]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/gfs2/ops_address.c
[GFS2] kernel changes to support new gfs2_grow command
[mirror_ubuntu-jammy-kernel.git] / fs / gfs2 / ops_address.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
7ae8fa84 3 * Copyright (C) 2004-2007 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
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/pagemap.h>
fd88de56 16#include <linux/pagevec.h>
9b124fbb 17#include <linux/mpage.h>
d1665e41 18#include <linux/fs.h>
a8d638e3 19#include <linux/writeback.h>
5c676f6d 20#include <linux/gfs2_ondisk.h>
7d308590 21#include <linux/lm_interface.h>
b3b94faa
DT
22
23#include "gfs2.h"
5c676f6d 24#include "incore.h"
b3b94faa
DT
25#include "bmap.h"
26#include "glock.h"
27#include "inode.h"
b3b94faa
DT
28#include "log.h"
29#include "meta_io.h"
30#include "ops_address.h"
b3b94faa
DT
31#include "quota.h"
32#include "trans.h"
18ec7d5c 33#include "rgrp.h"
61a30dcb 34#include "ops_file.h"
5c676f6d 35#include "util.h"
4340fe62 36#include "glops.h"
b3b94faa 37
ba7f7290
SW
38
39static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
40 unsigned int from, unsigned int to)
41{
42 struct buffer_head *head = page_buffers(page);
43 unsigned int bsize = head->b_size;
44 struct buffer_head *bh;
45 unsigned int start, end;
46
47 for (bh = head, start = 0; bh != head || !start;
48 bh = bh->b_this_page, start = end) {
49 end = start + bsize;
50 if (end <= from || start >= to)
51 continue;
52 gfs2_trans_add_bh(ip->i_gl, bh, 0);
53 }
54}
55
b3b94faa 56/**
4ff14670 57 * gfs2_get_block - Fills in a buffer head with details about a block
b3b94faa
DT
58 * @inode: The inode
59 * @lblock: The block number to look up
60 * @bh_result: The buffer head to return the result in
61 * @create: Non-zero if we may add block to the file
62 *
63 * Returns: errno
64 */
65
4ff14670
SW
66int gfs2_get_block(struct inode *inode, sector_t lblock,
67 struct buffer_head *bh_result, int create)
b3b94faa 68{
23591256 69 return gfs2_block_map(inode, lblock, create, bh_result);
b3b94faa
DT
70}
71
72/**
7a6bbacb 73 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
b3b94faa
DT
74 * @inode: The inode
75 * @lblock: The block number to look up
76 * @bh_result: The buffer head to return the result in
77 * @create: Non-zero if we may add block to the file
78 *
79 * Returns: errno
80 */
81
7a6bbacb
SW
82static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
83 struct buffer_head *bh_result, int create)
b3b94faa 84{
b3b94faa
DT
85 int error;
86
23591256 87 error = gfs2_block_map(inode, lblock, 0, bh_result);
b3b94faa
DT
88 if (error)
89 return error;
7a6bbacb
SW
90 if (bh_result->b_blocknr == 0)
91 return -EIO;
92 return 0;
b3b94faa
DT
93}
94
7a6bbacb
SW
95static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
96 struct buffer_head *bh_result, int create)
623d9355 97{
23591256 98 return gfs2_block_map(inode, lblock, 0, bh_result);
623d9355 99}
7a6bbacb 100
b3b94faa
DT
101/**
102 * gfs2_writepage - Write complete page
103 * @page: Page to write
104 *
105 * Returns: errno
106 *
18ec7d5c
SW
107 * Some of this is copied from block_write_full_page() although we still
108 * call it to do most of the work.
