blob: d310d12a9adc481187c78fb65bffba53a16a75e7 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
20 */
21
22#include <linux/fs.h>
23#include <linux/slab.h>
24#include <linux/highmem.h>
25#include <linux/pagemap.h>
26#include <asm/byteorder.h>
Mark Fasheh9517bac2007-02-09 20:24:12 -080027#include <linux/swap.h>
Mark Fasheh6af67d82007-03-06 17:24:46 -080028#include <linux/pipe_fs_i.h>
Mark Fasheh628a24f2007-10-30 12:08:32 -070029#include <linux/mpage.h>
Jan Karaa90714c2008-10-09 19:38:40 +020030#include <linux/quotaops.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080031
Mark Fashehccd979b2005-12-15 14:31:24 -080032#include <cluster/masklog.h>
33
34#include "ocfs2.h"
35
36#include "alloc.h"
37#include "aops.h"
38#include "dlmglue.h"
39#include "extent_map.h"
40#include "file.h"
41#include "inode.h"
42#include "journal.h"
Mark Fasheh9517bac2007-02-09 20:24:12 -080043#include "suballoc.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080044#include "super.h"
45#include "symlink.h"
Tao Ma293b2f72009-08-25 08:02:48 +080046#include "refcounttree.h"
Tao Ma95581562011-02-22 21:33:59 +080047#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080048
49#include "buffer_head_io.h"
50
51static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
52 struct buffer_head *bh_result, int create)
53{
54 int err = -EIO;
55 int status;
56 struct ocfs2_dinode *fe = NULL;
57 struct buffer_head *bh = NULL;
58 struct buffer_head *buffer_cache_bh = NULL;
59 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
60 void *kaddr;
61
Tao Ma95581562011-02-22 21:33:59 +080062 trace_ocfs2_symlink_get_block(
63 (unsigned long long)OCFS2_I(inode)->ip_blkno,
64 (unsigned long long)iblock, bh_result, create);
Mark Fashehccd979b2005-12-15 14:31:24 -080065
66 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
67
68 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
69 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
70 (unsigned long long)iblock);
71 goto bail;
72 }
73
Joel Beckerb657c952008-11-13 14:49:11 -080074 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -080075 if (status < 0) {
76 mlog_errno(status);
77 goto bail;
78 }
79 fe = (struct ocfs2_dinode *) bh->b_data;
80
Mark Fashehccd979b2005-12-15 14:31:24 -080081 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
82 le32_to_cpu(fe->i_clusters))) {
Rui Xiang7391a292013-11-12 15:06:54 -080083 err = -ENOMEM;
Mark Fashehccd979b2005-12-15 14:31:24 -080084 mlog(ML_ERROR, "block offset is outside the allocated size: "
85 "%llu\n", (unsigned long long)iblock);
86 goto bail;
87 }
88
89 /* We don't use the page cache to create symlink data, so if
90 * need be, copy it over from the buffer cache. */
91 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
92 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
93 iblock;
94 buffer_cache_bh = sb_getblk(osb->sb, blkno);
95 if (!buffer_cache_bh) {
Rui Xiang7391a292013-11-12 15:06:54 -080096 err = -ENOMEM;
Mark Fashehccd979b2005-12-15 14:31:24 -080097 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
98 goto bail;
99 }
100
101 /* we haven't locked out transactions, so a commit
102 * could've happened. Since we've got a reference on
103 * the bh, even if it commits while we're doing the
104 * copy, the data is still good. */
105 if (buffer_jbd(buffer_cache_bh)
106 && ocfs2_inode_is_new(inode)) {
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800107 kaddr = kmap_atomic(bh_result->b_page);
Mark Fashehccd979b2005-12-15 14:31:24 -0800108 if (!kaddr) {
109 mlog(ML_ERROR, "couldn't kmap!\n");
110 goto bail;
111 }
112 memcpy(kaddr + (bh_result->b_size * iblock),
113 buffer_cache_bh->b_data,
114 bh_result->b_size);
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800115 kunmap_atomic(kaddr);
Mark Fashehccd979b2005-12-15 14:31:24 -0800116 set_buffer_uptodate(bh_result);
117 }
118 brelse(buffer_cache_bh);
119 }
120
121 map_bh(bh_result, inode->i_sb,
122 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
123
124 err = 0;
125
126bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700127 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800128
Mark Fashehccd979b2005-12-15 14:31:24 -0800129 return err;
130}
131
Tao Ma6f70fa52009-08-25 08:05:12 +0800132int ocfs2_get_block(struct inode *inode, sector_t iblock,
133 struct buffer_head *bh_result, int create)
Mark Fashehccd979b2005-12-15 14:31:24 -0800134{
135 int err = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800136 unsigned int ext_flags;
Mark Fasheh628a24f2007-10-30 12:08:32 -0700137 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
138 u64 p_blkno, count, past_eof;
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800139 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800140
Tao Ma95581562011-02-22 21:33:59 +0800141 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
142 (unsigned long long)iblock, bh_result, create);
Mark Fashehccd979b2005-12-15 14:31:24 -0800143
144 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
145 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
146 inode, inode->i_ino);
147
148 if (S_ISLNK(inode->i_mode)) {
149 /* this always does I/O for some reason. */
150 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
151 goto bail;
152 }
153
Mark Fasheh628a24f2007-10-30 12:08:32 -0700154 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800155 &ext_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800156 if (err) {
157 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
Mark Fashehb06970532006-03-03 10:24:33 -0800158 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
159 (unsigned long long)p_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800160 goto bail;
161 }
162
Mark Fasheh628a24f2007-10-30 12:08:32 -0700163 if (max_blocks < count)
164 count = max_blocks;
165
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800166 /*
167 * ocfs2 never allocates in this function - the only time we
168 * need to use BH_New is when we're extending i_size on a file
169 * system which doesn't support holes, in which case BH_New
Christoph Hellwigebdec242010-10-06 10:47:23 +0200170 * allows __block_write_begin() to zero.
Coly Lic0420ad2008-06-30 18:45:45 +0800171 *
172 * If we see this on a sparse file system, then a truncate has
173 * raced us and removed the cluster. In this case, we clear
174 * the buffers dirty and uptodate bits and let the buffer code
175 * ignore it as a hole.
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800176 */
Coly Lic0420ad2008-06-30 18:45:45 +0800177 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
178 clear_buffer_dirty(bh_result);
179 clear_buffer_uptodate(bh_result);
180 goto bail;
181 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800182
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800183 /* Treat the unwritten extent as a hole for zeroing purposes. */
184 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800185 map_bh(bh_result, inode->i_sb, p_blkno);
186
Mark Fasheh628a24f2007-10-30 12:08:32 -0700187 bh_result->b_size = count << inode->i_blkbits;
188
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800189 if (!ocfs2_sparse_alloc(osb)) {
190 if (p_blkno == 0) {
191 err = -EIO;
192 mlog(ML_ERROR,
193 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
194 (unsigned long long)iblock,
195 (unsigned long long)p_blkno,
196 (unsigned long long)OCFS2_I(inode)->ip_blkno);
197 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
198 dump_stack();
Wengang Wang1f4cea32009-07-13 11:38:58 +0800199 goto bail;
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800200 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800201 }
202
Joel Becker56934862010-07-01 15:13:31 -0700203 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
Tao Ma95581562011-02-22 21:33:59 +0800204
205 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
206 (unsigned long long)past_eof);
Joel Becker56934862010-07-01 15:13:31 -0700207 if (create && (iblock >= past_eof))
208 set_buffer_new(bh_result);
209
Mark Fashehccd979b2005-12-15 14:31:24 -0800210bail:
211 if (err < 0)
212 err = -EIO;
213
Mark Fashehccd979b2005-12-15 14:31:24 -0800214 return err;
215}
216
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700217int ocfs2_read_inline_data(struct inode *inode, struct page *page,
218 struct buffer_head *di_bh)
Mark Fasheh6798d352007-09-07 14:05:51 -0700219{
220 void *kaddr;
Jan Karad2849fb2007-12-19 15:24:09 +0100221 loff_t size;
Mark Fasheh6798d352007-09-07 14:05:51 -0700222 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
223
224 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
225 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
226 (unsigned long long)OCFS2_I(inode)->ip_blkno);
227 return -EROFS;
228 }
229
230 size = i_size_read(inode);
231
232 if (size > PAGE_CACHE_SIZE ||
Tiger Yangd9ae49d2009-03-05 11:06:15 +0800233 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
Mark Fasheh6798d352007-09-07 14:05:51 -0700234 ocfs2_error(inode->i_sb,
Jan Karad2849fb2007-12-19 15:24:09 +0100235 "Inode %llu has with inline data has bad size: %Lu",
236 (unsigned long long)OCFS2_I(inode)->ip_blkno,
237 (unsigned long long)size);
Mark Fasheh6798d352007-09-07 14:05:51 -0700238 return -EROFS;
239 }
240
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800241 kaddr = kmap_atomic(page);
Mark Fasheh6798d352007-09-07 14:05:51 -0700242 if (size)
243 memcpy(kaddr, di->id2.i_data.id_data, size);
244 /* Clear the remaining part of the page */
245 memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
246 flush_dcache_page(page);
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800247 kunmap_atomic(kaddr);
Mark Fasheh6798d352007-09-07 14:05:51 -0700248
249 SetPageUptodate(page);
250
251 return 0;
252}
253
254static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
255{
256 int ret;
257 struct buffer_head *di_bh = NULL;
Mark Fasheh6798d352007-09-07 14:05:51 -0700258
259 BUG_ON(!PageLocked(page));
Julia Lawall86c838b2008-02-26 21:45:56 +0100260 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
Mark Fasheh6798d352007-09-07 14:05:51 -0700261
Joel Beckerb657c952008-11-13 14:49:11 -0800262 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh6798d352007-09-07 14:05:51 -0700263 if (ret) {
264 mlog_errno(ret);
265 goto out;
266 }
267
268 ret = ocfs2_read_inline_data(inode, page, di_bh);
269out:
270 unlock_page(page);
271
272 brelse(di_bh);
273 return ret;
274}
275
Mark Fashehccd979b2005-12-15 14:31:24 -0800276static int ocfs2_readpage(struct file *file, struct page *page)
277{
278 struct inode *inode = page->mapping->host;
Mark Fasheh6798d352007-09-07 14:05:51 -0700279 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800280 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
281 int ret, unlock = 1;
282
Tao Ma95581562011-02-22 21:33:59 +0800283 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
284 (page ? page->index : 0));
Mark Fashehccd979b2005-12-15 14:31:24 -0800285
Mark Fashehe63aecb62007-10-18 15:30:42 -0700286 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
Mark Fashehccd979b2005-12-15 14:31:24 -0800287 if (ret != 0) {
288 if (ret == AOP_TRUNCATED_PAGE)
289 unlock = 0;
290 mlog_errno(ret);
291 goto out;
292 }
293
Mark Fasheh6798d352007-09-07 14:05:51 -0700294 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
Jan Karac7e25e62011-06-23 22:51:47 +0200295 /*
296 * Unlock the page and cycle ip_alloc_sem so that we don't
297 * busyloop waiting for ip_alloc_sem to unlock
298 */
Mark Fashehe9dfc0b2007-05-14 11:38:51 -0700299 ret = AOP_TRUNCATED_PAGE;
Jan Karac7e25e62011-06-23 22:51:47 +0200300 unlock_page(page);
301 unlock = 0;
302 down_read(&oi->ip_alloc_sem);
303 up_read(&oi->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700304 goto out_inode_unlock;
Mark Fashehe9dfc0b2007-05-14 11:38:51 -0700305 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800306
307 /*
308 * i_size might have just been updated as we grabed the meta lock. We
309 * might now be discovering a truncate that hit on another node.
