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 | 45 | /** |
81cb277e | 46 | * gfs2_unstuffer_folio - unstuff a stuffed inode into a block cached by a folio |
f25ef0c1 SW |
47 | * @ip: the inode |
48 | * @dibh: the dinode buffer | |
49 | * @block: the block number that was allocated | |
81cb277e | 50 | * @folio: The folio. |
f25ef0c1 SW |
51 | * |
52 | * Returns: errno | |
53 | */ | |
81cb277e MWO |
54 | static int gfs2_unstuffer_folio(struct gfs2_inode *ip, struct buffer_head *dibh, |
55 | u64 block, struct folio *folio) | |
f25ef0c1 | 56 | { |
f25ef0c1 | 57 | struct inode *inode = &ip->i_inode; |
f25ef0c1 | 58 | |
81cb277e MWO |
59 | if (!folio_test_uptodate(folio)) { |
60 | void *kaddr = kmap_local_folio(folio, 0); | |
602c89d2 SW |
61 | u64 dsize = i_size_read(inode); |
62 | ||
602c89d2 | 63 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
81cb277e MWO |
64 | memset(kaddr + dsize, 0, folio_size(folio) - dsize); |
65 | kunmap_local(kaddr); | |
f25ef0c1 | 66 | |
81cb277e | 67 | folio_mark_uptodate(folio); |
f25ef0c1 SW |
68 | } |
69 | ||
2164f9b9 | 70 | if (gfs2_is_jdata(ip)) { |
81cb277e | 71 | struct buffer_head *bh = folio_buffers(folio); |
f25ef0c1 | 72 | |
81cb277e | 73 | if (!bh) |
0a88810d | 74 | bh = create_empty_buffers(folio, |
81cb277e | 75 | BIT(inode->i_blkbits), BIT(BH_Uptodate)); |
f25ef0c1 | 76 | |
2164f9b9 CH |
77 | if (!buffer_mapped(bh)) |
78 | map_bh(bh, inode->i_sb, block); | |
f25ef0c1 | 79 | |
2164f9b9 | 80 | set_buffer_uptodate(bh); |
350a9b0a | 81 | gfs2_trans_add_data(ip->i_gl, bh); |
2164f9b9 | 82 | } else { |
81cb277e | 83 | folio_mark_dirty(folio); |
845802b1 AG |
84 | gfs2_ordered_add_inode(ip); |
85 | } | |
f25ef0c1 | 86 | |
f25ef0c1 SW |
87 | return 0; |
88 | } | |
89 | ||
81cb277e | 90 | static int __gfs2_unstuff_inode(struct gfs2_inode *ip, struct folio *folio) |
b3b94faa DT |
91 | { |
92 | struct buffer_head *bh, *dibh; | |
48516ced | 93 | struct gfs2_dinode *di; |
cd915493 | 94 | u64 block = 0; |
18ec7d5c | 95 | int isdir = gfs2_is_dir(ip); |
b3b94faa DT |
96 | int error; |
97 | ||
b3b94faa DT |
98 | error = gfs2_meta_inode_buffer(ip, &dibh); |
99 | if (error) | |
7a607a41 | 100 | return error; |
907b9bce | 101 | |
a2e0f799 | 102 | if (i_size_read(&ip->i_inode)) { |
b3b94faa DT |
103 | /* Get a free block, fill it with the stuffed data, |
104 | and write it out to disk */ | |
105 | ||
b45e41d7 | 106 | unsigned int n = 1; |
6e87ed0f | 107 | error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL); |
09010978 SW |
108 | if (error) |
109 | goto out_brelse; | |
18ec7d5c | 110 | if (isdir) { |
fbb27873 | 111 | gfs2_trans_remove_revoke(GFS2_SB(&ip->i_inode), block, 1); |
61e085a8 | 112 | error = gfs2_dir_get_new_buffer(ip, block, &bh); |
b3b94faa DT |
113 | if (error) |
114 | goto out_brelse; | |
48516ced | 115 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header), |
b3b94faa DT |
116 | dibh, sizeof(struct gfs2_dinode)); |
117 | brelse(bh); | |
118 | } else { | |
81cb277e | 119 | error = gfs2_unstuffer_folio(ip, dibh, block, folio); |
b3b94faa DT |
120 | if (error) |
121 | goto out_brelse; | |
122 | } | |
123 | } | |
124 | ||
125 | /* Set up the pointer to the new block */ | |
126 | ||
350a9b0a | 127 | gfs2_trans_add_meta(ip->i_gl, dibh); |
48516ced | 128 | di = (struct gfs2_dinode *)dibh->b_data; |
b3b94faa DT |
129 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
130 | ||
a2e0f799 | 131 | if (i_size_read(&ip->i_inode)) { |
48516ced | 132 | *(__be64 *)(di + 1) = cpu_to_be64(block); |
77658aad SW |
133 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
134 | di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode)); | |
b3b94faa DT |
135 | } |
136 | ||
ecc30c79 | 137 | ip->i_height = 1; |
48516ced | 138 | di->di_height = cpu_to_be16(1); |
b3b94faa | 139 | |
a91ea69f | 140 | out_brelse: |
b3b94faa | 141 | brelse(dibh); |
7a607a41 AG |
142 | return error; |
143 | } | |
144 | ||
145 | /** | |
146 | * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big | |
147 | * @ip: The GFS2 inode to unstuff | |
148 | * | |
149 | * This routine unstuffs a dinode and returns it to a "normal" state such | |
150 | * that the height can be grown in the traditional way. | |
151 | * | |
152 | * Returns: errno | |
153 | */ | |
154 | ||
155 | int gfs2_unstuff_dinode(struct gfs2_inode *ip) | |
156 | { | |
157 | struct inode *inode = &ip->i_inode; | |
81cb277e | 158 | struct folio *folio; |
7a607a41 AG |
159 | int error; |
160 | ||
161 | down_write(&ip->i_rw_mutex); | |
81cb277e MWO |
162 | folio = filemap_grab_folio(inode->i_mapping, 0); |
163 | error = PTR_ERR(folio); | |
164 | if (IS_ERR(folio)) | |
7a607a41 | 165 | goto out; |
81cb277e MWO |
166 | error = __gfs2_unstuff_inode(ip, folio); |
167 | folio_unlock(folio); | |
168 | folio_put(folio); | |
a91ea69f | 169 | out: |
b3b94faa | 170 | up_write(&ip->i_rw_mutex); |
b3b94faa DT |
171 | return error; |
172 | } | |
173 | ||
b3b94faa DT |
174 | /** |
175 | * find_metapath - Find path through the metadata tree | |
9b8c81d1 | 176 | * @sdp: The superblock |
b3b94faa | 177 | * @block: The disk block to look up |
07e23d68 | 178 | * @mp: The metapath to return the result in |
9b8c81d1 | 179 | * @height: The pre-calculated height of the metadata tree |
b3b94faa DT |
180 | * |
181 | * This routine returns a struct metapath structure that defines a path | |
182 | * through the metadata of inode "ip" to get to block "block". | |
183 | * | |
184 | * Example: | |
185 | * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a | |
186 | * filesystem with a blocksize of 4096. | |
187 | * | |
188 | * find_metapath() would return a struct metapath structure set to: | |
07e23d68 | 189 | * mp_fheight = 3, mp_list[0] = 0, mp_list[1] = 48, and mp_list[2] = 165. |
b3b94faa DT |
190 | * |
191 | * That means that in order to get to the block containing the byte at | |
192 | * offset 101342453, we would load the indirect block pointed to by pointer | |
193 | * 0 in the dinode. We would then load the indirect block pointed to by | |
194 | * pointer 48 in that indirect block. We would then load the data block | |
195 | * pointed to by pointer 165 in that indirect block. | |
196 | * | |
197 | * ---------------------------------------- | |
198 | * | Dinode | | | |
199 | * | | 4| | |
200 | * | |0 1 2 3 4 5 9| | |
201 | * | | 6| | |
202 | * ---------------------------------------- | |
203 | * | | |
204 | * | | |
205 | * V | |
206 | * ---------------------------------------- | |
207 | * | Indirect Block | | |
208 | * | 5| | |
209 | * | 4 4 4 4 4 5 5 1| | |
210 | * |0 5 6 7 8 9 0 1 2| | |
211 | * ---------------------------------------- | |
212 | * | | |
213 | * | | |
214 | * V | |
215 | * ---------------------------------------- | |
216 | * | Indirect Block | | |
217 | * | 1 1 1 1 1 5| | |
218 | * | 6 6 6 6 6 1| | |
219 | * |0 3 4 5 6 7 2| | |
220 | * ---------------------------------------- | |
221 | * | | |
222 | * | | |
223 | * V | |
224 | * ---------------------------------------- | |
225 | * | Data block containing offset | | |
226 | * | 101342453 | | |
227 | * | | | |
228 | * | | | |
229 | * ---------------------------------------- | |
230 | * | |
231 | */ | |
232 | ||
9b8c81d1 SW |
233 | static void find_metapath(const struct gfs2_sbd *sdp, u64 block, |
234 | struct metapath *mp, unsigned int height) | |
b3b94faa | 235 | { |
b3b94faa DT |
236 | unsigned int i; |
237 | ||
5f8bd444 | 238 | mp->mp_fheight = height; |
9b8c81d1 | 239 | for (i = height; i--;) |
7eabb77e | 240 | mp->mp_list[i] = do_div(block, sdp->sd_inptrs); |
b3b94faa DT |
241 | } |
242 | ||
5af4e7a0 | 243 | static inline unsigned int metapath_branch_start(const struct metapath *mp) |
9b8c81d1 | 244 | { |
5af4e7a0 BM |
245 | if (mp->mp_list[0] == 0) |
246 | return 2; | |
247 | return 1; | |
9b8c81d1 SW |
248 | } |
249 | ||
d552a2b9 | 250 | /** |
20cdc193 | 251 | * metaptr1 - Return the first possible metadata pointer in a metapath buffer |
d552a2b9 BP |
252 | * @height: The metadata height (0 = dinode) |
253 | * @mp: The metapath | |
254 | */ | |
255 | static inline __be64 *metaptr1(unsigned int height, const struct metapath *mp) | |
256 | { | |
257 | struct buffer_head *bh = mp->mp_bh[height]; | |
258 | if (height == 0) | |
259 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_dinode))); | |
260 | return ((__be64 *)(bh->b_data + sizeof(struct gfs2_meta_header))); | |
261 | } | |
262 | ||
b3b94faa DT |
263 | /** |
264 | * metapointer - Return pointer to start of metadata in a buffer | |
b3b94faa DT |
265 | * @height: The metadata height (0 = dinode) |
266 | * @mp: The metapath | |
267 | * | |
268 | * Return a pointer to the block number of the next height of the metadata | |
269 | * tree given a buffer containing the pointer to the current height of the | |
270 | * metadata tree. | |
271 | */ | |
272 | ||
9b8c81d1 | 273 | static inline __be64 *metapointer(unsigned int height, const struct metapath *mp) |
b3b94faa | 274 | { |
d552a2b9 BP |
275 | __be64 *p = metaptr1(height, mp); |
276 | return p + mp->mp_list[height]; | |
b3b94faa DT |
277 | } |
278 | ||
7841b9f0 AG |
279 | static inline const __be64 *metaend(unsigned int height, const struct metapath *mp) |
280 | { | |
281 | const struct buffer_head *bh = mp->mp_bh[height]; | |
282 | return (const __be64 *)(bh->b_data + bh->b_size); | |
283 | } | |
284 | ||
285 | static void clone_metapath(struct metapath *clone, struct metapath *mp) | |
286 | { | |
287 | unsigned int hgt; | |
288 | ||
289 | *clone = *mp; | |
290 | for (hgt = 0; hgt < mp->mp_aheight; hgt++) | |
291 | get_bh(clone->mp_bh[hgt]); | |
292 | } | |
293 | ||
5cf26b1e | 294 | static void gfs2_metapath_ra(struct gfs2_glock *gl, __be64 *start, __be64 *end) |
b99b98dc | 295 | { |
b99b98dc SW |
296 | const __be64 *t; |
297 | ||
5cf26b1e | 298 | for (t = start; t < end; t++) { |
c3ce5aa9 AG |
299 | struct buffer_head *rabh; |
300 | ||
b99b98dc SW |
301 | if (!*t) |
302 | continue; | |
303 | ||
304 | rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE); | |
305 | if (trylock_buffer(rabh)) { | |
306 | if (!buffer_uptodate(rabh)) { | |
307 | rabh->b_end_io = end_buffer_read_sync; | |
1420c4a5 BVA |
308 | submit_bh(REQ_OP_READ | REQ_RAHEAD | REQ_META | |
309 | REQ_PRIO, rabh); | |
b99b98dc SW |
310 | continue; |
311 | } | |
312 | unlock_buffer(rabh); | |
313 | } | |
314 | brelse(rabh); | |
315 | } | |
316 | } | |
317 | ||
e8b43fe0 AG |
318 | static int __fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, |
319 | unsigned int x, unsigned int h) | |
d552a2b9 | 320 | { |
e8b43fe0 AG |
321 | for (; x < h; x++) { |
322 | __be64 *ptr = metapointer(x, mp); | |
323 | u64 dblock = be64_to_cpu(*ptr); | |
324 | int ret; | |
d552a2b9 | 325 | |
e8b43fe0 AG |
326 | if (!dblock) |
327 | break; | |
6d8da302 | 328 | ret = gfs2_meta_buffer(ip, GFS2_METATYPE_IN, dblock, &mp->mp_bh[x + 1]); |
e8b43fe0 AG |
329 | if (ret) |
330 | return ret; | |
331 | } | |
332 | mp->mp_aheight = x + 1; | |
333 | return 0; | |
d552a2b9 BP |
334 | } |
335 | ||
b3b94faa | 336 | /** |
9b8c81d1 SW |
337 | * lookup_metapath - Walk the metadata tree to a specific point |
338 | * @ip: The inode | |
b3b94faa | 339 | * @mp: The metapath |
b3b94faa | 340 | * |
9b8c81d1 SW |
341 | * Assumes that the inode's buffer has already been looked up and |
342 | * hooked onto mp->mp_bh[0] and that the metapath has been initialised | |
343 | * by find_metapath(). | |
344 | * | |
345 | * If this function encounters part of the tree which has not been | |
346 | * allocated, it returns the current height of the tree at the point | |
347 | * at which it found the unallocated block. Blocks which are found are | |
348 | * added to the mp->mp_bh[] list. | |
b3b94faa | 349 | * |
e8b43fe0 | 350 | * Returns: error |
b3b94faa DT |
351 | */ |
352 | ||
