blob: 7b8da941526749c9b4a566bf1d3fe7b76f417dd8 [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);
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100272 int i;
273 int ret;
274
Abhijith Das20b95bf2008-03-06 17:43:52 -0600275 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100276 if (ret < 0)
277 return ret;
278
279 for(i = 0; i < nr_pages; i++) {
280 struct page *page = pvec->pages[i];
281
282 lock_page(page);
283
284 if (unlikely(page->mapping != mapping)) {
285 unlock_page(page);
286 continue;
287 }
288
289 if (!wbc->range_cyclic && page->index > end) {
290 ret = 1;
291 unlock_page(page);
292 continue;
293 }
294
295 if (wbc->sync_mode != WB_SYNC_NONE)
296 wait_on_page_writeback(page);
297
298 if (PageWriteback(page) ||
299 !clear_page_dirty_for_io(page)) {
300 unlock_page(page);
301 continue;
302 }
303
304 /* Is the page fully outside i_size? (truncate in progress) */
305 if (page->index > end_index || (page->index == end_index && !offset)) {
306 page->mapping->a_ops->invalidatepage(page, 0);
307 unlock_page(page);
308 continue;
309 }
310
311 ret = __gfs2_jdata_writepage(page, wbc);
312
313 if (ret || (--(wbc->nr_to_write) <= 0))
314 ret = 1;
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100315 }
316 gfs2_trans_end(sdp);
317 return ret;
318}
319
320/**
321 * gfs2_write_cache_jdata - Like write_cache_pages but different
322 * @mapping: The mapping to write
323 * @wbc: The writeback control
324 * @writepage: The writepage function to call
325 * @data: The data to pass to writepage
326 *
327 * The reason that we use our own function here is that we need to
328 * start transactions before we grab page locks. This allows us
329 * to get the ordering right.
330 */
331
332static int gfs2_write_cache_jdata(struct address_space *mapping,
333 struct writeback_control *wbc)
334{
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100335 int ret = 0;
336 int done = 0;
337 struct pagevec pvec;
338 int nr_pages;
339 pgoff_t index;
340 pgoff_t end;
341 int scanned = 0;
342 int range_whole = 0;
343
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +0100344 pagevec_init(&pvec, 0);
345 if (wbc->range_cyclic) {
346 index = mapping->writeback_index; /* Start from prev offset */
347 end = -1;
348 } else {
349 index = wbc->range_start >> PAGE_CACHE_SHIFT;
350 end = wbc->range_end >> PAGE_CACHE_SHIFT;
351 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
352 range_whole = 1;
353 scanned = 1;
354 }
355
356retry:
357 while (!done && (index <= end) &&
358 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
359 PAGECACHE_TAG_DIRTY,
360 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
361 scanned = 1;
362 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
363 if (ret)
364 done = 1;
365 if (ret > 0)
366 ret = 0;
367
368 pagevec_release(&pvec);
369 cond_resched();
370 }
371
372 if (!scanned && !done) {
373 /*
374 * We hit the last page and there is more work to be done: wrap
375 * back to the start of the file
376 */
377 scanned = 1;
378 index = 0;
379 goto retry;
380 }
381
382 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
383 mapping->writeback_index = index;
384 return ret;
385}
386
387
388/**
389 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
390 * @mapping: The mapping to write
391 * @wbc: The writeback control
392 *
393 */
394
395static int gfs2_jdata_writepages(struct address_space *mapping,
396 struct writeback_control *wbc)
397{
398 struct gfs2_inode *ip = GFS2_I(mapping->host);
399 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
400 int ret;
401
402 ret = gfs2_write_cache_jdata(mapping, wbc);
403 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
404 gfs2_log_flush(sdp, ip->i_gl);
405 ret = gfs2_write_cache_jdata(mapping, wbc);
406 }
407 return ret;
408}
409
410/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000411 * stuffed_readpage - Fill in a Linux page with stuffed file data
412 * @ip: the inode
413 * @page: the page
414 *
415 * Returns: errno
416 */
417
418static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
419{
420 struct buffer_head *dibh;
421 void *kaddr;
422 int error;
423
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100424 /*
Nick Piggin3c18ddd2008-04-28 02:12:10 -0700425 * Due to the order of unstuffing files and ->fault(), we can be
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100426 * asked for a zero page in the case of a stuffed file being extended,
427 * so we need to supply one here. It doesn't happen often.