310 * block_read_full_page->get_block freaks out if it is asked to read
311 * beyond the end of a file, so we check here. Callers
Nick Piggin54cb8822007-07-19 01:46:59 -0700312 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
Mark Fashehccd979b2005-12-15 14:31:24 -0800313 * and notice that the page they just read isn't needed.
314 *
315 * XXX sys_readahead() seems to get that wrong?
316 */
317 if (start >= i_size_read(inode)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800318 zero_user(page, 0, PAGE_SIZE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800319 SetPageUptodate(page);
320 ret = 0;
321 goto out_alloc;
322 }
323
Mark Fasheh6798d352007-09-07 14:05:51 -0700324 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
325 ret = ocfs2_readpage_inline(inode, page);
326 else
327 ret = block_read_full_page(page, ocfs2_get_block);
Mark Fashehccd979b2005-12-15 14:31:24 -0800328 unlock = 0;
329
Mark Fashehccd979b2005-12-15 14:31:24 -0800330out_alloc:
331 up_read(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700332out_inode_unlock:
333 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800334out:
335 if (unlock)
336 unlock_page(page);
Mark Fashehccd979b2005-12-15 14:31:24 -0800337 return ret;
338}
339
Mark Fasheh628a24f2007-10-30 12:08:32 -0700340/*
341 * This is used only for read-ahead. Failures or difficult to handle
342 * situations are safe to ignore.
343 *
344 * Right now, we don't bother with BH_Boundary - in-inode extent lists
345 * are quite large (243 extents on 4k blocks), so most inodes don't
346 * grow out to a tree. If need be, detecting boundary extents could
347 * trivially be added in a future version of ocfs2_get_block().
348 */
349static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
350 struct list_head *pages, unsigned nr_pages)
351{
352 int ret, err = -EIO;
353 struct inode *inode = mapping->host;
354 struct ocfs2_inode_info *oi = OCFS2_I(inode);
355 loff_t start;
356 struct page *last;
357
358 /*
359 * Use the nonblocking flag for the dlm code to avoid page
360 * lock inversion, but don't bother with retrying.
361 */
362 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
363 if (ret)
364 return err;
365
366 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
367 ocfs2_inode_unlock(inode, 0);
368 return err;
369 }
370
371 /*
372 * Don't bother with inline-data. There isn't anything
373 * to read-ahead in that case anyway...
374 */
375 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
376 goto out_unlock;
377
378 /*
379 * Check whether a remote node truncated this file - we just
380 * drop out in that case as it's not worth handling here.
381 */
382 last = list_entry(pages->prev, struct page, lru);
383 start = (loff_t)last->index << PAGE_CACHE_SHIFT;
384 if (start >= i_size_read(inode))
385 goto out_unlock;
386
387 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
388
389out_unlock:
390 up_read(&oi->ip_alloc_sem);
391 ocfs2_inode_unlock(inode, 0);
392
393 return err;
394}
395
Mark Fashehccd979b2005-12-15 14:31:24 -0800396/* Note: Because we don't support holes, our allocation has
397 * already happened (allocation writes zeros to the file data)
398 * so we don't have to worry about ordered writes in
399 * ocfs2_writepage.
400 *
401 * ->writepage is called during the process of invalidating the page cache
402 * during blocked lock processing. It can't block on any cluster locks
403 * to during block mapping. It's relying on the fact that the block
404 * mapping can't have disappeared under the dirty pages that it is
405 * being asked to write back.
406 */
407static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
408{
Tao Ma95581562011-02-22 21:33:59 +0800409 trace_ocfs2_writepage(
410 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
411 page->index);
Mark Fashehccd979b2005-12-15 14:31:24 -0800412
Tao Ma95581562011-02-22 21:33:59 +0800413 return block_write_full_page(page, ocfs2_get_block, wbc);
Mark Fashehccd979b2005-12-15 14:31:24 -0800414}
415
Mark Fashehccd979b2005-12-15 14:31:24 -0800416/* Taken from ext3. We don't necessarily need the full blown
417 * functionality yet, but IMHO it's better to cut and paste the whole
418 * thing so we can avoid introducing our own bugs (and easily pick up
419 * their fixes when they happen) --Mark */
Mark Fasheh60b11392007-02-16 11:46:50 -0800420int walk_page_buffers( handle_t *handle,
421 struct buffer_head *head,
422 unsigned from,
423 unsigned to,
424 int *partial,
425 int (*fn)( handle_t *handle,
426 struct buffer_head *bh))
Mark Fashehccd979b2005-12-15 14:31:24 -0800427{
428 struct buffer_head *bh;
429 unsigned block_start, block_end;
430 unsigned blocksize = head->b_size;
431 int err, ret = 0;
432 struct buffer_head *next;
433
434 for ( bh = head, block_start = 0;
435 ret == 0 && (bh != head || !block_start);
436 block_start = block_end, bh = next)
437 {
438 next = bh->b_this_page;
439 block_end = block_start + blocksize;
440 if (block_end <= from || block_start >= to) {
441 if (partial && !buffer_uptodate(bh))
442 *partial = 1;
443 continue;
444 }
445 err = (*fn)(handle, bh);
446 if (!ret)
447 ret = err;
448 }
449 return ret;
450}
451
Mark Fashehccd979b2005-12-15 14:31:24 -0800452static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
453{
454 sector_t status;
455 u64 p_blkno = 0;
456 int err = 0;
457 struct inode *inode = mapping->host;
458
Tao Ma95581562011-02-22 21:33:59 +0800459 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
460 (unsigned long long)block);
Mark Fashehccd979b2005-12-15 14:31:24 -0800461
462 /* We don't need to lock journal system files, since they aren't
463 * accessed concurrently from multiple nodes.
464 */
465 if (!INODE_JOURNAL(inode)) {
Mark Fashehe63aecb62007-10-18 15:30:42 -0700466 err = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800467 if (err) {
468 if (err != -ENOENT)
469 mlog_errno(err);
470 goto bail;
471 }
472 down_read(&OCFS2_I(inode)->ip_alloc_sem);
473 }
474
Mark Fasheh6798d352007-09-07 14:05:51 -0700475 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
476 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
477 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800478
479 if (!INODE_JOURNAL(inode)) {
480 up_read(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700481 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800482 }
483
484 if (err) {
485 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
486 (unsigned long long)block);
487 mlog_errno(err);
488 goto bail;
489 }
490
Mark Fashehccd979b2005-12-15 14:31:24 -0800491bail:
492 status = err ? 0 : p_blkno;
493
Mark Fashehccd979b2005-12-15 14:31:24 -0800494 return status;
495}
496
497/*
498 * TODO: Make this into a generic get_blocks function.
499 *
500 * From do_direct_io in direct-io.c:
501 * "So what we do is to permit the ->get_blocks function to populate
502 * bh.b_size with the size of IO which is permitted at this offset and
503 * this i_blkbits."
504 *
505 * This function is called directly from get_more_blocks in direct-io.c.
506 *
507 * called like this: dio->get_blocks(dio->inode, fs_startblk,
508 * fs_count, map_bh, dio->rw == WRITE);
Christoph Hellwig5fe878a2009-12-15 16:47:50 -0800509 *
510 * Note that we never bother to allocate blocks here, and thus ignore the
511 * create argument.
