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, "get_blocks() f !! 175 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
160 "block: %llu\n", inode, ( !! 176 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
>> 177 (unsigned long long)p_blkno);
161 goto bail; 178 goto bail;
162 } 179 }
163 180
164 if (max_blocks < count) 181 if (max_blocks < count)
165 count = max_blocks; 182 count = max_blocks;
166 183
167 /* 184 /*
168 * ocfs2 never allocates in this funct 185 * ocfs2 never allocates in this function - the only time we
169 * 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
170 * system which doesn't support holes, 187 * system which doesn't support holes, in which case BH_New
171 * allows __block_write_begin() to zer 188 * allows __block_write_begin() to zero.
172 * 189 *
173 * If we see this on a sparse file sys 190 * If we see this on a sparse file system, then a truncate has
174 * raced us and removed the cluster. I 191 * raced us and removed the cluster. In this case, we clear
175 * the buffers dirty and uptodate bits 192 * the buffers dirty and uptodate bits and let the buffer code
176 * ignore it as a hole. 193 * ignore it as a hole.
177 */ 194 */
178 if (create && p_blkno == 0 && ocfs2_sp 195 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
179 clear_buffer_dirty(bh_result); 196 clear_buffer_dirty(bh_result);
180 clear_buffer_uptodate(bh_resul 197 clear_buffer_uptodate(bh_result);
181 goto bail; 198 goto bail;
182 } 199 }
183 200
184 /* Treat the unwritten extent as a hol 201 /* Treat the unwritten extent as a hole for zeroing purposes. */
185 if (p_blkno && !(ext_flags & OCFS2_EXT 202 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
186 map_bh(bh_result, inode->i_sb, 203 map_bh(bh_result, inode->i_sb, p_blkno);
187 204
188 bh_result->b_size = count << inode->i_ 205 bh_result->b_size = count << inode->i_blkbits;
189 206
190 if (!ocfs2_sparse_alloc(osb)) { 207 if (!ocfs2_sparse_alloc(osb)) {
191 if (p_blkno == 0) { 208 if (p_blkno == 0) {
192 err = -EIO; 209 err = -EIO;
193 mlog(ML_ERROR, 210 mlog(ML_ERROR,
194 "iblock = %llu p_ 211 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
195 (unsigned long lo 212 (unsigned long long)iblock,
196 (unsigned long lo 213 (unsigned long long)p_blkno,
197 (unsigned long lo 214 (unsigned long long)OCFS2_I(inode)->ip_blkno);
198 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);
199 dump_stack(); 216 dump_stack();
200 goto bail; 217 goto bail;
201 } 218 }
202 } 219 }
203 220
204 past_eof = ocfs2_blocks_for_bytes(inod 221 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
205 222
206 trace_ocfs2_get_block_end((unsigned lo 223 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
207 (unsigned lo 224 (unsigned long long)past_eof);
208 if (create && (iblock >= past_eof)) 225 if (create && (iblock >= past_eof))
209 set_buffer_new(bh_result); 226 set_buffer_new(bh_result);
210 227
211 bail: 228 bail:
212 if (err < 0) 229 if (err < 0)
213 err = -EIO; 230 err = -EIO;
214 231
215 return err; 232 return err;
216 } 233 }
217 234
218 int ocfs2_read_inline_data(struct inode *inode 235 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
219 struct buffer_head 236 struct buffer_head *di_bh)
220 { 237 {
221 void *kaddr; 238 void *kaddr;
222 loff_t size; 239 loff_t size;
223 struct ocfs2_dinode *di = (struct ocfs 240 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
224 241
225 if (!(le16_to_cpu(di->i_dyn_features) 242 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
226 ocfs2_error(inode->i_sb, "Inod 243 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
227 (unsigned long lon 244 (unsigned long long)OCFS2_I(inode)->ip_blkno);
228 return -EROFS; 245 return -EROFS;
229 } 246 }
230 247
231 size = i_size_read(inode); 248 size = i_size_read(inode);
232 249
233 if (size > PAGE_SIZE || 250 if (size > PAGE_SIZE ||
234 size > ocfs2_max_inline_data_with_ 251 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
235 ocfs2_error(inode->i_sb, 252 ocfs2_error(inode->i_sb,
236 "Inode %llu has wi 253 "Inode %llu has with inline data has bad size: %Lu\n",
237 (unsigned long lon 254 (unsigned long long)OCFS2_I(inode)->ip_blkno,
238 (unsigned long lon 255 (unsigned long long)size);
239 return -EROFS; 256 return -EROFS;
240 } 257 }
241 258
242 kaddr = kmap_atomic(page); 259 kaddr = kmap_atomic(page);
243 if (size) 260 if (size)
244 memcpy(kaddr, di->id2.i_data.i 261 memcpy(kaddr, di->id2.i_data.id_data, size);
245 /* Clear the remaining part of the pag 262 /* Clear the remaining part of the page */
246 memset(kaddr + size, 0, PAGE_SIZE - si 263 memset(kaddr + size, 0, PAGE_SIZE - size);
247 flush_dcache_page(page); 264 flush_dcache_page(page);
248 kunmap_atomic(kaddr); 265 kunmap_atomic(kaddr);
249 266
250 SetPageUptodate(page); 267 SetPageUptodate(page);
251 268
252 return 0; 269 return 0;
253 } 270 }
254 271
255 static int ocfs2_readpage_inline(struct inode 272 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
256 { 273 {
257 int ret; 274 int ret;
258 struct buffer_head *di_bh = NULL; 275 struct buffer_head *di_bh = NULL;
259 276
260 BUG_ON(!PageLocked(page)); 277 BUG_ON(!PageLocked(page));
261 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 278 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
262 279
263 ret = ocfs2_read_inode_block(inode, &d 280 ret = ocfs2_read_inode_block(inode, &di_bh);
264 if (ret) { 281 if (ret) {
265 mlog_errno(ret); 282 mlog_errno(ret);
266 goto out; 283 goto out;
267 } 284 }
268 285
269 ret = ocfs2_read_inline_data(inode, pa 286 ret = ocfs2_read_inline_data(inode, page, di_bh);
270 out: 287 out:
271 unlock_page(page); 288 unlock_page(page);
272 289
273 brelse(di_bh); 290 brelse(di_bh);
274 return ret; 291 return ret;
275 } 292 }
276 293
277 static int ocfs2_read_folio(struct file *file, !! 294 static int ocfs2_readpage(struct file *file, struct page *page)
278 { 295 {
279 struct inode *inode = folio->mapping-> !! 296 struct inode *inode = page->mapping->host;
280 struct ocfs2_inode_info *oi = OCFS2_I( 297 struct ocfs2_inode_info *oi = OCFS2_I(inode);
281 loff_t start = folio_pos(folio); !! 298 loff_t start = (loff_t)page->index << PAGE_SHIFT;
282 int ret, unlock = 1; 299 int ret, unlock = 1;
283 300
284 trace_ocfs2_readpage((unsigned long lo !! 301 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
>> 302 (page ? page->index : 0));
285 303
286 ret = ocfs2_inode_lock_with_page(inode !! 304 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
287 if (ret != 0) { 305 if (ret != 0) {
288 if (ret == AOP_TRUNCATED_PAGE) 306 if (ret == AOP_TRUNCATED_PAGE)
289 unlock = 0; 307 unlock = 0;
290 mlog_errno(ret); 308 mlog_errno(ret);
291 goto out; 309 goto out;
292 } 310 }
293 311
294 if (down_read_trylock(&oi->ip_alloc_se 312 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
295 /* 313 /*
296 * Unlock the folio and cycle !! 314 * Unlock the page and cycle ip_alloc_sem so that we don't
297 * busyloop waiting for ip_all 315 * busyloop waiting for ip_alloc_sem to unlock
298 */ 316 */
299 ret = AOP_TRUNCATED_PAGE; 317 ret = AOP_TRUNCATED_PAGE;
300 folio_unlock(folio); !! 318 unlock_page(page);
301 unlock = 0; 319 unlock = 0;
302 down_read(&oi->ip_alloc_sem); 320 down_read(&oi->ip_alloc_sem);
303 up_read(&oi->ip_alloc_sem); 321 up_read(&oi->ip_alloc_sem);
304 goto out_inode_unlock; 322 goto out_inode_unlock;
305 } 323 }
306 324
307 /* 325 /*
308 * i_size might have just been updated 326 * i_size might have just been updated as we grabed the meta lock. We
309 * might now be discovering a truncate 327 * might now be discovering a truncate that hit on another node.
310 * block_read_full_folio->get_block fr !! 328 * block_read_full_page->get_block freaks out if it is asked to read
311 * beyond the end of a file, so we che 329 * beyond the end of a file, so we check here. Callers
312 * (generic_file_read, vm_ops->fault) 330 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
313 * and notice that the folio they just !! 331 * and notice that the page they just read isn't needed.
314 * 332 *
315 * XXX sys_readahead() seems to get th 333 * XXX sys_readahead() seems to get that wrong?
316 */ 334 */
317 if (start >= i_size_read(inode)) { 335 if (start >= i_size_read(inode)) {
318 folio_zero_segment(folio, 0, f !! 336 zero_user(page, 0, PAGE_SIZE);
319 folio_mark_uptodate(folio); !! 337 SetPageUptodate(page);
320 ret = 0; 338 ret = 0;
321 goto out_alloc; 339 goto out_alloc;
322 } 340 }
323 341
324 if (oi->ip_dyn_features & OCFS2_INLINE 342 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
325 ret = ocfs2_readpage_inline(in !! 343 ret = ocfs2_readpage_inline(inode, page);
326 else 344 else
327 ret = block_read_full_folio(fo !! 345 ret = block_read_full_page(page, ocfs2_get_block);
328 unlock = 0; 346 unlock = 0;
329 347
330 out_alloc: 348 out_alloc:
331 up_read(&oi->ip_alloc_sem); !! 349 up_read(&OCFS2_I(inode)->ip_alloc_sem);
332 out_inode_unlock: 350 out_inode_unlock:
333 ocfs2_inode_unlock(inode, 0); 351 ocfs2_inode_unlock(inode, 0);
334 out: 352 out:
335 if (unlock) 353 if (unlock)
336 folio_unlock(folio); !! 354 unlock_page(page);
337 return ret; 355 return ret;
338 } 356 }
339 357
340 /* 358 /*
341 * This is used only for read-ahead. Failures 359 * This is used only for read-ahead. Failures or difficult to handle
342 * situations are safe to ignore. 360 * situations are safe to ignore.
343 * 361 *
344 * Right now, we don't bother with BH_Boundary 362 * Right now, we don't bother with BH_Boundary - in-inode extent lists
345 * are quite large (243 extents on 4k blocks), 363 * are quite large (243 extents on 4k blocks), so most inodes don't
346 * grow out to a tree. If need be, detecting b 364 * grow out to a tree. If need be, detecting boundary extents could
347 * trivially be added in a future version of o 365 * trivially be added in a future version of ocfs2_get_block().
348 */ 366 */
349 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)
350 { 369 {
351 int ret; !! 370 int ret, err = -EIO;
352 struct inode *inode = rac->mapping->ho !! 371 struct inode *inode = mapping->host;
353 struct ocfs2_inode_info *oi = OCFS2_I( 372 struct ocfs2_inode_info *oi = OCFS2_I(inode);
>> 373 loff_t start;
>> 374 struct page *last;
354 375
355 /* 376 /*
356 * Use the nonblocking flag for the dl 377 * Use the nonblocking flag for the dlm code to avoid page
357 * lock inversion, but don't bother wi 378 * lock inversion, but don't bother with retrying.
358 */ 379 */
359 ret = ocfs2_inode_lock_full(inode, NUL 380 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
360 if (ret) 381 if (ret)
361 return; !! 382 return err;
362 383
363 if (down_read_trylock(&oi->ip_alloc_se !! 384 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
364 goto out_unlock; !! 385 ocfs2_inode_unlock(inode, 0);
>> 386 return err;
>> 387 }
365 388
366 /* 389 /*
367 * Don't bother with inline-data. Ther 390 * Don't bother with inline-data. There isn't anything
368 * to read-ahead in that case anyway.. 391 * to read-ahead in that case anyway...
369 */ 392 */
370 if (oi->ip_dyn_features & OCFS2_INLINE 393 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
371 goto out_up; !! 394 goto out_unlock;
372 395
373 /* 396 /*
374 * Check whether a remote node truncat 397 * Check whether a remote node truncated this file - we just
375 * drop out in that case as it's not w 398 * drop out in that case as it's not worth handling here.
376 */ 399 */
377 if (readahead_pos(rac) >= i_size_read( !! 400 last = list_entry(pages->prev, struct page, lru);
378 goto out_up; !! 401 start = (loff_t)last->index << PAGE_SHIFT;
>> 402 if (start >= i_size_read(inode))
>> 403 goto out_unlock;
379 404
380 mpage_readahead(rac, ocfs2_get_block); !! 405 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
381 406
382 out_up: <<
383 up_read(&oi->ip_alloc_sem); <<
384 out_unlock: 407 out_unlock:
>> 408 up_read(&oi->ip_alloc_sem);
385 ocfs2_inode_unlock(inode, 0); 409 ocfs2_inode_unlock(inode, 0);
>> 410
>> 411 return err;
386 } 412 }
387 413
388 /* Note: Because we don't support holes, our a 414 /* Note: Because we don't support holes, our allocation has
389 * already happened (allocation writes zeros t 415 * already happened (allocation writes zeros to the file data)
390 * so we don't have to worry about ordered wri 416 * so we don't have to worry about ordered writes in
391 * ocfs2_writepages. !! 417 * ocfs2_writepage.
392 * 418 *
393 * ->writepages is called during the process o !! 419 * ->writepage is called during the process of invalidating the page cache
394 * during blocked lock processing. It can't b 420 * during blocked lock processing. It can't block on any cluster locks
395 * to during block mapping. It's relying on t 421 * to during block mapping. It's relying on the fact that the block
396 * mapping can't have disappeared under the di 422 * mapping can't have disappeared under the dirty pages that it is
397 * being asked to write back. 423 * being asked to write back.