428 */
429 if (unlikely(page->index)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800430 zero_user(page, 0, PAGE_CACHE_SIZE);
Abhijith Das0a7ab792009-01-07 16:03:37 -0600431 SetPageUptodate(page);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100432 return 0;
433 }
Steven Whitehousefd88de562006-05-05 16:59:11 -0400434
David Teiglandb3b94fa2006-01-16 16:50:04 +0000435 error = gfs2_meta_inode_buffer(ip, &dibh);
436 if (error)
437 return error;
438
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000439 kaddr = kmap_atomic(page, KM_USER0);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400440 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
Steven Whitehousec9e98882008-11-04 09:47:33 +0000441 ip->i_disksize);
442 memset(kaddr + ip->i_disksize, 0, PAGE_CACHE_SIZE - ip->i_disksize);
Russell Cattelanc312c4f2006-10-12 09:23:41 -0400443 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100444 flush_dcache_page(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000445 brelse(dibh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000446 SetPageUptodate(page);
447
448 return 0;
449}
450
David Teiglandb3b94fa2006-01-16 16:50:04 +0000451
452/**
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100453 * __gfs2_readpage - readpage
454 * @file: The file to read a page for
David Teiglandb3b94fa2006-01-16 16:50:04 +0000455 * @page: The page to read
456 *
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100457 * This is the core of gfs2's readpage. Its used by the internal file
458 * reading code as in that case we already hold the glock. Also its
459 * called by gfs2_readpage() once the required lock has been granted.
460 *
461 */
462
463static int __gfs2_readpage(void *file, struct page *page)
464{
465 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
466 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
467 int error;
468
469 if (gfs2_is_stuffed(ip)) {
470 error = stuffed_readpage(ip, page);
471 unlock_page(page);
472 } else {
Bob Petersone9e1ef22007-12-10 14:13:27 -0600473 error = mpage_readpage(page, gfs2_block_map);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100474 }
475
476 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
477 return -EIO;
478
479 return error;
480}
481
482/**
483 * gfs2_readpage - read a page of a file
484 * @file: The file to read
485 * @page: The page of the file
486 *
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100487 * This deals with the locking required. We have to unlock and
488 * relock the page in order to get the locking in the right
489 * order.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000490 */
491
492static int gfs2_readpage(struct file *file, struct page *page)
493{
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100494 struct address_space *mapping = page->mapping;
495 struct gfs2_inode *ip = GFS2_I(mapping->host);
Steven Whitehouse6802e342008-05-21 17:03:22 +0100496 struct gfs2_holder gh;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000497 int error;
498
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100499 unlock_page(page);
Steven Whitehouse719ee342008-09-18 13:53:59 +0100500 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
501 error = gfs2_glock_nq(&gh);
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100502 if (unlikely(error))
Steven Whitehouse6802e342008-05-21 17:03:22 +0100503 goto out;
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100504 error = AOP_TRUNCATED_PAGE;
505 lock_page(page);
506 if (page->mapping == mapping && !PageUptodate(page))
507 error = __gfs2_readpage(file, page);
508 else
509 unlock_page(page);
Steven Whitehouse6802e342008-05-21 17:03:22 +0100510 gfs2_glock_dq(&gh);
Steven Whitehouse7afd88d2008-02-22 16:07:18 +0000511out:
Steven Whitehouse6802e342008-05-21 17:03:22 +0100512 gfs2_holder_uninit(&gh);
Steven Whitehouse01b7c7a2008-06-02 09:14:54 +0100513 if (error && error != AOP_TRUNCATED_PAGE)
514 lock_page(page);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100515 return error;
516}
517
518/**
519 * gfs2_internal_read - read an internal file
520 * @ip: The gfs2 inode
521 * @ra_state: The readahead state (or NULL for no readahead)
522 * @buf: The buffer to fill
523 * @pos: The file position
524 * @size: The amount to read
525 *
526 */
527
528int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
529 char *buf, loff_t *pos, unsigned size)
530{
531 struct address_space *mapping = ip->i_inode.i_mapping;
532 unsigned long index = *pos / PAGE_CACHE_SIZE;
533 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
534 unsigned copied = 0;
535 unsigned amt;
536 struct page *page;
537 void *p;
538
539 do {
540 amt = size - copied;
541 if (offset + size > PAGE_CACHE_SIZE)
542 amt = PAGE_CACHE_SIZE - offset;
543 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
544 if (IS_ERR(page))
545 return PTR_ERR(page);
546 p = kmap_atomic(page, KM_USER0);
547 memcpy(buf + copied, p + offset, amt);
548 kunmap_atomic(p, KM_USER0);
549 mark_page_accessed(page);
550 page_cache_release(page);
551 copied += amt;
552 index++;
553 offset = 0;
554 } while(copied < size);
555 (*pos) += size;
556 return size;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400557}
558
Steven Whitehousefd88de562006-05-05 16:59:11 -0400559/**
560 * gfs2_readpages - Read a bunch of pages at once
561 *
562 * Some notes:
563 * 1. This is only for readahead, so we can simply ignore any things
564 * which are slightly inconvenient (such as locking conflicts between
565 * the page lock and the glock) and return having done no I/O. Its
566 * obviously not something we'd want to do on too regular a basis.
