blob: 694b5d48f0366e1b535abb102cbf456f3ae2e2c6 [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
139 ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc);
140 if (ret == -EAGAIN)
141 ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
142 return ret;
143}
144
145/**
146 * gfs2_ordered_writepage - Write page for ordered data files
147 * @page: The page to write
148 * @wbc: The writeback control
149 *
150 */
151
152static int gfs2_ordered_writepage(struct page *page,
153 struct writeback_control *wbc)
154{
155 struct inode *inode = page->mapping->host;
156 struct gfs2_inode *ip = GFS2_I(inode);
157 int ret;
158
159 ret = gfs2_writepage_common(page, wbc);
160 if (ret <= 0)
161 return ret;
162
163 if (!page_has_buffers(page)) {
164 create_empty_buffers(page, inode->i_sb->s_blocksize,
165 (1 << BH_Dirty)|(1 << BH_Uptodate));
166 }
167 gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
168 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
169}
170
171/**
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100172 * __gfs2_jdata_writepage - The core of jdata writepage
173 * @page: The page to write
174 * @wbc: The writeback control
175 *
176 * This is shared between writepage and writepages and implements the
177 * core of the writepage operation. If a transaction is required then
178 * PageChecked will have been set and the transaction will have
179 * already been started before this is called.
180 */
181
182static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
183{
184 struct inode *inode = page->mapping->host;
185 struct gfs2_inode *ip = GFS2_I(inode);
186 struct gfs2_sbd *sdp = GFS2_SB(inode);
187
188 if (PageChecked(page)) {
189 ClearPageChecked(page);
190 if (!page_has_buffers(page)) {
191 create_empty_buffers(page, inode->i_sb->s_blocksize,
192 (1 << BH_Dirty)|(1 << BH_Uptodate));
193 }
194 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
195 }
196 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
197}
198
199/**
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100200 * gfs2_jdata_writepage - Write complete page
201 * @page: Page to write
202 *
203 * Returns: errno
204 *
205 */
206
207static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
208{
209 struct inode *inode = page->mapping->host;
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +0100210 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100211 int ret;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000212 int done_trans = 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000213
Steven Whitehousebf36a712007-10-17 08:35:19 +0100214 if (PageChecked(page)) {
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100215 if (wbc->sync_mode != WB_SYNC_ALL)
216 goto out_ignore;
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100217 ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
218 if (ret)
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000219 goto out_ignore;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000220 done_trans = 1;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000221 }
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100222 ret = gfs2_writepage_common(page, wbc);
223 if (ret > 0)
224 ret = __gfs2_jdata_writepage(page, wbc);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000225 if (done_trans)
226 gfs2_trans_end(sdp);
Steven Whitehouse1bb73222008-10-15 09:46:39 +0100227 return ret;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000228
229out_ignore:
230 redirty_page_for_writepage(wbc, page);
231 unlock_page(page);
232 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000233}
234
235/**
Steven Whitehouse55610932007-10-17 08:47:38 +0100236 * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000237 * @mapping: The mapping to write
238 * @wbc: Write-back control
239 *
Steven Whitehouse55610932007-10-17 08:47:38 +0100240 * For the data=writeback case we can already ignore buffer heads
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000241 * and write whole extents at once. This is a big reduction in the
242 * number of I/O requests we send and the bmap calls we make in this case.