398 */ 424 */
399 static int ocfs2_writepages(struct address_spa !! 425 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
400 struct writeback_control *wbc) <<
401 { 426 {
402 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);
403 } 432 }
404 433
405 /* Taken from ext3. We don't necessarily need 434 /* Taken from ext3. We don't necessarily need the full blown
406 * functionality yet, but IMHO it's better to 435 * functionality yet, but IMHO it's better to cut and paste the whole
407 * thing so we can avoid introducing our own b 436 * thing so we can avoid introducing our own bugs (and easily pick up
408 * their fixes when they happen) --Mark */ 437 * their fixes when they happen) --Mark */
409 int walk_page_buffers( handle_t *handle, 438 int walk_page_buffers( handle_t *handle,
410 struct buffer_head *he 439 struct buffer_head *head,
411 unsigned from, 440 unsigned from,
412 unsigned to, 441 unsigned to,
413 int *partial, 442 int *partial,
414 int (*fn)( handle 443 int (*fn)( handle_t *handle,
415 struct 444 struct buffer_head *bh))
416 { 445 {
417 struct buffer_head *bh; 446 struct buffer_head *bh;
418 unsigned block_start, block_end; 447 unsigned block_start, block_end;
419 unsigned blocksize = head->b_size; 448 unsigned blocksize = head->b_size;
420 int err, ret = 0; 449 int err, ret = 0;
421 struct buffer_head *next; 450 struct buffer_head *next;
422 451
423 for ( bh = head, block_start = 0; 452 for ( bh = head, block_start = 0;
424 ret == 0 && (bh != head || !bl 453 ret == 0 && (bh != head || !block_start);
425 block_start = block_end, bh = 454 block_start = block_end, bh = next)
426 { 455 {
427 next = bh->b_this_page; 456 next = bh->b_this_page;
428 block_end = block_start + bloc 457 block_end = block_start + blocksize;
429 if (block_end <= from || block 458 if (block_end <= from || block_start >= to) {
430 if (partial && !buffer 459 if (partial && !buffer_uptodate(bh))
431 *partial = 1; 460 *partial = 1;
432 continue; 461 continue;
433 } 462 }
434 err = (*fn)(handle, bh); 463 err = (*fn)(handle, bh);
435 if (!ret) 464 if (!ret)
436 ret = err; 465 ret = err;
437 } 466 }
438 return ret; 467 return ret;
439 } 468 }
440 469
441 static sector_t ocfs2_bmap(struct address_spac 470 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
442 { 471 {
443 sector_t status; 472 sector_t status;
444 u64 p_blkno = 0; 473 u64 p_blkno = 0;
445 int err = 0; 474 int err = 0;
446 struct inode *inode = mapping->host; 475 struct inode *inode = mapping->host;
447 476
448 trace_ocfs2_bmap((unsigned long long)O 477 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
449 (unsigned long long)b 478 (unsigned long long)block);
450 479
451 /* 480 /*
452 * The swap code (ab-)uses ->bmap to g 481 * The swap code (ab-)uses ->bmap to get a block mapping and then
453 * bypasseѕ the file system for actua 482 * bypasseѕ the file system for actual I/O. We really can't allow
454 * that on refcounted inodes, so we ha 483 * that on refcounted inodes, so we have to skip out here. And yes,
455 * 0 is the magic code for a bmap erro 484 * 0 is the magic code for a bmap error..
456 */ 485 */
457 if (ocfs2_is_refcount_inode(inode)) 486 if (ocfs2_is_refcount_inode(inode))
458 return 0; 487 return 0;
459 488
460 /* We don't need to lock journal syste 489 /* We don't need to lock journal system files, since they aren't
461 * accessed concurrently from multiple 490 * accessed concurrently from multiple nodes.
462 */ 491 */
463 if (!INODE_JOURNAL(inode)) { 492 if (!INODE_JOURNAL(inode)) {
464 err = ocfs2_inode_lock(inode, 493 err = ocfs2_inode_lock(inode, NULL, 0);
465 if (err) { 494 if (err) {
466 if (err != -ENOENT) 495 if (err != -ENOENT)
467 mlog_errno(err 496 mlog_errno(err);
468 goto bail; 497 goto bail;
469 } 498 }
470 down_read(&OCFS2_I(inode)->ip_ 499 down_read(&OCFS2_I(inode)->ip_alloc_sem);
471 } 500 }
472 501
473 if (!(OCFS2_I(inode)->ip_dyn_features 502 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
474 err = ocfs2_extent_map_get_blo 503 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
475 504 NULL);
476 505
477 if (!INODE_JOURNAL(inode)) { 506 if (!INODE_JOURNAL(inode)) {
478 up_read(&OCFS2_I(inode)->ip_al 507 up_read(&OCFS2_I(inode)->ip_alloc_sem);
479 ocfs2_inode_unlock(inode, 0); 508 ocfs2_inode_unlock(inode, 0);
480 } 509 }
481 510
482 if (err) { 511 if (err) {
483 mlog(ML_ERROR, "get_blocks() f 512 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
484 (unsigned long long)block 513 (unsigned long long)block);
485 mlog_errno(err); 514 mlog_errno(err);
486 goto bail; 515 goto bail;
487 } 516 }
488 517
489 bail: 518 bail:
490 status = err ? 0 : p_blkno; 519 status = err ? 0 : p_blkno;
491 520
492 return status; 521 return status;
493 } 522 }
494 523
495 static bool ocfs2_release_folio(struct folio * !! 524 static int ocfs2_releasepage(struct page *page, gfp_t wait)
496 { 525 {
497 if (!folio_buffers(folio)) !! 526 if (!page_has_buffers(page))
498 return false; !! 527 return 0;
499 return try_to_free_buffers(folio); !! 528 return try_to_free_buffers(page);
500 } 529 }
501 530
502 static void ocfs2_figure_cluster_boundaries(st 531 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
503 u3 532 u32 cpos,
504 un 533 unsigned int *start,
505 un 534 unsigned int *end)
506 { 535 {
507 unsigned int cluster_start = 0, cluste 536 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
508 537
509 if (unlikely(PAGE_SHIFT > osb->s_clust 538 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
510 unsigned int cpp; 539 unsigned int cpp;
511 540
512 cpp = 1 << (PAGE_SHIFT - osb-> 541 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
513 542
514 cluster_start = cpos % cpp; 543 cluster_start = cpos % cpp;
515 cluster_start = cluster_start 544 cluster_start = cluster_start << osb->s_clustersize_bits;
516 545
517 cluster_end = cluster_start + 546 cluster_end = cluster_start + osb->s_clustersize;
518 } 547 }
519 548
520 BUG_ON(cluster_start > PAGE_SIZE); 549 BUG_ON(cluster_start > PAGE_SIZE);
521 BUG_ON(cluster_end > PAGE_SIZE); 550 BUG_ON(cluster_end > PAGE_SIZE);
522 551
523 if (start) 552 if (start)
524 *start = cluster_start; 553 *start = cluster_start;
525 if (end) 554 if (end)
526 *end = cluster_end; 555 *end = cluster_end;
527 } 556 }
528 557
529 /* 558 /*
530 * 'from' and 'to' are the region in the page 559 * 'from' and 'to' are the region in the page to avoid zeroing.
531 * 560 *
532 * If pagesize > clustersize, this function wi 561 * If pagesize > clustersize, this function will avoid zeroing outside
533 * of the cluster boundary. 562 * of the cluster boundary.
534 * 563 *
535 * from == to == 0 is code for "zero the entir 564 * from == to == 0 is code for "zero the entire cluster region"
536 */ 565 */
537 static void ocfs2_clear_page_regions(struct pa 566 static void ocfs2_clear_page_regions(struct page *page,
538 struct oc 567 struct ocfs2_super *osb, u32 cpos,
539 unsigned 568 unsigned from, unsigned to)
540 { 569 {
541 void *kaddr; 570 void *kaddr;
542 unsigned int cluster_start, cluster_en 571 unsigned int cluster_start, cluster_end;
543 572
544 ocfs2_figure_cluster_boundaries(osb, c 573 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
545 574
546 kaddr = kmap_atomic(page); 575 kaddr = kmap_atomic(page);
547 576
548 if (from || to) { 577 if (from || to) {
549 if (from > cluster_start) 578 if (from > cluster_start)
550 memset(kaddr + cluster 579 memset(kaddr + cluster_start, 0, from - cluster_start);
551 if (to < cluster_end) 580 if (to < cluster_end)
552 memset(kaddr + to, 0, 581 memset(kaddr + to, 0, cluster_end - to);
553 } else { 582 } else {
554 memset(kaddr + cluster_start, 583 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
555 } 584 }
556 585
557 kunmap_atomic(kaddr); 586 kunmap_atomic(kaddr);
558 } 587 }
559 588
560 /* 589 /*
561 * Nonsparse file systems fully allocate befor 590 * Nonsparse file systems fully allocate before we get to the write
562 * code. This prevents ocfs2_write() from tagg 591 * code. This prevents ocfs2_write() from tagging the write as an
563 * allocating one, which means ocfs2_map_page_ 592 * allocating one, which means ocfs2_map_page_blocks() might try to
564 * read-in the blocks at the tail of our file. 593 * read-in the blocks at the tail of our file. Avoid reading them by
565 * testing i_size against each block offset. 594 * testing i_size against each block offset.
566 */ 595 */
567 static int ocfs2_should_read_blk(struct inode !! 596 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
568 unsigned int 597 unsigned int block_start)
569 { 598 {
570 u64 offset = folio_pos(folio) + block_ !! 599 u64 offset = page_offset(page) + block_start;
571 600
572 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 601 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
573 return 1; 602 return 1;
574 603
575 if (i_size_read(inode) > offset) 604 if (i_size_read(inode) > offset)
576 return 1; 605 return 1;
577 606
578 return 0; 607 return 0;
579 } 608 }
580 609
581 /* 610 /*
582 * Some of this taken from __block_write_begin 611 * Some of this taken from __block_write_begin(). We already have our
583 * mapping by now though, and the entire write 612 * mapping by now though, and the entire write will be allocating or
584 * it won't, so not much need to use BH_New. 613 * it won't, so not much need to use BH_New.
585 * 614 *
586 * This will also skip zeroing, which is handl 615 * This will also skip zeroing, which is handled externally.
587 */ 616 */
588 int ocfs2_map_page_blocks(struct page *page, u 617 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
589 struct inode *inode, 618 struct inode *inode, unsigned int from,
590 unsigned int to, int 619 unsigned int to, int new)
591 { 620 {
592 struct folio *folio = page_folio(page) <<
593 int ret = 0; 621 int ret = 0;
594 struct buffer_head *head, *bh, *wait[2 622 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
595 unsigned int block_end, block_start; 623 unsigned int block_end, block_start;
596 unsigned int bsize = i_blocksize(inode 624 unsigned int bsize = i_blocksize(inode);
597 625
598 head = folio_buffers(folio); !! 626 if (!page_has_buffers(page))
599 if (!head) !! 627 create_empty_buffers(page, bsize, 0);
600 head = create_empty_buffers(fo <<
601 628
>> 629 head = page_buffers(page);
602 for (bh = head, block_start = 0; bh != 630 for (bh = head, block_start = 0; bh != head || !block_start;
603 bh = bh->b_this_page, block_start 631 bh = bh->b_this_page, block_start += bsize) {
604 block_end = block_start + bsiz 632 block_end = block_start + bsize;
605 633
606 clear_buffer_new(bh); 634 clear_buffer_new(bh);
607 635
608 /* 636 /*
609 * Ignore blocks outside of ou 637 * Ignore blocks outside of our i/o range -
610 * they may belong to unalloca 638 * they may belong to unallocated clusters.
611 */ 639 */
612 if (block_start >= to || block 640 if (block_start >= to || block_end <= from) {
613 if (folio_test_uptodat !! 641 if (PageUptodate(page))
614 set_buffer_upt 642 set_buffer_uptodate(bh);
615 continue; 643 continue;
616 } 644 }
617 645
618 /* 646 /*
619 * For an allocating write wit 647 * For an allocating write with cluster size >= page
620 * size, we always write the e 648 * size, we always write the entire page.
621 */ 649 */
622 if (new) 650 if (new)
623 set_buffer_new(bh); 651 set_buffer_new(bh);
624 652
625 if (!buffer_mapped(bh)) { 653 if (!buffer_mapped(bh)) {
626 map_bh(bh, inode->i_sb 654 map_bh(bh, inode->i_sb, *p_blkno);
627 clean_bdev_bh_alias(bh 655 clean_bdev_bh_alias(bh);
628 } 656 }
629 657
630 if (folio_test_uptodate(folio) !! 658 if (PageUptodate(page)) {
631 set_buffer_uptodate(bh !! 659 if (!buffer_uptodate(bh))
>> 660 set_buffer_uptodate(bh);
632 } else if (!buffer_uptodate(bh 661 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
633 !buffer_new(bh) && 662 !buffer_new(bh) &&
634 ocfs2_should_read_b !! 663 ocfs2_should_read_blk(inode, page, block_start) &&
635 (block_start < from 664 (block_start < from || block_end > to)) {
636 bh_read_nowait(bh, 0); !! 665 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
637 *wait_bh++=bh; 666 *wait_bh++=bh;
638 } 667 }
639 668
640 *p_blkno = *p_blkno + 1; 669 *p_blkno = *p_blkno + 1;
641 } 670 }
642 671
643 /* 672 /*
644 * If we issued read requests - let th 673 * If we issued read requests - let them complete.
645 */ 674 */
646 while(wait_bh > wait) { 675 while(wait_bh > wait) {
647 wait_on_buffer(*--wait_bh); 676 wait_on_buffer(*--wait_bh);
648 if (!buffer_uptodate(*wait_bh) 677 if (!buffer_uptodate(*wait_bh))
649 ret = -EIO; 678 ret = -EIO;
650 } 679 }
651 680
652 if (ret == 0 || !new) 681 if (ret == 0 || !new)
653 return ret; 682 return ret;
654 683
655 /* 684 /*
656 * If we get -EIO above, zero out any 685 * If we get -EIO above, zero out any newly allocated blocks
657 * to avoid exposing stale data. 686 * to avoid exposing stale data.
658 */ 687 */
659 bh = head; 688 bh = head;
660 block_start = 0; 689 block_start = 0;
661 do { 690 do {
662 block_end = block_start + bsiz 691 block_end = block_start + bsize;
663 if (block_end <= from) 692 if (block_end <= from)
664 goto next_bh; 693 goto next_bh;
665 if (block_start >= to) 694 if (block_start >= to)
666 break; 695 break;
667 696
668 folio_zero_range(folio, block_ !! 697 zero_user(page, block_start, bh->b_size);
669 set_buffer_uptodate(bh); 698 set_buffer_uptodate(bh);
670 mark_buffer_dirty(bh); 699 mark_buffer_dirty(bh);
671 700
672 next_bh: 701 next_bh:
673 block_start = block_end; 702 block_start = block_end;
674 bh = bh->b_this_page; 703 bh = bh->b_this_page;
675 } while (bh != head); 704 } while (bh != head);
676 705
677 return ret; 706 return ret;
678 } 707 }
679 708
680 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 709 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
681 #define OCFS2_MAX_CTXT_PAGES 1 710 #define OCFS2_MAX_CTXT_PAGES 1
682 #else 711 #else
683 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 712 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
684 #endif 713 #endif
685 714
686 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 715 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
687 716
688 struct ocfs2_unwritten_extent { 717 struct ocfs2_unwritten_extent {
689 struct list_head ue_node; 718 struct list_head ue_node;
690 struct list_head ue_ip_node; 719 struct list_head ue_ip_node;
691 u32 ue_cpos; 720 u32 ue_cpos;
692 u32 ue_phys; 721 u32 ue_phys;
693 }; 722 };
694 723
695 /* 724 /*
696 * Describe the state of a single cluster to b 725 * Describe the state of a single cluster to be written to.
697 */ 726 */
698 struct ocfs2_write_cluster_desc { 727 struct ocfs2_write_cluster_desc {
699 u32 c_cpos; 728 u32 c_cpos;
700 u32 c_phys; 729 u32 c_phys;
701 /* 730 /*
702 * Give this a unique field because c_ 731 * Give this a unique field because c_phys eventually gets
703 * filled. 732 * filled.
