blob: 30c15622174fdbc59f3b94ab8cec331001b2a943 [file] [log] [blame]
David Teiglandb3b94fa2006-01-16 16:50:04 +00001/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
Steven Whitehouse3a8a9a12006-05-18 15:09:15 -04003 * Copyright (C) 2004-2006 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 Whitehouse5c676f62006-02-27 17:23:27 -050020#include <linux/gfs2_ondisk.h>
Fabio Massimo Di Nitto7d308592006-09-19 07:56:29 +020021#include <linux/lm_interface.h>
David Teiglandb3b94fa2006-01-16 16:50:04 +000022
23#include "gfs2.h"
Steven Whitehouse5c676f62006-02-27 17:23:27 -050024#include "incore.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000025#include "bmap.h"
26#include "glock.h"
27#include "inode.h"
David Teiglandb3b94fa2006-01-16 16:50:04 +000028#include "log.h"
29#include "meta_io.h"
30#include "ops_address.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"
Steven Whitehouse61a30dc2006-02-15 10:15:18 +000034#include "ops_file.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;
52 gfs2_trans_add_bh(ip->i_gl, bh, 0);
53 }
54}
55
David Teiglandb3b94fa2006-01-16 16:50:04 +000056/**
Steven Whitehouse4ff14672006-01-30 09:39:10 +000057 * gfs2_get_block - Fills in a buffer head with details about a block
David Teiglandb3b94fa2006-01-16 16:50:04 +000058 * @inode: The inode
59 * @lblock: The block number to look up
60 * @bh_result: The buffer head to return the result in
61 * @create: Non-zero if we may add block to the file
62 *
63 * Returns: errno
64 */
65
Steven Whitehouse4ff14672006-01-30 09:39:10 +000066int gfs2_get_block(struct inode *inode, sector_t lblock,
67 struct buffer_head *bh_result, int create)
David Teiglandb3b94fa2006-01-16 16:50:04 +000068{
Steven Whitehouse23591252006-10-13 17:25:45 -040069 return gfs2_block_map(inode, lblock, create, bh_result);
David Teiglandb3b94fa2006-01-16 16:50:04 +000070}
71
72/**
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040073 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
David Teiglandb3b94fa2006-01-16 16:50:04 +000074 * @inode: The inode
75 * @lblock: The block number to look up
76 * @bh_result: The buffer head to return the result in
77 * @create: Non-zero if we may add block to the file
78 *
79 * Returns: errno
80 */
81
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040082static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
83 struct buffer_head *bh_result, int create)
David Teiglandb3b94fa2006-01-16 16:50:04 +000084{
David Teiglandb3b94fa2006-01-16 16:50:04 +000085 int error;
86
Steven Whitehouse23591252006-10-13 17:25:45 -040087 error = gfs2_block_map(inode, lblock, 0, bh_result);
David Teiglandb3b94fa2006-01-16 16:50:04 +000088 if (error)
89 return error;
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040090 if (bh_result->b_blocknr == 0)
91 return -EIO;
Steven Whitehouse623d9352006-08-31 12:14:44 -040092 return 0;
93}
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040094
95static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
96 struct buffer_head *bh_result, int create)
97{
Steven Whitehouse23591252006-10-13 17:25:45 -040098 return gfs2_block_map(inode, lblock, 0, bh_result);
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -040099}
100
David Teiglandb3b94fa2006-01-16 16:50:04 +0000101/**
102 * gfs2_writepage - Write complete page
103 * @page: Page to write
104 *
105 * Returns: errno
106 *
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000107 * Some of this is copied from block_write_full_page() although we still
108 * call it to do most of the work.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000109 */
110
111static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
112{
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000113 struct inode *inode = page->mapping->host;
Steven Whitehousef4387142006-08-08 13:23:19 -0400114 struct gfs2_inode *ip = GFS2_I(inode);
115 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000116 loff_t i_size = i_size_read(inode);
117 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
118 unsigned offset;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000119 int error;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000120 int done_trans = 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000121
David Teiglandb3b94fa2006-01-16 16:50:04 +0000122 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
123 unlock_page(page);
124 return -EIO;
125 }
Steven Whitehouse5c676f62006-02-27 17:23:27 -0500126 if (current->journal_info)
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000127 goto out_ignore;
128
129 /* Is the page fully outside i_size? (truncate in progress) */
130 offset = i_size & (PAGE_CACHE_SIZE-1);
