1 // SPDX-License-Identifier: GPL-2.0-or-later !! 1 /* -*- mode: c; c-basic-offset: 8; -*-
2 /* !! 2 * vim: noexpandtab sw=8 ts=8 sts=0:
>> 3 *
3 * Copyright (C) 2002, 2004 Oracle. All right 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.
4 */ 20 */
5 21
6 #include <linux/fs.h> 22 #include <linux/fs.h>
7 #include <linux/slab.h> 23 #include <linux/slab.h>
8 #include <linux/highmem.h> 24 #include <linux/highmem.h>
9 #include <linux/pagemap.h> 25 #include <linux/pagemap.h>
10 #include <asm/byteorder.h> 26 #include <asm/byteorder.h>
11 #include <linux/swap.h> 27 #include <linux/swap.h>
>> 28 #include <linux/pipe_fs_i.h>
12 #include <linux/mpage.h> 29 #include <linux/mpage.h>
13 #include <linux/quotaops.h> 30 #include <linux/quotaops.h>
14 #include <linux/blkdev.h> 31 #include <linux/blkdev.h>
15 #include <linux/uio.h> 32 #include <linux/uio.h>
16 #include <linux/mm.h> <<
17 33
18 #include <cluster/masklog.h> 34 #include <cluster/masklog.h>
19 35
20 #include "ocfs2.h" 36 #include "ocfs2.h"
21 37
22 #include "alloc.h" 38 #include "alloc.h"
23 #include "aops.h" 39 #include "aops.h"
24 #include "dlmglue.h" 40 #include "dlmglue.h"
25 #include "extent_map.h" 41 #include "extent_map.h"
26 #include "file.h" 42 #include "file.h"
27 #include "inode.h" 43 #include "inode.h"
28 #include "journal.h" 44 #include "journal.h"
29 #include "suballoc.h" 45 #include "suballoc.h"
30 #include "super.h" 46 #include "super.h"
31 #include "symlink.h" 47 #include "symlink.h"
32 #include "refcounttree.h" 48 #include "refcounttree.h"
33 #include "ocfs2_trace.h" 49 #include "ocfs2_trace.h"
34 50
35 #include "buffer_head_io.h" 51 #include "buffer_head_io.h"
36 #include "dir.h" 52 #include "dir.h"
37 #include "namei.h" 53 #include "namei.h"
38 #include "sysfile.h" 54 #include "sysfile.h"
39 55
40 static int ocfs2_symlink_get_block(struct inod 56 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
41 struct buff 57 struct buffer_head *bh_result, int create)
42 { 58 {
43 int err = -EIO; 59 int err = -EIO;
44 int status; 60 int status;
45 struct ocfs2_dinode *fe = NULL; 61 struct ocfs2_dinode *fe = NULL;
46 struct buffer_head *bh = NULL; 62 struct buffer_head *bh = NULL;
47 struct buffer_head *buffer_cache_bh = 63 struct buffer_head *buffer_cache_bh = NULL;
48 struct ocfs2_super *osb = OCFS2_SB(ino 64 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
49 void *kaddr; 65 void *kaddr;
50 66
51 trace_ocfs2_symlink_get_block( 67 trace_ocfs2_symlink_get_block(
52 (unsigned long long)OC 68 (unsigned long long)OCFS2_I(inode)->ip_blkno,
53 (unsigned long long)ib 69 (unsigned long long)iblock, bh_result, create);
54 70
55 BUG_ON(ocfs2_inode_is_fast_symlink(ino 71 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
56 72
57 if ((iblock << inode->i_sb->s_blocksiz 73 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
58 mlog(ML_ERROR, "block offset > 74 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
59 (unsigned long long)ibloc 75 (unsigned long long)iblock);
60 goto bail; 76 goto bail;
61 } 77 }
62 78
63 status = ocfs2_read_inode_block(inode, 79 status = ocfs2_read_inode_block(inode, &bh);
64 if (status < 0) { 80 if (status < 0) {
65 mlog_errno(status); 81 mlog_errno(status);
66 goto bail; 82 goto bail;
67 } 83 }
68 fe = (struct ocfs2_dinode *) bh->b_dat 84 fe = (struct ocfs2_dinode *) bh->b_data;
69 85
70 if ((u64)iblock >= ocfs2_clusters_to_b 86 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
71 87 le32_to_cpu(fe->i_clusters))) {
72 err = -ENOMEM; 88 err = -ENOMEM;
73 mlog(ML_ERROR, "block offset i 89 mlog(ML_ERROR, "block offset is outside the allocated size: "
74 "%llu\n", (unsigned long 90 "%llu\n", (unsigned long long)iblock);
75 goto bail; 91 goto bail;
76 } 92 }
77 93
78 /* We don't use the page cache to crea 94 /* We don't use the page cache to create symlink data, so if
79 * need be, copy it over from the buff 95 * need be, copy it over from the buffer cache. */
80 if (!buffer_uptodate(bh_result) && ocf 96 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
81 u64 blkno = le64_to_cpu(fe->id 97 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
82 iblock; 98 iblock;
83 buffer_cache_bh = sb_getblk(os 99 buffer_cache_bh = sb_getblk(osb->sb, blkno);
84 if (!buffer_cache_bh) { 100 if (!buffer_cache_bh) {
85 err = -ENOMEM; 101 err = -ENOMEM;
86 mlog(ML_ERROR, "couldn 102 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
87 goto bail; 103 goto bail;
88 } 104 }
89 105
90 /* we haven't locked out trans 106 /* we haven't locked out transactions, so a commit
91 * could've happened. Since we 107 * could've happened. Since we've got a reference on
92 * the bh, even if it commits 108 * the bh, even if it commits while we're doing the
93 * copy, the data is still goo 109 * copy, the data is still good. */
94 if (buffer_jbd(buffer_cache_bh 110 if (buffer_jbd(buffer_cache_bh)
95 && ocfs2_inode_is_new(inod 111 && ocfs2_inode_is_new(inode)) {
96 kaddr = kmap_atomic(bh 112 kaddr = kmap_atomic(bh_result->b_page);
97 if (!kaddr) { 113 if (!kaddr) {
98 mlog(ML_ERROR, 114 mlog(ML_ERROR, "couldn't kmap!\n");
99 goto bail; 115 goto bail;
100 } 116 }
101 memcpy(kaddr + (bh_res 117 memcpy(kaddr + (bh_result->b_size * iblock),
102 buffer_cache_bh 118 buffer_cache_bh->b_data,
103 bh_result->b_si 119 bh_result->b_size);
104 kunmap_atomic(kaddr); 120 kunmap_atomic(kaddr);
105 set_buffer_uptodate(bh 121 set_buffer_uptodate(bh_result);
106 } 122 }
107 brelse(buffer_cache_bh); 123 brelse(buffer_cache_bh);
108 } 124 }
109 125
110 map_bh(bh_result, inode->i_sb, 126 map_bh(bh_result, inode->i_sb,
111 le64_to_cpu(fe->id2.i_list.l_re 127 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
112 128
113 err = 0; 129 err = 0;
114 130
115 bail: 131 bail:
116 brelse(bh); 132 brelse(bh);
117 133
118 return err; 134 return err;
119 } 135 }
120 136
121 static int ocfs2_lock_get_block(struct inode * 137 static int ocfs2_lock_get_block(struct inode *inode, sector_t iblock,
122 struct buffer_head *bh_res 138 struct buffer_head *bh_result, int create)
123 { 139 {
124 int ret = 0; 140 int ret = 0;
125 struct ocfs2_inode_info *oi = OCFS2_I( 141 struct ocfs2_inode_info *oi = OCFS2_I(inode);
126 142
127 down_read(&oi->ip_alloc_sem); 143 down_read(&oi->ip_alloc_sem);
128 ret = ocfs2_get_block(inode, iblock, b 144 ret = ocfs2_get_block(inode, iblock, bh_result, create);
129 up_read(&oi->ip_alloc_sem); 145 up_read(&oi->ip_alloc_sem);
130 146
131 return ret; 147 return ret;
132 } 148 }
133 149
134 int ocfs2_get_block(struct inode *inode, secto 150 int ocfs2_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh_res 151 struct buffer_head *bh_result, int create)
136 { 152 {
137 int err = 0; 153 int err = 0;
138 unsigned int ext_flags; 154 unsigned int ext_flags;
139 u64 max_blocks = bh_result->b_size >> 155 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
140 u64 p_blkno, count, past_eof; 156 u64 p_blkno, count, past_eof;
141 struct ocfs2_super *osb = OCFS2_SB(ino 157 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
142 158
143 trace_ocfs2_get_block((unsigned long l 159 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
144 (unsigned long l 160 (unsigned long long)iblock, bh_result, create);
145 161
146 if (OCFS2_I(inode)->ip_flags & OCFS2_I 162 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
147 mlog(ML_NOTICE, "get_block on 163 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
148 inode, inode->i_ino); 164 inode, inode->i_ino);
149 165
150 if (S_ISLNK(inode->i_mode)) { 166 if (S_ISLNK(inode->i_mode)) {
151 /* this always does I/O for so 167 /* this always does I/O for some reason. */
152 err = ocfs2_symlink_get_block( 168 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
153 goto bail; 169 goto bail;
154 } 170 }
155 171
156 err = ocfs2_extent_map_get_blocks(inod 172 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
157 &ext 173 &ext_flags);
158 if (err) { 174 if (err) {
159 mlog(ML_ERROR, "Error %d from 175 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
160 "%llu, NULL)\n", err, ino 176 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
161 (unsigned long long)p_blk 177 (unsigned long long)p_blkno);
162 goto bail; 178 goto bail;
163 } 179 }
164 180
165 if (max_blocks < count) 181 if (max_blocks < count)
166 count = max_blocks; 182 count = max_blocks;
167 183
168 /* 184 /*
169 * ocfs2 never allocates in this funct 185 * ocfs2 never allocates in this function - the only time we
170 * need to use BH_New is when we're ex 186 * need to use BH_New is when we're extending i_size on a file
171 * system which doesn't support holes, 187 * system which doesn't support holes, in which case BH_New
172 * allows __block_write_begin() to zer 188 * allows __block_write_begin() to zero.
173 * 189 *
174 * If we see this on a sparse file sys 190 * If we see this on a sparse file system, then a truncate has
175 * raced us and removed the cluster. I 191 * raced us and removed the cluster. In this case, we clear
176 * the buffers dirty and uptodate bits 192 * the buffers dirty and uptodate bits and let the buffer code
177 * ignore it as a hole. 193 * ignore it as a hole.
178 */ 194 */
179 if (create && p_blkno == 0 && ocfs2_sp 195 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
180 clear_buffer_dirty(bh_result); 196 clear_buffer_dirty(bh_result);
181 clear_buffer_uptodate(bh_resul 197 clear_buffer_uptodate(bh_result);
182 goto bail; 198 goto bail;
183 } 199 }
184 200
185 /* Treat the unwritten extent as a hol 201 /* Treat the unwritten extent as a hole for zeroing purposes. */
186 if (p_blkno && !(ext_flags & OCFS2_EXT 202 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
187 map_bh(bh_result, inode->i_sb, 203 map_bh(bh_result, inode->i_sb, p_blkno);
188 204
189 bh_result->b_size = count << inode->i_ 205 bh_result->b_size = count << inode->i_blkbits;
190 206
191 if (!ocfs2_sparse_alloc(osb)) { 207 if (!ocfs2_sparse_alloc(osb)) {
192 if (p_blkno == 0) { 208 if (p_blkno == 0) {
193 err = -EIO; 209 err = -EIO;
194 mlog(ML_ERROR, 210 mlog(ML_ERROR,
195 "iblock = %llu p_ 211 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
196 (unsigned long lo 212 (unsigned long long)iblock,
197 (unsigned long lo 213 (unsigned long long)p_blkno,
198 (unsigned long lo 214 (unsigned long long)OCFS2_I(inode)->ip_blkno);
199 mlog(ML_ERROR, "Size % 215 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
200 dump_stack(); 216 dump_stack();
201 goto bail; 217 goto bail;
202 } 218 }
203 } 219 }
204 220
205 past_eof = ocfs2_blocks_for_bytes(inod 221 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
206 222
207 trace_ocfs2_get_block_end((unsigned lo 223 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
208 (unsigned lo 224 (unsigned long long)past_eof);
209 if (create && (iblock >= past_eof)) 225 if (create && (iblock >= past_eof))
210 set_buffer_new(bh_result); 226 set_buffer_new(bh_result);
211 227
212 bail: 228 bail:
213 if (err < 0) 229 if (err < 0)
214 err = -EIO; 230 err = -EIO;
215 231
216 return err; 232 return err;
217 } 233 }
218 234
219 int ocfs2_read_inline_data(struct inode *inode 235 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
220 struct buffer_head 236 struct buffer_head *di_bh)
221 { 237 {
222 void *kaddr; 238 void *kaddr;
223 loff_t size; 239 loff_t size;
224 struct ocfs2_dinode *di = (struct ocfs 240 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
225 241
226 if (!(le16_to_cpu(di->i_dyn_features) 242 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
227 ocfs2_error(inode->i_sb, "Inod 243 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
228 (unsigned long lon 244 (unsigned long long)OCFS2_I(inode)->ip_blkno);
229 return -EROFS; 245 return -EROFS;
230 } 246 }
231 247
232 size = i_size_read(inode); 248 size = i_size_read(inode);
233 249
234 if (size > PAGE_SIZE || 250 if (size > PAGE_SIZE ||
235 size > ocfs2_max_inline_data_with_ 251 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
236 ocfs2_error(inode->i_sb, 252 ocfs2_error(inode->i_sb,
237 "Inode %llu has wi 253 "Inode %llu has with inline data has bad size: %Lu\n",
238 (unsigned long lon 254 (unsigned long long)OCFS2_I(inode)->ip_blkno,
239 (unsigned long lon 255 (unsigned long long)size);
240 return -EROFS; 256 return -EROFS;
241 } 257 }
242 258
243 kaddr = kmap_atomic(page); 259 kaddr = kmap_atomic(page);
244 if (size) 260 if (size)
245 memcpy(kaddr, di->id2.i_data.i 261 memcpy(kaddr, di->id2.i_data.id_data, size);
246 /* Clear the remaining part of the pag 262 /* Clear the remaining part of the page */
247 memset(kaddr + size, 0, PAGE_SIZE - si 263 memset(kaddr + size, 0, PAGE_SIZE - size);
248 flush_dcache_page(page); 264 flush_dcache_page(page);
249 kunmap_atomic(kaddr); 265 kunmap_atomic(kaddr);
250 266
251 SetPageUptodate(page); 267 SetPageUptodate(page);
252 268
253 return 0; 269 return 0;
254 } 270 }
255 271
256 static int ocfs2_readpage_inline(struct inode 272 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
257 { 273 {
258 int ret; 274 int ret;
259 struct buffer_head *di_bh = NULL; 275 struct buffer_head *di_bh = NULL;
260 276
261 BUG_ON(!PageLocked(page)); 277 BUG_ON(!PageLocked(page));
262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 278 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
263 279
264 ret = ocfs2_read_inode_block(inode, &d 280 ret = ocfs2_read_inode_block(inode, &di_bh);
265 if (ret) { 281 if (ret) {
266 mlog_errno(ret); 282 mlog_errno(ret);
267 goto out; 283 goto out;
268 } 284 }
269 285
270 ret = ocfs2_read_inline_data(inode, pa 286 ret = ocfs2_read_inline_data(inode, page, di_bh);
271 out: 287 out:
272 unlock_page(page); 288 unlock_page(page);
273 289
274 brelse(di_bh); 290 brelse(di_bh);
275 return ret; 291 return ret;
276 } 292 }
277 293
278 static int ocfs2_read_folio(struct file *file, !! 294 static int ocfs2_readpage(struct file *file, struct page *page)
279 { 295 {
280 struct inode *inode = folio->mapping-> !! 296 struct inode *inode = page->mapping->host;
281 struct ocfs2_inode_info *oi = OCFS2_I( 297 struct ocfs2_inode_info *oi = OCFS2_I(inode);
282 loff_t start = folio_pos(folio); !! 298 loff_t start = (loff_t)page->index << PAGE_SHIFT;
283 int ret, unlock = 1; 299 int ret, unlock = 1;
284 300
285 trace_ocfs2_readpage((unsigned long lo !! 301 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
>> 302 (page ? page->index : 0));
286 303
287 ret = ocfs2_inode_lock_with_page(inode !! 304 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
288 if (ret != 0) { 305 if (ret != 0) {
289 if (ret == AOP_TRUNCATED_PAGE) 306 if (ret == AOP_TRUNCATED_PAGE)
290 unlock = 0; 307 unlock = 0;
291 mlog_errno(ret); 308 mlog_errno(ret);
292 goto out; 309 goto out;
293 } 310 }
294 311
295 if (down_read_trylock(&oi->ip_alloc_se 312 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
296 /* 313 /*
297 * Unlock the folio and cycle !! 314 * Unlock the page and cycle ip_alloc_sem so that we don't
298 * busyloop waiting for ip_all 315 * busyloop waiting for ip_alloc_sem to unlock
299 */ 316 */
300 ret = AOP_TRUNCATED_PAGE; 317 ret = AOP_TRUNCATED_PAGE;
301 folio_unlock(folio); !! 318 unlock_page(page);
302 unlock = 0; 319 unlock = 0;
303 down_read(&oi->ip_alloc_sem); 320 down_read(&oi->ip_alloc_sem);
304 up_read(&oi->ip_alloc_sem); 321 up_read(&oi->ip_alloc_sem);
305 goto out_inode_unlock; 322 goto out_inode_unlock;
306 } 323 }
307 324
308 /* 325 /*
309 * i_size might have just been updated 326 * i_size might have just been updated as we grabed the meta lock. We
310 * might now be discovering a truncate 327 * might now be discovering a truncate that hit on another node.
