Merge tag 'sh-for-4.9' of git://git.libc.org/linux-sh
[linux-2.6-block.git] / fs / gfs2 / bmap.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3a8a9a10 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
b3b94faa
DT
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
e9fc2aa0 7 * of the GNU General Public License version 2.
b3b94faa
DT
8 */
9
b3b94faa
DT
10#include <linux/spinlock.h>
11#include <linux/completion.h>
12#include <linux/buffer_head.h>
64dd153c 13#include <linux/blkdev.h>
5c676f6d 14#include <linux/gfs2_ondisk.h>
71b86f56 15#include <linux/crc32.h>
b3b94faa
DT
16
17#include "gfs2.h"
5c676f6d 18#include "incore.h"
b3b94faa
DT
19#include "bmap.h"
20#include "glock.h"
21#include "inode.h"
b3b94faa 22#include "meta_io.h"
b3b94faa
DT
23#include "quota.h"
24#include "rgrp.h"
45138990 25#include "log.h"
4c16c36a 26#include "super.h"
b3b94faa 27#include "trans.h"
18ec7d5c 28#include "dir.h"
5c676f6d 29#include "util.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 */
36struct metapath {
dbac6710 37 struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
b3b94faa
DT
38 __u16 mp_list[GFS2_MAX_META_HEIGHT];
39};
40
b3b94faa
DT
41struct strip_mine {
42 int sm_first;
43 unsigned int sm_height;
44};
45
f25ef0c1
SW
46/**
47 * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
48 * @ip: the inode
49 * @dibh: the dinode buffer
50 * @block: the block number that was allocated
ff8f33c8 51 * @page: The (optional) page. This is looked up if @page is NULL
f25ef0c1
SW
52 *
53 * Returns: errno
54 */
55
56static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
cd915493 57 u64 block, struct page *page)
f25ef0c1 58{
f25ef0c1
SW
59 struct inode *inode = &ip->i_inode;
60 struct buffer_head *bh;
61 int release = 0;
62
63 if (!page || page->index) {
220cca2a 64 page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
f25ef0c1
SW
65 if (!page)
66 return -ENOMEM;
67 release = 1;
68 }
69
70 if (!PageUptodate(page)) {
71 void *kaddr = kmap(page);
602c89d2
SW
72 u64 dsize = i_size_read(inode);
73
74 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
75 dsize = dibh->b_size - sizeof(struct gfs2_dinode);
f25ef0c1 76
602c89d2 77 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
09cbfeaf 78 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
f25ef0c1
SW
79 kunmap(page);
80
81 SetPageUptodate(page);
82 }
83
84 if (!page_has_buffers(page))
47a9a527
FF
85 create_empty_buffers(page, BIT(inode->i_blkbits),
86 BIT(BH_Uptodate));
f25ef0c1
SW
87
88 bh = page_buffers(page);
89
90 if (!buffer_mapped(bh))
91 map_bh(bh, inode->i_sb, block);
92
93 set_buffer_uptodate(bh);
eaf96527
SW
94 if (!gfs2_is_jdata(ip))
95 mark_buffer_dirty(bh);
bf36a713 96 if (!gfs2_is_writeback(ip))
350a9b0a 97 gfs2_trans_add_data(ip->i_gl, bh);
f25ef0c1
SW
98
99 if (release) {
100 unlock_page(page);
09cbfeaf 101 put_page(page);
f25ef0c1
SW
102 }
103
104 return 0;
105}
106
b3b94faa
DT
107/**
108 * gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
109 * @ip: The GFS2 inode to unstuff
ff8f33c8 110 * @page: The (optional) page. This is looked up if the @page is NULL
b3b94faa
DT
111 *
112 * This routine unstuffs a dinode and returns it to a "normal" state such
113 * that the height can be grown in the traditional way.
114 *
115 * Returns: errno
116 */
117
f25ef0c1 118int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
b3b94faa
DT
119{
120 struct buffer_head *bh, *dibh;
48516ced 121 struct gfs2_dinode *di;
cd915493 122 u64 block = 0;
18ec7d5c 123 int isdir = gfs2_is_dir(ip);
b3b94faa
DT
124 int error;
125
126 down_write(&ip->i_rw_mutex);
127
128 error = gfs2_meta_inode_buffer(ip, &dibh);
129 if (error)
130 goto out;
907b9bce 131
a2e0f799 132 if (i_size_read(&ip->i_inode)) {
b3b94faa
DT
133 /* Get a free block, fill it with the stuffed data,
134 and write it out to disk */
135
b45e41d7 136 unsigned int n = 1;
6e87ed0f 137 error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
09010978
SW
138 if (error)
139 goto out_brelse;
18ec7d5c 140 if (isdir) {
5731be53 141 gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
61e085a8 142 error = gfs2_dir_get_new_buffer(ip, block, &bh);
b3b94faa
DT
143 if (error)
144 goto out_brelse;
48516ced 145 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
b3b94faa
DT
146 dibh, sizeof(struct gfs2_dinode));
147 brelse(bh);
148 } else {
f25ef0c1 149 error = gfs2_unstuffer_page(ip, dibh, block, page);
b3b94faa
DT
150 if (error)
151 goto out_brelse;
152 }
153 }
154
155 /* Set up the pointer to the new block */
156
350a9b0a 157 gfs2_trans_add_meta(ip->i_gl, dibh);
48516ced 158 di = (struct gfs2_dinode *)dibh->b_data;
b3b94faa
DT
159 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
160
a2e0f799 161 if (i_size_read(&ip->i_inode)) {
48516ced 162 *(__be64 *)(di + 1) = cpu_to_be64(block);
77658aad
SW
163 gfs2_add_inode_blocks(&ip->i_inode, 1);
164 di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
b3b94faa
DT
165 }
166
ecc30c79 167 ip->i_height = 1;
48516ced 168 di->di_height = cpu_to_be16(1);
b3b94faa 169
a91ea69f 170out_brelse:
b3b94faa 171 brelse(dibh);
a91ea69f 172out:
b3b94faa 173 up_write(&ip->i_rw_mutex);
b3b94faa
DT
174 return error;
175}
176
b3b94faa
DT
177
178/**
179 * find_metapath - Find path through the metadata tree
9b8c81d1 180 * @sdp: The superblock
b3b94faa
DT
181 * @mp: The metapath to return the result in
182 * @block: The disk block to look up
9b8c81d1 183 * @height: The pre-calculated height of the metadata tree
b3b94faa
DT
184 *
185 * This routine returns a struct metapath structure that defines a path
186 * through the metadata of inode "ip" to get to block "block".
187 *
188 * Example:
189 * Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
190 * filesystem with a blocksize of 4096.
