blob: 194fe16d8418a332a274a74769b15277ff2d6858 [file] [log] [blame]
David Teiglandb3b94fa2006-01-16 16:50:04 +00001/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
Bob Peterson7eabb772008-01-28 11:24:35 -06003 * Copyright (C) 2004-2008 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>
Steven Whitehouse47e83b52007-10-18 11:15:50 +010022#include <linux/backing-dev.h>
David Teiglandb3b94fa2006-01-16 16:50:04 +000023
24#include "gfs2.h"
Steven Whitehouse5c676f62006-02-27 17:23:27 -050025#include "incore.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000026#include "bmap.h"
27#include "glock.h"
28#include "inode.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000029#include "log.h"
30#include "meta_io.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000031#include "quota.h"
32#include "trans.h"
Steven Whitehouse18ec7d52006-02-08 11:50:51 +000033#include "rgrp.h"
Robert Petersoncd81a4b2007-05-14 12:42:18 -050034#include "super.h"
Steven Whitehouse5c676f62006-02-27 17:23:27 -050035#include "util.h"
Steven Whitehouse4340fe62006-07-11 09:46:33 -040036#include "glops.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000037
Steven Whitehouseba7f7292006-07-26 11:27:10 -040038
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;
Benjamin Marzinskiddf4b422007-06-01 14:21:38 -050052 if (gfs2_is_jdata(ip))
53 set_buffer_uptodate(bh);
Steven Whitehouseba7f7292006-07-26 11:27:10 -040054 gfs2_trans_add_bh(ip->i_gl, bh, 0);
55 }
56}
57
David Teiglandb3b94fa2006-01-16 16:50:04 +000058/**
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040059 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
David Teiglandb3b94fa2006-01-16 16:50:04 +000060 * @inode: The inode
61 * @lblock: The block number to look up
62 * @bh_result: The buffer head to return the result in
63 * @create: Non-zero if we may add block to the file
64 *
65 * Returns: errno
66 */
67
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040068static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
69 struct buffer_head *bh_result, int create)
David Teiglandb3b94fa2006-01-16 16:50:04 +000070{
David Teiglandb3b94fa2006-01-16 16:50:04 +000071 int error;
72
Bob Petersone9e1ef22007-12-10 14:13:27 -060073 error = gfs2_block_map(inode, lblock, bh_result, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +000074 if (error)
75 return error;
Wendy Chengde986e82007-09-18 09:19:13 -040076 if (!buffer_mapped(bh_result))
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040077 return -EIO;
Steven Whitehouse623d9352006-08-31 12:14:44 -040078 return 0;
79}
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040080
81static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
82 struct buffer_head *bh_result, int create)
83{
Bob Petersone9e1ef22007-12-10 14:13:27 -060084 return gfs2_block_map(inode, lblock, bh_result, 0);
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040085}
86
David Teiglandb3b94fa2006-01-16 16:50:04 +000087/**
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +010088 * gfs2_writepage_common - Common bits of writepage
89 * @page: The page to be written
90 * @wbc: The writeback control
David Teiglandb3b94fa2006-01-16 16:50:04 +000091 *
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +010092 * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
David Teiglandb3b94fa2006-01-16 16:50:04 +000093 */
94
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +010095static int gfs2_writepage_common(struct page *page,
96 struct writeback_control *wbc)
David Teiglandb3b94fa2006-01-16 16:50:04 +000097{
Steven Whitehouse18ec7d52006-02-08 11:50:51 +000098 struct inode *inode = page->mapping->host;
Steven Whitehousef4387142006-08-08 13:23:19 -040099 struct gfs2_inode *ip = GFS2_I(inode);
100 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000101 loff_t i_size = i_size_read(inode);
102 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
103 unsigned offset;
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100104
105 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
106 goto out;
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100107 if (current->journal_info)
108 goto redirty;
109 /* Is the page fully outside i_size? (truncate in progress) */
110 offset = i_size & (PAGE_CACHE_SIZE-1);
111 if (page->index > end_index || (page->index == end_index && !offset)) {
112 page->mapping->a_ops->invalidatepage(page, 0);
113 goto out;
114 }
115 return 1;
116redirty:
117 redirty_page_for_writepage(wbc, page);
118out:
119 unlock_page(page);
120 return 0;
121}
122
123/**
124 * gfs2_writeback_writepage - Write page for writeback mappings
125 * @page: The page
126 * @wbc: The writeback control
127 *
128 */
129
130static int gfs2_writeback_writepage(struct page *page,
131 struct writeback_control *wbc)
132{
133 int ret;
134
135 ret = gfs2_writepage_common(page, wbc);
136 if (ret <= 0)
137 return ret;
138
Steven Whitehouse30116ff2010-06-14 09:58:41 +0100139 return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100140}
141
142/**
143 * gfs2_ordered_writepage - Write page for ordered data files
144 * @page: The page to write
145 * @wbc: The writeback control
146 *
147 */
148
149static int gfs2_ordered_writepage(struct page *page,
150 struct writeback_control *wbc)
151{
152 struct inode *inode = page->mapping->host;
153 struct gfs2_inode *ip = GFS2_I(inode);
154 int ret;
155
156 ret = gfs2_writepage_common(page, wbc);
157 if (ret <= 0)
158 return ret;
159
160 if (!page_has_buffers(page)) {
161 create_empty_buffers(page, inode->i_sb->s_blocksize,
162 (1 << BH_Dirty)|(1 << BH_Uptodate));
163 }
164 gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
165 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
166}
167
168/**
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100169 * __gfs2_jdata_writepage - The core of jdata writepage
170 * @page: The page to write
171 * @wbc: The writeback control
172 *
173 * This is shared between writepage and writepages and implements the
174 * core of the writepage operation. If a transaction is required then
175 * PageChecked will have been set and the transaction will have
176 * already been started before this is called.