b3b94faa
DT
109 */
110
111static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
112{
18ec7d5c 113 struct inode *inode = page->mapping->host;
f4387149
SW
114 struct gfs2_inode *ip = GFS2_I(inode);
115 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
116 loff_t i_size = i_size_read(inode);
117 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
118 unsigned offset;
b3b94faa 119 int error;
18ec7d5c 120 int done_trans = 0;
b3b94faa 121
b3b94faa
DT
122 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
123 unlock_page(page);
124 return -EIO;
125 }
5c676f6d 126 if (current->journal_info)
18ec7d5c
SW
127 goto out_ignore;
128
129 /* Is the page fully outside i_size? (truncate in progress) */
130 offset = i_size & (PAGE_CACHE_SIZE-1);
d2d7b8a2 131 if (page->index > end_index || (page->index == end_index && !offset)) {
18ec7d5c 132 page->mapping->a_ops->invalidatepage(page, 0);
b3b94faa 133 unlock_page(page);
18ec7d5c 134 return 0; /* don't care */
b3b94faa
DT
135 }
136
18ec7d5c
SW
137 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
138 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
139 if (error)
140 goto out_ignore;
f4387149
SW
141 if (!page_has_buffers(page)) {
142 create_empty_buffers(page, inode->i_sb->s_blocksize,
143 (1 << BH_Dirty)|(1 << BH_Uptodate));
144 }
18ec7d5c
SW
145 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
146 done_trans = 1;
147 }
7a6bbacb 148 error = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
18ec7d5c
SW
149 if (done_trans)
150 gfs2_trans_end(sdp);
b3b94faa 151 gfs2_meta_cache_flush(ip);
b3b94faa 152 return error;
18ec7d5c
SW
153
154out_ignore:
155 redirty_page_for_writepage(wbc, page);
156 unlock_page(page);
157 return 0;
b3b94faa
DT
158}
159
a8d638e3
SW
160/**
161 * gfs2_writepages - Write a bunch of dirty pages back to disk
162 * @mapping: The mapping to write
163 * @wbc: Write-back control
164 *
165 * For journaled files and/or ordered writes this just falls back to the
166 * kernel's default writepages path for now. We will probably want to change
167 * that eventually (i.e. when we look at allocate on flush).
168 *
169 * For the data=writeback case though we can already ignore buffer heads
170 * and write whole extents at once. This is a big reduction in the
171 * number of I/O requests we send and the bmap calls we make in this case.
172 */
a2cf8222
AB
173static int gfs2_writepages(struct address_space *mapping,
174 struct writeback_control *wbc)
a8d638e3
SW
175{
176 struct inode *inode = mapping->host;
177 struct gfs2_inode *ip = GFS2_I(inode);
178 struct gfs2_sbd *sdp = GFS2_SB(inode);
179
180 if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip))
181 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
182
183 return generic_writepages(mapping, wbc);
184}
185
b3b94faa
DT
186/**
187 * stuffed_readpage - Fill in a Linux page with stuffed file data
188 * @ip: the inode
189 * @page: the page
190 *
191 * Returns: errno
192 */
193
194static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
195{
196 struct buffer_head *dibh;
197 void *kaddr;
198 int error;
199
bf126aee
SW
200 /*
201 * Due to the order of unstuffing files and ->nopage(), we can be
202 * asked for a zero page in the case of a stuffed file being extended,
203 * so we need to supply one here. It doesn't happen often.
204 */
205 if (unlikely(page->index)) {
206 kaddr = kmap_atomic(page, KM_USER0);
207 memset(kaddr, 0, PAGE_CACHE_SIZE);
208 kunmap_atomic(kaddr, KM_USER0);
209 flush_dcache_page(page);
210 SetPageUptodate(page);
211 return 0;
212 }
fd88de56 213
b3b94faa
DT
214 error = gfs2_meta_inode_buffer(ip, &dibh);
215 if (error)
216 return error;
217
5c4e9e03 218 kaddr = kmap_atomic(page, KM_USER0);
fd88de56 219 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
b3b94faa 220 ip->i_di.di_size);
fd88de56 221 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
c312c4fd 222 kunmap_atomic(kaddr, KM_USER0);
bf126aee 223 flush_dcache_page(page);
b3b94faa 224 brelse(dibh);
b3b94faa
DT
225 SetPageUptodate(page);
226
227 return 0;
228}
229
b3b94faa 230
b3b94faa
DT
231/**
232 * gfs2_readpage - readpage with locking
18ec7d5c
SW
233 * @file: The file to read a page for. N.B. This may be NULL if we are
234 * reading an internal file.