191 *
192 * find_metapath() would return a struct metapath structure set to:
193 * mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
194 * and mp_list[2] = 165.
195 *
196 * That means that in order to get to the block containing the byte at
197 * offset 101342453, we would load the indirect block pointed to by pointer
198 * 0 in the dinode. We would then load the indirect block pointed to by
199 * pointer 48 in that indirect block. We would then load the data block
200 * pointed to by pointer 165 in that indirect block.
201 *
202 * ----------------------------------------
203 * | Dinode | |
204 * | | 4|
205 * | |0 1 2 3 4 5 9|
206 * | | 6|
207 * ----------------------------------------
208 * |
209 * |
210 * V
211 * ----------------------------------------
212 * | Indirect Block |
213 * | 5|
214 * | 4 4 4 4 4 5 5 1|
215 * |0 5 6 7 8 9 0 1 2|
216 * ----------------------------------------
217 * |
218 * |
219 * V
220 * ----------------------------------------
221 * | Indirect Block |
222 * | 1 1 1 1 1 5|
223 * | 6 6 6 6 6 1|
224 * |0 3 4 5 6 7 2|
225 * ----------------------------------------
226 * |
227 * |
228 * V
229 * ----------------------------------------
230 * | Data block containing offset |
231 * | 101342453 |
232 * | |
233 * | |
234 * ----------------------------------------
235 *
236 */
237
9b8c81d1
SW
238static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
239 struct metapath *mp, unsigned int height)
b3b94faa 240{
b3b94faa
DT
241 unsigned int i;
242
9b8c81d1 243 for (i = height; i--;)
7eabb77e 244 mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
b3b94faa
DT
245
246}
247
5af4e7a0 248static inline unsigned int metapath_branch_start(const struct metapath *mp)
9b8c81d1 249{
5af4e7a0
BM
250 if (mp->mp_list[0] == 0)
251 return 2;
252 return 1;
9b8c81d1
SW
253}
254
b3b94faa
DT
255/**
256 * metapointer - Return pointer to start of metadata in a buffer
b3b94faa
DT
257 * @height: The metadata height (0 = dinode)
258 * @mp: The metapath
259 *
260 * Return a pointer to the block number of the next height of the metadata
261 * tree given a buffer containing the pointer to the current height of the
262 * metadata tree.
263 */
264
9b8c81d1 265static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
b3b94faa 266{
dbac6710 267 struct buffer_head *bh = mp->mp_bh[height];
b3b94faa
DT
268 unsigned int head_size = (height > 0) ?
269 sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
9b8c81d1 270 return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
b3b94faa
DT
271}
272
b99b98dc
SW
273static void gfs2_metapath_ra(struct gfs2_glock *gl,
274 const struct buffer_head *bh, const __be64 *pos)
275{
276 struct buffer_head *rabh;
277 const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size);
278 const __be64 *t;
279
280 for (t = pos; t < endp; t++) {
281 if (!*t)
282 continue;
283
284 rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
285 if (trylock_buffer(rabh)) {
286 if (!buffer_uptodate(rabh)) {
287 rabh->b_end_io = end_buffer_read_sync;
70246286
CH
288 submit_bh(REQ_OP_READ, REQ_RAHEAD | REQ_META,
289 rabh);
b99b98dc
SW
290 continue;
291 }
292 unlock_buffer(rabh);
293 }
294 brelse(rabh);
295 }
296}
297
b3b94faa 298/**
9b8c81d1
SW
299 * lookup_metapath - Walk the metadata tree to a specific point
300 * @ip: The inode
b3b94faa 301 * @mp: The metapath
b3b94faa 302 *
9b8c81d1
SW
303 * Assumes that the inode's buffer has already been looked up and
304 * hooked onto mp->mp_bh[0] and that the metapath has been initialised
305 * by find_metapath().
306 *
307 * If this function encounters part of the tree which has not been
308 * allocated, it returns the current height of the tree at the point
309 * at which it found the unallocated block. Blocks which are found are
310 * added to the mp->mp_bh[] list.
b3b94faa 311 *
9b8c81d1 312 * Returns: error or height of metadata tree
b3b94faa
DT
313 */
314
9b8c81d1 315static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
11707ea0 316{
11707ea0
SW
317 unsigned int end_of_metadata = ip->i_height - 1;
318 unsigned int x;
9b8c81d1
SW
319 __be64 *ptr;
320 u64 dblock;
e23159d2 321 int ret;
11707ea0
SW
322
323 for (x = 0; x < end_of_metadata; x++) {
9b8c81d1
SW
324 ptr = metapointer(x, mp);
325 dblock = be64_to_cpu(*ptr);
326 if (!dblock)
327 return x + 1;
11707ea0 328
f2f9c812 329 ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, &mp->mp_bh[x+1]);
11707ea0
SW
330 if (ret)
331 return ret;
332 }
333
9b8c81d1 334 return ip->i_height;
dbac6710
SW
335}
336
9b8c81d1 337static inline void release_metapath(struct metapath *mp)
dbac6710
SW
338{
339 int i;
340
9b8c81d1
SW
341 for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
342 if (mp->mp_bh[i] == NULL)
343 break;
344 brelse(mp->mp_bh[i]);
345 }
11707ea0
SW
346}
347
30cbf189
SW
348/**
349 * gfs2_extent_length - Returns length of an extent of blocks
350 * @start: Start of the buffer
351 * @len: Length of the buffer in bytes
352 * @ptr: Current position in the buffer
353 * @limit: Max extent length to return (0 = unlimited)
354 * @eob: Set to 1 if we hit "end of block"
355 *
356 * If the first block is zero (unallocated) it will return the number of
357 * unallocated blocks in the extent, otherwise it will return the number
358 * of contiguous blocks in the extent.