177 */
178
179static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
180{
181 struct inode *inode = page->mapping->host;
182 struct gfs2_inode *ip = GFS2_I(inode);
183 struct gfs2_sbd *sdp = GFS2_SB(inode);
184
185 if (PageChecked(page)) {
186 ClearPageChecked(page);
187 if (!page_has_buffers(page)) {
188 create_empty_buffers(page, inode->i_sb->s_blocksize,
189 (1 << BH_Dirty)|(1 << BH_Uptodate));
190 }
191 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
192 }
193 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
194}
195
196/**
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100197 * gfs2_jdata_writepage - Write complete page
198 * @page: Page to write
199 *
200 * Returns: errno
201 *
202 */
203
204static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
205{
206 struct inode *inode = page->mapping->host;
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100207 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100208 int ret;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000209 int done_trans = 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000210
Steven Whitehousebf36a712007-10-17 08:35:19 +0100211 if (PageChecked(page)) {
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100212 if (wbc->sync_mode != WB_SYNC_ALL)
213 goto out_ignore;
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100214 ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
215 if (ret)
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000216 goto out_ignore;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000217 done_trans = 1;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000218 }
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100219 ret = gfs2_writepage_common(page, wbc);
220 if (ret > 0)
221 ret = __gfs2_jdata_writepage(page, wbc);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000222 if (done_trans)
223 gfs2_trans_end(sdp);
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100224 return ret;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000225
226out_ignore:
227 redirty_page_for_writepage(wbc, page);
228 unlock_page(page);
229 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000230}
231
232/**
Steven Whitehouse55610932007-10-17 08:47:38 +0100233 * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000234 * @mapping: The mapping to write
235 * @wbc: Write-back control
236 *
Steven Whitehouse55610932007-10-17 08:47:38 +0100237 * For the data=writeback case we can already ignore buffer heads
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000238 * and write whole extents at once. This is a big reduction in the
239 * number of I/O requests we send and the bmap calls we make in this case.
240 */
Steven Whitehouse55610932007-10-17 08:47:38 +0100241static int gfs2_writeback_writepages(struct address_space *mapping,
242 struct writeback_control *wbc)
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000243{
Steven Whitehouse55610932007-10-17 08:47:38 +0100244 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000245}
246
247/**
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100248 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
249 * @mapping: The mapping
250 * @wbc: The writeback control
251 * @writepage: The writepage function to call for each page
252 * @pvec: The vector of pages
253 * @nr_pages: The number of pages to write
254 *
255 * Returns: non-zero if loop should terminate, zero otherwise
256 */
257
258static int gfs2_write_jdata_pagevec(struct address_space *mapping,
259 struct writeback_control *wbc,
260 struct pagevec *pvec,
261 int nr_pages, pgoff_t end)
262{
263 struct inode *inode = mapping->host;
264 struct gfs2_sbd *sdp = GFS2_SB(inode);
265 loff_t i_size = i_size_read(inode);
266 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
267 unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
268 unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100269 int i;
270 int ret;
271
Abhijith Das20b95bf2008-03-06 17:43:52 -0600272 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100273 if (ret < 0)
274 return ret;
275
276 for(i = 0; i < nr_pages; i++) {
277 struct page *page = pvec->pages[i];
278
279 lock_page(page);
280
281 if (unlikely(page->mapping != mapping)) {
282 unlock_page(page);
283 continue;
284 }
285
286 if (!wbc->range_cyclic && page->index > end) {
287 ret = 1;
288 unlock_page(page);
289 continue;
290 }
291
292 if (wbc->sync_mode != WB_SYNC_NONE)
293 wait_on_page_writeback(page);
294
295 if (PageWriteback(page) ||
296 !clear_page_dirty_for_io(page)) {
297 unlock_page(page);
298 continue;
299 }
300
301 /* Is the page fully outside i_size? (truncate in progress) */
302 if (page->index > end_index || (page->index == end_index && !offset)) {
303 page->mapping->a_ops->invalidatepage(page, 0);
304 unlock_page(page);
305 continue;
306 }
307
308 ret = __gfs2_jdata_writepage(page, wbc);
309
310 if (ret || (--(wbc->nr_to_write) <= 0))
311 ret = 1;
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100312 }
313 gfs2_trans_end(sdp);
314 return ret;
315}
316
317/**
318 * gfs2_write_cache_jdata - Like write_cache_pages but different
319 * @mapping: The mapping to write
320 * @wbc: The writeback control
321 * @writepage: The writepage function to call
322 * @data: The data to pass to writepage
323 *
324 * The reason that we use our own function here is that we need to
325 * start transactions before we grab page locks. This allows us
326 * to get the ordering right.
