG |
1195 | default: |
1196 | return 0; | |
1197 | } | |
64bc06bb | 1198 | |
d0a22a4b | 1199 | if (!gfs2_is_stuffed(ip)) |
64bc06bb AG |
1200 | gfs2_ordered_add_inode(ip); |
1201 | ||
d0a22a4b | 1202 | if (inode == sdp->sd_rindex) |
64bc06bb | 1203 | adjust_fs_space(inode); |
64bc06bb | 1204 | |
64bc06bb AG |
1205 | gfs2_inplace_release(ip); |
1206 | ||
7009fa9c AG |
1207 | if (ip->i_qadata && ip->i_qadata->qa_qd_num) |
1208 | gfs2_quota_unlock(ip); | |
1209 | ||
64bc06bb AG |
1210 | if (length != written && (iomap->flags & IOMAP_F_NEW)) { |
1211 | /* Deallocate blocks that were just allocated. */ | |
1212 | loff_t blockmask = i_blocksize(inode) - 1; | |
1213 | loff_t end = (pos + length) & ~blockmask; | |
1214 | ||
1215 | pos = (pos + written + blockmask) & ~blockmask; | |
1216 | if (pos < end) { | |
1217 | truncate_pagecache_range(inode, pos, end - 1); | |
1218 | punch_hole(ip, pos, end - pos); | |
1219 | } | |
1220 | } | |
1221 | ||
706cb549 AG |
1222 | if (unlikely(!written)) |
1223 | goto out_unlock; | |
1224 | ||
8d3e72a1 AG |
1225 | if (iomap->flags & IOMAP_F_SIZE_CHANGED) |
1226 | mark_inode_dirty(inode); | |
706cb549 | 1227 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
64bc06bb | 1228 | |
706cb549 | 1229 | out_unlock: |
34aad20b AG |
1230 | if (gfs2_iomap_need_write_lock(flags)) |
1231 | gfs2_write_unlock(inode); | |
64bc06bb AG |
1232 | return 0; |
1233 | } | |
1234 | ||
628e366d AG |
1235 | const struct iomap_ops gfs2_iomap_ops = { |
1236 | .iomap_begin = gfs2_iomap_begin, | |
64bc06bb | 1237 | .iomap_end = gfs2_iomap_end, |
628e366d AG |
1238 | }; |
1239 | ||
3974320c | 1240 | /** |
d39d18e0 | 1241 | * gfs2_block_map - Map one or more blocks of an inode to a disk block |
3974320c BP |
1242 | * @inode: The inode |
1243 | * @lblock: The logical block number | |
1244 | * @bh_map: The bh to be mapped | |
1245 | * @create: True if its ok to alloc blocks to satify the request | |
1246 | * | |
d39d18e0 AG |
1247 | * The size of the requested mapping is defined in bh_map->b_size. |
1248 | * | |
1249 | * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged | |
1250 | * when @lblock is not mapped. Sets buffer_mapped(bh_map) and | |
1251 | * bh_map->b_size to indicate the size of the mapping when @lblock and | |
1252 | * successive blocks are mapped, up to the requested size. | |
1253 | * | |
1254 | * Sets buffer_boundary() if a read of metadata will be required | |
1255 | * before the next block can be mapped. Sets buffer_new() if new | |
1256 | * blocks were allocated. | |
3974320c BP |
1257 | * |
1258 | * Returns: errno | |
1259 | */ | |
1260 | ||
1261 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
1262 | struct buffer_head *bh_map, int create) | |
1263 | { | |
1264 | struct gfs2_inode *ip = GFS2_I(inode); | |
628e366d AG |
1265 | loff_t pos = (loff_t)lblock << inode->i_blkbits; |
1266 | loff_t length = bh_map->b_size; | |
628e366d AG |
1267 | struct iomap iomap = { }; |
1268 | int ret; | |
3974320c BP |
1269 | |
1270 | clear_buffer_mapped(bh_map); | |
1271 | clear_buffer_new(bh_map); | |
1272 | clear_buffer_boundary(bh_map); | |
1273 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
1274 | ||
54992257 AG |
1275 | if (!create) |
1276 | ret = gfs2_iomap_get(inode, pos, length, &iomap); | |
1277 | else | |
1278 | ret = gfs2_iomap_alloc(inode, pos, length, &iomap); | |
628e366d AG |
1279 | if (ret) |
1280 | goto out; | |
3974320c BP |
1281 | |
1282 | if (iomap.length > bh_map->b_size) { | |
1283 | iomap.length = bh_map->b_size; | |
7ee66c03 | 1284 | iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY; |
5f8bd444 | 1285 | } |
3974320c BP |
1286 | if (iomap.addr != IOMAP_NULL_ADDR) |
1287 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
1288 | bh_map->b_size = iomap.length; | |
7ee66c03 | 1289 | if (iomap.flags & IOMAP_F_GFS2_BOUNDARY) |
3974320c BP |
1290 | set_buffer_boundary(bh_map); |
1291 | if (iomap.flags & IOMAP_F_NEW) | |
1292 | set_buffer_new(bh_map); | |
1293 | ||
1294 | out: | |
1295 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
1296 | return ret; | |
fd88de56 SW |
1297 | } |
1298 | ||
9153dac1 AG |
1299 | int gfs2_get_extent(struct inode *inode, u64 lblock, u64 *dblock, |
1300 | unsigned int *extlen) | |
fd88de56 | 1301 | { |
9153dac1 AG |
1302 | unsigned int blkbits = inode->i_blkbits; |
1303 | struct iomap iomap = { }; | |
1304 | unsigned int len; | |
7a6bbacb | 1305 | int ret; |
9153dac1 AG |
1306 | |
1307 | ret = gfs2_iomap_get(inode, lblock << blkbits, *extlen << blkbits, | |
1308 | &iomap); | |
1309 | if (ret) | |
1310 | return ret; | |
1311 | if (iomap.type != IOMAP_MAPPED) | |
1312 | return -EIO; | |
1313 | *dblock = iomap.addr >> blkbits; | |
1314 | len = iomap.length >> blkbits; | |
1315 | if (len < *extlen) | |
1316 | *extlen = len; | |
1317 | return 0; | |
1318 | } | |
1319 | ||
1320 | int gfs2_alloc_extent(struct inode *inode, u64 lblock, u64 *dblock, | |
1321 | unsigned int *extlen, bool *new) | |
1322 | { | |
1323 | unsigned int blkbits = inode->i_blkbits; | |
1324 | struct iomap iomap = { }; | |
1325 | unsigned int len; | |
1326 | int ret; | |
1327 | ||
1328 | ret = gfs2_iomap_alloc(inode, lblock << blkbits, *extlen << blkbits, | |
1329 | &iomap); | |
1330 | if (ret) | |
1331 | return ret; | |
1332 | if (iomap.type != IOMAP_MAPPED) | |
1333 | return -EIO; | |
1334 | *dblock = iomap.addr >> blkbits; | |
1335 | len = iomap.length >> blkbits; | |
1336 | if (len < *extlen) | |
1337 | *extlen = len; | |
1338 | *new = iomap.flags & IOMAP_F_NEW; | |
1339 | return 0; | |
b3b94faa DT |
1340 | } |
1341 | ||
70499cdf BP |
1342 | /* |
1343 | * NOTE: Never call gfs2_block_zero_range with an open transaction because it | |
1344 | * uses iomap write to perform its actions, which begin their own transactions | |
1345 | * (iomap_begin, page_prepare, etc.) | |
1346 | */ | |
bdba0d5e AG |
1347 | static int gfs2_block_zero_range(struct inode *inode, loff_t from, |
1348 | unsigned int length) | |
ba7f7290 | 1349 | { |
70499cdf | 1350 | BUG_ON(current->journal_info); |
2257e468 | 1351 | return iomap_zero_range(inode, from, length, NULL, &gfs2_iomap_ops); |
ba7f7290 SW |
1352 | } |
1353 | ||
c62baf65 FF |
1354 | #define GFS2_JTRUNC_REVOKES 8192 |
1355 | ||
fa731fc4 SW |
1356 | /** |
1357 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
1358 | * @inode: The inode being truncated | |
1359 | * @oldsize: The original (larger) size | |
1360 | * @newsize: The new smaller size | |
