blob: 38dbe99a30ede1ac388596278ef02ca9e9590bae [file] [log] [blame]
David Teiglandb3b94fa2006-01-16 16:50:04 +00001/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
Robert Peterson7ae8fa82007-05-09 09:37:57 -05003 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
David Teiglandb3b94fa2006-01-16 16:50:04 +00004 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
Steven Whitehousee9fc2aa2006-09-01 11:05:15 -04007 * of the GNU General Public License version 2.
David Teiglandb3b94fa2006-01-16 16:50:04 +00008 */
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>
Steven Whitehousefd88de562006-05-05 16:59:11 -040016#include <linux/pagevec.h>
Steven Whitehouse9b124fb2006-01-30 11:55:32 +000017#include <linux/mpage.h>
Steven Whitehoused1665e42006-02-14 11:54:42 +000018#include <linux/fs.h>
Steven Whitehousea8d638e2007-01-15 13:52:17 +000019#include <linux/writeback.h>
Steven Whitehouse7765ec22007-10-16 01:25:07 -070020#include <linux/swap.h>
Steven Whitehouse5c676f62006-02-27 17:23:27 -050021#include <linux/gfs2_ondisk.h>
Fabio Massimo Di Nitto7d308592006-09-19 07:56:29 +020022#include <linux/lm_interface.h>
Steven Whitehouse47e83b52007-10-18 11:15:50 +010023#include <linux/backing-dev.h>
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +010024#include <linux/pagevec.h>
David Teiglandb3b94fa2006-01-16 16:50:04 +000025
26#include "gfs2.h"
Steven Whitehouse5c676f62006-02-27 17:23:27 -050027#include "incore.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000028#include "bmap.h"
29#include "glock.h"
30#include "inode.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000031#include "log.h"
32#include "meta_io.h"
33#include "ops_address.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000034#include "quota.h"
35#include "trans.h"
Steven Whitehouse18ec7d52006-02-08 11:50:51 +000036#include "rgrp.h"
Robert Petersoncd81a4b2007-05-14 12:42:18 -050037#include "super.h"
Steven Whitehouse5c676f62006-02-27 17:23:27 -050038#include "util.h"
Steven Whitehouse4340fe62006-07-11 09:46:33 -040039#include "glops.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000040
Steven Whitehouseba7f7292006-07-26 11:27:10 -040041
42static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
43 unsigned int from, unsigned int to)
44{
45 struct buffer_head *head = page_buffers(page);
46 unsigned int bsize = head->b_size;
47 struct buffer_head *bh;
48 unsigned int start, end;
49
50 for (bh = head, start = 0; bh != head || !start;
51 bh = bh->b_this_page, start = end) {
52 end = start + bsize;
53 if (end <= from || start >= to)
54 continue;
Benjamin Marzinskiddf4b422007-06-01 14:21:38 -050055 if (gfs2_is_jdata(ip))
56 set_buffer_uptodate(bh);
Steven Whitehouseba7f7292006-07-26 11:27:10 -040057 gfs2_trans_add_bh(ip->i_gl, bh, 0);
58 }
59}
60
David Teiglandb3b94fa2006-01-16 16:50:04 +000061/**
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040062 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
David Teiglandb3b94fa2006-01-16 16:50:04 +000063 * @inode: The inode
64 * @lblock: The block number to look up
65 * @bh_result: The buffer head to return the result in
66 * @create: Non-zero if we may add block to the file
67 *
68 * Returns: errno
69 */
70
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040071static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
72 struct buffer_head *bh_result, int create)
David Teiglandb3b94fa2006-01-16 16:50:04 +000073{
David Teiglandb3b94fa2006-01-16 16:50:04 +000074 int error;
75
Bob Petersone9e1ef22007-12-10 14:13:27 -060076 error = gfs2_block_map(inode, lblock, bh_result, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +000077 if (error)
78 return error;
Wendy Chengde986e82007-09-18 09:19:13 -040079 if (!buffer_mapped(bh_result))
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040080 return -EIO;
Steven Whitehouse623d9352006-08-31 12:14:44 -040081 return 0;
82}
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040083
84static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
85 struct buffer_head *bh_result, int create)
86{
Bob Petersone9e1ef22007-12-10 14:13:27 -060087 return gfs2_block_map(inode, lblock, bh_result, 0);
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040088}
89
David Teiglandb3b94fa2006-01-16 16:50:04 +000090/**
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +010091 * gfs2_writepage_common - Common bits of writepage
92 * @page: The page to be written
93 * @wbc: The writeback control
David Teiglandb3b94fa2006-01-16 16:50:04 +000094 *
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +010095 * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
David Teiglandb3b94fa2006-01-16 16:50:04 +000096 */
97
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +010098static int gfs2_writepage_common(struct page *page,
99 struct writeback_control *wbc)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000100{
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000101 struct inode *inode = page->mapping->host;
Steven Whitehousef4387142006-08-08 13:23:19 -0400102 struct gfs2_inode *ip = GFS2_I(inode);
103 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000104 loff_t i_size = i_size_read(inode);