327 */
328
329static int gfs2_write_cache_jdata(struct address_space *mapping,
330 struct writeback_control *wbc)
331{
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100332 int ret = 0;
333 int done = 0;
334 struct pagevec pvec;
335 int nr_pages;
336 pgoff_t index;
337 pgoff_t end;
338 int scanned = 0;
339 int range_whole = 0;
340
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100341 pagevec_init(&pvec, 0);
342 if (wbc->range_cyclic) {
343 index = mapping->writeback_index; /* Start from prev offset */
344 end = -1;
345 } else {
346 index = wbc->range_start >> PAGE_CACHE_SHIFT;
347 end = wbc->range_end >> PAGE_CACHE_SHIFT;
348 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
349 range_whole = 1;
350 scanned = 1;
351 }
352
353retry:
354 while (!done && (index <= end) &&
355 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
356 PAGECACHE_TAG_DIRTY,
357 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
358 scanned = 1;
359 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
360 if (ret)
361 done = 1;
362 if (ret > 0)
363 ret = 0;
364
365 pagevec_release(&pvec);
366 cond_resched();
367 }
368
369 if (!scanned && !done) {
370 /*
371 * We hit the last page and there is more work to be done: wrap
372 * back to the start of the file
373 */
374 scanned = 1;
375 index = 0;
376 goto retry;
377 }
378
379 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
380 mapping->writeback_index = index;
381 return ret;
382}
383
384
385/**
386 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
387 * @mapping: The mapping to write
388 * @wbc: The writeback control
389 *
390 */
391
392static int gfs2_jdata_writepages(struct address_space *mapping,
393 struct writeback_control *wbc)
394{
395 struct gfs2_inode *ip = GFS2_I(mapping->host);
396 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
397 int ret;
398
399 ret = gfs2_write_cache_jdata(mapping, wbc);
400 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
401 gfs2_log_flush(sdp, ip->i_gl);
402 ret = gfs2_write_cache_jdata(mapping, wbc);
403 }
404 return ret;
405}
406
407/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000408 * stuffed_readpage - Fill in a Linux page with stuffed file data
409 * @ip: the inode
410 * @page: the page
411 *
412 * Returns: errno
413 */
414
415static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
416{
417 struct buffer_head *dibh;
Steven Whitehouse602c89d2010-03-25 14:32:43 +0000418 u64 dsize = i_size_read(&ip->i_inode);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000419 void *kaddr;
420 int error;
421
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100422 /*
Nick Piggin3c18ddd2008-04-28 02:12:10 -0700423 * Due to the order of unstuffing files and ->fault(), we can be
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100424 * asked for a zero page in the case of a stuffed file being extended,
425 * so we need to supply one here. It doesn't happen often.
426 */
427 if (unlikely(page->index)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800428 zero_user(page, 0, PAGE_CACHE_SIZE);
Abhijith Das0a7ab792009-01-07 16:03:37 -0600429 SetPageUptodate(page);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100430 return 0;
431 }
Steven Whitehousefd88de562006-05-05 16:59:11 -0400432
David Teiglandb3b94fa2006-01-16 16:50:04 +0000433 error = gfs2_meta_inode_buffer(ip, &dibh);
434 if (error)
435 return error;
436
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000437 kaddr = kmap_atomic(page, KM_USER0);
Steven Whitehouse602c89d2010-03-25 14:32:43 +0000438 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
439 dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
440 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
441 memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
Russell Cattelanc312c4f2006-10-12 09:23:41 -0400442 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100443 flush_dcache_page(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000444 brelse(dibh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000445 SetPageUptodate(page);
446
447 return 0;
448}
449
David Teiglandb3b94fa2006-01-16 16:50:04 +0000450
451/**
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100452 * __gfs2_readpage - readpage
453 * @file: The file to read a page for
David Teiglandb3b94fa2006-01-16 16:50:04 +0000454 * @page: The page to read
455 *
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100456 * This is the core of gfs2's readpage. Its used by the internal file
457 * reading code as in that case we already hold the glock. Also its
458 * called by gfs2_readpage() once the required lock has been granted.