1361 | * | |
1362 | * With jdata files, we have to journal a revoke for each block which is | |
1363 | * truncated. As a result, we need to split this into separate transactions | |
1364 | * if the number of pages being truncated gets too large. | |
1365 | */ | |
1366 | ||
fa731fc4 SW |
1367 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
1368 | { | |
1369 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1370 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1371 | u64 chunk; | |
1372 | int error; | |
1373 | ||
1374 | while (oldsize != newsize) { | |
e7fdf004 AG |
1375 | struct gfs2_trans *tr; |
1376 | unsigned int offs; | |
1377 | ||
fa731fc4 SW |
1378 | chunk = oldsize - newsize; |
1379 | if (chunk > max_chunk) | |
1380 | chunk = max_chunk; | |
e7fdf004 AG |
1381 | |
1382 | offs = oldsize & ~PAGE_MASK; | |
1383 | if (offs && chunk > PAGE_SIZE) | |
1384 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1385 | ||
7caef267 | 1386 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1387 | oldsize -= chunk; |
e7fdf004 AG |
1388 | |
1389 | tr = current->journal_info; | |
1390 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1391 | continue; | |
1392 | ||
fa731fc4 SW |
1393 | gfs2_trans_end(sdp); |
1394 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1395 | if (error) | |
1396 | return error; | |
1397 | } | |
1398 | ||
1399 | return 0; | |
1400 | } | |
1401 | ||
8b5860a3 | 1402 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1403 | { |
ff8f33c8 SW |
1404 | struct gfs2_inode *ip = GFS2_I(inode); |
1405 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
80990f40 | 1406 | struct buffer_head *dibh = NULL; |
b3b94faa | 1407 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1408 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1409 | int error; |
1410 | ||
70499cdf BP |
1411 | if (!gfs2_is_stuffed(ip)) { |
1412 | unsigned int blocksize = i_blocksize(inode); | |
1413 | unsigned int offs = newsize & (blocksize - 1); | |
1414 | if (offs) { | |
1415 | error = gfs2_block_zero_range(inode, newsize, | |
1416 | blocksize - offs); | |
1417 | if (error) | |
1418 | return error; | |
1419 | } | |
1420 | } | |
fa731fc4 SW |
1421 | if (journaled) |
1422 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1423 | else | |
1424 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1425 | if (error) |
1426 | return error; | |
1427 | ||
1428 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1429 | if (error) | |
1430 | goto out; | |
1431 | ||
350a9b0a | 1432 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1433 | |
70499cdf | 1434 | if (gfs2_is_stuffed(ip)) |
ff8f33c8 | 1435 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
70499cdf | 1436 | else |
ff8f33c8 | 1437 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1438 | |
ff8f33c8 | 1439 | i_size_write(inode, newsize); |
078cd827 | 1440 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
ff8f33c8 | 1441 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1442 | |
fa731fc4 SW |
1443 | if (journaled) |
1444 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1445 | else | |
7caef267 | 1446 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1447 | |
a91ea69f | 1448 | out: |
80990f40 AG |
1449 | brelse(dibh); |
1450 | if (current->journal_info) | |
1451 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1452 | return error; |
1453 | } | |
1454 | ||
54992257 AG |
1455 | int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length, |
1456 | struct iomap *iomap) | |
1457 | { | |
1458 | struct metapath mp = { .mp_aheight = 1, }; | |
1459 | int ret; | |
1460 | ||
1461 | ret = __gfs2_iomap_get(inode, pos, length, 0, iomap, &mp); | |
1462 | release_metapath(&mp); | |
1463 | return ret; | |
1464 | } | |
1465 | ||
1466 | int gfs2_iomap_alloc(struct inode *inode, loff_t pos, loff_t length, | |
1467 | struct iomap *iomap) | |
628e366d AG |
1468 | { |
1469 | struct metapath mp = { .mp_aheight = 1, }; | |
1470 | int ret; | |
1471 | ||
54992257 | 1472 | ret = __gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp); |
628e366d | 1473 | if (!ret && iomap->type == IOMAP_HOLE) |
54992257 | 1474 | ret = __gfs2_iomap_alloc(inode, iomap, &mp); |
628e366d AG |
1475 | release_metapath(&mp); |
1476 | return ret; | |
1477 | } | |
1478 | ||
d552a2b9 BP |
1479 | /** |
1480 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1481 | * @ip: inode | |
c551f66c | 1482 | * @rd_gh: holder of resource group glock |
5cf26b1e AG |
1483 | * @bh: buffer head to sweep |
1484 | * @start: starting point in bh | |
1485 | * @end: end point in bh | |
1486 | * @meta: true if bh points to metadata (rather than data) | |
d552a2b9 | 1487 | * @btotal: place to keep count of total blocks freed |
d552a2b9 BP |
1488 | * |
1489 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1490 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1491 | * block has references to multiple rgrps, we break it into individual | |
1492 | * transactions. This allows other processes to use the rgrps while we're | |
1493 | * focused on a single one, for better concurrency / performance. | |
1494 | * At every transaction boundary, we rewrite the inode into the journal. | |
1495 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1496 | * if we're interrupted by power-outages. | |
1497 | * | |
1498 | * Returns: 0, or return code if an error occurred. | |
1499 | * *btotal has the total number of blocks freed | |
1500 | */ | |
1501 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
5cf26b1e AG |
1502 | struct buffer_head *bh, __be64 *start, __be64 *end, |
1503 | bool meta, u32 *btotal) | |
b3b94faa | 1504 | { |
9b8c81d1 | 1505 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1506 | struct gfs2_rgrpd *rgd; |
1507 | struct gfs2_trans *tr; | |
5cf26b1e | 1508 | __be64 *p; |
d552a2b9 BP |
1509 | int blks_outside_rgrp; |
1510 | u64 bn, bstart, isize_blks; | |
1511 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
d552a2b9 BP |
1512 | int ret = 0; |
1513 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1514 | ||
d552a2b9 | 1515 | more_rgrps: |
5cf26b1e AG |
1516 | rgd = NULL; |
1517 | if (gfs2_holder_initialized(rd_gh)) { | |
1518 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); | |
1519 | gfs2_assert_withdraw(sdp, | |
1520 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1521 | } | |
d552a2b9 BP |
1522 | blks_outside_rgrp = 0; |
1523 | bstart = 0; | |
1524 | blen = 0; | |