105 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
106 unsigned offset;
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100107 int ret = -EIO;
108
109 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
110 goto out;
111 ret = 0;
112 if (current->journal_info)
113 goto redirty;
114 /* Is the page fully outside i_size? (truncate in progress) */
115 offset = i_size & (PAGE_CACHE_SIZE-1);
116 if (page->index > end_index || (page->index == end_index && !offset)) {
117 page->mapping->a_ops->invalidatepage(page, 0);
118 goto out;
119 }
120 return 1;
121redirty:
122 redirty_page_for_writepage(wbc, page);
123out:
124 unlock_page(page);
125 return 0;
126}
127
128/**
129 * gfs2_writeback_writepage - Write page for writeback mappings
130 * @page: The page
131 * @wbc: The writeback control
132 *
133 */
134
135static int gfs2_writeback_writepage(struct page *page,
136 struct writeback_control *wbc)
137{
138 int ret;
139
140 ret = gfs2_writepage_common(page, wbc);
141 if (ret <= 0)
142 return ret;
143
144 ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc);
145 if (ret == -EAGAIN)
146 ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
147 return ret;
148}
149
150/**
151 * gfs2_ordered_writepage - Write page for ordered data files
152 * @page: The page to write
153 * @wbc: The writeback control
154 *
155 */
156
157static int gfs2_ordered_writepage(struct page *page,
158 struct writeback_control *wbc)
159{
160 struct inode *inode = page->mapping->host;
161 struct gfs2_inode *ip = GFS2_I(inode);
162 int ret;
163
164 ret = gfs2_writepage_common(page, wbc);
165 if (ret <= 0)
166 return ret;
167
168 if (!page_has_buffers(page)) {
169 create_empty_buffers(page, inode->i_sb->s_blocksize,
170 (1 << BH_Dirty)|(1 << BH_Uptodate));
171 }
172 gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
173 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
174}
175
176/**
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100177 * __gfs2_jdata_writepage - The core of jdata writepage
178 * @page: The page to write
179 * @wbc: The writeback control
180 *
181 * This is shared between writepage and writepages and implements the
182 * core of the writepage operation. If a transaction is required then
183 * PageChecked will have been set and the transaction will have
184 * already been started before this is called.
185 */
186
187static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
188{
189 struct inode *inode = page->mapping->host;
190 struct gfs2_inode *ip = GFS2_I(inode);
191 struct gfs2_sbd *sdp = GFS2_SB(inode);
192
193 if (PageChecked(page)) {
194 ClearPageChecked(page);
195 if (!page_has_buffers(page)) {
196 create_empty_buffers(page, inode->i_sb->s_blocksize,
197 (1 << BH_Dirty)|(1 << BH_Uptodate));
198 }
199 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
200 }
201 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
202}
203
204/**
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100205 * gfs2_jdata_writepage - Write complete page
206 * @page: Page to write
207 *
208 * Returns: errno
209 *
210 */
211
212static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
213{
214 struct inode *inode = page->mapping->host;
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100215 struct gfs2_sbd *sdp = GFS2_SB(inode);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000216 int error;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000217 int done_trans = 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000218
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100219 error = gfs2_writepage_common(page, wbc);
220 if (error <= 0)
221 return error;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000222
Steven Whitehousebf36a712007-10-17 08:35:19 +0100223 if (PageChecked(page)) {
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100224 if (wbc->sync_mode != WB_SYNC_ALL)
225 goto out_ignore;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000226 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
227 if (error)
228 goto out_ignore;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000229 done_trans = 1;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000230 }
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100231 error = __gfs2_jdata_writepage(page, wbc);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000232 if (done_trans)
233 gfs2_trans_end(sdp);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000234 return error;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000235
236out_ignore:
237 redirty_page_for_writepage(wbc, page);
238 unlock_page(page);
239 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000240}
241
242/**
Steven Whitehouse55610932007-10-17 08:47:38 +0100243 * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000244 * @mapping: The mapping to write
245 * @wbc: Write-back control
246 *
Steven Whitehouse55610932007-10-17 08:47:38 +0100247 * For the data=writeback case we can already ignore buffer heads
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000248 * and write whole extents at once. This is a big reduction in the
249 * number of I/O requests we send and the bmap calls we make in this case.