459 *
460 */
461
462static int __gfs2_readpage(void *file, struct page *page)
463{
464 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
465 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
466 int error;
467
468 if (gfs2_is_stuffed(ip)) {
469 error = stuffed_readpage(ip, page);
470 unlock_page(page);
471 } else {
Bob Petersone9e1ef22007-12-10 14:13:27 -0600472 error = mpage_readpage(page, gfs2_block_map);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100473 }
474
475 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
476 return -EIO;
477
478 return error;
479}
480
481/**
482 * gfs2_readpage - read a page of a file
483 * @file: The file to read
484 * @page: The page of the file
485 *
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100486 * This deals with the locking required. We have to unlock and
487 * relock the page in order to get the locking in the right
488 * order.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000489 */
490
491static int gfs2_readpage(struct file *file, struct page *page)
492{
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100493 struct address_space *mapping = page->mapping;
494 struct gfs2_inode *ip = GFS2_I(mapping->host);
Steven Whitehouse6802e342008-05-21 17:03:22 +0100495 struct gfs2_holder gh;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000496 int error;
497
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100498 unlock_page(page);
Steven Whitehouse719ee342008-09-18 13:53:59 +0100499 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
500 error = gfs2_glock_nq(&gh);
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100501 if (unlikely(error))
Steven Whitehouse6802e342008-05-21 17:03:22 +0100502 goto out;
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100503 error = AOP_TRUNCATED_PAGE;
504 lock_page(page);
505 if (page->mapping == mapping && !PageUptodate(page))
506 error = __gfs2_readpage(file, page);
507 else
508 unlock_page(page);
Steven Whitehouse6802e342008-05-21 17:03:22 +0100509 gfs2_glock_dq(&gh);
Steven Whitehouse7afd88d2008-02-22 16:07:18 +0000510out:
Steven Whitehouse6802e342008-05-21 17:03:22 +0100511 gfs2_holder_uninit(&gh);
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100512 if (error && error != AOP_TRUNCATED_PAGE)
513 lock_page(page);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100514 return error;
515}
516
517/**
518 * gfs2_internal_read - read an internal file
519 * @ip: The gfs2 inode
520 * @ra_state: The readahead state (or NULL for no readahead)
521 * @buf: The buffer to fill
522 * @pos: The file position
523 * @size: The amount to read
524 *
525 */
526
527int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
528 char *buf, loff_t *pos, unsigned size)
529{
530 struct address_space *mapping = ip->i_inode.i_mapping;
531 unsigned long index = *pos / PAGE_CACHE_SIZE;
532 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
533 unsigned copied = 0;
534 unsigned amt;
535 struct page *page;
536 void *p;
537
538 do {
539 amt = size - copied;
540 if (offset + size > PAGE_CACHE_SIZE)
541 amt = PAGE_CACHE_SIZE - offset;
542 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
543 if (IS_ERR(page))
544 return PTR_ERR(page);
545 p = kmap_atomic(page, KM_USER0);
546 memcpy(buf + copied, p + offset, amt);
547 kunmap_atomic(p, KM_USER0);
548 mark_page_accessed(page);
549 page_cache_release(page);
550 copied += amt;
551 index++;
552 offset = 0;
553 } while(copied < size);
554 (*pos) += size;
555 return size;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400556}
557
Steven Whitehousefd88de562006-05-05 16:59:11 -0400558/**
559 * gfs2_readpages - Read a bunch of pages at once
560 *
561 * Some notes:
562 * 1. This is only for readahead, so we can simply ignore any things
563 * which are slightly inconvenient (such as locking conflicts between
564 * the page lock and the glock) and return having done no I/O. Its
565 * obviously not something we'd want to do on too regular a basis.
566 * Any I/O we ignore at this time will be done via readpage later.
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500567 * 2. We don't handle stuffed files here we let readpage do the honours.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400568 * 3. mpage_readpages() does most of the heavy lifting in the common case.
Bob Petersone9e1ef22007-12-10 14:13:27 -0600569 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400570 */
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100571
Steven Whitehousefd88de562006-05-05 16:59:11 -0400572static int gfs2_readpages(struct file *file, struct address_space *mapping,
573 struct list_head *pages, unsigned nr_pages)
574{
575 struct inode *inode = mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400576 struct gfs2_inode *ip = GFS2_I(inode);
577 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400578 struct gfs2_holder gh;
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100579 int ret;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400580
Steven Whitehouse719ee342008-09-18 13:53:59 +0100581 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
582 ret = gfs2_glock_nq(&gh);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100583 if (unlikely(ret))
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100584 goto out_uninit;
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500585 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600586 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100587 gfs2_glock_dq(&gh);
588out_uninit:
589 gfs2_holder_uninit(&gh);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400590 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
591 ret = -EIO;
592 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000593}
594
595/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700596 * gfs2_write_begin - Begin to write to a file
David Teiglandb3b94fa2006-01-16 16:50:04 +0000597 * @file: The file to write to
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700598 * @mapping: The mapping in which to write
599 * @pos: The file offset at which to start writing
600 * @len: Length of the write
601 * @flags: Various flags
602 * @pagep: Pointer to return the page
603 * @fsdata: Pointer to return fs data (unused by GFS2)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000604 *
605 * Returns: errno
606 */
607
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700608static int gfs2_write_begin(struct file *file, struct address_space *mapping,
609 loff_t pos, unsigned len, unsigned flags,
610 struct page **pagep, void **fsdata)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000611{
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700612 struct gfs2_inode *ip = GFS2_I(mapping->host);