9b8c81d1 | 353 | static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp) |
11707ea0 | 354 | { |
e8b43fe0 | 355 | return __fillup_metapath(ip, mp, 0, ip->i_height - 1); |
dbac6710 SW |
356 | } |
357 | ||
d552a2b9 BP |
358 | /** |
359 | * fillup_metapath - fill up buffers for the metadata path to a specific height | |
360 | * @ip: The inode | |
361 | * @mp: The metapath | |
362 | * @h: The height to which it should be mapped | |
363 | * | |
364 | * Similar to lookup_metapath, but does lookups for a range of heights | |
365 | * | |
c3ce5aa9 | 366 | * Returns: error or the number of buffers filled |
d552a2b9 BP |
367 | */ |
368 | ||
369 | static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h) | |
370 | { | |
e8b43fe0 | 371 | unsigned int x = 0; |
c3ce5aa9 | 372 | int ret; |
d552a2b9 BP |
373 | |
374 | if (h) { | |
375 | /* find the first buffer we need to look up. */ | |
e8b43fe0 AG |
376 | for (x = h - 1; x > 0; x--) { |
377 | if (mp->mp_bh[x]) | |
378 | break; | |
d552a2b9 BP |
379 | } |
380 | } | |
c3ce5aa9 AG |
381 | ret = __fillup_metapath(ip, mp, x, h); |
382 | if (ret) | |
383 | return ret; | |
384 | return mp->mp_aheight - x - 1; | |
d552a2b9 BP |
385 | } |
386 | ||
a27a0c9b AG |
387 | static sector_t metapath_to_block(struct gfs2_sbd *sdp, struct metapath *mp) |
388 | { | |
389 | sector_t factor = 1, block = 0; | |
390 | int hgt; | |
391 | ||
392 | for (hgt = mp->mp_fheight - 1; hgt >= 0; hgt--) { | |
393 | if (hgt < mp->mp_aheight) | |
394 | block += mp->mp_list[hgt] * factor; | |
395 | factor *= sdp->sd_inptrs; | |
396 | } | |
397 | return block; | |
398 | } | |
399 | ||
64bc06bb | 400 | static void release_metapath(struct metapath *mp) |
dbac6710 SW |
401 | { |
402 | int i; | |
403 | ||
9b8c81d1 SW |
404 | for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) { |
405 | if (mp->mp_bh[i] == NULL) | |
406 | break; | |
407 | brelse(mp->mp_bh[i]); | |
64bc06bb | 408 | mp->mp_bh[i] = NULL; |
9b8c81d1 | 409 | } |
11707ea0 SW |
410 | } |
411 | ||
30cbf189 SW |
412 | /** |
413 | * gfs2_extent_length - Returns length of an extent of blocks | |
bcfe9413 AG |
414 | * @bh: The metadata block |
415 | * @ptr: Current position in @bh | |
416 | * @limit: Max extent length to return | |
30cbf189 SW |
417 | * @eob: Set to 1 if we hit "end of block" |
418 | * | |
30cbf189 SW |
419 | * Returns: The length of the extent (minimum of one block) |
420 | */ | |
421 | ||
bcfe9413 | 422 | static inline unsigned int gfs2_extent_length(struct buffer_head *bh, __be64 *ptr, size_t limit, int *eob) |
30cbf189 | 423 | { |
bcfe9413 | 424 | const __be64 *end = (__be64 *)(bh->b_data + bh->b_size); |
30cbf189 SW |
425 | const __be64 *first = ptr; |
426 | u64 d = be64_to_cpu(*ptr); | |
427 | ||
428 | *eob = 0; | |
429 | do { | |
430 | ptr++; | |
431 | if (ptr >= end) | |
432 | break; | |
bcfe9413 | 433 | d++; |
30cbf189 SW |
434 | } while(be64_to_cpu(*ptr) == d); |
435 | if (ptr >= end) | |
436 | *eob = 1; | |
bcfe9413 | 437 | return ptr - first; |
30cbf189 SW |
438 | } |
439 | ||
a27a0c9b AG |
440 | enum walker_status { WALK_STOP, WALK_FOLLOW, WALK_CONTINUE }; |
441 | ||
442 | /* | |
443 | * gfs2_metadata_walker - walk an indirect block | |
444 | * @mp: Metapath to indirect block | |
445 | * @ptrs: Number of pointers to look at | |
446 | * | |
447 | * When returning WALK_FOLLOW, the walker must update @mp to point at the right | |
448 | * indirect block to follow. | |
449 | */ | |
450 | typedef enum walker_status (*gfs2_metadata_walker)(struct metapath *mp, | |
451 | unsigned int ptrs); | |
7841b9f0 | 452 | |
a27a0c9b AG |
453 | /* |
454 | * gfs2_walk_metadata - walk a tree of indirect blocks | |
455 | * @inode: The inode | |
456 | * @mp: Starting point of walk | |
457 | * @max_len: Maximum number of blocks to walk | |
458 | * @walker: Called during the walk | |
459 | * | |
460 | * Returns 1 if the walk was stopped by @walker, 0 if we went past @max_len or | |
461 | * past the end of metadata, and a negative error code otherwise. | |
462 | */ | |
7841b9f0 | 463 | |
a27a0c9b AG |
464 | static int gfs2_walk_metadata(struct inode *inode, struct metapath *mp, |
465 | u64 max_len, gfs2_metadata_walker walker) | |
7841b9f0 | 466 | { |
7841b9f0 AG |
467 | struct gfs2_inode *ip = GFS2_I(inode); |
468 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7841b9f0 AG |
469 | u64 factor = 1; |
470 | unsigned int hgt; | |
a27a0c9b | 471 | int ret; |
7841b9f0 | 472 | |
a27a0c9b AG |
473 | /* |
474 | * The walk starts in the lowest allocated indirect block, which may be | |
475 | * before the position indicated by @mp. Adjust @max_len accordingly | |
476 | * to avoid a short walk. | |
477 | */ | |
478 | for (hgt = mp->mp_fheight - 1; hgt >= mp->mp_aheight; hgt--) { | |
479 | max_len += mp->mp_list[hgt] * factor; | |
480 | mp->mp_list[hgt] = 0; | |
7841b9f0 | 481 | factor *= sdp->sd_inptrs; |
a27a0c9b | 482 | } |
7841b9f0 AG |
483 | |
484 | for (;;) { | |
a27a0c9b AG |
485 | u16 start = mp->mp_list[hgt]; |
486 | enum walker_status status; | |
487 | unsigned int ptrs; | |
488 | u64 len; | |
7841b9f0 AG |
489 | |
490 | /* Walk indirect block. */ | |
a27a0c9b AG |
491 | ptrs = (hgt >= 1 ? sdp->sd_inptrs : sdp->sd_diptrs) - start; |
492 | len = ptrs * factor; | |
493 | if (len > max_len) | |
494 | ptrs = DIV_ROUND_UP_ULL(max_len, factor); | |
495 | status = walker(mp, ptrs); | |
496 | switch (status) { | |
497 | case WALK_STOP: | |
498 | return 1; | |
499 | case WALK_FOLLOW: | |
500 | BUG_ON(mp->mp_aheight == mp->mp_fheight); | |
501 | ptrs = mp->mp_list[hgt] - start; | |
502 | len = ptrs * factor; | |
7841b9f0 | 503 | break; |
a27a0c9b | 504 | case WALK_CONTINUE: |
7841b9f0 | 505 | break; |
7841b9f0 | 506 | } |
a27a0c9b AG |
507 | if (len >= max_len) |
508 | break; | |
509 | max_len -= len; | |
510 | if (status == WALK_FOLLOW) | |
511 | goto fill_up_metapath; | |
7841b9f0 AG |
512 | |
513 | lower_metapath: | |
514 | /* Decrease height of metapath. */ | |
7841b9f0 AG |
515 | brelse(mp->mp_bh[hgt]); |
516 | mp->mp_bh[hgt] = NULL; | |
a27a0c9b | 517 | mp->mp_list[hgt] = 0; |
7841b9f0 AG |
518 | if (!hgt) |
519 | break; | |
520 | hgt--; | |
521 | factor *= sdp->sd_inptrs; | |
522 | ||
523 | /* Advance in metadata tree. */ | |
524 | (mp->mp_list[hgt])++; | |
566a2ab3 AG |
525 | if (hgt) { |
526 | if (mp->mp_list[hgt] >= sdp->sd_inptrs) | |
527 | goto lower_metapath; | |
528 | } else { | |
529 | if (mp->mp_list[hgt] >= sdp->sd_diptrs) | |
7841b9f0 | 530 | break; |
7841b9f0 AG |
531 | } |
532 | ||
533 | fill_up_metapath: | |
534 | /* Increase height of metapath. */ | |
7841b9f0 AG |
535 | ret = fillup_metapath(ip, mp, ip->i_height - 1); |
536 | if (ret < 0) | |
a27a0c9b | 537 | return ret; |
7841b9f0 AG |
538 | hgt += ret; |
539 | for (; ret; ret--) | |
540 | do_div(factor, sdp->sd_inptrs); | |
541 | mp->mp_aheight = hgt + 1; | |
542 | } | |
a27a0c9b | 543 | return 0; |
7841b9f0 AG |
544 | } |
545 | ||
a27a0c9b AG |
546 | static enum walker_status gfs2_hole_walker(struct metapath *mp, |
547 | unsigned int ptrs) | |
7841b9f0 | 548 | { |
a27a0c9b AG |
549 | const __be64 *start, *ptr, *end; |
550 | unsigned int hgt; | |
551 | ||
552 | hgt = mp->mp_aheight - 1; | |
553 | start = metapointer(hgt, mp); | |
554 | end = start + ptrs; | |
7841b9f0 AG |
555 | |
556 | for (ptr = start; ptr < end; ptr++) { | |
557 | if (*ptr) { | |
a27a0c9b | 558 | mp->mp_list[hgt] += ptr - start; |
7841b9f0 AG |
559 | if (mp->mp_aheight == mp->mp_fheight) |
560 | return WALK_STOP; | |
a27a0c9b | 561 | return WALK_FOLLOW; |
7841b9f0 AG |
562 | } |
563 | } | |
a27a0c9b | 564 | return WALK_CONTINUE; |
7841b9f0 AG |
565 | } |
566 | ||
567 | /** | |
568 | * gfs2_hole_size - figure out the size of a hole | |
569 | * @inode: The inode | |
570 | * @lblock: The logical starting block number | |
571 | * @len: How far to look (in blocks) | |
572 | * @mp: The metapath at lblock | |
573 | * @iomap: The iomap to store the hole size in | |
574 | * | |
575 | * This function modifies @mp. | |
576 | * | |
577 | * Returns: errno on error | |
578 | */ | |
579 | static int gfs2_hole_size(struct inode *inode, sector_t lblock, u64 len, | |
580 | struct metapath *mp, struct iomap *iomap) | |
581 | { | |
a27a0c9b AG |
582 | struct metapath clone; |
583 | u64 hole_size; | |
584 | int ret; | |
585 | ||
586 | clone_metapath(&clone, mp); | |
587 | ret = gfs2_walk_metadata(inode, &clone, len, gfs2_hole_walker); | |
588 | if (ret < 0) | |
589 | goto out; | |
7841b9f0 | 590 | |
a27a0c9b AG |
591 | if (ret == 1) |
592 | hole_size = metapath_to_block(GFS2_SB(inode), &clone) - lblock; | |
593 | else | |
594 | hole_size = len; | |
595 | iomap->length = hole_size << inode->i_blkbits; | |
596 | ret = 0; | |
597 | ||
598 | out: | |
599 | release_metapath(&clone); | |
7841b9f0 AG |
600 | return ret; |
601 | } | |
602 | ||
b2963932 BP |
603 | static inline void gfs2_indirect_init(struct metapath *mp, |
604 | struct gfs2_glock *gl, unsigned int i, | |
605 | unsigned offset, u64 bn) | |
9b8c81d1 SW |
606 | { |
607 | __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data + | |
608 | ((i > 1) ? sizeof(struct gfs2_meta_header) : | |
609 | sizeof(struct gfs2_dinode))); | |
610 | BUG_ON(i < 1); | |
611 | BUG_ON(mp->mp_bh[i] != NULL); | |
612 | mp->mp_bh[i] = gfs2_meta_new(gl, bn); | |
350a9b0a | 613 | gfs2_trans_add_meta(gl, mp->mp_bh[i]); |
9b8c81d1 SW |
614 | gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
615 | gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header)); | |
616 | ptr += offset; | |
617 | *ptr = cpu_to_be64(bn); | |
9b8c81d1 SW |
618 | } |
619 | ||
620 | enum alloc_state { | |
621 | ALLOC_DATA = 0, | |
622 | ALLOC_GROW_DEPTH = 1, | |
623 | ALLOC_GROW_HEIGHT = 2, | |
624 | /* ALLOC_UNSTUFF = 3, TBD and rather complicated */ | |
625 | }; | |
626 | ||
627 | /** | |
54992257 | 628 | * __gfs2_iomap_alloc - Build a metadata tree of the requested height |
9b8c81d1 | 629 | * @inode: The GFS2 inode |
628e366d | 630 | * @iomap: The iomap structure |
5f8bd444 | 631 | * @mp: The metapath, with proper height information calculated |
9b8c81d1 SW |
632 | * |
633 | * In this routine we may have to alloc: | |
634 | * i) Indirect blocks to grow the metadata tree height | |
635 | * ii) Indirect blocks to fill in lower part of the metadata tree | |
636 | * iii) Data blocks | |
637 | * | |
54992257 | 638 | * This function is called after __gfs2_iomap_get, which works out the |
64bc06bb AG |
639 | * total number of blocks which we need via gfs2_alloc_size. |
640 | * | |
641 | * We then do the actual allocation asking for an extent at a time (if | |
642 | * enough contiguous free blocks are available, there will only be one | |
643 | * allocation request per call) and uses the state machine to initialise | |
644 | * the blocks in order. | |
9b8c81d1 | 645 | * |
628e366d AG |
646 | * Right now, this function will allocate at most one indirect block |
647 | * worth of data -- with a default block size of 4K, that's slightly | |
648 | * less than 2M. If this limitation is ever removed to allow huge | |
649 | * allocations, we would probably still want to limit the iomap size we | |
650 | * return to avoid stalling other tasks during huge writes; the next | |
651 | * iomap iteration would then find the blocks already allocated. | |
652 | * | |
9b8c81d1 SW |
653 | * Returns: errno on error |
654 | */ | |
655 | ||
54992257 AG |
656 | static int __gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap, |
657 | struct metapath *mp) | |
9b8c81d1 SW |
658 | { |
659 | struct gfs2_inode *ip = GFS2_I(inode); | |
660 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
661 | struct buffer_head *dibh = mp->mp_bh[0]; | |
5f8bd444 | 662 | u64 bn; |
5af4e7a0 | 663 | unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0; |
64bc06bb | 664 | size_t dblks = iomap->length >> inode->i_blkbits; |