567 * Any I/O we ignore at this time will be done via readpage later.
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500568 * 2. We don't handle stuffed files here we let readpage do the honours.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400569 * 3. mpage_readpages() does most of the heavy lifting in the common case.
Bob Petersone9e1ef22007-12-10 14:13:27 -0600570 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400571 */
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100572
Steven Whitehousefd88de562006-05-05 16:59:11 -0400573static int gfs2_readpages(struct file *file, struct address_space *mapping,
574 struct list_head *pages, unsigned nr_pages)
575{
576 struct inode *inode = mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400577 struct gfs2_inode *ip = GFS2_I(inode);
578 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400579 struct gfs2_holder gh;
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100580 int ret;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400581
Steven Whitehouse719ee342008-09-18 13:53:59 +0100582 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
583 ret = gfs2_glock_nq(&gh);
Steven Whitehouse51ff87b2007-10-15 14:42:35 +0100584 if (unlikely(ret))
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100585 goto out_uninit;
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500586 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600587 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
Steven Whitehouse3cc3f712007-10-15 15:40:33 +0100588 gfs2_glock_dq(&gh);
589out_uninit:
590 gfs2_holder_uninit(&gh);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400591 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
592 ret = -EIO;
593 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000594}
595
596/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700597 * gfs2_write_begin - Begin to write to a file
David Teiglandb3b94fa2006-01-16 16:50:04 +0000598 * @file: The file to write to
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700599 * @mapping: The mapping in which to write
600 * @pos: The file offset at which to start writing
601 * @len: Length of the write
602 * @flags: Various flags
603 * @pagep: Pointer to return the page
604 * @fsdata: Pointer to return fs data (unused by GFS2)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000605 *
606 * Returns: errno
607 */
608
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700609static int gfs2_write_begin(struct file *file, struct address_space *mapping,
610 loff_t pos, unsigned len, unsigned flags,
611 struct page **pagep, void **fsdata)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000612{
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700613 struct gfs2_inode *ip = GFS2_I(mapping->host);
614 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500615 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7ed122e2008-12-10 10:28:10 +0000616 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000617 int alloc_required;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000618 int error = 0;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000619 struct gfs2_alloc *al;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700620 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
621 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
622 unsigned to = from + len;
623 struct page *page;
Russell Cattelan52ae7b72006-10-09 12:11:54 -0500624
Steven Whitehouse719ee342008-09-18 13:53:59 +0100625 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
626 error = gfs2_glock_nq(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700627 if (unlikely(error))
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000628 goto out_uninit;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500629 if (&ip->i_inode == sdp->sd_rindex) {
630 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
631 GL_NOCACHE, &m_ip->i_gh);
632 if (unlikely(error)) {
633 gfs2_glock_dq(&ip->i_gh);
634 goto out_uninit;
635 }
636 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000637
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700638 error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
639 if (error)
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100640 goto out_unlock;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000641
Steven Whitehouse7ed122e2008-12-10 10:28:10 +0000642 if (alloc_required || gfs2_is_jdata(ip))
643 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
644
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000645 if (alloc_required) {
646 al = gfs2_alloc_get(ip);
Cyrill Gorcunov182fe5a2008-03-03 21:54:21 +0300647 if (!al) {
648 error = -ENOMEM;
649 goto out_unlock;
650 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000651