Mark Fashehccd979b2005-12-15 14:31:24 -0800512 */
513static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
Mark Fashehccd979b2005-12-15 14:31:24 -0800514 struct buffer_head *bh_result, int create)
515{
516 int ret;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800517 u64 p_blkno, inode_blocks, contig_blocks;
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800518 unsigned int ext_flags;
Florin Malita184d7d22006-06-03 19:30:10 -0400519 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -0800520 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
Mark Fashehccd979b2005-12-15 14:31:24 -0800521
Mark Fashehccd979b2005-12-15 14:31:24 -0800522 /* This function won't even be called if the request isn't all
523 * nicely aligned and of the right size, so there's no need
524 * for us to check any of that. */
525
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800526 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
Mark Fasheh564f8a32006-12-14 13:01:05 -0800527
Mark Fashehccd979b2005-12-15 14:31:24 -0800528 /* This figures out the size of the next contiguous block, and
529 * our logical offset */
Mark Fasheh363041a2007-01-17 12:31:35 -0800530 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800531 &contig_blocks, &ext_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800532 if (ret) {
533 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
534 (unsigned long long)iblock);
535 ret = -EIO;
536 goto bail;
537 }
538
Tao Macbaee472010-02-26 10:54:52 +0800539 /* We should already CoW the refcounted extent in case of create. */
540 BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED));
541
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800542 /*
543 * get_more_blocks() expects us to describe a hole by clearing
544 * the mapped bit on bh_result().
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800545 *
546 * Consider an unwritten extent as a hole.
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800547 */
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800548 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800549 map_bh(bh_result, inode->i_sb, p_blkno);
Christoph Hellwig5fe878a2009-12-15 16:47:50 -0800550 else
Mark Fasheh25baf2d2007-02-14 15:30:30 -0800551 clear_buffer_mapped(bh_result);
Mark Fashehccd979b2005-12-15 14:31:24 -0800552
553 /* make sure we don't map more than max_blocks blocks here as
554 that's all the kernel will handle at this point. */
555 if (max_blocks < contig_blocks)
556 contig_blocks = max_blocks;
557 bh_result->b_size = contig_blocks << blocksize_bits;
558bail:
559 return ret;
560}
561
Sunil Mushran2bd63212010-01-25 16:57:38 -0800562/*
Mark Fashehccd979b2005-12-15 14:31:24 -0800563 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400564 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
565 * to protect io on one node from truncation on another.
Mark Fashehccd979b2005-12-15 14:31:24 -0800566 */
567static void ocfs2_dio_end_io(struct kiocb *iocb,
568 loff_t offset,
569 ssize_t bytes,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200570 void *private)
Mark Fashehccd979b2005-12-15 14:31:24 -0800571{
Al Viro496ad9a2013-01-23 17:07:38 -0500572 struct inode *inode = file_inode(iocb->ki_filp);
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -0700573 int level;
Mark Fashehccd979b2005-12-15 14:31:24 -0800574
575 /* this io's submitter should not have unlocked this before we could */
576 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -0700577
Christoph Hellwigdf2d6f22011-06-24 14:29:46 -0400578 if (ocfs2_iocb_is_sem_locked(iocb))
Tristan Ye39c99f12010-12-07 14:35:07 +0800579 ocfs2_iocb_clear_sem_locked(iocb);
Tristan Ye39c99f12010-12-07 14:35:07 +0800580
Mark Fasheha11f7e62011-06-22 14:23:38 -0700581 if (ocfs2_iocb_is_unaligned_aio(iocb)) {
582 ocfs2_iocb_clear_unaligned_aio(iocb);
583
Wengang Wangc18ceab2014-04-03 14:46:46 -0700584 mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
Mark Fasheha11f7e62011-06-22 14:23:38 -0700585 }
586
Mark Fashehccd979b2005-12-15 14:31:24 -0800587 ocfs2_iocb_clear_rw_locked(iocb);
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -0700588
589 level = ocfs2_iocb_rw_locked_level(iocb);
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -0700590 ocfs2_rw_unlock(inode, level);
Mark Fashehccd979b2005-12-15 14:31:24 -0800591}
592
Joel Becker03f981c2007-01-04 14:54:41 -0800593static int ocfs2_releasepage(struct page *page, gfp_t wait)
594{
Joel Becker03f981c2007-01-04 14:54:41 -0800595 if (!page_has_buffers(page))
596 return 0;
Jan Kara41ecc342013-11-12 15:07:08 -0800597 return try_to_free_buffers(page);
Joel Becker03f981c2007-01-04 14:54:41 -0800598}
599
Mark Fashehccd979b2005-12-15 14:31:24 -0800600static ssize_t ocfs2_direct_IO(int rw,
601 struct kiocb *iocb,
602 const struct iovec *iov,
603 loff_t offset,
604 unsigned long nr_segs)
605{
606 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500607 struct inode *inode = file_inode(file)->i_mapping->host;
Mark Fasheh53013cb2006-05-05 19:04:03 -0700608
Mark Fasheh6798d352007-09-07 14:05:51 -0700609 /*
610 * Fallback to buffered I/O if we see an inode without
611 * extents.
612 */
613 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
614 return 0;
615
Tao Mab80474b2009-09-10 15:28:47 +0800616 /* Fallback to buffered I/O if we are appending. */
617 if (i_size_read(inode) <= offset)
618 return 0;
619
Tao Mac1e8d352011-03-07 16:43:21 +0800620 return __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev,
621 iov, offset, nr_segs,
622 ocfs2_direct_IO_get_blocks,
623 ocfs2_dio_end_io, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800624}
625
Mark Fasheh9517bac2007-02-09 20:24:12 -0800626static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
627 u32 cpos,
628 unsigned int *start,
629 unsigned int *end)
630{
631 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
632
633 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
634 unsigned int cpp;
635
636 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
637
638 cluster_start = cpos % cpp;
639 cluster_start = cluster_start << osb->s_clustersize_bits;
640
641 cluster_end = cluster_start + osb->s_clustersize;
642 }
643
644 BUG_ON(cluster_start > PAGE_SIZE);
645 BUG_ON(cluster_end > PAGE_SIZE);
646
647 if (start)
648 *start = cluster_start;
649 if (end)
650 *end = cluster_end;
651}
652
653/*
654 * 'from' and 'to' are the region in the page to avoid zeroing.
655 *
656 * If pagesize > clustersize, this function will avoid zeroing outside
657 * of the cluster boundary.
658 *
659 * from == to == 0 is code for "zero the entire cluster region"
660 */
661static void ocfs2_clear_page_regions(struct page *page,
662 struct ocfs2_super *osb, u32 cpos,
663 unsigned from, unsigned to)
664{
665 void *kaddr;
666 unsigned int cluster_start, cluster_end;
667
668 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
669
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800670 kaddr = kmap_atomic(page);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800671
672 if (from || to) {
673 if (from > cluster_start)
674 memset(kaddr + cluster_start, 0, from - cluster_start);
675 if (to < cluster_end)
676 memset(kaddr + to, 0, cluster_end - to);
677 } else {
678 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
679 }
680
Cong Wangc4bc8dc2011-11-25 23:14:34 +0800681 kunmap_atomic(kaddr);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800682}
683
684/*
Mark Fasheh4e9563f2007-11-01 11:37:48 -0700685 * Nonsparse file systems fully allocate before we get to the write
686 * code. This prevents ocfs2_write() from tagging the write as an
687 * allocating one, which means ocfs2_map_page_blocks() might try to
688 * read-in the blocks at the tail of our file. Avoid reading them by
689 * testing i_size against each block offset.
690 */
691static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
692 unsigned int block_start)
693{
694 u64 offset = page_offset(page) + block_start;
695
696 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
697 return 1;
698
699 if (i_size_read(inode) > offset)
700 return 1;
701
702 return 0;
703}
704
705/*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200706 * Some of this taken from __block_write_begin(). We already have our
Mark Fasheh9517bac2007-02-09 20:24:12 -0800707 * mapping by now though, and the entire write will be allocating or
708 * it won't, so not much need to use BH_New.
709 *
710 * This will also skip zeroing, which is handled externally.
711 */
Mark Fasheh60b11392007-02-16 11:46:50 -0800712int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
713 struct inode *inode, unsigned int from,
714 unsigned int to, int new)
Mark Fasheh9517bac2007-02-09 20:24:12 -0800715{
716 int ret = 0;
717 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
718 unsigned int block_end, block_start;
719 unsigned int bsize = 1 << inode->i_blkbits;
720
721 if (!page_has_buffers(page))
722 create_empty_buffers(page, bsize, 0);
723
724 head = page_buffers(page);
725 for (bh = head, block_start = 0; bh != head || !block_start;
726 bh = bh->b_this_page, block_start += bsize) {
727 block_end = block_start + bsize;
728
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700729 clear_buffer_new(bh);
730
Mark Fasheh9517bac2007-02-09 20:24:12 -0800731 /*
732 * Ignore blocks outside of our i/o range -
733 * they may belong to unallocated clusters.
734 */
Mark Fasheh60b11392007-02-16 11:46:50 -0800735 if (block_start >= to || block_end <= from) {
Mark Fasheh9517bac2007-02-09 20:24:12 -0800736 if (PageUptodate(page))
737 set_buffer_uptodate(bh);
738 continue;
739 }
740
741 /*
742 * For an allocating write with cluster size >= page
743 * size, we always write the entire page.
744 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700745 if (new)
746 set_buffer_new(bh);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800747
748 if (!buffer_mapped(bh)) {
749 map_bh(bh, inode->i_sb, *p_blkno);
750 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
751 }
752
753 if (PageUptodate(page)) {
754 if (!buffer_uptodate(bh))
755 set_buffer_uptodate(bh);
756 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
Mark Fashehbce99762007-06-18 11:12:36 -0700757 !buffer_new(bh) &&
Mark Fasheh4e9563f2007-11-01 11:37:48 -0700758 ocfs2_should_read_blk(inode, page, block_start) &&
Mark Fashehbce99762007-06-18 11:12:36 -0700759 (block_start < from || block_end > to)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -0800760 ll_rw_block(READ, 1, &bh);
761 *wait_bh++=bh;
762 }
763
764 *p_blkno = *p_blkno + 1;
765 }
766
767 /*
768 * If we issued read requests - let them complete.