250 */
Steven Whitehouse55610932007-10-17 08:47:38 +0100251static int gfs2_writeback_writepages(struct address_space *mapping,
252 struct writeback_control *wbc)
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000253{
Steven Whitehouse55610932007-10-17 08:47:38 +0100254 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000255}
256
257/**
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100258 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
259 * @mapping: The mapping
260 * @wbc: The writeback control
261 * @writepage: The writepage function to call for each page
262 * @pvec: The vector of pages
263 * @nr_pages: The number of pages to write
264 *
265 * Returns: non-zero if loop should terminate, zero otherwise
266 */
267
268static int gfs2_write_jdata_pagevec(struct address_space *mapping,
269 struct writeback_control *wbc,
270 struct pagevec *pvec,
271 int nr_pages, pgoff_t end)
272{
273 struct inode *inode = mapping->host;
274 struct gfs2_sbd *sdp = GFS2_SB(inode);
275 loff_t i_size = i_size_read(inode);
276 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
277 unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
278 unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
279 struct backing_dev_info *bdi = mapping->backing_dev_info;
280 int i;
281 int ret;
282
283 ret = gfs2_trans_begin(sdp, nrblocks, 0);
284 if (ret < 0)
285 return ret;
286
287 for(i = 0; i < nr_pages; i++) {
288 struct page *page = pvec->pages[i];
289
290 lock_page(page);
291
292 if (unlikely(page->mapping != mapping)) {
293 unlock_page(page);
294 continue;
295 }
296
297 if (!wbc->range_cyclic && page->index > end) {
298 ret = 1;
299 unlock_page(page);
300 continue;
301 }
302
303 if (wbc->sync_mode != WB_SYNC_NONE)
304 wait_on_page_writeback(page);
305
306 if (PageWriteback(page) ||
307 !clear_page_dirty_for_io(page)) {
308 unlock_page(page);
309 continue;
310 }
311
312 /* Is the page fully outside i_size? (truncate in progress) */
313 if (page->index > end_index || (page->index == end_index && !offset)) {
314 page->mapping->a_ops->invalidatepage(page, 0);
315 unlock_page(page);
316 continue;
317 }
318
319 ret = __gfs2_jdata_writepage(page, wbc);
320
321 if (ret || (--(wbc->nr_to_write) <= 0))
322 ret = 1;
323 if (wbc->nonblocking && bdi_write_congested(bdi)) {
324 wbc->encountered_congestion = 1;
325 ret = 1;
326 }
327
328 }
329 gfs2_trans_end(sdp);
330 return ret;
331}
332
333/**
334 * gfs2_write_cache_jdata - Like write_cache_pages but different
335 * @mapping: The mapping to write
336 * @wbc: The writeback control
337 * @writepage: The writepage function to call
338 * @data: The data to pass to writepage
339 *
340 * The reason that we use our own function here is that we need to
341 * start transactions before we grab page locks. This allows us
342 * to get the ordering right.
343 */
344
345static int gfs2_write_cache_jdata(struct address_space *mapping,
346 struct writeback_control *wbc)
347{
348 struct backing_dev_info *bdi = mapping->backing_dev_info;
349 int ret = 0;
350 int done = 0;
351 struct pagevec pvec;
352 int nr_pages;
353 pgoff_t index;
354 pgoff_t end;
355 int scanned = 0;
356 int range_whole = 0;
357
358 if (wbc->nonblocking && bdi_write_congested(bdi)) {
359 wbc->encountered_congestion = 1;
360 return 0;
361 }
362
363 pagevec_init(&pvec, 0);
364 if (wbc->range_cyclic) {
365 index = mapping->writeback_index; /* Start from prev offset */
366 end = -1;
367 } else {
368 index = wbc->range_start >> PAGE_CACHE_SHIFT;
369 end = wbc->range_end >> PAGE_CACHE_SHIFT;
370 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
371 range_whole = 1;
372 scanned = 1;
373 }
374
375retry:
376 while (!done && (index <= end) &&
377 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
378 PAGECACHE_TAG_DIRTY,
379 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
380 scanned = 1;
381 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
382 if (ret)
383 done = 1;
384 if (ret > 0)
385 ret = 0;
386
387 pagevec_release(&pvec);
388 cond_resched();
389 }
390
391 if (!scanned && !done) {
392 /*
393 * We hit the last page and there is more work to be done: wrap
394 * back to the start of the file
395 */
396 scanned = 1;
397 index = 0;
398 goto retry;
399 }
400
401 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
402 mapping->writeback_index = index;
403 return ret;
404}
405
406
407/**
408 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
409 * @mapping: The mapping to write
410 * @wbc: The writeback control
411 *
412 */
413
414static int gfs2_jdata_writepages(struct address_space *mapping,
415 struct writeback_control *wbc)
416{
417 struct gfs2_inode *ip = GFS2_I(mapping->host);
418 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
419 int ret;
420
421 ret = gfs2_write_cache_jdata(mapping, wbc);
422 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
423 gfs2_log_flush(sdp, ip->i_gl);
424 ret = gfs2_write_cache_jdata(mapping, wbc);
425 }
426 return ret;
427}
428
429/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000430 * stuffed_readpage - Fill in a Linux page with stuffed file data
431 * @ip: the inode
432 * @page: the page
433 *
434 * Returns: errno
435 */
436
437static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
438{
439 struct buffer_head *dibh;
440 void *kaddr;
441 int error;
442
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100443 /*
444 * Due to the order of unstuffing files and ->nopage(), we can be
445 * asked for a zero page in the case of a stuffed file being extended,
446 * so we need to supply one here. It doesn't happen often.