243 */
Steven Whitehouse55610932007-10-17 08:47:38 +0100244static int gfs2_writeback_writepages(struct address_space *mapping,
245 struct writeback_control *wbc)
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000246{
Steven Whitehouse55610932007-10-17 08:47:38 +0100247 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000248}
249
250/**
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100251 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
252 * @mapping: The mapping
253 * @wbc: The writeback control
254 * @writepage: The writepage function to call for each page
255 * @pvec: The vector of pages
256 * @nr_pages: The number of pages to write
257 *
258 * Returns: non-zero if loop should terminate, zero otherwise
259 */
260
261static int gfs2_write_jdata_pagevec(struct address_space *mapping,
262 struct writeback_control *wbc,
263 struct pagevec *pvec,
264 int nr_pages, pgoff_t end)
265{
266 struct inode *inode = mapping->host;
267 struct gfs2_sbd *sdp = GFS2_SB(inode);
268 loff_t i_size = i_size_read(inode);
269 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
270 unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
271 unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
272 struct backing_dev_info *bdi = mapping->backing_dev_info;
273 int i;
274 int ret;
275
Abhijith Das20b95bf2008-03-06 17:43:52 -0600276 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100277 if (ret < 0)
278 return ret;
279
280 for(i = 0; i < nr_pages; i++) {
281 struct page *page = pvec->pages[i];
282
283 lock_page(page);
284
285 if (unlikely(page->mapping != mapping)) {
286 unlock_page(page);
287 continue;
288 }
289
290 if (!wbc->range_cyclic && page->index > end) {
291 ret = 1;
292 unlock_page(page);
293 continue;
294 }
295
296 if (wbc->sync_mode != WB_SYNC_NONE)
297 wait_on_page_writeback(page);
298
299 if (PageWriteback(page) ||
300 !clear_page_dirty_for_io(page)) {
301 unlock_page(page);
302 continue;
303 }
304
305 /* Is the page fully outside i_size? (truncate in progress) */
306 if (page->index > end_index || (page->index == end_index && !offset)) {
307 page->mapping->a_ops->invalidatepage(page, 0);
308 unlock_page(page);
309 continue;
310 }
311
312 ret = __gfs2_jdata_writepage(page, wbc);
313
314 if (ret || (--(wbc->nr_to_write) <= 0))
315 ret = 1;
316 if (wbc->nonblocking && bdi_write_congested(bdi)) {
317 wbc->encountered_congestion = 1;
318 ret = 1;
319 }
320
321 }
322 gfs2_trans_end(sdp);
323 return ret;
324}
325
326/**
327 * gfs2_write_cache_jdata - Like write_cache_pages but different
328 * @mapping: The mapping to write
329 * @wbc: The writeback control
330 * @writepage: The writepage function to call
331 * @data: The data to pass to writepage
332 *
333 * The reason that we use our own function here is that we need to
334 * start transactions before we grab page locks. This allows us
335 * to get the ordering right.
336 */
337
338static int gfs2_write_cache_jdata(struct address_space *mapping,
339 struct writeback_control *wbc)
340{
341 struct backing_dev_info *bdi = mapping->backing_dev_info;
342 int ret = 0;
343 int done = 0;
344 struct pagevec pvec;
345 int nr_pages;
346 pgoff_t index;
347 pgoff_t end;
348 int scanned = 0;
349 int range_whole = 0;
350
351 if (wbc->nonblocking && bdi_write_congested(bdi)) {
352 wbc->encountered_congestion = 1;
353 return 0;
354 }
355
356 pagevec_init(&pvec, 0);
357 if (wbc->range_cyclic) {
358 index = mapping->writeback_index; /* Start from prev offset */
359 end = -1;
360 } else {
361 index = wbc->range_start >> PAGE_CACHE_SHIFT;
362 end = wbc->range_end >> PAGE_CACHE_SHIFT;
363 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
364 range_whole = 1;
365 scanned = 1;
366 }
367
368retry:
369 while (!done && (index <= end) &&
370 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
371 PAGECACHE_TAG_DIRTY,
372 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
373 scanned = 1;
374 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
375 if (ret)
376 done = 1;
377 if (ret > 0)
378 ret = 0;
379
380 pagevec_release(&pvec);
381 cond_resched();
382 }
383
384 if (!scanned && !done) {
385 /*
386 * We hit the last page and there is more work to be done: wrap
387 * back to the start of the file
388 */
389 scanned = 1;
390 index = 0;
391 goto retry;
392 }
393
394 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
395 mapping->writeback_index = index;
396 return ret;
397}
398
399
400/**
401 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
402 * @mapping: The mapping to write
403 * @wbc: The writeback control
404 *
405 */
406
407static int gfs2_jdata_writepages(struct address_space *mapping,
408 struct writeback_control *wbc)
409{
410 struct gfs2_inode *ip = GFS2_I(mapping->host);
411 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
412 int ret;
413
414 ret = gfs2_write_cache_jdata(mapping, wbc);
415 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
416 gfs2_log_flush(sdp, ip->i_gl);
417 ret = gfs2_write_cache_jdata(mapping, wbc);
418 }
419 return ret;
420}
421
422/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000423 * stuffed_readpage - Fill in a Linux page with stuffed file data
424 * @ip: the inode
425 * @page: the page
426 *
427 * Returns: errno
428 */
429
430static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
431{
432 struct buffer_head *dibh;
433 void *kaddr;
434 int error;
435
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100436 /*
Nick Piggin3c18ddd2008-04-28 02:12:10 -0700437 * Due to the order of unstuffing files and ->fault(), we can be
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100438 * asked for a zero page in the case of a stuffed file being extended,
439 * so we need to supply one here. It doesn't happen often.