311 * block_read_full_folio->get_block fr !! 328 * block_read_full_page->get_block freaks out if it is asked to read
312 * beyond the end of a file, so we che 329 * beyond the end of a file, so we check here. Callers
313 * (generic_file_read, vm_ops->fault) 330 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
314 * and notice that the folio they just !! 331 * and notice that the page they just read isn't needed.
315 * 332 *
316 * XXX sys_readahead() seems to get th 333 * XXX sys_readahead() seems to get that wrong?
317 */ 334 */
318 if (start >= i_size_read(inode)) { 335 if (start >= i_size_read(inode)) {
319 folio_zero_segment(folio, 0, f !! 336 zero_user(page, 0, PAGE_SIZE);
320 folio_mark_uptodate(folio); !! 337 SetPageUptodate(page);
321 ret = 0; 338 ret = 0;
322 goto out_alloc; 339 goto out_alloc;
323 } 340 }
324 341
325 if (oi->ip_dyn_features & OCFS2_INLINE 342 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
326 ret = ocfs2_readpage_inline(in !! 343 ret = ocfs2_readpage_inline(inode, page);
327 else 344 else
328 ret = block_read_full_folio(fo !! 345 ret = block_read_full_page(page, ocfs2_get_block);
329 unlock = 0; 346 unlock = 0;
330 347
331 out_alloc: 348 out_alloc:
332 up_read(&oi->ip_alloc_sem); !! 349 up_read(&OCFS2_I(inode)->ip_alloc_sem);
333 out_inode_unlock: 350 out_inode_unlock:
334 ocfs2_inode_unlock(inode, 0); 351 ocfs2_inode_unlock(inode, 0);
335 out: 352 out:
336 if (unlock) 353 if (unlock)
337 folio_unlock(folio); !! 354 unlock_page(page);
338 return ret; 355 return ret;
339 } 356 }
340 357
341 /* 358 /*
342 * This is used only for read-ahead. Failures 359 * This is used only for read-ahead. Failures or difficult to handle
343 * situations are safe to ignore. 360 * situations are safe to ignore.
344 * 361 *
345 * Right now, we don't bother with BH_Boundary 362 * Right now, we don't bother with BH_Boundary - in-inode extent lists
346 * are quite large (243 extents on 4k blocks), 363 * are quite large (243 extents on 4k blocks), so most inodes don't
347 * grow out to a tree. If need be, detecting b 364 * grow out to a tree. If need be, detecting boundary extents could
348 * trivially be added in a future version of o 365 * trivially be added in a future version of ocfs2_get_block().
349 */ 366 */
350 static void ocfs2_readahead(struct readahead_c !! 367 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
>> 368 struct list_head *pages, unsigned nr_pages)
351 { 369 {
352 int ret; !! 370 int ret, err = -EIO;
353 struct inode *inode = rac->mapping->ho !! 371 struct inode *inode = mapping->host;
354 struct ocfs2_inode_info *oi = OCFS2_I( 372 struct ocfs2_inode_info *oi = OCFS2_I(inode);
>> 373 loff_t start;
>> 374 struct page *last;
355 375
356 /* 376 /*
357 * Use the nonblocking flag for the dl 377 * Use the nonblocking flag for the dlm code to avoid page
358 * lock inversion, but don't bother wi 378 * lock inversion, but don't bother with retrying.
359 */ 379 */
360 ret = ocfs2_inode_lock_full(inode, NUL 380 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
361 if (ret) 381 if (ret)
362 return; !! 382 return err;
363 383
364 if (down_read_trylock(&oi->ip_alloc_se !! 384 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
365 goto out_unlock; !! 385 ocfs2_inode_unlock(inode, 0);
>> 386 return err;
>> 387 }
366 388
367 /* 389 /*
368 * Don't bother with inline-data. Ther 390 * Don't bother with inline-data. There isn't anything
369 * to read-ahead in that case anyway.. 391 * to read-ahead in that case anyway...
370 */ 392 */
371 if (oi->ip_dyn_features & OCFS2_INLINE 393 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
372 goto out_up; !! 394 goto out_unlock;
373 395
374 /* 396 /*
375 * Check whether a remote node truncat 397 * Check whether a remote node truncated this file - we just
376 * drop out in that case as it's not w 398 * drop out in that case as it's not worth handling here.
377 */ 399 */
378 if (readahead_pos(rac) >= i_size_read( !! 400 last = list_entry(pages->prev, struct page, lru);
379 goto out_up; !! 401 start = (loff_t)last->index << PAGE_SHIFT;
>> 402 if (start >= i_size_read(inode))
>> 403 goto out_unlock;
380 404
381 mpage_readahead(rac, ocfs2_get_block); !! 405 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
382 406
383 out_up: <<
384 up_read(&oi->ip_alloc_sem); <<
385 out_unlock: 407 out_unlock:
>> 408 up_read(&oi->ip_alloc_sem);
386 ocfs2_inode_unlock(inode, 0); 409 ocfs2_inode_unlock(inode, 0);
>> 410
>> 411 return err;
387 } 412 }
388 413
389 /* Note: Because we don't support holes, our a 414 /* Note: Because we don't support holes, our allocation has
390 * already happened (allocation writes zeros t 415 * already happened (allocation writes zeros to the file data)
391 * so we don't have to worry about ordered wri 416 * so we don't have to worry about ordered writes in
392 * ocfs2_writepages. !! 417 * ocfs2_writepage.
393 * 418 *
394 * ->writepages is called during the process o !! 419 * ->writepage is called during the process of invalidating the page cache
395 * during blocked lock processing. It can't b 420 * during blocked lock processing. It can't block on any cluster locks
396 * to during block mapping. It's relying on t 421 * to during block mapping. It's relying on the fact that the block
397 * mapping can't have disappeared under the di 422 * mapping can't have disappeared under the dirty pages that it is
398 * being asked to write back. 423 * being asked to write back.
399 */ 424 */
400 static int ocfs2_writepages(struct address_spa !! 425 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
401 struct writeback_control *wbc) <<
402 { 426 {
403 return mpage_writepages(mapping, wbc, !! 427 trace_ocfs2_writepage(
>> 428 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
>> 429 page->index);
>> 430
>> 431 return block_write_full_page(page, ocfs2_get_block, wbc);
404 } 432 }
405 433
406 /* Taken from ext3. We don't necessarily need 434 /* Taken from ext3. We don't necessarily need the full blown
407 * functionality yet, but IMHO it's better to 435 * functionality yet, but IMHO it's better to cut and paste the whole
408 * thing so we can avoid introducing our own b 436 * thing so we can avoid introducing our own bugs (and easily pick up
409 * their fixes when they happen) --Mark */ 437 * their fixes when they happen) --Mark */
410 int walk_page_buffers( handle_t *handle, 438 int walk_page_buffers( handle_t *handle,
411 struct buffer_head *he 439 struct buffer_head *head,
412 unsigned from, 440 unsigned from,
413 unsigned to, 441 unsigned to,
414 int *partial, 442 int *partial,
415 int (*fn)( handle 443 int (*fn)( handle_t *handle,
416 struct 444 struct buffer_head *bh))
417 { 445 {
418 struct buffer_head *bh; 446 struct buffer_head *bh;
419 unsigned block_start, block_end; 447 unsigned block_start, block_end;
420 unsigned blocksize = head->b_size; 448 unsigned blocksize = head->b_size;
421 int err, ret = 0; 449 int err, ret = 0;
422 struct buffer_head *next; 450 struct buffer_head *next;
423 451
424 for ( bh = head, block_start = 0; 452 for ( bh = head, block_start = 0;
425 ret == 0 && (bh != head || !bl 453 ret == 0 && (bh != head || !block_start);
426 block_start = block_end, bh = 454 block_start = block_end, bh = next)
427 { 455 {
428 next = bh->b_this_page; 456 next = bh->b_this_page;
429 block_end = block_start + bloc 457 block_end = block_start + blocksize;
430 if (block_end <= from || block 458 if (block_end <= from || block_start >= to) {
431 if (partial && !buffer 459 if (partial && !buffer_uptodate(bh))
432 *partial = 1; 460 *partial = 1;
433 continue; 461 continue;
434 } 462 }
435 err = (*fn)(handle, bh); 463 err = (*fn)(handle, bh);
436 if (!ret) 464 if (!ret)
437 ret = err; 465 ret = err;
438 } 466 }
439 return ret; 467 return ret;
440 } 468 }
441 469
442 static sector_t ocfs2_bmap(struct address_spac 470 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
443 { 471 {
444 sector_t status; 472 sector_t status;
445 u64 p_blkno = 0; 473 u64 p_blkno = 0;
446 int err = 0; 474 int err = 0;
447 struct inode *inode = mapping->host; 475 struct inode *inode = mapping->host;
448 476
449 trace_ocfs2_bmap((unsigned long long)O 477 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
450 (unsigned long long)b 478 (unsigned long long)block);
451 479
452 /* <<
453 * The swap code (ab-)uses ->bmap to g <<
454 * bypasseѕ the file system for actua <<
455 * that on refcounted inodes, so we ha <<
456 * 0 is the magic code for a bmap erro <<
457 */ <<
458 if (ocfs2_is_refcount_inode(inode)) <<
459 return 0; <<
460 <<
461 /* We don't need to lock journal syste 480 /* We don't need to lock journal system files, since they aren't
462 * accessed concurrently from multiple 481 * accessed concurrently from multiple nodes.
463 */ 482 */
464 if (!INODE_JOURNAL(inode)) { 483 if (!INODE_JOURNAL(inode)) {
465 err = ocfs2_inode_lock(inode, 484 err = ocfs2_inode_lock(inode, NULL, 0);
466 if (err) { 485 if (err) {
467 if (err != -ENOENT) 486 if (err != -ENOENT)
468 mlog_errno(err 487 mlog_errno(err);
469 goto bail; 488 goto bail;
470 } 489 }
471 down_read(&OCFS2_I(inode)->ip_ 490 down_read(&OCFS2_I(inode)->ip_alloc_sem);
472 } 491 }
473 492
474 if (!(OCFS2_I(inode)->ip_dyn_features 493 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
475 err = ocfs2_extent_map_get_blo 494 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
476 495 NULL);
477 496
478 if (!INODE_JOURNAL(inode)) { 497 if (!INODE_JOURNAL(inode)) {
479 up_read(&OCFS2_I(inode)->ip_al 498 up_read(&OCFS2_I(inode)->ip_alloc_sem);
480 ocfs2_inode_unlock(inode, 0); 499 ocfs2_inode_unlock(inode, 0);
481 } 500 }
482 501
483 if (err) { 502 if (err) {
484 mlog(ML_ERROR, "get_blocks() f 503 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
485 (unsigned long long)block 504 (unsigned long long)block);
486 mlog_errno(err); 505 mlog_errno(err);
487 goto bail; 506 goto bail;
488 } 507 }
489 508
490 bail: 509 bail:
491 status = err ? 0 : p_blkno; 510 status = err ? 0 : p_blkno;
492 511
493 return status; 512 return status;
494 } 513 }
495 514
496 static bool ocfs2_release_folio(struct folio * !! 515 static int ocfs2_releasepage(struct page *page, gfp_t wait)
497 { 516 {
498 if (!folio_buffers(folio)) !! 517 if (!page_has_buffers(page))
499 return false; !! 518 return 0;
500 return try_to_free_buffers(folio); !! 519 return try_to_free_buffers(page);
501 } 520 }
502 521
503 static void ocfs2_figure_cluster_boundaries(st 522 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
504 u3 523 u32 cpos,
505 un 524 unsigned int *start,
506 un 525 unsigned int *end)
507 { 526 {
508 unsigned int cluster_start = 0, cluste 527 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
509 528
510 if (unlikely(PAGE_SHIFT > osb->s_clust 529 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
511 unsigned int cpp; 530 unsigned int cpp;
512 531
513 cpp = 1 << (PAGE_SHIFT - osb-> 532 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
514 533
515 cluster_start = cpos % cpp; 534 cluster_start = cpos % cpp;
516 cluster_start = cluster_start 535 cluster_start = cluster_start << osb->s_clustersize_bits;
517 536
518 cluster_end = cluster_start + 537 cluster_end = cluster_start + osb->s_clustersize;
519 } 538 }
520 539
521 BUG_ON(cluster_start > PAGE_SIZE); 540 BUG_ON(cluster_start > PAGE_SIZE);
522 BUG_ON(cluster_end > PAGE_SIZE); 541 BUG_ON(cluster_end > PAGE_SIZE);
523 542
524 if (start) 543 if (start)
525 *start = cluster_start; 544 *start = cluster_start;
526 if (end) 545 if (end)
527 *end = cluster_end; 546 *end = cluster_end;
528 } 547 }
529 548
530 /* 549 /*
531 * 'from' and 'to' are the region in the page 550 * 'from' and 'to' are the region in the page to avoid zeroing.
532 * 551 *
533 * If pagesize > clustersize, this function wi 552 * If pagesize > clustersize, this function will avoid zeroing outside
534 * of the cluster boundary. 553 * of the cluster boundary.
535 * 554 *
536 * from == to == 0 is code for "zero the entir 555 * from == to == 0 is code for "zero the entire cluster region"
537 */ 556 */
538 static void ocfs2_clear_page_regions(struct pa 557 static void ocfs2_clear_page_regions(struct page *page,
539 struct oc 558 struct ocfs2_super *osb, u32 cpos,
540 unsigned 559 unsigned from, unsigned to)
541 { 560 {
542 void *kaddr; 561 void *kaddr;
543 unsigned int cluster_start, cluster_en 562 unsigned int cluster_start, cluster_end;
544 563
545 ocfs2_figure_cluster_boundaries(osb, c 564 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
546 565
547 kaddr = kmap_atomic(page); 566 kaddr = kmap_atomic(page);
548 567
549 if (from || to) { 568 if (from || to) {
550 if (from > cluster_start) 569 if (from > cluster_start)
551 memset(kaddr + cluster 570 memset(kaddr + cluster_start, 0, from - cluster_start);
552 if (to < cluster_end) 571 if (to < cluster_end)
553 memset(kaddr + to, 0, 572 memset(kaddr + to, 0, cluster_end - to);
554 } else { 573 } else {
555 memset(kaddr + cluster_start, 574 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
556 } 575 }
557 576
558 kunmap_atomic(kaddr); 577 kunmap_atomic(kaddr);
559 } 578 }
560 579
561 /* 580 /*
562 * Nonsparse file systems fully allocate befor 581 * Nonsparse file systems fully allocate before we get to the write
563 * code. This prevents ocfs2_write() from tagg 582 * code. This prevents ocfs2_write() from tagging the write as an
564 * allocating one, which means ocfs2_map_page_ 583 * allocating one, which means ocfs2_map_page_blocks() might try to
565 * read-in the blocks at the tail of our file. 584 * read-in the blocks at the tail of our file. Avoid reading them by
566 * testing i_size against each block offset. 585 * testing i_size against each block offset.
567 */ 586 */
568 static int ocfs2_should_read_blk(struct inode !! 587 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
569 unsigned int 588 unsigned int block_start)
570 { 589 {
571 u64 offset = folio_pos(folio) + block_ !! 590 u64 offset = page_offset(page) + block_start;
572 591
573 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 592 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
574 return 1; 593 return 1;
575 594
576 if (i_size_read(inode) > offset) 595 if (i_size_read(inode) > offset)
577 return 1; 596 return 1;
578 597
579 return 0; 598 return 0;
580 } 599 }
581 600
582 /* 601 /*
583 * Some of this taken from __block_write_begin 602 * Some of this taken from __block_write_begin(). We already have our
584 * mapping by now though, and the entire write 603 * mapping by now though, and the entire write will be allocating or
585 * it won't, so not much need to use BH_New. 604 * it won't, so not much need to use BH_New.
586 * 605 *
587 * This will also skip zeroing, which is handl 606 * This will also skip zeroing, which is handled externally.
588 */ 607 */
589 int ocfs2_map_page_blocks(struct page *page, u 608 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
590 struct inode *inode, 609 struct inode *inode, unsigned int from,
591 unsigned int to, int 610 unsigned int to, int new)
592 { 611 {
593 struct folio *folio = page_folio(page) <<
594 int ret = 0; 612 int ret = 0;
595 struct buffer_head *head, *bh, *wait[2 613 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
596 unsigned int block_end, block_start; 614 unsigned int block_end, block_start;
597 unsigned int bsize = i_blocksize(inode 615 unsigned int bsize = i_blocksize(inode);
598 616
599 head = folio_buffers(folio); !! 617 if (!page_has_buffers(page))
600 if (!head) !! 618 create_empty_buffers(page, bsize, 0);
601 head = create_empty_buffers(fo <<
602 619
>> 620 head = page_buffers(page);
603 for (bh = head, block_start = 0; bh != 621 for (bh = head, block_start = 0; bh != head || !block_start;
604 bh = bh->b_this_page, block_start 622 bh = bh->b_this_page, block_start += bsize) {
605 block_end = block_start + bsiz 623 block_end = block_start + bsize;
606 624
607 clear_buffer_new(bh); 625 clear_buffer_new(bh);
608 626
609 /* 627 /*
610 * Ignore blocks outside of ou 628 * Ignore blocks outside of our i/o range -
611 * they may belong to unalloca 629 * they may belong to unallocated clusters.
612 */ 630 */
613 if (block_start >= to || block 631 if (block_start >= to || block_end <= from) {
614 if (folio_test_uptodat !! 632 if (PageUptodate(page))
615 set_buffer_upt 633 set_buffer_uptodate(bh);
616 continue; 634 continue;
617 } 635 }
618 636
619 /* 637 /*
620 * For an allocating write wit 638 * For an allocating write with cluster size >= page
621 * size, we always write the e 639 * size, we always write the entire page.