Steven Whitehoused82661d2008-03-10 15:34:50 +0000652 error = gfs2_quota_lock_check(ip);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000653 if (error)
654 goto out_alloc_put;
655
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000656 al->al_requested = data_blocks + ind_blocks;
657 error = gfs2_inplace_reserve(ip);
658 if (error)
659 goto out_qunlock;
660 }
661
662 rblocks = RES_DINODE + ind_blocks;
663 if (gfs2_is_jdata(ip))
664 rblocks += data_blocks ? data_blocks : 1;
665 if (ind_blocks || data_blocks)
666 rblocks += RES_STATFS + RES_QUOTA;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500667 if (&ip->i_inode == sdp->sd_rindex)
668 rblocks += 2 * RES_STATFS;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000669
Steven Whitehouse16615be2007-09-17 10:59:52 +0100670 error = gfs2_trans_begin(sdp, rblocks,
671 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000672 if (error)
Steven Whitehousea867bb22007-07-17 10:29:02 +0100673 goto out_trans_fail;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000674
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100675 error = -ENOMEM;
Steven Whitehousee4fefba2009-01-06 10:08:33 +0000676 flags |= AOP_FLAG_NOFS;
Nick Piggin54566b22009-01-04 12:00:53 -0800677 page = grab_cache_page_write_begin(mapping, index, flags);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100678 *pagep = page;
679 if (unlikely(!page))
680 goto out_endtrans;
681
David Teiglandb3b94fa2006-01-16 16:50:04 +0000682 if (gfs2_is_stuffed(ip)) {
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100683 error = 0;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700684 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
Steven Whitehousef25ef0c2006-07-26 10:51:20 -0400685 error = gfs2_unstuff_dinode(ip, page);
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000686 if (error == 0)
687 goto prepare_write;
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100688 } else if (!PageUptodate(page)) {
David Teiglandb3b94fa2006-01-16 16:50:04 +0000689 error = stuffed_readpage(ip, page);
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100690 }
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000691 goto out;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000692 }
693
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000694prepare_write:
Bob Petersone9e1ef22007-12-10 14:13:27 -0600695 error = block_prepare_write(page, from, to, gfs2_block_map);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000696out:
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100697 if (error == 0)
698 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000699
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100700 page_cache_release(page);
701 if (pos + len > ip->i_inode.i_size)
702 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
703out_endtrans:
704 gfs2_trans_end(sdp);
705out_trans_fail:
706 if (alloc_required) {
707 gfs2_inplace_release(ip);
708out_qunlock:
709 gfs2_quota_unlock(ip);
710out_alloc_put:
711 gfs2_alloc_put(ip);
712 }
713out_unlock:
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500714 if (&ip->i_inode == sdp->sd_rindex) {
715 gfs2_glock_dq(&m_ip->i_gh);
716 gfs2_holder_uninit(&m_ip->i_gh);
717 }
Steven Whitehousec41d4f02007-10-17 14:05:41 +0100718 gfs2_glock_dq(&ip->i_gh);
719out_uninit:
720 gfs2_holder_uninit(&ip->i_gh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000721 return error;
722}
723
724/**
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500725 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
726 * @inode: the rindex inode
727 */
728static void adjust_fs_space(struct inode *inode)
729{
730 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500731 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
732 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500733 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
734 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500735 struct buffer_head *m_bh, *l_bh;
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500736 u64 fs_total, new_free;
737
738 /* Total up the file system space, according to the latest rindex. */
739 fs_total = gfs2_ri_total(sdp);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500740 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
741 return;
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500742
743 spin_lock(&sdp->sd_statfs_spin);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500744 gfs2_statfs_change_in(m_sc, m_bh->b_data +
745 sizeof(struct gfs2_dinode));
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500746 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
747 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
748 else
749 new_free = 0;
750 spin_unlock(&sdp->sd_statfs_spin);
Robert Peterson6c532672007-05-10 16:54:38 -0500751 fs_warn(sdp, "File system extended by %llu blocks.\n",