359 *
360 * Returns: The length of the extent (minimum of one block)
361 */
362
b650738c 363static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob)
30cbf189
SW
364{
365 const __be64 *end = (start + len);
366 const __be64 *first = ptr;
367 u64 d = be64_to_cpu(*ptr);
368
369 *eob = 0;
370 do {
371 ptr++;
372 if (ptr >= end)
373 break;
374 if (limit && --limit == 0)
375 break;
376 if (d)
377 d++;
378 } while(be64_to_cpu(*ptr) == d);
379 if (ptr >= end)
380 *eob = 1;
381 return (ptr - first);
382}
383
9b8c81d1 384static inline void bmap_lock(struct gfs2_inode *ip, int create)
4cf1ed81 385{
4cf1ed81
SW
386 if (create)
387 down_write(&ip->i_rw_mutex);
388 else
389 down_read(&ip->i_rw_mutex);
390}
391
9b8c81d1 392static inline void bmap_unlock(struct gfs2_inode *ip, int create)
4cf1ed81 393{
4cf1ed81
SW
394 if (create)
395 up_write(&ip->i_rw_mutex);
396 else
397 up_read(&ip->i_rw_mutex);
398}
399
9b8c81d1
SW
400static inline __be64 *gfs2_indirect_init(struct metapath *mp,
401 struct gfs2_glock *gl, unsigned int i,
402 unsigned offset, u64 bn)
403{
404 __be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
405 ((i > 1) ? sizeof(struct gfs2_meta_header) :
406 sizeof(struct gfs2_dinode)));
407 BUG_ON(i < 1);
408 BUG_ON(mp->mp_bh[i] != NULL);
409 mp->mp_bh[i] = gfs2_meta_new(gl, bn);
350a9b0a 410 gfs2_trans_add_meta(gl, mp->mp_bh[i]);
9b8c81d1
SW
411 gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
412 gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
413 ptr += offset;
414 *ptr = cpu_to_be64(bn);
415 return ptr;
416}
417
418enum alloc_state {
419 ALLOC_DATA = 0,
420 ALLOC_GROW_DEPTH = 1,
421 ALLOC_GROW_HEIGHT = 2,
422 /* ALLOC_UNSTUFF = 3, TBD and rather complicated */
423};
424
425/**
426 * gfs2_bmap_alloc - Build a metadata tree of the requested height
427 * @inode: The GFS2 inode
428 * @lblock: The logical starting block of the extent
429 * @bh_map: This is used to return the mapping details
430 * @mp: The metapath
431 * @sheight: The starting height (i.e. whats already mapped)
432 * @height: The height to build to
433 * @maxlen: The max number of data blocks to alloc
434 *
435 * In this routine we may have to alloc:
436 * i) Indirect blocks to grow the metadata tree height
437 * ii) Indirect blocks to fill in lower part of the metadata tree
438 * iii) Data blocks
439 *
440 * The function is in two parts. The first part works out the total
441 * number of blocks which we need. The second part does the actual
442 * allocation asking for an extent at a time (if enough contiguous free
443 * blocks are available, there will only be one request per bmap call)
444 * and uses the state machine to initialise the blocks in order.
445 *
446 * Returns: errno on error
447 */
448
449static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
450 struct buffer_head *bh_map, struct metapath *mp,
451 const unsigned int sheight,
452 const unsigned int height,
b650738c 453 const size_t maxlen)
9b8c81d1
SW
454{
455 struct gfs2_inode *ip = GFS2_I(inode);
456 struct gfs2_sbd *sdp = GFS2_SB(inode);
64dd153c 457 struct super_block *sb = sdp->sd_vfs;
9b8c81d1
SW
458 struct buffer_head *dibh = mp->mp_bh[0];
459 u64 bn, dblock = 0;
5af4e7a0 460 unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
9b8c81d1
SW
461 unsigned dblks = 0;
462 unsigned ptrs_per_blk;
463 const unsigned end_of_metadata = height - 1;
64dd153c 464 int ret;
9b8c81d1
SW
465 int eob = 0;
466 enum alloc_state state;
467 __be64 *ptr;
468 __be64 zero_bn = 0;
469
470 BUG_ON(sheight < 1);
471 BUG_ON(dibh == NULL);
472
350a9b0a 473 gfs2_trans_add_meta(ip->i_gl, dibh);
9b8c81d1
SW
474
475 if (height == sheight) {
476 struct buffer_head *bh;
477 /* Bottom indirect block exists, find unalloced extent size */
478 ptr = metapointer(end_of_metadata, mp);
479 bh = mp->mp_bh[end_of_metadata];
480 dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
481 &eob);
482 BUG_ON(dblks < 1);
483 state = ALLOC_DATA;
484 } else {
485 /* Need to allocate indirect blocks */
486 ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
b650738c
BP
487 dblks = min(maxlen, (size_t)(ptrs_per_blk -
488 mp->mp_list[end_of_metadata]));
9b8c81d1
SW
489 if (height == ip->i_height) {
490 /* Writing into existing tree, extend tree down */
491 iblks = height - sheight;
492 state = ALLOC_GROW_DEPTH;
493 } else {
494 /* Building up tree height */
495 state = ALLOC_GROW_HEIGHT;
496 iblks = height - ip->i_height;
5af4e7a0
BM
497 branch_start = metapath_branch_start(mp);
498 iblks += (height - branch_start);
9b8c81d1
SW
499 }
500 }
501
502 /* start of the second part of the function (state machine) */
503
504 blks = dblks + iblks;
505 i = sheight;
506 do {
09010978 507 int error;
9b8c81d1 508 n = blks - alloced;
6e87ed0f 509 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
09010978
SW
510 if (error)
511 return error;
9b8c81d1
SW
512 alloced += n;
513 if (state != ALLOC_DATA || gfs2_is_jdata(ip))
514 gfs2_trans_add_unrevoke(sdp, bn, n);
515 switch (state) {
516 /* Growing height of tree */
517 case ALLOC_GROW_HEIGHT:
518 if (i == 1) {
519 ptr = (__be64 *)(dibh->b_data +
520 sizeof(struct gfs2_dinode));
521 zero_bn = *ptr;
522 }
523 for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
524 gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
525 if (i - 1 == height - ip->i_height) {
526 i--;
527 gfs2_buffer_copy_tail(mp->mp_bh[i],
528 sizeof(struct gfs2_meta_header),
529 dibh, sizeof(struct gfs2_dinode));
530 gfs2_buffer_clear_tail(dibh,
531 sizeof(struct gfs2_dinode) +
532 sizeof(__be64));
533 ptr = (__be64 *)(mp->mp_bh[i]->b_data +
534 sizeof(struct gfs2_meta_header));
535 *ptr = zero_bn;
536 state = ALLOC_GROW_DEPTH;
5af4e7a0 537 for(i = branch_start; i < height; i++) {
9b8c81d1
SW
538 if (mp->mp_bh[i] == NULL)
539 break;
540 brelse(mp->mp_bh[i]);
541 mp->mp_bh[i] = NULL;
542 }
5af4e7a0 543 i = branch_start;
9b8c81d1
SW
544 }
545 if (n == 0)
546 break;
547 /* Branching from existing tree */
548 case ALLOC_GROW_DEPTH:
549 if (i > 1 && i < height)
350a9b0a 550 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
9b8c81d1
SW
551 for (; i < height && n > 0; i++, n--)
552 gfs2_indirect_init(mp, ip->i_gl, i,
553 mp->mp_list[i-1], bn++);
554 if (i == height)
555 state = ALLOC_DATA;
556 if (n == 0)
557 break;
558 /* Tree complete, adding data blocks */
559 case ALLOC_DATA:
560 BUG_ON(n > dblks);
561 BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
350a9b0a 562 gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
9b8c81d1
SW
563 dblks = n;
564 ptr = metapointer(end_of_metadata, mp);
565 dblock = bn;
566 while (n-- > 0)
567 *ptr++ = cpu_to_be64(bn++);
64dd153c
BM
568 if (buffer_zeronew(bh_map)) {
569 ret = sb_issue_zeroout(sb, dblock, dblks,
570 GFP_NOFS);
571 if (ret) {
572 fs_err(sdp,
573 "Failed to zero data buffers\n");
574 clear_buffer_zeronew(bh_map);
575 }
576 }
9b8c81d1
SW
577 break;
578 }
07ccb7bf 579 } while ((state != ALLOC_DATA) || !dblock);
9b8c81d1
SW
580
581 ip->i_height = height;
582 gfs2_add_inode_blocks(&ip->i_inode, alloced);
583 gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
584 map_bh(bh_map, inode->i_sb, dblock);
585 bh_map->b_size = dblks << inode->i_blkbits;
586 set_buffer_new(bh_map);
587 return 0;
588}
589
b3b94faa 590/**
4cf1ed81 591 * gfs2_block_map - Map a block from an inode to a disk block
fd88de56 592 * @inode: The inode
b3b94faa 593 * @lblock: The logical block number
4cf1ed81 594 * @bh_map: The bh to be mapped
9b8c81d1 595 * @create: True if its ok to alloc blocks to satify the request
b3b94faa 596 *
9b8c81d1
SW
597 * Sets buffer_mapped() if successful, sets buffer_boundary() if a
598 * read of metadata will be required before the next block can be
599 * mapped. Sets buffer_new() if new blocks were allocated.