447 */
448 if (unlikely(page->index)) {
Steven Whitehouse28405012007-06-18 16:31:42 +0100449 zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100450 return 0;
451 }
Steven Whitehousefd88de562006-05-05 16:59:11 -0400452
David Teiglandb3b94fa2006-01-16 16:50:04 +0000453 error = gfs2_meta_inode_buffer(ip, &dibh);
454 if (error)
455 return error;
456
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000457 kaddr = kmap_atomic(page, KM_USER0);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400458 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
David Teiglandb3b94fa2006-01-16 16:50:04 +0000459 ip->i_di.di_size);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400460 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
Russell Cattelanc312c4f2006-10-12 09:23:41 -0400461 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100462 flush_dcache_page(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000463 brelse(dibh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000464 SetPageUptodate(page);
465
466 return 0;
467}
468
David Teiglandb3b94fa2006-01-16 16:50:04 +0000469
470/**
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100471 * __gfs2_readpage - readpage
472 * @file: The file to read a page for
David Teiglandb3b94fa2006-01-16 16:50:04 +0000473 * @page: The page to read
474 *
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100475 * This is the core of gfs2's readpage. Its used by the internal file
476 * reading code as in that case we already hold the glock. Also its
477 * called by gfs2_readpage() once the required lock has been granted.
478 *
479 */
480
481static int __gfs2_readpage(void *file, struct page *page)
482{
483 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
484 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
485 int error;
486
487 if (gfs2_is_stuffed(ip)) {
488 error = stuffed_readpage(ip, page);
489 unlock_page(page);
490 } else {
Bob Petersone9e1ef22007-12-10 14:13:27 -0600491 error = mpage_readpage(page, gfs2_block_map);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100492 }
493
494 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
495 return -EIO;
496
497 return error;
498}
499
500/**
501 * gfs2_readpage - read a page of a file
502 * @file: The file to read
503 * @page: The page of the file
504 *
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100505 * This deals with the locking required. We use a trylock in order to
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100506 * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE
507 * in the event that we are unable to get the lock.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000508 */
509
510static int gfs2_readpage(struct file *file, struct page *page)
511{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400512 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000513 struct gfs2_holder gh;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000514 int error;
515
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100516 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
517 error = gfs2_glock_nq_atime(&gh);
518 if (unlikely(error)) {
Steven Whitehousefd88de562006-05-05 16:59:11 -0400519 unlock_page(page);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100520 goto out;
Steven Whitehouse61a30dc2006-02-15 10:15:18 +0000521 }
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100522 error = __gfs2_readpage(file, page);
523 gfs2_glock_dq(&gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000524out:
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100525 gfs2_holder_uninit(&gh);
Josef Whitera13cbe32007-02-23 12:49:51 -0500526 if (error == GLR_TRYFAILED) {
Josef Whitera13cbe32007-02-23 12:49:51 -0500527 yield();
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100528 return AOP_TRUNCATED_PAGE;
Josef Whitera13cbe32007-02-23 12:49:51 -0500529 }
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100530 return error;
531}
532
533/**
534 * gfs2_internal_read - read an internal file
535 * @ip: The gfs2 inode
536 * @ra_state: The readahead state (or NULL for no readahead)
537 * @buf: The buffer to fill
538 * @pos: The file position
539 * @size: The amount to read
540 *
541 */
542
543int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
544 char *buf, loff_t *pos, unsigned size)
545{
546 struct address_space *mapping = ip->i_inode.i_mapping;
547 unsigned long index = *pos / PAGE_CACHE_SIZE;
548 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
549 unsigned copied = 0;
550 unsigned amt;
551 struct page *page;
552 void *p;
553
554 do {
555 amt = size - copied;
556 if (offset + size > PAGE_CACHE_SIZE)
557 amt = PAGE_CACHE_SIZE - offset;
558 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
559 if (IS_ERR(page))
560 return PTR_ERR(page);
561 p = kmap_atomic(page, KM_USER0);
562 memcpy(buf + copied, p + offset, amt);
563 kunmap_atomic(p, KM_USER0);
564 mark_page_accessed(page);
565 page_cache_release(page);
566 copied += amt;
567 index++;
568 offset = 0;
569 } while(copied < size);
570 (*pos) += size;
571 return size;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400572}
573
Steven Whitehousefd88de562006-05-05 16:59:11 -0400574/**
575 * gfs2_readpages - Read a bunch of pages at once
576 *
577 * Some notes:
578 * 1. This is only for readahead, so we can simply ignore any things
579 * which are slightly inconvenient (such as locking conflicts between
580 * the page lock and the glock) and return having done no I/O. Its
581 * obviously not something we'd want to do on too regular a basis.
582 * Any I/O we ignore at this time will be done via readpage later.
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500583 * 2. We don't handle stuffed files here we let readpage do the honours.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400584 * 3. mpage_readpages() does most of the heavy lifting in the common case.