752 (unsigned long long)new_free);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500753 gfs2_statfs_change(sdp, new_free, new_free, 0);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500754
755 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
756 goto out;
757 update_statfs(sdp, m_bh, l_bh);
758 brelse(l_bh);
759out:
760 brelse(m_bh);
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500761}
762
763/**
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700764 * gfs2_stuffed_write_end - Write end for stuffed files
765 * @inode: The inode
766 * @dibh: The buffer_head containing the on-disk inode
767 * @pos: The file position
768 * @len: The length of the write
769 * @copied: How much was actually copied by the VFS
770 * @page: The page
771 *
772 * This copies the data from the page into the inode block after
773 * the inode data structure itself.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000774 *
775 * Returns: errno
776 */
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700777static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
778 loff_t pos, unsigned len, unsigned copied,
779 struct page *page)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000780{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400781 struct gfs2_inode *ip = GFS2_I(inode);
782 struct gfs2_sbd *sdp = GFS2_SB(inode);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500783 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700784 u64 to = pos + copied;
785 void *kaddr;
786 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
787 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000788
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700789 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
790 kaddr = kmap_atomic(page, KM_USER0);
791 memcpy(buf + pos, kaddr + pos, copied);
792 memset(kaddr + pos + copied, 0, len - copied);
793 flush_dcache_page(page);
794 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000795
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700796 if (!PageUptodate(page))
David Teiglandb3b94fa2006-01-16 16:50:04 +0000797 SetPageUptodate(page);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700798 unlock_page(page);
799 page_cache_release(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000800
Abhijith Das7537d81a2009-05-12 11:16:20 -0500801 if (copied) {
802 if (inode->i_size < to) {
803 i_size_write(inode, to);
804 ip->i_disksize = inode->i_size;
805 }
806 gfs2_dinode_out(ip, di);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700807 mark_inode_dirty(inode);
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400808 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000809
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600810 if (inode == sdp->sd_rindex) {
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500811 adjust_fs_space(inode);
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600812 ip->i_gh.gh_flags |= GL_NOCACHE;
813 }
Robert Peterson7ae8fa82007-05-09 09:37:57 -0500814
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000815 brelse(dibh);
816 gfs2_trans_end(sdp);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500817 if (inode == sdp->sd_rindex) {
818 gfs2_glock_dq(&m_ip->i_gh);
819 gfs2_holder_uninit(&m_ip->i_gh);
820 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700821 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000822 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700823 return copied;
824}
David Teiglandb3b94fa2006-01-16 16:50:04 +0000825
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700826/**
827 * gfs2_write_end
828 * @file: The file to write to
829 * @mapping: The address space to write to
830 * @pos: The file position
831 * @len: The length of the data
832 * @copied:
833 * @page: The page that has been written
834 * @fsdata: The fsdata (unused in GFS2)
835 *
836 * The main write_end function for GFS2. We have a separate one for
837 * stuffed files as they are slightly different, otherwise we just
838 * put our locking around the VFS provided functions.
839 *
840 * Returns: errno
841 */
842
843static int gfs2_write_end(struct file *file, struct address_space *mapping,
844 loff_t pos, unsigned len, unsigned copied,
845 struct page *page, void *fsdata)
846{
847 struct inode *inode = page->mapping->host;
848 struct gfs2_inode *ip = GFS2_I(inode);
849 struct gfs2_sbd *sdp = GFS2_SB(inode);
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500850 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700851 struct buffer_head *dibh;
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000852 struct gfs2_alloc *al = ip->i_alloc;
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700853 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