622 */ 640 */
623 if (new) 641 if (new)
624 set_buffer_new(bh); 642 set_buffer_new(bh);
625 643
626 if (!buffer_mapped(bh)) { 644 if (!buffer_mapped(bh)) {
627 map_bh(bh, inode->i_sb 645 map_bh(bh, inode->i_sb, *p_blkno);
628 clean_bdev_bh_alias(bh !! 646 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
629 } 647 }
630 648
631 if (folio_test_uptodate(folio) !! 649 if (PageUptodate(page)) {
632 set_buffer_uptodate(bh !! 650 if (!buffer_uptodate(bh))
>> 651 set_buffer_uptodate(bh);
633 } else if (!buffer_uptodate(bh 652 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
634 !buffer_new(bh) && 653 !buffer_new(bh) &&
635 ocfs2_should_read_b !! 654 ocfs2_should_read_blk(inode, page, block_start) &&
636 (block_start < from 655 (block_start < from || block_end > to)) {
637 bh_read_nowait(bh, 0); !! 656 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
638 *wait_bh++=bh; 657 *wait_bh++=bh;
639 } 658 }
640 659
641 *p_blkno = *p_blkno + 1; 660 *p_blkno = *p_blkno + 1;
642 } 661 }
643 662
644 /* 663 /*
645 * If we issued read requests - let th 664 * If we issued read requests - let them complete.
646 */ 665 */
647 while(wait_bh > wait) { 666 while(wait_bh > wait) {
648 wait_on_buffer(*--wait_bh); 667 wait_on_buffer(*--wait_bh);
649 if (!buffer_uptodate(*wait_bh) 668 if (!buffer_uptodate(*wait_bh))
650 ret = -EIO; 669 ret = -EIO;
651 } 670 }
652 671
653 if (ret == 0 || !new) 672 if (ret == 0 || !new)
654 return ret; 673 return ret;
655 674
656 /* 675 /*
657 * If we get -EIO above, zero out any 676 * If we get -EIO above, zero out any newly allocated blocks
658 * to avoid exposing stale data. 677 * to avoid exposing stale data.
659 */ 678 */
660 bh = head; 679 bh = head;
661 block_start = 0; 680 block_start = 0;
662 do { 681 do {
663 block_end = block_start + bsiz 682 block_end = block_start + bsize;
664 if (block_end <= from) 683 if (block_end <= from)
665 goto next_bh; 684 goto next_bh;
666 if (block_start >= to) 685 if (block_start >= to)
667 break; 686 break;
668 687
669 folio_zero_range(folio, block_ !! 688 zero_user(page, block_start, bh->b_size);
670 set_buffer_uptodate(bh); 689 set_buffer_uptodate(bh);
671 mark_buffer_dirty(bh); 690 mark_buffer_dirty(bh);
672 691
673 next_bh: 692 next_bh:
674 block_start = block_end; 693 block_start = block_end;
675 bh = bh->b_this_page; 694 bh = bh->b_this_page;
676 } while (bh != head); 695 } while (bh != head);
677 696
678 return ret; 697 return ret;
679 } 698 }
680 699
681 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 700 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
682 #define OCFS2_MAX_CTXT_PAGES 1 701 #define OCFS2_MAX_CTXT_PAGES 1
683 #else 702 #else
684 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 703 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
685 #endif 704 #endif
686 705
687 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 706 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
688 707
689 struct ocfs2_unwritten_extent { 708 struct ocfs2_unwritten_extent {
690 struct list_head ue_node; 709 struct list_head ue_node;
691 struct list_head ue_ip_node; 710 struct list_head ue_ip_node;
692 u32 ue_cpos; 711 u32 ue_cpos;
693 u32 ue_phys; 712 u32 ue_phys;
694 }; 713 };
695 714
696 /* 715 /*
697 * Describe the state of a single cluster to b 716 * Describe the state of a single cluster to be written to.
698 */ 717 */
699 struct ocfs2_write_cluster_desc { 718 struct ocfs2_write_cluster_desc {
700 u32 c_cpos; 719 u32 c_cpos;
701 u32 c_phys; 720 u32 c_phys;
702 /* 721 /*
703 * Give this a unique field because c_ 722 * Give this a unique field because c_phys eventually gets
704 * filled. 723 * filled.
705 */ 724 */
706 unsigned c_new; 725 unsigned c_new;
707 unsigned c_clear_unwritten; 726 unsigned c_clear_unwritten;
708 unsigned c_needs_zero; 727 unsigned c_needs_zero;
709 }; 728 };
710 729
711 struct ocfs2_write_ctxt { 730 struct ocfs2_write_ctxt {
712 /* Logical cluster position / len of w 731 /* Logical cluster position / len of write */
713 u32 w_cpos 732 u32 w_cpos;
714 u32 w_clen 733 u32 w_clen;
715 734
716 /* First cluster allocated in a nonspa 735 /* First cluster allocated in a nonsparse extend */
717 u32 w_firs 736 u32 w_first_new_cpos;
718 737
719 /* Type of caller. Must be one of buff 738 /* Type of caller. Must be one of buffer, mmap, direct. */
720 ocfs2_write_type_t w_type 739 ocfs2_write_type_t w_type;
721 740
722 struct ocfs2_write_cluster_desc w_desc 741 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
723 742
724 /* 743 /*
725 * This is true if page_size > cluster 744 * This is true if page_size > cluster_size.
726 * 745 *
727 * It triggers a set of special cases 746 * It triggers a set of special cases during write which might
728 * have to deal with allocating writes 747 * have to deal with allocating writes to partial pages.
729 */ 748 */
730 unsigned int w_larg 749 unsigned int w_large_pages;
731 750
732 /* 751 /*
733 * Pages involved in this write. 752 * Pages involved in this write.
734 * 753 *
735 * w_target_page is the page being wri 754 * w_target_page is the page being written to by the user.
736 * 755 *
737 * w_pages is an array of pages which 756 * w_pages is an array of pages which always contains
738 * w_target_page, and in the case of a 757 * w_target_page, and in the case of an allocating write with
739 * page_size < cluster size, it will c 758 * page_size < cluster size, it will contain zero'd and mapped
740 * pages adjacent to w_target_page whi 759 * pages adjacent to w_target_page which need to be written
741 * out in so that future reads from th 760 * out in so that future reads from that region will get
742 * zero's. 761 * zero's.
743 */ 762 */
744 unsigned int w_num_ 763 unsigned int w_num_pages;
745 struct page *w_pag 764 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
746 struct page *w_tar 765 struct page *w_target_page;
747 766
748 /* 767 /*
749 * w_target_locked is used for page_mk 768 * w_target_locked is used for page_mkwrite path indicating no unlocking
750 * against w_target_page in ocfs2_writ 769 * against w_target_page in ocfs2_write_end_nolock.
751 */ 770 */
752 unsigned int w_targ 771 unsigned int w_target_locked:1;
753 772
754 /* 773 /*
755 * ocfs2_write_end() uses this to know 774 * ocfs2_write_end() uses this to know what the real range to
756 * write in the target should be. 775 * write in the target should be.
757 */ 776 */
758 unsigned int w_targ 777 unsigned int w_target_from;
759 unsigned int w_targ 778 unsigned int w_target_to;
760 779
761 /* 780 /*
762 * We could use journal_current_handle 781 * We could use journal_current_handle() but this is cleaner,
763 * IMHO -Mark 782 * IMHO -Mark
764 */ 783 */
765 handle_t *w_han 784 handle_t *w_handle;
766 785
767 struct buffer_head *w_di_ 786 struct buffer_head *w_di_bh;
768 787
769 struct ocfs2_cached_dealloc_ctxt w_dea 788 struct ocfs2_cached_dealloc_ctxt w_dealloc;
770 789
771 struct list_head w_unwr 790 struct list_head w_unwritten_list;
772 unsigned int w_unwr <<
773 }; 791 };
774 792
775 void ocfs2_unlock_and_free_pages(struct page * 793 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
776 { 794 {
777 int i; 795 int i;
778 796
779 for(i = 0; i < num_pages; i++) { 797 for(i = 0; i < num_pages; i++) {
780 if (pages[i]) { 798 if (pages[i]) {
781 unlock_page(pages[i]); 799 unlock_page(pages[i]);
782 mark_page_accessed(pag 800 mark_page_accessed(pages[i]);
783 put_page(pages[i]); 801 put_page(pages[i]);
784 } 802 }
785 } 803 }
786 } 804 }
787 805
788 static void ocfs2_unlock_pages(struct ocfs2_wr 806 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
789 { 807 {
790 int i; 808 int i;
791 809
792 /* 810 /*
793 * w_target_locked is only set to true 811 * w_target_locked is only set to true in the page_mkwrite() case.
794 * The intent is to allow us to lock t 812 * The intent is to allow us to lock the target page from write_begin()
795 * to write_end(). The caller must hol 813 * to write_end(). The caller must hold a ref on w_target_page.
796 */ 814 */
797 if (wc->w_target_locked) { 815 if (wc->w_target_locked) {
798 BUG_ON(!wc->w_target_page); 816 BUG_ON(!wc->w_target_page);
799 for (i = 0; i < wc->w_num_page 817 for (i = 0; i < wc->w_num_pages; i++) {
800 if (wc->w_target_page 818 if (wc->w_target_page == wc->w_pages[i]) {
801 wc->w_pages[i] 819 wc->w_pages[i] = NULL;
802 break; 820 break;
803 } 821 }
804 } 822 }
805 mark_page_accessed(wc->w_targe 823 mark_page_accessed(wc->w_target_page);
806 put_page(wc->w_target_page); 824 put_page(wc->w_target_page);
807 } 825 }
808 ocfs2_unlock_and_free_pages(wc->w_page 826 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
809 } 827 }
810 828
811 static void ocfs2_free_unwritten_list(struct i 829 static void ocfs2_free_unwritten_list(struct inode *inode,
812 struct list_h 830 struct list_head *head)
813 { 831 {
814 struct ocfs2_inode_info *oi = OCFS2_I( 832 struct ocfs2_inode_info *oi = OCFS2_I(inode);
815 struct ocfs2_unwritten_extent *ue = NU 833 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
816 834
817 list_for_each_entry_safe(ue, tmp, head 835 list_for_each_entry_safe(ue, tmp, head, ue_node) {
818 list_del(&ue->ue_node); 836 list_del(&ue->ue_node);
819 spin_lock(&oi->ip_lock); 837 spin_lock(&oi->ip_lock);
820 list_del(&ue->ue_ip_node); 838 list_del(&ue->ue_ip_node);
821 spin_unlock(&oi->ip_lock); 839 spin_unlock(&oi->ip_lock);
822 kfree(ue); 840 kfree(ue);
823 } 841 }
824 } 842 }
825 843
826 static void ocfs2_free_write_ctxt(struct inode 844 static void ocfs2_free_write_ctxt(struct inode *inode,
827 struct ocfs2 845 struct ocfs2_write_ctxt *wc)
828 { 846 {
829 ocfs2_free_unwritten_list(inode, &wc-> 847 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
830 ocfs2_unlock_pages(wc); 848 ocfs2_unlock_pages(wc);
831 brelse(wc->w_di_bh); 849 brelse(wc->w_di_bh);
832 kfree(wc); 850 kfree(wc);
833 } 851 }
834 852
835 static int ocfs2_alloc_write_ctxt(struct ocfs2 853 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
836 struct ocfs2 854 struct ocfs2_super *osb, loff_t pos,
837 unsigned len 855 unsigned len, ocfs2_write_type_t type,
838 struct buffe 856 struct buffer_head *di_bh)
839 { 857 {
840 u32 cend; 858 u32 cend;
841 struct ocfs2_write_ctxt *wc; 859 struct ocfs2_write_ctxt *wc;
842 860
843 wc = kzalloc(sizeof(struct ocfs2_write 861 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
844 if (!wc) 862 if (!wc)
845 return -ENOMEM; 863 return -ENOMEM;
846 864
847 wc->w_cpos = pos >> osb->s_clustersize 865 wc->w_cpos = pos >> osb->s_clustersize_bits;
848 wc->w_first_new_cpos = UINT_MAX; 866 wc->w_first_new_cpos = UINT_MAX;
849 cend = (pos + len - 1) >> osb->s_clust 867 cend = (pos + len - 1) >> osb->s_clustersize_bits;
850 wc->w_clen = cend - wc->w_cpos + 1; 868 wc->w_clen = cend - wc->w_cpos + 1;
851 get_bh(di_bh); 869 get_bh(di_bh);
852 wc->w_di_bh = di_bh; 870 wc->w_di_bh = di_bh;
853 wc->w_type = type; 871 wc->w_type = type;
854 872
855 if (unlikely(PAGE_SHIFT > osb->s_clust 873 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
856 wc->w_large_pages = 1; 874 wc->w_large_pages = 1;
857 else 875 else
858 wc->w_large_pages = 0; 876 wc->w_large_pages = 0;
859 877
860 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 878 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
861 INIT_LIST_HEAD(&wc->w_unwritten_list); 879 INIT_LIST_HEAD(&wc->w_unwritten_list);
862 880
863 *wcp = wc; 881 *wcp = wc;
864 882
865 return 0; 883 return 0;
866 } 884 }
867 885
868 /* 886 /*
869 * If a page has any new buffers, zero them ou 887 * If a page has any new buffers, zero them out here, and mark them uptodate
870 * and dirty so they'll be written out (in ord 888 * and dirty so they'll be written out (in order to prevent uninitialised
871 * block data from leaking). And clear the new 889 * block data from leaking). And clear the new bit.
872 */ 890 */
873 static void ocfs2_zero_new_buffers(struct page 891 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
874 { 892 {
875 unsigned int block_start, block_end; 893 unsigned int block_start, block_end;
876 struct buffer_head *head, *bh; 894 struct buffer_head *head, *bh;
877 895
878 BUG_ON(!PageLocked(page)); 896 BUG_ON(!PageLocked(page));
879 if (!page_has_buffers(page)) 897 if (!page_has_buffers(page))
880 return; 898 return;
881 899
882 bh = head = page_buffers(page); 900 bh = head = page_buffers(page);
883 block_start = 0; 901 block_start = 0;
884 do { 902 do {
885 block_end = block_start + bh-> 903 block_end = block_start + bh->b_size;
886 904
887 if (buffer_new(bh)) { 905 if (buffer_new(bh)) {
888 if (block_end > from & 906 if (block_end > from && block_start < to) {
889 if (!PageUptod 907 if (!PageUptodate(page)) {
890 unsign 908 unsigned start, end;
891 909
892 start 910 start = max(from, block_start);
893 end = 911 end = min(to, block_end);
894 912
895 zero_u 913 zero_user_segment(page, start, end);
896 set_bu 914 set_buffer_uptodate(bh);
897 } 915 }
898 916
899 clear_buffer_n 917 clear_buffer_new(bh);
900 mark_buffer_di 918 mark_buffer_dirty(bh);
901 } 919 }
902 } 920 }
903 921
904 block_start = block_end; 922 block_start = block_end;
905 bh = bh->b_this_page; 923 bh = bh->b_this_page;
906 } while (bh != head); 924 } while (bh != head);
907 } 925 }
908 926
909 /* 927 /*
910 * Only called when we have a failure during a 928 * Only called when we have a failure during allocating write to write
911 * zero's to the newly allocated region. 929 * zero's to the newly allocated region.
912 */ 930 */
913 static void ocfs2_write_failure(struct inode * 931 static void ocfs2_write_failure(struct inode *inode,
914 struct ocfs2_w 932 struct ocfs2_write_ctxt *wc,
915 loff_t user_po 933 loff_t user_pos, unsigned user_len)
916 { 934 {
917 int i; 935 int i;
918 unsigned from = user_pos & (PAGE_SIZE 936 unsigned from = user_pos & (PAGE_SIZE - 1),
919 to = user_pos + user_len; 937 to = user_pos + user_len;
920 struct page *tmppage; 938 struct page *tmppage;
921 939
922 if (wc->w_target_page) 940 if (wc->w_target_page)
923 ocfs2_zero_new_buffers(wc->w_t 941 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
924 942
925 for(i = 0; i < wc->w_num_pages; i++) { 943 for(i = 0; i < wc->w_num_pages; i++) {
926 tmppage = wc->w_pages[i]; 944 tmppage = wc->w_pages[i];
927 945
928 if (tmppage && page_has_buffer 946 if (tmppage && page_has_buffers(tmppage)) {
929 if (ocfs2_should_order 947 if (ocfs2_should_order_data(inode))
930 ocfs2_jbd2_ino !! 948 ocfs2_jbd2_file_inode(wc->w_handle, inode);
931 <<
932 949
933 block_commit_write(tmp 950 block_commit_write(tmppage, from, to);
934 } 951 }
935 } 952 }
936 } 953 }
937 954
938 static int ocfs2_prepare_page_for_write(struct 955 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
939 struct 956 struct ocfs2_write_ctxt *wc,
940 struct 957 struct page *page, u32 cpos,
941 loff_t 958 loff_t user_pos, unsigned user_len,
942 int ne 959 int new)
943 { 960 {
944 int ret; 961 int ret;
945 unsigned int map_from = 0, map_to = 0; 962 unsigned int map_from = 0, map_to = 0;
946 unsigned int cluster_start, cluster_en 963 unsigned int cluster_start, cluster_end;
947 unsigned int user_data_from = 0, user_ 964 unsigned int user_data_from = 0, user_data_to = 0;
948 965
949 ocfs2_figure_cluster_boundaries(OCFS2_ 966 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
950 &clust 967 &cluster_start, &cluster_end);
951 968
952 /* treat the write as new if the a hol 969 /* treat the write as new if the a hole/lseek spanned across
953 * the page boundary. 970 * the page boundary.