Steven Whitehoused2d7b8a2006-05-02 12:09:42 -0400131 if (page->index > end_index || (page->index == end_index && !offset)) {
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000132 page->mapping->a_ops->invalidatepage(page, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000133 unlock_page(page);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000134 return 0; /* don't care */
135 }
136
137 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
138 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
139 if (error)
140 goto out_ignore;
Steven Whitehousef4387142006-08-08 13:23:19 -0400141 if (!page_has_buffers(page)) {
142 create_empty_buffers(page, inode->i_sb->s_blocksize,
143 (1 << BH_Dirty)|(1 << BH_Uptodate));
144 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000145 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
146 done_trans = 1;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000147 }
Steven Whitehouse7a6bbac2006-09-18 17:18:23 -0400148 error = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000149 if (done_trans)
150 gfs2_trans_end(sdp);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000151 gfs2_meta_cache_flush(ip);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000152 return error;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000153
154out_ignore:
155 redirty_page_for_writepage(wbc, page);
156 unlock_page(page);
157 return 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000158}
159
160/**
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000161 * gfs2_writepages - Write a bunch of dirty pages back to disk
162 * @mapping: The mapping to write
163 * @wbc: Write-back control
164 *
165 * For journaled files and/or ordered writes this just falls back to the
166 * kernel's default writepages path for now. We will probably want to change
167 * that eventually (i.e. when we look at allocate on flush).
168 *
169 * For the data=writeback case though we can already ignore buffer heads
170 * and write whole extents at once. This is a big reduction in the
171 * number of I/O requests we send and the bmap calls we make in this case.
172 */
Adrian Bunka2cf8222007-02-06 23:12:49 +0100173static int gfs2_writepages(struct address_space *mapping,
174 struct writeback_control *wbc)
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000175{
176 struct inode *inode = mapping->host;
177 struct gfs2_inode *ip = GFS2_I(inode);
178 struct gfs2_sbd *sdp = GFS2_SB(inode);
179
180 if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip))
181 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
182
183 return generic_writepages(mapping, wbc);
184}
185
186/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000187 * stuffed_readpage - Fill in a Linux page with stuffed file data
188 * @ip: the inode
189 * @page: the page
190 *
191 * Returns: errno
192 */
193
194static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
195{
196 struct buffer_head *dibh;
197 void *kaddr;
198 int error;
199
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100200 /*
201 * Due to the order of unstuffing files and ->nopage(), we can be
202 * asked for a zero page in the case of a stuffed file being extended,
203 * so we need to supply one here. It doesn't happen often.
204 */
205 if (unlikely(page->index)) {
206 kaddr = kmap_atomic(page, KM_USER0);
207 memset(kaddr, 0, PAGE_CACHE_SIZE);
208 kunmap_atomic(kaddr, KM_USER0);
209 flush_dcache_page(page);
210 SetPageUptodate(page);
211 return 0;
212 }
Steven Whitehousefd88de562006-05-05 16:59:11 -0400213
David Teiglandb3b94fa2006-01-16 16:50:04 +0000214 error = gfs2_meta_inode_buffer(ip, &dibh);
215 if (error)
216 return error;
217
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000218 kaddr = kmap_atomic(page, KM_USER0);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400219 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
David Teiglandb3b94fa2006-01-16 16:50:04 +0000220 ip->i_di.di_size);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400221 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
Russell Cattelanc312c4f2006-10-12 09:23:41 -0400222 kunmap_atomic(kaddr, KM_USER0);
Steven Whitehousebf126ae2007-04-20 09:18:30 +0100223 flush_dcache_page(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000224 brelse(dibh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000225 SetPageUptodate(page);
226
227 return 0;
228}
229
David Teiglandb3b94fa2006-01-16 16:50:04 +0000230
231/**
David Teiglandb3b94fa2006-01-16 16:50:04 +0000232 * gfs2_readpage - readpage with locking
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000233 * @file: The file to read a page for. N.B. This may be NULL if we are
234 * reading an internal file.