d552a2b9 | 1525 | |
5cf26b1e | 1526 | for (p = start; p < end; p++) { |
d552a2b9 BP |
1527 | if (!*p) |
1528 | continue; | |
1529 | bn = be64_to_cpu(*p); | |
5cf26b1e AG |
1530 | |
1531 | if (rgd) { | |
1532 | if (!rgrp_contains_block(rgd, bn)) { | |
1533 | blks_outside_rgrp++; | |
1534 | continue; | |
1535 | } | |
d552a2b9 | 1536 | } else { |
90bcab99 | 1537 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
5cf26b1e AG |
1538 | if (unlikely(!rgd)) { |
1539 | ret = -EIO; | |
1540 | goto out; | |
1541 | } | |
d552a2b9 | 1542 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
4fc7ec31 | 1543 | LM_FLAG_NODE_SCOPE, rd_gh); |
d552a2b9 BP |
1544 | if (ret) |
1545 | goto out; | |
1546 | ||
1547 | /* Must be done with the rgrp glock held: */ | |
1548 | if (gfs2_rs_active(&ip->i_res) && | |
c65b76b8 | 1549 | rgd == ip->i_res.rs_rgd) |
d552a2b9 BP |
1550 | gfs2_rs_deltree(&ip->i_res); |
1551 | } | |
1552 | ||
d552a2b9 BP |
1553 | /* The size of our transactions will be unknown until we |
1554 | actually process all the metadata blocks that relate to | |
1555 | the rgrp. So we estimate. We know it can't be more than | |
1556 | the dinode's i_blocks and we don't want to exceed the | |
1557 | journal flush threshold, sd_log_thresh2. */ | |
1558 | if (current->journal_info == NULL) { | |
1559 | unsigned int jblocks_rqsted, revokes; | |
1560 | ||
1561 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1562 | RES_INDIRECT; | |
1563 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1564 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1565 | jblocks_rqsted += | |
1566 | atomic_read(&sdp->sd_log_thresh2); | |
1567 | else | |
1568 | jblocks_rqsted += isize_blks; | |
1569 | revokes = jblocks_rqsted; | |
1570 | if (meta) | |
5cf26b1e | 1571 | revokes += end - start; |
d552a2b9 BP |
1572 | else if (ip->i_depth) |
1573 | revokes += sdp->sd_inptrs; | |
1574 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1575 | if (ret) | |
1576 | goto out_unlock; | |
1577 | down_write(&ip->i_rw_mutex); | |
1578 | } | |
1579 | /* check if we will exceed the transaction blocks requested */ | |
1580 | tr = current->journal_info; | |
1581 | if (tr->tr_num_buf_new + RES_STATFS + | |
1582 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1583 | /* We set blks_outside_rgrp to ensure the loop will | |
1584 | be repeated for the same rgrp, but with a new | |
1585 | transaction. */ | |
1586 | blks_outside_rgrp++; | |
1587 | /* This next part is tricky. If the buffer was added | |
1588 | to the transaction, we've already set some block | |
1589 | pointers to 0, so we better follow through and free | |
1590 | them, or we will introduce corruption (so break). | |
1591 | This may be impossible, or at least rare, but I | |
1592 | decided to cover the case regardless. | |
1593 | ||
1594 | If the buffer was not added to the transaction | |
1595 | (this call), doing so would exceed our transaction | |
1596 | size, so we need to end the transaction and start a | |
1597 | new one (so goto). */ | |
1598 | ||
1599 | if (buf_in_tr) | |
1600 | break; | |
1601 | goto out_unlock; | |
1602 | } | |
1603 | ||
1604 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1605 | buf_in_tr = true; | |
1606 | *p = 0; | |
1607 | if (bstart + blen == bn) { | |
1608 | blen++; | |
1609 | continue; | |
1610 | } | |
1611 | if (bstart) { | |
0ddeded4 | 1612 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1613 | (*btotal) += blen; |
1614 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1615 | } | |
1616 | bstart = bn; | |
1617 | blen = 1; | |
1618 | } | |
1619 | if (bstart) { | |
0ddeded4 | 1620 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1621 | (*btotal) += blen; |
1622 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1623 | } | |
1624 | out_unlock: | |
1625 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1626 | outside the rgrp we just processed, | |
1627 | do it all over again. */ | |
1628 | if (current->journal_info) { | |
5cf26b1e AG |
1629 | struct buffer_head *dibh; |
1630 | ||
1631 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1632 | if (ret) | |
1633 | goto out; | |
d552a2b9 BP |
1634 | |
1635 | /* Every transaction boundary, we rewrite the dinode | |
1636 | to keep its di_blocks current in case of failure. */ | |
1637 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = | |
b32c8c76 | 1638 | current_time(&ip->i_inode); |
d552a2b9 BP |
1639 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1640 | gfs2_dinode_out(ip, dibh->b_data); | |
5cf26b1e | 1641 | brelse(dibh); |
d552a2b9 BP |
1642 | up_write(&ip->i_rw_mutex); |
1643 | gfs2_trans_end(sdp); | |
f0b444b3 | 1644 | buf_in_tr = false; |
d552a2b9 BP |
1645 | } |
1646 | gfs2_glock_dq_uninit(rd_gh); | |
1647 | cond_resched(); | |
1648 | goto more_rgrps; | |
1649 | } | |
1650 | out: | |
1651 | return ret; | |
1652 | } | |
1653 | ||
10d2cf94 AG |
1654 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h) |
1655 | { | |
1656 | if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0]))) | |
1657 | return false; | |
1658 | return true; | |
1659 | } | |
1660 | ||
d552a2b9 BP |
1661 | /** |
1662 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
c551f66c | 1663 | * @sdp: The superblock |
d552a2b9 BP |
1664 | * @mp: starting metapath |
1665 | * @h: desired height to search | |
c551f66c LJ |
1666 | * @end_list: See punch_hole(). |
1667 | * @end_aligned: See punch_hole(). | |
d552a2b9 | 1668 | * |
10d2cf94 | 1669 | * Assumes the metapath is valid (with buffers) out to height h. |
d552a2b9 BP |
1670 | * Returns: true if a non-null pointer was found in the metapath buffer |
1671 | * false if all remaining pointers are NULL in the buffer | |
1672 | */ | |
1673 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
10d2cf94 AG |
1674 | unsigned int h, |
1675 | __u16 *end_list, unsigned int end_aligned) | |
d552a2b9 | 1676 | { |
10d2cf94 AG |
1677 | struct buffer_head *bh = mp->mp_bh[h]; |
1678 | __be64 *first, *ptr, *end; | |
1679 | ||
1680 | first = metaptr1(h, mp); | |
1681 | ptr = first + mp->mp_list[h]; | |
1682 | end = (__be64 *)(bh->b_data + bh->b_size); | |
1683 | if (end_list && mp_eq_to_hgt(mp, end_list, h)) { | |
1684 | bool keep_end = h < end_aligned; | |
1685 | end = first + end_list[h] + keep_end; | |
1686 | } | |
d552a2b9 | 1687 | |
10d2cf94 | 1688 | while (ptr < end) { |
c4a9d189 | 1689 | if (*ptr) { /* if we have a non-null pointer */ |