954 */ 971 */
955 new = new | ((i_size_read(inode) <= pa 972 new = new | ((i_size_read(inode) <= page_offset(page)) &&
956 (page_offset(page) <= 973 (page_offset(page) <= user_pos));
957 974
958 if (page == wc->w_target_page) { 975 if (page == wc->w_target_page) {
959 map_from = user_pos & (PAGE_SI 976 map_from = user_pos & (PAGE_SIZE - 1);
960 map_to = map_from + user_len; 977 map_to = map_from + user_len;
961 978
962 if (new) 979 if (new)
963 ret = ocfs2_map_page_b 980 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
964 981 cluster_start, cluster_end,
965 982 new);
966 else 983 else
967 ret = ocfs2_map_page_b 984 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
968 985 map_from, map_to, new);
969 if (ret) { 986 if (ret) {
970 mlog_errno(ret); 987 mlog_errno(ret);
971 goto out; 988 goto out;
972 } 989 }
973 990
974 user_data_from = map_from; 991 user_data_from = map_from;
975 user_data_to = map_to; 992 user_data_to = map_to;
976 if (new) { 993 if (new) {
977 map_from = cluster_sta 994 map_from = cluster_start;
978 map_to = cluster_end; 995 map_to = cluster_end;
979 } 996 }
980 } else { 997 } else {
981 /* 998 /*
982 * If we haven't allocated the 999 * If we haven't allocated the new page yet, we
983 * shouldn't be writing it out 1000 * shouldn't be writing it out without copying user
984 * data. This is likely a math 1001 * data. This is likely a math error from the caller.
985 */ 1002 */
986 BUG_ON(!new); 1003 BUG_ON(!new);
987 1004
988 map_from = cluster_start; 1005 map_from = cluster_start;
989 map_to = cluster_end; 1006 map_to = cluster_end;
990 1007
991 ret = ocfs2_map_page_blocks(pa 1008 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
992 cl 1009 cluster_start, cluster_end, new);
993 if (ret) { 1010 if (ret) {
994 mlog_errno(ret); 1011 mlog_errno(ret);
995 goto out; 1012 goto out;
996 } 1013 }
997 } 1014 }
998 1015
999 /* 1016 /*
1000 * Parts of newly allocated pages nee 1017 * Parts of newly allocated pages need to be zero'd.
1001 * 1018 *
1002 * Above, we have also rewritten 'to' 1019 * Above, we have also rewritten 'to' and 'from' - as far as
1003 * the rest of the function is concer 1020 * the rest of the function is concerned, the entire cluster
1004 * range inside of a page needs to be 1021 * range inside of a page needs to be written.
1005 * 1022 *
1006 * We can skip this if the page is up 1023 * We can skip this if the page is up to date - it's already
1007 * been zero'd from being read in as 1024 * been zero'd from being read in as a hole.
1008 */ 1025 */
1009 if (new && !PageUptodate(page)) 1026 if (new && !PageUptodate(page))
1010 ocfs2_clear_page_regions(page 1027 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1011 cpos 1028 cpos, user_data_from, user_data_to);
1012 1029
1013 flush_dcache_page(page); 1030 flush_dcache_page(page);
1014 1031
1015 out: 1032 out:
1016 return ret; 1033 return ret;
1017 } 1034 }
1018 1035
1019 /* 1036 /*
1020 * This function will only grab one clusters 1037 * This function will only grab one clusters worth of pages.
1021 */ 1038 */
1022 static int ocfs2_grab_pages_for_write(struct 1039 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1023 struct 1040 struct ocfs2_write_ctxt *wc,
1024 u32 cpo 1041 u32 cpos, loff_t user_pos,
1025 unsigne 1042 unsigned user_len, int new,
1026 struct 1043 struct page *mmap_page)
1027 { 1044 {
1028 int ret = 0, i; 1045 int ret = 0, i;
1029 unsigned long start, target_index, en 1046 unsigned long start, target_index, end_index, index;
1030 struct inode *inode = mapping->host; 1047 struct inode *inode = mapping->host;
1031 loff_t last_byte; 1048 loff_t last_byte;
1032 1049
1033 target_index = user_pos >> PAGE_SHIFT 1050 target_index = user_pos >> PAGE_SHIFT;
1034 1051
1035 /* 1052 /*
1036 * Figure out how many pages we'll be 1053 * Figure out how many pages we'll be manipulating here. For
1037 * non allocating write, we just chan 1054 * non allocating write, we just change the one
1038 * page. Otherwise, we'll need a whol 1055 * page. Otherwise, we'll need a whole clusters worth. If we're
1039 * writing past i_size, we only need 1056 * writing past i_size, we only need enough pages to cover the
1040 * last page of the write. 1057 * last page of the write.
1041 */ 1058 */
1042 if (new) { 1059 if (new) {
1043 wc->w_num_pages = ocfs2_pages 1060 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1044 start = ocfs2_align_clusters_ 1061 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1045 /* 1062 /*
1046 * We need the index *past* t 1063 * We need the index *past* the last page we could possibly
1047 * touch. This is the page p 1064 * touch. This is the page past the end of the write or
1048 * i_size, whichever is great 1065 * i_size, whichever is greater.
1049 */ 1066 */
1050 last_byte = max(user_pos + us 1067 last_byte = max(user_pos + user_len, i_size_read(inode));
1051 BUG_ON(last_byte < 1); 1068 BUG_ON(last_byte < 1);
1052 end_index = ((last_byte - 1) 1069 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
1053 if ((start + wc->w_num_pages) 1070 if ((start + wc->w_num_pages) > end_index)
1054 wc->w_num_pages = end 1071 wc->w_num_pages = end_index - start;
1055 } else { 1072 } else {
1056 wc->w_num_pages = 1; 1073 wc->w_num_pages = 1;
1057 start = target_index; 1074 start = target_index;
1058 } 1075 }
1059 end_index = (user_pos + user_len - 1) 1076 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
1060 1077
1061 for(i = 0; i < wc->w_num_pages; i++) 1078 for(i = 0; i < wc->w_num_pages; i++) {
1062 index = start + i; 1079 index = start + i;
1063 1080
1064 if (index >= target_index && 1081 if (index >= target_index && index <= end_index &&
1065 wc->w_type == OCFS2_WRITE 1082 wc->w_type == OCFS2_WRITE_MMAP) {
1066 /* 1083 /*
1067 * ocfs2_pagemkwrite( 1084 * ocfs2_pagemkwrite() is a little different
1068 * and wants us to di 1085 * and wants us to directly use the page
1069 * passed in. 1086 * passed in.
1070 */ 1087 */
1071 lock_page(mmap_page); 1088 lock_page(mmap_page);
1072 1089
1073 /* Exit and let the c 1090 /* Exit and let the caller retry */
1074 if (mmap_page->mappin 1091 if (mmap_page->mapping != mapping) {
1075 WARN_ON(mmap_ 1092 WARN_ON(mmap_page->mapping);
1076 unlock_page(m 1093 unlock_page(mmap_page);
1077 ret = -EAGAIN 1094 ret = -EAGAIN;
1078 goto out; 1095 goto out;
1079 } 1096 }
1080 1097
1081 get_page(mmap_page); 1098 get_page(mmap_page);
1082 wc->w_pages[i] = mmap 1099 wc->w_pages[i] = mmap_page;
1083 wc->w_target_locked = 1100 wc->w_target_locked = true;
1084 } else if (index >= target_in 1101 } else if (index >= target_index && index <= end_index &&
1085 wc->w_type == OCFS 1102 wc->w_type == OCFS2_WRITE_DIRECT) {
1086 /* Direct write has n 1103 /* Direct write has no mapping page. */
1087 wc->w_pages[i] = NULL 1104 wc->w_pages[i] = NULL;
1088 continue; 1105 continue;
1089 } else { 1106 } else {
1090 wc->w_pages[i] = find 1107 wc->w_pages[i] = find_or_create_page(mapping, index,
1091 1108 GFP_NOFS);
1092 if (!wc->w_pages[i]) 1109 if (!wc->w_pages[i]) {
1093 ret = -ENOMEM 1110 ret = -ENOMEM;
1094 mlog_errno(re 1111 mlog_errno(ret);
1095 goto out; 1112 goto out;
1096 } 1113 }
1097 } 1114 }
1098 wait_for_stable_page(wc->w_pa 1115 wait_for_stable_page(wc->w_pages[i]);
1099 1116
1100 if (index == target_index) 1117 if (index == target_index)
1101 wc->w_target_page = w 1118 wc->w_target_page = wc->w_pages[i];
1102 } 1119 }
1103 out: 1120 out:
1104 if (ret) 1121 if (ret)
1105 wc->w_target_locked = false; 1122 wc->w_target_locked = false;
1106 return ret; 1123 return ret;
1107 } 1124 }
1108 1125
1109 /* 1126 /*
1110 * Prepare a single cluster for write one clu 1127 * Prepare a single cluster for write one cluster into the file.
1111 */ 1128 */
1112 static int ocfs2_write_cluster(struct address 1129 static int ocfs2_write_cluster(struct address_space *mapping,
1113 u32 *phys, uns 1130 u32 *phys, unsigned int new,
1114 unsigned int c 1131 unsigned int clear_unwritten,
1115 unsigned int s 1132 unsigned int should_zero,
1116 struct ocfs2_a 1133 struct ocfs2_alloc_context *data_ac,
1117 struct ocfs2_a 1134 struct ocfs2_alloc_context *meta_ac,
1118 struct ocfs2_w 1135 struct ocfs2_write_ctxt *wc, u32 cpos,
1119 loff_t user_po 1136 loff_t user_pos, unsigned user_len)
1120 { 1137 {
1121 int ret, i; 1138 int ret, i;
1122 u64 p_blkno; 1139 u64 p_blkno;
1123 struct inode *inode = mapping->host; 1140 struct inode *inode = mapping->host;
1124 struct ocfs2_extent_tree et; 1141 struct ocfs2_extent_tree et;
1125 int bpc = ocfs2_clusters_to_blocks(in 1142 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1126 1143
1127 if (new) { 1144 if (new) {
1128 u32 tmp_pos; 1145 u32 tmp_pos;
1129 1146
1130 /* 1147 /*
1131 * This is safe to call with 1148 * This is safe to call with the page locks - it won't take
1132 * any additional semaphores 1149 * any additional semaphores or cluster locks.
1133 */ 1150 */
1134 tmp_pos = cpos; 1151 tmp_pos = cpos;
1135 ret = ocfs2_add_inode_data(OC 1152 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1136 &t 1153 &tmp_pos, 1, !clear_unwritten,
1137 wc 1154 wc->w_di_bh, wc->w_handle,
1138 da 1155 data_ac, meta_ac, NULL);
1139 /* 1156 /*
1140 * This shouldn't happen beca 1157 * This shouldn't happen because we must have already
1141 * calculated the correct met 1158 * calculated the correct meta data allocation required. The
1142 * internal tree allocation c 1159 * internal tree allocation code should know how to increase
1143 * transaction credits itself 1160 * transaction credits itself.
1144 * 1161 *
1145 * If need be, we could handl 1162 * If need be, we could handle -EAGAIN for a
1146 * RESTART_TRANS here. 1163 * RESTART_TRANS here.
1147 */ 1164 */
1148 mlog_bug_on_msg(ret == -EAGAI 1165 mlog_bug_on_msg(ret == -EAGAIN,
1149 "Inode %llu: 1166 "Inode %llu: EAGAIN return during allocation.\n",
1150 (unsigned lon 1167 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1151 if (ret < 0) { 1168 if (ret < 0) {
1152 mlog_errno(ret); 1169 mlog_errno(ret);
1153 goto out; 1170 goto out;
1154 } 1171 }
1155 } else if (clear_unwritten) { 1172 } else if (clear_unwritten) {
1156 ocfs2_init_dinode_extent_tree 1173 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1157 1174 wc->w_di_bh);
1158 ret = ocfs2_mark_extent_writt 1175 ret = ocfs2_mark_extent_written(inode, &et,
1159 1176 wc->w_handle, cpos, 1, *phys,
1160 1177 meta_ac, &wc->w_dealloc);
1161 if (ret < 0) { 1178 if (ret < 0) {
1162 mlog_errno(ret); 1179 mlog_errno(ret);
1163 goto out; 1180 goto out;
1164 } 1181 }
1165 } 1182 }
1166 1183
1167 /* 1184 /*
1168 * The only reason this should fail i 1185 * The only reason this should fail is due to an inability to
1169 * find the extent added. 1186 * find the extent added.
1170 */ 1187 */
1171 ret = ocfs2_get_clusters(inode, cpos, 1188 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
1172 if (ret < 0) { 1189 if (ret < 0) {
1173 mlog(ML_ERROR, "Get physical 1190 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1174 "at logical clust 1191 "at logical cluster %u",
1175 (unsigned long lo 1192 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
1176 goto out; 1193 goto out;
1177 } 1194 }
1178 1195
1179 BUG_ON(*phys == 0); 1196 BUG_ON(*phys == 0);
1180 1197
1181 p_blkno = ocfs2_clusters_to_blocks(in 1198 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1182 if (!should_zero) 1199 if (!should_zero)
1183 p_blkno += (user_pos >> inode 1200 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
1184 1201
1185 for(i = 0; i < wc->w_num_pages; i++) 1202 for(i = 0; i < wc->w_num_pages; i++) {
1186 int tmpret; 1203 int tmpret;
1187 1204
1188 /* This is the direct io targ 1205 /* This is the direct io target page. */
1189 if (wc->w_pages[i] == NULL) { 1206 if (wc->w_pages[i] == NULL) {
1190 p_blkno++; 1207 p_blkno++;
1191 continue; 1208 continue;
1192 } 1209 }
1193 1210
1194 tmpret = ocfs2_prepare_page_f 1211 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1195 1212 wc->w_pages[i], cpos,
1196 1213 user_pos, user_len,
1197 1214 should_zero);
1198 if (tmpret) { 1215 if (tmpret) {
1199 mlog_errno(tmpret); 1216 mlog_errno(tmpret);
1200 if (ret == 0) 1217 if (ret == 0)
1201 ret = tmpret; 1218 ret = tmpret;
1202 } 1219 }
1203 } 1220 }
1204 1221
1205 /* 1222 /*
1206 * We only have cleanup to do in case 1223 * We only have cleanup to do in case of allocating write.
1207 */ 1224 */
1208 if (ret && new) 1225 if (ret && new)
1209 ocfs2_write_failure(inode, wc 1226 ocfs2_write_failure(inode, wc, user_pos, user_len);
1210 1227
1211 out: 1228 out:
1212 1229
1213 return ret; 1230 return ret;
1214 } 1231 }
1215 1232
1216 static int ocfs2_write_cluster_by_desc(struct 1233 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1217 struct 1234 struct ocfs2_alloc_context *data_ac,
1218 struct 1235 struct ocfs2_alloc_context *meta_ac,
1219 struct 1236 struct ocfs2_write_ctxt *wc,
1220 loff_t 1237 loff_t pos, unsigned len)
1221 { 1238 {
1222 int ret, i; 1239 int ret, i;
1223 loff_t cluster_off; 1240 loff_t cluster_off;
1224 unsigned int local_len = len; 1241 unsigned int local_len = len;
1225 struct ocfs2_write_cluster_desc *desc 1242 struct ocfs2_write_cluster_desc *desc;
1226 struct ocfs2_super *osb = OCFS2_SB(ma 1243 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1227 1244
1228 for (i = 0; i < wc->w_clen; i++) { 1245 for (i = 0; i < wc->w_clen; i++) {
1229 desc = &wc->w_desc[i]; 1246 desc = &wc->w_desc[i];
1230 1247
1231 /* 1248 /*
1232 * We have to make sure that 1249 * We have to make sure that the total write passed in
1233 * doesn't extend past a sing 1250 * doesn't extend past a single cluster.
1234 */ 1251 */
1235 local_len = len; 1252 local_len = len;
1236 cluster_off = pos & (osb->s_c 1253 cluster_off = pos & (osb->s_clustersize - 1);
1237 if ((cluster_off + local_len) 1254 if ((cluster_off + local_len) > osb->s_clustersize)
1238 local_len = osb->s_cl 1255 local_len = osb->s_clustersize - cluster_off;
1239 1256
1240 ret = ocfs2_write_cluster(map 1257 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
1241 des 1258 desc->c_new,
1242 des 1259 desc->c_clear_unwritten,
1243 des 1260 desc->c_needs_zero,
1244 dat 1261 data_ac, meta_ac,
1245 wc, 1262 wc, desc->c_cpos, pos, local_len);
1246 if (ret) { 1263 if (ret) {
1247 mlog_errno(ret); 1264 mlog_errno(ret);
1248 goto out; 1265 goto out;
1249 } 1266 }
1250 1267
1251 len -= local_len; 1268 len -= local_len;
1252 pos += local_len; 1269 pos += local_len;
1253 } 1270 }
1254 1271
1255 ret = 0; 1272 ret = 0;
1256 out: 1273 out:
1257 return ret; 1274 return ret;
1258 } 1275 }
1259 1276
1260 /* 1277 /*
1261 * ocfs2_write_end() wants to know which part 1278 * ocfs2_write_end() wants to know which parts of the target page it
1262 * should complete the write on. It's easiest 1279 * should complete the write on. It's easiest to compute them ahead of
1263 * time when a more complete view of the writ 1280 * time when a more complete view of the write is available.
1264 */ 1281 */
1265 static void ocfs2_set_target_boundaries(struc 1282 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1266 struc 1283 struct ocfs2_write_ctxt *wc,
1267 loff_ 1284 loff_t pos, unsigned len, int alloc)
1268 { 1285 {
1269 struct ocfs2_write_cluster_desc *desc 1286 struct ocfs2_write_cluster_desc *desc;
1270 1287
1271 wc->w_target_from = pos & (PAGE_SIZE 1288 wc->w_target_from = pos & (PAGE_SIZE - 1);
1272 wc->w_target_to = wc->w_target_from + 1289 wc->w_target_to = wc->w_target_from + len;
1273 1290
1274 if (alloc == 0) 1291 if (alloc == 0)
1275 return; 1292 return;
1276 1293
1277 /* 1294 /*
1278 * Allocating write - we may have dif 1295 * Allocating write - we may have different boundaries based
1279 * on page size and cluster size. 1296 * on page size and cluster size.