613 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500614 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7ed122e2008-12-10 10:28:10 +0000615 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000616 int alloc_required;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000617 int error = 0;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000618 struct gfs2_alloc *al;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700619 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
620 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
621 unsigned to = from + len;
622 struct page *page;
Russell Cattelan52ae7b72006-10-09 12:11:54 -0500623
Steven Whitehouse719ee342008-09-18 13:53:59 +0100624 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
625 error = gfs2_glock_nq(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700626 if (unlikely(error))
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000627 goto out_uninit;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500628 if (&ip->i_inode == sdp->sd_rindex) {
629 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
630 GL_NOCACHE, &m_ip->i_gh);
631 if (unlikely(error)) {
632 gfs2_glock_dq(&ip->i_gh);
633 goto out_uninit;
634 }
635 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000636
Bob Peterson461cb412010-06-24 19:21:20 -0400637 alloc_required = gfs2_write_alloc_required(ip, pos, len);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000638
Steven Whitehouse7ed122e2008-12-10 10:28:10 +0000639 if (alloc_required || gfs2_is_jdata(ip))
640 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
641
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000642 if (alloc_required) {
643 al = gfs2_alloc_get(ip);
Cyrill Gorcunov182fe5a2008-03-03 21:54:21 +0300644 if (!al) {
645 error = -ENOMEM;
646 goto out_unlock;
647 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000648
Steven Whitehoused82661d2008-03-10 15:34:50 +0000649 error = gfs2_quota_lock_check(ip);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000650 if (error)
651 goto out_alloc_put;
652
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000653 al->al_requested = data_blocks + ind_blocks;
654 error = gfs2_inplace_reserve(ip);
655 if (error)
656 goto out_qunlock;
657 }
658
659 rblocks = RES_DINODE + ind_blocks;
660 if (gfs2_is_jdata(ip))
661 rblocks += data_blocks ? data_blocks : 1;
662 if (ind_blocks || data_blocks)
663 rblocks += RES_STATFS + RES_QUOTA;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500664 if (&ip->i_inode == sdp->sd_rindex)
665 rblocks += 2 * RES_STATFS;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000666
Steven Whitehouse16615be2007-09-17 10:59:52 +0100667 error = gfs2_trans_begin(sdp, rblocks,
668 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000669 if (error)
Steven Whitehousea867bb22007-07-17 10:29:02 +0100670 goto out_trans_fail;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000671
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100672 error = -ENOMEM;
Steven Whitehousee4fefba2009-01-06 10:08:33 +0000673 flags |= AOP_FLAG_NOFS;
Nick Piggin54566b22009-01-04 12:00:53 -0800674 page = grab_cache_page_write_begin(mapping, index, flags);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100675 *pagep = page;
676 if (unlikely(!page))
677 goto out_endtrans;
678
David Teiglandb3b94fa2006-01-16 16:50:04 +0000679 if (gfs2_is_stuffed(ip)) {
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100680 error = 0;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700681 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
Steven Whitehousef25ef0c2006-07-26 10:51:20 -0400682 error = gfs2_unstuff_dinode(ip, page);
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000683 if (error == 0)
684 goto prepare_write;
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100685 } else if (!PageUptodate(page)) {
David Teiglandb3b94fa2006-01-16 16:50:04 +0000686 error = stuffed_readpage(ip, page);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100687 }
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000688 goto out;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000689 }
690
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000691prepare_write:
Bob Petersone9e1ef22007-12-10 14:13:27 -0600692 error = block_prepare_write(page, from, to, gfs2_block_map);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000693out:
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100694 if (error == 0)
695 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000696
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100697 page_cache_release(page);
npiggin@suse.de15c6fd92010-05-27 01:05:34 +1000698
699 /*
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200700 * XXX(truncate): the call below should probably be replaced with
npiggin@suse.de15c6fd92010-05-27 01:05:34 +1000701 * a call to the gfs2-specific truncate blocks helper to actually
702 * release disk blocks..
703 */
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100704 if (pos + len > ip->i_inode.i_size)
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200705 truncate_setsize(&ip->i_inode, ip->i_inode.i_size);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100706out_endtrans:
707 gfs2_trans_end(sdp);
708out_trans_fail:
709 if (alloc_required) {
710 gfs2_inplace_release(ip);
711out_qunlock:
712 gfs2_quota_unlock(ip);
713out_alloc_put:
714 gfs2_alloc_put(ip);
715 }
716out_unlock:
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500717 if (&ip->i_inode == sdp->sd_rindex) {
718 gfs2_glock_dq(&m_ip->i_gh);
719 gfs2_holder_uninit(&m_ip->i_gh);
720 }
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100721 gfs2_glock_dq(&ip->i_gh);
722out_uninit:
723 gfs2_holder_uninit(&ip->i_gh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000724 return error;
725}
726
727/**
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500728 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
729 * @inode: the rindex inode
730 */
731static void adjust_fs_space(struct inode *inode)
732{
733 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500734 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
735 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500736 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
737 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500738 struct buffer_head *m_bh, *l_bh;
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500739 u64 fs_total, new_free;
740
741 /* Total up the file system space, according to the latest rindex. */
742 fs_total = gfs2_ri_total(sdp);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500743 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
744 return;
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500745
746 spin_lock(&sdp->sd_statfs_spin);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500747 gfs2_statfs_change_in(m_sc, m_bh->b_data +