5f8bd444 | 665 | const unsigned end_of_metadata = mp->mp_fheight - 1; |
628e366d | 666 | int ret; |
9b8c81d1 SW |
667 | enum alloc_state state; |
668 | __be64 *ptr; | |
669 | __be64 zero_bn = 0; | |
670 | ||
5f8bd444 | 671 | BUG_ON(mp->mp_aheight < 1); |
9b8c81d1 | 672 | BUG_ON(dibh == NULL); |
64bc06bb | 673 | BUG_ON(dblks < 1); |
9b8c81d1 | 674 | |
350a9b0a | 675 | gfs2_trans_add_meta(ip->i_gl, dibh); |
9b8c81d1 | 676 | |
628e366d AG |
677 | down_write(&ip->i_rw_mutex); |
678 | ||
5f8bd444 | 679 | if (mp->mp_fheight == mp->mp_aheight) { |
64bc06bb | 680 | /* Bottom indirect block exists */ |
9b8c81d1 SW |
681 | state = ALLOC_DATA; |
682 | } else { | |
683 | /* Need to allocate indirect blocks */ | |
5f8bd444 | 684 | if (mp->mp_fheight == ip->i_height) { |
9b8c81d1 | 685 | /* Writing into existing tree, extend tree down */ |
5f8bd444 | 686 | iblks = mp->mp_fheight - mp->mp_aheight; |
9b8c81d1 SW |
687 | state = ALLOC_GROW_DEPTH; |
688 | } else { | |
689 | /* Building up tree height */ | |
690 | state = ALLOC_GROW_HEIGHT; | |
5f8bd444 | 691 | iblks = mp->mp_fheight - ip->i_height; |
5af4e7a0 | 692 | branch_start = metapath_branch_start(mp); |
5f8bd444 | 693 | iblks += (mp->mp_fheight - branch_start); |
9b8c81d1 SW |
694 | } |
695 | } | |
696 | ||
697 | /* start of the second part of the function (state machine) */ | |
698 | ||
3974320c | 699 | blks = dblks + iblks; |
5f8bd444 | 700 | i = mp->mp_aheight; |
9b8c81d1 SW |
701 | do { |
702 | n = blks - alloced; | |
628e366d AG |
703 | ret = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
704 | if (ret) | |
705 | goto out; | |
9b8c81d1 SW |
706 | alloced += n; |
707 | if (state != ALLOC_DATA || gfs2_is_jdata(ip)) | |
fbb27873 | 708 | gfs2_trans_remove_revoke(sdp, bn, n); |
9b8c81d1 SW |
709 | switch (state) { |
710 | /* Growing height of tree */ | |
711 | case ALLOC_GROW_HEIGHT: | |
712 | if (i == 1) { | |
713 | ptr = (__be64 *)(dibh->b_data + | |
714 | sizeof(struct gfs2_dinode)); | |
715 | zero_bn = *ptr; | |
716 | } | |
5f8bd444 BP |
717 | for (; i - 1 < mp->mp_fheight - ip->i_height && n > 0; |
718 | i++, n--) | |
9b8c81d1 | 719 | gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++); |
5f8bd444 | 720 | if (i - 1 == mp->mp_fheight - ip->i_height) { |
9b8c81d1 SW |
721 | i--; |
722 | gfs2_buffer_copy_tail(mp->mp_bh[i], | |
723 | sizeof(struct gfs2_meta_header), | |
724 | dibh, sizeof(struct gfs2_dinode)); | |
725 | gfs2_buffer_clear_tail(dibh, | |
726 | sizeof(struct gfs2_dinode) + | |
727 | sizeof(__be64)); | |
728 | ptr = (__be64 *)(mp->mp_bh[i]->b_data + | |
729 | sizeof(struct gfs2_meta_header)); | |
730 | *ptr = zero_bn; | |
731 | state = ALLOC_GROW_DEPTH; | |
5f8bd444 | 732 | for(i = branch_start; i < mp->mp_fheight; i++) { |
9b8c81d1 SW |
733 | if (mp->mp_bh[i] == NULL) |
734 | break; | |
735 | brelse(mp->mp_bh[i]); | |
736 | mp->mp_bh[i] = NULL; | |
737 | } | |
5af4e7a0 | 738 | i = branch_start; |
9b8c81d1 SW |
739 | } |
740 | if (n == 0) | |
741 | break; | |
df561f66 | 742 | fallthrough; /* To branching from existing tree */ |
9b8c81d1 | 743 | case ALLOC_GROW_DEPTH: |
5f8bd444 | 744 | if (i > 1 && i < mp->mp_fheight) |
350a9b0a | 745 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]); |
5f8bd444 | 746 | for (; i < mp->mp_fheight && n > 0; i++, n--) |
9b8c81d1 SW |
747 | gfs2_indirect_init(mp, ip->i_gl, i, |
748 | mp->mp_list[i-1], bn++); | |
5f8bd444 | 749 | if (i == mp->mp_fheight) |
9b8c81d1 SW |
750 | state = ALLOC_DATA; |
751 | if (n == 0) | |
752 | break; | |
df561f66 | 753 | fallthrough; /* To tree complete, adding data blocks */ |
9b8c81d1 | 754 | case ALLOC_DATA: |
3974320c | 755 | BUG_ON(n > dblks); |
9b8c81d1 | 756 | BUG_ON(mp->mp_bh[end_of_metadata] == NULL); |
350a9b0a | 757 | gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]); |
3974320c | 758 | dblks = n; |
9b8c81d1 | 759 | ptr = metapointer(end_of_metadata, mp); |
3974320c | 760 | iomap->addr = bn << inode->i_blkbits; |
628e366d | 761 | iomap->flags |= IOMAP_F_MERGED | IOMAP_F_NEW; |
9b8c81d1 SW |
762 | while (n-- > 0) |
763 | *ptr++ = cpu_to_be64(bn++); | |
764 | break; | |
765 | } | |
3974320c | 766 | } while (iomap->addr == IOMAP_NULL_ADDR); |
9b8c81d1 | 767 | |
d505a96a | 768 | iomap->type = IOMAP_MAPPED; |
3974320c | 769 | iomap->length = (u64)dblks << inode->i_blkbits; |
5f8bd444 | 770 | ip->i_height = mp->mp_fheight; |
9b8c81d1 | 771 | gfs2_add_inode_blocks(&ip->i_inode, alloced); |
628e366d AG |
772 | gfs2_dinode_out(ip, dibh->b_data); |
773 | out: | |
774 | up_write(&ip->i_rw_mutex); | |
775 | return ret; | |
9b8c81d1 SW |
776 | } |
777 | ||
7ee66c03 CH |
778 | #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE |
779 | ||
64bc06bb AG |
780 | /** |
781 | * gfs2_alloc_size - Compute the maximum allocation size | |
782 | * @inode: The inode | |
783 | * @mp: The metapath | |
784 | * @size: Requested size in blocks | |
785 | * | |
786 | * Compute the maximum size of the next allocation at @mp. | |
787 | * | |
788 | * Returns: size in blocks | |
789 | */ | |
790 | static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size) | |
3974320c BP |
791 | { |
792 | struct gfs2_inode *ip = GFS2_I(inode); | |
64bc06bb AG |
793 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
794 | const __be64 *first, *ptr, *end; | |
795 | ||
796 | /* | |
797 | * For writes to stuffed files, this function is called twice via | |
54992257 | 798 | * __gfs2_iomap_get, before and after unstuffing. The size we return the |
64bc06bb AG |
799 | * first time needs to be large enough to get the reservation and |
800 | * allocation sizes right. The size we return the second time must | |
54992257 | 801 | * be exact or else __gfs2_iomap_alloc won't do the right thing. |
64bc06bb AG |
802 | */ |
803 | ||
804 | if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) { | |
805 | unsigned int maxsize = mp->mp_fheight > 1 ? | |
806 | sdp->sd_inptrs : sdp->sd_diptrs; | |
807 | maxsize -= mp->mp_list[mp->mp_fheight - 1]; | |
808 | if (size > maxsize) | |
809 | size = maxsize; | |
810 | return size; | |
811 | } | |
3974320c | 812 | |
64bc06bb AG |
813 | first = metapointer(ip->i_height - 1, mp); |
814 | end = metaend(ip->i_height - 1, mp); | |
815 | if (end - first > size) | |
816 | end = first + size; | |
817 | for (ptr = first; ptr < end; ptr++) { | |
818 | if (*ptr) | |
819 | break; | |
820 | } | |
821 | return ptr - first; | |
3974320c BP |
822 | } |
823 | ||
824 | /** | |
54992257 | 825 | * __gfs2_iomap_get - Map blocks from an inode to disk blocks |
3974320c BP |
826 | * @inode: The inode |
827 | * @pos: Starting position in bytes | |
828 | * @length: Length to map, in bytes | |
829 | * @flags: iomap flags | |
830 | * @iomap: The iomap structure | |
628e366d | 831 | * @mp: The metapath |
3974320c BP |
832 | * |
833 | * Returns: errno | |
834 | */ | |
54992257 AG |
835 | static int __gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length, |
836 | unsigned flags, struct iomap *iomap, | |
837 | struct metapath *mp) | |
b3b94faa | 838 | { |
feaa7bba SW |
839 | struct gfs2_inode *ip = GFS2_I(inode); |
840 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
d505a96a | 841 | loff_t size = i_size_read(inode); |
9b8c81d1 | 842 | __be64 *ptr; |
3974320c | 843 | sector_t lblock; |
628e366d AG |
844 | sector_t lblock_stop; |
845 | int ret; | |
9b8c81d1 | 846 | int eob; |
628e366d | 847 | u64 len; |
d505a96a | 848 | struct buffer_head *dibh = NULL, *bh; |
9b8c81d1 | 849 | u8 height; |
7276b3b0 | 850 | |
628e366d AG |
851 | if (!length) |
852 | return -EINVAL; | |
b3b94faa | 853 | |
d505a96a AG |
854 | down_read(&ip->i_rw_mutex); |
855 | ||
856 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
857 | if (ret) | |
858 | goto unlock; | |
c26b5aa8 | 859 | mp->mp_bh[0] = dibh; |
d505a96a | 860 | |
49edd5bf | 861 | if (gfs2_is_stuffed(ip)) { |
d505a96a AG |
862 | if (flags & IOMAP_WRITE) { |
863 | loff_t max_size = gfs2_max_stuffed_size(ip); | |
864 | ||
865 | if (pos + length > max_size) | |
866 | goto unstuff; | |
867 | iomap->length = max_size; | |
868 | } else { | |
869 | if (pos >= size) { | |
870 | if (flags & IOMAP_REPORT) { | |
871 | ret = -ENOENT; | |
872 | goto unlock; | |
873 | } else { | |
d505a96a AG |
874 | iomap->offset = pos; |
875 | iomap->length = length; | |
566a2ab3 | 876 | goto hole_found; |
d505a96a AG |
877 | } |
878 | } | |
879 | iomap->length = size; | |
49edd5bf | 880 | } |
d505a96a AG |
881 | iomap->addr = (ip->i_no_addr << inode->i_blkbits) + |
882 | sizeof(struct gfs2_dinode); | |
883 | iomap->type = IOMAP_INLINE; | |
64bc06bb | 884 | iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode); |
d505a96a | 885 | goto out; |
3974320c | 886 | } |
d505a96a AG |
887 | |
888 | unstuff: | |
3974320c | 889 | lblock = pos >> inode->i_blkbits; |
3974320c | 890 | iomap->offset = lblock << inode->i_blkbits; |
628e366d AG |
891 | lblock_stop = (pos + length - 1) >> inode->i_blkbits; |
892 | len = lblock_stop - lblock + 1; | |
d505a96a | 893 | iomap->length = len << inode->i_blkbits; |
628e366d | 894 | |
9b8c81d1 | 895 | height = ip->i_height; |
9a38662b | 896 | while ((lblock + 1) * sdp->sd_sb.sb_bsize > sdp->sd_heightsize[height]) |
9b8c81d1 | 897 | height++; |
628e366d | 898 | find_metapath(sdp, lblock, mp, height); |
9b8c81d1 SW |
899 | if (height > ip->i_height || gfs2_is_stuffed(ip)) |
900 | goto do_alloc; | |
3974320c | 901 | |
628e366d | 902 | ret = lookup_metapath(ip, mp); |
e8b43fe0 | 903 | if (ret) |
628e366d | 904 | goto unlock; |
3974320c | 905 | |
628e366d | 906 | if (mp->mp_aheight != ip->i_height) |
9b8c81d1 | 907 | goto do_alloc; |
3974320c | 908 | |
628e366d | 909 | ptr = metapointer(ip->i_height - 1, mp); |
9b8c81d1 SW |
910 | if (*ptr == 0) |
911 | goto do_alloc; | |
3974320c | 912 | |
628e366d | 913 | bh = mp->mp_bh[ip->i_height - 1]; |
bcfe9413 | 914 | len = gfs2_extent_length(bh, ptr, len, &eob); |
3974320c | 915 | |
628e366d AG |
916 | iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits; |
917 | iomap->length = len << inode->i_blkbits; | |
918 | iomap->type = IOMAP_MAPPED; | |
0ed91eca | 919 | iomap->flags |= IOMAP_F_MERGED; |
9b8c81d1 | 920 | if (eob) |
7ee66c03 | 921 | iomap->flags |= IOMAP_F_GFS2_BOUNDARY; |
3974320c | 922 | |
3974320c | 923 | out: |
628e366d AG |
924 | iomap->bdev = inode->i_sb->s_bdev; |
925 | unlock: | |
926 | up_read(&ip->i_rw_mutex); | |
9b8c81d1 | 927 | return ret; |
30cbf189 | 928 | |
9b8c81d1 | 929 | do_alloc: |
628e366d | 930 | if (flags & IOMAP_REPORT) { |
49edd5bf | 931 | if (pos >= size) |
3974320c | 932 | ret = -ENOENT; |
628e366d AG |
933 | else if (height == ip->i_height) |
934 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
49edd5bf | 935 | else |
f3506eee | 936 | iomap->length = size - iomap->offset; |
64bc06bb AG |
937 | } else if (flags & IOMAP_WRITE) { |
938 | u64 alloc_size; | |
939 | ||
967bcc91 AG |
940 | if (flags & IOMAP_DIRECT) |
941 | goto out; /* (see gfs2_file_direct_write) */ | |
942 | ||
64bc06bb AG |
943 | len = gfs2_alloc_size(inode, mp, len); |
944 | alloc_size = len << inode->i_blkbits; | |
945 | if (alloc_size < iomap->length) | |
946 | iomap->length = alloc_size; | |
947 | } else { | |
d505a96a AG |
948 | if (pos < size && height == ip->i_height) |
949 | ret = gfs2_hole_size(inode, lblock, len, mp, iomap); | |
b3b94faa | 950 | } |
566a2ab3 AG |
951 | hole_found: |
952 | iomap->addr = IOMAP_NULL_ADDR; | |
953 | iomap->type = IOMAP_HOLE; | |
628e366d | 954 | goto out; |
3974320c BP |
955 | } |
956 | ||
9060bc4d | 957 | static struct folio * |
c82abc23 | 958 | gfs2_iomap_get_folio(struct iomap_iter *iter, loff_t pos, unsigned len) |
d0a22a4b | 959 | { |
9060bc4d | 960 | struct inode *inode = iter->inode; |
2741b672 | 961 | unsigned int blockmask = i_blocksize(inode) - 1; |
d0a22a4b | 962 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