1280 * 1297 *
1281 * NOTE: We can no longer compute one 1298 * NOTE: We can no longer compute one value from the other as
1282 * the actual write length and user p 1299 * the actual write length and user provided length may be
1283 * different. 1300 * different.
1284 */ 1301 */
1285 1302
1286 if (wc->w_large_pages) { 1303 if (wc->w_large_pages) {
1287 /* 1304 /*
1288 * We only care about the 1st 1305 * We only care about the 1st and last cluster within
1289 * our range and whether they 1306 * our range and whether they should be zero'd or not. Either
1290 * value may be extended out 1307 * value may be extended out to the start/end of a
1291 * newly allocated cluster. 1308 * newly allocated cluster.
1292 */ 1309 */
1293 desc = &wc->w_desc[0]; 1310 desc = &wc->w_desc[0];
1294 if (desc->c_needs_zero) 1311 if (desc->c_needs_zero)
1295 ocfs2_figure_cluster_ 1312 ocfs2_figure_cluster_boundaries(osb,
1296 1313 desc->c_cpos,
1297 1314 &wc->w_target_from,
1298 1315 NULL);
1299 1316
1300 desc = &wc->w_desc[wc->w_clen 1317 desc = &wc->w_desc[wc->w_clen - 1];
1301 if (desc->c_needs_zero) 1318 if (desc->c_needs_zero)
1302 ocfs2_figure_cluster_ 1319 ocfs2_figure_cluster_boundaries(osb,
1303 1320 desc->c_cpos,
1304 1321 NULL,
1305 1322 &wc->w_target_to);
1306 } else { 1323 } else {
1307 wc->w_target_from = 0; 1324 wc->w_target_from = 0;
1308 wc->w_target_to = PAGE_SIZE; 1325 wc->w_target_to = PAGE_SIZE;
1309 } 1326 }
1310 } 1327 }
1311 1328
1312 /* 1329 /*
1313 * Check if this extent is marked UNWRITTEN b 1330 * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to
1314 * do the zero work. And should not to clear 1331 * do the zero work. And should not to clear UNWRITTEN since it will be cleared
1315 * by the direct io procedure. 1332 * by the direct io procedure.
1316 * If this is a new extent that allocated by 1333 * If this is a new extent that allocated by direct io, we should mark it in
1317 * the ip_unwritten_list. 1334 * the ip_unwritten_list.
1318 */ 1335 */
1319 static int ocfs2_unwritten_check(struct inode 1336 static int ocfs2_unwritten_check(struct inode *inode,
1320 struct ocfs2 1337 struct ocfs2_write_ctxt *wc,
1321 struct ocfs2 1338 struct ocfs2_write_cluster_desc *desc)
1322 { 1339 {
1323 struct ocfs2_inode_info *oi = OCFS2_I 1340 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1324 struct ocfs2_unwritten_extent *ue = N 1341 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
1325 int ret = 0; 1342 int ret = 0;
1326 1343
1327 if (!desc->c_needs_zero) 1344 if (!desc->c_needs_zero)
1328 return 0; 1345 return 0;
1329 1346
1330 retry: 1347 retry:
1331 spin_lock(&oi->ip_lock); 1348 spin_lock(&oi->ip_lock);
1332 /* Needs not to zero no metter buffer 1349 /* Needs not to zero no metter buffer or direct. The one who is zero
1333 * the cluster is doing zero. And he 1350 * the cluster is doing zero. And he will clear unwritten after all
1334 * cluster io finished. */ 1351 * cluster io finished. */
1335 list_for_each_entry(ue, &oi->ip_unwri 1352 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1336 if (desc->c_cpos == ue->ue_cp 1353 if (desc->c_cpos == ue->ue_cpos) {
1337 BUG_ON(desc->c_new); 1354 BUG_ON(desc->c_new);
1338 desc->c_needs_zero = 1355 desc->c_needs_zero = 0;
1339 desc->c_clear_unwritt 1356 desc->c_clear_unwritten = 0;
1340 goto unlock; 1357 goto unlock;
1341 } 1358 }
1342 } 1359 }
1343 1360
1344 if (wc->w_type != OCFS2_WRITE_DIRECT) 1361 if (wc->w_type != OCFS2_WRITE_DIRECT)
1345 goto unlock; 1362 goto unlock;
1346 1363
1347 if (new == NULL) { 1364 if (new == NULL) {
1348 spin_unlock(&oi->ip_lock); 1365 spin_unlock(&oi->ip_lock);
1349 new = kmalloc(sizeof(struct o 1366 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1350 GFP_NOFS); 1367 GFP_NOFS);
1351 if (new == NULL) { 1368 if (new == NULL) {
1352 ret = -ENOMEM; 1369 ret = -ENOMEM;
1353 goto out; 1370 goto out;
1354 } 1371 }
1355 goto retry; 1372 goto retry;
1356 } 1373 }
1357 /* This direct write will doing zero. 1374 /* This direct write will doing zero. */
1358 new->ue_cpos = desc->c_cpos; 1375 new->ue_cpos = desc->c_cpos;
1359 new->ue_phys = desc->c_phys; 1376 new->ue_phys = desc->c_phys;
1360 desc->c_clear_unwritten = 0; 1377 desc->c_clear_unwritten = 0;
1361 list_add_tail(&new->ue_ip_node, &oi-> 1378 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1362 list_add_tail(&new->ue_node, &wc->w_u 1379 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
1363 wc->w_unwritten_count++; <<
1364 new = NULL; 1380 new = NULL;
1365 unlock: 1381 unlock:
1366 spin_unlock(&oi->ip_lock); 1382 spin_unlock(&oi->ip_lock);
1367 out: 1383 out:
1368 kfree(new); !! 1384 if (new)
>> 1385 kfree(new);
1369 return ret; 1386 return ret;
1370 } 1387 }
1371 1388
1372 /* 1389 /*
1373 * Populate each single-cluster write descrip 1390 * Populate each single-cluster write descriptor in the write context
1374 * with information about the i/o to be done. 1391 * with information about the i/o to be done.
1375 * 1392 *
1376 * Returns the number of clusters that will h 1393 * Returns the number of clusters that will have to be allocated, as
1377 * well as a worst case estimate of the numbe 1394 * well as a worst case estimate of the number of extent records that
1378 * would have to be created during a write to 1395 * would have to be created during a write to an unwritten region.
1379 */ 1396 */
1380 static int ocfs2_populate_write_desc(struct i 1397 static int ocfs2_populate_write_desc(struct inode *inode,
1381 struct o 1398 struct ocfs2_write_ctxt *wc,
1382 unsigned 1399 unsigned int *clusters_to_alloc,
1383 unsigned 1400 unsigned int *extents_to_split)
1384 { 1401 {
1385 int ret; 1402 int ret;
1386 struct ocfs2_write_cluster_desc *desc 1403 struct ocfs2_write_cluster_desc *desc;
1387 unsigned int num_clusters = 0; 1404 unsigned int num_clusters = 0;
1388 unsigned int ext_flags = 0; 1405 unsigned int ext_flags = 0;
1389 u32 phys = 0; 1406 u32 phys = 0;
1390 int i; 1407 int i;
1391 1408
1392 *clusters_to_alloc = 0; 1409 *clusters_to_alloc = 0;
1393 *extents_to_split = 0; 1410 *extents_to_split = 0;
1394 1411
1395 for (i = 0; i < wc->w_clen; i++) { 1412 for (i = 0; i < wc->w_clen; i++) {
1396 desc = &wc->w_desc[i]; 1413 desc = &wc->w_desc[i];
1397 desc->c_cpos = wc->w_cpos + i 1414 desc->c_cpos = wc->w_cpos + i;
1398 1415
1399 if (num_clusters == 0) { 1416 if (num_clusters == 0) {
1400 /* 1417 /*
1401 * Need to look up th 1418 * Need to look up the next extent record.
1402 */ 1419 */
1403 ret = ocfs2_get_clust 1420 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1404 1421 &num_clusters, &ext_flags);
1405 if (ret) { 1422 if (ret) {
1406 mlog_errno(re 1423 mlog_errno(ret);
1407 goto out; 1424 goto out;
1408 } 1425 }
1409 1426
1410 /* We should already 1427 /* We should already CoW the refcountd extent. */
1411 BUG_ON(ext_flags & OC 1428 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1412 1429
1413 /* 1430 /*
1414 * Assume worst case 1431 * Assume worst case - that we're writing in
1415 * the middle of the 1432 * the middle of the extent.
1416 * 1433 *
1417 * We can assume that 1434 * We can assume that the write proceeds from
1418 * left to right, in 1435 * left to right, in which case the extent
1419 * insert code is sma 1436 * insert code is smart enough to coalesce the
1420 * next splits into t 1437 * next splits into the previous records created.
1421 */ 1438 */
1422 if (ext_flags & OCFS2 1439 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1423 *extents_to_s 1440 *extents_to_split = *extents_to_split + 2;
1424 } else if (phys) { 1441 } else if (phys) {
1425 /* 1442 /*
1426 * Only increment phy 1443 * Only increment phys if it doesn't describe
1427 * a hole. 1444 * a hole.
1428 */ 1445 */
1429 phys++; 1446 phys++;
1430 } 1447 }
1431 1448
1432 /* 1449 /*
1433 * If w_first_new_cpos is < U 1450 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1434 * file that got extended. w 1451 * file that got extended. w_first_new_cpos tells us
1435 * where the newly allocated 1452 * where the newly allocated clusters are so we can
1436 * zero them. 1453 * zero them.
1437 */ 1454 */
1438 if (desc->c_cpos >= wc->w_fir 1455 if (desc->c_cpos >= wc->w_first_new_cpos) {
1439 BUG_ON(phys == 0); 1456 BUG_ON(phys == 0);
1440 desc->c_needs_zero = 1457 desc->c_needs_zero = 1;
1441 } 1458 }
1442 1459
1443 desc->c_phys = phys; 1460 desc->c_phys = phys;
1444 if (phys == 0) { 1461 if (phys == 0) {
1445 desc->c_new = 1; 1462 desc->c_new = 1;
1446 desc->c_needs_zero = 1463 desc->c_needs_zero = 1;
1447 desc->c_clear_unwritt 1464 desc->c_clear_unwritten = 1;
1448 *clusters_to_alloc = 1465 *clusters_to_alloc = *clusters_to_alloc + 1;
1449 } 1466 }
1450 1467
1451 if (ext_flags & OCFS2_EXT_UNW 1468 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1452 desc->c_clear_unwritt 1469 desc->c_clear_unwritten = 1;
1453 desc->c_needs_zero = 1470 desc->c_needs_zero = 1;
1454 } 1471 }
1455 1472
1456 ret = ocfs2_unwritten_check(i 1473 ret = ocfs2_unwritten_check(inode, wc, desc);
1457 if (ret) { 1474 if (ret) {
1458 mlog_errno(ret); 1475 mlog_errno(ret);
1459 goto out; 1476 goto out;
1460 } 1477 }
1461 1478
1462 num_clusters--; 1479 num_clusters--;
1463 } 1480 }
1464 1481
1465 ret = 0; 1482 ret = 0;
1466 out: 1483 out:
1467 return ret; 1484 return ret;
1468 } 1485 }
1469 1486
1470 static int ocfs2_write_begin_inline(struct ad 1487 static int ocfs2_write_begin_inline(struct address_space *mapping,
1471 struct in 1488 struct inode *inode,
1472 struct oc 1489 struct ocfs2_write_ctxt *wc)
1473 { 1490 {
1474 int ret; 1491 int ret;
1475 struct ocfs2_super *osb = OCFS2_SB(in 1492 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1476 struct page *page; 1493 struct page *page;
1477 handle_t *handle; 1494 handle_t *handle;
1478 struct ocfs2_dinode *di = (struct ocf 1495 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1479 1496
1480 handle = ocfs2_start_trans(osb, OCFS2 1497 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1481 if (IS_ERR(handle)) { 1498 if (IS_ERR(handle)) {
1482 ret = PTR_ERR(handle); 1499 ret = PTR_ERR(handle);
1483 mlog_errno(ret); 1500 mlog_errno(ret);
1484 goto out; 1501 goto out;
1485 } 1502 }
1486 1503
1487 page = find_or_create_page(mapping, 0 1504 page = find_or_create_page(mapping, 0, GFP_NOFS);
1488 if (!page) { 1505 if (!page) {
1489 ocfs2_commit_trans(osb, handl 1506 ocfs2_commit_trans(osb, handle);
1490 ret = -ENOMEM; 1507 ret = -ENOMEM;
1491 mlog_errno(ret); 1508 mlog_errno(ret);
1492 goto out; 1509 goto out;
1493 } 1510 }
1494 /* 1511 /*
1495 * If we don't set w_num_pages then t 1512 * If we don't set w_num_pages then this page won't get unlocked
1496 * and freed on cleanup of the write 1513 * and freed on cleanup of the write context.
1497 */ 1514 */
1498 wc->w_pages[0] = wc->w_target_page = 1515 wc->w_pages[0] = wc->w_target_page = page;
1499 wc->w_num_pages = 1; 1516 wc->w_num_pages = 1;
1500 1517
1501 ret = ocfs2_journal_access_di(handle, 1518 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1502 OCFS2_J 1519 OCFS2_JOURNAL_ACCESS_WRITE);
1503 if (ret) { 1520 if (ret) {
1504 ocfs2_commit_trans(osb, handl 1521 ocfs2_commit_trans(osb, handle);
1505 1522
1506 mlog_errno(ret); 1523 mlog_errno(ret);
1507 goto out; 1524 goto out;
1508 } 1525 }
1509 1526
1510 if (!(OCFS2_I(inode)->ip_dyn_features 1527 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1511 ocfs2_set_inode_data_inline(i 1528 ocfs2_set_inode_data_inline(inode, di);
1512 1529
1513 if (!PageUptodate(page)) { 1530 if (!PageUptodate(page)) {
1514 ret = ocfs2_read_inline_data( 1531 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1515 if (ret) { 1532 if (ret) {
1516 ocfs2_commit_trans(os 1533 ocfs2_commit_trans(osb, handle);
1517 1534
1518 goto out; 1535 goto out;
1519 } 1536 }
1520 } 1537 }
1521 1538
1522 wc->w_handle = handle; 1539 wc->w_handle = handle;
1523 out: 1540 out:
1524 return ret; 1541 return ret;
1525 } 1542 }
1526 1543
1527 int ocfs2_size_fits_inline_data(struct buffer 1544 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1528 { 1545 {
1529 struct ocfs2_dinode *di = (struct ocf 1546 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1530 1547
1531 if (new_size <= le16_to_cpu(di->id2.i 1548 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1532 return 1; 1549 return 1;
1533 return 0; 1550 return 0;
1534 } 1551 }
1535 1552
1536 static int ocfs2_try_to_write_inline_data(str 1553 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1537 str 1554 struct inode *inode, loff_t pos,
1538 uns 1555 unsigned len, struct page *mmap_page,
1539 str 1556 struct ocfs2_write_ctxt *wc)
1540 { 1557 {
1541 int ret, written = 0; 1558 int ret, written = 0;
1542 loff_t end = pos + len; 1559 loff_t end = pos + len;
1543 struct ocfs2_inode_info *oi = OCFS2_I 1560 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1544 struct ocfs2_dinode *di = NULL; 1561 struct ocfs2_dinode *di = NULL;
1545 1562
1546 trace_ocfs2_try_to_write_inline_data( 1563 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1547 1564 len, (unsigned long long)pos,
1548 1565 oi->ip_dyn_features);
1549 1566
1550 /* 1567 /*
1551 * Handle inodes which already have i 1568 * Handle inodes which already have inline data 1st.
1552 */ 1569 */
1553 if (oi->ip_dyn_features & OCFS2_INLIN 1570 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1554 if (mmap_page == NULL && 1571 if (mmap_page == NULL &&
1555 ocfs2_size_fits_inline_da 1572 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1556 goto do_inline_write; 1573 goto do_inline_write;
1557 1574
1558 /* 1575 /*
1559 * The write won't fit - we h 1576 * The write won't fit - we have to give this inode an
1560 * inline extent list now. 1577 * inline extent list now.
1561 */ 1578 */
1562 ret = ocfs2_convert_inline_da 1579 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1563 if (ret) 1580 if (ret)
1564 mlog_errno(ret); 1581 mlog_errno(ret);
1565 goto out; 1582 goto out;
1566 } 1583 }
1567 1584
1568 /* 1585 /*
1569 * Check whether the inode can accept 1586 * Check whether the inode can accept inline data.
1570 */ 1587 */
1571 if (oi->ip_clusters != 0 || i_size_re 1588 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1572 return 0; 1589 return 0;
1573 1590
1574 /* 1591 /*
1575 * Check whether the write can fit. 1592 * Check whether the write can fit.
1576 */ 1593 */
1577 di = (struct ocfs2_dinode *)wc->w_di_ 1594 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1578 if (mmap_page || 1595 if (mmap_page ||
1579 end > ocfs2_max_inline_data_with_ 1596 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1580 return 0; 1597 return 0;
1581 1598
1582 do_inline_write: 1599 do_inline_write:
1583 ret = ocfs2_write_begin_inline(mappin 1600 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1584 if (ret) { 1601 if (ret) {
1585 mlog_errno(ret); 1602 mlog_errno(ret);
1586 goto out; 1603 goto out;
1587 } 1604 }
1588 1605
1589 /* 1606 /*
1590 * This signals to the caller that th 1607 * This signals to the caller that the data can be written
1591 * inline. 1608 * inline.