b3b94faa
DT
235 * @page: The page to read
236 *
237 * Returns: errno
238 */
239
240static int gfs2_readpage(struct file *file, struct page *page)
241{
feaa7bba
SW
242 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
243 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
dc41aeed 244 struct gfs2_file *gf = NULL;
18ec7d5c 245 struct gfs2_holder gh;
b3b94faa 246 int error;
59a1cc6b 247 int do_unlock = 0;
b3b94faa 248
dd538c83 249 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b 250 if (file) {
dc41aeed 251 gf = file->private_data;
59a1cc6b 252 if (test_bit(GFF_EXLOCK, &gf->f_flags))
dc41aeed 253 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
59a1cc6b
SW
254 goto skip_lock;
255 }
2ca99501 256 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
59a1cc6b 257 do_unlock = 1;
dcd24799 258 error = gfs2_glock_nq_atime(&gh);
fd88de56 259 if (unlikely(error))
61a30dcb
SW
260 goto out_unlock;
261 }
b3b94faa 262
59a1cc6b 263skip_lock:
18ec7d5c 264 if (gfs2_is_stuffed(ip)) {
fd88de56
SW
265 error = stuffed_readpage(ip, page);
266 unlock_page(page);
b3b94faa 267 } else
18ec7d5c 268 error = mpage_readpage(page, gfs2_get_block);
b3b94faa
DT
269
270 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
271 error = -EIO;
272
07903c02 273 if (do_unlock) {
61a30dcb
SW
274 gfs2_glock_dq_m(1, &gh);
275 gfs2_holder_uninit(&gh);
276 }
18ec7d5c 277out:
b3b94faa 278 return error;
18ec7d5c
SW
279out_unlock:
280 unlock_page(page);
a13cbe37
JW
281 if (error == GLR_TRYFAILED) {
282 error = AOP_TRUNCATED_PAGE;
283 yield();
284 }
59a1cc6b 285 if (do_unlock)
fd88de56
SW
286 gfs2_holder_uninit(&gh);
287 goto out;
288}
289
fd88de56
SW
290/**
291 * gfs2_readpages - Read a bunch of pages at once
292 *
293 * Some notes:
294 * 1. This is only for readahead, so we can simply ignore any things
295 * which are slightly inconvenient (such as locking conflicts between
296 * the page lock and the glock) and return having done no I/O. Its
297 * obviously not something we'd want to do on too regular a basis.
298 * Any I/O we ignore at this time will be done via readpage later.
e1d5b18a 299 * 2. We don't handle stuffed files here we let readpage do the honours.
fd88de56
SW
300 * 3. mpage_readpages() does most of the heavy lifting in the common case.
301 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
302 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
303 * well as read-ahead.
304 */
305static int gfs2_readpages(struct file *file, struct address_space *mapping,
306 struct list_head *pages, unsigned nr_pages)
307{
308 struct inode *inode = mapping->host;
feaa7bba
SW
309 struct gfs2_inode *ip = GFS2_I(inode);
310 struct gfs2_sbd *sdp = GFS2_SB(inode);
fd88de56 311 struct gfs2_holder gh;
e1d5b18a 312 int ret = 0;
59a1cc6b 313 int do_unlock = 0;
fd88de56 314
dd538c83 315 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b
SW
316 if (file) {
317 struct gfs2_file *gf = file->private_data;
318 if (test_bit(GFF_EXLOCK, &gf->f_flags))
319 goto skip_lock;
320 }
fd88de56 321 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
2ca99501 322 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
59a1cc6b 323 do_unlock = 1;
dcd24799 324 ret = gfs2_glock_nq_atime(&gh);
907b9bce 325 if (ret == GLR_TRYFAILED)
fd88de56
SW
326 goto out_noerror;
327 if (unlikely(ret))
328 goto out_unlock;
329 }
59a1cc6b 330skip_lock:
e1d5b18a 331 if (!gfs2_is_stuffed(ip))
fd88de56 332 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
fd88de56 333
59a1cc6b 334 if (do_unlock) {
fd88de56
SW