854 unsigned int to = from + len;
855 int ret;
856
Steven Whitehouse7afd88d2008-02-22 16:07:18 +0000857 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700858
859 ret = gfs2_meta_inode_buffer(ip, &dibh);
860 if (unlikely(ret)) {
861 unlock_page(page);
862 page_cache_release(page);
863 goto failed;
864 }
865
866 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
867
868 if (gfs2_is_stuffed(ip))
869 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
870
Steven Whitehousebf36a712007-10-17 08:35:19 +0100871 if (!gfs2_is_writeback(ip))
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700872 gfs2_page_add_databufs(ip, page, from, to);
873
874 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Abhijith Das7537d81a2009-05-12 11:16:20 -0500875 if (ret > 0) {
876 if (inode->i_size > ip->i_disksize)
877 ip->i_disksize = inode->i_size;
878 gfs2_dinode_out(ip, dibh->b_data);
Steven Whitehouse9656b2c2008-01-08 08:14:30 +0000879 mark_inode_dirty(inode);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700880 }
881
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600882 if (inode == sdp->sd_rindex) {
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700883 adjust_fs_space(inode);
Benjamin Marzinski9ae3c6d2009-11-10 12:54:56 -0600884 ip->i_gh.gh_flags |= GL_NOCACHE;
885 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700886
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000887 brelse(dibh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000888 gfs2_trans_end(sdp);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700889failed:
Steven Whitehouse6dbd8222008-01-10 15:18:55 +0000890 if (al) {
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000891 gfs2_inplace_release(ip);
892 gfs2_quota_unlock(ip);
893 gfs2_alloc_put(ip);
894 }
Benjamin Marzinski1946f702009-06-25 15:09:51 -0500895 if (inode == sdp->sd_rindex) {
896 gfs2_glock_dq(&m_ip->i_gh);
897 gfs2_holder_uninit(&m_ip->i_gh);
898 }
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700899 gfs2_glock_dq(&ip->i_gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000900 gfs2_holder_uninit(&ip->i_gh);
Steven Whitehouse7765ec22007-10-16 01:25:07 -0700901 return ret;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000902}
903
904/**
Robert Peterson8fb68592007-06-12 11:24:36 -0500905 * gfs2_set_page_dirty - Page dirtying function
906 * @page: The page to dirty
907 *
908 * Returns: 1 if it dirtyed the page, or 0 otherwise
909 */
910
911static int gfs2_set_page_dirty(struct page *page)
912{
Steven Whitehouse55610932007-10-17 08:47:38 +0100913 SetPageChecked(page);
Robert Peterson8fb68592007-06-12 11:24:36 -0500914 return __set_page_dirty_buffers(page);
915}
916
917/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000918 * gfs2_bmap - Block map function
919 * @mapping: Address space info
920 * @lblock: The block to map
921 *
922 * Returns: The disk address for the block or 0 on hole or error
923 */
924
925static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
926{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400927 struct gfs2_inode *ip = GFS2_I(mapping->host);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000928 struct gfs2_holder i_gh;
929 sector_t dblock = 0;
930 int error;
931
David Teiglandb3b94fa2006-01-16 16:50:04 +0000932 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
933 if (error)
934 return 0;
935
936 if (!gfs2_is_stuffed(ip))
Bob Petersone9e1ef22007-12-10 14:13:27 -0600937 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000938
939 gfs2_glock_dq_uninit(&i_gh);
940
941 return dblock;
942}
943
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100944static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
945{
946 struct gfs2_bufdata *bd;
947
948 lock_buffer(bh);
949 gfs2_log_lock(sdp);
950 clear_buffer_dirty(bh);
951 bd = bh->b_private;
952 if (bd) {
Steven Whitehouse16615be2007-09-17 10:59:52 +0100953 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
954 list_del_init(&bd->bd_le.le_list);
955 else
956 gfs2_remove_from_journal(bh, current->journal_info, 0);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100957 }
958 bh->b_bdev = NULL;
959 clear_buffer_mapped(bh);
960 clear_buffer_req(bh);
961 clear_buffer_new(bh);
962 gfs2_log_unlock(sdp);
963 unlock_buffer(bh);
964}
965
Steven Whitehouse8628de02006-03-31 16:48:41 -0500966static void gfs2_invalidatepage(struct page *page, unsigned long offset)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000967{
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100968 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
969 struct buffer_head *bh, *head;
970 unsigned long pos = 0;
971
David Teiglandb3b94fa2006-01-16 16:50:04 +0000972 BUG_ON(!PageLocked(page));
Robert Peterson8fb68592007-06-12 11:24:36 -0500973 if (offset == 0)
974 ClearPageChecked(page);