440 */
441 if (unlikely(page->index)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800442 zero_user(page, 0, PAGE_CACHE_SIZE);
Abhijith Das0a7ab792009-01-07 16:03:37 -0600443 SetPageUptodate(page);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100444 return 0;
445 }
Steven Whitehousefd88de562006-05-05 16:59:11 -0400446
David Teiglandb3b94fa2006-01-16 16:50:04 +0000447 error = gfs2_meta_inode_buffer(ip, &dibh);
448 if (error)
449 return error;
450
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000451 kaddr = kmap_atomic(page, KM_USER0);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400452 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
Steven Whitehousec9e98882008-11-04 09:47:33 +0000453 ip->i_disksize);
454 memset(kaddr + ip->i_disksize, 0, PAGE_CACHE_SIZE - ip->i_disksize);
Russell Cattelanc312c4f2006-10-12 09:23:41 -0400455 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100456 flush_dcache_page(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000457 brelse(dibh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000458 SetPageUptodate(page);
459
460 return 0;
461}
462
David Teiglandb3b94fa2006-01-16 16:50:04 +0000463
464/**
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100465 * __gfs2_readpage - readpage
466 * @file: The file to read a page for
David Teiglandb3b94fa2006-01-16 16:50:04 +0000467 * @page: The page to read
468 *
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100469 * This is the core of gfs2's readpage. Its used by the internal file
470 * reading code as in that case we already hold the glock. Also its
471 * called by gfs2_readpage() once the required lock has been granted.
472 *
473 */
474
475static int __gfs2_readpage(void *file, struct page *page)
476{
477 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
478 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
479 int error;
480
481 if (gfs2_is_stuffed(ip)) {
482 error = stuffed_readpage(ip, page);
483 unlock_page(page);
484 } else {
Bob Petersone9e1ef22007-12-10 14:13:27 -0600485 error = mpage_readpage(page, gfs2_block_map);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100486 }
487
488 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
489 return -EIO;
490
491 return error;
492}
493
494/**
495 * gfs2_readpage - read a page of a file
496 * @file: The file to read
497 * @page: The page of the file
498 *
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100499 * This deals with the locking required. We have to unlock and
500 * relock the page in order to get the locking in the right
501 * order.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000502 */
503
504static int gfs2_readpage(struct file *file, struct page *page)
505{
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100506 struct address_space *mapping = page->mapping;
507 struct gfs2_inode *ip = GFS2_I(mapping->host);
Steven Whitehouse6802e342008-05-21 17:03:22 +0100508 struct gfs2_holder gh;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000509 int error;
510
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100511 unlock_page(page);
Steven Whitehouse719ee342008-09-18 13:53:59 +0100512 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
513 error = gfs2_glock_nq(&gh);
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100514 if (unlikely(error))
Steven Whitehouse6802e342008-05-21 17:03:22 +0100515 goto out;
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100516 error = AOP_TRUNCATED_PAGE;
517 lock_page(page);
518 if (page->mapping == mapping && !PageUptodate(page))
519 error = __gfs2_readpage(file, page);
520 else
521 unlock_page(page);
Steven Whitehouse6802e342008-05-21 17:03:22 +0100522 gfs2_glock_dq(&gh);
Steven Whitehouse7afd88d2008-02-22 16:07:18 +0000523out:
Steven Whitehouse6802e342008-05-21 17:03:22 +0100524 gfs2_holder_uninit(&gh);
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100525 if (error && error != AOP_TRUNCATED_PAGE)
526 lock_page(page);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100527 return error;
528}
529
530/**
531 * gfs2_internal_read - read an internal file
532 * @ip: The gfs2 inode
533 * @ra_state: The readahead state (or NULL for no readahead)
534 * @buf: The buffer to fill
535 * @pos: The file position
536 * @size: The amount to read
537 *
538 */
539
540int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
541 char *buf, loff_t *pos, unsigned size)
542{
543 struct address_space *mapping = ip->i_inode.i_mapping;
544 unsigned long index = *pos / PAGE_CACHE_SIZE;
545 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
546 unsigned copied = 0;
547 unsigned amt;
548 struct page *page;
549 void *p;
550
551 do {
552 amt = size - copied;
553 if (offset + size > PAGE_CACHE_SIZE)
554 amt = PAGE_CACHE_SIZE - offset;
555 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
556 if (IS_ERR(page))
557 return PTR_ERR(page);
558 p = kmap_atomic(page, KM_USER0);
559 memcpy(buf + copied, p + offset, amt);
560 kunmap_atomic(p, KM_USER0);
561 mark_page_accessed(page);
562 page_cache_release(page);
563 copied += amt;
564 index++;
565 offset = 0;
566 } while(copied < size);
567 (*pos) += size;
568 return size;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400569}
570
Steven Whitehousefd88de562006-05-05 16:59:11 -0400571/**
572 * gfs2_readpages - Read a bunch of pages at once
573 *
574 * Some notes:
575 * 1. This is only for readahead, so we can simply ignore any things
576 * which are slightly inconvenient (such as locking conflicts between
577 * the page lock and the glock) and return having done no I/O. Its
578 * obviously not something we'd want to do on too regular a basis.