335 gfs2_glock_dq_m(1, &gh);
336 gfs2_holder_uninit(&gh);
337 }
338out:
339 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
340 ret = -EIO;
341 return ret;
342out_noerror:
343 ret = 0;
344out_unlock:
59a1cc6b 345 if (do_unlock)
fd88de56 346 gfs2_holder_uninit(&gh);
18ec7d5c 347 goto out;
b3b94faa
DT
348}
349
350/**
351 * gfs2_prepare_write - Prepare to write a page to a file
352 * @file: The file to write to
353 * @page: The page which is to be prepared for writing
354 * @from: From (byte range within page)
355 * @to: To (byte range within page)
356 *
357 * Returns: errno
358 */
359
360static int gfs2_prepare_write(struct file *file, struct page *page,
361 unsigned from, unsigned to)
362{
feaa7bba
SW
363 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
364 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
18ec7d5c
SW
365 unsigned int data_blocks, ind_blocks, rblocks;
366 int alloc_required;
b3b94faa 367 int error = 0;
18ec7d5c
SW
368 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
369 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
370 struct gfs2_alloc *al;
52ae7b79
RC
371 unsigned int write_len = to - from;
372
b3b94faa 373
2ca99501 374 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
dcd24799 375 error = gfs2_glock_nq_atime(&ip->i_gh);
2ca99501 376 if (unlikely(error)) {
2d72e710
SW
377 if (error == GLR_TRYFAILED) {
378 unlock_page(page);
2ca99501 379 error = AOP_TRUNCATED_PAGE;
a13cbe37 380 yield();
2d72e710 381 }
18ec7d5c 382 goto out_uninit;
2ca99501 383 }
b3b94faa 384
52ae7b79 385 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
18ec7d5c 386
52ae7b79 387 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
18ec7d5c
SW
388 if (error)
389 goto out_unlock;
b3b94faa 390
18ec7d5c 391
f5c54804 392 ip->i_alloc.al_requested = 0;
18ec7d5c
SW
393 if (alloc_required) {
394 al = gfs2_alloc_get(ip);
395
396 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
397 if (error)
398 goto out_alloc_put;
399
2933f925 400 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
18ec7d5c
SW
401 if (error)
402 goto out_qunlock;
403
404 al->al_requested = data_blocks + ind_blocks;
405 error = gfs2_inplace_reserve(ip);
406 if (error)
407 goto out_qunlock;
408 }
409
410 rblocks = RES_DINODE + ind_blocks;
411 if (gfs2_is_jdata(ip))
412 rblocks += data_blocks ? data_blocks : 1;
413 if (ind_blocks || data_blocks)
414 rblocks += RES_STATFS + RES_QUOTA;
415
416 error = gfs2_trans_begin(sdp, rblocks, 0);
417 if (error)
418 goto out;
419
420 if (gfs2_is_stuffed(ip)) {
421 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
f25ef0c1 422 error = gfs2_unstuff_dinode(ip, page);
5c4e9e03
SW
423 if (error == 0)
424 goto prepare_write;
425 } else if (!PageUptodate(page))
b3b94faa 426 error = stuffed_readpage(ip, page);
5c4e9e03 427 goto out;
18ec7d5c
SW
428 }
429
5c4e9e03 430prepare_write:
18ec7d5c
SW
431 error = block_prepare_write(page, from, to, gfs2_get_block);
432
433out:
434 if (error) {
435 gfs2_trans_end(sdp);
436 if (alloc_required) {
437 gfs2_inplace_release(ip);
438out_qunlock:
439 gfs2_quota_unlock(ip);
440out_alloc_put:
441 gfs2_alloc_put(ip);
442 }
443out_unlock:
444 gfs2_glock_dq_m(1, &ip->i_gh);
445out_uninit:
446 gfs2_holder_uninit(&ip->i_gh);
447 }
b3b94faa
DT
448
449 return error;
450}
451
7ae8fa84
RP
452/**
453 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
454 * @inode: the rindex inode
455 */
456static void adjust_fs_space(struct inode *inode)
457{
458 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