David Teiglandb3b94fa2006-01-16 16:50:04 +0000235 * @page: The page to read
236 *
237 * Returns: errno
238 */
239
240static int gfs2_readpage(struct file *file, struct page *page)
241{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400242 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
243 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
Russell Cattelandc41aee2006-09-18 17:26:48 -0400244 struct gfs2_file *gf = NULL;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000245 struct gfs2_holder gh;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000246 int error;
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400247 int do_unlock = 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000248
Steven Whitehousedd538c82006-09-04 14:53:30 -0400249 if (likely(file != &gfs2_internal_file_sentinel)) {
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400250 if (file) {
Russell Cattelandc41aee2006-09-18 17:26:48 -0400251 gf = file->private_data;
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400252 if (test_bit(GFF_EXLOCK, &gf->f_flags))
Russell Cattelandc41aee2006-09-18 17:26:48 -0400253 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400254 goto skip_lock;
255 }
Steven Whitehouse2ca99502006-11-08 10:26:54 -0500256 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400257 do_unlock = 1;
Steven Whitehousedcd24792006-11-16 11:08:16 -0500258 error = gfs2_glock_nq_atime(&gh);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400259 if (unlikely(error))
Steven Whitehouse61a30dc2006-02-15 10:15:18 +0000260 goto out_unlock;
261 }
David Teiglandb3b94fa2006-01-16 16:50:04 +0000262
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400263skip_lock:
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000264 if (gfs2_is_stuffed(ip)) {
Steven Whitehousefd88de562006-05-05 16:59:11 -0400265 error = stuffed_readpage(ip, page);
266 unlock_page(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000267 } else
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000268 error = mpage_readpage(page, gfs2_get_block);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000269
270 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
271 error = -EIO;
272
Steven Whitehouse07903c02006-09-18 17:30:05 -0400273 if (do_unlock) {
Steven Whitehouse61a30dc2006-02-15 10:15:18 +0000274 gfs2_glock_dq_m(1, &gh);
275 gfs2_holder_uninit(&gh);
276 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000277out:
David Teiglandb3b94fa2006-01-16 16:50:04 +0000278 return error;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000279out_unlock:
280 unlock_page(page);
Josef Whitera13cbe32007-02-23 12:49:51 -0500281 if (error == GLR_TRYFAILED) {
282 error = AOP_TRUNCATED_PAGE;
283 yield();
284 }
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400285 if (do_unlock)
Steven Whitehousefd88de562006-05-05 16:59:11 -0400286 gfs2_holder_uninit(&gh);
287 goto out;
288}
289
Steven Whitehousefd88de562006-05-05 16:59:11 -0400290/**
291 * gfs2_readpages - Read a bunch of pages at once
292 *
293 * Some notes:
294 * 1. This is only for readahead, so we can simply ignore any things
295 * which are slightly inconvenient (such as locking conflicts between
296 * the page lock and the glock) and return having done no I/O. Its
297 * obviously not something we'd want to do on too regular a basis.
298 * Any I/O we ignore at this time will be done via readpage later.
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500299 * 2. We don't handle stuffed files here we let readpage do the honours.
Steven Whitehousefd88de562006-05-05 16:59:11 -0400300 * 3. mpage_readpages() does most of the heavy lifting in the common case.
301 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
302 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
303 * well as read-ahead.