748 sizeof(struct gfs2_dinode));
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500749 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
750 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
751 else
752 new_free = 0;
753 spin_unlock(&sdp->sd_statfs_spin);
Robert Peterson6c532672007-05-10 16:54:38 -0500754 fs_warn(sdp, "File system extended by %llu blocks.\n",
755 (unsigned long long)new_free);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500756 gfs2_statfs_change(sdp, new_free, new_free, 0);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500757
758 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
759 goto out;
760 update_statfs(sdp, m_bh, l_bh);
761 brelse(l_bh);
762out:
763 brelse(m_bh);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500764}
765
766/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700767 * gfs2_stuffed_write_end - Write end for stuffed files
768 * @inode: The inode
769 * @dibh: The buffer_head containing the on-disk inode
770 * @pos: The file position
771 * @len: The length of the write
772 * @copied: How much was actually copied by the VFS
773 * @page: The page
774 *
775 * This copies the data from the page into the inode block after
776 * the inode data structure itself.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000777 *
778 * Returns: errno
779 */
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700780static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
781 loff_t pos, unsigned len, unsigned copied,
782 struct page *page)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000783{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400784 struct gfs2_inode *ip = GFS2_I(inode);
785 struct gfs2_sbd *sdp = GFS2_SB(inode);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500786 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700787 u64 to = pos + copied;
788 void *kaddr;
789 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
790 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000791
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700792 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
793 kaddr = kmap_atomic(page, KM_USER0);
794 memcpy(buf + pos, kaddr + pos, copied);
795 memset(kaddr + pos + copied, 0, len - copied);
796 flush_dcache_page(page);
797 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000798
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700799 if (!PageUptodate(page))
David Teiglandb3b94fa2006-01-16 16:50:04 +0000800 SetPageUptodate(page);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700801 unlock_page(page);
802 page_cache_release(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000803
Abhijith Das7537d81a2009-05-12 11:16:20 -0500804 if (copied) {
805 if (inode->i_size < to) {
806 i_size_write(inode, to);
807 ip->i_disksize = inode->i_size;
808 }
809 gfs2_dinode_out(ip, di);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700810 mark_inode_dirty(inode);
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400811 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000812
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600813 if (inode == sdp->sd_rindex) {
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500814 adjust_fs_space(inode);
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600815 ip->i_gh.gh_flags |= GL_NOCACHE;
816 }
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500817
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000818 brelse(dibh);
819 gfs2_trans_end(sdp);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500820 if (inode == sdp->sd_rindex) {
821 gfs2_glock_dq(&m_ip->i_gh);
822 gfs2_holder_uninit(&m_ip->i_gh);
823 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700824 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000825 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700826 return copied;
827}
David Teiglandb3b94fa2006-01-16 16:50:04 +0000828
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700829/**
830 * gfs2_write_end
831 * @file: The file to write to
832 * @mapping: The address space to write to
833 * @pos: The file position
834 * @len: The length of the data
835 * @copied:
836 * @page: The page that has been written
837 * @fsdata: The fsdata (unused in GFS2)
838 *
839 * The main write_end function for GFS2. We have a separate one for
840 * stuffed files as they are slightly different, otherwise we just
841 * put our locking around the VFS provided functions.
842 *
843 * Returns: errno
844 */
845
846static int gfs2_write_end(struct file *file, struct address_space *mapping,
847 loff_t pos, unsigned len, unsigned copied,
848 struct page *page, void *fsdata)
849{
850 struct inode *inode = page->mapping->host;
851 struct gfs2_inode *ip = GFS2_I(inode);
852 struct gfs2_sbd *sdp = GFS2_SB(inode);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500853 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700854 struct buffer_head *dibh;
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000855 struct gfs2_alloc *al = ip->i_alloc;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700856 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
857 unsigned int to = from + len;
858 int ret;
859
Steven Whitehouse7afd88d2008-02-22 16:07:18 +0000860 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700861
862 ret = gfs2_meta_inode_buffer(ip, &dibh);
863 if (unlikely(ret)) {
864 unlock_page(page);
865 page_cache_release(page);
866 goto failed;
867 }
868
869 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
870
871 if (gfs2_is_stuffed(ip))
872 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
873
Steven Whitehousebf36a712007-10-17 08:35:19 +0100874 if (!gfs2_is_writeback(ip))
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700875 gfs2_page_add_databufs(ip, page, from, to);
876
877 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Abhijith Das7537d81a2009-05-12 11:16:20 -0500878 if (ret > 0) {
879 if (inode->i_size > ip->i_disksize)
880 ip->i_disksize = inode->i_size;
881 gfs2_dinode_out(ip, dibh->b_data);
Steven Whitehouse9656b2c2008-01-08 08:14:30 +0000882 mark_inode_dirty(inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700883 }
884
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600885 if (inode == sdp->sd_rindex) {
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700886 adjust_fs_space(inode);
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600887 ip->i_gh.gh_flags |= GL_NOCACHE;
888 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700889
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000890 brelse(dibh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000891 gfs2_trans_end(sdp);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700892failed:
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000893 if (al) {