459 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
460 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
461 u64 fs_total, new_free;
462
463 /* Total up the file system space, according to the latest rindex. */
464 fs_total = gfs2_ri_total(sdp);
465
466 spin_lock(&sdp->sd_statfs_spin);
467 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
468 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
469 else
470 new_free = 0;
471 spin_unlock(&sdp->sd_statfs_spin);
472 fs_warn(sdp, "File system extended by %llu blocks.\n", new_free);
473 gfs2_statfs_change(sdp, new_free, new_free, 0);
474}
475
b3b94faa
DT
476/**
477 * gfs2_commit_write - Commit write to a file
478 * @file: The file to write to
479 * @page: The page containing the data
480 * @from: From (byte range within page)
481 * @to: To (byte range within page)
482 *
483 * Returns: errno
484 */
485
486static int gfs2_commit_write(struct file *file, struct page *page,
487 unsigned from, unsigned to)
488{
489 struct inode *inode = page->mapping->host;
feaa7bba
SW
490 struct gfs2_inode *ip = GFS2_I(inode);
491 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
492 int error = -EOPNOTSUPP;
493 struct buffer_head *dibh;
48516ced
SW
494 struct gfs2_alloc *al = &ip->i_alloc;
495 struct gfs2_dinode *di;
b3b94faa 496
18ec7d5c
SW
497 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
498 goto fail_nounlock;
499
500 error = gfs2_meta_inode_buffer(ip, &dibh);
501 if (error)
502 goto fail_endtrans;
503
504 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
48516ced 505 di = (struct gfs2_dinode *)dibh->b_data;
18ec7d5c 506
b3b94faa 507 if (gfs2_is_stuffed(ip)) {
cd915493 508 u64 file_size;
b3b94faa
DT
509 void *kaddr;
510
cd915493 511 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
b3b94faa 512
18ec7d5c 513 kaddr = kmap_atomic(page, KM_USER0);
b3b94faa 514 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
0bd5996a 515 kaddr + from, to - from);
c312c4fd 516 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
517
518 SetPageUptodate(page);
519
ae619320 520 if (inode->i_size < file_size) {
b3b94faa 521 i_size_write(inode, file_size);
ae619320
SW
522 mark_inode_dirty(inode);
523 }
b3b94faa 524 } else {
568f4c96
SW
525 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
526 gfs2_is_jdata(ip))
257f9b4e 527 gfs2_page_add_databufs(ip, page, from, to);
b3b94faa
DT
528 error = generic_commit_write(file, page, from, to);
529 if (error)
530 goto fail;
531 }
532
48516ced 533 if (ip->i_di.di_size < inode->i_size) {
18ec7d5c 534 ip->i_di.di_size = inode->i_size;
48516ced
SW
535 di->di_size = cpu_to_be64(inode->i_size);
536 }
537
7ae8fa84
RP
538 if (inode == sdp->sd_rindex)
539 adjust_fs_space(inode);
540
18ec7d5c
SW
541 brelse(dibh);
542 gfs2_trans_end(sdp);
543 if (al->al_requested) {
544 gfs2_inplace_release(ip);
545 gfs2_quota_unlock(ip);
546 gfs2_alloc_put(ip);
547 }
1de91390 548 unlock_page(page);
18ec7d5c 549 gfs2_glock_dq_m(1, &ip->i_gh);
1de91390 550 lock_page(page);
18ec7d5c 551 gfs2_holder_uninit(&ip->i_gh);
b3b94faa
DT
552 return 0;
553
18ec7d5c
SW
554fail:
555 brelse(dibh);
556fail_endtrans:
557 gfs2_trans_end(sdp);
558 if (al->al_requested) {
559 gfs2_inplace_release(ip);
560 gfs2_quota_unlock(ip);
561 gfs2_alloc_put(ip);
562 }
1de91390 563 unlock_page(page);
18ec7d5c 564 gfs2_glock_dq_m(1, &ip->i_gh);
1de91390 565 lock_page(page);
18ec7d5c
SW
566 gfs2_holder_uninit(&ip->i_gh);
567fail_nounlock:
b3b94faa 568 ClearPageUptodate(page);
b3b94faa
DT
569 return error;
570}
571
572/**