579 * Any I/O we ignore at this time will be done via readpage later.
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500580 * 2. We don't handle stuffed files here we let readpage do the honours.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400581 * 3. mpage_readpages() does most of the heavy lifting in the common case.
Bob Petersone9e1ef22007-12-10 14:13:27 -0600582 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400583 */
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100584
Steven Whitehousefd88de562006-05-05 16:59:11 -0400585static int gfs2_readpages(struct file *file, struct address_space *mapping,
586 struct list_head *pages, unsigned nr_pages)
587{
588 struct inode *inode = mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400589 struct gfs2_inode *ip = GFS2_I(inode);
590 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400591 struct gfs2_holder gh;
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100592 int ret;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400593
Steven Whitehouse719ee342008-09-18 13:53:59 +0100594 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
595 ret = gfs2_glock_nq(&gh);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100596 if (unlikely(ret))
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100597 goto out_uninit;
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500598 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600599 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100600 gfs2_glock_dq(&gh);
601out_uninit:
602 gfs2_holder_uninit(&gh);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400603 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
604 ret = -EIO;
605 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000606}
607
608/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700609 * gfs2_write_begin - Begin to write to a file
David Teiglandb3b94fa2006-01-16 16:50:04 +0000610 * @file: The file to write to
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700611 * @mapping: The mapping in which to write
612 * @pos: The file offset at which to start writing
613 * @len: Length of the write
614 * @flags: Various flags
615 * @pagep: Pointer to return the page
616 * @fsdata: Pointer to return fs data (unused by GFS2)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000617 *
618 * Returns: errno
619 */
620
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700621static int gfs2_write_begin(struct file *file, struct address_space *mapping,
622 loff_t pos, unsigned len, unsigned flags,
623 struct page **pagep, void **fsdata)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000624{
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700625 struct gfs2_inode *ip = GFS2_I(mapping->host);
626 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500627 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7ed122e2008-12-10 10:28:10 +0000628 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000629 int alloc_required;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000630 int error = 0;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000631 struct gfs2_alloc *al;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700632 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
633 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
634 unsigned to = from + len;
635 struct page *page;
Russell Cattelan52ae7b72006-10-09 12:11:54 -0500636
Steven Whitehouse719ee342008-09-18 13:53:59 +0100637 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
638 error = gfs2_glock_nq(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700639 if (unlikely(error))
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000640 goto out_uninit;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500641 if (&ip->i_inode == sdp->sd_rindex) {
642 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
643 GL_NOCACHE, &m_ip->i_gh);
644 if (unlikely(error)) {
645 gfs2_glock_dq(&ip->i_gh);
646 goto out_uninit;
647 }
648 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000649
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700650 error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
651 if (error)
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100652 goto out_unlock;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000653
Steven Whitehouse7ed122e2008-12-10 10:28:10 +0000654 if (alloc_required || gfs2_is_jdata(ip))
655 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
656
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000657 if (alloc_required) {
658 al = gfs2_alloc_get(ip);
Cyrill Gorcunov182fe5a2008-03-03 21:54:21 +0300659 if (!al) {
660 error = -ENOMEM;
661 goto out_unlock;
662 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000663
Steven Whitehoused82661d2008-03-10 15:34:50 +0000664 error = gfs2_quota_lock_check(ip);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000665 if (error)
666 goto out_alloc_put;
667
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000668 al->al_requested = data_blocks + ind_blocks;
669 error = gfs2_inplace_reserve(ip);
670 if (error)
671 goto out_qunlock;
672 }
673
674 rblocks = RES_DINODE + ind_blocks;
675 if (gfs2_is_jdata(ip))
676 rblocks += data_blocks ? data_blocks : 1;
677 if (ind_blocks || data_blocks)
678 rblocks += RES_STATFS + RES_QUOTA;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500679 if (&ip->i_inode == sdp->sd_rindex)
680 rblocks += 2 * RES_STATFS;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000681
Steven Whitehouse16615be2007-09-17 10:59:52 +0100682 error = gfs2_trans_begin(sdp, rblocks,
683 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000684 if (error)
Steven Whitehousea867bb22007-07-17 10:29:02 +0100685 goto out_trans_fail;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000686
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100687 error = -ENOMEM;
Steven Whitehousee4fefba2009-01-06 10:08:33 +0000688 flags |= AOP_FLAG_NOFS;
Nick Piggin54566b22009-01-04 12:00:53 -0800689 page = grab_cache_page_write_begin(mapping, index, flags);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100690 *pagep = page;
691 if (unlikely(!page))
692 goto out_endtrans;
693
David Teiglandb3b94fa2006-01-16 16:50:04 +0000694 if (gfs2_is_stuffed(ip)) {
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100695 error = 0;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700696 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
Steven Whitehousef25ef0c2006-07-26 10:51:20 -0400697 error = gfs2_unstuff_dinode(ip, page);
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000698 if (error == 0)
699 goto prepare_write;
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100700 } else if (!PageUptodate(page)) {