769 */
770 while(wait_bh > wait) {
771 wait_on_buffer(*--wait_bh);
772 if (!buffer_uptodate(*wait_bh))
773 ret = -EIO;
774 }
775
776 if (ret == 0 || !new)
777 return ret;
778
779 /*
780 * If we get -EIO above, zero out any newly allocated blocks
781 * to avoid exposing stale data.
782 */
783 bh = head;
784 block_start = 0;
785 do {
Mark Fasheh9517bac2007-02-09 20:24:12 -0800786 block_end = block_start + bsize;
787 if (block_end <= from)
788 goto next_bh;
789 if (block_start >= to)
790 break;
791
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800792 zero_user(page, block_start, bh->b_size);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800793 set_buffer_uptodate(bh);
794 mark_buffer_dirty(bh);
795
796next_bh:
797 block_start = block_end;
798 bh = bh->b_this_page;
799 } while (bh != head);
800
801 return ret;
802}
803
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700804#if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
805#define OCFS2_MAX_CTXT_PAGES 1
806#else
807#define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
808#endif
Mark Fasheh6af67d82007-03-06 17:24:46 -0800809
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700810#define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
Mark Fasheh6af67d82007-03-06 17:24:46 -0800811
812/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700813 * Describe the state of a single cluster to be written to.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800814 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700815struct ocfs2_write_cluster_desc {
816 u32 c_cpos;
817 u32 c_phys;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800818 /*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700819 * Give this a unique field because c_phys eventually gets
820 * filled.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800821 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700822 unsigned c_new;
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700823 unsigned c_unwritten;
Sunil Mushrane7432672009-08-06 16:12:58 -0700824 unsigned c_needs_zero;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700825};
Mark Fasheh9517bac2007-02-09 20:24:12 -0800826
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700827struct ocfs2_write_ctxt {
828 /* Logical cluster position / len of write */
829 u32 w_cpos;
830 u32 w_clen;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800831
Sunil Mushrane7432672009-08-06 16:12:58 -0700832 /* First cluster allocated in a nonsparse extend */
833 u32 w_first_new_cpos;
834
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700835 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
Mark Fasheh9517bac2007-02-09 20:24:12 -0800836
837 /*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700838 * This is true if page_size > cluster_size.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800839 *
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700840 * It triggers a set of special cases during write which might
841 * have to deal with allocating writes to partial pages.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800842 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700843 unsigned int w_large_pages;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800844
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700845 /*
846 * Pages involved in this write.
847 *
848 * w_target_page is the page being written to by the user.
849 *
850 * w_pages is an array of pages which always contains
851 * w_target_page, and in the case of an allocating write with
852 * page_size < cluster size, it will contain zero'd and mapped
853 * pages adjacent to w_target_page which need to be written
854 * out in so that future reads from that region will get
855 * zero's.
856 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700857 unsigned int w_num_pages;
Goldwyn Rodrigues83fd9c72010-06-10 17:21:36 -0500858 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700859 struct page *w_target_page;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800860
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700861 /*
Wengang Wang5cffff92011-07-24 10:36:54 -0700862 * w_target_locked is used for page_mkwrite path indicating no unlocking
863 * against w_target_page in ocfs2_write_end_nolock.
864 */
865 unsigned int w_target_locked:1;
866
867 /*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700868 * ocfs2_write_end() uses this to know what the real range to
869 * write in the target should be.
870 */
871 unsigned int w_target_from;
872 unsigned int w_target_to;
873
874 /*
875 * We could use journal_current_handle() but this is cleaner,
876 * IMHO -Mark
877 */
878 handle_t *w_handle;
879
880 struct buffer_head *w_di_bh;
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700881
882 struct ocfs2_cached_dealloc_ctxt w_dealloc;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700883};
884
Mark Fasheh1d410a62007-09-07 14:20:45 -0700885void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700886{
887 int i;
888
Mark Fasheh1d410a62007-09-07 14:20:45 -0700889 for(i = 0; i < num_pages; i++) {
890 if (pages[i]) {
891 unlock_page(pages[i]);
892 mark_page_accessed(pages[i]);
893 page_cache_release(pages[i]);
894 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700895 }
Mark Fasheh1d410a62007-09-07 14:20:45 -0700896}
897
898static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
899{
Wengang Wang5cffff92011-07-24 10:36:54 -0700900 int i;
901
902 /*
903 * w_target_locked is only set to true in the page_mkwrite() case.
904 * The intent is to allow us to lock the target page from write_begin()
905 * to write_end(). The caller must hold a ref on w_target_page.
906 */
907 if (wc->w_target_locked) {
908 BUG_ON(!wc->w_target_page);
909 for (i = 0; i < wc->w_num_pages; i++) {
910 if (wc->w_target_page == wc->w_pages[i]) {
911 wc->w_pages[i] = NULL;
912 break;
913 }
914 }
915 mark_page_accessed(wc->w_target_page);
916 page_cache_release(wc->w_target_page);
917 }
Mark Fasheh1d410a62007-09-07 14:20:45 -0700918 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700919
920 brelse(wc->w_di_bh);
921 kfree(wc);
922}
923
924static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
925 struct ocfs2_super *osb, loff_t pos,
Mark Fasheh607d44a2007-05-09 15:14:45 -0700926 unsigned len, struct buffer_head *di_bh)
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700927{
tao.ma@oracle.com30b85482007-09-06 08:02:25 +0800928 u32 cend;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700929 struct ocfs2_write_ctxt *wc;
930
931 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
932 if (!wc)
933 return -ENOMEM;
934
935 wc->w_cpos = pos >> osb->s_clustersize_bits;
Sunil Mushrane7432672009-08-06 16:12:58 -0700936 wc->w_first_new_cpos = UINT_MAX;
tao.ma@oracle.com30b85482007-09-06 08:02:25 +0800937 cend = (pos + len - 1) >> osb->s_clustersize_bits;
938 wc->w_clen = cend - wc->w_cpos + 1;
Mark Fasheh607d44a2007-05-09 15:14:45 -0700939 get_bh(di_bh);
940 wc->w_di_bh = di_bh;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700941
942 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
943 wc->w_large_pages = 1;
944 else
945 wc->w_large_pages = 0;
946
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700947 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
948
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700949 *wcp = wc;
950
951 return 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800952}
953
954/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700955 * If a page has any new buffers, zero them out here, and mark them uptodate
956 * and dirty so they'll be written out (in order to prevent uninitialised
957 * block data from leaking). And clear the new bit.
Mark Fasheh9517bac2007-02-09 20:24:12 -0800958 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700959static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
Mark Fasheh9517bac2007-02-09 20:24:12 -0800960{
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700961 unsigned int block_start, block_end;
962 struct buffer_head *head, *bh;
Mark Fasheh9517bac2007-02-09 20:24:12 -0800963
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700964 BUG_ON(!PageLocked(page));
965 if (!page_has_buffers(page))
966 return;
967
968 bh = head = page_buffers(page);
969 block_start = 0;
970 do {
971 block_end = block_start + bh->b_size;
972
973 if (buffer_new(bh)) {
974 if (block_end > from && block_start < to) {
975 if (!PageUptodate(page)) {
976 unsigned start, end;
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700977
978 start = max(from, block_start);
979 end = min(to, block_end);
980
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800981 zero_user_segment(page, start, end);
Mark Fasheh3a307ff2007-05-08 17:47:32 -0700982 set_buffer_uptodate(bh);
983 }
984
985 clear_buffer_new(bh);
986 mark_buffer_dirty(bh);
987 }
988 }
989
990 block_start = block_end;
991 bh = bh->b_this_page;
992 } while (bh != head);
993}
994
995/*
996 * Only called when we have a failure during allocating write to write
997 * zero's to the newly allocated region.
998 */
999static void ocfs2_write_failure(struct inode *inode,
1000 struct ocfs2_write_ctxt *wc,
1001 loff_t user_pos, unsigned user_len)
1002{
1003 int i;
Mark Fasheh5c26a7b2007-09-18 17:49:29 -07001004 unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1005 to = user_pos + user_len;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001006 struct page *tmppage;
1007
Mark Fasheh5c26a7b2007-09-18 17:49:29 -07001008 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001009
1010 for(i = 0; i < wc->w_num_pages; i++) {
1011 tmppage = wc->w_pages[i];
1012
Sunil Mushran961cecb2008-07-16 17:22:22 -07001013 if (page_has_buffers(tmppage)) {
Mark Fasheh53ef99c2008-11-18 16:53:43 -08001014 if (ocfs2_should_order_data(inode))
Joel Becker2b4e30f2008-09-03 20:03:41 -07001015 ocfs2_jbd2_file_inode(wc->w_handle, inode);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001016
Sunil Mushran961cecb2008-07-16 17:22:22 -07001017 block_commit_write(tmppage, from, to);
1018 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001019 }
1020}
1021
1022static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1023 struct ocfs2_write_ctxt *wc,
1024 struct page *page, u32 cpos,
1025 loff_t user_pos, unsigned user_len,
1026 int new)
1027{
1028 int ret;
1029 unsigned int map_from = 0, map_to = 0;
1030 unsigned int cluster_start, cluster_end;
1031 unsigned int user_data_from = 0, user_data_to = 0;
1032
1033 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001034 &cluster_start, &cluster_end);
1035
Goldwyn Rodrigues272b62c2011-02-17 09:44:40 -06001036 /* treat the write as new if the a hole/lseek spanned across
1037 * the page boundary.