573 * gfs2_bmap - Block map function
574 * @mapping: Address space info
575 * @lblock: The block to map
576 *
577 * Returns: The disk address for the block or 0 on hole or error
578 */
579
580static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
581{
feaa7bba 582 struct gfs2_inode *ip = GFS2_I(mapping->host);
b3b94faa
DT
583 struct gfs2_holder i_gh;
584 sector_t dblock = 0;
585 int error;
586
b3b94faa
DT
587 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
588 if (error)
589 return 0;
590
591 if (!gfs2_is_stuffed(ip))
4ff14670 592 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
b3b94faa
DT
593
594 gfs2_glock_dq_uninit(&i_gh);
595
596 return dblock;
597}
598
599static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
600{
64fb4eb7 601 struct gfs2_bufdata *bd;
b3b94faa
DT
602
603 gfs2_log_lock(sdp);
5c676f6d 604 bd = bh->b_private;
64fb4eb7
SW
605 if (bd) {
606 bd->bd_bh = NULL;
5c676f6d 607 bh->b_private = NULL;
15d00c0b
SW
608 }
609 gfs2_log_unlock(sdp);
b3b94faa
DT
610
611 lock_buffer(bh);
612 clear_buffer_dirty(bh);
613 bh->b_bdev = NULL;
614 clear_buffer_mapped(bh);
615 clear_buffer_req(bh);
616 clear_buffer_new(bh);
617 clear_buffer_delay(bh);
618 unlock_buffer(bh);
619}
620
8628de05 621static void gfs2_invalidatepage(struct page *page, unsigned long offset)
b3b94faa 622{
15d00c0b 623 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
b3b94faa
DT
624 struct buffer_head *head, *bh, *next;
625 unsigned int curr_off = 0;
b3b94faa
DT
626
627 BUG_ON(!PageLocked(page));
628 if (!page_has_buffers(page))
8628de05 629 return;
b3b94faa
DT
630
631 bh = head = page_buffers(page);
632 do {
633 unsigned int next_off = curr_off + bh->b_size;
634 next = bh->b_this_page;
635
636 if (offset <= curr_off)
637 discard_buffer(sdp, bh);
638
639 curr_off = next_off;
640 bh = next;
641 } while (bh != head);
642
643 if (!offset)
8628de05 644 try_to_release_page(page, 0);
b3b94faa 645
8628de05 646 return;
b3b94faa
DT
647}
648
c7b33834
SW
649/**
650 * gfs2_ok_for_dio - check that dio is valid on this file
651 * @ip: The inode
652 * @rw: READ or WRITE
653 * @offset: The offset at which we are reading or writing
654 *
655 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
656 * 1 (to accept the i/o request)
657 */
658static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
659{
660 /*
661 * Should we return an error here? I can't see that O_DIRECT for
662 * a journaled file makes any sense. For now we'll silently fall
663 * back to buffered I/O, likewise we do the same for stuffed
664 * files since they are (a) small and (b) unaligned.
665 */
666 if (gfs2_is_jdata(ip))
667 return 0;
668
669 if (gfs2_is_stuffed(ip))
670 return 0;
671
672 if (offset > i_size_read(&ip->i_inode))
673 return 0;
674 return 1;
675}
676
677
678
a9e5f4d0
SW
679static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
680 const struct iovec *iov, loff_t offset,
681 unsigned long nr_segs)
d1665e41
SW
682{
683 struct file *file = iocb->ki_filp;
684 struct inode *inode = file->f_mapping->host;
feaa7bba 685 struct gfs2_inode *ip = GFS2_I(inode);
d1665e41
SW
686 struct gfs2_holder gh;
687 int rv;
688
689 /*
c7b33834
SW
690 * Deferred lock, even if its a write, since we do no allocation
691 * on this path. All we need change is atime, and this lock mode
692 * ensures that other nodes have flushed their buffered read caches
693 * (i.e. their page cache entries for this inode). We do not,
694 * unfortunately have the option of only flushing a range like
695 * the VFS does.