704 */ 733 */
705 unsigned c_new; 734 unsigned c_new;
706 unsigned c_clear_unwritten; 735 unsigned c_clear_unwritten;
707 unsigned c_needs_zero; 736 unsigned c_needs_zero;
708 }; 737 };
709 738
710 struct ocfs2_write_ctxt { 739 struct ocfs2_write_ctxt {
711 /* Logical cluster position / len of w 740 /* Logical cluster position / len of write */
712 u32 w_cpos 741 u32 w_cpos;
713 u32 w_clen 742 u32 w_clen;
714 743
715 /* First cluster allocated in a nonspa 744 /* First cluster allocated in a nonsparse extend */
716 u32 w_firs 745 u32 w_first_new_cpos;
717 746
718 /* Type of caller. Must be one of buff 747 /* Type of caller. Must be one of buffer, mmap, direct. */
719 ocfs2_write_type_t w_type 748 ocfs2_write_type_t w_type;
720 749
721 struct ocfs2_write_cluster_desc w_desc 750 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
722 751
723 /* 752 /*
724 * This is true if page_size > cluster 753 * This is true if page_size > cluster_size.
725 * 754 *
726 * It triggers a set of special cases 755 * It triggers a set of special cases during write which might
727 * have to deal with allocating writes 756 * have to deal with allocating writes to partial pages.
728 */ 757 */
729 unsigned int w_larg 758 unsigned int w_large_pages;
730 759
731 /* 760 /*
732 * Pages involved in this write. 761 * Pages involved in this write.
733 * 762 *
734 * w_target_page is the page being wri 763 * w_target_page is the page being written to by the user.
735 * 764 *
736 * w_pages is an array of pages which 765 * w_pages is an array of pages which always contains
737 * w_target_page, and in the case of a 766 * w_target_page, and in the case of an allocating write with
738 * page_size < cluster size, it will c 767 * page_size < cluster size, it will contain zero'd and mapped
739 * pages adjacent to w_target_page whi 768 * pages adjacent to w_target_page which need to be written
740 * out in so that future reads from th 769 * out in so that future reads from that region will get
741 * zero's. 770 * zero's.
742 */ 771 */
743 unsigned int w_num_ 772 unsigned int w_num_pages;
744 struct page *w_pag 773 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
745 struct page *w_tar 774 struct page *w_target_page;
746 775
747 /* 776 /*
748 * w_target_locked is used for page_mk 777 * w_target_locked is used for page_mkwrite path indicating no unlocking
749 * against w_target_page in ocfs2_writ 778 * against w_target_page in ocfs2_write_end_nolock.
750 */ 779 */
751 unsigned int w_targ 780 unsigned int w_target_locked:1;
752 781
753 /* 782 /*
754 * ocfs2_write_end() uses this to know 783 * ocfs2_write_end() uses this to know what the real range to
755 * write in the target should be. 784 * write in the target should be.
756 */ 785 */
757 unsigned int w_targ 786 unsigned int w_target_from;
758 unsigned int w_targ 787 unsigned int w_target_to;
759 788
760 /* 789 /*
761 * We could use journal_current_handle 790 * We could use journal_current_handle() but this is cleaner,
762 * IMHO -Mark 791 * IMHO -Mark
763 */ 792 */
764 handle_t *w_han 793 handle_t *w_handle;
765 794
766 struct buffer_head *w_di_ 795 struct buffer_head *w_di_bh;
767 796
768 struct ocfs2_cached_dealloc_ctxt w_dea 797 struct ocfs2_cached_dealloc_ctxt w_dealloc;
769 798
770 struct list_head w_unwr 799 struct list_head w_unwritten_list;
771 unsigned int w_unwr <<
772 }; 800 };
773 801
774 void ocfs2_unlock_and_free_pages(struct page * 802 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
775 { 803 {
776 int i; 804 int i;
777 805
778 for(i = 0; i < num_pages; i++) { 806 for(i = 0; i < num_pages; i++) {
779 if (pages[i]) { 807 if (pages[i]) {
780 unlock_page(pages[i]); 808 unlock_page(pages[i]);
781 mark_page_accessed(pag 809 mark_page_accessed(pages[i]);
782 put_page(pages[i]); 810 put_page(pages[i]);
783 } 811 }
784 } 812 }
785 } 813 }
786 814
787 static void ocfs2_unlock_pages(struct ocfs2_wr 815 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
788 { 816 {
789 int i; 817 int i;
790 818
791 /* 819 /*
792 * w_target_locked is only set to true 820 * w_target_locked is only set to true in the page_mkwrite() case.
793 * The intent is to allow us to lock t 821 * The intent is to allow us to lock the target page from write_begin()
794 * to write_end(). The caller must hol 822 * to write_end(). The caller must hold a ref on w_target_page.
795 */ 823 */
796 if (wc->w_target_locked) { 824 if (wc->w_target_locked) {
797 BUG_ON(!wc->w_target_page); 825 BUG_ON(!wc->w_target_page);
798 for (i = 0; i < wc->w_num_page 826 for (i = 0; i < wc->w_num_pages; i++) {
799 if (wc->w_target_page 827 if (wc->w_target_page == wc->w_pages[i]) {
800 wc->w_pages[i] 828 wc->w_pages[i] = NULL;
801 break; 829 break;
802 } 830 }
803 } 831 }
804 mark_page_accessed(wc->w_targe 832 mark_page_accessed(wc->w_target_page);
805 put_page(wc->w_target_page); 833 put_page(wc->w_target_page);
806 } 834 }
807 ocfs2_unlock_and_free_pages(wc->w_page 835 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
808 } 836 }
809 837
810 static void ocfs2_free_unwritten_list(struct i 838 static void ocfs2_free_unwritten_list(struct inode *inode,
811 struct list_h 839 struct list_head *head)
812 { 840 {
813 struct ocfs2_inode_info *oi = OCFS2_I( 841 struct ocfs2_inode_info *oi = OCFS2_I(inode);
814 struct ocfs2_unwritten_extent *ue = NU 842 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
815 843
816 list_for_each_entry_safe(ue, tmp, head 844 list_for_each_entry_safe(ue, tmp, head, ue_node) {
817 list_del(&ue->ue_node); 845 list_del(&ue->ue_node);
818 spin_lock(&oi->ip_lock); 846 spin_lock(&oi->ip_lock);
819 list_del(&ue->ue_ip_node); 847 list_del(&ue->ue_ip_node);
820 spin_unlock(&oi->ip_lock); 848 spin_unlock(&oi->ip_lock);
821 kfree(ue); 849 kfree(ue);
822 } 850 }
823 } 851 }
824 852
825 static void ocfs2_free_write_ctxt(struct inode 853 static void ocfs2_free_write_ctxt(struct inode *inode,
826 struct ocfs2 854 struct ocfs2_write_ctxt *wc)
827 { 855 {
828 ocfs2_free_unwritten_list(inode, &wc-> 856 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
829 ocfs2_unlock_pages(wc); 857 ocfs2_unlock_pages(wc);
830 brelse(wc->w_di_bh); 858 brelse(wc->w_di_bh);
831 kfree(wc); 859 kfree(wc);
832 } 860 }
833 861
834 static int ocfs2_alloc_write_ctxt(struct ocfs2 862 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
835 struct ocfs2 863 struct ocfs2_super *osb, loff_t pos,
836 unsigned len 864 unsigned len, ocfs2_write_type_t type,
837 struct buffe 865 struct buffer_head *di_bh)
838 { 866 {
839 u32 cend; 867 u32 cend;
840 struct ocfs2_write_ctxt *wc; 868 struct ocfs2_write_ctxt *wc;
841 869
842 wc = kzalloc(sizeof(struct ocfs2_write 870 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
843 if (!wc) 871 if (!wc)
844 return -ENOMEM; 872 return -ENOMEM;
845 873
846 wc->w_cpos = pos >> osb->s_clustersize 874 wc->w_cpos = pos >> osb->s_clustersize_bits;
847 wc->w_first_new_cpos = UINT_MAX; 875 wc->w_first_new_cpos = UINT_MAX;
848 cend = (pos + len - 1) >> osb->s_clust 876 cend = (pos + len - 1) >> osb->s_clustersize_bits;
849 wc->w_clen = cend - wc->w_cpos + 1; 877 wc->w_clen = cend - wc->w_cpos + 1;
850 get_bh(di_bh); 878 get_bh(di_bh);
851 wc->w_di_bh = di_bh; 879 wc->w_di_bh = di_bh;
852 wc->w_type = type; 880 wc->w_type = type;
853 881
854 if (unlikely(PAGE_SHIFT > osb->s_clust 882 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
855 wc->w_large_pages = 1; 883 wc->w_large_pages = 1;
856 else 884 else
857 wc->w_large_pages = 0; 885 wc->w_large_pages = 0;
858 886
859 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 887 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
860 INIT_LIST_HEAD(&wc->w_unwritten_list); 888 INIT_LIST_HEAD(&wc->w_unwritten_list);
861 889
862 *wcp = wc; 890 *wcp = wc;
863 891
864 return 0; 892 return 0;
865 } 893 }
866 894
867 /* 895 /*
868 * If a page has any new buffers, zero them ou 896 * If a page has any new buffers, zero them out here, and mark them uptodate
869 * and dirty so they'll be written out (in ord 897 * and dirty so they'll be written out (in order to prevent uninitialised
870 * block data from leaking). And clear the new 898 * block data from leaking). And clear the new bit.
871 */ 899 */
872 static void ocfs2_zero_new_buffers(struct page 900 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
873 { 901 {
874 unsigned int block_start, block_end; 902 unsigned int block_start, block_end;
875 struct buffer_head *head, *bh; 903 struct buffer_head *head, *bh;
876 904
877 BUG_ON(!PageLocked(page)); 905 BUG_ON(!PageLocked(page));
878 if (!page_has_buffers(page)) 906 if (!page_has_buffers(page))
879 return; 907 return;
880 908
881 bh = head = page_buffers(page); 909 bh = head = page_buffers(page);
882 block_start = 0; 910 block_start = 0;
883 do { 911 do {
884 block_end = block_start + bh-> 912 block_end = block_start + bh->b_size;
885 913
886 if (buffer_new(bh)) { 914 if (buffer_new(bh)) {
887 if (block_end > from & 915 if (block_end > from && block_start < to) {
888 if (!PageUptod 916 if (!PageUptodate(page)) {
889 unsign 917 unsigned start, end;
890 918
891 start 919 start = max(from, block_start);
892 end = 920 end = min(to, block_end);
893 921
894 zero_u 922 zero_user_segment(page, start, end);
895 set_bu 923 set_buffer_uptodate(bh);
896 } 924 }
897 925
898 clear_buffer_n 926 clear_buffer_new(bh);
899 mark_buffer_di 927 mark_buffer_dirty(bh);
900 } 928 }
901 } 929 }
902 930
903 block_start = block_end; 931 block_start = block_end;
904 bh = bh->b_this_page; 932 bh = bh->b_this_page;
905 } while (bh != head); 933 } while (bh != head);
906 } 934 }
907 935
908 /* 936 /*
909 * Only called when we have a failure during a 937 * Only called when we have a failure during allocating write to write
910 * zero's to the newly allocated region. 938 * zero's to the newly allocated region.
911 */ 939 */
912 static void ocfs2_write_failure(struct inode * 940 static void ocfs2_write_failure(struct inode *inode,
913 struct ocfs2_w 941 struct ocfs2_write_ctxt *wc,
914 loff_t user_po 942 loff_t user_pos, unsigned user_len)
915 { 943 {
916 int i; 944 int i;
917 unsigned from = user_pos & (PAGE_SIZE 945 unsigned from = user_pos & (PAGE_SIZE - 1),
918 to = user_pos + user_len; 946 to = user_pos + user_len;
919 struct page *tmppage; 947 struct page *tmppage;
920 948
921 if (wc->w_target_page) 949 if (wc->w_target_page)
922 ocfs2_zero_new_buffers(wc->w_t 950 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
923 951
924 for(i = 0; i < wc->w_num_pages; i++) { 952 for(i = 0; i < wc->w_num_pages; i++) {
925 tmppage = wc->w_pages[i]; 953 tmppage = wc->w_pages[i];
926 954
927 if (tmppage && page_has_buffer 955 if (tmppage && page_has_buffers(tmppage)) {
928 if (ocfs2_should_order 956 if (ocfs2_should_order_data(inode))
929 ocfs2_jbd2_ino !! 957 ocfs2_jbd2_file_inode(wc->w_handle, inode);
930 <<
931 958
932 block_commit_write(tmp 959 block_commit_write(tmppage, from, to);
933 } 960 }
934 } 961 }
935 } 962 }
936 963
937 static int ocfs2_prepare_page_for_write(struct 964 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
938 struct 965 struct ocfs2_write_ctxt *wc,
939 struct 966 struct page *page, u32 cpos,
940 loff_t 967 loff_t user_pos, unsigned user_len,
941 int ne 968 int new)
942 { 969 {
943 int ret; 970 int ret;
944 unsigned int map_from = 0, map_to = 0; 971 unsigned int map_from = 0, map_to = 0;
945 unsigned int cluster_start, cluster_en 972 unsigned int cluster_start, cluster_end;
946 unsigned int user_data_from = 0, user_ 973 unsigned int user_data_from = 0, user_data_to = 0;
947 974
948 ocfs2_figure_cluster_boundaries(OCFS2_ 975 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
949 &clust 976 &cluster_start, &cluster_end);
950 977
951 /* treat the write as new if the a hol 978 /* treat the write as new if the a hole/lseek spanned across
952 * the page boundary. 979 * the page boundary.
953 */ 980 */
954 new = new | ((i_size_read(inode) <= pa 981 new = new | ((i_size_read(inode) <= page_offset(page)) &&
955 (page_offset(page) <= 982 (page_offset(page) <= user_pos));
956 983
957 if (page == wc->w_target_page) { 984 if (page == wc->w_target_page) {
958 map_from = user_pos & (PAGE_SI 985 map_from = user_pos & (PAGE_SIZE - 1);
959 map_to = map_from + user_len; 986 map_to = map_from + user_len;
960 987
961 if (new) 988 if (new)
962 ret = ocfs2_map_page_b 989 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
963 990 cluster_start, cluster_end,
964 991 new);
965 else 992 else
966 ret = ocfs2_map_page_b 993 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
967 994 map_from, map_to, new);
968 if (ret) { 995 if (ret) {
969 mlog_errno(ret); 996 mlog_errno(ret);
970 goto out; 997 goto out;
971 } 998 }
972 999
973 user_data_from = map_from; 1000 user_data_from = map_from;
974 user_data_to = map_to; 1001 user_data_to = map_to;
975 if (new) { 1002 if (new) {
976 map_from = cluster_sta 1003 map_from = cluster_start;
977 map_to = cluster_end; 1004 map_to = cluster_end;
978 } 1005 }
979 } else { 1006 } else {
980 /* 1007 /*
981 * If we haven't allocated the 1008 * If we haven't allocated the new page yet, we
982 * shouldn't be writing it out 1009 * shouldn't be writing it out without copying user
983 * data. This is likely a math 1010 * data. This is likely a math error from the caller.