1592 */ 1609 */
1593 written = 1; 1610 written = 1;
1594 out: 1611 out:
1595 return written ? written : ret; 1612 return written ? written : ret;
1596 } 1613 }
1597 1614
1598 /* 1615 /*
1599 * This function only does anything for file 1616 * This function only does anything for file systems which can't
1600 * handle sparse files. 1617 * handle sparse files.
1601 * 1618 *
1602 * What we want to do here is fill in any hol 1619 * What we want to do here is fill in any hole between the current end
1603 * of allocation and the end of our write. Th 1620 * of allocation and the end of our write. That way the rest of the
1604 * write path can treat it as an non-allocati 1621 * write path can treat it as an non-allocating write, which has no
1605 * special case code for sparse/nonsparse fil 1622 * special case code for sparse/nonsparse files.
1606 */ 1623 */
1607 static int ocfs2_expand_nonsparse_inode(struc 1624 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1608 struc 1625 struct buffer_head *di_bh,
1609 loff_ 1626 loff_t pos, unsigned len,
1610 struc 1627 struct ocfs2_write_ctxt *wc)
1611 { 1628 {
1612 int ret; 1629 int ret;
1613 loff_t newsize = pos + len; 1630 loff_t newsize = pos + len;
1614 1631
1615 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1632 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1616 1633
1617 if (newsize <= i_size_read(inode)) 1634 if (newsize <= i_size_read(inode))
1618 return 0; 1635 return 0;
1619 1636
1620 ret = ocfs2_extend_no_holes(inode, di 1637 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1621 if (ret) 1638 if (ret)
1622 mlog_errno(ret); 1639 mlog_errno(ret);
1623 1640
1624 /* There is no wc if this is call fro 1641 /* There is no wc if this is call from direct. */
1625 if (wc) 1642 if (wc)
1626 wc->w_first_new_cpos = 1643 wc->w_first_new_cpos =
1627 ocfs2_clusters_for_by 1644 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1628 1645
1629 return ret; 1646 return ret;
1630 } 1647 }
1631 1648
1632 static int ocfs2_zero_tail(struct inode *inod 1649 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1633 loff_t pos) 1650 loff_t pos)
1634 { 1651 {
1635 int ret = 0; 1652 int ret = 0;
1636 1653
1637 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1654 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1638 if (pos > i_size_read(inode)) 1655 if (pos > i_size_read(inode))
1639 ret = ocfs2_zero_extend(inode 1656 ret = ocfs2_zero_extend(inode, di_bh, pos);
1640 1657
1641 return ret; 1658 return ret;
1642 } 1659 }
1643 1660
1644 int ocfs2_write_begin_nolock(struct address_s 1661 int ocfs2_write_begin_nolock(struct address_space *mapping,
1645 loff_t pos, unsi 1662 loff_t pos, unsigned len, ocfs2_write_type_t type,
1646 struct page **pa 1663 struct page **pagep, void **fsdata,
1647 struct buffer_he 1664 struct buffer_head *di_bh, struct page *mmap_page)
1648 { 1665 {
1649 int ret, cluster_of_pages, credits = 1666 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1650 unsigned int clusters_to_alloc, exten 1667 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1651 struct ocfs2_write_ctxt *wc; 1668 struct ocfs2_write_ctxt *wc;
1652 struct inode *inode = mapping->host; 1669 struct inode *inode = mapping->host;
1653 struct ocfs2_super *osb = OCFS2_SB(in 1670 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1654 struct ocfs2_dinode *di; 1671 struct ocfs2_dinode *di;
1655 struct ocfs2_alloc_context *data_ac = 1672 struct ocfs2_alloc_context *data_ac = NULL;
1656 struct ocfs2_alloc_context *meta_ac = 1673 struct ocfs2_alloc_context *meta_ac = NULL;
1657 handle_t *handle; 1674 handle_t *handle;
1658 struct ocfs2_extent_tree et; 1675 struct ocfs2_extent_tree et;
1659 int try_free = 1, ret1; 1676 int try_free = 1, ret1;
1660 1677
1661 try_again: 1678 try_again:
1662 ret = ocfs2_alloc_write_ctxt(&wc, osb 1679 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
1663 if (ret) { 1680 if (ret) {
1664 mlog_errno(ret); 1681 mlog_errno(ret);
1665 return ret; 1682 return ret;
1666 } 1683 }
1667 1684
1668 if (ocfs2_supports_inline_data(osb)) 1685 if (ocfs2_supports_inline_data(osb)) {
1669 ret = ocfs2_try_to_write_inli 1686 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1670 1687 mmap_page, wc);
1671 if (ret == 1) { 1688 if (ret == 1) {
1672 ret = 0; 1689 ret = 0;
1673 goto success; 1690 goto success;
1674 } 1691 }
1675 if (ret < 0) { 1692 if (ret < 0) {
1676 mlog_errno(ret); 1693 mlog_errno(ret);
1677 goto out; 1694 goto out;
1678 } 1695 }
1679 } 1696 }
1680 1697
1681 /* Direct io change i_size late, shou 1698 /* Direct io change i_size late, should not zero tail here. */
1682 if (type != OCFS2_WRITE_DIRECT) { 1699 if (type != OCFS2_WRITE_DIRECT) {
1683 if (ocfs2_sparse_alloc(osb)) 1700 if (ocfs2_sparse_alloc(osb))
1684 ret = ocfs2_zero_tail 1701 ret = ocfs2_zero_tail(inode, di_bh, pos);
1685 else 1702 else
1686 ret = ocfs2_expand_no 1703 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1687 1704 len, wc);
1688 if (ret) { 1705 if (ret) {
1689 mlog_errno(ret); 1706 mlog_errno(ret);
1690 goto out; 1707 goto out;
1691 } 1708 }
1692 } 1709 }
1693 1710
1694 ret = ocfs2_check_range_for_refcount( 1711 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1695 if (ret < 0) { 1712 if (ret < 0) {
1696 mlog_errno(ret); 1713 mlog_errno(ret);
1697 goto out; 1714 goto out;
1698 } else if (ret == 1) { 1715 } else if (ret == 1) {
1699 clusters_need = wc->w_clen; 1716 clusters_need = wc->w_clen;
1700 ret = ocfs2_refcount_cow(inod 1717 ret = ocfs2_refcount_cow(inode, di_bh,
1701 wc-> 1718 wc->w_cpos, wc->w_clen, UINT_MAX);
1702 if (ret) { 1719 if (ret) {
1703 mlog_errno(ret); 1720 mlog_errno(ret);
1704 goto out; 1721 goto out;
1705 } 1722 }
1706 } 1723 }
1707 1724
1708 ret = ocfs2_populate_write_desc(inode 1725 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1709 &exte 1726 &extents_to_split);
1710 if (ret) { 1727 if (ret) {
1711 mlog_errno(ret); 1728 mlog_errno(ret);
1712 goto out; 1729 goto out;
1713 } 1730 }
1714 clusters_need += clusters_to_alloc; 1731 clusters_need += clusters_to_alloc;
1715 1732
1716 di = (struct ocfs2_dinode *)wc->w_di_ 1733 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1717 1734
1718 trace_ocfs2_write_begin_nolock( 1735 trace_ocfs2_write_begin_nolock(
1719 (unsigned long long)O 1736 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1720 (long long)i_size_rea 1737 (long long)i_size_read(inode),
1721 le32_to_cpu(di->i_clu 1738 le32_to_cpu(di->i_clusters),
1722 pos, len, type, mmap_ 1739 pos, len, type, mmap_page,
1723 clusters_to_alloc, ex 1740 clusters_to_alloc, extents_to_split);
1724 1741
1725 /* 1742 /*
1726 * We set w_target_from, w_target_to 1743 * We set w_target_from, w_target_to here so that
1727 * ocfs2_write_end() knows which rang 1744 * ocfs2_write_end() knows which range in the target page to
1728 * write out. An allocation requires 1745 * write out. An allocation requires that we write the entire
1729 * cluster range. 1746 * cluster range.
1730 */ 1747 */
1731 if (clusters_to_alloc || extents_to_s 1748 if (clusters_to_alloc || extents_to_split) {
1732 /* 1749 /*
1733 * XXX: We are stretching the 1750 * XXX: We are stretching the limits of
1734 * ocfs2_lock_allocators(). I 1751 * ocfs2_lock_allocators(). It greatly over-estimates
1735 * the work to be done. 1752 * the work to be done.
1736 */ 1753 */
1737 ocfs2_init_dinode_extent_tree 1754 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1738 1755 wc->w_di_bh);
1739 ret = ocfs2_lock_allocators(i 1756 ret = ocfs2_lock_allocators(inode, &et,
1740 c 1757 clusters_to_alloc, extents_to_split,
1741 & 1758 &data_ac, &meta_ac);
1742 if (ret) { 1759 if (ret) {
1743 mlog_errno(ret); 1760 mlog_errno(ret);
1744 goto out; 1761 goto out;
1745 } 1762 }
1746 1763
1747 if (data_ac) 1764 if (data_ac)
1748 data_ac->ac_resv = &O 1765 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1749 1766
1750 credits = ocfs2_calc_extend_c 1767 credits = ocfs2_calc_extend_credits(inode->i_sb,
1751 1768 &di->id2.i_list);
1752 } else if (type == OCFS2_WRITE_DIRECT 1769 } else if (type == OCFS2_WRITE_DIRECT)
1753 /* direct write needs not to 1770 /* direct write needs not to start trans if no extents alloc. */
1754 goto success; 1771 goto success;
1755 1772
1756 /* 1773 /*
1757 * We have to zero sparse allocated c 1774 * We have to zero sparse allocated clusters, unwritten extent clusters,
1758 * and non-sparse clusters we just ex 1775 * and non-sparse clusters we just extended. For non-sparse writes,
1759 * we know zeros will only be needed 1776 * we know zeros will only be needed in the first and/or last cluster.
1760 */ 1777 */
1761 if (wc->w_clen && (wc->w_desc[0].c_ne 1778 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1762 wc->w_desc[wc->w_c 1779 wc->w_desc[wc->w_clen - 1].c_needs_zero))
1763 cluster_of_pages = 1; 1780 cluster_of_pages = 1;
1764 else 1781 else
1765 cluster_of_pages = 0; 1782 cluster_of_pages = 0;
1766 1783
1767 ocfs2_set_target_boundaries(osb, wc, 1784 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1768 1785
1769 handle = ocfs2_start_trans(osb, credi 1786 handle = ocfs2_start_trans(osb, credits);
1770 if (IS_ERR(handle)) { 1787 if (IS_ERR(handle)) {
1771 ret = PTR_ERR(handle); 1788 ret = PTR_ERR(handle);
1772 mlog_errno(ret); 1789 mlog_errno(ret);
1773 goto out; 1790 goto out;
1774 } 1791 }
1775 1792
1776 wc->w_handle = handle; 1793 wc->w_handle = handle;
1777 1794
1778 if (clusters_to_alloc) { 1795 if (clusters_to_alloc) {
1779 ret = dquot_alloc_space_nodir 1796 ret = dquot_alloc_space_nodirty(inode,
1780 ocfs2_clusters_to_byt 1797 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1781 if (ret) 1798 if (ret)
1782 goto out_commit; 1799 goto out_commit;
1783 } 1800 }
1784 1801
1785 ret = ocfs2_journal_access_di(handle, 1802 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1786 OCFS2_J 1803 OCFS2_JOURNAL_ACCESS_WRITE);
1787 if (ret) { 1804 if (ret) {
1788 mlog_errno(ret); 1805 mlog_errno(ret);
1789 goto out_quota; 1806 goto out_quota;
1790 } 1807 }
1791 1808
1792 /* 1809 /*
1793 * Fill our page array first. That wa 1810 * Fill our page array first. That way we've grabbed enough so
1794 * that we can zero and flush if we e 1811 * that we can zero and flush if we error after adding the
1795 * extent. 1812 * extent.
1796 */ 1813 */
1797 ret = ocfs2_grab_pages_for_write(mapp 1814 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1798 clus 1815 cluster_of_pages, mmap_page);
1799 if (ret) { !! 1816 if (ret && ret != -EAGAIN) {
1800 /* <<
1801 * ocfs2_grab_pages_for_write <<
1802 * the target page. In this c <<
1803 * page. This will trigger th <<
1804 * the operation. <<
1805 */ <<
1806 if (type == OCFS2_WRITE_MMAP <<
1807 BUG_ON(wc->w_target_p <<
1808 ret = 0; <<
1809 goto out_quota; <<
1810 } <<
1811 <<
1812 mlog_errno(ret); 1817 mlog_errno(ret);
1813 goto out_quota; 1818 goto out_quota;
1814 } 1819 }
1815 1820
>> 1821 /*
>> 1822 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
>> 1823 * the target page. In this case, we exit with no error and no target
>> 1824 * page. This will trigger the caller, page_mkwrite(), to re-try
>> 1825 * the operation.
>> 1826 */
>> 1827 if (ret == -EAGAIN) {
>> 1828 BUG_ON(wc->w_target_page);
>> 1829 ret = 0;
>> 1830 goto out_quota;
>> 1831 }
>> 1832
1816 ret = ocfs2_write_cluster_by_desc(map 1833 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1817 len 1834 len);
1818 if (ret) { 1835 if (ret) {
1819 mlog_errno(ret); 1836 mlog_errno(ret);
1820 goto out_quota; 1837 goto out_quota;
1821 } 1838 }
1822 1839
1823 if (data_ac) 1840 if (data_ac)
1824 ocfs2_free_alloc_context(data 1841 ocfs2_free_alloc_context(data_ac);
1825 if (meta_ac) 1842 if (meta_ac)
1826 ocfs2_free_alloc_context(meta 1843 ocfs2_free_alloc_context(meta_ac);
1827 1844
1828 success: 1845 success:
1829 if (pagep) 1846 if (pagep)
1830 *pagep = wc->w_target_page; 1847 *pagep = wc->w_target_page;
1831 *fsdata = wc; 1848 *fsdata = wc;
1832 return 0; 1849 return 0;
1833 out_quota: 1850 out_quota:
1834 if (clusters_to_alloc) 1851 if (clusters_to_alloc)
1835 dquot_free_space(inode, 1852 dquot_free_space(inode,
1836 ocfs2_clusters_to_b 1853 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1837 out_commit: 1854 out_commit:
1838 ocfs2_commit_trans(osb, handle); 1855 ocfs2_commit_trans(osb, handle);
1839 1856
1840 out: 1857 out:
1841 /* 1858 /*
1842 * The mmapped page won't be unlocked 1859 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1843 * even in case of error here like EN 1860 * even in case of error here like ENOSPC and ENOMEM. So, we need
1844 * to unlock the target page manually 1861 * to unlock the target page manually to prevent deadlocks when
1845 * retrying again on ENOSPC, or when 1862 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1846 * to VM code. 1863 * to VM code.
1847 */ 1864 */
1848 if (wc->w_target_locked) 1865 if (wc->w_target_locked)
1849 unlock_page(mmap_page); 1866 unlock_page(mmap_page);
1850 1867
1851 ocfs2_free_write_ctxt(inode, wc); 1868 ocfs2_free_write_ctxt(inode, wc);
1852 1869
1853 if (data_ac) { 1870 if (data_ac) {
1854 ocfs2_free_alloc_context(data 1871 ocfs2_free_alloc_context(data_ac);
1855 data_ac = NULL; 1872 data_ac = NULL;
1856 } 1873 }
1857 if (meta_ac) { 1874 if (meta_ac) {
1858 ocfs2_free_alloc_context(meta 1875 ocfs2_free_alloc_context(meta_ac);
1859 meta_ac = NULL; 1876 meta_ac = NULL;
1860 } 1877 }
1861 1878
1862 if (ret == -ENOSPC && try_free) { 1879 if (ret == -ENOSPC && try_free) {
1863 /* 1880 /*
1864 * Try to free some truncate 1881 * Try to free some truncate log so that we can have enough
1865 * clusters to allocate. 1882 * clusters to allocate.
1866 */ 1883 */
1867 try_free = 0; 1884 try_free = 0;
1868 1885
1869 ret1 = ocfs2_try_to_free_trun 1886 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1870 if (ret1 == 1) 1887 if (ret1 == 1)
1871 goto try_again; 1888 goto try_again;
1872 1889
1873 if (ret1 < 0) 1890 if (ret1 < 0)
1874 mlog_errno(ret1); 1891 mlog_errno(ret1);
1875 } 1892 }
1876 1893
1877 return ret; 1894 return ret;
1878 } 1895 }
1879 1896
1880 static int ocfs2_write_begin(struct file *fil 1897 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1881 loff_t pos, unsi !! 1898 loff_t pos, unsigned len, unsigned flags,
1882 struct page **pa 1899 struct page **pagep, void **fsdata)
1883 { 1900 {
1884 int ret; 1901 int ret;
1885 struct buffer_head *di_bh = NULL; 1902 struct buffer_head *di_bh = NULL;
1886 struct inode *inode = mapping->host; 1903 struct inode *inode = mapping->host;
1887 1904
1888 ret = ocfs2_inode_lock(inode, &di_bh, 1905 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1889 if (ret) { 1906 if (ret) {
1890 mlog_errno(ret); 1907 mlog_errno(ret);
1891 return ret; 1908 return ret;
1892 } 1909 }
1893 1910
1894 /* 1911 /*
1895 * Take alloc sem here to prevent con 1912 * Take alloc sem here to prevent concurrent lookups. That way
1896 * the mapping, zeroing and tree mani 1913 * the mapping, zeroing and tree manipulation within
1897 * ocfs2_write() will be safe against !! 1914 * ocfs2_write() will be safe against ->readpage(). This
1898 * should also serve to lock out allo 1915 * should also serve to lock out allocation from a shared
1899 * writeable region. 1916 * writeable region.