Bob Petersone9e1ef22007-12-10 14:13:27 -0600585 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400586 */
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100587
Steven Whitehousefd88de562006-05-05 16:59:11 -0400588static int gfs2_readpages(struct file *file, struct address_space *mapping,
589 struct list_head *pages, unsigned nr_pages)
590{
591 struct inode *inode = mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400592 struct gfs2_inode *ip = GFS2_I(inode);
593 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400594 struct gfs2_holder gh;
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100595 int ret;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400596
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100597 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100598 ret = gfs2_glock_nq_atime(&gh);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100599 if (unlikely(ret))
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100600 goto out_uninit;
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500601 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600602 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100603 gfs2_glock_dq(&gh);
604out_uninit:
605 gfs2_holder_uninit(&gh);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400606 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
607 ret = -EIO;
608 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000609}
610
611/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700612 * gfs2_write_begin - Begin to write to a file
David Teiglandb3b94fa2006-01-16 16:50:04 +0000613 * @file: The file to write to
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700614 * @mapping: The mapping in which to write
615 * @pos: The file offset at which to start writing
616 * @len: Length of the write
617 * @flags: Various flags
618 * @pagep: Pointer to return the page
619 * @fsdata: Pointer to return fs data (unused by GFS2)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000620 *
621 * Returns: errno
622 */
623
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700624static int gfs2_write_begin(struct file *file, struct address_space *mapping,
625 loff_t pos, unsigned len, unsigned flags,
626 struct page **pagep, void **fsdata)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000627{
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700628 struct gfs2_inode *ip = GFS2_I(mapping->host);
629 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000630 unsigned int data_blocks, ind_blocks, rblocks;
631 int alloc_required;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000632 int error = 0;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000633 struct gfs2_alloc *al;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700634 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
635 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
636 unsigned to = from + len;
637 struct page *page;
Russell Cattelan52ae7b72006-10-09 12:11:54 -0500638
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700639 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
Steven Whitehousedcd24792006-11-16 11:08:16 -0500640 error = gfs2_glock_nq_atime(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700641 if (unlikely(error))
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000642 goto out_uninit;
643
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700644 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700645 error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
646 if (error)
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100647 goto out_unlock;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000648
649 if (alloc_required) {
650 al = gfs2_alloc_get(ip);
651
652 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
653 if (error)
654 goto out_alloc_put;
655
Steven Whitehouse2933f922006-11-01 13:23:29 -0500656 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000657 if (error)
658 goto out_qunlock;
659
660 al->al_requested = data_blocks + ind_blocks;
661 error = gfs2_inplace_reserve(ip);
662 if (error)
663 goto out_qunlock;
664 }
665
666 rblocks = RES_DINODE + ind_blocks;
667 if (gfs2_is_jdata(ip))
668 rblocks += data_blocks ? data_blocks : 1;
669 if (ind_blocks || data_blocks)
670 rblocks += RES_STATFS + RES_QUOTA;
671
Steven Whitehouse16615be2007-09-17 10:59:52 +0100672 error = gfs2_trans_begin(sdp, rblocks,
673 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000674 if (error)
Steven Whitehousea867bb22007-07-17 10:29:02 +0100675 goto out_trans_fail;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000676
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100677 error = -ENOMEM;
678 page = __grab_cache_page(mapping, index);
679 *pagep = page;
680 if (unlikely(!page))
681 goto out_endtrans;
682
David Teiglandb3b94fa2006-01-16 16:50:04 +0000683 if (gfs2_is_stuffed(ip)) {
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100684 error = 0;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700685 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
Steven Whitehousef25ef0c2006-07-26 10:51:20 -0400686 error = gfs2_unstuff_dinode(ip, page);
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000687 if (error == 0)
688 goto prepare_write;
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100689 } else if (!PageUptodate(page)) {
David Teiglandb3b94fa2006-01-16 16:50:04 +0000690 error = stuffed_readpage(ip, page);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100691 }
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000692 goto out;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000693 }
694
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000695prepare_write:
Bob Petersone9e1ef22007-12-10 14:13:27 -0600696 error = block_prepare_write(page, from, to, gfs2_block_map);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000697out:
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100698 if (error == 0)
699 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000700
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100701 page_cache_release(page);
702 if (pos + len > ip->i_inode.i_size)
703 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
704out_endtrans:
705 gfs2_trans_end(sdp);
706out_trans_fail:
707 if (alloc_required) {
708 gfs2_inplace_release(ip);
709out_qunlock:
710 gfs2_quota_unlock(ip);
711out_alloc_put:
712 gfs2_alloc_put(ip);
713 }
714out_unlock:
715 gfs2_glock_dq(&ip->i_gh);