304 */
305static int gfs2_readpages(struct file *file, struct address_space *mapping,
306 struct list_head *pages, unsigned nr_pages)
307{
308 struct inode *inode = mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400309 struct gfs2_inode *ip = GFS2_I(inode);
310 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400311 struct gfs2_holder gh;
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500312 int ret = 0;
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400313 int do_unlock = 0;
Steven Whitehousefd88de562006-05-05 16:59:11 -0400314
Steven Whitehousedd538c82006-09-04 14:53:30 -0400315 if (likely(file != &gfs2_internal_file_sentinel)) {
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400316 if (file) {
317 struct gfs2_file *gf = file->private_data;
318 if (test_bit(GFF_EXLOCK, &gf->f_flags))
319 goto skip_lock;
320 }
Steven Whitehousefd88de562006-05-05 16:59:11 -0400321 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
Steven Whitehouse2ca99502006-11-08 10:26:54 -0500322 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400323 do_unlock = 1;
Steven Whitehousedcd24792006-11-16 11:08:16 -0500324 ret = gfs2_glock_nq_atime(&gh);
Steven Whitehouse907b9bc2006-09-25 09:26:04 -0400325 if (ret == GLR_TRYFAILED)
Steven Whitehousefd88de562006-05-05 16:59:11 -0400326 goto out_noerror;
327 if (unlikely(ret))
328 goto out_unlock;
329 }
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400330skip_lock:
Steven Whitehousee1d5b182006-12-15 16:49:51 -0500331 if (!gfs2_is_stuffed(ip))
Steven Whitehousefd88de562006-05-05 16:59:11 -0400332 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
Steven Whitehousefd88de562006-05-05 16:59:11 -0400333
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400334 if (do_unlock) {
Steven Whitehousefd88de562006-05-05 16:59:11 -0400335 gfs2_glock_dq_m(1, &gh);
336 gfs2_holder_uninit(&gh);
337 }
338out:
339 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
340 ret = -EIO;
341 return ret;
342out_noerror:
343 ret = 0;
344out_unlock:
Steven Whitehouse59a1cc62006-08-04 15:41:22 -0400345 if (do_unlock)
Steven Whitehousefd88de562006-05-05 16:59:11 -0400346 gfs2_holder_uninit(&gh);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000347 goto out;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000348}
349
350/**
351 * gfs2_prepare_write - Prepare to write a page to a file
352 * @file: The file to write to
353 * @page: The page which is to be prepared for writing
354 * @from: From (byte range within page)
355 * @to: To (byte range within page)
356 *
357 * Returns: errno
358 */
359
360static int gfs2_prepare_write(struct file *file, struct page *page,
361 unsigned from, unsigned to)
362{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400363 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
364 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000365 unsigned int data_blocks, ind_blocks, rblocks;
366 int alloc_required;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000367 int error = 0;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000368 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
369 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
370 struct gfs2_alloc *al;
Russell Cattelan52ae7b72006-10-09 12:11:54 -0500371 unsigned int write_len = to - from;
372
David Teiglandb3b94fa2006-01-16 16:50:04 +0000373
Steven Whitehouse2ca99502006-11-08 10:26:54 -0500374 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
Steven Whitehousedcd24792006-11-16 11:08:16 -0500375 error = gfs2_glock_nq_atime(&ip->i_gh);
Steven Whitehouse2ca99502006-11-08 10:26:54 -0500376 if (unlikely(error)) {
Steven Whitehouse2d72e712007-02-07 10:25:59 -0500377 if (error == GLR_TRYFAILED) {
378 unlock_page(page);
Steven Whitehouse2ca99502006-11-08 10:26:54 -0500379 error = AOP_TRUNCATED_PAGE;
Josef Whitera13cbe32007-02-23 12:49:51 -0500380 yield();
Steven Whitehouse2d72e712007-02-07 10:25:59 -0500381 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000382 goto out_uninit;
Steven Whitehouse2ca99502006-11-08 10:26:54 -0500383 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000384
Russell Cattelan52ae7b72006-10-09 12:11:54 -0500385 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000386
Russell Cattelan52ae7b72006-10-09 12:11:54 -0500387 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000388 if (error)
389 goto out_unlock;
390
391
Steven Whitehousef5c54802006-10-10 13:45:15 -0400392 ip->i_alloc.al_requested = 0;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000393 if (alloc_required) {
394 al = gfs2_alloc_get(ip);
395
396 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
397 if (error)
398 goto out_alloc_put;
399
Steven Whitehouse2933f922006-11-01 13:23:29 -0500400 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000401 if (error)