2741b672 | 963 | unsigned int blocks; |
9060bc4d AG |
964 | struct folio *folio; |
965 | int status; | |
d0a22a4b | 966 | |
2741b672 | 967 | blocks = ((pos & blockmask) + len + blockmask) >> inode->i_blkbits; |
9060bc4d AG |
968 | status = gfs2_trans_begin(sdp, RES_DINODE + blocks, 0); |
969 | if (status) | |
970 | return ERR_PTR(status); | |
971 | ||
d6bb59a9 | 972 | folio = iomap_get_folio(iter, pos, len); |
9060bc4d AG |
973 | if (IS_ERR(folio)) |
974 | gfs2_trans_end(sdp); | |
975 | return folio; | |
d0a22a4b AG |
976 | } |
977 | ||
40405ddd | 978 | static void gfs2_iomap_put_folio(struct inode *inode, loff_t pos, |
80baab88 | 979 | unsigned copied, struct folio *folio) |
64bc06bb | 980 | { |
706cb549 | 981 | struct gfs2_trans *tr = current->journal_info; |
64bc06bb | 982 | struct gfs2_inode *ip = GFS2_I(inode); |
d0a22a4b | 983 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
64bc06bb | 984 | |
80baab88 | 985 | if (!gfs2_is_stuffed(ip)) |
c1b0c3cf AG |
986 | gfs2_trans_add_databufs(ip, folio, offset_in_folio(folio, pos), |
987 | copied); | |
80baab88 AG |
988 | |
989 | folio_unlock(folio); | |
990 | folio_put(folio); | |
706cb549 AG |
991 | |
992 | if (tr->tr_num_buf_new) | |
993 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
994 | ||
d0a22a4b | 995 | gfs2_trans_end(sdp); |
64bc06bb AG |
996 | } |
997 | ||
471859f5 | 998 | static const struct iomap_folio_ops gfs2_iomap_folio_ops = { |
c82abc23 | 999 | .get_folio = gfs2_iomap_get_folio, |
40405ddd | 1000 | .put_folio = gfs2_iomap_put_folio, |
df0db3ec AG |
1001 | }; |
1002 | ||
64bc06bb AG |
1003 | static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, |
1004 | loff_t length, unsigned flags, | |
c26b5aa8 AG |
1005 | struct iomap *iomap, |
1006 | struct metapath *mp) | |
64bc06bb | 1007 | { |
64bc06bb AG |
1008 | struct gfs2_inode *ip = GFS2_I(inode); |
1009 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
34aad20b | 1010 | bool unstuff; |
64bc06bb AG |
1011 | int ret; |
1012 | ||
64bc06bb AG |
1013 | unstuff = gfs2_is_stuffed(ip) && |
1014 | pos + length > gfs2_max_stuffed_size(ip); | |
1015 | ||
34aad20b AG |
1016 | if (unstuff || iomap->type == IOMAP_HOLE) { |
1017 | unsigned int data_blocks, ind_blocks; | |
1018 | struct gfs2_alloc_parms ap = {}; | |
1019 | unsigned int rblocks; | |
1020 | struct gfs2_trans *tr; | |
64bc06bb | 1021 | |
64bc06bb AG |
1022 | gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, |
1023 | &ind_blocks); | |
34aad20b | 1024 | ap.target = data_blocks + ind_blocks; |
64bc06bb AG |
1025 | ret = gfs2_quota_lock_check(ip, &ap); |
1026 | if (ret) | |
34aad20b | 1027 | return ret; |
64bc06bb AG |
1028 | |
1029 | ret = gfs2_inplace_reserve(ip, &ap); | |
1030 | if (ret) | |
1031 | goto out_qunlock; | |
64bc06bb | 1032 | |
34aad20b AG |
1033 | rblocks = RES_DINODE + ind_blocks; |
1034 | if (gfs2_is_jdata(ip)) | |
1035 | rblocks += data_blocks; | |
1036 | if (ind_blocks || data_blocks) | |
1037 | rblocks += RES_STATFS + RES_QUOTA; | |
1038 | if (inode == sdp->sd_rindex) | |
1039 | rblocks += 2 * RES_STATFS; | |
64bc06bb AG |
1040 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
1041 | ||
d0a22a4b AG |
1042 | ret = gfs2_trans_begin(sdp, rblocks, |
1043 | iomap->length >> inode->i_blkbits); | |
64bc06bb | 1044 | if (ret) |
d0a22a4b AG |
1045 | goto out_trans_fail; |
1046 | ||
1047 | if (unstuff) { | |
7a607a41 | 1048 | ret = gfs2_unstuff_dinode(ip); |
d0a22a4b AG |
1049 | if (ret) |
1050 | goto out_trans_end; | |
1051 | release_metapath(mp); | |
54992257 AG |
1052 | ret = __gfs2_iomap_get(inode, iomap->offset, |
1053 | iomap->length, flags, iomap, mp); | |
d0a22a4b AG |
1054 | if (ret) |
1055 | goto out_trans_end; | |
1056 | } | |
64bc06bb | 1057 | |
d0a22a4b | 1058 | if (iomap->type == IOMAP_HOLE) { |
54992257 | 1059 | ret = __gfs2_iomap_alloc(inode, iomap, mp); |
d0a22a4b AG |
1060 | if (ret) { |
1061 | gfs2_trans_end(sdp); | |
1062 | gfs2_inplace_release(ip); | |
1063 | punch_hole(ip, iomap->offset, iomap->length); | |
1064 | goto out_qunlock; | |
1065 | } | |
64bc06bb | 1066 | } |
d0a22a4b AG |
1067 | |
1068 | tr = current->journal_info; | |
1069 | if (tr->tr_num_buf_new) | |
1070 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
d0a22a4b AG |
1071 | |
1072 | gfs2_trans_end(sdp); | |
64bc06bb | 1073 | } |
d0a22a4b AG |
1074 | |
1075 | if (gfs2_is_stuffed(ip) || gfs2_is_jdata(ip)) | |
471859f5 | 1076 | iomap->folio_ops = &gfs2_iomap_folio_ops; |
64bc06bb AG |
1077 | return 0; |
1078 | ||
1079 | out_trans_end: | |
1080 | gfs2_trans_end(sdp); | |
1081 | out_trans_fail: | |
34aad20b | 1082 | gfs2_inplace_release(ip); |
64bc06bb | 1083 | out_qunlock: |
34aad20b | 1084 | gfs2_quota_unlock(ip); |
64bc06bb AG |
1085 | return ret; |
1086 | } | |
1087 | ||
628e366d | 1088 | static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length, |
c039b997 GR |
1089 | unsigned flags, struct iomap *iomap, |
1090 | struct iomap *srcmap) | |
628e366d AG |
1091 | { |
1092 | struct gfs2_inode *ip = GFS2_I(inode); | |
1093 | struct metapath mp = { .mp_aheight = 1, }; | |
1094 | int ret; | |
1095 | ||
2164f9b9 CH |
1096 | if (gfs2_is_jdata(ip)) |
1097 | iomap->flags |= IOMAP_F_BUFFER_HEAD; | |
0ed91eca | 1098 | |
628e366d | 1099 | trace_gfs2_iomap_start(ip, pos, length, flags); |
54992257 | 1100 | ret = __gfs2_iomap_get(inode, pos, length, flags, iomap, &mp); |
34aad20b AG |
1101 | if (ret) |
1102 | goto out_unlock; | |
1103 | ||
72d36d05 | 1104 | switch(flags & (IOMAP_WRITE | IOMAP_ZERO)) { |
34aad20b AG |
1105 | case IOMAP_WRITE: |
1106 | if (flags & IOMAP_DIRECT) { | |
1107 | /* | |
1108 | * Silently fall back to buffered I/O for stuffed files | |
1109 | * or if we've got a hole (see gfs2_file_direct_write). | |
1110 | */ | |
1111 | if (iomap->type != IOMAP_MAPPED) | |
1112 | ret = -ENOTBLK; | |
1113 | goto out_unlock; | |
1114 | } | |
1115 | break; | |
72d36d05 AG |
1116 | case IOMAP_ZERO: |
1117 | if (iomap->type == IOMAP_HOLE) | |
1118 | goto out_unlock; | |
1119 | break; | |
34aad20b AG |
1120 | default: |
1121 | goto out_unlock; | |
1122 | } | |
1123 | ||
1124 | ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap, &mp); | |
1125 | ||
1126 | out_unlock: | |
c26b5aa8 | 1127 | release_metapath(&mp); |
628e366d AG |
1128 | trace_gfs2_iomap_end(ip, iomap, ret); |
1129 | return ret; | |
1130 | } | |
1131 | ||
64bc06bb AG |
1132 | static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length, |
1133 | ssize_t written, unsigned flags, struct iomap *iomap) | |
1134 | { | |
1135 | struct gfs2_inode *ip = GFS2_I(inode); | |
1136 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
64bc06bb | 1137 | |
72d36d05 | 1138 | switch (flags & (IOMAP_WRITE | IOMAP_ZERO)) { |
34aad20b AG |
1139 | case IOMAP_WRITE: |
1140 | if (flags & IOMAP_DIRECT) | |
1141 | return 0; | |
1142 | break; | |
72d36d05 AG |
1143 | case IOMAP_ZERO: |
1144 | if (iomap->type == IOMAP_HOLE) | |
1145 | return 0; | |
1146 | break; | |
34aad20b AG |
1147 | default: |
1148 | return 0; | |
1149 | } | |
64bc06bb | 1150 | |
d0a22a4b | 1151 | if (!gfs2_is_stuffed(ip)) |
64bc06bb AG |
1152 | gfs2_ordered_add_inode(ip); |
1153 | ||
d0a22a4b | 1154 | if (inode == sdp->sd_rindex) |
64bc06bb | 1155 | adjust_fs_space(inode); |
64bc06bb | 1156 | |
64bc06bb AG |
1157 | gfs2_inplace_release(ip); |
1158 | ||
7009fa9c AG |
1159 | if (ip->i_qadata && ip->i_qadata->qa_qd_num) |
1160 | gfs2_quota_unlock(ip); | |
1161 | ||
64bc06bb AG |
1162 | if (length != written && (iomap->flags & IOMAP_F_NEW)) { |
1163 | /* Deallocate blocks that were just allocated. */ | |
d031a886 AG |
1164 | loff_t hstart = round_up(pos + written, i_blocksize(inode)); |
1165 | loff_t hend = iomap->offset + iomap->length; | |
64bc06bb | 1166 | |
d031a886 AG |
1167 | if (hstart < hend) { |
1168 | truncate_pagecache_range(inode, hstart, hend - 1); | |
1169 | punch_hole(ip, hstart, hend - hstart); | |
64bc06bb AG |
1170 | } |
1171 | } | |
1172 | ||
706cb549 | 1173 | if (unlikely(!written)) |
b924bdab | 1174 | return 0; |
706cb549 | 1175 | |
8d3e72a1 AG |
1176 | if (iomap->flags & IOMAP_F_SIZE_CHANGED) |
1177 | mark_inode_dirty(inode); | |
706cb549 | 1178 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
64bc06bb AG |
1179 | return 0; |
1180 | } | |
1181 | ||
628e366d AG |
1182 | const struct iomap_ops gfs2_iomap_ops = { |
1183 | .iomap_begin = gfs2_iomap_begin, | |
64bc06bb | 1184 | .iomap_end = gfs2_iomap_end, |
628e366d AG |
1185 | }; |
1186 | ||
3974320c | 1187 | /** |
d39d18e0 | 1188 | * gfs2_block_map - Map one or more blocks of an inode to a disk block |
3974320c BP |
1189 | * @inode: The inode |
1190 | * @lblock: The logical block number | |
1191 | * @bh_map: The bh to be mapped | |
1192 | * @create: True if its ok to alloc blocks to satify the request | |
1193 | * | |
d39d18e0 AG |
1194 | * The size of the requested mapping is defined in bh_map->b_size. |
1195 | * | |
1196 | * Clears buffer_mapped(bh_map) and leaves bh_map->b_size unchanged | |
1197 | * when @lblock is not mapped. Sets buffer_mapped(bh_map) and | |
1198 | * bh_map->b_size to indicate the size of the mapping when @lblock and | |
1199 | * successive blocks are mapped, up to the requested size. | |
1200 | * | |
1201 | * Sets buffer_boundary() if a read of metadata will be required | |
1202 | * before the next block can be mapped. Sets buffer_new() if new | |
1203 | * blocks were allocated. | |
3974320c BP |
1204 | * |
1205 | * Returns: errno | |
1206 | */ | |
1207 | ||
1208 | int gfs2_block_map(struct inode *inode, sector_t lblock, | |
1209 | struct buffer_head *bh_map, int create) | |
1210 | { | |
1211 | struct gfs2_inode *ip = GFS2_I(inode); | |
628e366d AG |
1212 | loff_t pos = (loff_t)lblock << inode->i_blkbits; |
1213 | loff_t length = bh_map->b_size; | |
628e366d AG |
1214 | struct iomap iomap = { }; |
1215 | int ret; | |
3974320c BP |
1216 | |
1217 | clear_buffer_mapped(bh_map); | |
1218 | clear_buffer_new(bh_map); | |
1219 | clear_buffer_boundary(bh_map); | |
1220 | trace_gfs2_bmap(ip, bh_map, lblock, create, 1); | |
1221 | ||
54992257 AG |
1222 | if (!create) |
1223 | ret = gfs2_iomap_get(inode, pos, length, &iomap); | |
1224 | else | |
1225 | ret = gfs2_iomap_alloc(inode, pos, length, &iomap); | |
628e366d AG |
1226 | if (ret) |
1227 | goto out; | |
3974320c BP |
1228 | |
1229 | if (iomap.length > bh_map->b_size) { | |
1230 | iomap.length = bh_map->b_size; | |
7ee66c03 | 1231 | iomap.flags &= ~IOMAP_F_GFS2_BOUNDARY; |
5f8bd444 | 1232 | } |
3974320c BP |
1233 | if (iomap.addr != IOMAP_NULL_ADDR) |
1234 | map_bh(bh_map, inode->i_sb, iomap.addr >> inode->i_blkbits); | |
1235 | bh_map->b_size = iomap.length; | |
7ee66c03 | 1236 | if (iomap.flags & IOMAP_F_GFS2_BOUNDARY) |
3974320c BP |
1237 | set_buffer_boundary(bh_map); |
1238 | if (iomap.flags & IOMAP_F_NEW) | |
1239 | set_buffer_new(bh_map); | |
1240 | ||
1241 | out: | |
1242 | trace_gfs2_bmap(ip, bh_map, lblock, create, ret); | |
1243 | return ret; | |
fd88de56 SW |
1244 | } |
1245 | ||
9153dac1 AG |
1246 | int gfs2_get_extent(struct inode *inode, u64 lblock, u64 *dblock, |
1247 | unsigned int *extlen) | |
fd88de56 | 1248 | { |
9153dac1 AG |
1249 | unsigned int blkbits = inode->i_blkbits; |
1250 | struct iomap iomap = { }; | |
1251 | unsigned int len; | |
7a6bbacb | 1252 | int ret; |
9153dac1 AG |
1253 | |
1254 | ret = gfs2_iomap_get(inode, lblock << blkbits, *extlen << blkbits, | |
1255 | &iomap); | |
1256 | if (ret) | |
1257 | return ret; | |
1258 | if (iomap.type != IOMAP_MAPPED) | |
1259 | return -EIO; | |
1260 | *dblock = iomap.addr >> blkbits; | |
1261 | len = iomap.length >> blkbits; | |
1262 | if (len < *extlen) | |
1263 | *extlen = len; | |
1264 | return 0; | |
1265 | } | |
1266 | ||
1267 | int gfs2_alloc_extent(struct inode *inode, u64 lblock, u64 *dblock, | |
1268 | unsigned int *extlen, bool *new) | |
1269 | { | |