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000894 gfs2_inplace_release(ip);
895 gfs2_quota_unlock(ip);
896 gfs2_alloc_put(ip);
897 }
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500898 if (inode == sdp->sd_rindex) {
899 gfs2_glock_dq(&m_ip->i_gh);
900 gfs2_holder_uninit(&m_ip->i_gh);
901 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700902 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000903 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700904 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000905}
906
907/**
Robert Peterson8fb68592007-06-12 11:24:36 -0500908 * gfs2_set_page_dirty - Page dirtying function
909 * @page: The page to dirty
910 *
911 * Returns: 1 if it dirtyed the page, or 0 otherwise
912 */
913
914static int gfs2_set_page_dirty(struct page *page)
915{
Steven Whitehouse55610932007-10-17 08:47:38 +0100916 SetPageChecked(page);
Robert Peterson8fb68592007-06-12 11:24:36 -0500917 return __set_page_dirty_buffers(page);
918}
919
920/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000921 * gfs2_bmap - Block map function
922 * @mapping: Address space info
923 * @lblock: The block to map
924 *
925 * Returns: The disk address for the block or 0 on hole or error
926 */
927
928static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
929{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400930 struct gfs2_inode *ip = GFS2_I(mapping->host);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000931 struct gfs2_holder i_gh;
932 sector_t dblock = 0;
933 int error;
934
David Teiglandb3b94fa2006-01-16 16:50:04 +0000935 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
936 if (error)
937 return 0;
938
939 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600940 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000941
942 gfs2_glock_dq_uninit(&i_gh);
943
944 return dblock;
945}
946
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100947static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
948{
949 struct gfs2_bufdata *bd;
950
951 lock_buffer(bh);
952 gfs2_log_lock(sdp);
953 clear_buffer_dirty(bh);
954 bd = bh->b_private;
955 if (bd) {
Steven Whitehouse16615be2007-09-17 10:59:52 +0100956 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
957 list_del_init(&bd->bd_le.le_list);
958 else
959 gfs2_remove_from_journal(bh, current->journal_info, 0);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100960 }
961 bh->b_bdev = NULL;
962 clear_buffer_mapped(bh);
963 clear_buffer_req(bh);
964 clear_buffer_new(bh);
965 gfs2_log_unlock(sdp);
966 unlock_buffer(bh);
967}
968
Steven Whitehouse8628de02006-03-31 16:48:41 -0500969static void gfs2_invalidatepage(struct page *page, unsigned long offset)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000970{
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100971 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
972 struct buffer_head *bh, *head;
973 unsigned long pos = 0;
974
David Teiglandb3b94fa2006-01-16 16:50:04 +0000975 BUG_ON(!PageLocked(page));
Robert Peterson8fb68592007-06-12 11:24:36 -0500976 if (offset == 0)
977 ClearPageChecked(page);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100978 if (!page_has_buffers(page))
979 goto out;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000980
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100981 bh = head = page_buffers(page);
982 do {
983 if (offset <= pos)
984 gfs2_discard(sdp, bh);
985 pos += bh->b_size;
986 bh = bh->b_this_page;
987 } while (bh != head);
988out:
989 if (offset == 0)
990 try_to_release_page(page, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000991}
992
Steven Whitehousec7b33832006-12-14 18:24:26 +0000993/**
994 * gfs2_ok_for_dio - check that dio is valid on this file
995 * @ip: The inode
996 * @rw: READ or WRITE
997 * @offset: The offset at which we are reading or writing
998 *
999 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
1000 * 1 (to accept the i/o request)
1001 */
1002static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
1003{
1004 /*
1005 * Should we return an error here? I can't see that O_DIRECT for
Steven Whitehouse55610932007-10-17 08:47:38 +01001006 * a stuffed file makes any sense. For now we'll silently fall
1007 * back to buffered I/O
Steven Whitehousec7b33832006-12-14 18:24:26 +00001008 */
Steven Whitehousec7b33832006-12-14 18:24:26 +00001009 if (gfs2_is_stuffed(ip))
1010 return 0;
1011
Bob Petersonacb57a32008-09-11 15:35:37 -04001012 if (offset >= i_size_read(&ip->i_inode))
Steven Whitehousec7b33832006-12-14 18:24:26 +00001013 return 0;
1014 return 1;
1015}
1016
1017
1018
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001019static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
1020 const struct iovec *iov, loff_t offset,
1021 unsigned long nr_segs)
Steven Whitehoused1665e42006-02-14 11:54:42 +00001022{
1023 struct file *file = iocb->ki_filp;
1024 struct inode *inode = file->f_mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -04001025 struct gfs2_inode *ip = GFS2_I(inode);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001026 struct gfs2_holder gh;
1027 int rv;
1028
1029 /*
Steven Whitehousec7b33832006-12-14 18:24:26 +00001030 * Deferred lock, even if its a write, since we do no allocation
1031 * on this path. All we need change is atime, and this lock mode
1032 * ensures that other nodes have flushed their buffered read caches
1033 * (i.e. their page cache entries for this inode). We do not,
1034 * unfortunately have the option of only flushing a range like
1035 * the VFS does.
Steven Whitehoused1665e42006-02-14 11:54:42 +00001036 */
Steven Whitehouse719ee342008-09-18 13:53:59 +01001037 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1038 rv = gfs2_glock_nq(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001039 if (rv)
Steven Whitehousec7b33832006-12-14 18:24:26 +00001040 return rv;
1041 rv = gfs2_ok_for_dio(ip, rw, offset);
1042 if (rv != 1)
1043 goto out; /* dio not valid, fall back to buffered i/o */
Steven Whitehoused1665e42006-02-14 11:54:42 +00001044
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001045 rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1046 offset, nr_segs, gfs2_get_block_direct,
1047 NULL, NULL, 0);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001048out:
1049 gfs2_glock_dq_m(1, &gh);
1050 gfs2_holder_uninit(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001051 return rv;
1052}
1053
1054/**
Steven Whitehouse623d9352006-08-31 12:14:44 -04001055 * gfs2_releasepage - free the metadata associated with a page
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001056 * @page: the page that's being released
1057 * @gfp_mask: passed from Linux VFS, ignored by us
1058 *
1059 * Call try_to_free_buffers() if the buffers in this page can be
1060 * released.