b3b94faa
DT
600 *
601 * Returns: errno
602 */
603
e9e1ef2b
BP
604int gfs2_block_map(struct inode *inode, sector_t lblock,
605 struct buffer_head *bh_map, int create)
b3b94faa 606{
feaa7bba
SW
607 struct gfs2_inode *ip = GFS2_I(inode);
608 struct gfs2_sbd *sdp = GFS2_SB(inode);
ecc30c79 609 unsigned int bsize = sdp->sd_sb.sb_bsize;
b650738c 610 const size_t maxlen = bh_map->b_size >> inode->i_blkbits;
ecc30c79 611 const u64 *arr = sdp->sd_heightsize;
9b8c81d1
SW
612 __be64 *ptr;
613 u64 size;
614 struct metapath mp;
615 int ret;
616 int eob;
617 unsigned int len;
618 struct buffer_head *bh;
619 u8 height;
7276b3b0 620
9b8c81d1 621 BUG_ON(maxlen == 0);
b3b94faa 622
dbac6710 623 memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
9b8c81d1 624 bmap_lock(ip, create);
4cf1ed81
SW
625 clear_buffer_mapped(bh_map);
626 clear_buffer_new(bh_map);
627 clear_buffer_boundary(bh_map);
63997775 628 trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
ecc30c79
SW
629 if (gfs2_is_dir(ip)) {
630 bsize = sdp->sd_jbsize;
631 arr = sdp->sd_jheightsize;
632 }
4cf1ed81 633
9b8c81d1
SW
634 ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
635 if (ret)
636 goto out;
b3b94faa 637
9b8c81d1
SW
638 height = ip->i_height;
639 size = (lblock + 1) * bsize;
640 while (size > arr[height])
641 height++;
642 find_metapath(sdp, lblock, &mp, height);
643 ret = 1;
644 if (height > ip->i_height || gfs2_is_stuffed(ip))
645 goto do_alloc;
646 ret = lookup_metapath(ip, &mp);
647 if (ret < 0)
648 goto out;
649 if (ret != ip->i_height)
650 goto do_alloc;
651 ptr = metapointer(ip->i_height - 1, &mp);
652 if (*ptr == 0)
653 goto do_alloc;
654 map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
655 bh = mp.mp_bh[ip->i_height - 1];
656 len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
657 bh_map->b_size = (len << inode->i_blkbits);
658 if (eob)
659 set_buffer_boundary(bh_map);
660 ret = 0;
661out:
662 release_metapath(&mp);
63997775 663 trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
9b8c81d1
SW
664 bmap_unlock(ip, create);
665 return ret;
30cbf189 666
9b8c81d1
SW
667do_alloc:
668 /* All allocations are done here, firstly check create flag */
669 if (!create) {
670 BUG_ON(gfs2_is_stuffed(ip));
671 ret = 0;
672 goto out;
b3b94faa 673 }
9b8c81d1
SW
674
675 /* At this point ret is the tree depth of already allocated blocks */
676 ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
677 goto out;
fd88de56
SW
678}
679
941e6d7d
SW
680/*
681 * Deprecated: do not use in new code
682 */
fd88de56
SW
683int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
684{
23591256 685 struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
7a6bbacb 686 int ret;
fd88de56
SW
687 int create = *new;
688
689 BUG_ON(!extlen);
690 BUG_ON(!dblock);
691 BUG_ON(!new);
692
47a9a527 693 bh.b_size = BIT(inode->i_blkbits + (create ? 0 : 5));
e9e1ef2b 694 ret = gfs2_block_map(inode, lblock, &bh, create);
7a6bbacb
SW
695 *extlen = bh.b_size >> inode->i_blkbits;
696 *dblock = bh.b_blocknr;
697 if (buffer_new(&bh))
698 *new = 1;
699 else
700 *new = 0;
701 return ret;
b3b94faa
DT
702}
703
b3b94faa
DT
704/**
705 * do_strip - Look for a layer a particular layer of the file and strip it off
706 * @ip: the inode
707 * @dibh: the dinode buffer
708 * @bh: A buffer of pointers
709 * @top: The first pointer in the buffer
710 * @bottom: One more than the last pointer
711 * @height: the height this buffer is at
c62baf65 712 * @sm: a pointer to a struct strip_mine
b3b94faa
DT
713 *
714 * Returns: errno
715 */
716
717static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
b44b84d7 718 struct buffer_head *bh, __be64 *top, __be64 *bottom,
d56fa8a1 719 unsigned int height, struct strip_mine *sm)
b3b94faa 720{
feaa7bba 721 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
b3b94faa 722 struct gfs2_rgrp_list rlist;
cd915493 723 u64 bn, bstart;
4c16c36a 724 u32 blen, btotal;
b44b84d7 725 __be64 *p;
b3b94faa
DT
726 unsigned int rg_blocks = 0;
727 int metadata;
728 unsigned int revokes = 0;
729 int x;
5e2f7d61
BP
730 int error;
731
732 error = gfs2_rindex_update(sdp);
733 if (error)
734 return error;
b3b94faa
DT
735
736 if (!*top)
737 sm->sm_first = 0;
738
739 if (height != sm->sm_height)
740 return 0;
741
742 if (sm->sm_first) {
743 top++;
744 sm->sm_first = 0;
745 }
746
ecc30c79 747 metadata = (height != ip->i_height - 1);
b3b94faa
DT
748 if (metadata)
749 revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
6d3117b4
SW
750 else if (ip->i_depth)
751 revokes = sdp->sd_inptrs;
b3b94faa 752
b3b94faa
DT
753 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