984 */ 1011 */
985 BUG_ON(!new); 1012 BUG_ON(!new);
986 1013
987 map_from = cluster_start; 1014 map_from = cluster_start;
988 map_to = cluster_end; 1015 map_to = cluster_end;
989 1016
990 ret = ocfs2_map_page_blocks(pa 1017 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
991 cl 1018 cluster_start, cluster_end, new);
992 if (ret) { 1019 if (ret) {
993 mlog_errno(ret); 1020 mlog_errno(ret);
994 goto out; 1021 goto out;
995 } 1022 }
996 } 1023 }
997 1024
998 /* 1025 /*
999 * Parts of newly allocated pages need 1026 * Parts of newly allocated pages need to be zero'd.
1000 * 1027 *
1001 * Above, we have also rewritten 'to' 1028 * Above, we have also rewritten 'to' and 'from' - as far as
1002 * the rest of the function is concer 1029 * the rest of the function is concerned, the entire cluster
1003 * range inside of a page needs to be 1030 * range inside of a page needs to be written.
1004 * 1031 *
1005 * We can skip this if the page is up 1032 * We can skip this if the page is up to date - it's already
1006 * been zero'd from being read in as 1033 * been zero'd from being read in as a hole.
1007 */ 1034 */
1008 if (new && !PageUptodate(page)) 1035 if (new && !PageUptodate(page))
1009 ocfs2_clear_page_regions(page 1036 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1010 cpos 1037 cpos, user_data_from, user_data_to);
1011 1038
1012 flush_dcache_page(page); 1039 flush_dcache_page(page);
1013 1040
1014 out: 1041 out:
1015 return ret; 1042 return ret;
1016 } 1043 }
1017 1044
1018 /* 1045 /*
1019 * This function will only grab one clusters 1046 * This function will only grab one clusters worth of pages.
1020 */ 1047 */
1021 static int ocfs2_grab_pages_for_write(struct 1048 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1022 struct 1049 struct ocfs2_write_ctxt *wc,
1023 u32 cpo 1050 u32 cpos, loff_t user_pos,
1024 unsigne 1051 unsigned user_len, int new,
1025 struct 1052 struct page *mmap_page)
1026 { 1053 {
1027 int ret = 0, i; 1054 int ret = 0, i;
1028 unsigned long start, target_index, en 1055 unsigned long start, target_index, end_index, index;
1029 struct inode *inode = mapping->host; 1056 struct inode *inode = mapping->host;
1030 loff_t last_byte; 1057 loff_t last_byte;
1031 1058
1032 target_index = user_pos >> PAGE_SHIFT 1059 target_index = user_pos >> PAGE_SHIFT;
1033 1060
1034 /* 1061 /*
1035 * Figure out how many pages we'll be 1062 * Figure out how many pages we'll be manipulating here. For
1036 * non allocating write, we just chan 1063 * non allocating write, we just change the one
1037 * page. Otherwise, we'll need a whol 1064 * page. Otherwise, we'll need a whole clusters worth. If we're
1038 * writing past i_size, we only need 1065 * writing past i_size, we only need enough pages to cover the
1039 * last page of the write. 1066 * last page of the write.
1040 */ 1067 */
1041 if (new) { 1068 if (new) {
1042 wc->w_num_pages = ocfs2_pages 1069 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1043 start = ocfs2_align_clusters_ 1070 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1044 /* 1071 /*
1045 * We need the index *past* t 1072 * We need the index *past* the last page we could possibly
1046 * touch. This is the page p 1073 * touch. This is the page past the end of the write or
1047 * i_size, whichever is great 1074 * i_size, whichever is greater.
1048 */ 1075 */
1049 last_byte = max(user_pos + us 1076 last_byte = max(user_pos + user_len, i_size_read(inode));
1050 BUG_ON(last_byte < 1); 1077 BUG_ON(last_byte < 1);
1051 end_index = ((last_byte - 1) 1078 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
1052 if ((start + wc->w_num_pages) 1079 if ((start + wc->w_num_pages) > end_index)
1053 wc->w_num_pages = end 1080 wc->w_num_pages = end_index - start;
1054 } else { 1081 } else {
1055 wc->w_num_pages = 1; 1082 wc->w_num_pages = 1;
1056 start = target_index; 1083 start = target_index;
1057 } 1084 }
1058 end_index = (user_pos + user_len - 1) 1085 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
1059 1086
1060 for(i = 0; i < wc->w_num_pages; i++) 1087 for(i = 0; i < wc->w_num_pages; i++) {
1061 index = start + i; 1088 index = start + i;
1062 1089
1063 if (index >= target_index && 1090 if (index >= target_index && index <= end_index &&
1064 wc->w_type == OCFS2_WRITE 1091 wc->w_type == OCFS2_WRITE_MMAP) {
1065 /* 1092 /*
1066 * ocfs2_pagemkwrite( 1093 * ocfs2_pagemkwrite() is a little different
1067 * and wants us to di 1094 * and wants us to directly use the page
1068 * passed in. 1095 * passed in.
1069 */ 1096 */
1070 lock_page(mmap_page); 1097 lock_page(mmap_page);
1071 1098
1072 /* Exit and let the c 1099 /* Exit and let the caller retry */
1073 if (mmap_page->mappin 1100 if (mmap_page->mapping != mapping) {
1074 WARN_ON(mmap_ 1101 WARN_ON(mmap_page->mapping);
1075 unlock_page(m 1102 unlock_page(mmap_page);
1076 ret = -EAGAIN 1103 ret = -EAGAIN;
1077 goto out; 1104 goto out;
1078 } 1105 }
1079 1106
1080 get_page(mmap_page); 1107 get_page(mmap_page);
1081 wc->w_pages[i] = mmap 1108 wc->w_pages[i] = mmap_page;
1082 wc->w_target_locked = 1109 wc->w_target_locked = true;
1083 } else if (index >= target_in 1110 } else if (index >= target_index && index <= end_index &&
1084 wc->w_type == OCFS 1111 wc->w_type == OCFS2_WRITE_DIRECT) {
1085 /* Direct write has n 1112 /* Direct write has no mapping page. */
1086 wc->w_pages[i] = NULL 1113 wc->w_pages[i] = NULL;
1087 continue; 1114 continue;
1088 } else { 1115 } else {
1089 wc->w_pages[i] = find 1116 wc->w_pages[i] = find_or_create_page(mapping, index,
1090 1117 GFP_NOFS);
1091 if (!wc->w_pages[i]) 1118 if (!wc->w_pages[i]) {
1092 ret = -ENOMEM 1119 ret = -ENOMEM;
1093 mlog_errno(re 1120 mlog_errno(ret);
1094 goto out; 1121 goto out;
1095 } 1122 }
1096 } 1123 }
1097 wait_for_stable_page(wc->w_pa 1124 wait_for_stable_page(wc->w_pages[i]);
1098 1125
1099 if (index == target_index) 1126 if (index == target_index)
1100 wc->w_target_page = w 1127 wc->w_target_page = wc->w_pages[i];
1101 } 1128 }
1102 out: 1129 out:
1103 if (ret) 1130 if (ret)
1104 wc->w_target_locked = false; 1131 wc->w_target_locked = false;
1105 return ret; 1132 return ret;
1106 } 1133 }
1107 1134
1108 /* 1135 /*
1109 * Prepare a single cluster for write one clu 1136 * Prepare a single cluster for write one cluster into the file.
1110 */ 1137 */
1111 static int ocfs2_write_cluster(struct address 1138 static int ocfs2_write_cluster(struct address_space *mapping,
1112 u32 *phys, uns 1139 u32 *phys, unsigned int new,
1113 unsigned int c 1140 unsigned int clear_unwritten,
1114 unsigned int s 1141 unsigned int should_zero,
1115 struct ocfs2_a 1142 struct ocfs2_alloc_context *data_ac,
1116 struct ocfs2_a 1143 struct ocfs2_alloc_context *meta_ac,
1117 struct ocfs2_w 1144 struct ocfs2_write_ctxt *wc, u32 cpos,
1118 loff_t user_po 1145 loff_t user_pos, unsigned user_len)
1119 { 1146 {
1120 int ret, i; 1147 int ret, i;
1121 u64 p_blkno; 1148 u64 p_blkno;
1122 struct inode *inode = mapping->host; 1149 struct inode *inode = mapping->host;
1123 struct ocfs2_extent_tree et; 1150 struct ocfs2_extent_tree et;
1124 int bpc = ocfs2_clusters_to_blocks(in 1151 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1125 1152
1126 if (new) { 1153 if (new) {
1127 u32 tmp_pos; 1154 u32 tmp_pos;
1128 1155
1129 /* 1156 /*
1130 * This is safe to call with 1157 * This is safe to call with the page locks - it won't take
1131 * any additional semaphores 1158 * any additional semaphores or cluster locks.
1132 */ 1159 */
1133 tmp_pos = cpos; 1160 tmp_pos = cpos;
1134 ret = ocfs2_add_inode_data(OC 1161 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1135 &t 1162 &tmp_pos, 1, !clear_unwritten,
1136 wc 1163 wc->w_di_bh, wc->w_handle,
1137 da 1164 data_ac, meta_ac, NULL);
1138 /* 1165 /*
1139 * This shouldn't happen beca 1166 * This shouldn't happen because we must have already
1140 * calculated the correct met 1167 * calculated the correct meta data allocation required. The
1141 * internal tree allocation c 1168 * internal tree allocation code should know how to increase
1142 * transaction credits itself 1169 * transaction credits itself.
1143 * 1170 *
1144 * If need be, we could handl 1171 * If need be, we could handle -EAGAIN for a
1145 * RESTART_TRANS here. 1172 * RESTART_TRANS here.
1146 */ 1173 */
1147 mlog_bug_on_msg(ret == -EAGAI 1174 mlog_bug_on_msg(ret == -EAGAIN,
1148 "Inode %llu: 1175 "Inode %llu: EAGAIN return during allocation.\n",
1149 (unsigned lon 1176 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1150 if (ret < 0) { 1177 if (ret < 0) {
1151 mlog_errno(ret); 1178 mlog_errno(ret);
1152 goto out; 1179 goto out;
1153 } 1180 }
1154 } else if (clear_unwritten) { 1181 } else if (clear_unwritten) {
1155 ocfs2_init_dinode_extent_tree 1182 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1156 1183 wc->w_di_bh);
1157 ret = ocfs2_mark_extent_writt 1184 ret = ocfs2_mark_extent_written(inode, &et,
1158 1185 wc->w_handle, cpos, 1, *phys,
1159 1186 meta_ac, &wc->w_dealloc);
1160 if (ret < 0) { 1187 if (ret < 0) {
1161 mlog_errno(ret); 1188 mlog_errno(ret);
1162 goto out; 1189 goto out;
1163 } 1190 }
1164 } 1191 }
1165 1192
1166 /* 1193 /*
1167 * The only reason this should fail i 1194 * The only reason this should fail is due to an inability to
1168 * find the extent added. 1195 * find the extent added.
1169 */ 1196 */
1170 ret = ocfs2_get_clusters(inode, cpos, 1197 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
1171 if (ret < 0) { 1198 if (ret < 0) {
1172 mlog(ML_ERROR, "Get physical 1199 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1173 "at logical clust 1200 "at logical cluster %u",
1174 (unsigned long lo 1201 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
1175 goto out; 1202 goto out;
1176 } 1203 }
1177 1204
1178 BUG_ON(*phys == 0); 1205 BUG_ON(*phys == 0);
1179 1206
1180 p_blkno = ocfs2_clusters_to_blocks(in 1207 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1181 if (!should_zero) 1208 if (!should_zero)
1182 p_blkno += (user_pos >> inode 1209 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
1183 1210
1184 for(i = 0; i < wc->w_num_pages; i++) 1211 for(i = 0; i < wc->w_num_pages; i++) {
1185 int tmpret; 1212 int tmpret;
1186 1213
1187 /* This is the direct io targ 1214 /* This is the direct io target page. */
1188 if (wc->w_pages[i] == NULL) { 1215 if (wc->w_pages[i] == NULL) {
1189 p_blkno++; 1216 p_blkno++;
1190 continue; 1217 continue;
1191 } 1218 }
1192 1219
1193 tmpret = ocfs2_prepare_page_f 1220 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1194 1221 wc->w_pages[i], cpos,
1195 1222 user_pos, user_len,
1196 1223 should_zero);
1197 if (tmpret) { 1224 if (tmpret) {
1198 mlog_errno(tmpret); 1225 mlog_errno(tmpret);
1199 if (ret == 0) 1226 if (ret == 0)
1200 ret = tmpret; 1227 ret = tmpret;
1201 } 1228 }
1202 } 1229 }
1203 1230
1204 /* 1231 /*
1205 * We only have cleanup to do in case 1232 * We only have cleanup to do in case of allocating write.
1206 */ 1233 */
1207 if (ret && new) 1234 if (ret && new)
1208 ocfs2_write_failure(inode, wc 1235 ocfs2_write_failure(inode, wc, user_pos, user_len);
1209 1236
1210 out: 1237 out:
1211 1238
1212 return ret; 1239 return ret;
1213 } 1240 }
1214 1241
1215 static int ocfs2_write_cluster_by_desc(struct 1242 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1216 struct 1243 struct ocfs2_alloc_context *data_ac,
1217 struct 1244 struct ocfs2_alloc_context *meta_ac,
1218 struct 1245 struct ocfs2_write_ctxt *wc,
1219 loff_t 1246 loff_t pos, unsigned len)
1220 { 1247 {
1221 int ret, i; 1248 int ret, i;
1222 loff_t cluster_off; 1249 loff_t cluster_off;
1223 unsigned int local_len = len; 1250 unsigned int local_len = len;
1224 struct ocfs2_write_cluster_desc *desc 1251 struct ocfs2_write_cluster_desc *desc;
1225 struct ocfs2_super *osb = OCFS2_SB(ma 1252 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1226 1253
1227 for (i = 0; i < wc->w_clen; i++) { 1254 for (i = 0; i < wc->w_clen; i++) {
1228 desc = &wc->w_desc[i]; 1255 desc = &wc->w_desc[i];
1229 1256
1230 /* 1257 /*
1231 * We have to make sure that 1258 * We have to make sure that the total write passed in
1232 * doesn't extend past a sing 1259 * doesn't extend past a single cluster.
1233 */ 1260 */
1234 local_len = len; 1261 local_len = len;
1235 cluster_off = pos & (osb->s_c 1262 cluster_off = pos & (osb->s_clustersize - 1);
1236 if ((cluster_off + local_len) 1263 if ((cluster_off + local_len) > osb->s_clustersize)
1237 local_len = osb->s_cl 1264 local_len = osb->s_clustersize - cluster_off;
1238 1265
1239 ret = ocfs2_write_cluster(map 1266 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
1240 des 1267 desc->c_new,
1241 des 1268 desc->c_clear_unwritten,
1242 des 1269 desc->c_needs_zero,
1243 dat 1270 data_ac, meta_ac,
1244 wc, 1271 wc, desc->c_cpos, pos, local_len);
1245 if (ret) { 1272 if (ret) {
1246 mlog_errno(ret); 1273 mlog_errno(ret);
1247 goto out; 1274 goto out;
1248 } 1275 }
1249 1276
1250 len -= local_len; 1277 len -= local_len;
1251 pos += local_len; 1278 pos += local_len;
1252 } 1279 }
1253 1280
1254 ret = 0; 1281 ret = 0;
1255 out: 1282 out:
1256 return ret; 1283 return ret;
1257 } 1284 }
1258 1285
1259 /* 1286 /*
1260 * ocfs2_write_end() wants to know which part 1287 * ocfs2_write_end() wants to know which parts of the target page it
1261 * should complete the write on. It's easiest 1288 * should complete the write on. It's easiest to compute them ahead of
1262 * time when a more complete view of the writ 1289 * time when a more complete view of the write is available.