1038 */
1039 new = new | ((i_size_read(inode) <= page_offset(page)) &&
1040 (page_offset(page) <= user_pos));
1041
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001042 if (page == wc->w_target_page) {
1043 map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1044 map_to = map_from + user_len;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001045
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001046 if (new)
1047 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1048 cluster_start, cluster_end,
1049 new);
1050 else
1051 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1052 map_from, map_to, new);
1053 if (ret) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001054 mlog_errno(ret);
1055 goto out;
1056 }
1057
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001058 user_data_from = map_from;
1059 user_data_to = map_to;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001060 if (new) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001061 map_from = cluster_start;
1062 map_to = cluster_end;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001063 }
1064 } else {
1065 /*
1066 * If we haven't allocated the new page yet, we
1067 * shouldn't be writing it out without copying user
1068 * data. This is likely a math error from the caller.
1069 */
1070 BUG_ON(!new);
1071
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001072 map_from = cluster_start;
1073 map_to = cluster_end;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001074
1075 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001076 cluster_start, cluster_end, new);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001077 if (ret) {
1078 mlog_errno(ret);
1079 goto out;
1080 }
1081 }
1082
1083 /*
1084 * Parts of newly allocated pages need to be zero'd.
1085 *
1086 * Above, we have also rewritten 'to' and 'from' - as far as
1087 * the rest of the function is concerned, the entire cluster
1088 * range inside of a page needs to be written.
1089 *
1090 * We can skip this if the page is up to date - it's already
1091 * been zero'd from being read in as a hole.
1092 */
1093 if (new && !PageUptodate(page))
1094 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001095 cpos, user_data_from, user_data_to);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001096
1097 flush_dcache_page(page);
1098
Mark Fasheh9517bac2007-02-09 20:24:12 -08001099out:
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001100 return ret;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001101}
1102
1103/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001104 * This function will only grab one clusters worth of pages.
Mark Fasheh9517bac2007-02-09 20:24:12 -08001105 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001106static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1107 struct ocfs2_write_ctxt *wc,
Joel Becker693c2412010-07-02 17:20:27 -07001108 u32 cpos, loff_t user_pos,
1109 unsigned user_len, int new,
Mark Fasheh7307de82007-05-09 15:16:19 -07001110 struct page *mmap_page)
Mark Fasheh9517bac2007-02-09 20:24:12 -08001111{
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001112 int ret = 0, i;
Joel Becker693c2412010-07-02 17:20:27 -07001113 unsigned long start, target_index, end_index, index;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001114 struct inode *inode = mapping->host;
Joel Becker693c2412010-07-02 17:20:27 -07001115 loff_t last_byte;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001116
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001117 target_index = user_pos >> PAGE_CACHE_SHIFT;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001118
1119 /*
1120 * Figure out how many pages we'll be manipulating here. For
Mark Fasheh60b11392007-02-16 11:46:50 -08001121 * non allocating write, we just change the one
Joel Becker693c2412010-07-02 17:20:27 -07001122 * page. Otherwise, we'll need a whole clusters worth. If we're
1123 * writing past i_size, we only need enough pages to cover the
1124 * last page of the write.
Mark Fasheh9517bac2007-02-09 20:24:12 -08001125 */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001126 if (new) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001127 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1128 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
Joel Becker693c2412010-07-02 17:20:27 -07001129 /*
1130 * We need the index *past* the last page we could possibly
1131 * touch. This is the page past the end of the write or
1132 * i_size, whichever is greater.
1133 */
1134 last_byte = max(user_pos + user_len, i_size_read(inode));
1135 BUG_ON(last_byte < 1);
1136 end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1;
1137 if ((start + wc->w_num_pages) > end_index)
1138 wc->w_num_pages = end_index - start;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001139 } else {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001140 wc->w_num_pages = 1;
1141 start = target_index;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001142 }
1143
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001144 for(i = 0; i < wc->w_num_pages; i++) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001145 index = start + i;
1146
Mark Fasheh7307de82007-05-09 15:16:19 -07001147 if (index == target_index && mmap_page) {
1148 /*
1149 * ocfs2_pagemkwrite() is a little different
1150 * and wants us to directly use the page
1151 * passed in.
1152 */
1153 lock_page(mmap_page);
1154
Wengang Wang5cffff92011-07-24 10:36:54 -07001155 /* Exit and let the caller retry */
Mark Fasheh7307de82007-05-09 15:16:19 -07001156 if (mmap_page->mapping != mapping) {
Wengang Wang5cffff92011-07-24 10:36:54 -07001157 WARN_ON(mmap_page->mapping);
Mark Fasheh7307de82007-05-09 15:16:19 -07001158 unlock_page(mmap_page);
Wengang Wang5cffff92011-07-24 10:36:54 -07001159 ret = -EAGAIN;
Mark Fasheh7307de82007-05-09 15:16:19 -07001160 goto out;
1161 }
1162
1163 page_cache_get(mmap_page);
1164 wc->w_pages[i] = mmap_page;
Wengang Wang5cffff92011-07-24 10:36:54 -07001165 wc->w_target_locked = true;
Mark Fasheh7307de82007-05-09 15:16:19 -07001166 } else {
1167 wc->w_pages[i] = find_or_create_page(mapping, index,
1168 GFP_NOFS);
1169 if (!wc->w_pages[i]) {
1170 ret = -ENOMEM;
1171 mlog_errno(ret);
1172 goto out;
1173 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001174 }
Jan Kara12695292013-02-21 16:42:57 -08001175 wait_for_stable_page(wc->w_pages[i]);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001176
1177 if (index == target_index)
1178 wc->w_target_page = wc->w_pages[i];
Mark Fasheh9517bac2007-02-09 20:24:12 -08001179 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001180out:
Wengang Wang5cffff92011-07-24 10:36:54 -07001181 if (ret)
1182 wc->w_target_locked = false;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001183 return ret;
1184}
1185
1186/*
1187 * Prepare a single cluster for write one cluster into the file.
1188 */
1189static int ocfs2_write_cluster(struct address_space *mapping,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001190 u32 phys, unsigned int unwritten,
Sunil Mushrane7432672009-08-06 16:12:58 -07001191 unsigned int should_zero,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001192 struct ocfs2_alloc_context *data_ac,
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001193 struct ocfs2_alloc_context *meta_ac,
1194 struct ocfs2_write_ctxt *wc, u32 cpos,
1195 loff_t user_pos, unsigned user_len)
1196{
Sunil Mushrane7432672009-08-06 16:12:58 -07001197 int ret, i, new;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001198 u64 v_blkno, p_blkno;
1199 struct inode *inode = mapping->host;
Joel Beckerf99b9b72008-08-20 19:36:33 -07001200 struct ocfs2_extent_tree et;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001201
1202 new = phys == 0 ? 1 : 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001203 if (new) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001204 u32 tmp_pos;
1205
Mark Fasheh9517bac2007-02-09 20:24:12 -08001206 /*
1207 * This is safe to call with the page locks - it won't take
1208 * any additional semaphores or cluster locks.
1209 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001210 tmp_pos = cpos;
Tao Ma0eb8d472008-08-18 17:38:45 +08001211 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1212 &tmp_pos, 1, 0, wc->w_di_bh,
1213 wc->w_handle, data_ac,
1214 meta_ac, NULL);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001215 /*
1216 * This shouldn't happen because we must have already
1217 * calculated the correct meta data allocation required. The
1218 * internal tree allocation code should know how to increase
1219 * transaction credits itself.
1220 *
1221 * If need be, we could handle -EAGAIN for a
1222 * RESTART_TRANS here.
1223 */
1224 mlog_bug_on_msg(ret == -EAGAIN,
1225 "Inode %llu: EAGAIN return during allocation.\n",
1226 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1227 if (ret < 0) {
1228 mlog_errno(ret);
1229 goto out;
1230 }
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001231 } else if (unwritten) {
Joel Becker5e404e92009-02-13 03:54:22 -08001232 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1233 wc->w_di_bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -07001234 ret = ocfs2_mark_extent_written(inode, &et,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001235 wc->w_handle, cpos, 1, phys,
Joel Beckerf99b9b72008-08-20 19:36:33 -07001236 meta_ac, &wc->w_dealloc);
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001237 if (ret < 0) {
1238 mlog_errno(ret);
1239 goto out;
1240 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001241 }
1242
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001243 if (should_zero)
1244 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1245 else
1246 v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1247
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001248 /*
1249 * The only reason this should fail is due to an inability to
1250 * find the extent added.
1251 */
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001252 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1253 NULL);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001254 if (ret < 0) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001255 ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
1256 "at logical block %llu",
1257 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1258 (unsigned long long)v_blkno);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001259 goto out;
1260 }
1261
1262 BUG_ON(p_blkno == 0);
1263
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001264 for(i = 0; i < wc->w_num_pages; i++) {
1265 int tmpret;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001266
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001267 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1268 wc->w_pages[i], cpos,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001269 user_pos, user_len,
1270 should_zero);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001271 if (tmpret) {
1272 mlog_errno(tmpret);
1273 if (ret == 0)
Wengang Wangcbfa9632009-07-13 11:38:23 +08001274 ret = tmpret;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001275 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001276 }
1277
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001278 /*
1279 * We only have cleanup to do in case of allocating write.