d1665e41 696 */
c7b33834 697 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
dcd24799 698 rv = gfs2_glock_nq_atime(&gh);
d1665e41 699 if (rv)
c7b33834
SW
700 return rv;
701 rv = gfs2_ok_for_dio(ip, rw, offset);
702 if (rv != 1)
703 goto out; /* dio not valid, fall back to buffered i/o */
704
705 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
706 iov, offset, nr_segs,
707 gfs2_get_block_direct, NULL);
d1665e41
SW
708out:
709 gfs2_glock_dq_m(1, &gh);
710 gfs2_holder_uninit(&gh);
d1665e41
SW
711 return rv;
712}
713
4340fe62
SW
714/**
715 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
716 * @bh: the buffer we're stuck on
717 *
718 */
719
720static void stuck_releasepage(struct buffer_head *bh)
721{
722 struct inode *inode = bh->b_page->mapping->host;
723 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
724 struct gfs2_bufdata *bd = bh->b_private;
725 struct gfs2_glock *gl;
166afccd
SW
726static unsigned limit = 0;
727
0bd5996a 728 if (limit > 3)
166afccd 729 return;
0bd5996a 730 limit++;
4340fe62
SW
731
732 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
733 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
734 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
735 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
736 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
737
738 if (!bd)
739 return;
740
741 gl = bd->bd_gl;
742
907b9bce 743 fs_warn(sdp, "gl = (%u, %llu)\n",
4340fe62
SW
744 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
745
746 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
747 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
748 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
749
750 if (gl->gl_ops == &gfs2_inode_glops) {
751 struct gfs2_inode *ip = gl->gl_object;
752 unsigned int x;
753
754 if (!ip)
755 return;
756
757 fs_warn(sdp, "ip = %llu %llu\n",
758 (unsigned long long)ip->i_num.no_formal_ino,
759 (unsigned long long)ip->i_num.no_addr);
760
761 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
762 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
763 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
764 }
765}
766
767/**
623d9355 768 * gfs2_releasepage - free the metadata associated with a page
4340fe62
SW
769 * @page: the page that's being released
770 * @gfp_mask: passed from Linux VFS, ignored by us
771 *
772 * Call try_to_free_buffers() if the buffers in this page can be
773 * released.
774 *
775 * Returns: 0
776 */
777
778int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
779{
780 struct inode *aspace = page->mapping->host;
781 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
782 struct buffer_head *bh, *head;
783 struct gfs2_bufdata *bd;
166afccd 784 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
4340fe62
SW
785
786 if (!page_has_buffers(page))
787 goto out;
788
789 head = bh = page_buffers(page);
790 do {
4340fe62 791 while (atomic_read(&bh->b_count)) {
166afccd
SW
792 if (!atomic_read(&aspace->i_writecount))
793 return 0;
794
61057c6b
RC
795 if (!(gfp_mask & __GFP_WAIT))
796 return 0;
797
166afccd
SW
798 if (time_after_eq(jiffies, t)) {
799 stuck_releasepage(bh);
800 /* should we withdraw here? */
801 return 0;
4340fe62
SW
802 }
803
166afccd 804 yield();
4340fe62
SW
805 }
806
807 gfs2_assert_warn(sdp, !buffer_pinned(bh));
623d9355 808 gfs2_assert_warn(sdp, !buffer_dirty(bh));
4340fe62 809
623d9355 810 gfs2_log_lock(sdp);
4340fe62
SW
811 bd = bh->b_private;
812 if (bd) {
813 gfs2_assert_warn(sdp, bd->bd_bh == bh);
814 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
4340fe62 815 gfs2_assert_warn(sdp, !bd->bd_ail);
623d9355
SW
816 bd->bd_bh = NULL;
817 if (!list_empty(&bd->bd_le.le_list))
818 bd = NULL;
4340fe62
SW
819 bh->b_private = NULL;
820 }
623d9355
SW
821 gfs2_log_unlock(sdp);
822 if (bd)
823 kmem_cache_free(gfs2_bufdata_cachep, bd);
4340fe62
SW
824
825 bh = bh->b_this_page;
166afccd 826 } while (bh != head);
4340fe62 827
a9e5f4d0 828out:
4340fe62
SW
829 return try_to_free_buffers(page);
830}
831
66de045d 832const struct address_space_operations gfs2_file_aops = {
b3b94faa 833 .writepage = gfs2_writepage,
a8d638e3 834 .writepages = gfs2_writepages,
b3b94faa 835 .readpage = gfs2_readpage,
fd88de56 836 .readpages = gfs2_readpages,
b3b94faa
DT
837 .sync_page = block_sync_page,
838 .prepare_write = gfs2_prepare_write,
839 .commit_write = gfs2_commit_write,
840 .bmap = gfs2_bmap,
841 .invalidatepage = gfs2_invalidatepage,
4340fe62 842 .releasepage = gfs2_releasepage,
b3b94faa
DT
843 .direct_IO = gfs2_direct_IO,
844};
845