1263 */ 1290 */
1264 static void ocfs2_set_target_boundaries(struc 1291 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1265 struc 1292 struct ocfs2_write_ctxt *wc,
1266 loff_ 1293 loff_t pos, unsigned len, int alloc)
1267 { 1294 {
1268 struct ocfs2_write_cluster_desc *desc 1295 struct ocfs2_write_cluster_desc *desc;
1269 1296
1270 wc->w_target_from = pos & (PAGE_SIZE 1297 wc->w_target_from = pos & (PAGE_SIZE - 1);
1271 wc->w_target_to = wc->w_target_from + 1298 wc->w_target_to = wc->w_target_from + len;
1272 1299
1273 if (alloc == 0) 1300 if (alloc == 0)
1274 return; 1301 return;
1275 1302
1276 /* 1303 /*
1277 * Allocating write - we may have dif 1304 * Allocating write - we may have different boundaries based
1278 * on page size and cluster size. 1305 * on page size and cluster size.
1279 * 1306 *
1280 * NOTE: We can no longer compute one 1307 * NOTE: We can no longer compute one value from the other as
1281 * the actual write length and user p 1308 * the actual write length and user provided length may be
1282 * different. 1309 * different.
1283 */ 1310 */
1284 1311
1285 if (wc->w_large_pages) { 1312 if (wc->w_large_pages) {
1286 /* 1313 /*
1287 * We only care about the 1st 1314 * We only care about the 1st and last cluster within
1288 * our range and whether they 1315 * our range and whether they should be zero'd or not. Either
1289 * value may be extended out 1316 * value may be extended out to the start/end of a
1290 * newly allocated cluster. 1317 * newly allocated cluster.
1291 */ 1318 */
1292 desc = &wc->w_desc[0]; 1319 desc = &wc->w_desc[0];
1293 if (desc->c_needs_zero) 1320 if (desc->c_needs_zero)
1294 ocfs2_figure_cluster_ 1321 ocfs2_figure_cluster_boundaries(osb,
1295 1322 desc->c_cpos,
1296 1323 &wc->w_target_from,
1297 1324 NULL);
1298 1325
1299 desc = &wc->w_desc[wc->w_clen 1326 desc = &wc->w_desc[wc->w_clen - 1];
1300 if (desc->c_needs_zero) 1327 if (desc->c_needs_zero)
1301 ocfs2_figure_cluster_ 1328 ocfs2_figure_cluster_boundaries(osb,
1302 1329 desc->c_cpos,
1303 1330 NULL,
1304 1331 &wc->w_target_to);
1305 } else { 1332 } else {
1306 wc->w_target_from = 0; 1333 wc->w_target_from = 0;
1307 wc->w_target_to = PAGE_SIZE; 1334 wc->w_target_to = PAGE_SIZE;
1308 } 1335 }
1309 } 1336 }
1310 1337
1311 /* 1338 /*
1312 * Check if this extent is marked UNWRITTEN b 1339 * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to
1313 * do the zero work. And should not to clear 1340 * do the zero work. And should not to clear UNWRITTEN since it will be cleared
1314 * by the direct io procedure. 1341 * by the direct io procedure.
1315 * If this is a new extent that allocated by 1342 * If this is a new extent that allocated by direct io, we should mark it in
1316 * the ip_unwritten_list. 1343 * the ip_unwritten_list.
1317 */ 1344 */
1318 static int ocfs2_unwritten_check(struct inode 1345 static int ocfs2_unwritten_check(struct inode *inode,
1319 struct ocfs2 1346 struct ocfs2_write_ctxt *wc,
1320 struct ocfs2 1347 struct ocfs2_write_cluster_desc *desc)
1321 { 1348 {
1322 struct ocfs2_inode_info *oi = OCFS2_I 1349 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1323 struct ocfs2_unwritten_extent *ue = N 1350 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
1324 int ret = 0; 1351 int ret = 0;
1325 1352
1326 if (!desc->c_needs_zero) 1353 if (!desc->c_needs_zero)
1327 return 0; 1354 return 0;
1328 1355
1329 retry: 1356 retry:
1330 spin_lock(&oi->ip_lock); 1357 spin_lock(&oi->ip_lock);
1331 /* Needs not to zero no metter buffer 1358 /* Needs not to zero no metter buffer or direct. The one who is zero
1332 * the cluster is doing zero. And he 1359 * the cluster is doing zero. And he will clear unwritten after all
1333 * cluster io finished. */ 1360 * cluster io finished. */
1334 list_for_each_entry(ue, &oi->ip_unwri 1361 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1335 if (desc->c_cpos == ue->ue_cp 1362 if (desc->c_cpos == ue->ue_cpos) {
1336 BUG_ON(desc->c_new); 1363 BUG_ON(desc->c_new);
1337 desc->c_needs_zero = 1364 desc->c_needs_zero = 0;
1338 desc->c_clear_unwritt 1365 desc->c_clear_unwritten = 0;
1339 goto unlock; 1366 goto unlock;
1340 } 1367 }
1341 } 1368 }
1342 1369
1343 if (wc->w_type != OCFS2_WRITE_DIRECT) 1370 if (wc->w_type != OCFS2_WRITE_DIRECT)
1344 goto unlock; 1371 goto unlock;
1345 1372
1346 if (new == NULL) { 1373 if (new == NULL) {
1347 spin_unlock(&oi->ip_lock); 1374 spin_unlock(&oi->ip_lock);
1348 new = kmalloc(sizeof(struct o 1375 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1349 GFP_NOFS); 1376 GFP_NOFS);
1350 if (new == NULL) { 1377 if (new == NULL) {
1351 ret = -ENOMEM; 1378 ret = -ENOMEM;
1352 goto out; 1379 goto out;
1353 } 1380 }
1354 goto retry; 1381 goto retry;
1355 } 1382 }
1356 /* This direct write will doing zero. 1383 /* This direct write will doing zero. */
1357 new->ue_cpos = desc->c_cpos; 1384 new->ue_cpos = desc->c_cpos;
1358 new->ue_phys = desc->c_phys; 1385 new->ue_phys = desc->c_phys;
1359 desc->c_clear_unwritten = 0; 1386 desc->c_clear_unwritten = 0;
1360 list_add_tail(&new->ue_ip_node, &oi-> 1387 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1361 list_add_tail(&new->ue_node, &wc->w_u 1388 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
1362 wc->w_unwritten_count++; <<
1363 new = NULL; 1389 new = NULL;
1364 unlock: 1390 unlock:
1365 spin_unlock(&oi->ip_lock); 1391 spin_unlock(&oi->ip_lock);
1366 out: 1392 out:
1367 kfree(new); !! 1393 if (new)
>> 1394 kfree(new);
1368 return ret; 1395 return ret;
1369 } 1396 }
1370 1397
1371 /* 1398 /*
1372 * Populate each single-cluster write descrip 1399 * Populate each single-cluster write descriptor in the write context
1373 * with information about the i/o to be done. 1400 * with information about the i/o to be done.
1374 * 1401 *
1375 * Returns the number of clusters that will h 1402 * Returns the number of clusters that will have to be allocated, as
1376 * well as a worst case estimate of the numbe 1403 * well as a worst case estimate of the number of extent records that
1377 * would have to be created during a write to 1404 * would have to be created during a write to an unwritten region.
1378 */ 1405 */
1379 static int ocfs2_populate_write_desc(struct i 1406 static int ocfs2_populate_write_desc(struct inode *inode,
1380 struct o 1407 struct ocfs2_write_ctxt *wc,
1381 unsigned 1408 unsigned int *clusters_to_alloc,
1382 unsigned 1409 unsigned int *extents_to_split)
1383 { 1410 {
1384 int ret; 1411 int ret;
1385 struct ocfs2_write_cluster_desc *desc 1412 struct ocfs2_write_cluster_desc *desc;
1386 unsigned int num_clusters = 0; 1413 unsigned int num_clusters = 0;
1387 unsigned int ext_flags = 0; 1414 unsigned int ext_flags = 0;
1388 u32 phys = 0; 1415 u32 phys = 0;
1389 int i; 1416 int i;
1390 1417
1391 *clusters_to_alloc = 0; 1418 *clusters_to_alloc = 0;
1392 *extents_to_split = 0; 1419 *extents_to_split = 0;
1393 1420
1394 for (i = 0; i < wc->w_clen; i++) { 1421 for (i = 0; i < wc->w_clen; i++) {
1395 desc = &wc->w_desc[i]; 1422 desc = &wc->w_desc[i];
1396 desc->c_cpos = wc->w_cpos + i 1423 desc->c_cpos = wc->w_cpos + i;
1397 1424
1398 if (num_clusters == 0) { 1425 if (num_clusters == 0) {
1399 /* 1426 /*
1400 * Need to look up th 1427 * Need to look up the next extent record.
1401 */ 1428 */
1402 ret = ocfs2_get_clust 1429 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1403 1430 &num_clusters, &ext_flags);
1404 if (ret) { 1431 if (ret) {
1405 mlog_errno(re 1432 mlog_errno(ret);
1406 goto out; 1433 goto out;
1407 } 1434 }
1408 1435
1409 /* We should already 1436 /* We should already CoW the refcountd extent. */
1410 BUG_ON(ext_flags & OC 1437 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1411 1438
1412 /* 1439 /*
1413 * Assume worst case 1440 * Assume worst case - that we're writing in
1414 * the middle of the 1441 * the middle of the extent.
1415 * 1442 *
1416 * We can assume that 1443 * We can assume that the write proceeds from
1417 * left to right, in 1444 * left to right, in which case the extent
1418 * insert code is sma 1445 * insert code is smart enough to coalesce the
1419 * next splits into t 1446 * next splits into the previous records created.
1420 */ 1447 */
1421 if (ext_flags & OCFS2 1448 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1422 *extents_to_s 1449 *extents_to_split = *extents_to_split + 2;
1423 } else if (phys) { 1450 } else if (phys) {
1424 /* 1451 /*
1425 * Only increment phy 1452 * Only increment phys if it doesn't describe
1426 * a hole. 1453 * a hole.
1427 */ 1454 */
1428 phys++; 1455 phys++;
1429 } 1456 }
1430 1457
1431 /* 1458 /*
1432 * If w_first_new_cpos is < U 1459 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1433 * file that got extended. w 1460 * file that got extended. w_first_new_cpos tells us
1434 * where the newly allocated 1461 * where the newly allocated clusters are so we can
1435 * zero them. 1462 * zero them.
1436 */ 1463 */
1437 if (desc->c_cpos >= wc->w_fir 1464 if (desc->c_cpos >= wc->w_first_new_cpos) {
1438 BUG_ON(phys == 0); 1465 BUG_ON(phys == 0);
1439 desc->c_needs_zero = 1466 desc->c_needs_zero = 1;
1440 } 1467 }
1441 1468
1442 desc->c_phys = phys; 1469 desc->c_phys = phys;
1443 if (phys == 0) { 1470 if (phys == 0) {
1444 desc->c_new = 1; 1471 desc->c_new = 1;
1445 desc->c_needs_zero = 1472 desc->c_needs_zero = 1;
1446 desc->c_clear_unwritt 1473 desc->c_clear_unwritten = 1;
1447 *clusters_to_alloc = 1474 *clusters_to_alloc = *clusters_to_alloc + 1;
1448 } 1475 }
1449 1476
1450 if (ext_flags & OCFS2_EXT_UNW 1477 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1451 desc->c_clear_unwritt 1478 desc->c_clear_unwritten = 1;
1452 desc->c_needs_zero = 1479 desc->c_needs_zero = 1;
1453 } 1480 }
1454 1481
1455 ret = ocfs2_unwritten_check(i 1482 ret = ocfs2_unwritten_check(inode, wc, desc);
1456 if (ret) { 1483 if (ret) {
1457 mlog_errno(ret); 1484 mlog_errno(ret);
1458 goto out; 1485 goto out;
1459 } 1486 }
1460 1487
1461 num_clusters--; 1488 num_clusters--;
1462 } 1489 }
1463 1490
1464 ret = 0; 1491 ret = 0;
1465 out: 1492 out:
1466 return ret; 1493 return ret;
1467 } 1494 }
1468 1495
1469 static int ocfs2_write_begin_inline(struct ad 1496 static int ocfs2_write_begin_inline(struct address_space *mapping,
1470 struct in 1497 struct inode *inode,
1471 struct oc 1498 struct ocfs2_write_ctxt *wc)
1472 { 1499 {
1473 int ret; 1500 int ret;
1474 struct ocfs2_super *osb = OCFS2_SB(in 1501 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1475 struct page *page; 1502 struct page *page;
1476 handle_t *handle; 1503 handle_t *handle;
1477 struct ocfs2_dinode *di = (struct ocf 1504 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1478 1505
1479 handle = ocfs2_start_trans(osb, OCFS2 1506 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1480 if (IS_ERR(handle)) { 1507 if (IS_ERR(handle)) {
1481 ret = PTR_ERR(handle); 1508 ret = PTR_ERR(handle);
1482 mlog_errno(ret); 1509 mlog_errno(ret);
1483 goto out; 1510 goto out;
1484 } 1511 }
1485 1512
1486 page = find_or_create_page(mapping, 0 1513 page = find_or_create_page(mapping, 0, GFP_NOFS);
1487 if (!page) { 1514 if (!page) {
1488 ocfs2_commit_trans(osb, handl 1515 ocfs2_commit_trans(osb, handle);
1489 ret = -ENOMEM; 1516 ret = -ENOMEM;
1490 mlog_errno(ret); 1517 mlog_errno(ret);
1491 goto out; 1518 goto out;
1492 } 1519 }
1493 /* 1520 /*
1494 * If we don't set w_num_pages then t 1521 * If we don't set w_num_pages then this page won't get unlocked
1495 * and freed on cleanup of the write 1522 * and freed on cleanup of the write context.