402 goto out_qunlock;
403
404 al->al_requested = data_blocks + ind_blocks;
405 error = gfs2_inplace_reserve(ip);
406 if (error)
407 goto out_qunlock;
408 }
409
410 rblocks = RES_DINODE + ind_blocks;
411 if (gfs2_is_jdata(ip))
412 rblocks += data_blocks ? data_blocks : 1;
413 if (ind_blocks || data_blocks)
414 rblocks += RES_STATFS + RES_QUOTA;
415
416 error = gfs2_trans_begin(sdp, rblocks, 0);
417 if (error)
418 goto out;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000419
420 if (gfs2_is_stuffed(ip)) {
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000421 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
Steven Whitehousef25ef0c2006-07-26 10:51:20 -0400422 error = gfs2_unstuff_dinode(ip, page);
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000423 if (error == 0)
424 goto prepare_write;
425 } else if (!PageUptodate(page))
David Teiglandb3b94fa2006-01-16 16:50:04 +0000426 error = stuffed_readpage(ip, page);
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000427 goto out;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000428 }
429
Steven Whitehouse5c4e9e02006-02-15 12:26:19 +0000430prepare_write:
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000431 error = block_prepare_write(page, from, to, gfs2_get_block);
432
433out:
434 if (error) {
435 gfs2_trans_end(sdp);
436 if (alloc_required) {
437 gfs2_inplace_release(ip);
438out_qunlock:
439 gfs2_quota_unlock(ip);
440out_alloc_put:
441 gfs2_alloc_put(ip);
442 }
443out_unlock:
444 gfs2_glock_dq_m(1, &ip->i_gh);
445out_uninit:
446 gfs2_holder_uninit(&ip->i_gh);
447 }
David Teiglandb3b94fa2006-01-16 16:50:04 +0000448
449 return error;
450}
451
452/**
453 * gfs2_commit_write - Commit write to a file
454 * @file: The file to write to
455 * @page: The page containing the data
456 * @from: From (byte range within page)
457 * @to: To (byte range within page)
458 *
459 * Returns: errno
460 */
461
462static int gfs2_commit_write(struct file *file, struct page *page,
463 unsigned from, unsigned to)
464{
465 struct inode *inode = page->mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400466 struct gfs2_inode *ip = GFS2_I(inode);
467 struct gfs2_sbd *sdp = GFS2_SB(inode);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000468 int error = -EOPNOTSUPP;
469 struct buffer_head *dibh;
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400470 struct gfs2_alloc *al = &ip->i_alloc;
471 struct gfs2_dinode *di;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000472
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000473 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
474 goto fail_nounlock;
475
476 error = gfs2_meta_inode_buffer(ip, &dibh);
477 if (error)
478 goto fail_endtrans;
479
480 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400481 di = (struct gfs2_dinode *)dibh->b_data;
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000482
David Teiglandb3b94fa2006-01-16 16:50:04 +0000483 if (gfs2_is_stuffed(ip)) {
Steven Whitehousecd915492006-09-04 12:49:07 -0400484 u64 file_size;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000485 void *kaddr;
486
Steven Whitehousecd915492006-09-04 12:49:07 -0400487 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000488
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000489 kaddr = kmap_atomic(page, KM_USER0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000490 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
Steven Whitehouse0bd59962006-09-04 14:59:35 -0400491 kaddr + from, to - from);
Russell Cattelanc312c4f2006-10-12 09:23:41 -0400492 kunmap_atomic(kaddr, KM_USER0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000493
494 SetPageUptodate(page);
495
Steven Whitehouseae619322006-11-22 11:28:47 -0500496 if (inode->i_size < file_size) {
David Teiglandb3b94fa2006-01-16 16:50:04 +0000497 i_size_write(inode, file_size);
Steven Whitehouseae619322006-11-22 11:28:47 -0500498 mark_inode_dirty(inode);
499 }
David Teiglandb3b94fa2006-01-16 16:50:04 +0000500 } else {
Steven Whitehouse568f4c92006-02-27 12:00:42 -0500501 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
502 gfs2_is_jdata(ip))
Steven Whitehouse257f9b42006-01-31 10:00:25 +0000503 gfs2_page_add_databufs(ip, page, from, to);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000504 error = generic_commit_write(file, page, from, to);
505 if (error)
506 goto fail;
507 }
508
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400509 if (ip->i_di.di_size < inode->i_size) {
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000510 ip->i_di.di_size = inode->i_size;
Steven Whitehouse48516ce2006-10-02 12:39:19 -0400511 di->di_size = cpu_to_be64(inode->i_size);
512 }
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000513