1061 *
1062 * Returns: 0
1063 */
1064
1065int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1066{
Steven Whitehouse009d8512009-12-08 12:12:13 +00001067 struct address_space *mapping = page->mapping;
1068 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001069 struct buffer_head *bh, *head;
1070 struct gfs2_bufdata *bd;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001071
1072 if (!page_has_buffers(page))
Steven Whitehouse891ba6d2007-09-20 15:26:33 +01001073 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001074
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001075 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001076 head = bh = page_buffers(page);
1077 do {
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001078 if (atomic_read(&bh->b_count))
1079 goto cannot_release;
1080 bd = bh->b_private;
1081 if (bd && bd->bd_ail)
1082 goto cannot_release;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001083 gfs2_assert_warn(sdp, !buffer_pinned(bh));
Steven Whitehouse623d9352006-08-31 12:14:44 -04001084 gfs2_assert_warn(sdp, !buffer_dirty(bh));
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001085 bh = bh->b_this_page;
1086 } while(bh != head);
1087 gfs2_log_unlock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001088
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001089 head = bh = page_buffers(page);
1090 do {
Steven Whitehouse623d9352006-08-31 12:14:44 -04001091 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001092 bd = bh->b_private;
1093 if (bd) {
1094 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1095 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
Steven Whitehoused7b616e2007-09-02 10:48:13 +01001096 if (!list_empty(&bd->bd_le.le_list)) {
1097 if (!buffer_pinned(bh))
1098 list_del_init(&bd->bd_le.le_list);
1099 else
1100 bd = NULL;
1101 }
1102 if (bd)
1103 bd->bd_bh = NULL;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001104 bh->b_private = NULL;
1105 }
Steven Whitehouse623d9352006-08-31 12:14:44 -04001106 gfs2_log_unlock(sdp);
1107 if (bd)
1108 kmem_cache_free(gfs2_bufdata_cachep, bd);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001109
1110 bh = bh->b_this_page;
Steven Whitehouse166afcc2006-08-24 15:59:40 -04001111 } while (bh != head);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001112
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001113 return try_to_free_buffers(page);
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001114cannot_release:
1115 gfs2_log_unlock(sdp);
1116 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001117}
1118
Steven Whitehouse55610932007-10-17 08:47:38 +01001119static const struct address_space_operations gfs2_writeback_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001120 .writepage = gfs2_writeback_writepage,
Steven Whitehouse55610932007-10-17 08:47:38 +01001121 .writepages = gfs2_writeback_writepages,
1122 .readpage = gfs2_readpage,
1123 .readpages = gfs2_readpages,
1124 .sync_page = block_sync_page,
1125 .write_begin = gfs2_write_begin,
1126 .write_end = gfs2_write_end,
1127 .bmap = gfs2_bmap,
1128 .invalidatepage = gfs2_invalidatepage,
1129 .releasepage = gfs2_releasepage,
1130 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001131 .migratepage = buffer_migrate_page,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001132 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001133 .error_remove_page = generic_error_remove_page,
Steven Whitehouse55610932007-10-17 08:47:38 +01001134};
1135
1136static const struct address_space_operations gfs2_ordered_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001137 .writepage = gfs2_ordered_writepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001138 .readpage = gfs2_readpage,
Steven Whitehousefd88de562006-05-05 16:59:11 -04001139 .readpages = gfs2_readpages,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001140 .sync_page = block_sync_page,
Steven Whitehouse7765ec22007-10-16 01:25:07 -07001141 .write_begin = gfs2_write_begin,
1142 .write_end = gfs2_write_end,
Robert Peterson8fb68592007-06-12 11:24:36 -05001143 .set_page_dirty = gfs2_set_page_dirty,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001144 .bmap = gfs2_bmap,
1145 .invalidatepage = gfs2_invalidatepage,
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001146 .releasepage = gfs2_releasepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001147 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001148 .migratepage = buffer_migrate_page,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001149 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001150 .error_remove_page = generic_error_remove_page,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001151};
1152
Steven Whitehouse55610932007-10-17 08:47:38 +01001153static const struct address_space_operations gfs2_jdata_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001154 .writepage = gfs2_jdata_writepage,
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +01001155 .writepages = gfs2_jdata_writepages,
Steven Whitehouse55610932007-10-17 08:47:38 +01001156 .readpage = gfs2_readpage,
1157 .readpages = gfs2_readpages,
1158 .sync_page = block_sync_page,
1159 .write_begin = gfs2_write_begin,
1160 .write_end = gfs2_write_end,
1161 .set_page_dirty = gfs2_set_page_dirty,
1162 .bmap = gfs2_bmap,
1163 .invalidatepage = gfs2_invalidatepage,
1164 .releasepage = gfs2_releasepage,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001165 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001166 .error_remove_page = generic_error_remove_page,
Steven Whitehouse55610932007-10-17 08:47:38 +01001167};
1168
1169void gfs2_set_aops(struct inode *inode)
1170{
1171 struct gfs2_inode *ip = GFS2_I(inode);
1172
1173 if (gfs2_is_writeback(ip))
1174 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1175 else if (gfs2_is_ordered(ip))
1176 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1177 else if (gfs2_is_jdata(ip))
1178 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1179 else
1180 BUG();
1181}
1182