754 bstart = 0;
755 blen = 0;
756
757 for (p = top; p < bottom; p++) {
758 if (!*p)
759 continue;
760
761 bn = be64_to_cpu(*p);
762
763 if (bstart + blen == bn)
764 blen++;
765 else {
766 if (bstart)
70b0c365 767 gfs2_rlist_add(ip, &rlist, bstart);
b3b94faa
DT
768
769 bstart = bn;
770 blen = 1;
771 }
772 }
773
774 if (bstart)
70b0c365 775 gfs2_rlist_add(ip, &rlist, bstart);
b3b94faa
DT
776 else
777 goto out; /* Nothing to do */
778
fe6c991c 779 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
b3b94faa
DT
780
781 for (x = 0; x < rlist.rl_rgrps; x++) {
782 struct gfs2_rgrpd *rgd;
5c676f6d 783 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
bb8d8a6f 784 rg_blocks += rgd->rd_length;
b3b94faa
DT
785 }
786
787 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
788 if (error)
789 goto out_rlist;
790
a097dc7e
BP
791 if (gfs2_rs_active(&ip->i_res)) /* needs to be done with the rgrp glock held */
792 gfs2_rs_deltree(&ip->i_res);
8e2e0047 793
b3b94faa
DT
794 error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
795 RES_INDIRECT + RES_STATFS + RES_QUOTA,
796 revokes);
797 if (error)
798 goto out_rg_gunlock;
799
800 down_write(&ip->i_rw_mutex);
801
350a9b0a
SW
802 gfs2_trans_add_meta(ip->i_gl, dibh);
803 gfs2_trans_add_meta(ip->i_gl, bh);
b3b94faa
DT
804
805 bstart = 0;
806 blen = 0;
4c16c36a 807 btotal = 0;
b3b94faa
DT
808
809 for (p = top; p < bottom; p++) {
810 if (!*p)
811 continue;
812
813 bn = be64_to_cpu(*p);
814
815 if (bstart + blen == bn)
816 blen++;
817 else {
818 if (bstart) {
46fcb2ed 819 __gfs2_free_blocks(ip, bstart, blen, metadata);
4c16c36a 820 btotal += blen;
b3b94faa
DT
821 }
822
823 bstart = bn;
824 blen = 1;
825 }
826
827 *p = 0;
77658aad 828 gfs2_add_inode_blocks(&ip->i_inode, -1);
b3b94faa
DT
829 }
830 if (bstart) {
46fcb2ed 831 __gfs2_free_blocks(ip, bstart, blen, metadata);
4c16c36a 832 btotal += blen;
b3b94faa
DT
833 }
834
4c16c36a
BP
835 gfs2_statfs_change(sdp, 0, +btotal, 0);
836 gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
837 ip->i_inode.i_gid);
838
078cd827 839 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
b3b94faa 840
539e5d6b 841 gfs2_dinode_out(ip, dibh->b_data);
b3b94faa
DT
842
843 up_write(&ip->i_rw_mutex);
844
845 gfs2_trans_end(sdp);
846
a91ea69f 847out_rg_gunlock:
b3b94faa 848 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
a91ea69f 849out_rlist:
b3b94faa 850 gfs2_rlist_free(&rlist);
a91ea69f 851out:
b3b94faa
DT
852 return error;
853}
854
d56fa8a1
SW
855/**
856 * recursive_scan - recursively scan through the end of a file
857 * @ip: the inode
858 * @dibh: the dinode buffer
859 * @mp: the path through the metadata to the point to start
860 * @height: the height the recursion is at
861 * @block: the indirect block to look at
862 * @first: 1 if this is the first block
863 * @sm: data opaque to this function to pass to @bc
864 *
865 * When this is first called @height and @block should be zero and
866 * @first should be 1.
867 *
868 * Returns: errno
869 */
870
871static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
872 struct metapath *mp, unsigned int height,
873 u64 block, int first, struct strip_mine *sm)
874{
875 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
876 struct buffer_head *bh = NULL;
b99b98dc 877 __be64 *top, *bottom;
d56fa8a1
SW
878 u64 bn;
879 int error;
880 int mh_size = sizeof(struct gfs2_meta_header);
881
882 if (!height) {
883 error = gfs2_meta_inode_buffer(ip, &bh);
884 if (error)
885 return error;
886 dibh = bh;
887
888 top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
889 bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
890 } else {
f2f9c812 891 error = gfs2_meta_indirect_buffer(ip, height, block, &bh);
d56fa8a1
SW
892 if (error)
893 return error;
894
895 top = (__be64 *)(bh->b_data + mh_size) +
896 (first ? mp->mp_list[height] : 0);
897
898 bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
899 }
900
901 error = do_strip(ip, dibh, bh, top, bottom, height, sm);
902 if (error)
903 goto out;
904
bd5437a7 905 if (height < ip->i_height - 1) {
bd5437a7 906
b99b98dc 907 gfs2_metapath_ra(ip->i_gl, bh, top);
bd5437a7 908
d56fa8a1
SW
909 for (; top < bottom; top++, first = 0) {
910 if (!*top)
911 continue;
912
913 bn = be64_to_cpu(*top);
914
915 error = recursive_scan(ip, dibh, mp, height + 1, bn,
916 first, sm);
917 if (error)
918 break;
919 }
bd5437a7 920 }
d56fa8a1
SW
921out:
922 brelse(bh);
923 return error;
924}
925
926
ba7f7290
SW
927/**
928 * gfs2_block_truncate_page - Deal with zeroing out data for truncate
929 *
930 * This is partly borrowed from ext3.