1280 */
1281 if (ret && new)
1282 ocfs2_write_failure(inode, wc, user_pos, user_len);
1283
Mark Fasheh9517bac2007-02-09 20:24:12 -08001284out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08001285
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001286 return ret;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001287}
1288
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001289static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1290 struct ocfs2_alloc_context *data_ac,
1291 struct ocfs2_alloc_context *meta_ac,
1292 struct ocfs2_write_ctxt *wc,
1293 loff_t pos, unsigned len)
1294{
1295 int ret, i;
Mark Fashehdb562462007-09-17 09:06:29 -07001296 loff_t cluster_off;
1297 unsigned int local_len = len;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001298 struct ocfs2_write_cluster_desc *desc;
Mark Fashehdb562462007-09-17 09:06:29 -07001299 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001300
1301 for (i = 0; i < wc->w_clen; i++) {
1302 desc = &wc->w_desc[i];
1303
Mark Fashehdb562462007-09-17 09:06:29 -07001304 /*
1305 * We have to make sure that the total write passed in
1306 * doesn't extend past a single cluster.
1307 */
1308 local_len = len;
1309 cluster_off = pos & (osb->s_clustersize - 1);
1310 if ((cluster_off + local_len) > osb->s_clustersize)
1311 local_len = osb->s_clustersize - cluster_off;
1312
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001313 ret = ocfs2_write_cluster(mapping, desc->c_phys,
Sunil Mushrane7432672009-08-06 16:12:58 -07001314 desc->c_unwritten,
1315 desc->c_needs_zero,
1316 data_ac, meta_ac,
Mark Fashehdb562462007-09-17 09:06:29 -07001317 wc, desc->c_cpos, pos, local_len);
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001318 if (ret) {
1319 mlog_errno(ret);
1320 goto out;
1321 }
Mark Fashehdb562462007-09-17 09:06:29 -07001322
1323 len -= local_len;
1324 pos += local_len;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001325 }
1326
1327 ret = 0;
1328out:
1329 return ret;
1330}
1331
Mark Fasheh9517bac2007-02-09 20:24:12 -08001332/*
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001333 * ocfs2_write_end() wants to know which parts of the target page it
1334 * should complete the write on. It's easiest to compute them ahead of
1335 * time when a more complete view of the write is available.
Mark Fasheh9517bac2007-02-09 20:24:12 -08001336 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001337static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1338 struct ocfs2_write_ctxt *wc,
1339 loff_t pos, unsigned len, int alloc)
Mark Fasheh9517bac2007-02-09 20:24:12 -08001340{
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001341 struct ocfs2_write_cluster_desc *desc;
1342
1343 wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1344 wc->w_target_to = wc->w_target_from + len;
1345
1346 if (alloc == 0)
1347 return;
1348
1349 /*
1350 * Allocating write - we may have different boundaries based
1351 * on page size and cluster size.
1352 *
1353 * NOTE: We can no longer compute one value from the other as
1354 * the actual write length and user provided length may be
1355 * different.
1356 */
1357
1358 if (wc->w_large_pages) {
1359 /*
1360 * We only care about the 1st and last cluster within
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001361 * our range and whether they should be zero'd or not. Either
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001362 * value may be extended out to the start/end of a
1363 * newly allocated cluster.
1364 */
1365 desc = &wc->w_desc[0];
Sunil Mushrane7432672009-08-06 16:12:58 -07001366 if (desc->c_needs_zero)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001367 ocfs2_figure_cluster_boundaries(osb,
1368 desc->c_cpos,
1369 &wc->w_target_from,
1370 NULL);
1371
1372 desc = &wc->w_desc[wc->w_clen - 1];
Sunil Mushrane7432672009-08-06 16:12:58 -07001373 if (desc->c_needs_zero)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001374 ocfs2_figure_cluster_boundaries(osb,
1375 desc->c_cpos,
1376 NULL,
1377 &wc->w_target_to);
1378 } else {
1379 wc->w_target_from = 0;
1380 wc->w_target_to = PAGE_CACHE_SIZE;
1381 }
1382}
1383
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001384/*
1385 * Populate each single-cluster write descriptor in the write context
1386 * with information about the i/o to be done.
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001387 *
1388 * Returns the number of clusters that will have to be allocated, as
1389 * well as a worst case estimate of the number of extent records that
1390 * would have to be created during a write to an unwritten region.
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001391 */
1392static int ocfs2_populate_write_desc(struct inode *inode,
1393 struct ocfs2_write_ctxt *wc,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001394 unsigned int *clusters_to_alloc,
1395 unsigned int *extents_to_split)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001396{
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001397 int ret;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001398 struct ocfs2_write_cluster_desc *desc;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001399 unsigned int num_clusters = 0;
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001400 unsigned int ext_flags = 0;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001401 u32 phys = 0;
1402 int i;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001403
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001404 *clusters_to_alloc = 0;
1405 *extents_to_split = 0;
1406
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001407 for (i = 0; i < wc->w_clen; i++) {
1408 desc = &wc->w_desc[i];
1409 desc->c_cpos = wc->w_cpos + i;
1410
1411 if (num_clusters == 0) {
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001412 /*
1413 * Need to look up the next extent record.
1414 */
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001415 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001416 &num_clusters, &ext_flags);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001417 if (ret) {
1418 mlog_errno(ret);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001419 goto out;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001420 }
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001421
Tao Ma293b2f72009-08-25 08:02:48 +08001422 /* We should already CoW the refcountd extent. */
1423 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1424
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001425 /*
1426 * Assume worst case - that we're writing in
1427 * the middle of the extent.
1428 *
1429 * We can assume that the write proceeds from
1430 * left to right, in which case the extent
1431 * insert code is smart enough to coalesce the
1432 * next splits into the previous records created.
1433 */
1434 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1435 *extents_to_split = *extents_to_split + 2;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001436 } else if (phys) {
1437 /*
1438 * Only increment phys if it doesn't describe
1439 * a hole.
1440 */
1441 phys++;
1442 }
1443
Sunil Mushrane7432672009-08-06 16:12:58 -07001444 /*
1445 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1446 * file that got extended. w_first_new_cpos tells us
1447 * where the newly allocated clusters are so we can
1448 * zero them.
1449 */
1450 if (desc->c_cpos >= wc->w_first_new_cpos) {
1451 BUG_ON(phys == 0);
1452 desc->c_needs_zero = 1;
1453 }
1454
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001455 desc->c_phys = phys;
1456 if (phys == 0) {
1457 desc->c_new = 1;
Sunil Mushrane7432672009-08-06 16:12:58 -07001458 desc->c_needs_zero = 1;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001459 *clusters_to_alloc = *clusters_to_alloc + 1;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001460 }
Sunil Mushrane7432672009-08-06 16:12:58 -07001461
1462 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001463 desc->c_unwritten = 1;
Sunil Mushrane7432672009-08-06 16:12:58 -07001464 desc->c_needs_zero = 1;
1465 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001466
1467 num_clusters--;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001468 }
1469
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001470 ret = 0;
1471out:
1472 return ret;
1473}
1474
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001475static int ocfs2_write_begin_inline(struct address_space *mapping,
1476 struct inode *inode,
1477 struct ocfs2_write_ctxt *wc)
1478{
1479 int ret;
1480 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1481 struct page *page;
1482 handle_t *handle;
1483 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1484
1485 page = find_or_create_page(mapping, 0, GFP_NOFS);
1486 if (!page) {
1487 ret = -ENOMEM;
1488 mlog_errno(ret);
1489 goto out;
1490 }
1491 /*
1492 * If we don't set w_num_pages then this page won't get unlocked
1493 * and freed on cleanup of the write context.
1494 */
1495 wc->w_pages[0] = wc->w_target_page = page;
1496 wc->w_num_pages = 1;
1497
1498 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1499 if (IS_ERR(handle)) {
1500 ret = PTR_ERR(handle);
1501 mlog_errno(ret);
1502 goto out;
1503 }
1504
Joel Becker0cf2f762009-02-12 16:41:25 -08001505 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001506 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001507 if (ret) {
1508 ocfs2_commit_trans(osb, handle);
1509
1510 mlog_errno(ret);
1511 goto out;
1512 }
1513
1514 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1515 ocfs2_set_inode_data_inline(inode, di);
1516
1517 if (!PageUptodate(page)) {
1518 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1519 if (ret) {
1520 ocfs2_commit_trans(osb, handle);
1521
1522 goto out;
1523 }
1524 }
1525
1526 wc->w_handle = handle;
1527out:
1528 return ret;
1529}
1530
1531int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1532{
1533 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1534
Mark Fasheh0d8a4e02007-11-20 11:48:41 -08001535 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001536 return 1;
1537 return 0;
1538}
1539
1540static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1541 struct inode *inode, loff_t pos,
1542 unsigned len, struct page *mmap_page,
1543 struct ocfs2_write_ctxt *wc)
1544{
1545 int ret, written = 0;
1546 loff_t end = pos + len;
1547 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Tiger Yangd9ae49d2009-03-05 11:06:15 +08001548 struct ocfs2_dinode *di = NULL;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001549
Tao Ma95581562011-02-22 21:33:59 +08001550 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1551 len, (unsigned long long)pos,
1552 oi->ip_dyn_features);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001553
1554 /*
1555 * Handle inodes which already have inline data 1st.
1556 */
1557 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1558 if (mmap_page == NULL &&
1559 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1560 goto do_inline_write;
1561
1562 /*
1563 * The write won't fit - we have to give this inode an
1564 * inline extent list now.
1565 */
1566 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1567 if (ret)
1568 mlog_errno(ret);
1569 goto out;
1570 }
1571
1572 /*
1573 * Check whether the inode can accept inline data.
1574 */
1575 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1576 return 0;
1577
1578 /*
1579 * Check whether the write can fit.