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000514 brelse(dibh);
515 gfs2_trans_end(sdp);
516 if (al->al_requested) {
517 gfs2_inplace_release(ip);
518 gfs2_quota_unlock(ip);
519 gfs2_alloc_put(ip);
520 }
Josef Whiter1de91392007-03-12 16:55:07 -0500521 unlock_page(page);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000522 gfs2_glock_dq_m(1, &ip->i_gh);
Josef Whiter1de91392007-03-12 16:55:07 -0500523 lock_page(page);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000524 gfs2_holder_uninit(&ip->i_gh);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000525 return 0;
526
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000527fail:
528 brelse(dibh);
529fail_endtrans:
530 gfs2_trans_end(sdp);
531 if (al->al_requested) {
532 gfs2_inplace_release(ip);
533 gfs2_quota_unlock(ip);
534 gfs2_alloc_put(ip);
535 }
Josef Whiter1de91392007-03-12 16:55:07 -0500536 unlock_page(page);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000537 gfs2_glock_dq_m(1, &ip->i_gh);
Josef Whiter1de91392007-03-12 16:55:07 -0500538 lock_page(page);
Steven Whitehouse18ec7d52006-02-08 11:50:51 +0000539 gfs2_holder_uninit(&ip->i_gh);
540fail_nounlock:
David Teiglandb3b94fa2006-01-16 16:50:04 +0000541 ClearPageUptodate(page);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000542 return error;
543}
544
545/**
546 * gfs2_bmap - Block map function
547 * @mapping: Address space info
548 * @lblock: The block to map
549 *
550 * Returns: The disk address for the block or 0 on hole or error
551 */
552
553static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
554{
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400555 struct gfs2_inode *ip = GFS2_I(mapping->host);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000556 struct gfs2_holder i_gh;
557 sector_t dblock = 0;
558 int error;
559
David Teiglandb3b94fa2006-01-16 16:50:04 +0000560 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
561 if (error)
562 return 0;
563
564 if (!gfs2_is_stuffed(ip))
Steven Whitehouse4ff14672006-01-30 09:39:10 +0000565 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000566
567 gfs2_glock_dq_uninit(&i_gh);
568
569 return dblock;
570}
571
572static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
573{
Steven Whitehouse64fb4eb2006-01-18 13:14:40 +0000574 struct gfs2_bufdata *bd;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000575
576 gfs2_log_lock(sdp);
Steven Whitehouse5c676f62006-02-27 17:23:27 -0500577 bd = bh->b_private;
Steven Whitehouse64fb4eb2006-01-18 13:14:40 +0000578 if (bd) {
579 bd->bd_bh = NULL;
Steven Whitehouse5c676f62006-02-27 17:23:27 -0500580 bh->b_private = NULL;
Steven Whitehouse15d00c02006-08-18 15:51:09 -0400581 }
582 gfs2_log_unlock(sdp);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000583
584 lock_buffer(bh);
585 clear_buffer_dirty(bh);
586 bh->b_bdev = NULL;
587 clear_buffer_mapped(bh);
588 clear_buffer_req(bh);
589 clear_buffer_new(bh);
590 clear_buffer_delay(bh);
591 unlock_buffer(bh);
592}
593
Steven Whitehouse8628de02006-03-31 16:48:41 -0500594static void gfs2_invalidatepage(struct page *page, unsigned long offset)
David Teiglandb3b94fa2006-01-16 16:50:04 +0000595{
Steven Whitehouse15d00c02006-08-18 15:51:09 -0400596 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000597 struct buffer_head *head, *bh, *next;
598 unsigned int curr_off = 0;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000599
600 BUG_ON(!PageLocked(page));
601 if (!page_has_buffers(page))
Steven Whitehouse8628de02006-03-31 16:48:41 -0500602 return;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000603
604 bh = head = page_buffers(page);
605 do {
606 unsigned int next_off = curr_off + bh->b_size;
607 next = bh->b_this_page;
608
609 if (offset <= curr_off)
610 discard_buffer(sdp, bh);
611
612 curr_off = next_off;
613 bh = next;
614 } while (bh != head);
615
616 if (!offset)
Steven Whitehouse8628de02006-03-31 16:48:41 -0500617 try_to_release_page(page, 0);
David Teiglandb3b94fa2006-01-16 16:50:04 +0000618
Steven Whitehouse8628de02006-03-31 16:48:41 -0500619 return;
David Teiglandb3b94fa2006-01-16 16:50:04 +0000620}
621
Steven Whitehousec7b33832006-12-14 18:24:26 +0000622/**
623 * gfs2_ok_for_dio - check that dio is valid on this file
624 * @ip: The inode
625 * @rw: READ or WRITE
626 * @offset: The offset at which we are reading or writing
627 *
628 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
629 * 1 (to accept the i/o request)
630 */
631static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
632{
633 /*
634 * Should we return an error here? I can't see that O_DIRECT for
635 * a journaled file makes any sense. For now we'll silently fall
636 * back to buffered I/O, likewise we do the same for stuffed
637 * files since they are (a) small and (b) unaligned.