1496 */ 1523 */
1497 wc->w_pages[0] = wc->w_target_page = 1524 wc->w_pages[0] = wc->w_target_page = page;
1498 wc->w_num_pages = 1; 1525 wc->w_num_pages = 1;
1499 1526
1500 ret = ocfs2_journal_access_di(handle, 1527 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1501 OCFS2_J 1528 OCFS2_JOURNAL_ACCESS_WRITE);
1502 if (ret) { 1529 if (ret) {
1503 ocfs2_commit_trans(osb, handl 1530 ocfs2_commit_trans(osb, handle);
1504 1531
1505 mlog_errno(ret); 1532 mlog_errno(ret);
1506 goto out; 1533 goto out;
1507 } 1534 }
1508 1535
1509 if (!(OCFS2_I(inode)->ip_dyn_features 1536 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1510 ocfs2_set_inode_data_inline(i 1537 ocfs2_set_inode_data_inline(inode, di);
1511 1538
1512 if (!PageUptodate(page)) { 1539 if (!PageUptodate(page)) {
1513 ret = ocfs2_read_inline_data( 1540 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1514 if (ret) { 1541 if (ret) {
1515 ocfs2_commit_trans(os 1542 ocfs2_commit_trans(osb, handle);
1516 1543
1517 goto out; 1544 goto out;
1518 } 1545 }
1519 } 1546 }
1520 1547
1521 wc->w_handle = handle; 1548 wc->w_handle = handle;
1522 out: 1549 out:
1523 return ret; 1550 return ret;
1524 } 1551 }
1525 1552
1526 int ocfs2_size_fits_inline_data(struct buffer 1553 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1527 { 1554 {
1528 struct ocfs2_dinode *di = (struct ocf 1555 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1529 1556
1530 if (new_size <= le16_to_cpu(di->id2.i 1557 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1531 return 1; 1558 return 1;
1532 return 0; 1559 return 0;
1533 } 1560 }
1534 1561
1535 static int ocfs2_try_to_write_inline_data(str 1562 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1536 str 1563 struct inode *inode, loff_t pos,
1537 uns 1564 unsigned len, struct page *mmap_page,
1538 str 1565 struct ocfs2_write_ctxt *wc)
1539 { 1566 {
1540 int ret, written = 0; 1567 int ret, written = 0;
1541 loff_t end = pos + len; 1568 loff_t end = pos + len;
1542 struct ocfs2_inode_info *oi = OCFS2_I 1569 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1543 struct ocfs2_dinode *di = NULL; 1570 struct ocfs2_dinode *di = NULL;
1544 1571
1545 trace_ocfs2_try_to_write_inline_data( 1572 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1546 1573 len, (unsigned long long)pos,
1547 1574 oi->ip_dyn_features);
1548 1575
1549 /* 1576 /*
1550 * Handle inodes which already have i 1577 * Handle inodes which already have inline data 1st.
1551 */ 1578 */
1552 if (oi->ip_dyn_features & OCFS2_INLIN 1579 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1553 if (mmap_page == NULL && 1580 if (mmap_page == NULL &&
1554 ocfs2_size_fits_inline_da 1581 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1555 goto do_inline_write; 1582 goto do_inline_write;
1556 1583
1557 /* 1584 /*
1558 * The write won't fit - we h 1585 * The write won't fit - we have to give this inode an
1559 * inline extent list now. 1586 * inline extent list now.
1560 */ 1587 */
1561 ret = ocfs2_convert_inline_da 1588 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1562 if (ret) 1589 if (ret)
1563 mlog_errno(ret); 1590 mlog_errno(ret);
1564 goto out; 1591 goto out;
1565 } 1592 }
1566 1593
1567 /* 1594 /*
1568 * Check whether the inode can accept 1595 * Check whether the inode can accept inline data.
1569 */ 1596 */
1570 if (oi->ip_clusters != 0 || i_size_re 1597 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1571 return 0; 1598 return 0;
1572 1599
1573 /* 1600 /*
1574 * Check whether the write can fit. 1601 * Check whether the write can fit.
1575 */ 1602 */
1576 di = (struct ocfs2_dinode *)wc->w_di_ 1603 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1577 if (mmap_page || 1604 if (mmap_page ||
1578 end > ocfs2_max_inline_data_with_ 1605 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1579 return 0; 1606 return 0;
1580 1607
1581 do_inline_write: 1608 do_inline_write:
1582 ret = ocfs2_write_begin_inline(mappin 1609 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1583 if (ret) { 1610 if (ret) {
1584 mlog_errno(ret); 1611 mlog_errno(ret);
1585 goto out; 1612 goto out;
1586 } 1613 }
1587 1614
1588 /* 1615 /*
1589 * This signals to the caller that th 1616 * This signals to the caller that the data can be written
1590 * inline. 1617 * inline.
1591 */ 1618 */
1592 written = 1; 1619 written = 1;
1593 out: 1620 out:
1594 return written ? written : ret; 1621 return written ? written : ret;
1595 } 1622 }
1596 1623
1597 /* 1624 /*
1598 * This function only does anything for file 1625 * This function only does anything for file systems which can't
1599 * handle sparse files. 1626 * handle sparse files.
1600 * 1627 *
1601 * What we want to do here is fill in any hol 1628 * What we want to do here is fill in any hole between the current end
1602 * of allocation and the end of our write. Th 1629 * of allocation and the end of our write. That way the rest of the
1603 * write path can treat it as an non-allocati 1630 * write path can treat it as an non-allocating write, which has no
1604 * special case code for sparse/nonsparse fil 1631 * special case code for sparse/nonsparse files.
1605 */ 1632 */
1606 static int ocfs2_expand_nonsparse_inode(struc 1633 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1607 struc 1634 struct buffer_head *di_bh,
1608 loff_ 1635 loff_t pos, unsigned len,
1609 struc 1636 struct ocfs2_write_ctxt *wc)
1610 { 1637 {
1611 int ret; 1638 int ret;
1612 loff_t newsize = pos + len; 1639 loff_t newsize = pos + len;
1613 1640
1614 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1641 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1615 1642
1616 if (newsize <= i_size_read(inode)) 1643 if (newsize <= i_size_read(inode))
1617 return 0; 1644 return 0;
1618 1645
1619 ret = ocfs2_extend_no_holes(inode, di 1646 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1620 if (ret) 1647 if (ret)
1621 mlog_errno(ret); 1648 mlog_errno(ret);
1622 1649
1623 /* There is no wc if this is call fro 1650 /* There is no wc if this is call from direct. */
1624 if (wc) 1651 if (wc)
1625 wc->w_first_new_cpos = 1652 wc->w_first_new_cpos =
1626 ocfs2_clusters_for_by 1653 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1627 1654
1628 return ret; 1655 return ret;
1629 } 1656 }
1630 1657
1631 static int ocfs2_zero_tail(struct inode *inod 1658 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1632 loff_t pos) 1659 loff_t pos)
1633 { 1660 {
1634 int ret = 0; 1661 int ret = 0;
1635 1662
1636 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1663 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1637 if (pos > i_size_read(inode)) 1664 if (pos > i_size_read(inode))
1638 ret = ocfs2_zero_extend(inode 1665 ret = ocfs2_zero_extend(inode, di_bh, pos);
1639 1666
1640 return ret; 1667 return ret;
1641 } 1668 }
1642 1669
1643 int ocfs2_write_begin_nolock(struct address_s 1670 int ocfs2_write_begin_nolock(struct address_space *mapping,
1644 loff_t pos, unsi 1671 loff_t pos, unsigned len, ocfs2_write_type_t type,
1645 struct page **pa 1672 struct page **pagep, void **fsdata,
1646 struct buffer_he 1673 struct buffer_head *di_bh, struct page *mmap_page)
1647 { 1674 {
1648 int ret, cluster_of_pages, credits = 1675 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1649 unsigned int clusters_to_alloc, exten 1676 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1650 struct ocfs2_write_ctxt *wc; 1677 struct ocfs2_write_ctxt *wc;
1651 struct inode *inode = mapping->host; 1678 struct inode *inode = mapping->host;
1652 struct ocfs2_super *osb = OCFS2_SB(in 1679 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1653 struct ocfs2_dinode *di; 1680 struct ocfs2_dinode *di;
1654 struct ocfs2_alloc_context *data_ac = 1681 struct ocfs2_alloc_context *data_ac = NULL;
1655 struct ocfs2_alloc_context *meta_ac = 1682 struct ocfs2_alloc_context *meta_ac = NULL;
1656 handle_t *handle; 1683 handle_t *handle;
1657 struct ocfs2_extent_tree et; 1684 struct ocfs2_extent_tree et;
1658 int try_free = 1, ret1; 1685 int try_free = 1, ret1;
1659 1686
1660 try_again: 1687 try_again:
1661 ret = ocfs2_alloc_write_ctxt(&wc, osb 1688 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
1662 if (ret) { 1689 if (ret) {
1663 mlog_errno(ret); 1690 mlog_errno(ret);
1664 return ret; 1691 return ret;
1665 } 1692 }
1666 1693
1667 if (ocfs2_supports_inline_data(osb)) 1694 if (ocfs2_supports_inline_data(osb)) {
1668 ret = ocfs2_try_to_write_inli 1695 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1669 1696 mmap_page, wc);
1670 if (ret == 1) { 1697 if (ret == 1) {
1671 ret = 0; 1698 ret = 0;
1672 goto success; 1699 goto success;
1673 } 1700 }
1674 if (ret < 0) { 1701 if (ret < 0) {
1675 mlog_errno(ret); 1702 mlog_errno(ret);
1676 goto out; 1703 goto out;
1677 } 1704 }
1678 } 1705 }
1679 1706
1680 /* Direct io change i_size late, shou 1707 /* Direct io change i_size late, should not zero tail here. */
1681 if (type != OCFS2_WRITE_DIRECT) { 1708 if (type != OCFS2_WRITE_DIRECT) {
1682 if (ocfs2_sparse_alloc(osb)) 1709 if (ocfs2_sparse_alloc(osb))
1683 ret = ocfs2_zero_tail 1710 ret = ocfs2_zero_tail(inode, di_bh, pos);
1684 else 1711 else
1685 ret = ocfs2_expand_no 1712 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1686 1713 len, wc);
1687 if (ret) { 1714 if (ret) {
1688 mlog_errno(ret); 1715 mlog_errno(ret);
1689 goto out; 1716 goto out;
1690 } 1717 }
1691 } 1718 }
1692 1719
1693 ret = ocfs2_check_range_for_refcount( 1720 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1694 if (ret < 0) { 1721 if (ret < 0) {
1695 mlog_errno(ret); 1722 mlog_errno(ret);
1696 goto out; 1723 goto out;
1697 } else if (ret == 1) { 1724 } else if (ret == 1) {
1698 clusters_need = wc->w_clen; 1725 clusters_need = wc->w_clen;
1699 ret = ocfs2_refcount_cow(inod 1726 ret = ocfs2_refcount_cow(inode, di_bh,
1700 wc-> 1727 wc->w_cpos, wc->w_clen, UINT_MAX);
1701 if (ret) { 1728 if (ret) {
1702 mlog_errno(ret); 1729 mlog_errno(ret);
1703 goto out; 1730 goto out;
1704 } 1731 }
1705 } 1732 }
1706 1733
1707 ret = ocfs2_populate_write_desc(inode 1734 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1708 &exte 1735 &extents_to_split);
1709 if (ret) { 1736 if (ret) {
1710 mlog_errno(ret); 1737 mlog_errno(ret);
1711 goto out; 1738 goto out;
1712 } 1739 }
1713 clusters_need += clusters_to_alloc; 1740 clusters_need += clusters_to_alloc;
1714 1741
1715 di = (struct ocfs2_dinode *)wc->w_di_ 1742 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1716 1743
1717 trace_ocfs2_write_begin_nolock( 1744 trace_ocfs2_write_begin_nolock(
1718 (unsigned long long)O 1745 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1719 (long long)i_size_rea 1746 (long long)i_size_read(inode),
1720 le32_to_cpu(di->i_clu 1747 le32_to_cpu(di->i_clusters),
1721 pos, len, type, mmap_ 1748 pos, len, type, mmap_page,
1722 clusters_to_alloc, ex 1749 clusters_to_alloc, extents_to_split);
1723 1750
1724 /* 1751 /*
1725 * We set w_target_from, w_target_to 1752 * We set w_target_from, w_target_to here so that
1726 * ocfs2_write_end() knows which rang 1753 * ocfs2_write_end() knows which range in the target page to
1727 * write out. An allocation requires 1754 * write out. An allocation requires that we write the entire
1728 * cluster range. 1755 * cluster range.
1729 */ 1756 */
1730 if (clusters_to_alloc || extents_to_s 1757 if (clusters_to_alloc || extents_to_split) {
1731 /* 1758 /*
1732 * XXX: We are stretching the 1759 * XXX: We are stretching the limits of
1733 * ocfs2_lock_allocators(). I 1760 * ocfs2_lock_allocators(). It greatly over-estimates
1734 * the work to be done. 1761 * the work to be done.
1735 */ 1762 */
1736 ocfs2_init_dinode_extent_tree 1763 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1737 1764 wc->w_di_bh);
1738 ret = ocfs2_lock_allocators(i 1765 ret = ocfs2_lock_allocators(inode, &et,
1739 c 1766 clusters_to_alloc, extents_to_split,
1740 & 1767 &data_ac, &meta_ac);
1741 if (ret) { 1768 if (ret) {
1742 mlog_errno(ret); 1769 mlog_errno(ret);
1743 goto out; 1770 goto out;
1744 } 1771 }
1745 1772
1746 if (data_ac) 1773 if (data_ac)
1747 data_ac->ac_resv = &O 1774 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1748 1775
1749 credits = ocfs2_calc_extend_c 1776 credits = ocfs2_calc_extend_credits(inode->i_sb,
1750 1777 &di->id2.i_list);
1751 } else if (type == OCFS2_WRITE_DIRECT 1778 } else if (type == OCFS2_WRITE_DIRECT)
1752 /* direct write needs not to 1779 /* direct write needs not to start trans if no extents alloc. */
1753 goto success; 1780 goto success;
1754 1781
1755 /* 1782 /*
1756 * We have to zero sparse allocated c 1783 * We have to zero sparse allocated clusters, unwritten extent clusters,
1757 * and non-sparse clusters we just ex 1784 * and non-sparse clusters we just extended. For non-sparse writes,
1758 * we know zeros will only be needed 1785 * we know zeros will only be needed in the first and/or last cluster.
1759 */ 1786 */
1760 if (wc->w_clen && (wc->w_desc[0].c_ne 1787 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1761 wc->w_desc[wc->w_c 1788 wc->w_desc[wc->w_clen - 1].c_needs_zero))
1762 cluster_of_pages = 1; 1789 cluster_of_pages = 1;
1763 else 1790 else
1764 cluster_of_pages = 0; 1791 cluster_of_pages = 0;
1765 1792
1766 ocfs2_set_target_boundaries(osb, wc, 1793 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1767 1794
1768 handle = ocfs2_start_trans(osb, credi 1795 handle = ocfs2_start_trans(osb, credits);
1769 if (IS_ERR(handle)) { 1796 if (IS_ERR(handle)) {
1770 ret = PTR_ERR(handle); 1797 ret = PTR_ERR(handle);
1771 mlog_errno(ret); 1798 mlog_errno(ret);
1772 goto out; 1799 goto out;
1773 } 1800 }
1774 1801
1775 wc->w_handle = handle; 1802 wc->w_handle = handle;
1776 1803
1777 if (clusters_to_alloc) { 1804 if (clusters_to_alloc) {
1778 ret = dquot_alloc_space_nodir 1805 ret = dquot_alloc_space_nodirty(inode,
1779 ocfs2_clusters_to_byt 1806 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1780 if (ret) 1807 if (ret)
1781 goto out_commit; 1808 goto out_commit;
1782 } 1809 }
1783 1810
1784 ret = ocfs2_journal_access_di(handle, 1811 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1785 OCFS2_J 1812 OCFS2_JOURNAL_ACCESS_WRITE);
1786 if (ret) { 1813 if (ret) {
1787 mlog_errno(ret); 1814 mlog_errno(ret);
1788 goto out_quota; 1815 goto out_quota;
1789 } 1816 }
1790 1817
1791 /* 1818 /*
1792 * Fill our page array first. That wa 1819 * Fill our page array first. That way we've grabbed enough so
1793 * that we can zero and flush if we e 1820 * that we can zero and flush if we error after adding the
1794 * extent. 1821 * extent.