1580 */
Tiger Yangd9ae49d2009-03-05 11:06:15 +08001581 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1582 if (mmap_page ||
1583 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001584 return 0;
1585
1586do_inline_write:
1587 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1588 if (ret) {
1589 mlog_errno(ret);
1590 goto out;
1591 }
1592
1593 /*
1594 * This signals to the caller that the data can be written
1595 * inline.
1596 */
1597 written = 1;
1598out:
1599 return written ? written : ret;
1600}
1601
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001602/*
1603 * This function only does anything for file systems which can't
1604 * handle sparse files.
1605 *
1606 * What we want to do here is fill in any hole between the current end
1607 * of allocation and the end of our write. That way the rest of the
1608 * write path can treat it as an non-allocating write, which has no
1609 * special case code for sparse/nonsparse files.
1610 */
Joel Becker56934862010-07-01 15:13:31 -07001611static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1612 struct buffer_head *di_bh,
1613 loff_t pos, unsigned len,
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001614 struct ocfs2_write_ctxt *wc)
1615{
1616 int ret;
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001617 loff_t newsize = pos + len;
1618
Joel Becker56934862010-07-01 15:13:31 -07001619 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001620
1621 if (newsize <= i_size_read(inode))
1622 return 0;
1623
Joel Becker56934862010-07-01 15:13:31 -07001624 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001625 if (ret)
1626 mlog_errno(ret);
1627
Sunil Mushrane7432672009-08-06 16:12:58 -07001628 wc->w_first_new_cpos =
1629 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1630
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001631 return ret;
1632}
1633
Joel Becker56934862010-07-01 15:13:31 -07001634static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1635 loff_t pos)
1636{
1637 int ret = 0;
1638
1639 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1640 if (pos > i_size_read(inode))
1641 ret = ocfs2_zero_extend(inode, di_bh, pos);
1642
1643 return ret;
1644}
1645
Tao Ma50308d82010-11-04 15:14:11 +08001646/*
1647 * Try to flush truncate logs if we can free enough clusters from it.
1648 * As for return value, "< 0" means error, "0" no space and "1" means
1649 * we have freed enough spaces and let the caller try to allocate again.
1650 */
1651static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
1652 unsigned int needed)
1653{
1654 tid_t target;
1655 int ret = 0;
1656 unsigned int truncated_clusters;
1657
1658 mutex_lock(&osb->osb_tl_inode->i_mutex);
1659 truncated_clusters = osb->truncated_clusters;
1660 mutex_unlock(&osb->osb_tl_inode->i_mutex);
1661
1662 /*
1663 * Check whether we can succeed in allocating if we free
1664 * the truncate log.
1665 */
1666 if (truncated_clusters < needed)
1667 goto out;
1668
1669 ret = ocfs2_flush_truncate_log(osb);
1670 if (ret) {
1671 mlog_errno(ret);
1672 goto out;
1673 }
1674
1675 if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
1676 jbd2_log_wait_commit(osb->journal->j_journal, target);
1677 ret = 1;
1678 }
1679out:
1680 return ret;
1681}
1682
Tao Ma0378da0f2010-08-12 10:25:28 +08001683int ocfs2_write_begin_nolock(struct file *filp,
1684 struct address_space *mapping,
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001685 loff_t pos, unsigned len, unsigned flags,
1686 struct page **pagep, void **fsdata,
1687 struct buffer_head *di_bh, struct page *mmap_page)
1688{
Sunil Mushrane7432672009-08-06 16:12:58 -07001689 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
Tao Ma50308d82010-11-04 15:14:11 +08001690 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001691 struct ocfs2_write_ctxt *wc;
1692 struct inode *inode = mapping->host;
1693 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1694 struct ocfs2_dinode *di;
1695 struct ocfs2_alloc_context *data_ac = NULL;
1696 struct ocfs2_alloc_context *meta_ac = NULL;
1697 handle_t *handle;
Joel Beckerf99b9b72008-08-20 19:36:33 -07001698 struct ocfs2_extent_tree et;
Tao Ma50308d82010-11-04 15:14:11 +08001699 int try_free = 1, ret1;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001700
Tao Ma50308d82010-11-04 15:14:11 +08001701try_again:
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001702 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1703 if (ret) {
1704 mlog_errno(ret);
1705 return ret;
1706 }
1707
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001708 if (ocfs2_supports_inline_data(osb)) {
1709 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1710 mmap_page, wc);
1711 if (ret == 1) {
1712 ret = 0;
1713 goto success;
1714 }
1715 if (ret < 0) {
1716 mlog_errno(ret);
1717 goto out;
1718 }
1719 }
1720
Joel Becker56934862010-07-01 15:13:31 -07001721 if (ocfs2_sparse_alloc(osb))
1722 ret = ocfs2_zero_tail(inode, di_bh, pos);
1723 else
1724 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len,
1725 wc);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001726 if (ret) {
1727 mlog_errno(ret);
1728 goto out;
1729 }
1730
Tao Ma293b2f72009-08-25 08:02:48 +08001731 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1732 if (ret < 0) {
1733 mlog_errno(ret);
1734 goto out;
1735 } else if (ret == 1) {
Tao Ma50308d82010-11-04 15:14:11 +08001736 clusters_need = wc->w_clen;
Tiger Yangc7dd3392013-08-13 16:00:58 -07001737 ret = ocfs2_refcount_cow(inode, di_bh,
Tao Ma37f8a2b2009-08-26 09:47:28 +08001738 wc->w_cpos, wc->w_clen, UINT_MAX);
Tao Ma293b2f72009-08-25 08:02:48 +08001739 if (ret) {
1740 mlog_errno(ret);
1741 goto out;
1742 }
1743 }
1744
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001745 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1746 &extents_to_split);
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001747 if (ret) {
1748 mlog_errno(ret);
1749 goto out;
1750 }
Tao Ma50308d82010-11-04 15:14:11 +08001751 clusters_need += clusters_to_alloc;
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001752
1753 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1754
Tao Ma95581562011-02-22 21:33:59 +08001755 trace_ocfs2_write_begin_nolock(
1756 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1757 (long long)i_size_read(inode),
1758 le32_to_cpu(di->i_clusters),
1759 pos, len, flags, mmap_page,
1760 clusters_to_alloc, extents_to_split);
1761
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001762 /*
1763 * We set w_target_from, w_target_to here so that
1764 * ocfs2_write_end() knows which range in the target page to
1765 * write out. An allocation requires that we write the entire
1766 * cluster range.
1767 */
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001768 if (clusters_to_alloc || extents_to_split) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001769 /*
1770 * XXX: We are stretching the limits of
Mark Fashehb27b7cb2007-06-18 11:22:56 -07001771 * ocfs2_lock_allocators(). It greatly over-estimates
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001772 * the work to be done.
1773 */
Joel Becker5e404e92009-02-13 03:54:22 -08001774 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1775 wc->w_di_bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -07001776 ret = ocfs2_lock_allocators(inode, &et,
Tao Ma231b87d2008-08-18 17:38:42 +08001777 clusters_to_alloc, extents_to_split,
Joel Beckerf99b9b72008-08-20 19:36:33 -07001778 &data_ac, &meta_ac);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001779 if (ret) {
1780 mlog_errno(ret);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001781 goto out;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001782 }
1783
Mark Fasheh4fe370a2009-12-07 13:15:40 -08001784 if (data_ac)
1785 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1786
Tao Ma811f9332008-08-18 17:38:43 +08001787 credits = ocfs2_calc_extend_credits(inode->i_sb,
Goldwyn Rodrigues06f9da62013-11-12 15:06:52 -08001788 &di->id2.i_list);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001789
Mark Fasheh9517bac2007-02-09 20:24:12 -08001790 }
1791
Sunil Mushrane7432672009-08-06 16:12:58 -07001792 /*
1793 * We have to zero sparse allocated clusters, unwritten extent clusters,
1794 * and non-sparse clusters we just extended. For non-sparse writes,
1795 * we know zeros will only be needed in the first and/or last cluster.
1796 */
1797 if (clusters_to_alloc || extents_to_split ||
Sunil Mushran8379e7c2009-09-04 11:12:01 -07001798 (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1799 wc->w_desc[wc->w_clen - 1].c_needs_zero)))
Sunil Mushrane7432672009-08-06 16:12:58 -07001800 cluster_of_pages = 1;
1801 else
1802 cluster_of_pages = 0;
1803
1804 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001805
Mark Fasheh9517bac2007-02-09 20:24:12 -08001806 handle = ocfs2_start_trans(osb, credits);
1807 if (IS_ERR(handle)) {
1808 ret = PTR_ERR(handle);
1809 mlog_errno(ret);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001810 goto out;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001811 }
1812
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001813 wc->w_handle = handle;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001814
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001815 if (clusters_to_alloc) {
1816 ret = dquot_alloc_space_nodirty(inode,
1817 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1818 if (ret)
1819 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02001820 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001821 /*
1822 * We don't want this to fail in ocfs2_write_end(), so do it
1823 * here.
1824 */
Joel Becker0cf2f762009-02-12 16:41:25 -08001825 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001826 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001827 if (ret) {
1828 mlog_errno(ret);
Jan Karaa90714c2008-10-09 19:38:40 +02001829 goto out_quota;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001830 }
1831
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001832 /*
1833 * Fill our page array first. That way we've grabbed enough so
1834 * that we can zero and flush if we error after adding the
1835 * extent.