1900 */ 1917 */
1901 down_write(&OCFS2_I(inode)->ip_alloc_ 1918 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1902 1919
1903 ret = ocfs2_write_begin_nolock(mappin 1920 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1904 pagep, 1921 pagep, fsdata, di_bh, NULL);
1905 if (ret) { 1922 if (ret) {
1906 mlog_errno(ret); 1923 mlog_errno(ret);
1907 goto out_fail; 1924 goto out_fail;
1908 } 1925 }
1909 1926
1910 brelse(di_bh); 1927 brelse(di_bh);
1911 1928
1912 return 0; 1929 return 0;
1913 1930
1914 out_fail: 1931 out_fail:
1915 up_write(&OCFS2_I(inode)->ip_alloc_se 1932 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1916 1933
1917 brelse(di_bh); 1934 brelse(di_bh);
1918 ocfs2_inode_unlock(inode, 1); 1935 ocfs2_inode_unlock(inode, 1);
1919 1936
1920 return ret; 1937 return ret;
1921 } 1938 }
1922 1939
1923 static void ocfs2_write_end_inline(struct ino 1940 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1924 unsigned l 1941 unsigned len, unsigned *copied,
1925 struct ocf 1942 struct ocfs2_dinode *di,
1926 struct ocf 1943 struct ocfs2_write_ctxt *wc)
1927 { 1944 {
1928 void *kaddr; 1945 void *kaddr;
1929 1946
1930 if (unlikely(*copied < len)) { 1947 if (unlikely(*copied < len)) {
1931 if (!PageUptodate(wc->w_targe 1948 if (!PageUptodate(wc->w_target_page)) {
1932 *copied = 0; 1949 *copied = 0;
1933 return; 1950 return;
1934 } 1951 }
1935 } 1952 }
1936 1953
1937 kaddr = kmap_atomic(wc->w_target_page 1954 kaddr = kmap_atomic(wc->w_target_page);
1938 memcpy(di->id2.i_data.id_data + pos, 1955 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1939 kunmap_atomic(kaddr); 1956 kunmap_atomic(kaddr);
1940 1957
1941 trace_ocfs2_write_end_inline( 1958 trace_ocfs2_write_end_inline(
1942 (unsigned long long)OCFS2_I(inod 1959 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1943 (unsigned long long)pos, *copied 1960 (unsigned long long)pos, *copied,
1944 le16_to_cpu(di->id2.i_data.id_co 1961 le16_to_cpu(di->id2.i_data.id_count),
1945 le16_to_cpu(di->i_dyn_features)) 1962 le16_to_cpu(di->i_dyn_features));
1946 } 1963 }
1947 1964
1948 int ocfs2_write_end_nolock(struct address_spa 1965 int ocfs2_write_end_nolock(struct address_space *mapping,
1949 loff_t pos, unsign !! 1966 loff_t pos, unsigned len, unsigned copied,
>> 1967 struct page *page, void *fsdata)
1950 { 1968 {
1951 int i, ret; 1969 int i, ret;
1952 unsigned from, to, start = pos & (PAG 1970 unsigned from, to, start = pos & (PAGE_SIZE - 1);
1953 struct inode *inode = mapping->host; 1971 struct inode *inode = mapping->host;
1954 struct ocfs2_super *osb = OCFS2_SB(in 1972 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1955 struct ocfs2_write_ctxt *wc = fsdata; 1973 struct ocfs2_write_ctxt *wc = fsdata;
1956 struct ocfs2_dinode *di = (struct ocf 1974 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1957 handle_t *handle = wc->w_handle; 1975 handle_t *handle = wc->w_handle;
1958 struct page *tmppage; 1976 struct page *tmppage;
1959 1977
1960 BUG_ON(!list_empty(&wc->w_unwritten_l 1978 BUG_ON(!list_empty(&wc->w_unwritten_list));
1961 1979
1962 if (handle) { 1980 if (handle) {
1963 ret = ocfs2_journal_access_di 1981 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1964 wc->w_di_bh, 1982 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1965 if (ret) { 1983 if (ret) {
1966 copied = ret; 1984 copied = ret;
1967 mlog_errno(ret); 1985 mlog_errno(ret);
1968 goto out; 1986 goto out;
1969 } 1987 }
1970 } 1988 }
1971 1989
1972 if (OCFS2_I(inode)->ip_dyn_features & 1990 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1973 ocfs2_write_end_inline(inode, 1991 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1974 goto out_write_size; 1992 goto out_write_size;
1975 } 1993 }
1976 1994
1977 if (unlikely(copied < len) && wc->w_t 1995 if (unlikely(copied < len) && wc->w_target_page) {
1978 loff_t new_isize; <<
1979 <<
1980 if (!PageUptodate(wc->w_targe 1996 if (!PageUptodate(wc->w_target_page))
1981 copied = 0; 1997 copied = 0;
1982 1998
1983 new_isize = max_t(loff_t, i_s !! 1999 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1984 if (new_isize > page_offset(w !! 2000 start+len);
1985 ocfs2_zero_new_buffer <<
1986 <<
1987 else { <<
1988 /* <<
1989 * When page is fully <<
1990 * failed), do not bo <<
1991 * it instead so that <<
1992 * put page & buffer <<
1993 * state. <<
1994 */ <<
1995 block_invalidate_foli <<
1996 <<
1997 } <<
1998 } 2001 }
1999 if (wc->w_target_page) 2002 if (wc->w_target_page)
2000 flush_dcache_page(wc->w_targe 2003 flush_dcache_page(wc->w_target_page);
2001 2004
2002 for(i = 0; i < wc->w_num_pages; i++) 2005 for(i = 0; i < wc->w_num_pages; i++) {
2003 tmppage = wc->w_pages[i]; 2006 tmppage = wc->w_pages[i];
2004 2007
2005 /* This is the direct io targ 2008 /* This is the direct io target page. */
2006 if (tmppage == NULL) 2009 if (tmppage == NULL)
2007 continue; 2010 continue;
2008 2011
2009 if (tmppage == wc->w_target_p 2012 if (tmppage == wc->w_target_page) {
2010 from = wc->w_target_f 2013 from = wc->w_target_from;
2011 to = wc->w_target_to; 2014 to = wc->w_target_to;
2012 2015
2013 BUG_ON(from > PAGE_SI 2016 BUG_ON(from > PAGE_SIZE ||
2014 to > PAGE_SIZE 2017 to > PAGE_SIZE ||
2015 to < from); 2018 to < from);
2016 } else { 2019 } else {
2017 /* 2020 /*
2018 * Pages adjacent to 2021 * Pages adjacent to the target (if any) imply
2019 * a hole-filling wri 2022 * a hole-filling write in which case we want
2020 * to flush their ent 2023 * to flush their entire range.
2021 */ 2024 */
2022 from = 0; 2025 from = 0;
2023 to = PAGE_SIZE; 2026 to = PAGE_SIZE;
2024 } 2027 }
2025 2028
2026 if (page_has_buffers(tmppage) 2029 if (page_has_buffers(tmppage)) {
2027 if (handle && ocfs2_s !! 2030 if (handle && ocfs2_should_order_data(inode))
2028 loff_t start_ !! 2031 ocfs2_jbd2_file_inode(handle, inode);
2029 ((lof <<
2030 from; <<
2031 loff_t length <<
2032 ocfs2_jbd2_in <<
2033 <<
2034 } <<
2035 block_commit_write(tm 2032 block_commit_write(tmppage, from, to);
2036 } 2033 }
2037 } 2034 }
2038 2035
2039 out_write_size: 2036 out_write_size:
2040 /* Direct io do not update i_size her 2037 /* Direct io do not update i_size here. */
2041 if (wc->w_type != OCFS2_WRITE_DIRECT) 2038 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2042 pos += copied; 2039 pos += copied;
2043 if (pos > i_size_read(inode)) 2040 if (pos > i_size_read(inode)) {
2044 i_size_write(inode, p 2041 i_size_write(inode, pos);
2045 mark_inode_dirty(inod 2042 mark_inode_dirty(inode);
2046 } 2043 }
2047 inode->i_blocks = ocfs2_inode 2044 inode->i_blocks = ocfs2_inode_sector_count(inode);
2048 di->i_size = cpu_to_le64((u64 2045 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2049 inode_set_mtime_to_ts(inode, !! 2046 inode->i_mtime = inode->i_ctime = current_time(inode);
2050 di->i_mtime = di->i_ctime = c !! 2047 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2051 di->i_mtime_nsec = di->i_ctim !! 2048 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2052 if (handle) 2049 if (handle)
2053 ocfs2_update_inode_fs 2050 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2054 } 2051 }
2055 if (handle) 2052 if (handle)
2056 ocfs2_journal_dirty(handle, w 2053 ocfs2_journal_dirty(handle, wc->w_di_bh);
2057 2054
2058 out: 2055 out:
2059 /* unlock pages before dealloc since 2056 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2060 * lock, or it will cause a deadlock 2057 * lock, or it will cause a deadlock since journal commit threads holds
2061 * this lock and will ask for the pag 2058 * this lock and will ask for the page lock when flushing the data.
2062 * put it here to preserve the unlock 2059 * put it here to preserve the unlock order.
2063 */ 2060 */
2064 ocfs2_unlock_pages(wc); 2061 ocfs2_unlock_pages(wc);
2065 2062
2066 if (handle) 2063 if (handle)
2067 ocfs2_commit_trans(osb, handl 2064 ocfs2_commit_trans(osb, handle);
2068 2065
2069 ocfs2_run_deallocs(osb, &wc->w_deallo 2066 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2070 2067
2071 brelse(wc->w_di_bh); 2068 brelse(wc->w_di_bh);
2072 kfree(wc); 2069 kfree(wc);
2073 2070
2074 return copied; 2071 return copied;
2075 } 2072 }
2076 2073
2077 static int ocfs2_write_end(struct file *file, 2074 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2078 loff_t pos, unsign 2075 loff_t pos, unsigned len, unsigned copied,
2079 struct page *page, 2076 struct page *page, void *fsdata)
2080 { 2077 {
2081 int ret; 2078 int ret;
2082 struct inode *inode = mapping->host; 2079 struct inode *inode = mapping->host;
2083 2080
2084 ret = ocfs2_write_end_nolock(mapping, !! 2081 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
2085 2082
2086 up_write(&OCFS2_I(inode)->ip_alloc_se 2083 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2087 ocfs2_inode_unlock(inode, 1); 2084 ocfs2_inode_unlock(inode, 1);
2088 2085
2089 return ret; 2086 return ret;
2090 } 2087 }
2091 2088
2092 struct ocfs2_dio_write_ctxt { 2089 struct ocfs2_dio_write_ctxt {
2093 struct list_head dw_zero_list; 2090 struct list_head dw_zero_list;
2094 unsigned dw_zero_count 2091 unsigned dw_zero_count;
2095 int dw_orphaned; 2092 int dw_orphaned;
2096 pid_t dw_writer_pid 2093 pid_t dw_writer_pid;
2097 }; 2094 };
2098 2095
2099 static struct ocfs2_dio_write_ctxt * 2096 static struct ocfs2_dio_write_ctxt *
2100 ocfs2_dio_alloc_write_ctx(struct buffer_head 2097 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2101 { 2098 {
2102 struct ocfs2_dio_write_ctxt *dwc = NU 2099 struct ocfs2_dio_write_ctxt *dwc = NULL;
2103 2100
2104 if (bh->b_private) 2101 if (bh->b_private)
2105 return bh->b_private; 2102 return bh->b_private;
2106 2103
2107 dwc = kmalloc(sizeof(struct ocfs2_dio 2104 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2108 if (dwc == NULL) 2105 if (dwc == NULL)
2109 return NULL; 2106 return NULL;
2110 INIT_LIST_HEAD(&dwc->dw_zero_list); 2107 INIT_LIST_HEAD(&dwc->dw_zero_list);
2111 dwc->dw_zero_count = 0; 2108 dwc->dw_zero_count = 0;
2112 dwc->dw_orphaned = 0; 2109 dwc->dw_orphaned = 0;
2113 dwc->dw_writer_pid = task_pid_nr(curr 2110 dwc->dw_writer_pid = task_pid_nr(current);
2114 bh->b_private = dwc; 2111 bh->b_private = dwc;
2115 *alloc = 1; 2112 *alloc = 1;
2116 2113
2117 return dwc; 2114 return dwc;
2118 } 2115 }
2119 2116
2120 static void ocfs2_dio_free_write_ctx(struct i 2117 static void ocfs2_dio_free_write_ctx(struct inode *inode,
2121 struct o 2118 struct ocfs2_dio_write_ctxt *dwc)
2122 { 2119 {
2123 ocfs2_free_unwritten_list(inode, &dwc 2120 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2124 kfree(dwc); 2121 kfree(dwc);
2125 } 2122 }
2126 2123
2127 /* 2124 /*
2128 * TODO: Make this into a generic get_blocks 2125 * TODO: Make this into a generic get_blocks function.
2129 * 2126 *
2130 * From do_direct_io in direct-io.c: 2127 * From do_direct_io in direct-io.c:
2131 * "So what we do is to permit the ->get_blo 2128 * "So what we do is to permit the ->get_blocks function to populate
2132 * bh.b_size with the size of IO which is p 2129 * bh.b_size with the size of IO which is permitted at this offset and
2133 * this i_blkbits." 2130 * this i_blkbits."
2134 * 2131 *
2135 * This function is called directly from get_ 2132 * This function is called directly from get_more_blocks in direct-io.c.
2136 * 2133 *
2137 * called like this: dio->get_blocks(dio->ino 2134 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2138 * fs_co 2135 * fs_count, map_bh, dio->rw == WRITE);
2139 */ 2136 */
2140 static int ocfs2_dio_wr_get_block(struct inod 2137 static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock,
2141 struct buffer_ 2138 struct buffer_head *bh_result, int create)
2142 { 2139 {
2143 struct ocfs2_super *osb = OCFS2_SB(in 2140 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2144 struct ocfs2_inode_info *oi = OCFS2_I 2141 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2145 struct ocfs2_write_ctxt *wc; 2142 struct ocfs2_write_ctxt *wc;
2146 struct ocfs2_write_cluster_desc *desc 2143 struct ocfs2_write_cluster_desc *desc = NULL;
2147 struct ocfs2_dio_write_ctxt *dwc = NU 2144 struct ocfs2_dio_write_ctxt *dwc = NULL;
2148 struct buffer_head *di_bh = NULL; 2145 struct buffer_head *di_bh = NULL;
2149 u64 p_blkno; 2146 u64 p_blkno;
2150 unsigned int i_blkbits = inode->i_sb- 2147 unsigned int i_blkbits = inode->i_sb->s_blocksize_bits;
2151 loff_t pos = iblock << i_blkbits; 2148 loff_t pos = iblock << i_blkbits;
2152 sector_t endblk = (i_size_read(inode) 2149 sector_t endblk = (i_size_read(inode) - 1) >> i_blkbits;
2153 unsigned len, total_len = bh_result-> 2150 unsigned len, total_len = bh_result->b_size;
2154 int ret = 0, first_get_block = 0; 2151 int ret = 0, first_get_block = 0;
2155 2152
2156 len = osb->s_clustersize - (pos & (os 2153 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2157 len = min(total_len, len); 2154 len = min(total_len, len);
2158 2155
2159 /* 2156 /*
2160 * bh_result->b_size is count in get_ 2157 * bh_result->b_size is count in get_more_blocks according to write
2161 * "pos" and "end", we need map twice 2158 * "pos" and "end", we need map twice to return different buffer state:
2162 * 1. area in file size, not set NEW; 2159 * 1. area in file size, not set NEW;
2163 * 2. area out file size, set NEW. 2160 * 2. area out file size, set NEW.
2164 * 2161 *
2165 * iblock endblk 2162 * iblock endblk
2166 * |--------|---------|---------|---- 2163 * |--------|---------|---------|---------
2167 * |<-------area in file------->| 2164 * |<-------area in file------->|
2168 */ 2165 */
2169 2166
2170 if ((iblock <= endblk) && 2167 if ((iblock <= endblk) &&
2171 ((iblock + ((len - 1) >> i_blkbit 2168 ((iblock + ((len - 1) >> i_blkbits)) > endblk))
2172 len = (endblk - iblock + 1) < 2169 len = (endblk - iblock + 1) << i_blkbits;
2173 2170
2174 mlog(0, "get block of %lu at %llu:%u 2171 mlog(0, "get block of %lu at %llu:%u req %u\n",
2175 inode->i_ino, pos, le 2172 inode->i_ino, pos, len, total_len);
2176 2173
2177 /* 2174 /*
2178 * Because we need to change file siz 2175 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2179 * we may need to add it to orphan di 2176 * we may need to add it to orphan dir. So can not fall to fast path
2180 * while file size will be changed. 2177 * while file size will be changed.
2181 */ 2178 */
2182 if (pos + total_len <= i_size_read(in 2179 if (pos + total_len <= i_size_read(inode)) {
2183 2180
2184 /* This is the fast path for 2181 /* This is the fast path for re-write. */
2185 ret = ocfs2_lock_get_block(in 2182 ret = ocfs2_lock_get_block(inode, iblock, bh_result, create);
2186 if (buffer_mapped(bh_result) 2183 if (buffer_mapped(bh_result) &&
2187 !buffer_new(bh_result) && 2184 !buffer_new(bh_result) &&
2188 ret == 0) 2185 ret == 0)
2189 goto out; 2186 goto out;
2190 2187
2191 /* Clear state set by ocfs2_g 2188 /* Clear state set by ocfs2_get_block. */
2192 bh_result->b_state = 0; 2189 bh_result->b_state = 0;
2193 } 2190 }
2194 2191
2195 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2192 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2196 if (unlikely(dwc == NULL)) { 2193 if (unlikely(dwc == NULL)) {
2197 ret = -ENOMEM; 2194 ret = -ENOMEM;
2198 mlog_errno(ret); 2195 mlog_errno(ret);
2199 goto out; 2196 goto out;
2200 } 2197 }
2201 2198
2202 if (ocfs2_clusters_for_bytes(inode->i 2199 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2203 ocfs2_clusters_for_bytes(inode->i 2200 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2204 !dwc->dw_orphaned) { 2201 !dwc->dw_orphaned) {
2205 /* 2202 /*
2206 * when we are going to alloc 2203 * when we are going to alloc extents beyond file size, add the
2207 * inode to orphan dir, so we 2204 * inode to orphan dir, so we can recall those spaces when
2208 * system crashed during writ 2205 * system crashed during write.