1795 */ 1822 */
1796 ret = ocfs2_grab_pages_for_write(mapp 1823 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1797 clus 1824 cluster_of_pages, mmap_page);
1798 if (ret) { !! 1825 if (ret && ret != -EAGAIN) {
1799 /* <<
1800 * ocfs2_grab_pages_for_write <<
1801 * the target page. In this c <<
1802 * page. This will trigger th <<
1803 * the operation. <<
1804 */ <<
1805 if (type == OCFS2_WRITE_MMAP <<
1806 BUG_ON(wc->w_target_p <<
1807 ret = 0; <<
1808 goto out_quota; <<
1809 } <<
1810 <<
1811 mlog_errno(ret); 1826 mlog_errno(ret);
1812 goto out_quota; 1827 goto out_quota;
1813 } 1828 }
1814 1829
>> 1830 /*
>> 1831 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
>> 1832 * the target page. In this case, we exit with no error and no target
>> 1833 * page. This will trigger the caller, page_mkwrite(), to re-try
>> 1834 * the operation.
>> 1835 */
>> 1836 if (ret == -EAGAIN) {
>> 1837 BUG_ON(wc->w_target_page);
>> 1838 ret = 0;
>> 1839 goto out_quota;
>> 1840 }
>> 1841
1815 ret = ocfs2_write_cluster_by_desc(map 1842 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1816 len 1843 len);
1817 if (ret) { 1844 if (ret) {
1818 mlog_errno(ret); 1845 mlog_errno(ret);
1819 goto out_quota; 1846 goto out_quota;
1820 } 1847 }
1821 1848
1822 if (data_ac) 1849 if (data_ac)
1823 ocfs2_free_alloc_context(data 1850 ocfs2_free_alloc_context(data_ac);
1824 if (meta_ac) 1851 if (meta_ac)
1825 ocfs2_free_alloc_context(meta 1852 ocfs2_free_alloc_context(meta_ac);
1826 1853
1827 success: 1854 success:
1828 if (pagep) 1855 if (pagep)
1829 *pagep = wc->w_target_page; 1856 *pagep = wc->w_target_page;
1830 *fsdata = wc; 1857 *fsdata = wc;
1831 return 0; 1858 return 0;
1832 out_quota: 1859 out_quota:
1833 if (clusters_to_alloc) 1860 if (clusters_to_alloc)
1834 dquot_free_space(inode, 1861 dquot_free_space(inode,
1835 ocfs2_clusters_to_b 1862 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1836 out_commit: 1863 out_commit:
1837 ocfs2_commit_trans(osb, handle); 1864 ocfs2_commit_trans(osb, handle);
1838 1865
1839 out: 1866 out:
1840 /* 1867 /*
1841 * The mmapped page won't be unlocked 1868 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1842 * even in case of error here like EN 1869 * even in case of error here like ENOSPC and ENOMEM. So, we need
1843 * to unlock the target page manually 1870 * to unlock the target page manually to prevent deadlocks when
1844 * retrying again on ENOSPC, or when 1871 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1845 * to VM code. 1872 * to VM code.
1846 */ 1873 */
1847 if (wc->w_target_locked) 1874 if (wc->w_target_locked)
1848 unlock_page(mmap_page); 1875 unlock_page(mmap_page);
1849 1876
1850 ocfs2_free_write_ctxt(inode, wc); 1877 ocfs2_free_write_ctxt(inode, wc);
1851 1878
1852 if (data_ac) { 1879 if (data_ac) {
1853 ocfs2_free_alloc_context(data 1880 ocfs2_free_alloc_context(data_ac);
1854 data_ac = NULL; 1881 data_ac = NULL;
1855 } 1882 }
1856 if (meta_ac) { 1883 if (meta_ac) {
1857 ocfs2_free_alloc_context(meta 1884 ocfs2_free_alloc_context(meta_ac);
1858 meta_ac = NULL; 1885 meta_ac = NULL;
1859 } 1886 }
1860 1887
1861 if (ret == -ENOSPC && try_free) { 1888 if (ret == -ENOSPC && try_free) {
1862 /* 1889 /*
1863 * Try to free some truncate 1890 * Try to free some truncate log so that we can have enough
1864 * clusters to allocate. 1891 * clusters to allocate.
1865 */ 1892 */
1866 try_free = 0; 1893 try_free = 0;
1867 1894
1868 ret1 = ocfs2_try_to_free_trun 1895 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1869 if (ret1 == 1) 1896 if (ret1 == 1)
1870 goto try_again; 1897 goto try_again;
1871 1898
1872 if (ret1 < 0) 1899 if (ret1 < 0)
1873 mlog_errno(ret1); 1900 mlog_errno(ret1);
1874 } 1901 }
1875 1902
1876 return ret; 1903 return ret;
1877 } 1904 }
1878 1905
1879 static int ocfs2_write_begin(struct file *fil 1906 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1880 loff_t pos, unsi !! 1907 loff_t pos, unsigned len, unsigned flags,
1881 struct page **pa 1908 struct page **pagep, void **fsdata)
1882 { 1909 {
1883 int ret; 1910 int ret;
1884 struct buffer_head *di_bh = NULL; 1911 struct buffer_head *di_bh = NULL;
1885 struct inode *inode = mapping->host; 1912 struct inode *inode = mapping->host;
1886 1913
1887 ret = ocfs2_inode_lock(inode, &di_bh, 1914 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1888 if (ret) { 1915 if (ret) {
1889 mlog_errno(ret); 1916 mlog_errno(ret);
1890 return ret; 1917 return ret;
1891 } 1918 }
1892 1919
1893 /* 1920 /*
1894 * Take alloc sem here to prevent con 1921 * Take alloc sem here to prevent concurrent lookups. That way
1895 * the mapping, zeroing and tree mani 1922 * the mapping, zeroing and tree manipulation within
1896 * ocfs2_write() will be safe against !! 1923 * ocfs2_write() will be safe against ->readpage(). This
1897 * should also serve to lock out allo 1924 * should also serve to lock out allocation from a shared
1898 * writeable region. 1925 * writeable region.
1899 */ 1926 */
1900 down_write(&OCFS2_I(inode)->ip_alloc_ 1927 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1901 1928
1902 ret = ocfs2_write_begin_nolock(mappin 1929 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1903 pagep, 1930 pagep, fsdata, di_bh, NULL);
1904 if (ret) { 1931 if (ret) {
1905 mlog_errno(ret); 1932 mlog_errno(ret);
1906 goto out_fail; 1933 goto out_fail;
1907 } 1934 }
1908 1935
1909 brelse(di_bh); 1936 brelse(di_bh);
1910 1937
1911 return 0; 1938 return 0;
1912 1939
1913 out_fail: 1940 out_fail:
1914 up_write(&OCFS2_I(inode)->ip_alloc_se 1941 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1915 1942
1916 brelse(di_bh); 1943 brelse(di_bh);
1917 ocfs2_inode_unlock(inode, 1); 1944 ocfs2_inode_unlock(inode, 1);
1918 1945
1919 return ret; 1946 return ret;
1920 } 1947 }
1921 1948
1922 static void ocfs2_write_end_inline(struct ino 1949 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1923 unsigned l 1950 unsigned len, unsigned *copied,
1924 struct ocf 1951 struct ocfs2_dinode *di,
1925 struct ocf 1952 struct ocfs2_write_ctxt *wc)
1926 { 1953 {
1927 void *kaddr; 1954 void *kaddr;
1928 1955
1929 if (unlikely(*copied < len)) { 1956 if (unlikely(*copied < len)) {
1930 if (!PageUptodate(wc->w_targe 1957 if (!PageUptodate(wc->w_target_page)) {
1931 *copied = 0; 1958 *copied = 0;
1932 return; 1959 return;
1933 } 1960 }
1934 } 1961 }
1935 1962
1936 kaddr = kmap_atomic(wc->w_target_page 1963 kaddr = kmap_atomic(wc->w_target_page);
1937 memcpy(di->id2.i_data.id_data + pos, 1964 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1938 kunmap_atomic(kaddr); 1965 kunmap_atomic(kaddr);
1939 1966
1940 trace_ocfs2_write_end_inline( 1967 trace_ocfs2_write_end_inline(
1941 (unsigned long long)OCFS2_I(inod 1968 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1942 (unsigned long long)pos, *copied 1969 (unsigned long long)pos, *copied,
1943 le16_to_cpu(di->id2.i_data.id_co 1970 le16_to_cpu(di->id2.i_data.id_count),
1944 le16_to_cpu(di->i_dyn_features)) 1971 le16_to_cpu(di->i_dyn_features));
1945 } 1972 }
1946 1973
1947 int ocfs2_write_end_nolock(struct address_spa 1974 int ocfs2_write_end_nolock(struct address_space *mapping,
1948 loff_t pos, unsign 1975 loff_t pos, unsigned len, unsigned copied, void *fsdata)
1949 { 1976 {
1950 int i, ret; 1977 int i, ret;
1951 unsigned from, to, start = pos & (PAG 1978 unsigned from, to, start = pos & (PAGE_SIZE - 1);
1952 struct inode *inode = mapping->host; 1979 struct inode *inode = mapping->host;
1953 struct ocfs2_super *osb = OCFS2_SB(in 1980 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1954 struct ocfs2_write_ctxt *wc = fsdata; 1981 struct ocfs2_write_ctxt *wc = fsdata;
1955 struct ocfs2_dinode *di = (struct ocf 1982 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1956 handle_t *handle = wc->w_handle; 1983 handle_t *handle = wc->w_handle;
1957 struct page *tmppage; 1984 struct page *tmppage;
1958 1985
1959 BUG_ON(!list_empty(&wc->w_unwritten_l 1986 BUG_ON(!list_empty(&wc->w_unwritten_list));
1960 1987
1961 if (handle) { 1988 if (handle) {
1962 ret = ocfs2_journal_access_di 1989 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1963 wc->w_di_bh, 1990 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1964 if (ret) { 1991 if (ret) {
1965 copied = ret; 1992 copied = ret;
1966 mlog_errno(ret); 1993 mlog_errno(ret);
1967 goto out; 1994 goto out;
1968 } 1995 }
1969 } 1996 }
1970 1997
1971 if (OCFS2_I(inode)->ip_dyn_features & 1998 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1972 ocfs2_write_end_inline(inode, 1999 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1973 goto out_write_size; 2000 goto out_write_size;
1974 } 2001 }
1975 2002
1976 if (unlikely(copied < len) && wc->w_t 2003 if (unlikely(copied < len) && wc->w_target_page) {
1977 loff_t new_isize; <<
1978 <<
1979 if (!PageUptodate(wc->w_targe 2004 if (!PageUptodate(wc->w_target_page))
1980 copied = 0; 2005 copied = 0;
1981 2006
1982 new_isize = max_t(loff_t, i_s !! 2007 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1983 if (new_isize > page_offset(w !! 2008 start+len);
1984 ocfs2_zero_new_buffer <<
1985 <<
1986 else { <<
1987 /* <<
1988 * When page is fully <<
1989 * failed), do not bo <<
1990 * it instead so that <<
1991 * put page & buffer <<
1992 * state. <<
1993 */ <<
1994 block_invalidate_foli <<
1995 <<
1996 } <<
1997 } 2009 }
1998 if (wc->w_target_page) 2010 if (wc->w_target_page)
1999 flush_dcache_page(wc->w_targe 2011 flush_dcache_page(wc->w_target_page);
2000 2012
2001 for(i = 0; i < wc->w_num_pages; i++) 2013 for(i = 0; i < wc->w_num_pages; i++) {
2002 tmppage = wc->w_pages[i]; 2014 tmppage = wc->w_pages[i];
2003 2015
2004 /* This is the direct io targ 2016 /* This is the direct io target page. */
2005 if (tmppage == NULL) 2017 if (tmppage == NULL)
2006 continue; 2018 continue;
2007 2019
2008 if (tmppage == wc->w_target_p 2020 if (tmppage == wc->w_target_page) {
2009 from = wc->w_target_f 2021 from = wc->w_target_from;
2010 to = wc->w_target_to; 2022 to = wc->w_target_to;
2011 2023
2012 BUG_ON(from > PAGE_SI 2024 BUG_ON(from > PAGE_SIZE ||
2013 to > PAGE_SIZE 2025 to > PAGE_SIZE ||
2014 to < from); 2026 to < from);
2015 } else { 2027 } else {
2016 /* 2028 /*
2017 * Pages adjacent to 2029 * Pages adjacent to the target (if any) imply
2018 * a hole-filling wri 2030 * a hole-filling write in which case we want
2019 * to flush their ent 2031 * to flush their entire range.
2020 */ 2032 */
2021 from = 0; 2033 from = 0;
2022 to = PAGE_SIZE; 2034 to = PAGE_SIZE;
2023 } 2035 }
2024 2036
2025 if (page_has_buffers(tmppage) 2037 if (page_has_buffers(tmppage)) {
2026 if (handle && ocfs2_s !! 2038 if (handle && ocfs2_should_order_data(inode))
2027 loff_t start_ !! 2039 ocfs2_jbd2_file_inode(handle, inode);
2028 ((lof <<
2029 from; <<
2030 loff_t length <<
2031 ocfs2_jbd2_in <<
2032 <<
2033 } <<
2034 block_commit_write(tm 2040 block_commit_write(tmppage, from, to);
2035 } 2041 }
2036 } 2042 }
2037 2043
2038 out_write_size: 2044 out_write_size:
2039 /* Direct io do not update i_size her 2045 /* Direct io do not update i_size here. */
2040 if (wc->w_type != OCFS2_WRITE_DIRECT) 2046 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2041 pos += copied; 2047 pos += copied;
2042 if (pos > i_size_read(inode)) 2048 if (pos > i_size_read(inode)) {
2043 i_size_write(inode, p 2049 i_size_write(inode, pos);
2044 mark_inode_dirty(inod 2050 mark_inode_dirty(inode);
2045 } 2051 }
2046 inode->i_blocks = ocfs2_inode 2052 inode->i_blocks = ocfs2_inode_sector_count(inode);
2047 di->i_size = cpu_to_le64((u64 2053 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2048 inode_set_mtime_to_ts(inode, !! 2054 inode->i_mtime = inode->i_ctime = current_time(inode);
2049 di->i_mtime = di->i_ctime = c !! 2055 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2050 di->i_mtime_nsec = di->i_ctim !! 2056 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2051 if (handle) !! 2057 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2052 ocfs2_update_inode_fs <<
2053 } 2058 }
2054 if (handle) 2059 if (handle)
2055 ocfs2_journal_dirty(handle, w 2060 ocfs2_journal_dirty(handle, wc->w_di_bh);
2056 2061
2057 out: 2062 out:
2058 /* unlock pages before dealloc since 2063 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2059 * lock, or it will cause a deadlock 2064 * lock, or it will cause a deadlock since journal commit threads holds
2060 * this lock and will ask for the pag 2065 * this lock and will ask for the page lock when flushing the data.