716out_uninit:
717 gfs2_holder_uninit(&ip->i_gh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000718 return error;
719}
720
721/**
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500722 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
723 * @inode: the rindex inode
724 */
725static void adjust_fs_space(struct inode *inode)
726{
727 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
728 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
729 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
730 u64 fs_total, new_free;
731
732 /* Total up the file system space, according to the latest rindex. */
733 fs_total = gfs2_ri_total(sdp);
734
735 spin_lock(&sdp->sd_statfs_spin);
736 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
737 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
738 else
739 new_free = 0;
740 spin_unlock(&sdp->sd_statfs_spin);
Robert Peterson6c532672007-05-10 16:54:38 -0500741 fs_warn(sdp, "File system extended by %llu blocks.\n",
742 (unsigned long long)new_free);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500743 gfs2_statfs_change(sdp, new_free, new_free, 0);
744}
745
746/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700747 * gfs2_stuffed_write_end - Write end for stuffed files
748 * @inode: The inode
749 * @dibh: The buffer_head containing the on-disk inode
750 * @pos: The file position
751 * @len: The length of the write
752 * @copied: How much was actually copied by the VFS
753 * @page: The page
754 *
755 * This copies the data from the page into the inode block after
756 * the inode data structure itself.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000757 *
758 * Returns: errno
759 */
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700760static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
761 loff_t pos, unsigned len, unsigned copied,
762 struct page *page)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000763{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400764 struct gfs2_inode *ip = GFS2_I(inode);
765 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700766 u64 to = pos + copied;
767 void *kaddr;
768 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
769 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000770
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700771 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
772 kaddr = kmap_atomic(page, KM_USER0);
773 memcpy(buf + pos, kaddr + pos, copied);
774 memset(kaddr + pos + copied, 0, len - copied);
775 flush_dcache_page(page);
776 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000777
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700778 if (!PageUptodate(page))
David Teiglandb3b94fa2006-01-16 16:50:04 +0000779 SetPageUptodate(page);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700780 unlock_page(page);
781 page_cache_release(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000782
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700783 if (inode->i_size < to) {
784 i_size_write(inode, to);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000785 ip->i_di.di_size = inode->i_size;
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400786 di->di_size = cpu_to_be64(inode->i_size);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700787 mark_inode_dirty(inode);
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400788 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000789
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500790 if (inode == sdp->sd_rindex)
791 adjust_fs_space(inode);
792
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000793 brelse(dibh);
794 gfs2_trans_end(sdp);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700795 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000796 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700797 return copied;
798}
David Teiglandb3b94fa2006-01-16 16:50:04 +0000799
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700800/**
801 * gfs2_write_end
802 * @file: The file to write to
803 * @mapping: The address space to write to
804 * @pos: The file position
805 * @len: The length of the data
806 * @copied:
807 * @page: The page that has been written
808 * @fsdata: The fsdata (unused in GFS2)
809 *
810 * The main write_end function for GFS2. We have a separate one for
811 * stuffed files as they are slightly different, otherwise we just
812 * put our locking around the VFS provided functions.
813 *
814 * Returns: errno
815 */
816
817static int gfs2_write_end(struct file *file, struct address_space *mapping,
818 loff_t pos, unsigned len, unsigned copied,
819 struct page *page, void *fsdata)
820{
821 struct inode *inode = page->mapping->host;
822 struct gfs2_inode *ip = GFS2_I(inode);
823 struct gfs2_sbd *sdp = GFS2_SB(inode);
824 struct buffer_head *dibh;
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000825 struct gfs2_alloc *al = ip->i_alloc;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700826 struct gfs2_dinode *di;
827 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
828 unsigned int to = from + len;
829 int ret;
830
831 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == 0);
832
833 ret = gfs2_meta_inode_buffer(ip, &dibh);
834 if (unlikely(ret)) {
835 unlock_page(page);
836 page_cache_release(page);
837 goto failed;
838 }
839
840 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
841
842 if (gfs2_is_stuffed(ip))
843 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
844
Steven Whitehousebf36a712007-10-17 08:35:19 +0100845 if (!gfs2_is_writeback(ip))
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700846 gfs2_page_add_databufs(ip, page, from, to);
847
848 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
849
Steven Whitehouse9656b2c2008-01-08 08:14:30 +0000850 if (likely(ret >= 0) && (inode->i_size > ip->i_di.di_size)) {
851 di = (struct gfs2_dinode *)dibh->b_data;
852 ip->i_di.di_size = inode->i_size;
853 di->di_size = cpu_to_be64(inode->i_size);
854 mark_inode_dirty(inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700855 }
856
857 if (inode == sdp->sd_rindex)
858 adjust_fs_space(inode);
859
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000860 brelse(dibh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000861 gfs2_trans_end(sdp);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700862failed:
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000863 if (al) {
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000864 gfs2_inplace_release(ip);