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100975 if (!page_has_buffers(page))
976 goto out;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000977
Steven Whitehoused7b616e2007-09-02 10:48:13 +0100978 bh = head = page_buffers(page);
979 do {
980 if (offset <= pos)
981 gfs2_discard(sdp, bh);
982 pos += bh->b_size;
983 bh = bh->b_this_page;
984 } while (bh != head);
985out:
986 if (offset == 0)
987 try_to_release_page(page, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000988}
989
Steven Whitehousec7b33832006-12-14 18:24:26 +0000990/**
991 * gfs2_ok_for_dio - check that dio is valid on this file
992 * @ip: The inode
993 * @rw: READ or WRITE
994 * @offset: The offset at which we are reading or writing
995 *
996 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
997 * 1 (to accept the i/o request)
998 */
999static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
1000{
1001 /*
1002 * Should we return an error here? I can't see that O_DIRECT for
Steven Whitehouse55610932007-10-17 08:47:38 +01001003 * a stuffed file makes any sense. For now we'll silently fall
1004 * back to buffered I/O
Steven Whitehousec7b33832006-12-14 18:24:26 +00001005 */
Steven Whitehousec7b33832006-12-14 18:24:26 +00001006 if (gfs2_is_stuffed(ip))
1007 return 0;
1008
Bob Petersonacb57a32008-09-11 15:35:37 -04001009 if (offset >= i_size_read(&ip->i_inode))
Steven Whitehousec7b33832006-12-14 18:24:26 +00001010 return 0;
1011 return 1;
1012}
1013
1014
1015
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -04001016static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
1017 const struct iovec *iov, loff_t offset,
1018 unsigned long nr_segs)
Steven Whitehoused1665e42006-02-14 11:54:42 +00001019{
1020 struct file *file = iocb->ki_filp;
1021 struct inode *inode = file->f_mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -04001022 struct gfs2_inode *ip = GFS2_I(inode);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001023 struct gfs2_holder gh;
1024 int rv;
1025
1026 /*
Steven Whitehousec7b33832006-12-14 18:24:26 +00001027 * Deferred lock, even if its a write, since we do no allocation
1028 * on this path. All we need change is atime, and this lock mode
1029 * ensures that other nodes have flushed their buffered read caches
1030 * (i.e. their page cache entries for this inode). We do not,
1031 * unfortunately have the option of only flushing a range like
1032 * the VFS does.
Steven Whitehoused1665e42006-02-14 11:54:42 +00001033 */
Steven Whitehouse719ee342008-09-18 13:53:59 +01001034 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1035 rv = gfs2_glock_nq(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001036 if (rv)
Steven Whitehousec7b33832006-12-14 18:24:26 +00001037 return rv;
1038 rv = gfs2_ok_for_dio(ip, rw, offset);
1039 if (rv != 1)
1040 goto out; /* dio not valid, fall back to buffered i/o */
Steven Whitehoused1665e42006-02-14 11:54:42 +00001041
Steven Whitehousec7b33832006-12-14 18:24:26 +00001042 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
1043 iov, offset, nr_segs,
1044 gfs2_get_block_direct, NULL);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001045out:
1046 gfs2_glock_dq_m(1, &gh);
1047 gfs2_holder_uninit(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +00001048 return rv;
1049}
1050
1051/**
Steven Whitehouse623d9352006-08-31 12:14:44 -04001052 * gfs2_releasepage - free the metadata associated with a page
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001053 * @page: the page that's being released
1054 * @gfp_mask: passed from Linux VFS, ignored by us
1055 *
1056 * Call try_to_free_buffers() if the buffers in this page can be
1057 * released.
1058 *
1059 * Returns: 0
1060 */
1061
1062int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1063{
1064 struct inode *aspace = page->mapping->host;
1065 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
1066 struct buffer_head *bh, *head;
1067 struct gfs2_bufdata *bd;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001068
1069 if (!page_has_buffers(page))
Steven Whitehouse891ba6d2007-09-20 15:26:33 +01001070 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001071
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001072 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001073 head = bh = page_buffers(page);
1074 do {
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001075 if (atomic_read(&bh->b_count))
1076 goto cannot_release;
1077 bd = bh->b_private;
1078 if (bd && bd->bd_ail)
1079 goto cannot_release;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001080 gfs2_assert_warn(sdp, !buffer_pinned(bh));
Steven Whitehouse623d9352006-08-31 12:14:44 -04001081 gfs2_assert_warn(sdp, !buffer_dirty(bh));
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001082 bh = bh->b_this_page;
1083 } while(bh != head);