David Teiglandb3b94fa2006-01-16 16:50:04 +0000701 error = stuffed_readpage(ip, page);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100702 }
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000703 goto out;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000704 }
705
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000706prepare_write:
Bob Petersone9e1ef22007-12-10 14:13:27 -0600707 error = block_prepare_write(page, from, to, gfs2_block_map);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000708out:
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100709 if (error == 0)
710 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000711
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100712 page_cache_release(page);
713 if (pos + len > ip->i_inode.i_size)
714 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
715out_endtrans:
716 gfs2_trans_end(sdp);
717out_trans_fail:
718 if (alloc_required) {
719 gfs2_inplace_release(ip);
720out_qunlock:
721 gfs2_quota_unlock(ip);
722out_alloc_put:
723 gfs2_alloc_put(ip);
724 }
725out_unlock:
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500726 if (&ip->i_inode == sdp->sd_rindex) {
727 gfs2_glock_dq(&m_ip->i_gh);
728 gfs2_holder_uninit(&m_ip->i_gh);
729 }
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100730 gfs2_glock_dq(&ip->i_gh);
731out_uninit:
732 gfs2_holder_uninit(&ip->i_gh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000733 return error;
734}
735
736/**
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500737 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
738 * @inode: the rindex inode
739 */
740static void adjust_fs_space(struct inode *inode)
741{
742 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500743 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
744 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500745 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
746 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500747 struct buffer_head *m_bh, *l_bh;
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500748 u64 fs_total, new_free;
749
750 /* Total up the file system space, according to the latest rindex. */
751 fs_total = gfs2_ri_total(sdp);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500752 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
753 return;
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500754
755 spin_lock(&sdp->sd_statfs_spin);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500756 gfs2_statfs_change_in(m_sc, m_bh->b_data +
757 sizeof(struct gfs2_dinode));
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500758 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
759 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
760 else
761 new_free = 0;
762 spin_unlock(&sdp->sd_statfs_spin);
Robert Peterson6c532672007-05-10 16:54:38 -0500763 fs_warn(sdp, "File system extended by %llu blocks.\n",
764 (unsigned long long)new_free);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500765 gfs2_statfs_change(sdp, new_free, new_free, 0);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500766
767 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
768 goto out;
769 update_statfs(sdp, m_bh, l_bh);
770 brelse(l_bh);
771out:
772 brelse(m_bh);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500773}
774
775/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700776 * gfs2_stuffed_write_end - Write end for stuffed files
777 * @inode: The inode
778 * @dibh: The buffer_head containing the on-disk inode
779 * @pos: The file position
780 * @len: The length of the write
781 * @copied: How much was actually copied by the VFS
782 * @page: The page
783 *
784 * This copies the data from the page into the inode block after
785 * the inode data structure itself.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000786 *
787 * Returns: errno
788 */
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700789static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
790 loff_t pos, unsigned len, unsigned copied,
791 struct page *page)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000792{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400793 struct gfs2_inode *ip = GFS2_I(inode);
794 struct gfs2_sbd *sdp = GFS2_SB(inode);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500795 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700796 u64 to = pos + copied;
797 void *kaddr;
798 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
799 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000800
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700801 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
802 kaddr = kmap_atomic(page, KM_USER0);
803 memcpy(buf + pos, kaddr + pos, copied);
804 memset(kaddr + pos + copied, 0, len - copied);
805 flush_dcache_page(page);
806 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000807
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700808 if (!PageUptodate(page))
David Teiglandb3b94fa2006-01-16 16:50:04 +0000809 SetPageUptodate(page);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700810 unlock_page(page);
811 page_cache_release(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000812
Abhijith Das7537d81a2009-05-12 11:16:20 -0500813 if (copied) {
814 if (inode->i_size < to) {
815 i_size_write(inode, to);
816 ip->i_disksize = inode->i_size;
817 }
818 gfs2_dinode_out(ip, di);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700819 mark_inode_dirty(inode);
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400820 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000821
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500822 if (inode == sdp->sd_rindex)
823 adjust_fs_space(inode);
824
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000825 brelse(dibh);
826 gfs2_trans_end(sdp);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500827 if (inode == sdp->sd_rindex) {
828 gfs2_glock_dq(&m_ip->i_gh);
829 gfs2_holder_uninit(&m_ip->i_gh);
830 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700831 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000832 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700833 return copied;
834}
David Teiglandb3b94fa2006-01-16 16:50:04 +0000835
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700836/**
837 * gfs2_write_end
838 * @file: The file to write to
839 * @mapping: The address space to write to
840 * @pos: The file position
841 * @len: The length of the data
842 * @copied:
843 * @page: The page that has been written
844 * @fsdata: The fsdata (unused in GFS2)
845 *
846 * The main write_end function for GFS2. We have a separate one for
847 * stuffed files as they are slightly different, otherwise we just
848 * put our locking around the VFS provided functions.