1270 | unsigned int blkbits = inode->i_blkbits; | |
1271 | struct iomap iomap = { }; | |
1272 | unsigned int len; | |
1273 | int ret; | |
1274 | ||
1275 | ret = gfs2_iomap_alloc(inode, lblock << blkbits, *extlen << blkbits, | |
1276 | &iomap); | |
1277 | if (ret) | |
1278 | return ret; | |
1279 | if (iomap.type != IOMAP_MAPPED) | |
1280 | return -EIO; | |
1281 | *dblock = iomap.addr >> blkbits; | |
1282 | len = iomap.length >> blkbits; | |
1283 | if (len < *extlen) | |
1284 | *extlen = len; | |
1285 | *new = iomap.flags & IOMAP_F_NEW; | |
1286 | return 0; | |
b3b94faa DT |
1287 | } |
1288 | ||
70499cdf BP |
1289 | /* |
1290 | * NOTE: Never call gfs2_block_zero_range with an open transaction because it | |
1291 | * uses iomap write to perform its actions, which begin their own transactions | |
c82abc23 | 1292 | * (iomap_begin, get_folio, etc.) |
70499cdf | 1293 | */ |
bdba0d5e AG |
1294 | static int gfs2_block_zero_range(struct inode *inode, loff_t from, |
1295 | unsigned int length) | |
ba7f7290 | 1296 | { |
70499cdf | 1297 | BUG_ON(current->journal_info); |
2257e468 | 1298 | return iomap_zero_range(inode, from, length, NULL, &gfs2_iomap_ops); |
ba7f7290 SW |
1299 | } |
1300 | ||
c62baf65 FF |
1301 | #define GFS2_JTRUNC_REVOKES 8192 |
1302 | ||
fa731fc4 SW |
1303 | /** |
1304 | * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files | |
1305 | * @inode: The inode being truncated | |
1306 | * @oldsize: The original (larger) size | |
1307 | * @newsize: The new smaller size | |
1308 | * | |
1309 | * With jdata files, we have to journal a revoke for each block which is | |
1310 | * truncated. As a result, we need to split this into separate transactions | |
1311 | * if the number of pages being truncated gets too large. | |
1312 | */ | |
1313 | ||
fa731fc4 SW |
1314 | static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize) |
1315 | { | |
1316 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1317 | u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
1318 | u64 chunk; | |
1319 | int error; | |
1320 | ||
1321 | while (oldsize != newsize) { | |
e7fdf004 AG |
1322 | struct gfs2_trans *tr; |
1323 | unsigned int offs; | |
1324 | ||
fa731fc4 SW |
1325 | chunk = oldsize - newsize; |
1326 | if (chunk > max_chunk) | |
1327 | chunk = max_chunk; | |
e7fdf004 AG |
1328 | |
1329 | offs = oldsize & ~PAGE_MASK; | |
1330 | if (offs && chunk > PAGE_SIZE) | |
1331 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
1332 | ||
7caef267 | 1333 | truncate_pagecache(inode, oldsize - chunk); |
fa731fc4 | 1334 | oldsize -= chunk; |
e7fdf004 AG |
1335 | |
1336 | tr = current->journal_info; | |
1337 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
1338 | continue; | |
1339 | ||
fa731fc4 SW |
1340 | gfs2_trans_end(sdp); |
1341 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
1342 | if (error) | |
1343 | return error; | |
1344 | } | |
1345 | ||
1346 | return 0; | |
1347 | } | |
1348 | ||
8b5860a3 | 1349 | static int trunc_start(struct inode *inode, u64 newsize) |
b3b94faa | 1350 | { |
ff8f33c8 SW |
1351 | struct gfs2_inode *ip = GFS2_I(inode); |
1352 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
80990f40 | 1353 | struct buffer_head *dibh = NULL; |
b3b94faa | 1354 | int journaled = gfs2_is_jdata(ip); |
8b5860a3 | 1355 | u64 oldsize = inode->i_size; |
b3b94faa DT |
1356 | int error; |
1357 | ||
70499cdf BP |
1358 | if (!gfs2_is_stuffed(ip)) { |
1359 | unsigned int blocksize = i_blocksize(inode); | |
1360 | unsigned int offs = newsize & (blocksize - 1); | |
1361 | if (offs) { | |
1362 | error = gfs2_block_zero_range(inode, newsize, | |
1363 | blocksize - offs); | |
1364 | if (error) | |
1365 | return error; | |
1366 | } | |
1367 | } | |
fa731fc4 SW |
1368 | if (journaled) |
1369 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES); | |
1370 | else | |
1371 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
b3b94faa DT |
1372 | if (error) |
1373 | return error; | |
1374 | ||
1375 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1376 | if (error) | |
1377 | goto out; | |
1378 | ||
350a9b0a | 1379 | gfs2_trans_add_meta(ip->i_gl, dibh); |
ff8f33c8 | 1380 | |
70499cdf | 1381 | if (gfs2_is_stuffed(ip)) |
ff8f33c8 | 1382 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize); |
70499cdf | 1383 | else |
ff8f33c8 | 1384 | ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1385 | |
ff8f33c8 | 1386 | i_size_write(inode, newsize); |
8a8b8d91 | 1387 | ip->i_inode.i_mtime = inode_set_ctime_current(&ip->i_inode); |
ff8f33c8 | 1388 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa | 1389 | |
fa731fc4 SW |
1390 | if (journaled) |
1391 | error = gfs2_journaled_truncate(inode, oldsize, newsize); | |
1392 | else | |
7caef267 | 1393 | truncate_pagecache(inode, newsize); |
fa731fc4 | 1394 | |
a91ea69f | 1395 | out: |
80990f40 AG |
1396 | brelse(dibh); |
1397 | if (current->journal_info) | |
1398 | gfs2_trans_end(sdp); | |
b3b94faa DT |
1399 | return error; |
1400 | } | |
1401 | ||
54992257 AG |
1402 | int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length, |
1403 | struct iomap *iomap) | |
1404 | { | |
1405 | struct metapath mp = { .mp_aheight = 1, }; | |
1406 | int ret; | |
1407 | ||
1408 | ret = __gfs2_iomap_get(inode, pos, length, 0, iomap, &mp); | |
1409 | release_metapath(&mp); | |
1410 | return ret; | |
1411 | } | |
1412 | ||
1413 | int gfs2_iomap_alloc(struct inode *inode, loff_t pos, loff_t length, | |
1414 | struct iomap *iomap) | |
628e366d AG |
1415 | { |
1416 | struct metapath mp = { .mp_aheight = 1, }; | |
1417 | int ret; | |
1418 | ||
54992257 | 1419 | ret = __gfs2_iomap_get(inode, pos, length, IOMAP_WRITE, iomap, &mp); |
628e366d | 1420 | if (!ret && iomap->type == IOMAP_HOLE) |
54992257 | 1421 | ret = __gfs2_iomap_alloc(inode, iomap, &mp); |
628e366d AG |
1422 | release_metapath(&mp); |
1423 | return ret; | |
1424 | } | |
1425 | ||
d552a2b9 BP |
1426 | /** |
1427 | * sweep_bh_for_rgrps - find an rgrp in a meta buffer and free blocks therein | |
1428 | * @ip: inode | |
c551f66c | 1429 | * @rd_gh: holder of resource group glock |
5cf26b1e AG |
1430 | * @bh: buffer head to sweep |
1431 | * @start: starting point in bh | |
1432 | * @end: end point in bh | |
1433 | * @meta: true if bh points to metadata (rather than data) | |
d552a2b9 | 1434 | * @btotal: place to keep count of total blocks freed |
d552a2b9 BP |
1435 | * |
1436 | * We sweep a metadata buffer (provided by the metapath) for blocks we need to | |
1437 | * free, and free them all. However, we do it one rgrp at a time. If this | |
1438 | * block has references to multiple rgrps, we break it into individual | |
1439 | * transactions. This allows other processes to use the rgrps while we're | |
1440 | * focused on a single one, for better concurrency / performance. | |
1441 | * At every transaction boundary, we rewrite the inode into the journal. | |
1442 | * That way the bitmaps are kept consistent with the inode and we can recover | |
1443 | * if we're interrupted by power-outages. | |
1444 | * | |
1445 | * Returns: 0, or return code if an error occurred. | |
1446 | * *btotal has the total number of blocks freed | |
1447 | */ | |
1448 | static int sweep_bh_for_rgrps(struct gfs2_inode *ip, struct gfs2_holder *rd_gh, | |
5cf26b1e AG |
1449 | struct buffer_head *bh, __be64 *start, __be64 *end, |
1450 | bool meta, u32 *btotal) | |
b3b94faa | 1451 | { |
9b8c81d1 | 1452 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d552a2b9 BP |
1453 | struct gfs2_rgrpd *rgd; |
1454 | struct gfs2_trans *tr; | |
5cf26b1e | 1455 | __be64 *p; |
d552a2b9 BP |
1456 | int blks_outside_rgrp; |
1457 | u64 bn, bstart, isize_blks; | |
1458 | s64 blen; /* needs to be s64 or gfs2_add_inode_blocks breaks */ | |
d552a2b9 BP |
1459 | int ret = 0; |
1460 | bool buf_in_tr = false; /* buffer was added to transaction */ | |
1461 | ||
d552a2b9 | 1462 | more_rgrps: |
5cf26b1e AG |
1463 | rgd = NULL; |
1464 | if (gfs2_holder_initialized(rd_gh)) { | |
1465 | rgd = gfs2_glock2rgrp(rd_gh->gh_gl); | |
1466 | gfs2_assert_withdraw(sdp, | |
1467 | gfs2_glock_is_locked_by_me(rd_gh->gh_gl)); | |
1468 | } | |
d552a2b9 BP |
1469 | blks_outside_rgrp = 0; |
1470 | bstart = 0; | |
1471 | blen = 0; | |
d552a2b9 | 1472 | |
5cf26b1e | 1473 | for (p = start; p < end; p++) { |
d552a2b9 BP |
1474 | if (!*p) |
1475 | continue; | |
1476 | bn = be64_to_cpu(*p); | |
5cf26b1e AG |
1477 | |
1478 | if (rgd) { | |
1479 | if (!rgrp_contains_block(rgd, bn)) { | |
1480 | blks_outside_rgrp++; | |
1481 | continue; | |
1482 | } | |
d552a2b9 | 1483 | } else { |
90bcab99 | 1484 | rgd = gfs2_blk2rgrpd(sdp, bn, true); |
5cf26b1e AG |
1485 | if (unlikely(!rgd)) { |
1486 | ret = -EIO; | |
1487 | goto out; | |
1488 | } | |
d552a2b9 | 1489 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
4fc7ec31 | 1490 | LM_FLAG_NODE_SCOPE, rd_gh); |
d552a2b9 BP |
1491 | if (ret) |
1492 | goto out; | |
1493 | ||
1494 | /* Must be done with the rgrp glock held: */ | |
1495 | if (gfs2_rs_active(&ip->i_res) && | |
c65b76b8 | 1496 | rgd == ip->i_res.rs_rgd) |
d552a2b9 BP |
1497 | gfs2_rs_deltree(&ip->i_res); |
1498 | } | |
1499 | ||
d552a2b9 BP |
1500 | /* The size of our transactions will be unknown until we |
1501 | actually process all the metadata blocks that relate to | |
1502 | the rgrp. So we estimate. We know it can't be more than | |
1503 | the dinode's i_blocks and we don't want to exceed the | |
1504 | journal flush threshold, sd_log_thresh2. */ | |
1505 | if (current->journal_info == NULL) { | |
1506 | unsigned int jblocks_rqsted, revokes; | |
1507 | ||
1508 | jblocks_rqsted = rgd->rd_length + RES_DINODE + | |
1509 | RES_INDIRECT; | |
1510 | isize_blks = gfs2_get_inode_blocks(&ip->i_inode); | |
1511 | if (isize_blks > atomic_read(&sdp->sd_log_thresh2)) | |
1512 | jblocks_rqsted += | |
1513 | atomic_read(&sdp->sd_log_thresh2); | |
1514 | else | |
1515 | jblocks_rqsted += isize_blks; | |
1516 | revokes = jblocks_rqsted; | |
1517 | if (meta) | |
5cf26b1e | 1518 | revokes += end - start; |
d552a2b9 BP |
1519 | else if (ip->i_depth) |
1520 | revokes += sdp->sd_inptrs; | |
1521 | ret = gfs2_trans_begin(sdp, jblocks_rqsted, revokes); | |
1522 | if (ret) | |
1523 | goto out_unlock; | |
1524 | down_write(&ip->i_rw_mutex); | |
1525 | } | |
1526 | /* check if we will exceed the transaction blocks requested */ | |
1527 | tr = current->journal_info; | |
1528 | if (tr->tr_num_buf_new + RES_STATFS + | |
1529 | RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) { | |
1530 | /* We set blks_outside_rgrp to ensure the loop will | |
1531 | be repeated for the same rgrp, but with a new | |
1532 | transaction. */ | |
1533 | blks_outside_rgrp++; | |
1534 | /* This next part is tricky. If the buffer was added | |
1535 | to the transaction, we've already set some block | |
1536 | pointers to 0, so we better follow through and free | |
1537 | them, or we will introduce corruption (so break). | |
1538 | This may be impossible, or at least rare, but I | |
1539 | decided to cover the case regardless. | |
1540 | ||
1541 | If the buffer was not added to the transaction | |
1542 | (this call), doing so would exceed our transaction | |
1543 | size, so we need to end the transaction and start a | |
1544 | new one (so goto). */ | |
1545 | ||
1546 | if (buf_in_tr) | |
1547 | break; | |
1548 | goto out_unlock; | |
1549 | } | |
1550 | ||
1551 | gfs2_trans_add_meta(ip->i_gl, bh); | |
1552 | buf_in_tr = true; | |
1553 | *p = 0; | |
1554 | if (bstart + blen == bn) { | |
1555 | blen++; | |
1556 | continue; | |
1557 | } | |
1558 | if (bstart) { | |
0ddeded4 | 1559 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1560 | (*btotal) += blen; |
1561 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1562 | } | |
1563 | bstart = bn; | |
1564 | blen = 1; | |
1565 | } | |
1566 | if (bstart) { | |