1836 */
Joel Becker693c2412010-07-02 17:20:27 -07001837 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
Sunil Mushrane7432672009-08-06 16:12:58 -07001838 cluster_of_pages, mmap_page);
Wengang Wang5cffff92011-07-24 10:36:54 -07001839 if (ret && ret != -EAGAIN) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001840 mlog_errno(ret);
Jan Karaa90714c2008-10-09 19:38:40 +02001841 goto out_quota;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001842 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001843
Wengang Wang5cffff92011-07-24 10:36:54 -07001844 /*
1845 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
1846 * the target page. In this case, we exit with no error and no target
1847 * page. This will trigger the caller, page_mkwrite(), to re-try
1848 * the operation.
1849 */
1850 if (ret == -EAGAIN) {
1851 BUG_ON(wc->w_target_page);
1852 ret = 0;
1853 goto out_quota;
1854 }
1855
Mark Fasheh0d172ba2007-05-14 18:09:54 -07001856 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1857 len);
1858 if (ret) {
1859 mlog_errno(ret);
Jan Karaa90714c2008-10-09 19:38:40 +02001860 goto out_quota;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001861 }
1862
1863 if (data_ac)
1864 ocfs2_free_alloc_context(data_ac);
1865 if (meta_ac)
1866 ocfs2_free_alloc_context(meta_ac);
1867
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001868success:
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001869 *pagep = wc->w_target_page;
1870 *fsdata = wc;
1871 return 0;
Jan Karaa90714c2008-10-09 19:38:40 +02001872out_quota:
1873 if (clusters_to_alloc)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001874 dquot_free_space(inode,
Jan Karaa90714c2008-10-09 19:38:40 +02001875 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
Mark Fasheh9517bac2007-02-09 20:24:12 -08001876out_commit:
1877 ocfs2_commit_trans(osb, handle);
1878
Mark Fasheh9517bac2007-02-09 20:24:12 -08001879out:
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001880 ocfs2_free_write_ctxt(wc);
1881
Xue jiufeib1214e42013-11-12 15:07:06 -08001882 if (data_ac) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001883 ocfs2_free_alloc_context(data_ac);
Xue jiufeib1214e42013-11-12 15:07:06 -08001884 data_ac = NULL;
1885 }
1886 if (meta_ac) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001887 ocfs2_free_alloc_context(meta_ac);
Xue jiufeib1214e42013-11-12 15:07:06 -08001888 meta_ac = NULL;
1889 }
Tao Ma50308d82010-11-04 15:14:11 +08001890
1891 if (ret == -ENOSPC && try_free) {
1892 /*
1893 * Try to free some truncate log so that we can have enough
1894 * clusters to allocate.
1895 */
1896 try_free = 0;
1897
1898 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1899 if (ret1 == 1)
1900 goto try_again;
1901
1902 if (ret1 < 0)
1903 mlog_errno(ret1);
1904 }
1905
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001906 return ret;
1907}
Mark Fasheh9517bac2007-02-09 20:24:12 -08001908
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001909static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1910 loff_t pos, unsigned len, unsigned flags,
1911 struct page **pagep, void **fsdata)
Mark Fasheh607d44a2007-05-09 15:14:45 -07001912{
1913 int ret;
1914 struct buffer_head *di_bh = NULL;
1915 struct inode *inode = mapping->host;
1916
Mark Fashehe63aecb62007-10-18 15:30:42 -07001917 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001918 if (ret) {
1919 mlog_errno(ret);
1920 return ret;
1921 }
1922
1923 /*
1924 * Take alloc sem here to prevent concurrent lookups. That way
1925 * the mapping, zeroing and tree manipulation within
1926 * ocfs2_write() will be safe against ->readpage(). This
1927 * should also serve to lock out allocation from a shared
1928 * writeable region.
1929 */
1930 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1931
Tao Ma0378da0f2010-08-12 10:25:28 +08001932 ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep,
Mark Fasheh7307de82007-05-09 15:16:19 -07001933 fsdata, di_bh, NULL);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001934 if (ret) {
1935 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001936 goto out_fail;
Mark Fasheh607d44a2007-05-09 15:14:45 -07001937 }
1938
1939 brelse(di_bh);
1940
1941 return 0;
1942
Mark Fasheh607d44a2007-05-09 15:14:45 -07001943out_fail:
1944 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1945
1946 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001947 ocfs2_inode_unlock(inode, 1);
Mark Fasheh607d44a2007-05-09 15:14:45 -07001948
1949 return ret;
1950}
1951
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001952static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1953 unsigned len, unsigned *copied,
1954 struct ocfs2_dinode *di,
1955 struct ocfs2_write_ctxt *wc)
1956{
1957 void *kaddr;
1958
1959 if (unlikely(*copied < len)) {
1960 if (!PageUptodate(wc->w_target_page)) {
1961 *copied = 0;
1962 return;
1963 }
1964 }
1965
Cong Wangc4bc8dc2011-11-25 23:14:34 +08001966 kaddr = kmap_atomic(wc->w_target_page);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001967 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
Cong Wangc4bc8dc2011-11-25 23:14:34 +08001968 kunmap_atomic(kaddr);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001969
Tao Ma95581562011-02-22 21:33:59 +08001970 trace_ocfs2_write_end_inline(
1971 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001972 (unsigned long long)pos, *copied,
1973 le16_to_cpu(di->id2.i_data.id_count),
1974 le16_to_cpu(di->i_dyn_features));
1975}
1976
Mark Fasheh7307de82007-05-09 15:16:19 -07001977int ocfs2_write_end_nolock(struct address_space *mapping,
1978 loff_t pos, unsigned len, unsigned copied,
1979 struct page *page, void *fsdata)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001980{
1981 int i;
1982 unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
1983 struct inode *inode = mapping->host;
1984 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1985 struct ocfs2_write_ctxt *wc = fsdata;
1986 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1987 handle_t *handle = wc->w_handle;
1988 struct page *tmppage;
1989
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001990 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1991 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1992 goto out_write_size;
1993 }
1994
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001995 if (unlikely(copied < len)) {
1996 if (!PageUptodate(wc->w_target_page))
1997 copied = 0;
1998
1999 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
2000 start+len);
2001 }
2002 flush_dcache_page(wc->w_target_page);
2003
2004 for(i = 0; i < wc->w_num_pages; i++) {
2005 tmppage = wc->w_pages[i];
2006
2007 if (tmppage == wc->w_target_page) {
2008 from = wc->w_target_from;
2009 to = wc->w_target_to;
2010
2011 BUG_ON(from > PAGE_CACHE_SIZE ||
2012 to > PAGE_CACHE_SIZE ||
2013 to < from);
2014 } else {
2015 /*
2016 * Pages adjacent to the target (if any) imply
2017 * a hole-filling write in which case we want
2018 * to flush their entire range.
2019 */
2020 from = 0;
2021 to = PAGE_CACHE_SIZE;
2022 }
2023
Sunil Mushran961cecb2008-07-16 17:22:22 -07002024 if (page_has_buffers(tmppage)) {
Mark Fasheh53ef99c2008-11-18 16:53:43 -08002025 if (ocfs2_should_order_data(inode))
Joel Becker2b4e30f2008-09-03 20:03:41 -07002026 ocfs2_jbd2_file_inode(wc->w_handle, inode);
Sunil Mushran961cecb2008-07-16 17:22:22 -07002027 block_commit_write(tmppage, from, to);
2028 }
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002029 }
2030
Mark Fasheh1afc32b2007-09-07 14:46:51 -07002031out_write_size:
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002032 pos += copied;
Junxiao Bif17c20d2013-09-11 14:19:45 -07002033 if (pos > i_size_read(inode)) {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002034 i_size_write(inode, pos);
2035 mark_inode_dirty(inode);
2036 }
2037 inode->i_blocks = ocfs2_inode_sector_count(inode);
2038 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2039 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2040 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2041 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
Darrick J. Wong2931cdc2014-04-03 14:46:48 -07002042 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002043 ocfs2_journal_dirty(handle, wc->w_di_bh);
2044
2045 ocfs2_commit_trans(osb, handle);
Mark Fasheh59a5e412007-06-22 15:52:36 -07002046
Mark Fashehb27b7cb2007-06-18 11:22:56 -07002047 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2048
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002049 ocfs2_free_write_ctxt(wc);
2050
2051 return copied;
Mark Fasheh9517bac2007-02-09 20:24:12 -08002052}
2053
Nick Pigginb6af1bc2007-10-16 01:25:24 -07002054static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2055 loff_t pos, unsigned len, unsigned copied,
2056 struct page *page, void *fsdata)
Mark Fasheh607d44a2007-05-09 15:14:45 -07002057{
2058 int ret;
2059 struct inode *inode = mapping->host;
2060
2061 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
2062
Mark Fasheh607d44a2007-05-09 15:14:45 -07002063 up_write(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002064 ocfs2_inode_unlock(inode, 1);
Mark Fasheh607d44a2007-05-09 15:14:45 -07002065
2066 return ret;
2067}
2068
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002069const struct address_space_operations ocfs2_aops = {
Hisashi Hifumi1fca3a02009-03-05 17:22:21 +09002070 .readpage = ocfs2_readpage,
2071 .readpages = ocfs2_readpages,
2072 .writepage = ocfs2_writepage,
2073 .write_begin = ocfs2_write_begin,
2074 .write_end = ocfs2_write_end,
2075 .bmap = ocfs2_bmap,
Hisashi Hifumi1fca3a02009-03-05 17:22:21 +09002076 .direct_IO = ocfs2_direct_IO,
Jan Kara41ecc342013-11-12 15:07:08 -08002077 .invalidatepage = block_invalidatepage,
Hisashi Hifumi1fca3a02009-03-05 17:22:21 +09002078 .releasepage = ocfs2_releasepage,
2079 .migratepage = buffer_migrate_page,
2080 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02002081 .error_remove_page = generic_error_remove_page,
Mark Fashehccd979b2005-12-15 14:31:24 -08002082};