849 *
850 * Returns: errno
851 */
852
853static int gfs2_write_end(struct file *file, struct address_space *mapping,
854 loff_t pos, unsigned len, unsigned copied,
855 struct page *page, void *fsdata)
856{
857 struct inode *inode = page->mapping->host;
858 struct gfs2_inode *ip = GFS2_I(inode);
859 struct gfs2_sbd *sdp = GFS2_SB(inode);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500860 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700861 struct buffer_head *dibh;
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000862 struct gfs2_alloc *al = ip->i_alloc;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700863 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
864 unsigned int to = from + len;
865 int ret;
866
Steven Whitehouse7afd88d2008-02-22 16:07:18 +0000867 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700868
869 ret = gfs2_meta_inode_buffer(ip, &dibh);
870 if (unlikely(ret)) {
871 unlock_page(page);
872 page_cache_release(page);
873 goto failed;
874 }
875
876 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
877
878 if (gfs2_is_stuffed(ip))
879 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
880
Steven Whitehousebf36a712007-10-17 08:35:19 +0100881 if (!gfs2_is_writeback(ip))
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700882 gfs2_page_add_databufs(ip, page, from, to);
883
884 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Abhijith Das7537d81a2009-05-12 11:16:20 -0500885 if (ret > 0) {
886 if (inode->i_size > ip->i_disksize)
887 ip->i_disksize = inode->i_size;
888 gfs2_dinode_out(ip, dibh->b_data);
Steven Whitehouse9656b2c2008-01-08 08:14:30 +0000889 mark_inode_dirty(inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700890 }
891
892 if (inode == sdp->sd_rindex)
893 adjust_fs_space(inode);
894
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000895 brelse(dibh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000896 gfs2_trans_end(sdp);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700897failed:
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000898 if (al) {
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000899 gfs2_inplace_release(ip);
900 gfs2_quota_unlock(ip);
901 gfs2_alloc_put(ip);
902 }
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500903 if (inode == sdp->sd_rindex) {
904 gfs2_glock_dq(&m_ip->i_gh);
905 gfs2_holder_uninit(&m_ip->i_gh);
906 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700907 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000908 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700909 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000910}
911
912/**
Robert Peterson8fb68592007-06-12 11:24:36 -0500913 * gfs2_set_page_dirty - Page dirtying function
914 * @page: The page to dirty
915 *
916 * Returns: 1 if it dirtyed the page, or 0 otherwise
917 */
918
919static int gfs2_set_page_dirty(struct page *page)
920{
Steven Whitehouse55610932007-10-17 08:47:38 +0100921 SetPageChecked(page);
Robert Peterson8fb68592007-06-12 11:24:36 -0500922 return __set_page_dirty_buffers(page);
923}
924
925/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000926 * gfs2_bmap - Block map function
927 * @mapping: Address space info
928 * @lblock: The block to map
929 *
930 * Returns: The disk address for the block or 0 on hole or error
931 */
932
933static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
934{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400935 struct gfs2_inode *ip = GFS2_I(mapping->host);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000936 struct gfs2_holder i_gh;
937 sector_t dblock = 0;
938 int error;
939
David Teiglandb3b94fa2006-01-16 16:50:04 +0000940 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
941 if (error)
942 return 0;
943
944 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600945 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000946
947 gfs2_glock_dq_uninit(&i_gh);
948
949 return dblock;
950}
951
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100952static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
953{
954 struct gfs2_bufdata *bd;
955
956 lock_buffer(bh);
957 gfs2_log_lock(sdp);
958 clear_buffer_dirty(bh);
959 bd = bh->b_private;
960 if (bd) {
Steven Whitehouse16615be2007-09-17 10:59:52 +0100961 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
962 list_del_init(&bd->bd_le.le_list);
963 else