10d2cf94 | 1690 | mp->mp_list[h] = ptr - first; |
c4a9d189 BP |
1691 | h++; |
1692 | if (h < GFS2_MAX_META_HEIGHT) | |
10d2cf94 | 1693 | mp->mp_list[h] = 0; |
d552a2b9 | 1694 | return true; |
c4a9d189 | 1695 | } |
10d2cf94 | 1696 | ptr++; |
d552a2b9 | 1697 | } |
10d2cf94 | 1698 | return false; |
d552a2b9 BP |
1699 | } |
1700 | ||
1701 | enum dealloc_states { | |
1702 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1703 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1704 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1705 | DEALLOC_DONE = 3, /* process complete */ | |
1706 | }; | |
b3b94faa | 1707 | |
5cf26b1e AG |
1708 | static inline void |
1709 | metapointer_range(struct metapath *mp, int height, | |
1710 | __u16 *start_list, unsigned int start_aligned, | |
10d2cf94 | 1711 | __u16 *end_list, unsigned int end_aligned, |
5cf26b1e AG |
1712 | __be64 **start, __be64 **end) |
1713 | { | |
1714 | struct buffer_head *bh = mp->mp_bh[height]; | |
1715 | __be64 *first; | |
1716 | ||
1717 | first = metaptr1(height, mp); | |
1718 | *start = first; | |
1719 | if (mp_eq_to_hgt(mp, start_list, height)) { | |
1720 | bool keep_start = height < start_aligned; | |
1721 | *start = first + start_list[height] + keep_start; | |
1722 | } | |
1723 | *end = (__be64 *)(bh->b_data + bh->b_size); | |
10d2cf94 AG |
1724 | if (end_list && mp_eq_to_hgt(mp, end_list, height)) { |
1725 | bool keep_end = height < end_aligned; | |
1726 | *end = first + end_list[height] + keep_end; | |
1727 | } | |
1728 | } | |
1729 | ||
1730 | static inline bool walk_done(struct gfs2_sbd *sdp, | |
1731 | struct metapath *mp, int height, | |
1732 | __u16 *end_list, unsigned int end_aligned) | |
1733 | { | |
1734 | __u16 end; | |
1735 | ||
1736 | if (end_list) { | |
1737 | bool keep_end = height < end_aligned; | |
1738 | if (!mp_eq_to_hgt(mp, end_list, height)) | |
1739 | return false; | |
1740 | end = end_list[height] + keep_end; | |
1741 | } else | |
1742 | end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
1743 | return mp->mp_list[height] >= end; | |
5cf26b1e AG |
1744 | } |
1745 | ||
d552a2b9 | 1746 | /** |
10d2cf94 | 1747 | * punch_hole - deallocate blocks in a file |
d552a2b9 | 1748 | * @ip: inode to truncate |
10d2cf94 AG |
1749 | * @offset: the start of the hole |
1750 | * @length: the size of the hole (or 0 for truncate) | |
1751 | * | |
1752 | * Punch a hole into a file or truncate a file at a given position. This | |
1753 | * function operates in whole blocks (@offset and @length are rounded | |
1754 | * accordingly); partially filled blocks must be cleared otherwise. | |
d552a2b9 | 1755 | * |
10d2cf94 AG |
1756 | * This function works from the bottom up, and from the right to the left. In |
1757 | * other words, it strips off the highest layer (data) before stripping any of | |
1758 | * the metadata. Doing it this way is best in case the operation is interrupted | |
1759 | * by power failure, etc. The dinode is rewritten in every transaction to | |
1760 | * guarantee integrity. | |
d552a2b9 | 1761 | */ |
10d2cf94 | 1762 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length) |
d552a2b9 BP |
1763 | { |
1764 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
bb491ce6 | 1765 | u64 maxsize = sdp->sd_heightsize[ip->i_height]; |
10d2cf94 | 1766 | struct metapath mp = {}; |
d552a2b9 BP |
1767 | struct buffer_head *dibh, *bh; |
1768 | struct gfs2_holder rd_gh; | |
cb7f0903 | 1769 | unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; |
10d2cf94 AG |
1770 | u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift; |
1771 | __u16 start_list[GFS2_MAX_META_HEIGHT]; | |
1772 | __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL; | |
3f649ab7 | 1773 | unsigned int start_aligned, end_aligned; |
d552a2b9 BP |
1774 | unsigned int strip_h = ip->i_height - 1; |
1775 | u32 btotal = 0; | |
1776 | int ret, state; | |
1777 | int mp_h; /* metapath buffers are read in to this height */ | |
d552a2b9 | 1778 | u64 prev_bnr = 0; |
5cf26b1e | 1779 | __be64 *start, *end; |
b3b94faa | 1780 | |
bb491ce6 AG |
1781 | if (offset >= maxsize) { |
1782 | /* | |
1783 | * The starting point lies beyond the allocated meta-data; | |
1784 | * there are no blocks do deallocate. | |
1785 | */ | |
1786 | return 0; | |
1787 | } | |
1788 | ||
10d2cf94 AG |
1789 | /* |
1790 | * The start position of the hole is defined by lblock, start_list, and | |
1791 | * start_aligned. The end position of the hole is defined by lend, | |
1792 | * end_list, and end_aligned. | |
1793 | * | |
1794 | * start_aligned and end_aligned define down to which height the start | |
1795 | * and end positions are aligned to the metadata tree (i.e., the | |
1796 | * position is a multiple of the metadata granularity at the height | |
1797 | * above). This determines at which heights additional meta pointers | |
1798 | * needs to be preserved for the remaining data. | |
1799 | */ | |
b3b94faa | 1800 | |
10d2cf94 | 1801 | if (length) { |
10d2cf94 AG |
1802 | u64 end_offset = offset + length; |
1803 | u64 lend; | |
1804 | ||
1805 | /* | |
1806 | * Clip the end at the maximum file size for the given height: | |
1807 | * that's how far the metadata goes; files bigger than that | |
1808 | * will have additional layers of indirection. | |
1809 | */ | |
1810 | if (end_offset > maxsize) | |
1811 | end_offset = maxsize; | |
1812 | lend = end_offset >> bsize_shift; | |
1813 | ||
1814 | if (lblock >= lend) | |
1815 | return 0; | |
1816 | ||
1817 | find_metapath(sdp, lend, &mp, ip->i_height); | |
1818 | end_list = __end_list; | |
1819 | memcpy(end_list, mp.mp_list, sizeof(mp.mp_list)); | |
1820 | ||
1821 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { | |
1822 | if (end_list[mp_h]) | |
1823 | break; | |
1824 | } | |
1825 | end_aligned = mp_h; | |
1826 | } | |
1827 | ||
1828 | find_metapath(sdp, lblock, &mp, ip->i_height); | |
cb7f0903 AG |
1829 | memcpy(start_list, mp.mp_list, sizeof(start_list)); |
1830 | ||
cb7f0903 AG |
1831 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { |
1832 | if (start_list[mp_h]) | |
1833 | break; | |
1834 | } | |
1835 | start_aligned = mp_h; | |
d552a2b9 BP |
1836 | |
1837 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1838 | if (ret) | |
1839 | return ret; | |
b3b94faa | 1840 | |
d552a2b9 BP |
1841 | mp.mp_bh[0] = dibh; |
1842 | ret = lookup_metapath(ip, &mp); | |
e8b43fe0 AG |
1843 | if (ret) |
1844 | goto out_metapath; | |
c3ce5aa9 AG |
1845 | |