1084 gfs2_log_unlock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001085
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001086 head = bh = page_buffers(page);
1087 do {
Steven Whitehouse623d9352006-08-31 12:14:44 -04001088 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001089 bd = bh->b_private;
1090 if (bd) {
1091 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1092 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
Steven Whitehoused7b616e2007-09-02 10:48:13 +01001093 if (!list_empty(&bd->bd_le.le_list)) {
1094 if (!buffer_pinned(bh))
1095 list_del_init(&bd->bd_le.le_list);
1096 else
1097 bd = NULL;
1098 }
1099 if (bd)
1100 bd->bd_bh = NULL;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001101 bh->b_private = NULL;
1102 }
Steven Whitehouse623d9352006-08-31 12:14:44 -04001103 gfs2_log_unlock(sdp);
1104 if (bd)
1105 kmem_cache_free(gfs2_bufdata_cachep, bd);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001106
1107 bh = bh->b_this_page;
Steven Whitehouse166afcc2006-08-24 15:59:40 -04001108 } while (bh != head);
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001109
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001110 return try_to_free_buffers(page);
Steven Whitehousebb3b0e32007-08-16 16:03:57 +01001111cannot_release:
1112 gfs2_log_unlock(sdp);
1113 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001114}
1115
Steven Whitehouse55610932007-10-17 08:47:38 +01001116static const struct address_space_operations gfs2_writeback_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001117 .writepage = gfs2_writeback_writepage,
Steven Whitehouse55610932007-10-17 08:47:38 +01001118 .writepages = gfs2_writeback_writepages,
1119 .readpage = gfs2_readpage,
1120 .readpages = gfs2_readpages,
1121 .sync_page = block_sync_page,
1122 .write_begin = gfs2_write_begin,
1123 .write_end = gfs2_write_end,
1124 .bmap = gfs2_bmap,
1125 .invalidatepage = gfs2_invalidatepage,
1126 .releasepage = gfs2_releasepage,
1127 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001128 .migratepage = buffer_migrate_page,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001129 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001130 .error_remove_page = generic_error_remove_page,
Steven Whitehouse55610932007-10-17 08:47:38 +01001131};
1132
1133static const struct address_space_operations gfs2_ordered_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001134 .writepage = gfs2_ordered_writepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001135 .readpage = gfs2_readpage,
Steven Whitehousefd88de562006-05-05 16:59:11 -04001136 .readpages = gfs2_readpages,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001137 .sync_page = block_sync_page,
Steven Whitehouse7765ec22007-10-16 01:25:07 -07001138 .write_begin = gfs2_write_begin,
1139 .write_end = gfs2_write_end,
Robert Peterson8fb68592007-06-12 11:24:36 -05001140 .set_page_dirty = gfs2_set_page_dirty,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001141 .bmap = gfs2_bmap,
1142 .invalidatepage = gfs2_invalidatepage,
Steven Whitehouse4340fe62006-07-11 09:46:33 -04001143 .releasepage = gfs2_releasepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001144 .direct_IO = gfs2_direct_IO,
Steven Whitehousee5d9dc22008-01-03 11:31:38 +00001145 .migratepage = buffer_migrate_page,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001146 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001147 .error_remove_page = generic_error_remove_page,
David Teiglandb3b94fa2006-01-16 16:50:04 +00001148};
1149
Steven Whitehouse55610932007-10-17 08:47:38 +01001150static const struct address_space_operations gfs2_jdata_aops = {
Steven Whitehouse9ff8ec32007-09-28 13:49:05 +01001151 .writepage = gfs2_jdata_writepage,
Steven Whitehouseb8e7cbb2007-10-17 09:04:24 +01001152 .writepages = gfs2_jdata_writepages,
Steven Whitehouse55610932007-10-17 08:47:38 +01001153 .readpage = gfs2_readpage,
1154 .readpages = gfs2_readpages,
1155 .sync_page = block_sync_page,
1156 .write_begin = gfs2_write_begin,
1157 .write_end = gfs2_write_end,
1158 .set_page_dirty = gfs2_set_page_dirty,
1159 .bmap = gfs2_bmap,
1160 .invalidatepage = gfs2_invalidatepage,
1161 .releasepage = gfs2_releasepage,
Hisashi Hifumi229615d2009-03-03 11:45:20 +09001162 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001163 .error_remove_page = generic_error_remove_page,
Steven Whitehouse55610932007-10-17 08:47:38 +01001164};
1165
1166void gfs2_set_aops(struct inode *inode)
1167{
1168 struct gfs2_inode *ip = GFS2_I(inode);
1169
1170 if (gfs2_is_writeback(ip))
1171 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1172 else if (gfs2_is_ordered(ip))
1173 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1174 else if (gfs2_is_jdata(ip))
1175 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1176 else
1177 BUG();
1178}
1179