964 gfs2_remove_from_journal(bh, current->journal_info, 0);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100965 }
966 bh->b_bdev = NULL;
967 clear_buffer_mapped(bh);
968 clear_buffer_req(bh);
969 clear_buffer_new(bh);
970 gfs2_log_unlock(sdp);
971 unlock_buffer(bh);
972}
973
Steven Whitehouse8628de02006-03-31 16:48:41 -0500974static void gfs2_invalidatepage(struct page *page, unsigned long offset)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000975{
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100976 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
977 struct buffer_head *bh, *head;
978 unsigned long pos = 0;
979
David Teiglandb3b94fa2006-01-16 16:50:04 +0000980 BUG_ON(!PageLocked(page));
Robert Peterson8fb68592007-06-12 11:24:36 -0500981 if (offset == 0)
982 ClearPageChecked(page);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100983 if (!page_has_buffers(page))
984 goto out;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000985
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100986 bh = head = page_buffers(page);
987 do {
988 if (offset <= pos)
989 gfs2_discard(sdp, bh);
990 pos += bh->b_size;
991 bh = bh->b_this_page;
992 } while (bh != head);
993out:
994 if (offset == 0)
995 try_to_release_page(page, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000996}
997
Steven Whitehousec7b33832006-12-14 18:24:26 +0000998/**
999 * gfs2_ok_for_dio - check that dio is valid on this file
1000 * @ip: The inode
1001 * @rw: READ or WRITE
1002 * @offset: The offset at which we are reading or writing
1003 *
1004 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
1005 * 1 (to accept the i/o request)
1006 */
1007static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
1008{
1009 /*
1010 * Should we return an error here? I can't see that O_DIRECT for
Steven Whitehouse55610932007-10-17 08:47:38 +01001011 * a stuffed file makes any sense. For now we'll silently fall
1012 * back to buffered I/O
Steven Whitehousec7b33832006-12-14 18:24:26 +00001013 */
Steven Whitehousec7b33832006-12-14 18:24:26 +00001014 if (gfs2_is_stuffed(ip))
1015 return 0;
1016
Bob Petersonacb57a32008-09-11 15:35:37 -04001017 if (offset >= i_size_read(&ip->i_inode))
Steven Whitehousec7b33832006-12-14 18:24:26 +00001018 return 0;
1019 return 1;
1020}
1021
1022
1023
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001024static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
1025 const struct iovec *iov, loff_t offset,
1026 unsigned long nr_segs)
Steven Whitehoused1665e42006-02-14 11:54:42 +00001027{
1028 struct file *file = iocb->ki_filp;
1029 struct inode *inode = file->f_mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -04001030 struct gfs2_inode *ip = GFS2_I(inode);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001031 struct gfs2_holder gh;
1032 int rv;
1033
1034 /*
Steven Whitehousec7b33832006-12-14 18:24:26 +00001035 * Deferred lock, even if its a write, since we do no allocation
1036 * on this path. All we need change is atime, and this lock mode
1037 * ensures that other nodes have flushed their buffered read caches
1038 * (i.e. their page cache entries for this inode). We do not,
1039 * unfortunately have the option of only flushing a range like
1040 * the VFS does.
Steven Whitehoused1665e42006-02-14 11:54:42 +00001041 */
Steven Whitehouse719ee342008-09-18 13:53:59 +01001042 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1043 rv = gfs2_glock_nq(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001044 if (rv)
Steven Whitehousec7b33832006-12-14 18:24:26 +00001045 return rv;
1046 rv = gfs2_ok_for_dio(ip, rw, offset);
1047 if (rv != 1)
1048 goto out; /* dio not valid, fall back to buffered i/o */
Steven Whitehoused1665e42006-02-14 11:54:42 +00001049
Steven Whitehousec7b33832006-12-14 18:24:26 +00001050 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
1051 iov, offset, nr_segs,
1052 gfs2_get_block_direct, NULL);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001053out:
1054 gfs2_glock_dq_m(1, &gh);
1055 gfs2_holder_uninit(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001056 return rv;
1057}
1058
1059/**
Steven Whitehouse623d9352006-08-31 12:14:44 -04001060 * gfs2_releasepage - free the metadata associated with a page
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001061 * @page: the page that's being released
1062 * @gfp_mask: passed from Linux VFS, ignored by us
1063 *
1064 * Call try_to_free_buffers() if the buffers in this page can be
1065 * released.