1846 | /* issue read-ahead on metadata */ | |
5cf26b1e AG |
1847 | for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) { |
1848 | metapointer_range(&mp, mp_h, start_list, start_aligned, | |
10d2cf94 | 1849 | end_list, end_aligned, &start, &end); |
5cf26b1e AG |
1850 | gfs2_metapath_ra(ip->i_gl, start, end); |
1851 | } | |
c3ce5aa9 | 1852 | |
e8b43fe0 | 1853 | if (mp.mp_aheight == ip->i_height) |
d552a2b9 BP |
1854 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ |
1855 | else | |
1856 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1857 | |
d552a2b9 BP |
1858 | ret = gfs2_rindex_update(sdp); |
1859 | if (ret) | |
1860 | goto out_metapath; | |
1861 | ||
1862 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1863 | if (ret) | |
1864 | goto out_metapath; | |
1865 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1866 | ||
1867 | mp_h = strip_h; | |
1868 | ||
1869 | while (state != DEALLOC_DONE) { | |
1870 | switch (state) { | |
1871 | /* Truncate a full metapath at the given strip height. | |
1872 | * Note that strip_h == mp_h in order to be in this state. */ | |
1873 | case DEALLOC_MP_FULL: | |
d552a2b9 BP |
1874 | bh = mp.mp_bh[mp_h]; |
1875 | gfs2_assert_withdraw(sdp, bh); | |
1876 | if (gfs2_assert_withdraw(sdp, | |
1877 | prev_bnr != bh->b_blocknr)) { | |
f29e62ee BP |
1878 | fs_emerg(sdp, "inode %llu, block:%llu, i_h:%u," |
1879 | "s_h:%u, mp_h:%u\n", | |
d552a2b9 BP |
1880 | (unsigned long long)ip->i_no_addr, |
1881 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1882 | } | |
1883 | prev_bnr = bh->b_blocknr; | |
cb7f0903 | 1884 | |
5cf26b1e AG |
1885 | if (gfs2_metatype_check(sdp, bh, |
1886 | (mp_h ? GFS2_METATYPE_IN : | |
1887 | GFS2_METATYPE_DI))) { | |
1888 | ret = -EIO; | |
1889 | goto out; | |
1890 | } | |
1891 | ||
10d2cf94 AG |
1892 | /* |
1893 | * Below, passing end_aligned as 0 gives us the | |
1894 | * metapointer range excluding the end point: the end | |
1895 | * point is the first metapath we must not deallocate! | |
1896 | */ | |
1897 | ||
5cf26b1e | 1898 | metapointer_range(&mp, mp_h, start_list, start_aligned, |
10d2cf94 | 1899 | end_list, 0 /* end_aligned */, |
5cf26b1e AG |
1900 | &start, &end); |
1901 | ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h], | |
1902 | start, end, | |
1903 | mp_h != ip->i_height - 1, | |
1904 | &btotal); | |
cb7f0903 | 1905 | |
d552a2b9 BP |
1906 | /* If we hit an error or just swept dinode buffer, |
1907 | just exit. */ | |
1908 | if (ret || !mp_h) { | |
1909 | state = DEALLOC_DONE; | |
1910 | break; | |
1911 | } | |
1912 | state = DEALLOC_MP_LOWER; | |
1913 | break; | |
1914 | ||
1915 | /* lower the metapath strip height */ | |
1916 | case DEALLOC_MP_LOWER: | |
1917 | /* We're done with the current buffer, so release it, | |
1918 | unless it's the dinode buffer. Then back up to the | |
1919 | previous pointer. */ | |
1920 | if (mp_h) { | |
1921 | brelse(mp.mp_bh[mp_h]); | |
1922 | mp.mp_bh[mp_h] = NULL; | |
1923 | } | |
1924 | /* If we can't get any lower in height, we've stripped | |
1925 | off all we can. Next step is to back up and start | |
1926 | stripping the previous level of metadata. */ | |
1927 | if (mp_h == 0) { | |
1928 | strip_h--; | |
cb7f0903 | 1929 | memcpy(mp.mp_list, start_list, sizeof(start_list)); |
d552a2b9 BP |
1930 | mp_h = strip_h; |
1931 | state = DEALLOC_FILL_MP; | |
1932 | break; | |
1933 | } | |
1934 | mp.mp_list[mp_h] = 0; | |
1935 | mp_h--; /* search one metadata height down */ | |
d552a2b9 | 1936 | mp.mp_list[mp_h]++; |
10d2cf94 AG |
1937 | if (walk_done(sdp, &mp, mp_h, end_list, end_aligned)) |
1938 | break; | |
d552a2b9 BP |
1939 | /* Here we've found a part of the metapath that is not |
1940 | * allocated. We need to search at that height for the | |
1941 | * next non-null pointer. */ | |
10d2cf94 | 1942 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) { |
d552a2b9 BP |
1943 | state = DEALLOC_FILL_MP; |
1944 | mp_h++; | |
1945 | } | |
1946 | /* No more non-null pointers at this height. Back up | |
1947 | to the previous height and try again. */ | |
1948 | break; /* loop around in the same state */ | |
1949 | ||
1950 | /* Fill the metapath with buffers to the given height. */ | |
1951 | case DEALLOC_FILL_MP: | |
1952 | /* Fill the buffers out to the current height. */ | |
1953 | ret = fillup_metapath(ip, &mp, mp_h); | |
c3ce5aa9 | 1954 | if (ret < 0) |
d552a2b9 | 1955 | goto out; |
c3ce5aa9 | 1956 | |
e7445ced AG |
1957 | /* On the first pass, issue read-ahead on metadata. */ |
1958 | if (mp.mp_aheight > 1 && strip_h == ip->i_height - 1) { | |
1959 | unsigned int height = mp.mp_aheight - 1; | |
1960 | ||
1961 | /* No read-ahead for data blocks. */ | |
1962 | if (mp.mp_aheight - 1 == strip_h) | |
1963 | height--; | |
1964 | ||
1965 | for (; height >= mp.mp_aheight - ret; height--) { | |
1966 | metapointer_range(&mp, height, | |
5cf26b1e | 1967 | start_list, start_aligned, |
10d2cf94 | 1968 | end_list, end_aligned, |
5cf26b1e AG |
1969 | &start, &end); |
1970 | gfs2_metapath_ra(ip->i_gl, start, end); | |
1971 | } | |
c3ce5aa9 | 1972 | } |
d552a2b9 BP |
1973 | |
1974 | /* If buffers found for the entire strip height */ | |
e8b43fe0 | 1975 | if (mp.mp_aheight - 1 == strip_h) { |
d552a2b9 BP |
1976 | state = DEALLOC_MP_FULL; |
1977 | break; | |
1978 | } | |
e8b43fe0 AG |
1979 | if (mp.mp_aheight < ip->i_height) /* We have a partial height */ |
1980 | mp_h = mp.mp_aheight - 1; | |
d552a2b9 BP |
1981 | |
1982 | /* If we find a non-null block pointer, crawl a bit | |
1983 | higher up in the metapath and try again, otherwise | |
1984 | we need to look lower for a new starting point. */ | |
10d2cf94 | 1985 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) |
d552a2b9 BP |
1986 | mp_h++; |
1987 | else | |
1988 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1989 | break; |
d552a2b9 | 1990 | } |
b3b94faa DT |
1991 | } |
1992 | ||
d552a2b9 BP |
1993 | if (btotal) { |
1994 | if (current->journal_info == NULL) { | |
1995 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1996 | RES_QUOTA, 0); | |
1997 | if (ret) | |
1998 | goto out; | |
1999 | down_write(&ip->i_rw_mutex); | |
2000 | } | |
2001 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
2002 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
2003 | ip->i_inode.i_gid); | |
b32c8c76 | 2004 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
d552a2b9 BP |
2005 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2006 | gfs2_dinode_out(ip, dibh->b_data); | |