865 gfs2_quota_unlock(ip);
866 gfs2_alloc_put(ip);
867 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700868 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000869 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700870 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000871}
872
873/**
Robert Peterson8fb68592007-06-12 11:24:36 -0500874 * gfs2_set_page_dirty - Page dirtying function
875 * @page: The page to dirty
876 *
877 * Returns: 1 if it dirtyed the page, or 0 otherwise
878 */
879
880static int gfs2_set_page_dirty(struct page *page)
881{
Steven Whitehouse55610932007-10-17 08:47:38 +0100882 SetPageChecked(page);
Robert Peterson8fb68592007-06-12 11:24:36 -0500883 return __set_page_dirty_buffers(page);
884}
885
886/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000887 * gfs2_bmap - Block map function
888 * @mapping: Address space info
889 * @lblock: The block to map
890 *
891 * Returns: The disk address for the block or 0 on hole or error
892 */
893
894static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
895{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400896 struct gfs2_inode *ip = GFS2_I(mapping->host);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000897 struct gfs2_holder i_gh;
898 sector_t dblock = 0;
899 int error;
900
David Teiglandb3b94fa2006-01-16 16:50:04 +0000901 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
902 if (error)
903 return 0;
904
905 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600906 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000907
908 gfs2_glock_dq_uninit(&i_gh);
909
910 return dblock;
911}
912
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100913static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
914{
915 struct gfs2_bufdata *bd;
916
917 lock_buffer(bh);
918 gfs2_log_lock(sdp);
919 clear_buffer_dirty(bh);
920 bd = bh->b_private;
921 if (bd) {
Steven Whitehouse16615be2007-09-17 10:59:52 +0100922 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
923 list_del_init(&bd->bd_le.le_list);
924 else
925 gfs2_remove_from_journal(bh, current->journal_info, 0);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100926 }
927 bh->b_bdev = NULL;
928 clear_buffer_mapped(bh);
929 clear_buffer_req(bh);
930 clear_buffer_new(bh);
931 gfs2_log_unlock(sdp);
932 unlock_buffer(bh);
933}
934
Steven Whitehouse8628de02006-03-31 16:48:41 -0500935static void gfs2_invalidatepage(struct page *page, unsigned long offset)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000936{
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100937 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
938 struct buffer_head *bh, *head;
939 unsigned long pos = 0;
940
David Teiglandb3b94fa2006-01-16 16:50:04 +0000941 BUG_ON(!PageLocked(page));
Robert Peterson8fb68592007-06-12 11:24:36 -0500942 if (offset == 0)
943 ClearPageChecked(page);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100944 if (!page_has_buffers(page))
945 goto out;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000946
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100947 bh = head = page_buffers(page);
948 do {
949 if (offset <= pos)
950 gfs2_discard(sdp, bh);
951 pos += bh->b_size;
952 bh = bh->b_this_page;
953 } while (bh != head);
954out:
955 if (offset == 0)
956 try_to_release_page(page, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000957}
958
Steven Whitehousec7b33832006-12-14 18:24:26 +0000959/**
960 * gfs2_ok_for_dio - check that dio is valid on this file
961 * @ip: The inode
962 * @rw: READ or WRITE
963 * @offset: The offset at which we are reading or writing
964 *
965 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
966 * 1 (to accept the i/o request)
967 */
968static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
969{
970 /*
971 * Should we return an error here? I can't see that O_DIRECT for
Steven Whitehouse55610932007-10-17 08:47:38 +0100972 * a stuffed file makes any sense. For now we'll silently fall
973 * back to buffered I/O
Steven Whitehousec7b33832006-12-14 18:24:26 +0000974 */
Steven Whitehousec7b33832006-12-14 18:24:26 +0000975 if (gfs2_is_stuffed(ip))
976 return 0;
977
978 if (offset > i_size_read(&ip->i_inode))
979 return 0;
980 return 1;
981}
982
983
984
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -0400985static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
986 const struct iovec *iov, loff_t offset,
987 unsigned long nr_segs)
Steven Whitehoused1665e42006-02-14 11:54:42 +0000988{
989 struct file *file = iocb->ki_filp;
990 struct inode *inode = file->f_mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400991 struct gfs2_inode *ip = GFS2_I(inode);
Steven Whitehoused1665e42006-02-14 11:54:42 +0000992 struct gfs2_holder gh;
993 int rv;
994
995 /*
Steven Whitehousec7b33832006-12-14 18:24:26 +0000996 * Deferred lock, even if its a write, since we do no allocation
997 * on this path. All we need change is atime, and this lock mode
998 * ensures that other nodes have flushed their buffered read caches
999 * (i.e. their page cache entries for this inode). We do not,
1000 * unfortunately have the option of only flushing a range like
1001 * the VFS does.
Steven Whitehoused1665e42006-02-14 11:54:42 +00001002 */
Steven Whitehousec7b33832006-12-14 18:24:26 +00001003 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
Steven Whitehousedcd24792006-11-16 11:08:16 -05001004 rv = gfs2_glock_nq_atime(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001005 if (rv)
Steven Whitehousec7b33832006-12-14 18:24:26 +00001006 return rv;
1007 rv = gfs2_ok_for_dio(ip, rw, offset);
1008 if (rv != 1)
1009 goto out; /* dio not valid, fall back to buffered i/o */
Steven Whitehoused1665e42006-02-14 11:54:42 +00001010
Steven Whitehousec7b33832006-12-14 18:24:26 +00001011 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
1012 iov, offset, nr_segs,
1013 gfs2_get_block_direct, NULL);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001014out:
1015 gfs2_glock_dq_m(1, &gh);
1016 gfs2_holder_uninit(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001017 return rv;
1018}
1019
1020/**
Steven Whitehouse623d9352006-08-31 12:14:44 -04001021 * gfs2_releasepage - free the metadata associated with a page
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001022 * @page: the page that's being released
1023 * @gfp_mask: passed from Linux VFS, ignored by us
1024 *
1025 * Call try_to_free_buffers() if the buffers in this page can be
1026 * released.