931 */
ff8f33c8 932static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
ba7f7290
SW
933{
934 struct inode *inode = mapping->host;
935 struct gfs2_inode *ip = GFS2_I(inode);
09cbfeaf
KS
936 unsigned long index = from >> PAGE_SHIFT;
937 unsigned offset = from & (PAGE_SIZE-1);
ba7f7290
SW
938 unsigned blocksize, iblock, length, pos;
939 struct buffer_head *bh;
940 struct page *page;
ba7f7290
SW
941 int err;
942
220cca2a 943 page = find_or_create_page(mapping, index, GFP_NOFS);
ba7f7290
SW
944 if (!page)
945 return 0;
946
947 blocksize = inode->i_sb->s_blocksize;
948 length = blocksize - (offset & (blocksize - 1));
09cbfeaf 949 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
ba7f7290
SW
950
951 if (!page_has_buffers(page))
952 create_empty_buffers(page, blocksize, 0);
953
954 /* Find the buffer that contains "offset" */
955 bh = page_buffers(page);
956 pos = blocksize;
957 while (offset >= pos) {
958 bh = bh->b_this_page;
959 iblock++;
960 pos += blocksize;
961 }
962
963 err = 0;
964
965 if (!buffer_mapped(bh)) {
e9e1ef2b 966 gfs2_block_map(inode, iblock, bh, 0);
ba7f7290
SW
967 /* unmapped? It's a hole - nothing to do */
968 if (!buffer_mapped(bh))
969 goto unlock;
970 }
971
972 /* Ok, it's mapped. Make sure it's up-to-date */
973 if (PageUptodate(page))
974 set_buffer_uptodate(bh);
975
976 if (!buffer_uptodate(bh)) {
977 err = -EIO;
dfec8a14 978 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
ba7f7290
SW
979 wait_on_buffer(bh);
980 /* Uhhuh. Read error. Complain and punt. */
981 if (!buffer_uptodate(bh))
982 goto unlock;
1875f2f3 983 err = 0;
ba7f7290
SW
984 }
985
bf36a713 986 if (!gfs2_is_writeback(ip))
350a9b0a 987 gfs2_trans_add_data(ip->i_gl, bh);
ba7f7290 988
eebd2aa3 989 zero_user(page, offset, length);
40bc9a27 990 mark_buffer_dirty(bh);
ba7f7290
SW
991unlock:
992 unlock_page(page);
09cbfeaf 993 put_page(page);
ba7f7290
SW
994 return err;
995}
996
c62baf65
FF
997#define GFS2_JTRUNC_REVOKES 8192
998
fa731fc4
SW
999/**
1000 * gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
1001 * @inode: The inode being truncated
1002 * @oldsize: The original (larger) size
1003 * @newsize: The new smaller size
1004 *
1005 * With jdata files, we have to journal a revoke for each block which is
1006 * truncated. As a result, we need to split this into separate transactions
1007 * if the number of pages being truncated gets too large.
1008 */
1009
fa731fc4
SW
1010static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
1011{
1012 struct gfs2_sbd *sdp = GFS2_SB(inode);
1013 u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
1014 u64 chunk;
1015 int error;
1016
1017 while (oldsize != newsize) {
1018 chunk = oldsize - newsize;
1019 if (chunk > max_chunk)
1020 chunk = max_chunk;
7caef267 1021 truncate_pagecache(inode, oldsize - chunk);
fa731fc4
SW
1022 oldsize -= chunk;
1023 gfs2_trans_end(sdp);
1024 error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
1025 if (error)
1026 return error;
1027 }
1028
1029 return 0;
1030}
1031
ff8f33c8 1032static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
b3b94faa 1033{
ff8f33c8
SW
1034 struct gfs2_inode *ip = GFS2_I(inode);
1035 struct gfs2_sbd *sdp = GFS2_SB(inode);
1036 struct address_space *mapping = inode->i_mapping;
b3b94faa
DT
1037 struct buffer_head *dibh;
1038 int journaled = gfs2_is_jdata(ip);
1039 int error;
1040
fa731fc4
SW
1041 if (journaled)
1042 error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
1043 else
1044 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
b3b94faa
DT
1045 if (error)
1046 return error;
1047
1048 error = gfs2_meta_inode_buffer(ip, &dibh);
1049 if (error)
1050 goto out;
1051
350a9b0a 1052 gfs2_trans_add_meta(ip->i_gl, dibh);
ff8f33c8 1053
b3b94faa 1054 if (gfs2_is_stuffed(ip)) {
ff8f33c8 1055 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
b3b94faa 1056 } else {
ff8f33c8
SW
1057 if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
1058 error = gfs2_block_truncate_page(mapping, newsize);
1059 if (error)
1060 goto out_brelse;
b3b94faa 1061 }
ff8f33c8 1062 ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
b3b94faa
DT
1063 }
1064
ff8f33c8 1065 i_size_write(inode, newsize);
078cd827 1066 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
ff8f33c8 1067 gfs2_dinode_out(ip, dibh->b_data);
b3b94faa 1068
fa731fc4
SW
1069 if (journaled)
1070 error = gfs2_journaled_truncate(inode, oldsize, newsize);
1071 else
7caef267 1072 truncate_pagecache(inode, newsize);
fa731fc4
SW
1073
1074 if (error) {
1075 brelse(dibh);
1076 return error;
1077 }
1078
ff8f33c8
SW
1079out_brelse:
1080 brelse(dibh);
a91ea69f 1081out:
b3b94faa 1082 gfs2_trans_end(sdp);
b3b94faa
DT
1083 return error;
1084}
1085
cd915493 1086static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
b3b94faa 1087{
9b8c81d1 1088 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
ecc30c79 1089 unsigned int height = ip->i_height;
cd915493 1090 u64 lblock;
b3b94faa
DT
1091 struct metapath mp;
1092 int error;
1093
1094 if (!size)
1095 lblock = 0;
18ec7d5c 1096 else
9b8c81d1 1097 lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
b3b94faa 1098
9b8c81d1 1099 find_metapath(sdp, lblock, &mp, ip->i_height);
5407e242
BP
1100 error = gfs2_rindex_update(sdp);
1101 if (error)
1102 return error;
b3b94faa 1103
f4108a60 1104 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
b3b94faa 1105 if (error)
5407e242 1106 return error;
b3b94faa
DT
1107
1108 while (height--) {
1109 struct strip_mine sm;
1110 sm.sm_first = !!size;
1111 sm.sm_height = height;
1112
d56fa8a1 1113 error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm);
b3b94faa
DT
1114 if (error)
1115 break;
1116 }
1117
1118 gfs2_quota_unhold(ip);
1119
b3b94faa
DT
1120 return error;
1121}
1122
1123static int trunc_end(struct gfs2_inode *ip)
1124{
feaa7bba 1125 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
b3b94faa
DT
1126 struct buffer_head *dibh;
1127 int error;
1128
1129 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1130 if (error)
1131 return error;
1132
1133 down_write(&ip->i_rw_mutex);
1134
1135 error = gfs2_meta_inode_buffer(ip, &dibh);
1136 if (error)
1137 goto out;
1138
a2e0f799 1139 if (!i_size_read(&ip->i_inode)) {
ecc30c79 1140 ip->i_height = 0;
ce276b06 1141 ip->i_goal = ip->i_no_addr;
b3b94faa 1142 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
45138990 1143 gfs2_ordered_del_inode(ip);
b3b94faa 1144 }
078cd827 1145 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
383f01fb 1146 ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
b3b94faa 1147
350a9b0a 1148 gfs2_trans_add_meta(ip->i_gl, dibh);
539e5d6b 1149 gfs2_dinode_out(ip, dibh->b_data);
b3b94faa
DT
1150 brelse(dibh);
1151
a91ea69f 1152out:
b3b94faa 1153 up_write(&ip->i_rw_mutex);
b3b94faa 1154 gfs2_trans_end(sdp);
b3b94faa
DT
1155 return error;
1156}
1157
1158/**
1159 * do_shrink - make a file smaller
ff8f33c8
SW
1160 * @inode: the inode
1161 * @oldsize: the current inode size
1162 * @newsize: the size to make the file
b3b94faa 1163 *
ff8f33c8
SW
1164 * Called with an exclusive lock on @inode. The @size must
1165 * be equal to or smaller than the current inode size.