2007 | up_write(&ip->i_rw_mutex); | |
2008 | gfs2_trans_end(sdp); | |
2009 | } | |
b3b94faa | 2010 | |
d552a2b9 BP |
2011 | out: |
2012 | if (gfs2_holder_initialized(&rd_gh)) | |
2013 | gfs2_glock_dq_uninit(&rd_gh); | |
2014 | if (current->journal_info) { | |
2015 | up_write(&ip->i_rw_mutex); | |
2016 | gfs2_trans_end(sdp); | |
2017 | cond_resched(); | |
2018 | } | |
2019 | gfs2_quota_unhold(ip); | |
2020 | out_metapath: | |
2021 | release_metapath(&mp); | |
2022 | return ret; | |
b3b94faa DT |
2023 | } |
2024 | ||
2025 | static int trunc_end(struct gfs2_inode *ip) | |
2026 | { | |
feaa7bba | 2027 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2028 | struct buffer_head *dibh; |
2029 | int error; | |
2030 | ||
2031 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2032 | if (error) | |
2033 | return error; | |
2034 | ||
2035 | down_write(&ip->i_rw_mutex); | |
2036 | ||
2037 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2038 | if (error) | |
2039 | goto out; | |
2040 | ||
a2e0f799 | 2041 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 2042 | ip->i_height = 0; |
ce276b06 | 2043 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 2044 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 2045 | gfs2_ordered_del_inode(ip); |
b3b94faa | 2046 | } |
078cd827 | 2047 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
383f01fb | 2048 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 2049 | |
350a9b0a | 2050 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 2051 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
2052 | brelse(dibh); |
2053 | ||
a91ea69f | 2054 | out: |
b3b94faa | 2055 | up_write(&ip->i_rw_mutex); |
b3b94faa | 2056 | gfs2_trans_end(sdp); |
b3b94faa DT |
2057 | return error; |
2058 | } | |
2059 | ||
2060 | /** | |
2061 | * do_shrink - make a file smaller | |
ff8f33c8 | 2062 | * @inode: the inode |
ff8f33c8 | 2063 | * @newsize: the size to make the file |
b3b94faa | 2064 | * |
ff8f33c8 SW |
2065 | * Called with an exclusive lock on @inode. The @size must |
2066 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
2067 | * |
2068 | * Returns: errno | |
2069 | */ | |
2070 | ||
8b5860a3 | 2071 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 2072 | { |
ff8f33c8 | 2073 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
2074 | int error; |
2075 | ||
8b5860a3 | 2076 | error = trunc_start(inode, newsize); |
b3b94faa DT |
2077 | if (error < 0) |
2078 | return error; | |
ff8f33c8 | 2079 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
2080 | return 0; |
2081 | ||
10d2cf94 | 2082 | error = punch_hole(ip, newsize, 0); |
ff8f33c8 | 2083 | if (error == 0) |
b3b94faa DT |
2084 | error = trunc_end(ip); |
2085 | ||
2086 | return error; | |
2087 | } | |
2088 | ||
ff8f33c8 | 2089 | void gfs2_trim_blocks(struct inode *inode) |
a13b8c5f | 2090 | { |
ff8f33c8 SW |
2091 | int ret; |
2092 | ||
8b5860a3 | 2093 | ret = do_shrink(inode, inode->i_size); |
ff8f33c8 SW |
2094 | WARN_ON(ret != 0); |
2095 | } | |
2096 | ||
2097 | /** | |
2098 | * do_grow - Touch and update inode size | |
2099 | * @inode: The inode | |
2100 | * @size: The new size | |
2101 | * | |
2102 | * This function updates the timestamps on the inode and | |
2103 | * may also increase the size of the inode. This function | |
2104 | * must not be called with @size any smaller than the current | |
2105 | * inode size. | |
2106 | * | |
2107 | * Although it is not strictly required to unstuff files here, | |
2108 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
2109 | * code which will result in a buffer overrun if the size is larger | |
2110 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 2111 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
2112 | * just update the inode size directly. |
2113 | * | |
2114 | * Returns: 0 on success, or -ve on error | |
2115 | */ | |
2116 | ||
2117 | static int do_grow(struct inode *inode, u64 size) | |
2118 | { | |
2119 | struct gfs2_inode *ip = GFS2_I(inode); | |
2120 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 2121 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
2122 | struct buffer_head *dibh; |
2123 | int error; | |
2f7ee358 | 2124 | int unstuff = 0; |
a13b8c5f | 2125 | |
235628c5 | 2126 | if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) { |
b8fbf471 | 2127 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 2128 | if (error) |
5407e242 | 2129 | return error; |
ff8f33c8 | 2130 | |
7b9cff46 | 2131 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
2132 | if (error) |
2133 | goto do_grow_qunlock; | |
2f7ee358 | 2134 | unstuff = 1; |
ff8f33c8 SW |
2135 | } |
2136 | ||
a01aedfe | 2137 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
bc020561 BP |
2138 | (unstuff && |
2139 | gfs2_is_jdata(ip) ? RES_JDATA : 0) + | |
a01aedfe BP |
2140 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? |
2141 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 2142 | if (error) |
ff8f33c8 | 2143 | goto do_grow_release; |
a13b8c5f | 2144 | |
2f7ee358 | 2145 | if (unstuff) { |
7a607a41 | 2146 | error = gfs2_unstuff_dinode(ip); |
ff8f33c8 SW |
2147 | if (error) |
2148 | goto do_end_trans; | |
2149 | } | |
a13b8c5f WC |
2150 | |
2151 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2152 | if (error) | |
ff8f33c8 | 2153 | goto do_end_trans; |
a13b8c5f | 2154 | |
b473bc2d | 2155 | truncate_setsize(inode, size); |
078cd827 | 2156 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
350a9b0a | 2157 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
2158 | gfs2_dinode_out(ip, dibh->b_data); |
2159 | brelse(dibh); | |
2160 | ||
ff8f33c8 | 2161 | do_end_trans: |
a13b8c5f | 2162 | gfs2_trans_end(sdp); |
ff8f33c8 | 2163 | do_grow_release: |
2f7ee358 | 2164 | if (unstuff) { |
ff8f33c8 SW |
2165 | gfs2_inplace_release(ip); |
2166 | do_grow_qunlock: | |
2167 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 2168 | } |
a13b8c5f WC |
2169 | return error; |
2170 | } | |
2171 | ||
b3b94faa | 2172 | /** |
ff8f33c8 SW |
2173 | * gfs2_setattr_size - make a file a given size |
2174 | * @inode: the inode | |
2175 | * @newsize: the size to make the file | |
b3b94faa | 2176 | * |
ff8f33c8 | 2177 | * The file size can grow, shrink, or stay the same size. This |
3e7aafc3 | 2178 | * is called holding i_rwsem and an exclusive glock on the inode |
ff8f33c8 | 2179 | * in question. |
b3b94faa DT |
2180 | * |
2181 | * Returns: errno | |
2182 | */ | |
2183 | ||