1027 *
1028 * Returns: 0
1029 */
1030
1031int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1032{
1033 struct inode *aspace = page->mapping->host;
1034 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
1035 struct buffer_head *bh, *head;
1036 struct gfs2_bufdata *bd;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001037
1038 if (!page_has_buffers(page))
Steven Whitehouse891ba6d2007-09-20 15:26:33 +01001039 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001040
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001041 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001042 head = bh = page_buffers(page);
1043 do {
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001044 if (atomic_read(&bh->b_count))
1045 goto cannot_release;
1046 bd = bh->b_private;
1047 if (bd && bd->bd_ail)
1048 goto cannot_release;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001049 gfs2_assert_warn(sdp, !buffer_pinned(bh));
Steven Whitehouse623d9352006-08-31 12:14:44 -04001050 gfs2_assert_warn(sdp, !buffer_dirty(bh));
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001051 bh = bh->b_this_page;
1052 } while(bh != head);
1053 gfs2_log_unlock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001054
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001055 head = bh = page_buffers(page);
1056 do {
Steven Whitehouse623d9352006-08-31 12:14:44 -04001057 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001058 bd = bh->b_private;
1059 if (bd) {
1060 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1061 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
Steven Whitehoused7b616e2007-09-02 10:48:13 +01001062 if (!list_empty(&bd->bd_le.le_list)) {
1063 if (!buffer_pinned(bh))
1064 list_del_init(&bd->bd_le.le_list);
1065 else
1066 bd = NULL;
1067 }
1068 if (bd)
1069 bd->bd_bh = NULL;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001070 bh->b_private = NULL;
1071 }
Steven Whitehouse623d9352006-08-31 12:14:44 -04001072 gfs2_log_unlock(sdp);
1073 if (bd)
1074 kmem_cache_free(gfs2_bufdata_cachep, bd);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001075
1076 bh = bh->b_this_page;
Steven Whitehouse166afcc2006-08-24 15:59:40 -04001077 } while (bh != head);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001078
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001079 return try_to_free_buffers(page);
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001080cannot_release:
1081 gfs2_log_unlock(sdp);
1082 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001083}
1084
Steven Whitehouse55610932007-10-17 08:47:38 +01001085static const struct address_space_operations gfs2_writeback_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001086 .writepage = gfs2_writeback_writepage,
Steven Whitehouse55610932007-10-17 08:47:38 +01001087 .writepages = gfs2_writeback_writepages,
1088 .readpage = gfs2_readpage,
1089 .readpages = gfs2_readpages,
1090 .sync_page = block_sync_page,
1091 .write_begin = gfs2_write_begin,
1092 .write_end = gfs2_write_end,
1093 .bmap = gfs2_bmap,
1094 .invalidatepage = gfs2_invalidatepage,
1095 .releasepage = gfs2_releasepage,
1096 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001097 .migratepage = buffer_migrate_page,
Steven Whitehouse55610932007-10-17 08:47:38 +01001098};
1099
1100static const struct address_space_operations gfs2_ordered_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001101 .writepage = gfs2_ordered_writepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001102 .readpage = gfs2_readpage,
Steven Whitehousefd88de562006-05-05 16:59:11 -04001103 .readpages = gfs2_readpages,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001104 .sync_page = block_sync_page,
Steven Whitehouse7765ec22007-10-16 01:25:07 -07001105 .write_begin = gfs2_write_begin,
1106 .write_end = gfs2_write_end,
Robert Peterson8fb68592007-06-12 11:24:36 -05001107 .set_page_dirty = gfs2_set_page_dirty,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001108 .bmap = gfs2_bmap,
1109 .invalidatepage = gfs2_invalidatepage,
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001110 .releasepage = gfs2_releasepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001111 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001112 .migratepage = buffer_migrate_page,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001113};
1114
Steven Whitehouse55610932007-10-17 08:47:38 +01001115static const struct address_space_operations gfs2_jdata_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001116 .writepage = gfs2_jdata_writepage,
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +01001117 .writepages = gfs2_jdata_writepages,
Steven Whitehouse55610932007-10-17 08:47:38 +01001118 .readpage = gfs2_readpage,
1119 .readpages = gfs2_readpages,
1120 .sync_page = block_sync_page,
1121 .write_begin = gfs2_write_begin,
1122 .write_end = gfs2_write_end,
1123 .set_page_dirty = gfs2_set_page_dirty,
1124 .bmap = gfs2_bmap,
1125 .invalidatepage = gfs2_invalidatepage,
1126 .releasepage = gfs2_releasepage,
1127};
1128
1129void gfs2_set_aops(struct inode *inode)
1130{
1131 struct gfs2_inode *ip = GFS2_I(inode);
1132
1133 if (gfs2_is_writeback(ip))
1134 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1135 else if (gfs2_is_ordered(ip))
1136 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1137 else if (gfs2_is_jdata(ip))
1138 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1139 else
1140 BUG();
1141}
1142