0ddeded4 | 1567 | __gfs2_free_blocks(ip, rgd, bstart, (u32)blen, meta); |
d552a2b9 BP |
1568 | (*btotal) += blen; |
1569 | gfs2_add_inode_blocks(&ip->i_inode, -blen); | |
1570 | } | |
1571 | out_unlock: | |
1572 | if (!ret && blks_outside_rgrp) { /* If buffer still has non-zero blocks | |
1573 | outside the rgrp we just processed, | |
1574 | do it all over again. */ | |
1575 | if (current->journal_info) { | |
5cf26b1e AG |
1576 | struct buffer_head *dibh; |
1577 | ||
1578 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1579 | if (ret) | |
1580 | goto out; | |
d552a2b9 BP |
1581 | |
1582 | /* Every transaction boundary, we rewrite the dinode | |
1583 | to keep its di_blocks current in case of failure. */ | |
8a8b8d91 | 1584 | ip->i_inode.i_mtime = inode_set_ctime_current(&ip->i_inode); |
d552a2b9 BP |
1585 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1586 | gfs2_dinode_out(ip, dibh->b_data); | |
5cf26b1e | 1587 | brelse(dibh); |
d552a2b9 BP |
1588 | up_write(&ip->i_rw_mutex); |
1589 | gfs2_trans_end(sdp); | |
f0b444b3 | 1590 | buf_in_tr = false; |
d552a2b9 BP |
1591 | } |
1592 | gfs2_glock_dq_uninit(rd_gh); | |
1593 | cond_resched(); | |
1594 | goto more_rgrps; | |
1595 | } | |
1596 | out: | |
1597 | return ret; | |
1598 | } | |
1599 | ||
10d2cf94 AG |
1600 | static bool mp_eq_to_hgt(struct metapath *mp, __u16 *list, unsigned int h) |
1601 | { | |
1602 | if (memcmp(mp->mp_list, list, h * sizeof(mp->mp_list[0]))) | |
1603 | return false; | |
1604 | return true; | |
1605 | } | |
1606 | ||
d552a2b9 BP |
1607 | /** |
1608 | * find_nonnull_ptr - find a non-null pointer given a metapath and height | |
c551f66c | 1609 | * @sdp: The superblock |
d552a2b9 BP |
1610 | * @mp: starting metapath |
1611 | * @h: desired height to search | |
c551f66c LJ |
1612 | * @end_list: See punch_hole(). |
1613 | * @end_aligned: See punch_hole(). | |
d552a2b9 | 1614 | * |
10d2cf94 | 1615 | * Assumes the metapath is valid (with buffers) out to height h. |
d552a2b9 BP |
1616 | * Returns: true if a non-null pointer was found in the metapath buffer |
1617 | * false if all remaining pointers are NULL in the buffer | |
1618 | */ | |
1619 | static bool find_nonnull_ptr(struct gfs2_sbd *sdp, struct metapath *mp, | |
10d2cf94 AG |
1620 | unsigned int h, |
1621 | __u16 *end_list, unsigned int end_aligned) | |
d552a2b9 | 1622 | { |
10d2cf94 AG |
1623 | struct buffer_head *bh = mp->mp_bh[h]; |
1624 | __be64 *first, *ptr, *end; | |
1625 | ||
1626 | first = metaptr1(h, mp); | |
1627 | ptr = first + mp->mp_list[h]; | |
1628 | end = (__be64 *)(bh->b_data + bh->b_size); | |
1629 | if (end_list && mp_eq_to_hgt(mp, end_list, h)) { | |
1630 | bool keep_end = h < end_aligned; | |
1631 | end = first + end_list[h] + keep_end; | |
1632 | } | |
d552a2b9 | 1633 | |
10d2cf94 | 1634 | while (ptr < end) { |
c4a9d189 | 1635 | if (*ptr) { /* if we have a non-null pointer */ |
10d2cf94 | 1636 | mp->mp_list[h] = ptr - first; |
c4a9d189 BP |
1637 | h++; |
1638 | if (h < GFS2_MAX_META_HEIGHT) | |
10d2cf94 | 1639 | mp->mp_list[h] = 0; |
d552a2b9 | 1640 | return true; |
c4a9d189 | 1641 | } |
10d2cf94 | 1642 | ptr++; |
d552a2b9 | 1643 | } |
10d2cf94 | 1644 | return false; |
d552a2b9 BP |
1645 | } |
1646 | ||
1647 | enum dealloc_states { | |
1648 | DEALLOC_MP_FULL = 0, /* Strip a metapath with all buffers read in */ | |
1649 | DEALLOC_MP_LOWER = 1, /* lower the metapath strip height */ | |
1650 | DEALLOC_FILL_MP = 2, /* Fill in the metapath to the given height. */ | |
1651 | DEALLOC_DONE = 3, /* process complete */ | |
1652 | }; | |
b3b94faa | 1653 | |
5cf26b1e AG |
1654 | static inline void |
1655 | metapointer_range(struct metapath *mp, int height, | |
1656 | __u16 *start_list, unsigned int start_aligned, | |
10d2cf94 | 1657 | __u16 *end_list, unsigned int end_aligned, |
5cf26b1e AG |
1658 | __be64 **start, __be64 **end) |
1659 | { | |
1660 | struct buffer_head *bh = mp->mp_bh[height]; | |
1661 | __be64 *first; | |
1662 | ||
1663 | first = metaptr1(height, mp); | |
1664 | *start = first; | |
1665 | if (mp_eq_to_hgt(mp, start_list, height)) { | |
1666 | bool keep_start = height < start_aligned; | |
1667 | *start = first + start_list[height] + keep_start; | |
1668 | } | |
1669 | *end = (__be64 *)(bh->b_data + bh->b_size); | |
10d2cf94 AG |
1670 | if (end_list && mp_eq_to_hgt(mp, end_list, height)) { |
1671 | bool keep_end = height < end_aligned; | |
1672 | *end = first + end_list[height] + keep_end; | |
1673 | } | |
1674 | } | |
1675 | ||
1676 | static inline bool walk_done(struct gfs2_sbd *sdp, | |
1677 | struct metapath *mp, int height, | |
1678 | __u16 *end_list, unsigned int end_aligned) | |
1679 | { | |
1680 | __u16 end; | |
1681 | ||
1682 | if (end_list) { | |
1683 | bool keep_end = height < end_aligned; | |
1684 | if (!mp_eq_to_hgt(mp, end_list, height)) | |
1685 | return false; | |
1686 | end = end_list[height] + keep_end; | |
1687 | } else | |
1688 | end = (height > 0) ? sdp->sd_inptrs : sdp->sd_diptrs; | |
1689 | return mp->mp_list[height] >= end; | |
5cf26b1e AG |
1690 | } |
1691 | ||
d552a2b9 | 1692 | /** |
10d2cf94 | 1693 | * punch_hole - deallocate blocks in a file |
d552a2b9 | 1694 | * @ip: inode to truncate |
10d2cf94 AG |
1695 | * @offset: the start of the hole |
1696 | * @length: the size of the hole (or 0 for truncate) | |
1697 | * | |
1698 | * Punch a hole into a file or truncate a file at a given position. This | |
1699 | * function operates in whole blocks (@offset and @length are rounded | |
1700 | * accordingly); partially filled blocks must be cleared otherwise. | |
d552a2b9 | 1701 | * |
10d2cf94 AG |
1702 | * This function works from the bottom up, and from the right to the left. In |
1703 | * other words, it strips off the highest layer (data) before stripping any of | |
1704 | * the metadata. Doing it this way is best in case the operation is interrupted | |
1705 | * by power failure, etc. The dinode is rewritten in every transaction to | |
1706 | * guarantee integrity. | |
d552a2b9 | 1707 | */ |
10d2cf94 | 1708 | static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length) |
d552a2b9 BP |
1709 | { |
1710 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
bb491ce6 | 1711 | u64 maxsize = sdp->sd_heightsize[ip->i_height]; |
10d2cf94 | 1712 | struct metapath mp = {}; |
d552a2b9 BP |
1713 | struct buffer_head *dibh, *bh; |
1714 | struct gfs2_holder rd_gh; | |
cb7f0903 | 1715 | unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; |
10d2cf94 AG |
1716 | u64 lblock = (offset + (1 << bsize_shift) - 1) >> bsize_shift; |
1717 | __u16 start_list[GFS2_MAX_META_HEIGHT]; | |
1718 | __u16 __end_list[GFS2_MAX_META_HEIGHT], *end_list = NULL; | |
3f649ab7 | 1719 | unsigned int start_aligned, end_aligned; |
d552a2b9 BP |
1720 | unsigned int strip_h = ip->i_height - 1; |
1721 | u32 btotal = 0; | |
1722 | int ret, state; | |
1723 | int mp_h; /* metapath buffers are read in to this height */ | |
d552a2b9 | 1724 | u64 prev_bnr = 0; |
5cf26b1e | 1725 | __be64 *start, *end; |
b3b94faa | 1726 | |
bb491ce6 AG |
1727 | if (offset >= maxsize) { |
1728 | /* | |
e4f82bf2 BP |
1729 | * The starting point lies beyond the allocated metadata; |
1730 | * there are no blocks to deallocate. | |
bb491ce6 AG |
1731 | */ |
1732 | return 0; | |
1733 | } | |
1734 | ||
10d2cf94 AG |
1735 | /* |
1736 | * The start position of the hole is defined by lblock, start_list, and | |
1737 | * start_aligned. The end position of the hole is defined by lend, | |
1738 | * end_list, and end_aligned. | |
1739 | * | |
1740 | * start_aligned and end_aligned define down to which height the start | |
1741 | * and end positions are aligned to the metadata tree (i.e., the | |
1742 | * position is a multiple of the metadata granularity at the height | |
1743 | * above). This determines at which heights additional meta pointers | |
1744 | * needs to be preserved for the remaining data. | |
1745 | */ | |
b3b94faa | 1746 | |
10d2cf94 | 1747 | if (length) { |
10d2cf94 AG |
1748 | u64 end_offset = offset + length; |
1749 | u64 lend; | |
1750 | ||
1751 | /* | |
1752 | * Clip the end at the maximum file size for the given height: | |
1753 | * that's how far the metadata goes; files bigger than that | |
1754 | * will have additional layers of indirection. | |
1755 | */ | |
1756 | if (end_offset > maxsize) | |
1757 | end_offset = maxsize; | |
1758 | lend = end_offset >> bsize_shift; | |
1759 | ||
1760 | if (lblock >= lend) | |
1761 | return 0; | |
1762 | ||
1763 | find_metapath(sdp, lend, &mp, ip->i_height); | |
1764 | end_list = __end_list; | |
1765 | memcpy(end_list, mp.mp_list, sizeof(mp.mp_list)); | |
1766 | ||
1767 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { | |
1768 | if (end_list[mp_h]) | |
1769 | break; | |
1770 | } | |
1771 | end_aligned = mp_h; | |
1772 | } | |
1773 | ||
1774 | find_metapath(sdp, lblock, &mp, ip->i_height); | |
cb7f0903 AG |
1775 | memcpy(start_list, mp.mp_list, sizeof(start_list)); |
1776 | ||
cb7f0903 AG |
1777 | for (mp_h = ip->i_height - 1; mp_h > 0; mp_h--) { |
1778 | if (start_list[mp_h]) | |
1779 | break; | |
1780 | } | |
1781 | start_aligned = mp_h; | |
d552a2b9 BP |
1782 | |
1783 | ret = gfs2_meta_inode_buffer(ip, &dibh); | |
1784 | if (ret) | |
1785 | return ret; | |
b3b94faa | 1786 | |
d552a2b9 BP |
1787 | mp.mp_bh[0] = dibh; |
1788 | ret = lookup_metapath(ip, &mp); | |
e8b43fe0 AG |
1789 | if (ret) |
1790 | goto out_metapath; | |
c3ce5aa9 AG |
1791 | |
1792 | /* issue read-ahead on metadata */ | |
5cf26b1e AG |
1793 | for (mp_h = 0; mp_h < mp.mp_aheight - 1; mp_h++) { |
1794 | metapointer_range(&mp, mp_h, start_list, start_aligned, | |
10d2cf94 | 1795 | end_list, end_aligned, &start, &end); |
5cf26b1e AG |
1796 | gfs2_metapath_ra(ip->i_gl, start, end); |
1797 | } | |
c3ce5aa9 | 1798 | |
e8b43fe0 | 1799 | if (mp.mp_aheight == ip->i_height) |
d552a2b9 BP |
1800 | state = DEALLOC_MP_FULL; /* We have a complete metapath */ |
1801 | else | |
1802 | state = DEALLOC_FILL_MP; /* deal with partial metapath */ | |
b3b94faa | 1803 | |
d552a2b9 BP |
1804 | ret = gfs2_rindex_update(sdp); |
1805 | if (ret) | |
1806 | goto out_metapath; | |
1807 | ||
1808 | ret = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); | |
1809 | if (ret) | |
1810 | goto out_metapath; | |
1811 | gfs2_holder_mark_uninitialized(&rd_gh); | |
1812 | ||
1813 | mp_h = strip_h; | |
1814 | ||
1815 | while (state != DEALLOC_DONE) { | |
1816 | switch (state) { | |
1817 | /* Truncate a full metapath at the given strip height. | |
1818 | * Note that strip_h == mp_h in order to be in this state. */ | |
1819 | case DEALLOC_MP_FULL: | |
d552a2b9 BP |
1820 | bh = mp.mp_bh[mp_h]; |
1821 | gfs2_assert_withdraw(sdp, bh); | |
1822 | if (gfs2_assert_withdraw(sdp, | |
1823 | prev_bnr != bh->b_blocknr)) { | |
f29e62ee BP |
1824 | fs_emerg(sdp, "inode %llu, block:%llu, i_h:%u," |
1825 | "s_h:%u, mp_h:%u\n", | |
d552a2b9 BP |
1826 | (unsigned long long)ip->i_no_addr, |
1827 | prev_bnr, ip->i_height, strip_h, mp_h); | |
1828 | } | |
1829 | prev_bnr = bh->b_blocknr; | |
cb7f0903 | 1830 | |
5cf26b1e AG |
1831 | if (gfs2_metatype_check(sdp, bh, |
1832 | (mp_h ? GFS2_METATYPE_IN : | |
1833 | GFS2_METATYPE_DI))) { | |
1834 | ret = -EIO; | |
1835 | goto out; | |
1836 | } | |
1837 | ||
10d2cf94 AG |
1838 | /* |
1839 | * Below, passing end_aligned as 0 gives us the | |
1840 | * metapointer range excluding the end point: the end | |
1841 | * point is the first metapath we must not deallocate! | |
1842 | */ | |
1843 | ||
5cf26b1e | 1844 | metapointer_range(&mp, mp_h, start_list, start_aligned, |
10d2cf94 | 1845 | end_list, 0 /* end_aligned */, |
5cf26b1e AG |
1846 | &start, &end); |
1847 | ret = sweep_bh_for_rgrps(ip, &rd_gh, mp.mp_bh[mp_h], | |
1848 | start, end, | |
1849 | mp_h != ip->i_height - 1, | |
1850 | &btotal); | |
cb7f0903 | 1851 | |
d552a2b9 BP |