ff8f33c8 | 2184 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 2185 | { |
af5c2697 | 2186 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 2187 | int ret; |
b3b94faa | 2188 | |
ff8f33c8 | 2189 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 2190 | |
ff8f33c8 SW |
2191 | ret = inode_newsize_ok(inode, newsize); |
2192 | if (ret) | |
2193 | return ret; | |
b3b94faa | 2194 | |
562c72aa CH |
2195 | inode_dio_wait(inode); |
2196 | ||
2fba46a0 | 2197 | ret = gfs2_qa_get(ip); |
d2b47cfb | 2198 | if (ret) |
2b3dcf35 | 2199 | goto out; |
d2b47cfb | 2200 | |
8b5860a3 | 2201 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
2202 | ret = do_grow(inode, newsize); |
2203 | goto out; | |
2204 | } | |
ff8f33c8 | 2205 | |
8b5860a3 | 2206 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 2207 | out: |
1595548f AG |
2208 | gfs2_rs_delete(ip, NULL); |
2209 | gfs2_qa_put(ip); | |
2b3dcf35 | 2210 | return ret; |
b3b94faa DT |
2211 | } |
2212 | ||
2213 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
2214 | { | |
2215 | int error; | |
10d2cf94 | 2216 | error = punch_hole(ip, i_size_read(&ip->i_inode), 0); |
b3b94faa DT |
2217 | if (!error) |
2218 | error = trunc_end(ip); | |
2219 | return error; | |
2220 | } | |
2221 | ||
2222 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
2223 | { | |
10d2cf94 | 2224 | return punch_hole(ip, 0, 0); |
b3b94faa DT |
2225 | } |
2226 | ||
b50f227b SW |
2227 | /** |
2228 | * gfs2_free_journal_extents - Free cached journal bmap info | |
2229 | * @jd: The journal | |
2230 | * | |
2231 | */ | |
2232 | ||
2233 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
2234 | { | |
2235 | struct gfs2_journal_extent *jext; | |
2236 | ||
2237 | while(!list_empty(&jd->extent_list)) { | |
969183bc | 2238 | jext = list_first_entry(&jd->extent_list, struct gfs2_journal_extent, list); |
b50f227b SW |
2239 | list_del(&jext->list); |
2240 | kfree(jext); | |
2241 | } | |
2242 | } | |
2243 | ||
2244 | /** | |
2245 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
2246 | * @jd: The journal descriptor | |
2247 | * @lblock: The logical block at start of new extent | |
c62baf65 | 2248 | * @dblock: The physical block at start of new extent |
b50f227b SW |
2249 | * @blocks: Size of extent in fs blocks |
2250 | * | |
2251 | * Returns: 0 on success or -ENOMEM | |
2252 | */ | |
2253 | ||
2254 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
2255 | { | |
2256 | struct gfs2_journal_extent *jext; | |
2257 | ||
2258 | if (!list_empty(&jd->extent_list)) { | |
969183bc | 2259 | jext = list_last_entry(&jd->extent_list, struct gfs2_journal_extent, list); |
b50f227b SW |
2260 | if ((jext->dblock + jext->blocks) == dblock) { |
2261 | jext->blocks += blocks; | |
2262 | return 0; | |
2263 | } | |
2264 | } | |
2265 | ||
2266 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
2267 | if (jext == NULL) | |
2268 | return -ENOMEM; | |
2269 | jext->dblock = dblock; | |
2270 | jext->lblock = lblock; | |
2271 | jext->blocks = blocks; | |
2272 | list_add_tail(&jext->list, &jd->extent_list); | |
2273 | jd->nr_extents++; | |
2274 | return 0; | |
2275 | } | |
2276 | ||
2277 | /** | |
2278 | * gfs2_map_journal_extents - Cache journal bmap info | |
2279 | * @sdp: The super block | |
2280 | * @jd: The journal to map | |
2281 | * | |
2282 | * Create a reusable "extent" mapping from all logical | |
2283 | * blocks to all physical blocks for the given journal. This will save | |
2284 | * us time when writing journal blocks. Most journals will have only one | |
2285 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
2286 | * arranges the journal blocks sequentially to maximize performance. | |
2287 | * So the extent would map the first block for the entire file length. | |
2288 | * However, gfs2_jadd can happen while file activity is happening, so | |
2289 | * those journals may not be sequential. Less likely is the case where | |
2290 | * the users created their own journals by mounting the metafs and | |
2291 | * laying it out. But it's still possible. These journals might have | |
2292 | * several extents. | |
2293 | * | |
2294 | * Returns: 0 on success, or error on failure | |
2295 | */ | |
2296 | ||
2297 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
2298 | { | |
2299 | u64 lblock = 0; | |
2300 | u64 lblock_stop; | |
2301 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
2302 | struct buffer_head bh; | |
2303 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
2304 | u64 size; | |
2305 | int rc; | |
98583b3e | 2306 | ktime_t start, end; |
b50f227b | 2307 | |
98583b3e | 2308 | start = ktime_get(); |
b50f227b SW |
2309 | lblock_stop = i_size_read(jd->jd_inode) >> shift; |
2310 | size = (lblock_stop - lblock) << shift; | |
2311 | jd->nr_extents = 0; | |
2312 | WARN_ON(!list_empty(&jd->extent_list)); | |
2313 | ||
2314 | do { | |
2315 | bh.b_state = 0; | |
2316 | bh.b_blocknr = 0; | |
2317 | bh.b_size = size; | |
2318 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
2319 | if (rc || !buffer_mapped(&bh)) | |
2320 | goto fail; | |
2321 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
2322 | if (rc) | |
2323 | goto fail; | |
2324 | size -= bh.b_size; | |
2325 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2326 | } while(size > 0); | |
2327 | ||
98583b3e AD |
2328 | end = ktime_get(); |
2329 | fs_info(sdp, "journal %d mapped with %u extents in %lldms\n", jd->jd_jid, | |
2330 | jd->nr_extents, ktime_ms_delta(end, start)); | |
b50f227b SW |
2331 | return 0; |
2332 | ||
2333 | fail: | |
2334 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
2335 | rc, jd->jd_jid, | |
2336 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
2337 | jd->nr_extents); | |
2338 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
2339 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
2340 | bh.b_state, (unsigned long long)bh.b_size); | |
2341 | gfs2_free_journal_extents(jd); | |
2342 | return rc; | |
2343 | } | |
2344 | ||
b3b94faa DT |
2345 | /** |
2346 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
2347 | * @ip: the file being written to | |
2348 | * @offset: the offset to write to | |
2349 | * @len: the number of bytes being written | |
b3b94faa | 2350 | * |
461cb419 | 2351 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
2352 | */ |
2353 | ||
cd915493 | 2354 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 2355 | unsigned int len) |
b3b94faa | 2356 | { |