638 */
639 if (gfs2_is_jdata(ip))
640 return 0;
641
642 if (gfs2_is_stuffed(ip))
643 return 0;
644
645 if (offset > i_size_read(&ip->i_inode))
646 return 0;
647 return 1;
648}
649
650
651
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -0400652static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
653 const struct iovec *iov, loff_t offset,
654 unsigned long nr_segs)
Steven Whitehoused1665e42006-02-14 11:54:42 +0000655{
656 struct file *file = iocb->ki_filp;
657 struct inode *inode = file->f_mapping->host;
Steven Whitehousefeaa7bb2006-06-14 15:32:57 -0400658 struct gfs2_inode *ip = GFS2_I(inode);
Steven Whitehoused1665e42006-02-14 11:54:42 +0000659 struct gfs2_holder gh;
660 int rv;
661
662 /*
Steven Whitehousec7b33832006-12-14 18:24:26 +0000663 * Deferred lock, even if its a write, since we do no allocation
664 * on this path. All we need change is atime, and this lock mode
665 * ensures that other nodes have flushed their buffered read caches
666 * (i.e. their page cache entries for this inode). We do not,
667 * unfortunately have the option of only flushing a range like
668 * the VFS does.
Steven Whitehoused1665e42006-02-14 11:54:42 +0000669 */
Steven Whitehousec7b33832006-12-14 18:24:26 +0000670 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
Steven Whitehousedcd24792006-11-16 11:08:16 -0500671 rv = gfs2_glock_nq_atime(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +0000672 if (rv)
Steven Whitehousec7b33832006-12-14 18:24:26 +0000673 return rv;
674 rv = gfs2_ok_for_dio(ip, rw, offset);
675 if (rv != 1)
676 goto out; /* dio not valid, fall back to buffered i/o */
Steven Whitehoused1665e42006-02-14 11:54:42 +0000677
Steven Whitehousec7b33832006-12-14 18:24:26 +0000678 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
679 iov, offset, nr_segs,
680 gfs2_get_block_direct, NULL);
Steven Whitehoused1665e42006-02-14 11:54:42 +0000681out:
682 gfs2_glock_dq_m(1, &gh);
683 gfs2_holder_uninit(&gh);
Steven Whitehoused1665e42006-02-14 11:54:42 +0000684 return rv;
685}
686
687/**
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400688 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
689 * @bh: the buffer we're stuck on
690 *
691 */
692
693static void stuck_releasepage(struct buffer_head *bh)
694{
695 struct inode *inode = bh->b_page->mapping->host;
696 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
697 struct gfs2_bufdata *bd = bh->b_private;
698 struct gfs2_glock *gl;
Steven Whitehouse166afcc2006-08-24 15:59:40 -0400699static unsigned limit = 0;
700
Steven Whitehouse0bd59962006-09-04 14:59:35 -0400701 if (limit > 3)
Steven Whitehouse166afcc2006-08-24 15:59:40 -0400702 return;
Steven Whitehouse0bd59962006-09-04 14:59:35 -0400703 limit++;
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400704
705 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
706 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
707 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
708 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
709 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
710
711 if (!bd)
712 return;
713
714 gl = bd->bd_gl;
715
Steven Whitehouse907b9bc2006-09-25 09:26:04 -0400716 fs_warn(sdp, "gl = (%u, %llu)\n",
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400717 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
718
719 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
720 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
721 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
722
723 if (gl->gl_ops == &gfs2_inode_glops) {
724 struct gfs2_inode *ip = gl->gl_object;
725 unsigned int x;
726
727 if (!ip)
728 return;
729
730 fs_warn(sdp, "ip = %llu %llu\n",
731 (unsigned long long)ip->i_num.no_formal_ino,
732 (unsigned long long)ip->i_num.no_addr);
733
734 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
735 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
736 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
737 }
738}
739
740/**
Steven Whitehouse623d9352006-08-31 12:14:44 -0400741 * gfs2_releasepage - free the metadata associated with a page
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400742 * @page: the page that's being released
743 * @gfp_mask: passed from Linux VFS, ignored by us
744 *
745 * Call try_to_free_buffers() if the buffers in this page can be
746 * released.