1852 | /* If we hit an error or just swept dinode buffer, |
1853 | just exit. */ | |
1854 | if (ret || !mp_h) { | |
1855 | state = DEALLOC_DONE; | |
1856 | break; | |
1857 | } | |
1858 | state = DEALLOC_MP_LOWER; | |
1859 | break; | |
1860 | ||
1861 | /* lower the metapath strip height */ | |
1862 | case DEALLOC_MP_LOWER: | |
1863 | /* We're done with the current buffer, so release it, | |
1864 | unless it's the dinode buffer. Then back up to the | |
1865 | previous pointer. */ | |
1866 | if (mp_h) { | |
1867 | brelse(mp.mp_bh[mp_h]); | |
1868 | mp.mp_bh[mp_h] = NULL; | |
1869 | } | |
1870 | /* If we can't get any lower in height, we've stripped | |
1871 | off all we can. Next step is to back up and start | |
1872 | stripping the previous level of metadata. */ | |
1873 | if (mp_h == 0) { | |
1874 | strip_h--; | |
cb7f0903 | 1875 | memcpy(mp.mp_list, start_list, sizeof(start_list)); |
d552a2b9 BP |
1876 | mp_h = strip_h; |
1877 | state = DEALLOC_FILL_MP; | |
1878 | break; | |
1879 | } | |
1880 | mp.mp_list[mp_h] = 0; | |
1881 | mp_h--; /* search one metadata height down */ | |
d552a2b9 | 1882 | mp.mp_list[mp_h]++; |
10d2cf94 AG |
1883 | if (walk_done(sdp, &mp, mp_h, end_list, end_aligned)) |
1884 | break; | |
d552a2b9 BP |
1885 | /* Here we've found a part of the metapath that is not |
1886 | * allocated. We need to search at that height for the | |
1887 | * next non-null pointer. */ | |
10d2cf94 | 1888 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) { |
d552a2b9 BP |
1889 | state = DEALLOC_FILL_MP; |
1890 | mp_h++; | |
1891 | } | |
1892 | /* No more non-null pointers at this height. Back up | |
1893 | to the previous height and try again. */ | |
1894 | break; /* loop around in the same state */ | |
1895 | ||
1896 | /* Fill the metapath with buffers to the given height. */ | |
1897 | case DEALLOC_FILL_MP: | |
1898 | /* Fill the buffers out to the current height. */ | |
1899 | ret = fillup_metapath(ip, &mp, mp_h); | |
c3ce5aa9 | 1900 | if (ret < 0) |
d552a2b9 | 1901 | goto out; |
c3ce5aa9 | 1902 | |
e7445ced AG |
1903 | /* On the first pass, issue read-ahead on metadata. */ |
1904 | if (mp.mp_aheight > 1 && strip_h == ip->i_height - 1) { | |
1905 | unsigned int height = mp.mp_aheight - 1; | |
1906 | ||
1907 | /* No read-ahead for data blocks. */ | |
1908 | if (mp.mp_aheight - 1 == strip_h) | |
1909 | height--; | |
1910 | ||
1911 | for (; height >= mp.mp_aheight - ret; height--) { | |
1912 | metapointer_range(&mp, height, | |
5cf26b1e | 1913 | start_list, start_aligned, |
10d2cf94 | 1914 | end_list, end_aligned, |
5cf26b1e AG |
1915 | &start, &end); |
1916 | gfs2_metapath_ra(ip->i_gl, start, end); | |
1917 | } | |
c3ce5aa9 | 1918 | } |
d552a2b9 BP |
1919 | |
1920 | /* If buffers found for the entire strip height */ | |
e8b43fe0 | 1921 | if (mp.mp_aheight - 1 == strip_h) { |
d552a2b9 BP |
1922 | state = DEALLOC_MP_FULL; |
1923 | break; | |
1924 | } | |
e8b43fe0 AG |
1925 | if (mp.mp_aheight < ip->i_height) /* We have a partial height */ |
1926 | mp_h = mp.mp_aheight - 1; | |
d552a2b9 BP |
1927 | |
1928 | /* If we find a non-null block pointer, crawl a bit | |
1929 | higher up in the metapath and try again, otherwise | |
1930 | we need to look lower for a new starting point. */ | |
10d2cf94 | 1931 | if (find_nonnull_ptr(sdp, &mp, mp_h, end_list, end_aligned)) |
d552a2b9 BP |
1932 | mp_h++; |
1933 | else | |
1934 | state = DEALLOC_MP_LOWER; | |
b3b94faa | 1935 | break; |
d552a2b9 | 1936 | } |
b3b94faa DT |
1937 | } |
1938 | ||
d552a2b9 BP |
1939 | if (btotal) { |
1940 | if (current->journal_info == NULL) { | |
1941 | ret = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + | |
1942 | RES_QUOTA, 0); | |
1943 | if (ret) | |
1944 | goto out; | |
1945 | down_write(&ip->i_rw_mutex); | |
1946 | } | |
1947 | gfs2_statfs_change(sdp, 0, +btotal, 0); | |
1948 | gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid, | |
1949 | ip->i_inode.i_gid); | |
8a8b8d91 | 1950 | ip->i_inode.i_mtime = inode_set_ctime_current(&ip->i_inode); |
d552a2b9 BP |
1951 | gfs2_trans_add_meta(ip->i_gl, dibh); |
1952 | gfs2_dinode_out(ip, dibh->b_data); | |
1953 | up_write(&ip->i_rw_mutex); | |
1954 | gfs2_trans_end(sdp); | |
1955 | } | |
b3b94faa | 1956 | |
d552a2b9 BP |
1957 | out: |
1958 | if (gfs2_holder_initialized(&rd_gh)) | |
1959 | gfs2_glock_dq_uninit(&rd_gh); | |
1960 | if (current->journal_info) { | |
1961 | up_write(&ip->i_rw_mutex); | |
1962 | gfs2_trans_end(sdp); | |
1963 | cond_resched(); | |
1964 | } | |
1965 | gfs2_quota_unhold(ip); | |
1966 | out_metapath: | |
1967 | release_metapath(&mp); | |
1968 | return ret; | |
b3b94faa DT |
1969 | } |
1970 | ||
1971 | static int trunc_end(struct gfs2_inode *ip) | |
1972 | { | |
feaa7bba | 1973 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1974 | struct buffer_head *dibh; |
1975 | int error; | |
1976 | ||
1977 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1978 | if (error) | |
1979 | return error; | |
1980 | ||
1981 | down_write(&ip->i_rw_mutex); | |
1982 | ||
1983 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1984 | if (error) | |
1985 | goto out; | |
1986 | ||
a2e0f799 | 1987 | if (!i_size_read(&ip->i_inode)) { |
ecc30c79 | 1988 | ip->i_height = 0; |
ce276b06 | 1989 | ip->i_goal = ip->i_no_addr; |
b3b94faa | 1990 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
45138990 | 1991 | gfs2_ordered_del_inode(ip); |
b3b94faa | 1992 | } |
8a8b8d91 | 1993 | ip->i_inode.i_mtime = inode_set_ctime_current(&ip->i_inode); |
383f01fb | 1994 | ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG; |
b3b94faa | 1995 | |
350a9b0a | 1996 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 1997 | gfs2_dinode_out(ip, dibh->b_data); |
b3b94faa DT |
1998 | brelse(dibh); |
1999 | ||
a91ea69f | 2000 | out: |
b3b94faa | 2001 | up_write(&ip->i_rw_mutex); |
b3b94faa | 2002 | gfs2_trans_end(sdp); |
b3b94faa DT |
2003 | return error; |
2004 | } | |
2005 | ||
2006 | /** | |
2007 | * do_shrink - make a file smaller | |
ff8f33c8 | 2008 | * @inode: the inode |
ff8f33c8 | 2009 | * @newsize: the size to make the file |
b3b94faa | 2010 | * |
ff8f33c8 SW |
2011 | * Called with an exclusive lock on @inode. The @size must |
2012 | * be equal to or smaller than the current inode size. | |
b3b94faa DT |
2013 | * |
2014 | * Returns: errno | |
2015 | */ | |
2016 | ||
8b5860a3 | 2017 | static int do_shrink(struct inode *inode, u64 newsize) |
b3b94faa | 2018 | { |
ff8f33c8 | 2019 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
2020 | int error; |
2021 | ||
8b5860a3 | 2022 | error = trunc_start(inode, newsize); |
b3b94faa DT |
2023 | if (error < 0) |
2024 | return error; | |
ff8f33c8 | 2025 | if (gfs2_is_stuffed(ip)) |
b3b94faa DT |
2026 | return 0; |
2027 | ||
10d2cf94 | 2028 | error = punch_hole(ip, newsize, 0); |
ff8f33c8 | 2029 | if (error == 0) |
b3b94faa DT |
2030 | error = trunc_end(ip); |
2031 | ||
2032 | return error; | |
2033 | } | |
2034 | ||
ff8f33c8 SW |
2035 | /** |
2036 | * do_grow - Touch and update inode size | |
2037 | * @inode: The inode | |
2038 | * @size: The new size | |
2039 | * | |
2040 | * This function updates the timestamps on the inode and | |
2041 | * may also increase the size of the inode. This function | |
2042 | * must not be called with @size any smaller than the current | |
2043 | * inode size. | |
2044 | * | |
2045 | * Although it is not strictly required to unstuff files here, | |
2046 | * earlier versions of GFS2 have a bug in the stuffed file reading | |
2047 | * code which will result in a buffer overrun if the size is larger | |
2048 | * than the max stuffed file size. In order to prevent this from | |
25985edc | 2049 | * occurring, such files are unstuffed, but in other cases we can |
ff8f33c8 SW |
2050 | * just update the inode size directly. |
2051 | * | |
2052 | * Returns: 0 on success, or -ve on error | |
2053 | */ | |
2054 | ||
2055 | static int do_grow(struct inode *inode, u64 size) | |
2056 | { | |
2057 | struct gfs2_inode *ip = GFS2_I(inode); | |
2058 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 2059 | struct gfs2_alloc_parms ap = { .target = 1, }; |
a13b8c5f WC |
2060 | struct buffer_head *dibh; |
2061 | int error; | |
2f7ee358 | 2062 | int unstuff = 0; |
a13b8c5f | 2063 | |
235628c5 | 2064 | if (gfs2_is_stuffed(ip) && size > gfs2_max_stuffed_size(ip)) { |
b8fbf471 | 2065 | error = gfs2_quota_lock_check(ip, &ap); |
ff8f33c8 | 2066 | if (error) |
5407e242 | 2067 | return error; |
ff8f33c8 | 2068 | |
7b9cff46 | 2069 | error = gfs2_inplace_reserve(ip, &ap); |
ff8f33c8 SW |
2070 | if (error) |
2071 | goto do_grow_qunlock; | |
2f7ee358 | 2072 | unstuff = 1; |
ff8f33c8 SW |
2073 | } |
2074 | ||
a01aedfe | 2075 | error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT + |
bc020561 BP |
2076 | (unstuff && |
2077 | gfs2_is_jdata(ip) ? RES_JDATA : 0) + | |
a01aedfe BP |
2078 | (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ? |
2079 | 0 : RES_QUOTA), 0); | |
a13b8c5f | 2080 | if (error) |
ff8f33c8 | 2081 | goto do_grow_release; |
a13b8c5f | 2082 | |
2f7ee358 | 2083 | if (unstuff) { |
7a607a41 | 2084 | error = gfs2_unstuff_dinode(ip); |
ff8f33c8 SW |
2085 | if (error) |
2086 | goto do_end_trans; | |
2087 | } | |
a13b8c5f WC |
2088 | |
2089 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2090 | if (error) | |
ff8f33c8 | 2091 | goto do_end_trans; |
a13b8c5f | 2092 | |
b473bc2d | 2093 | truncate_setsize(inode, size); |
8a8b8d91 | 2094 | ip->i_inode.i_mtime = inode_set_ctime_current(&ip->i_inode); |
350a9b0a | 2095 | gfs2_trans_add_meta(ip->i_gl, dibh); |
a13b8c5f WC |
2096 | gfs2_dinode_out(ip, dibh->b_data); |
2097 | brelse(dibh); | |
2098 | ||
ff8f33c8 | 2099 | do_end_trans: |
a13b8c5f | 2100 | gfs2_trans_end(sdp); |
ff8f33c8 | 2101 | do_grow_release: |
2f7ee358 | 2102 | if (unstuff) { |
ff8f33c8 SW |
2103 | gfs2_inplace_release(ip); |
2104 | do_grow_qunlock: | |
2105 | gfs2_quota_unlock(ip); | |
ff8f33c8 | 2106 | } |
a13b8c5f WC |
2107 | return error; |
2108 | } | |
2109 | ||
b3b94faa | 2110 | /** |
ff8f33c8 SW |
2111 | * gfs2_setattr_size - make a file a given size |
2112 | * @inode: the inode | |
2113 | * @newsize: the size to make the file | |
b3b94faa | 2114 | * |
ff8f33c8 | 2115 | * The file size can grow, shrink, or stay the same size. This |
3e7aafc3 | 2116 | * is called holding i_rwsem and an exclusive glock on the inode |
ff8f33c8 | 2117 | * in question. |
b3b94faa DT |
2118 | * |
2119 | * Returns: errno | |
2120 | */ | |
2121 | ||
ff8f33c8 | 2122 | int gfs2_setattr_size(struct inode *inode, u64 newsize) |
b3b94faa | 2123 | { |
af5c2697 | 2124 | struct gfs2_inode *ip = GFS2_I(inode); |
ff8f33c8 | 2125 | int ret; |
b3b94faa | 2126 | |
ff8f33c8 | 2127 | BUG_ON(!S_ISREG(inode->i_mode)); |
b3b94faa | 2128 | |
ff8f33c8 SW |
2129 | ret = inode_newsize_ok(inode, newsize); |
2130 | if (ret) | |
2131 | return ret; | |
b3b94faa | 2132 | |
562c72aa CH |
2133 | inode_dio_wait(inode); |
2134 | ||
2fba46a0 | 2135 | ret = gfs2_qa_get(ip); |
d2b47cfb | 2136 | if (ret) |
2b3dcf35 | 2137 | goto out; |
d2b47cfb | 2138 | |
8b5860a3 | 2139 | if (newsize >= inode->i_size) { |
2b3dcf35 BP |
2140 | ret = do_grow(inode, newsize); |
2141 | goto out; | |
2142 | } | |
ff8f33c8 | 2143 | |
8b5860a3 | 2144 | ret = do_shrink(inode, newsize); |
2b3dcf35 | 2145 | out: |
7336905a | 2146 | gfs2_rs_delete(ip); |
1595548f | 2147 | gfs2_qa_put(ip); |
2b3dcf35 | 2148 | return ret; |
b3b94faa DT |
2149 | } |
2150 | ||
2151 | int gfs2_truncatei_resume(struct gfs2_inode *ip) | |
2152 | { | |
2153 | int error; | |
10d2cf94 | 2154 | error = punch_hole(ip, i_size_read(&ip->i_inode), 0); |
b3b94faa DT |
2155 | if (!error) |
2156 | error = trunc_end(ip); | |
2157 | return error; | |
2158 | } | |
2159 | ||
2160 | int gfs2_file_dealloc(struct gfs2_inode *ip) | |
2161 | { | |
10d2cf94 | 2162 | return punch_hole(ip, 0, 0); |
b3b94faa DT |