feaa7bba | 2357 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
2358 | struct buffer_head bh; |
2359 | unsigned int shift; | |
2360 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 2361 | u64 end_of_file; |
b3b94faa | 2362 | |
b3b94faa DT |
2363 | if (!len) |
2364 | return 0; | |
2365 | ||
2366 | if (gfs2_is_stuffed(ip)) { | |
235628c5 | 2367 | if (offset + len > gfs2_max_stuffed_size(ip)) |
461cb419 | 2368 | return 1; |
b3b94faa DT |
2369 | return 0; |
2370 | } | |
2371 | ||
941e6d7d | 2372 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 2373 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 2374 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
2375 | lblock = offset >> shift; |
2376 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
77612578 | 2377 | if (lblock_stop > end_of_file && ip != GFS2_I(sdp->sd_rindex)) |
461cb419 | 2378 | return 1; |
b3b94faa | 2379 | |
941e6d7d SW |
2380 | size = (lblock_stop - lblock) << shift; |
2381 | do { | |
2382 | bh.b_state = 0; | |
2383 | bh.b_size = size; | |
2384 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
2385 | if (!buffer_mapped(&bh)) | |
461cb419 | 2386 | return 1; |
941e6d7d SW |
2387 | size -= bh.b_size; |
2388 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2389 | } while(size > 0); | |
b3b94faa DT |
2390 | |
2391 | return 0; | |
2392 | } | |
2393 | ||
4e56a641 AG |
2394 | static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length) |
2395 | { | |
2396 | struct gfs2_inode *ip = GFS2_I(inode); | |
2397 | struct buffer_head *dibh; | |
2398 | int error; | |
2399 | ||
2400 | if (offset >= inode->i_size) | |
2401 | return 0; | |
2402 | if (offset + length > inode->i_size) | |
2403 | length = inode->i_size - offset; | |
2404 | ||
2405 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2406 | if (error) | |
2407 | return error; | |
2408 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
2409 | memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0, | |
2410 | length); | |
2411 | brelse(dibh); | |
2412 | return 0; | |
2413 | } | |
2414 | ||
2415 | static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset, | |
2416 | loff_t length) | |
2417 | { | |
2418 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2419 | loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
2420 | int error; | |
2421 | ||
2422 | while (length) { | |
2423 | struct gfs2_trans *tr; | |
2424 | loff_t chunk; | |
2425 | unsigned int offs; | |
2426 | ||
2427 | chunk = length; | |
2428 | if (chunk > max_chunk) | |
2429 | chunk = max_chunk; | |
2430 | ||
2431 | offs = offset & ~PAGE_MASK; | |
2432 | if (offs && chunk > PAGE_SIZE) | |
2433 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
2434 | ||
2435 | truncate_pagecache_range(inode, offset, chunk); | |
2436 | offset += chunk; | |
2437 | length -= chunk; | |
2438 | ||
2439 | tr = current->journal_info; | |
2440 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
2441 | continue; | |
2442 | ||
2443 | gfs2_trans_end(sdp); | |
2444 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
2445 | if (error) | |
2446 | return error; | |
2447 | } | |
2448 | return 0; | |
2449 | } | |
2450 | ||
2451 | int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length) | |
2452 | { | |
2453 | struct inode *inode = file_inode(file); | |
2454 | struct gfs2_inode *ip = GFS2_I(inode); | |
2455 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
39c3a948 AG |
2456 | unsigned int blocksize = i_blocksize(inode); |
2457 | loff_t start, end; | |
4e56a641 AG |
2458 | int error; |
2459 | ||
70499cdf | 2460 | if (!gfs2_is_stuffed(ip)) { |
39c3a948 | 2461 | unsigned int start_off, end_len; |
4e56a641 | 2462 | |
4e56a641 | 2463 | start_off = offset & (blocksize - 1); |
00251a16 | 2464 | end_len = (offset + length) & (blocksize - 1); |
4e56a641 AG |
2465 | if (start_off) { |
2466 | unsigned int len = length; | |
2467 | if (length > blocksize - start_off) | |
2468 | len = blocksize - start_off; | |
2469 | error = gfs2_block_zero_range(inode, offset, len); | |
2470 | if (error) | |
2471 | goto out; | |
2472 | if (start_off + length < blocksize) | |
00251a16 | 2473 | end_len = 0; |
4e56a641 | 2474 | } |
00251a16 | 2475 | if (end_len) { |
4e56a641 | 2476 | error = gfs2_block_zero_range(inode, |
00251a16 | 2477 | offset + length - end_len, end_len); |
4e56a641 AG |
2478 | if (error) |
2479 | goto out; | |
2480 | } | |
2481 | } | |
2482 | ||
70499cdf BP |
2483 | start = round_down(offset, blocksize); |
2484 | end = round_up(offset + length, blocksize) - 1; | |
2485 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
2486 | if (error) | |
2487 | return error; | |
2488 | ||
2489 | if (gfs2_is_jdata(ip)) | |
2490 | error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA, | |
2491 | GFS2_JTRUNC_REVOKES); | |
2492 | else | |
2493 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2494 | if (error) | |
2495 | return error; | |
2496 | ||
2497 | if (gfs2_is_stuffed(ip)) { | |
2498 | error = stuffed_zero_range(inode, offset, length); | |
2499 | if (error) | |
2500 | goto out; | |
2501 | } | |
2502 | ||
4e56a641 AG |
2503 | if (gfs2_is_jdata(ip)) { |
2504 | BUG_ON(!current->journal_info); | |
2505 | gfs2_journaled_truncate_range(inode, offset, length); | |
2506 | } else | |
2507 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2508 | ||
2509 | file_update_time(file); | |
2510 | mark_inode_dirty(inode); | |
2511 | ||
2512 | if (current->journal_info) | |
2513 | gfs2_trans_end(sdp); | |
2514 | ||
2515 | if (!gfs2_is_stuffed(ip)) | |
2516 | error = punch_hole(ip, offset, length); | |
2517 | ||
2518 | out: | |
2519 | if (current->journal_info) | |
2520 | gfs2_trans_end(sdp); | |
2521 | return error; | |
2522 | } | |
2164f9b9 CH |
2523 | |
2524 | static int gfs2_map_blocks(struct iomap_writepage_ctx *wpc, struct inode *inode, | |
2525 | loff_t offset) | |
2526 | { | |
2164f9b9 CH |
2527 | int ret; |
2528 | ||
2529 | if (WARN_ON_ONCE(gfs2_is_stuffed(GFS2_I(inode)))) | |
2530 | return -EIO; | |
2531 | ||
2532 | if (offset >= wpc->iomap.offset && | |
2533 | offset < wpc->iomap.offset + wpc->iomap.length) | |
2534 | return 0; | |
2535 | ||
2536 | memset(&wpc->iomap, 0, sizeof(wpc->iomap)); | |
54992257 | 2537 | ret = gfs2_iomap_get(inode, offset, INT_MAX, &wpc->iomap); |
2164f9b9 CH |
2538 | return ret; |
2539 | } | |
2540 | ||
2541 | const struct iomap_writeback_ops gfs2_writeback_ops = { | |
2542 | .map_blocks = gfs2_map_blocks, | |
2543 | }; |