2209 */ 2206 */
2210 ret = ocfs2_add_inode_to_orph 2207 ret = ocfs2_add_inode_to_orphan(osb, inode);
2211 if (ret < 0) { 2208 if (ret < 0) {
2212 mlog_errno(ret); 2209 mlog_errno(ret);
2213 goto out; 2210 goto out;
2214 } 2211 }
2215 dwc->dw_orphaned = 1; 2212 dwc->dw_orphaned = 1;
2216 } 2213 }
2217 2214
2218 ret = ocfs2_inode_lock(inode, &di_bh, 2215 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2219 if (ret) { 2216 if (ret) {
2220 mlog_errno(ret); 2217 mlog_errno(ret);
2221 goto out; 2218 goto out;
2222 } 2219 }
2223 2220
2224 down_write(&oi->ip_alloc_sem); 2221 down_write(&oi->ip_alloc_sem);
2225 2222
2226 if (first_get_block) { 2223 if (first_get_block) {
2227 if (ocfs2_sparse_alloc(osb)) !! 2224 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
2228 ret = ocfs2_zero_tail 2225 ret = ocfs2_zero_tail(inode, di_bh, pos);
2229 else 2226 else
2230 ret = ocfs2_expand_no 2227 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2231 2228 total_len, NULL);
2232 if (ret < 0) { 2229 if (ret < 0) {
2233 mlog_errno(ret); 2230 mlog_errno(ret);
2234 goto unlock; 2231 goto unlock;
2235 } 2232 }
2236 } 2233 }
2237 2234
2238 ret = ocfs2_write_begin_nolock(inode- 2235 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2239 OCFS2_ 2236 OCFS2_WRITE_DIRECT, NULL,
2240 (void 2237 (void **)&wc, di_bh, NULL);
2241 if (ret) { 2238 if (ret) {
2242 mlog_errno(ret); 2239 mlog_errno(ret);
2243 goto unlock; 2240 goto unlock;
2244 } 2241 }
2245 2242
2246 desc = &wc->w_desc[0]; 2243 desc = &wc->w_desc[0];
2247 2244
2248 p_blkno = ocfs2_clusters_to_blocks(in 2245 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2249 BUG_ON(p_blkno == 0); 2246 BUG_ON(p_blkno == 0);
2250 p_blkno += iblock & (u64)(ocfs2_clust 2247 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2251 2248
2252 map_bh(bh_result, inode->i_sb, p_blkn 2249 map_bh(bh_result, inode->i_sb, p_blkno);
2253 bh_result->b_size = len; 2250 bh_result->b_size = len;
2254 if (desc->c_needs_zero) 2251 if (desc->c_needs_zero)
2255 set_buffer_new(bh_result); 2252 set_buffer_new(bh_result);
2256 2253
2257 if (iblock > endblk) 2254 if (iblock > endblk)
2258 set_buffer_new(bh_result); 2255 set_buffer_new(bh_result);
2259 2256
2260 /* May sleep in end_io. It should not 2257 /* May sleep in end_io. It should not happen in a irq context. So defer
2261 * it to dio work queue. */ 2258 * it to dio work queue. */
2262 set_buffer_defer_completion(bh_result 2259 set_buffer_defer_completion(bh_result);
2263 2260
2264 if (!list_empty(&wc->w_unwritten_list 2261 if (!list_empty(&wc->w_unwritten_list)) {
2265 struct ocfs2_unwritten_extent 2262 struct ocfs2_unwritten_extent *ue = NULL;
2266 2263
2267 ue = list_first_entry(&wc->w_ 2264 ue = list_first_entry(&wc->w_unwritten_list,
2268 struct 2265 struct ocfs2_unwritten_extent,
2269 ue_node 2266 ue_node);
2270 BUG_ON(ue->ue_cpos != desc->c 2267 BUG_ON(ue->ue_cpos != desc->c_cpos);
2271 /* The physical address may b 2268 /* The physical address may be 0, fill it. */
2272 ue->ue_phys = desc->c_phys; 2269 ue->ue_phys = desc->c_phys;
2273 2270
2274 list_splice_tail_init(&wc->w_ 2271 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
2275 dwc->dw_zero_count += wc->w_u !! 2272 dwc->dw_zero_count++;
2276 } 2273 }
2277 2274
2278 ret = ocfs2_write_end_nolock(inode->i !! 2275 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, NULL, wc);
2279 BUG_ON(ret != len); 2276 BUG_ON(ret != len);
2280 ret = 0; 2277 ret = 0;
2281 unlock: 2278 unlock:
2282 up_write(&oi->ip_alloc_sem); 2279 up_write(&oi->ip_alloc_sem);
2283 ocfs2_inode_unlock(inode, 1); 2280 ocfs2_inode_unlock(inode, 1);
2284 brelse(di_bh); 2281 brelse(di_bh);
2285 out: 2282 out:
>> 2283 if (ret < 0)
>> 2284 ret = -EIO;
2286 return ret; 2285 return ret;
2287 } 2286 }
2288 2287
2289 static int ocfs2_dio_end_io_write(struct inod !! 2288 static void ocfs2_dio_end_io_write(struct inode *inode,
2290 struct ocfs !! 2289 struct ocfs2_dio_write_ctxt *dwc,
2291 loff_t offs !! 2290 loff_t offset,
2292 ssize_t byt !! 2291 ssize_t bytes)
2293 { 2292 {
2294 struct ocfs2_cached_dealloc_ctxt deal 2293 struct ocfs2_cached_dealloc_ctxt dealloc;
2295 struct ocfs2_extent_tree et; 2294 struct ocfs2_extent_tree et;
2296 struct ocfs2_super *osb = OCFS2_SB(in 2295 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2297 struct ocfs2_inode_info *oi = OCFS2_I 2296 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2298 struct ocfs2_unwritten_extent *ue = N 2297 struct ocfs2_unwritten_extent *ue = NULL;
2299 struct buffer_head *di_bh = NULL; 2298 struct buffer_head *di_bh = NULL;
2300 struct ocfs2_dinode *di; 2299 struct ocfs2_dinode *di;
2301 struct ocfs2_alloc_context *data_ac = 2300 struct ocfs2_alloc_context *data_ac = NULL;
2302 struct ocfs2_alloc_context *meta_ac = 2301 struct ocfs2_alloc_context *meta_ac = NULL;
2303 handle_t *handle = NULL; 2302 handle_t *handle = NULL;
2304 loff_t end = offset + bytes; 2303 loff_t end = offset + bytes;
2305 int ret = 0, credits = 0; 2304 int ret = 0, credits = 0;
2306 2305
2307 ocfs2_init_dealloc_ctxt(&dealloc); 2306 ocfs2_init_dealloc_ctxt(&dealloc);
2308 2307
2309 /* We do clear unwritten, delete orph 2308 /* We do clear unwritten, delete orphan, change i_size here. If neither
2310 * of these happen, we can skip all t 2309 * of these happen, we can skip all this. */
2311 if (list_empty(&dwc->dw_zero_list) && 2310 if (list_empty(&dwc->dw_zero_list) &&
2312 end <= i_size_read(inode) && 2311 end <= i_size_read(inode) &&
2313 !dwc->dw_orphaned) 2312 !dwc->dw_orphaned)
2314 goto out; 2313 goto out;
2315 2314
2316 ret = ocfs2_inode_lock(inode, &di_bh, 2315 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2317 if (ret < 0) { 2316 if (ret < 0) {
2318 mlog_errno(ret); 2317 mlog_errno(ret);
2319 goto out; 2318 goto out;
2320 } 2319 }
2321 2320
2322 down_write(&oi->ip_alloc_sem); 2321 down_write(&oi->ip_alloc_sem);
2323 2322
2324 /* Delete orphan before acquire i_rws !! 2323 /* Delete orphan before acquire i_mutex. */
2325 if (dwc->dw_orphaned) { 2324 if (dwc->dw_orphaned) {
2326 BUG_ON(dwc->dw_writer_pid != 2325 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2327 2326
2328 end = end > i_size_read(inode 2327 end = end > i_size_read(inode) ? end : 0;
2329 2328
2330 ret = ocfs2_del_inode_from_or 2329 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2331 !!end, end); 2330 !!end, end);
2332 if (ret < 0) 2331 if (ret < 0)
2333 mlog_errno(ret); 2332 mlog_errno(ret);
2334 } 2333 }
2335 2334
2336 di = (struct ocfs2_dinode *)di_bh->b_ !! 2335 di = (struct ocfs2_dinode *)di_bh;
2337 2336
2338 ocfs2_init_dinode_extent_tree(&et, IN 2337 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2339 2338
2340 /* Attach dealloc with extent tree in <<
2341 * which are already unlinked from cu <<
2342 * rotation and merging. <<
2343 */ <<
2344 et.et_dealloc = &dealloc; <<
2345 <<
2346 ret = ocfs2_lock_allocators(inode, &e 2339 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2347 &data_ac, 2340 &data_ac, &meta_ac);
2348 if (ret) { 2341 if (ret) {
2349 mlog_errno(ret); 2342 mlog_errno(ret);
2350 goto unlock; 2343 goto unlock;
2351 } 2344 }
2352 2345
2353 credits = ocfs2_calc_extend_credits(i 2346 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2354 2347
2355 handle = ocfs2_start_trans(osb, credi 2348 handle = ocfs2_start_trans(osb, credits);
2356 if (IS_ERR(handle)) { 2349 if (IS_ERR(handle)) {
2357 ret = PTR_ERR(handle); 2350 ret = PTR_ERR(handle);
2358 mlog_errno(ret); 2351 mlog_errno(ret);
2359 goto unlock; 2352 goto unlock;
2360 } 2353 }
2361 ret = ocfs2_journal_access_di(handle, 2354 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2362 OCFS2_J 2355 OCFS2_JOURNAL_ACCESS_WRITE);
2363 if (ret) { 2356 if (ret) {
2364 mlog_errno(ret); 2357 mlog_errno(ret);
2365 goto commit; 2358 goto commit;
2366 } 2359 }
2367 2360
2368 list_for_each_entry(ue, &dwc->dw_zero 2361 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) {
2369 ret = ocfs2_assure_trans_cred <<
2370 if (ret < 0) { <<
2371 mlog_errno(ret); <<
2372 break; <<
2373 } <<
2374 ret = ocfs2_mark_extent_writt 2362 ret = ocfs2_mark_extent_written(inode, &et, handle,
2375 2363 ue->ue_cpos, 1,
2376 2364 ue->ue_phys,
2377 2365 meta_ac, &dealloc);
2378 if (ret < 0) { 2366 if (ret < 0) {
2379 mlog_errno(ret); 2367 mlog_errno(ret);
2380 break; 2368 break;
2381 } 2369 }
2382 } 2370 }
2383 2371
2384 if (end > i_size_read(inode)) { 2372 if (end > i_size_read(inode)) {
2385 ret = ocfs2_set_inode_size(ha 2373 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2386 if (ret < 0) 2374 if (ret < 0)
2387 mlog_errno(ret); 2375 mlog_errno(ret);
2388 } 2376 }
2389 commit: 2377 commit:
2390 ocfs2_commit_trans(osb, handle); 2378 ocfs2_commit_trans(osb, handle);
2391 unlock: 2379 unlock:
2392 up_write(&oi->ip_alloc_sem); 2380 up_write(&oi->ip_alloc_sem);
2393 ocfs2_inode_unlock(inode, 1); 2381 ocfs2_inode_unlock(inode, 1);
2394 brelse(di_bh); 2382 brelse(di_bh);
2395 out: 2383 out:
2396 if (data_ac) 2384 if (data_ac)
2397 ocfs2_free_alloc_context(data 2385 ocfs2_free_alloc_context(data_ac);
2398 if (meta_ac) 2386 if (meta_ac)
2399 ocfs2_free_alloc_context(meta 2387 ocfs2_free_alloc_context(meta_ac);
2400 ocfs2_run_deallocs(osb, &dealloc); 2388 ocfs2_run_deallocs(osb, &dealloc);
2401 ocfs2_dio_free_write_ctx(inode, dwc); 2389 ocfs2_dio_free_write_ctx(inode, dwc);
2402 <<
2403 return ret; <<
2404 } 2390 }
2405 2391
2406 /* 2392 /*
2407 * ocfs2_dio_end_io is called by the dio core 2393 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
2408 * particularly interested in the aio/dio cas 2394 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2409 * to protect io on one node from truncation 2395 * to protect io on one node from truncation on another.
2410 */ 2396 */
2411 static int ocfs2_dio_end_io(struct kiocb *ioc 2397 static int ocfs2_dio_end_io(struct kiocb *iocb,
2412 loff_t offset, 2398 loff_t offset,
2413 ssize_t bytes, 2399 ssize_t bytes,
2414 void *private) 2400 void *private)
2415 { 2401 {
2416 struct inode *inode = file_inode(iocb 2402 struct inode *inode = file_inode(iocb->ki_filp);
2417 int level; 2403 int level;
2418 int ret = 0; !! 2404
>> 2405 if (bytes <= 0)
>> 2406 return 0;
2419 2407
2420 /* this io's submitter should not hav 2408 /* this io's submitter should not have unlocked this before we could */
2421 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2409 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2422 2410
2423 if (bytes <= 0) !! 2411 if (private)
2424 mlog_ratelimited(ML_ERROR, "D !! 2412 ocfs2_dio_end_io_write(inode, private, offset, bytes);
2425 (long long)b <<
2426 if (private) { <<
2427 if (bytes > 0) <<
2428 ret = ocfs2_dio_end_i <<
2429 <<
2430 else <<
2431 ocfs2_dio_free_write_ <<
2432 } <<
2433 2413
2434 ocfs2_iocb_clear_rw_locked(iocb); 2414 ocfs2_iocb_clear_rw_locked(iocb);
2435 2415
2436 level = ocfs2_iocb_rw_locked_level(io 2416 level = ocfs2_iocb_rw_locked_level(iocb);
2437 ocfs2_rw_unlock(inode, level); 2417 ocfs2_rw_unlock(inode, level);
2438 return ret; !! 2418 return 0;
2439 } 2419 }
2440 2420
2441 static ssize_t ocfs2_direct_IO(struct kiocb * 2421 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2442 { 2422 {
2443 struct file *file = iocb->ki_filp; 2423 struct file *file = iocb->ki_filp;
2444 struct inode *inode = file->f_mapping 2424 struct inode *inode = file->f_mapping->host;
2445 struct ocfs2_super *osb = OCFS2_SB(in 2425 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2446 get_block_t *get_block; 2426 get_block_t *get_block;
2447 2427
2448 /* 2428 /*
2449 * Fallback to buffered I/O if we see 2429 * Fallback to buffered I/O if we see an inode without
2450 * extents. 2430 * extents.
2451 */ 2431 */
2452 if (OCFS2_I(inode)->ip_dyn_features & 2432 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2453 return 0; 2433 return 0;
2454 2434
2455 /* Fallback to buffered I/O if we do 2435 /* Fallback to buffered I/O if we do not support append dio. */
2456 if (iocb->ki_pos + iter->count > i_si 2436 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2457 !ocfs2_supports_append_dio(osb)) 2437 !ocfs2_supports_append_dio(osb))
2458 return 0; 2438 return 0;
2459 2439
2460 if (iov_iter_rw(iter) == READ) 2440 if (iov_iter_rw(iter) == READ)
2461 get_block = ocfs2_lock_get_bl 2441 get_block = ocfs2_lock_get_block;
2462 else 2442 else
2463 get_block = ocfs2_dio_wr_get_ 2443 get_block = ocfs2_dio_wr_get_block;
2464 2444
2465 return __blockdev_direct_IO(iocb, ino 2445 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2466 iter, get 2446 iter, get_block,
2467 ocfs2_dio !! 2447 ocfs2_dio_end_io, NULL, 0);
2468 } 2448 }
2469 2449
2470 const struct address_space_operations ocfs2_a 2450 const struct address_space_operations ocfs2_aops = {
2471 .dirty_folio = block_dirty !! 2451 .readpage = ocfs2_readpage,
2472 .read_folio = ocfs2_read_ !! 2452 .readpages = ocfs2_readpages,
2473 .readahead = ocfs2_reada !! 2453 .writepage = ocfs2_writepage,
2474 .writepages = ocfs2_write <<
2475 .write_begin = ocfs2_write 2454 .write_begin = ocfs2_write_begin,
2476 .write_end = ocfs2_write 2455 .write_end = ocfs2_write_end,
2477 .bmap = ocfs2_bmap, 2456 .bmap = ocfs2_bmap,
2478 .direct_IO = ocfs2_direc 2457 .direct_IO = ocfs2_direct_IO,
2479 .invalidate_folio = block_inval !! 2458 .invalidatepage = block_invalidatepage,
2480 .release_folio = ocfs2_relea !! 2459 .releasepage = ocfs2_releasepage,
2481 .migrate_folio = buffer_migr !! 2460 .migratepage = buffer_migrate_page,
2482 .is_partially_uptodate = block_is_pa 2461 .is_partially_uptodate = block_is_partially_uptodate,
2483 .error_remove_folio = generic_err !! 2462 .error_remove_page = generic_error_remove_page,
2484 }; 2463 };
2485 2464
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.