747 *
748 * Returns: 0
749 */
750
751int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
752{
753 struct inode *aspace = page->mapping->host;
754 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
755 struct buffer_head *bh, *head;
756 struct gfs2_bufdata *bd;
Steven Whitehouse166afcc2006-08-24 15:59:40 -0400757 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400758
759 if (!page_has_buffers(page))
760 goto out;
761
762 head = bh = page_buffers(page);
763 do {
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400764 while (atomic_read(&bh->b_count)) {
Steven Whitehouse166afcc2006-08-24 15:59:40 -0400765 if (!atomic_read(&aspace->i_writecount))
766 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400767
Russell Cattelan61057c62006-11-09 11:42:33 -0500768 if (!(gfp_mask & __GFP_WAIT))
769 return 0;
770
Steven Whitehouse166afcc2006-08-24 15:59:40 -0400771 if (time_after_eq(jiffies, t)) {
772 stuck_releasepage(bh);
773 /* should we withdraw here? */
774 return 0;
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400775 }
776
Steven Whitehouse166afcc2006-08-24 15:59:40 -0400777 yield();
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400778 }
779
780 gfs2_assert_warn(sdp, !buffer_pinned(bh));
Steven Whitehouse623d9352006-08-31 12:14:44 -0400781 gfs2_assert_warn(sdp, !buffer_dirty(bh));
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400782
Steven Whitehouse623d9352006-08-31 12:14:44 -0400783 gfs2_log_lock(sdp);
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400784 bd = bh->b_private;
785 if (bd) {
786 gfs2_assert_warn(sdp, bd->bd_bh == bh);
787 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400788 gfs2_assert_warn(sdp, !bd->bd_ail);
Steven Whitehouse623d9352006-08-31 12:14:44 -0400789 bd->bd_bh = NULL;
790 if (!list_empty(&bd->bd_le.le_list))
791 bd = NULL;
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400792 bh->b_private = NULL;
793 }
Steven Whitehouse623d9352006-08-31 12:14:44 -0400794 gfs2_log_unlock(sdp);
795 if (bd)
796 kmem_cache_free(gfs2_bufdata_cachep, bd);
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400797
798 bh = bh->b_this_page;
Steven Whitehouse166afcc2006-08-24 15:59:40 -0400799 } while (bh != head);
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400800
Steven Whitehousea9e5f4d2006-07-25 17:24:12 -0400801out:
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400802 return try_to_free_buffers(page);
803}
804
Steven Whitehouse66de0452006-07-03 13:37:30 -0400805const struct address_space_operations gfs2_file_aops = {
David Teiglandb3b94fa2006-01-16 16:50:04 +0000806 .writepage = gfs2_writepage,
Steven Whitehousea8d638e2007-01-15 13:52:17 +0000807 .writepages = gfs2_writepages,
David Teiglandb3b94fa2006-01-16 16:50:04 +0000808 .readpage = gfs2_readpage,
Steven Whitehousefd88de562006-05-05 16:59:11 -0400809 .readpages = gfs2_readpages,
David Teiglandb3b94fa2006-01-16 16:50:04 +0000810 .sync_page = block_sync_page,
811 .prepare_write = gfs2_prepare_write,
812 .commit_write = gfs2_commit_write,
813 .bmap = gfs2_bmap,
814 .invalidatepage = gfs2_invalidatepage,
Steven Whitehouse4340fe62006-07-11 09:46:33 -0400815 .releasepage = gfs2_releasepage,
David Teiglandb3b94fa2006-01-16 16:50:04 +0000816 .direct_IO = gfs2_direct_IO,
817};
818