2163 | } |
2164 | ||
b50f227b SW |
2165 | /** |
2166 | * gfs2_free_journal_extents - Free cached journal bmap info | |
2167 | * @jd: The journal | |
2168 | * | |
2169 | */ | |
2170 | ||
2171 | void gfs2_free_journal_extents(struct gfs2_jdesc *jd) | |
2172 | { | |
2173 | struct gfs2_journal_extent *jext; | |
2174 | ||
2175 | while(!list_empty(&jd->extent_list)) { | |
969183bc | 2176 | jext = list_first_entry(&jd->extent_list, struct gfs2_journal_extent, list); |
b50f227b SW |
2177 | list_del(&jext->list); |
2178 | kfree(jext); | |
2179 | } | |
2180 | } | |
2181 | ||
2182 | /** | |
2183 | * gfs2_add_jextent - Add or merge a new extent to extent cache | |
2184 | * @jd: The journal descriptor | |
2185 | * @lblock: The logical block at start of new extent | |
c62baf65 | 2186 | * @dblock: The physical block at start of new extent |
b50f227b SW |
2187 | * @blocks: Size of extent in fs blocks |
2188 | * | |
2189 | * Returns: 0 on success or -ENOMEM | |
2190 | */ | |
2191 | ||
2192 | static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks) | |
2193 | { | |
2194 | struct gfs2_journal_extent *jext; | |
2195 | ||
2196 | if (!list_empty(&jd->extent_list)) { | |
969183bc | 2197 | jext = list_last_entry(&jd->extent_list, struct gfs2_journal_extent, list); |
b50f227b SW |
2198 | if ((jext->dblock + jext->blocks) == dblock) { |
2199 | jext->blocks += blocks; | |
2200 | return 0; | |
2201 | } | |
2202 | } | |
2203 | ||
2204 | jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS); | |
2205 | if (jext == NULL) | |
2206 | return -ENOMEM; | |
2207 | jext->dblock = dblock; | |
2208 | jext->lblock = lblock; | |
2209 | jext->blocks = blocks; | |
2210 | list_add_tail(&jext->list, &jd->extent_list); | |
2211 | jd->nr_extents++; | |
2212 | return 0; | |
2213 | } | |
2214 | ||
2215 | /** | |
2216 | * gfs2_map_journal_extents - Cache journal bmap info | |
2217 | * @sdp: The super block | |
2218 | * @jd: The journal to map | |
2219 | * | |
2220 | * Create a reusable "extent" mapping from all logical | |
2221 | * blocks to all physical blocks for the given journal. This will save | |
2222 | * us time when writing journal blocks. Most journals will have only one | |
2223 | * extent that maps all their logical blocks. That's because gfs2.mkfs | |
2224 | * arranges the journal blocks sequentially to maximize performance. | |
2225 | * So the extent would map the first block for the entire file length. | |
2226 | * However, gfs2_jadd can happen while file activity is happening, so | |
2227 | * those journals may not be sequential. Less likely is the case where | |
2228 | * the users created their own journals by mounting the metafs and | |
2229 | * laying it out. But it's still possible. These journals might have | |
2230 | * several extents. | |
2231 | * | |
2232 | * Returns: 0 on success, or error on failure | |
2233 | */ | |
2234 | ||
2235 | int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd) | |
2236 | { | |
2237 | u64 lblock = 0; | |
2238 | u64 lblock_stop; | |
2239 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); | |
2240 | struct buffer_head bh; | |
2241 | unsigned int shift = sdp->sd_sb.sb_bsize_shift; | |
2242 | u64 size; | |
2243 | int rc; | |
98583b3e | 2244 | ktime_t start, end; |
b50f227b | 2245 | |
98583b3e | 2246 | start = ktime_get(); |
b50f227b SW |
2247 | lblock_stop = i_size_read(jd->jd_inode) >> shift; |
2248 | size = (lblock_stop - lblock) << shift; | |
2249 | jd->nr_extents = 0; | |
2250 | WARN_ON(!list_empty(&jd->extent_list)); | |
2251 | ||
2252 | do { | |
2253 | bh.b_state = 0; | |
2254 | bh.b_blocknr = 0; | |
2255 | bh.b_size = size; | |
2256 | rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0); | |
2257 | if (rc || !buffer_mapped(&bh)) | |
2258 | goto fail; | |
2259 | rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift); | |
2260 | if (rc) | |
2261 | goto fail; | |
2262 | size -= bh.b_size; | |
2263 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2264 | } while(size > 0); | |
2265 | ||
98583b3e AD |
2266 | end = ktime_get(); |
2267 | fs_info(sdp, "journal %d mapped with %u extents in %lldms\n", jd->jd_jid, | |
2268 | jd->nr_extents, ktime_ms_delta(end, start)); | |
b50f227b SW |
2269 | return 0; |
2270 | ||
2271 | fail: | |
2272 | fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n", | |
2273 | rc, jd->jd_jid, | |
2274 | (unsigned long long)(i_size_read(jd->jd_inode) - size), | |
2275 | jd->nr_extents); | |
2276 | fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n", | |
2277 | rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr, | |
2278 | bh.b_state, (unsigned long long)bh.b_size); | |
2279 | gfs2_free_journal_extents(jd); | |
2280 | return rc; | |
2281 | } | |
2282 | ||
b3b94faa DT |
2283 | /** |
2284 | * gfs2_write_alloc_required - figure out if a write will require an allocation | |
2285 | * @ip: the file being written to | |
2286 | * @offset: the offset to write to | |
2287 | * @len: the number of bytes being written | |
b3b94faa | 2288 | * |
461cb419 | 2289 | * Returns: 1 if an alloc is required, 0 otherwise |
b3b94faa DT |
2290 | */ |
2291 | ||
cd915493 | 2292 | int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset, |
461cb419 | 2293 | unsigned int len) |
b3b94faa | 2294 | { |
feaa7bba | 2295 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
941e6d7d SW |
2296 | struct buffer_head bh; |
2297 | unsigned int shift; | |
2298 | u64 lblock, lblock_stop, size; | |
7ed122e4 | 2299 | u64 end_of_file; |
b3b94faa | 2300 | |
b3b94faa DT |
2301 | if (!len) |
2302 | return 0; | |
2303 | ||
2304 | if (gfs2_is_stuffed(ip)) { | |
235628c5 | 2305 | if (offset + len > gfs2_max_stuffed_size(ip)) |
461cb419 | 2306 | return 1; |
b3b94faa DT |
2307 | return 0; |
2308 | } | |
2309 | ||
941e6d7d | 2310 | shift = sdp->sd_sb.sb_bsize_shift; |
7ed122e4 | 2311 | BUG_ON(gfs2_is_dir(ip)); |
a2e0f799 | 2312 | end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift; |
7ed122e4 SW |
2313 | lblock = offset >> shift; |
2314 | lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift; | |
77612578 | 2315 | if (lblock_stop > end_of_file && ip != GFS2_I(sdp->sd_rindex)) |
461cb419 | 2316 | return 1; |
b3b94faa | 2317 | |
941e6d7d SW |
2318 | size = (lblock_stop - lblock) << shift; |
2319 | do { | |
2320 | bh.b_state = 0; | |
2321 | bh.b_size = size; | |
2322 | gfs2_block_map(&ip->i_inode, lblock, &bh, 0); | |
2323 | if (!buffer_mapped(&bh)) | |
461cb419 | 2324 | return 1; |
941e6d7d SW |
2325 | size -= bh.b_size; |
2326 | lblock += (bh.b_size >> ip->i_inode.i_blkbits); | |
2327 | } while(size > 0); | |
b3b94faa DT |
2328 | |
2329 | return 0; | |
2330 | } | |
2331 | ||
4e56a641 AG |
2332 | static int stuffed_zero_range(struct inode *inode, loff_t offset, loff_t length) |
2333 | { | |
2334 | struct gfs2_inode *ip = GFS2_I(inode); | |
2335 | struct buffer_head *dibh; | |
2336 | int error; | |
2337 | ||
2338 | if (offset >= inode->i_size) | |
2339 | return 0; | |
2340 | if (offset + length > inode->i_size) | |
2341 | length = inode->i_size - offset; | |
2342 | ||
2343 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
2344 | if (error) | |
2345 | return error; | |
2346 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
2347 | memset(dibh->b_data + sizeof(struct gfs2_dinode) + offset, 0, | |
2348 | length); | |
2349 | brelse(dibh); | |
2350 | return 0; | |
2351 | } | |
2352 | ||
2353 | static int gfs2_journaled_truncate_range(struct inode *inode, loff_t offset, | |
2354 | loff_t length) | |
2355 | { | |
2356 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2357 | loff_t max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize; | |
2358 | int error; | |
2359 | ||
2360 | while (length) { | |
2361 | struct gfs2_trans *tr; | |
2362 | loff_t chunk; | |
2363 | unsigned int offs; | |
2364 | ||
2365 | chunk = length; | |
2366 | if (chunk > max_chunk) | |
2367 | chunk = max_chunk; | |
2368 | ||
2369 | offs = offset & ~PAGE_MASK; | |
2370 | if (offs && chunk > PAGE_SIZE) | |
2371 | chunk = offs + ((chunk - offs) & PAGE_MASK); | |
2372 | ||
2373 | truncate_pagecache_range(inode, offset, chunk); | |
2374 | offset += chunk; | |
2375 | length -= chunk; | |
2376 | ||
2377 | tr = current->journal_info; | |
2378 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) | |
2379 | continue; | |
2380 | ||
2381 | gfs2_trans_end(sdp); | |
2382 | error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES); | |
2383 | if (error) | |
2384 | return error; | |
2385 | } | |
2386 | return 0; | |
2387 | } | |
2388 | ||
2389 | int __gfs2_punch_hole(struct file *file, loff_t offset, loff_t length) | |
2390 | { | |
2391 | struct inode *inode = file_inode(file); | |
2392 | struct gfs2_inode *ip = GFS2_I(inode); | |
2393 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
39c3a948 AG |
2394 | unsigned int blocksize = i_blocksize(inode); |
2395 | loff_t start, end; | |
4e56a641 AG |
2396 | int error; |
2397 | ||
70499cdf | 2398 | if (!gfs2_is_stuffed(ip)) { |
39c3a948 | 2399 | unsigned int start_off, end_len; |
4e56a641 | 2400 | |
4e56a641 | 2401 | start_off = offset & (blocksize - 1); |
00251a16 | 2402 | end_len = (offset + length) & (blocksize - 1); |
4e56a641 AG |
2403 | if (start_off) { |
2404 | unsigned int len = length; | |
2405 | if (length > blocksize - start_off) | |
2406 | len = blocksize - start_off; | |
2407 | error = gfs2_block_zero_range(inode, offset, len); | |
2408 | if (error) | |
2409 | goto out; | |
2410 | if (start_off + length < blocksize) | |
00251a16 | 2411 | end_len = 0; |
4e56a641 | 2412 | } |
00251a16 | 2413 | if (end_len) { |
4e56a641 | 2414 | error = gfs2_block_zero_range(inode, |
00251a16 | 2415 | offset + length - end_len, end_len); |
4e56a641 AG |
2416 | if (error) |
2417 | goto out; | |
2418 | } | |
2419 | } | |
2420 | ||
70499cdf BP |
2421 | start = round_down(offset, blocksize); |
2422 | end = round_up(offset + length, blocksize) - 1; | |
2423 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
2424 | if (error) | |
2425 | return error; | |
2426 | ||
2427 | if (gfs2_is_jdata(ip)) | |
2428 | error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_JDATA, | |
2429 | GFS2_JTRUNC_REVOKES); | |
2430 | else | |
2431 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
2432 | if (error) | |
2433 | return error; | |
2434 | ||
2435 | if (gfs2_is_stuffed(ip)) { | |
2436 | error = stuffed_zero_range(inode, offset, length); | |
2437 | if (error) | |
2438 | goto out; | |
2439 | } | |
2440 | ||
4e56a641 AG |
2441 | if (gfs2_is_jdata(ip)) { |
2442 | BUG_ON(!current->journal_info); | |
2443 | gfs2_journaled_truncate_range(inode, offset, length); | |
2444 | } else | |
2445 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2446 | ||
2447 | file_update_time(file); | |
2448 | mark_inode_dirty(inode); | |
2449 | ||
2450 | if (current->journal_info) | |
2451 | gfs2_trans_end(sdp); | |
2452 | ||
2453 | if (!gfs2_is_stuffed(ip)) | |
2454 | error = punch_hole(ip, offset, length); | |
2455 | ||
2456 | out: | |
2457 | if (current->journal_info) | |
2458 | gfs2_trans_end(sdp); | |
2459 | return error; | |
2460 | } | |
2164f9b9 CH |
2461 | |
2462 | static int gfs2_map_blocks(struct iomap_writepage_ctx *wpc, struct inode *inode, | |
2463 | loff_t offset) | |
2464 | { | |
2164f9b9 CH |
2465 | int ret; |
2466 | ||
2467 | if (WARN_ON_ONCE(gfs2_is_stuffed(GFS2_I(inode)))) | |
2468 | return -EIO; | |
2469 | ||
2470 | if (offset >= wpc->iomap.offset && | |
2471 | offset < wpc->iomap.offset + wpc->iomap.length) | |
2472 | return 0; | |
2473 | ||
2474 | memset(&wpc->iomap, 0, sizeof(wpc->iomap)); | |
54992257 | 2475 | ret = gfs2_iomap_get(inode, offset, INT_MAX, &wpc->iomap); |
2164f9b9 CH |
2476 | return ret; |
2477 | } | |
2478 | ||
2479 | const struct iomap_writeback_ops gfs2_writeback_ops = { | |
2480 | .map_blocks = gfs2_map_blocks, | |
2481 | }; |