b3b94faa
DT
1166 *
1167 * Returns: errno
1168 */
1169
ff8f33c8 1170static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
b3b94faa 1171{
ff8f33c8 1172 struct gfs2_inode *ip = GFS2_I(inode);
b3b94faa
DT
1173 int error;
1174
ff8f33c8 1175 error = trunc_start(inode, oldsize, newsize);
b3b94faa
DT
1176 if (error < 0)
1177 return error;
ff8f33c8 1178 if (gfs2_is_stuffed(ip))
b3b94faa
DT
1179 return 0;
1180
ff8f33c8
SW
1181 error = trunc_dealloc(ip, newsize);
1182 if (error == 0)
b3b94faa
DT
1183 error = trunc_end(ip);
1184
1185 return error;
1186}
1187
ff8f33c8 1188void gfs2_trim_blocks(struct inode *inode)
a13b8c5f 1189{
ff8f33c8
SW
1190 u64 size = inode->i_size;
1191 int ret;
1192
1193 ret = do_shrink(inode, size, size);
1194 WARN_ON(ret != 0);
1195}
1196
1197/**
1198 * do_grow - Touch and update inode size
1199 * @inode: The inode
1200 * @size: The new size
1201 *
1202 * This function updates the timestamps on the inode and
1203 * may also increase the size of the inode. This function
1204 * must not be called with @size any smaller than the current
1205 * inode size.
1206 *
1207 * Although it is not strictly required to unstuff files here,
1208 * earlier versions of GFS2 have a bug in the stuffed file reading
1209 * code which will result in a buffer overrun if the size is larger
1210 * than the max stuffed file size. In order to prevent this from
25985edc 1211 * occurring, such files are unstuffed, but in other cases we can
ff8f33c8
SW
1212 * just update the inode size directly.
1213 *
1214 * Returns: 0 on success, or -ve on error
1215 */
1216
1217static int do_grow(struct inode *inode, u64 size)
1218{
1219 struct gfs2_inode *ip = GFS2_I(inode);
1220 struct gfs2_sbd *sdp = GFS2_SB(inode);
7b9cff46 1221 struct gfs2_alloc_parms ap = { .target = 1, };
a13b8c5f
WC
1222 struct buffer_head *dibh;
1223 int error;
2f7ee358 1224 int unstuff = 0;
a13b8c5f 1225
ff8f33c8
SW
1226 if (gfs2_is_stuffed(ip) &&
1227 (size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
b8fbf471 1228 error = gfs2_quota_lock_check(ip, &ap);
ff8f33c8 1229 if (error)
5407e242 1230 return error;
ff8f33c8 1231
7b9cff46 1232 error = gfs2_inplace_reserve(ip, &ap);
ff8f33c8
SW
1233 if (error)
1234 goto do_grow_qunlock;
2f7ee358 1235 unstuff = 1;
ff8f33c8
SW
1236 }
1237
a01aedfe
BP
1238 error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
1239 (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
1240 0 : RES_QUOTA), 0);
a13b8c5f 1241 if (error)
ff8f33c8 1242 goto do_grow_release;
a13b8c5f 1243
2f7ee358 1244 if (unstuff) {
ff8f33c8
SW
1245 error = gfs2_unstuff_dinode(ip, NULL);
1246 if (error)
1247 goto do_end_trans;
1248 }
a13b8c5f
WC
1249
1250 error = gfs2_meta_inode_buffer(ip, &dibh);
1251 if (error)
ff8f33c8 1252 goto do_end_trans;
a13b8c5f 1253
ff8f33c8 1254 i_size_write(inode, size);
078cd827 1255 ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
350a9b0a 1256 gfs2_trans_add_meta(ip->i_gl, dibh);
a13b8c5f
WC
1257 gfs2_dinode_out(ip, dibh->b_data);
1258 brelse(dibh);
1259
ff8f33c8 1260do_end_trans:
a13b8c5f 1261 gfs2_trans_end(sdp);
ff8f33c8 1262do_grow_release:
2f7ee358 1263 if (unstuff) {
ff8f33c8
SW
1264 gfs2_inplace_release(ip);
1265do_grow_qunlock:
1266 gfs2_quota_unlock(ip);
ff8f33c8 1267 }
a13b8c5f
WC
1268 return error;
1269}
1270
b3b94faa 1271/**
ff8f33c8
SW
1272 * gfs2_setattr_size - make a file a given size
1273 * @inode: the inode
1274 * @newsize: the size to make the file
b3b94faa 1275 *
ff8f33c8
SW
1276 * The file size can grow, shrink, or stay the same size. This
1277 * is called holding i_mutex and an exclusive glock on the inode
1278 * in question.