1066 *
1067 * Returns: 0
1068 */
1069
1070int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1071{
1072 struct inode *aspace = page->mapping->host;
1073 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
1074 struct buffer_head *bh, *head;
1075 struct gfs2_bufdata *bd;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001076
1077 if (!page_has_buffers(page))
Steven Whitehouse891ba6d2007-09-20 15:26:33 +01001078 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001079
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001080 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001081 head = bh = page_buffers(page);
1082 do {
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001083 if (atomic_read(&bh->b_count))
1084 goto cannot_release;
1085 bd = bh->b_private;
1086 if (bd && bd->bd_ail)
1087 goto cannot_release;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001088 gfs2_assert_warn(sdp, !buffer_pinned(bh));
Steven Whitehouse623d9352006-08-31 12:14:44 -04001089 gfs2_assert_warn(sdp, !buffer_dirty(bh));
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001090 bh = bh->b_this_page;
1091 } while(bh != head);
1092 gfs2_log_unlock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001093
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001094 head = bh = page_buffers(page);
1095 do {
Steven Whitehouse623d9352006-08-31 12:14:44 -04001096 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001097 bd = bh->b_private;
1098 if (bd) {
1099 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1100 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
Steven Whitehoused7b616e2007-09-02 10:48:13 +01001101 if (!list_empty(&bd->bd_le.le_list)) {
1102 if (!buffer_pinned(bh))
1103 list_del_init(&bd->bd_le.le_list);
1104 else
1105 bd = NULL;
1106 }
1107 if (bd)
1108 bd->bd_bh = NULL;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001109 bh->b_private = NULL;
1110 }
Steven Whitehouse623d9352006-08-31 12:14:44 -04001111 gfs2_log_unlock(sdp);
1112 if (bd)
1113 kmem_cache_free(gfs2_bufdata_cachep, bd);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001114
1115 bh = bh->b_this_page;
Steven Whitehouse166afcc2006-08-24 15:59:40 -04001116 } while (bh != head);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001117
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001118 return try_to_free_buffers(page);
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001119cannot_release:
1120 gfs2_log_unlock(sdp);
1121 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001122}
1123
Steven Whitehouse55610932007-10-17 08:47:38 +01001124static const struct address_space_operations gfs2_writeback_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001125 .writepage = gfs2_writeback_writepage,
Steven Whitehouse55610932007-10-17 08:47:38 +01001126 .writepages = gfs2_writeback_writepages,
1127 .readpage = gfs2_readpage,
1128 .readpages = gfs2_readpages,
1129 .sync_page = block_sync_page,
1130 .write_begin = gfs2_write_begin,
1131 .write_end = gfs2_write_end,
1132 .bmap = gfs2_bmap,
1133 .invalidatepage = gfs2_invalidatepage,
1134 .releasepage = gfs2_releasepage,
1135 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001136 .migratepage = buffer_migrate_page,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001137 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001138 .error_remove_page = generic_error_remove_page,
Steven Whitehouse55610932007-10-17 08:47:38 +01001139};
1140
1141static const struct address_space_operations gfs2_ordered_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001142 .writepage = gfs2_ordered_writepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001143 .readpage = gfs2_readpage,
Steven Whitehousefd88de562006-05-05 16:59:11 -04001144 .readpages = gfs2_readpages,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001145 .sync_page = block_sync_page,
Steven Whitehouse7765ec22007-10-16 01:25:07 -07001146 .write_begin = gfs2_write_begin,
1147 .write_end = gfs2_write_end,
Robert Peterson8fb68592007-06-12 11:24:36 -05001148 .set_page_dirty = gfs2_set_page_dirty,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001149 .bmap = gfs2_bmap,
1150 .invalidatepage = gfs2_invalidatepage,
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001151 .releasepage = gfs2_releasepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001152 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001153 .migratepage = buffer_migrate_page,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001154 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001155 .error_remove_page = generic_error_remove_page,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001156};
1157
Steven Whitehouse55610932007-10-17 08:47:38 +01001158static const struct address_space_operations gfs2_jdata_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001159 .writepage = gfs2_jdata_writepage,
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +01001160 .writepages = gfs2_jdata_writepages,
Steven Whitehouse55610932007-10-17 08:47:38 +01001161 .readpage = gfs2_readpage,
1162 .readpages = gfs2_readpages,
1163 .sync_page = block_sync_page,
1164 .write_begin = gfs2_write_begin,
1165 .write_end = gfs2_write_end,
1166 .set_page_dirty = gfs2_set_page_dirty,
1167 .bmap = gfs2_bmap,
1168 .invalidatepage = gfs2_invalidatepage,
1169 .releasepage = gfs2_releasepage,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001170 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001171 .error_remove_page = generic_error_remove_page,
Steven Whitehouse55610932007-10-17 08:47:38 +01001172};
1173
1174void gfs2_set_aops(struct inode *inode)
1175{
1176 struct gfs2_inode *ip = GFS2_I(inode);
1177
1178 if (gfs2_is_writeback(ip))
1179 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1180 else if (gfs2_is_ordered(ip))
1181 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1182 else if (gfs2_is_jdata(ip))
1183 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1184 else
1185 BUG();
1186}
1187