b3b94faa
DT
1279 *
1280 * Returns: errno
1281 */
1282
ff8f33c8 1283int gfs2_setattr_size(struct inode *inode, u64 newsize)
b3b94faa 1284{
af5c2697 1285 struct gfs2_inode *ip = GFS2_I(inode);
ff8f33c8
SW
1286 int ret;
1287 u64 oldsize;
b3b94faa 1288
ff8f33c8 1289 BUG_ON(!S_ISREG(inode->i_mode));
b3b94faa 1290
ff8f33c8
SW
1291 ret = inode_newsize_ok(inode, newsize);
1292 if (ret)
1293 return ret;
b3b94faa 1294
562c72aa
CH
1295 inode_dio_wait(inode);
1296
b54e9a0b 1297 ret = gfs2_rsqa_alloc(ip);
d2b47cfb 1298 if (ret)
2b3dcf35 1299 goto out;
d2b47cfb 1300
ff8f33c8 1301 oldsize = inode->i_size;
2b3dcf35
BP
1302 if (newsize >= oldsize) {
1303 ret = do_grow(inode, newsize);
1304 goto out;
1305 }
ff8f33c8 1306
2b3dcf35
BP
1307 ret = do_shrink(inode, oldsize, newsize);
1308out:
a097dc7e 1309 gfs2_rsqa_delete(ip, NULL);
2b3dcf35 1310 return ret;
b3b94faa
DT
1311}
1312
1313int gfs2_truncatei_resume(struct gfs2_inode *ip)
1314{
1315 int error;
a2e0f799 1316 error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
b3b94faa
DT
1317 if (!error)
1318 error = trunc_end(ip);
1319 return error;
1320}
1321
1322int gfs2_file_dealloc(struct gfs2_inode *ip)
1323{
1324 return trunc_dealloc(ip, 0);
1325}
1326
b50f227b
SW
1327/**
1328 * gfs2_free_journal_extents - Free cached journal bmap info
1329 * @jd: The journal
1330 *
1331 */
1332
1333void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
1334{
1335 struct gfs2_journal_extent *jext;
1336
1337 while(!list_empty(&jd->extent_list)) {
1338 jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
1339 list_del(&jext->list);
1340 kfree(jext);
1341 }
1342}
1343
1344/**
1345 * gfs2_add_jextent - Add or merge a new extent to extent cache
1346 * @jd: The journal descriptor
1347 * @lblock: The logical block at start of new extent
c62baf65 1348 * @dblock: The physical block at start of new extent
b50f227b
SW
1349 * @blocks: Size of extent in fs blocks
1350 *
1351 * Returns: 0 on success or -ENOMEM
1352 */
1353
1354static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
1355{
1356 struct gfs2_journal_extent *jext;
1357
1358 if (!list_empty(&jd->extent_list)) {
1359 jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
1360 if ((jext->dblock + jext->blocks) == dblock) {
1361 jext->blocks += blocks;
1362 return 0;
1363 }
1364 }
1365
1366 jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
1367 if (jext == NULL)
1368 return -ENOMEM;
1369 jext->dblock = dblock;
1370 jext->lblock = lblock;
1371 jext->blocks = blocks;
1372 list_add_tail(&jext->list, &jd->extent_list);
1373 jd->nr_extents++;
1374 return 0;
1375}
1376
1377/**
1378 * gfs2_map_journal_extents - Cache journal bmap info
1379 * @sdp: The super block
1380 * @jd: The journal to map
1381 *
1382 * Create a reusable "extent" mapping from all logical
1383 * blocks to all physical blocks for the given journal. This will save
1384 * us time when writing journal blocks. Most journals will have only one
1385 * extent that maps all their logical blocks. That's because gfs2.mkfs
1386 * arranges the journal blocks sequentially to maximize performance.
1387 * So the extent would map the first block for the entire file length.
1388 * However, gfs2_jadd can happen while file activity is happening, so
1389 * those journals may not be sequential. Less likely is the case where
1390 * the users created their own journals by mounting the metafs and
1391 * laying it out. But it's still possible. These journals might have
1392 * several extents.
1393 *
1394 * Returns: 0 on success, or error on failure
1395 */
1396
1397int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
1398{
1399 u64 lblock = 0;
1400 u64 lblock_stop;
1401 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1402 struct buffer_head bh;
1403 unsigned int shift = sdp->sd_sb.sb_bsize_shift;
1404 u64 size;
1405 int rc;
1406
1407 lblock_stop = i_size_read(jd->jd_inode) >> shift;
1408 size = (lblock_stop - lblock) << shift;
1409 jd->nr_extents = 0;
1410 WARN_ON(!list_empty(&jd->extent_list));
1411
1412 do {
1413 bh.b_state = 0;
1414 bh.b_blocknr = 0;
1415 bh.b_size = size;
1416 rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
1417 if (rc || !buffer_mapped(&bh))
1418 goto fail;
1419 rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
1420 if (rc)
1421 goto fail;
1422 size -= bh.b_size;
1423 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1424 } while(size > 0);
1425
1426 fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid,
1427 jd->nr_extents);
1428 return 0;
1429
1430fail:
1431 fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
1432 rc, jd->jd_jid,
1433 (unsigned long long)(i_size_read(jd->jd_inode) - size),
1434 jd->nr_extents);
1435 fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
1436 rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
1437 bh.b_state, (unsigned long long)bh.b_size);
1438 gfs2_free_journal_extents(jd);
1439 return rc;
1440}
1441
b3b94faa
DT
1442/**
1443 * gfs2_write_alloc_required - figure out if a write will require an allocation
1444 * @ip: the file being written to
1445 * @offset: the offset to write to
1446 * @len: the number of bytes being written
b3b94faa 1447 *
461cb419 1448 * Returns: 1 if an alloc is required, 0 otherwise
b3b94faa
DT
1449 */
1450
cd915493 1451int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
461cb419 1452 unsigned int len)
b3b94faa 1453{
feaa7bba 1454 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
941e6d7d
SW
1455 struct buffer_head bh;
1456 unsigned int shift;
1457 u64 lblock, lblock_stop, size;
7ed122e4 1458 u64 end_of_file;
b3b94faa 1459
b3b94faa
DT
1460 if (!len)
1461 return 0;
1462
1463 if (gfs2_is_stuffed(ip)) {
1464 if (offset + len >
1465 sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
461cb419 1466 return 1;
b3b94faa
DT
1467 return 0;
1468 }
1469
941e6d7d 1470 shift = sdp->sd_sb.sb_bsize_shift;
7ed122e4 1471 BUG_ON(gfs2_is_dir(ip));
a2e0f799 1472 end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
7ed122e4
SW
1473 lblock = offset >> shift;
1474 lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
1475 if (lblock_stop > end_of_file)
461cb419 1476 return 1;
b3b94faa 1477
941e6d7d
SW
1478 size = (lblock_stop - lblock) << shift;
1479 do {
1480 bh.b_state = 0;
1481 bh.b_size = size;
1482 gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
1483 if (!buffer_mapped(&bh))
461cb419 1484 return 1;
941e6d7d
SW
1485 size -= bh.b_size;
1486 lblock += (bh.b_size >> ip->i_inode.i_blkbits);
1487 } while(size > 0);
b3b94faa
DT
1488
1489 return 0;
1490}
1491