2061 * put it here to preserve the unlock 2066 * put it here to preserve the unlock order.
2062 */ 2067 */
2063 ocfs2_unlock_pages(wc); 2068 ocfs2_unlock_pages(wc);
2064 2069
2065 if (handle) 2070 if (handle)
2066 ocfs2_commit_trans(osb, handl 2071 ocfs2_commit_trans(osb, handle);
2067 2072
2068 ocfs2_run_deallocs(osb, &wc->w_deallo 2073 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2069 2074
2070 brelse(wc->w_di_bh); 2075 brelse(wc->w_di_bh);
2071 kfree(wc); 2076 kfree(wc);
2072 2077
2073 return copied; 2078 return copied;
2074 } 2079 }
2075 2080
2076 static int ocfs2_write_end(struct file *file, 2081 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2077 loff_t pos, unsign 2082 loff_t pos, unsigned len, unsigned copied,
2078 struct page *page, 2083 struct page *page, void *fsdata)
2079 { 2084 {
2080 int ret; 2085 int ret;
2081 struct inode *inode = mapping->host; 2086 struct inode *inode = mapping->host;
2082 2087
2083 ret = ocfs2_write_end_nolock(mapping, 2088 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata);
2084 2089
2085 up_write(&OCFS2_I(inode)->ip_alloc_se 2090 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2086 ocfs2_inode_unlock(inode, 1); 2091 ocfs2_inode_unlock(inode, 1);
2087 2092
2088 return ret; 2093 return ret;
2089 } 2094 }
2090 2095
2091 struct ocfs2_dio_write_ctxt { 2096 struct ocfs2_dio_write_ctxt {
2092 struct list_head dw_zero_list; 2097 struct list_head dw_zero_list;
2093 unsigned dw_zero_count 2098 unsigned dw_zero_count;
2094 int dw_orphaned; 2099 int dw_orphaned;
2095 pid_t dw_writer_pid 2100 pid_t dw_writer_pid;
2096 }; 2101 };
2097 2102
2098 static struct ocfs2_dio_write_ctxt * 2103 static struct ocfs2_dio_write_ctxt *
2099 ocfs2_dio_alloc_write_ctx(struct buffer_head 2104 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2100 { 2105 {
2101 struct ocfs2_dio_write_ctxt *dwc = NU 2106 struct ocfs2_dio_write_ctxt *dwc = NULL;
2102 2107
2103 if (bh->b_private) 2108 if (bh->b_private)
2104 return bh->b_private; 2109 return bh->b_private;
2105 2110
2106 dwc = kmalloc(sizeof(struct ocfs2_dio 2111 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2107 if (dwc == NULL) 2112 if (dwc == NULL)
2108 return NULL; 2113 return NULL;
2109 INIT_LIST_HEAD(&dwc->dw_zero_list); 2114 INIT_LIST_HEAD(&dwc->dw_zero_list);
2110 dwc->dw_zero_count = 0; 2115 dwc->dw_zero_count = 0;
2111 dwc->dw_orphaned = 0; 2116 dwc->dw_orphaned = 0;
2112 dwc->dw_writer_pid = task_pid_nr(curr 2117 dwc->dw_writer_pid = task_pid_nr(current);
2113 bh->b_private = dwc; 2118 bh->b_private = dwc;
2114 *alloc = 1; 2119 *alloc = 1;
2115 2120
2116 return dwc; 2121 return dwc;
2117 } 2122 }
2118 2123
2119 static void ocfs2_dio_free_write_ctx(struct i 2124 static void ocfs2_dio_free_write_ctx(struct inode *inode,
2120 struct o 2125 struct ocfs2_dio_write_ctxt *dwc)
2121 { 2126 {
2122 ocfs2_free_unwritten_list(inode, &dwc 2127 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2123 kfree(dwc); 2128 kfree(dwc);
2124 } 2129 }
2125 2130
2126 /* 2131 /*
2127 * TODO: Make this into a generic get_blocks 2132 * TODO: Make this into a generic get_blocks function.
2128 * 2133 *
2129 * From do_direct_io in direct-io.c: 2134 * From do_direct_io in direct-io.c:
2130 * "So what we do is to permit the ->get_blo 2135 * "So what we do is to permit the ->get_blocks function to populate
2131 * bh.b_size with the size of IO which is p 2136 * bh.b_size with the size of IO which is permitted at this offset and
2132 * this i_blkbits." 2137 * this i_blkbits."
2133 * 2138 *
2134 * This function is called directly from get_ 2139 * This function is called directly from get_more_blocks in direct-io.c.
2135 * 2140 *
2136 * called like this: dio->get_blocks(dio->ino 2141 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2137 * fs_co 2142 * fs_count, map_bh, dio->rw == WRITE);
2138 */ 2143 */
2139 static int ocfs2_dio_wr_get_block(struct inod 2144 static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock,
2140 struct buffer_ 2145 struct buffer_head *bh_result, int create)
2141 { 2146 {
2142 struct ocfs2_super *osb = OCFS2_SB(in 2147 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2143 struct ocfs2_inode_info *oi = OCFS2_I 2148 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2144 struct ocfs2_write_ctxt *wc; 2149 struct ocfs2_write_ctxt *wc;
2145 struct ocfs2_write_cluster_desc *desc 2150 struct ocfs2_write_cluster_desc *desc = NULL;
2146 struct ocfs2_dio_write_ctxt *dwc = NU 2151 struct ocfs2_dio_write_ctxt *dwc = NULL;
2147 struct buffer_head *di_bh = NULL; 2152 struct buffer_head *di_bh = NULL;
2148 u64 p_blkno; 2153 u64 p_blkno;
2149 unsigned int i_blkbits = inode->i_sb- !! 2154 loff_t pos = iblock << inode->i_sb->s_blocksize_bits;
2150 loff_t pos = iblock << i_blkbits; <<
2151 sector_t endblk = (i_size_read(inode) <<
2152 unsigned len, total_len = bh_result-> 2155 unsigned len, total_len = bh_result->b_size;
2153 int ret = 0, first_get_block = 0; 2156 int ret = 0, first_get_block = 0;
2154 2157
2155 len = osb->s_clustersize - (pos & (os 2158 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2156 len = min(total_len, len); 2159 len = min(total_len, len);
2157 2160
2158 /* <<
2159 * bh_result->b_size is count in get_ <<
2160 * "pos" and "end", we need map twice <<
2161 * 1. area in file size, not set NEW; <<
2162 * 2. area out file size, set NEW. <<
2163 * <<
2164 * iblock endblk <<
2165 * |--------|---------|---------|---- <<
2166 * |<-------area in file------->| <<
2167 */ <<
2168 <<
2169 if ((iblock <= endblk) && <<
2170 ((iblock + ((len - 1) >> i_blkbit <<
2171 len = (endblk - iblock + 1) < <<
2172 <<
2173 mlog(0, "get block of %lu at %llu:%u 2161 mlog(0, "get block of %lu at %llu:%u req %u\n",
2174 inode->i_ino, pos, le 2162 inode->i_ino, pos, len, total_len);
2175 2163
2176 /* 2164 /*
2177 * Because we need to change file siz 2165 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2178 * we may need to add it to orphan di 2166 * we may need to add it to orphan dir. So can not fall to fast path
2179 * while file size will be changed. 2167 * while file size will be changed.
2180 */ 2168 */
2181 if (pos + total_len <= i_size_read(in 2169 if (pos + total_len <= i_size_read(inode)) {
2182 2170
2183 /* This is the fast path for 2171 /* This is the fast path for re-write. */
2184 ret = ocfs2_lock_get_block(in 2172 ret = ocfs2_lock_get_block(inode, iblock, bh_result, create);
2185 if (buffer_mapped(bh_result) 2173 if (buffer_mapped(bh_result) &&
2186 !buffer_new(bh_result) && 2174 !buffer_new(bh_result) &&
2187 ret == 0) 2175 ret == 0)
2188 goto out; 2176 goto out;
2189 2177
2190 /* Clear state set by ocfs2_g 2178 /* Clear state set by ocfs2_get_block. */
2191 bh_result->b_state = 0; 2179 bh_result->b_state = 0;
2192 } 2180 }
2193 2181
2194 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2182 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2195 if (unlikely(dwc == NULL)) { 2183 if (unlikely(dwc == NULL)) {
2196 ret = -ENOMEM; 2184 ret = -ENOMEM;
2197 mlog_errno(ret); 2185 mlog_errno(ret);
2198 goto out; 2186 goto out;
2199 } 2187 }
2200 2188
2201 if (ocfs2_clusters_for_bytes(inode->i 2189 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2202 ocfs2_clusters_for_bytes(inode->i 2190 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2203 !dwc->dw_orphaned) { 2191 !dwc->dw_orphaned) {
2204 /* 2192 /*
2205 * when we are going to alloc 2193 * when we are going to alloc extents beyond file size, add the
2206 * inode to orphan dir, so we 2194 * inode to orphan dir, so we can recall those spaces when
2207 * system crashed during writ 2195 * system crashed during write.
2208 */ 2196 */
2209 ret = ocfs2_add_inode_to_orph 2197 ret = ocfs2_add_inode_to_orphan(osb, inode);
2210 if (ret < 0) { 2198 if (ret < 0) {
2211 mlog_errno(ret); 2199 mlog_errno(ret);
2212 goto out; 2200 goto out;
2213 } 2201 }
2214 dwc->dw_orphaned = 1; 2202 dwc->dw_orphaned = 1;
2215 } 2203 }
2216 2204
2217 ret = ocfs2_inode_lock(inode, &di_bh, 2205 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2218 if (ret) { 2206 if (ret) {
2219 mlog_errno(ret); 2207 mlog_errno(ret);
2220 goto out; 2208 goto out;
2221 } 2209 }
2222 2210
2223 down_write(&oi->ip_alloc_sem); 2211 down_write(&oi->ip_alloc_sem);
2224 2212
2225 if (first_get_block) { 2213 if (first_get_block) {
2226 if (ocfs2_sparse_alloc(osb)) !! 2214 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
2227 ret = ocfs2_zero_tail 2215 ret = ocfs2_zero_tail(inode, di_bh, pos);
2228 else 2216 else
2229 ret = ocfs2_expand_no 2217 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2230 2218 total_len, NULL);
2231 if (ret < 0) { 2219 if (ret < 0) {
2232 mlog_errno(ret); 2220 mlog_errno(ret);
2233 goto unlock; 2221 goto unlock;
2234 } 2222 }
2235 } 2223 }
2236 2224
2237 ret = ocfs2_write_begin_nolock(inode- 2225 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2238 OCFS2_ 2226 OCFS2_WRITE_DIRECT, NULL,
2239 (void 2227 (void **)&wc, di_bh, NULL);
2240 if (ret) { 2228 if (ret) {
2241 mlog_errno(ret); 2229 mlog_errno(ret);
2242 goto unlock; 2230 goto unlock;
2243 } 2231 }
2244 2232
2245 desc = &wc->w_desc[0]; 2233 desc = &wc->w_desc[0];
2246 2234
2247 p_blkno = ocfs2_clusters_to_blocks(in 2235 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2248 BUG_ON(p_blkno == 0); 2236 BUG_ON(p_blkno == 0);
2249 p_blkno += iblock & (u64)(ocfs2_clust 2237 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2250 2238
2251 map_bh(bh_result, inode->i_sb, p_blkn 2239 map_bh(bh_result, inode->i_sb, p_blkno);
2252 bh_result->b_size = len; 2240 bh_result->b_size = len;
2253 if (desc->c_needs_zero) 2241 if (desc->c_needs_zero)
2254 set_buffer_new(bh_result); 2242 set_buffer_new(bh_result);
2255 2243
2256 if (iblock > endblk) <<
2257 set_buffer_new(bh_result); <<
2258 <<
2259 /* May sleep in end_io. It should not 2244 /* May sleep in end_io. It should not happen in a irq context. So defer
2260 * it to dio work queue. */ 2245 * it to dio work queue. */
2261 set_buffer_defer_completion(bh_result 2246 set_buffer_defer_completion(bh_result);
2262 2247
2263 if (!list_empty(&wc->w_unwritten_list 2248 if (!list_empty(&wc->w_unwritten_list)) {
2264 struct ocfs2_unwritten_extent 2249 struct ocfs2_unwritten_extent *ue = NULL;
2265 2250
2266 ue = list_first_entry(&wc->w_ 2251 ue = list_first_entry(&wc->w_unwritten_list,
2267 struct 2252 struct ocfs2_unwritten_extent,
2268 ue_node 2253 ue_node);
2269 BUG_ON(ue->ue_cpos != desc->c 2254 BUG_ON(ue->ue_cpos != desc->c_cpos);
2270 /* The physical address may b 2255 /* The physical address may be 0, fill it. */
2271 ue->ue_phys = desc->c_phys; 2256 ue->ue_phys = desc->c_phys;
2272 2257
2273 list_splice_tail_init(&wc->w_ 2258 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
2274 dwc->dw_zero_count += wc->w_u !! 2259 dwc->dw_zero_count++;
2275 } 2260 }
2276 2261
2277 ret = ocfs2_write_end_nolock(inode->i 2262 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc);
2278 BUG_ON(ret != len); 2263 BUG_ON(ret != len);
2279 ret = 0; 2264 ret = 0;
2280 unlock: 2265 unlock:
2281 up_write(&oi->ip_alloc_sem); 2266 up_write(&oi->ip_alloc_sem);
2282 ocfs2_inode_unlock(inode, 1); 2267 ocfs2_inode_unlock(inode, 1);
2283 brelse(di_bh); 2268 brelse(di_bh);
2284 out: 2269 out:
>> 2270 if (ret < 0)
>> 2271 ret = -EIO;
2285 return ret; 2272 return ret;
2286 } 2273 }
2287 2274
2288 static int ocfs2_dio_end_io_write(struct inod 2275 static int ocfs2_dio_end_io_write(struct inode *inode,
2289 struct ocfs 2276 struct ocfs2_dio_write_ctxt *dwc,
2290 loff_t offs 2277 loff_t offset,
2291 ssize_t byt 2278 ssize_t bytes)
2292 { 2279 {
2293 struct ocfs2_cached_dealloc_ctxt deal 2280 struct ocfs2_cached_dealloc_ctxt dealloc;
2294 struct ocfs2_extent_tree et; 2281 struct ocfs2_extent_tree et;
2295 struct ocfs2_super *osb = OCFS2_SB(in 2282 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2296 struct ocfs2_inode_info *oi = OCFS2_I 2283 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2297 struct ocfs2_unwritten_extent *ue = N 2284 struct ocfs2_unwritten_extent *ue = NULL;
2298 struct buffer_head *di_bh = NULL; 2285 struct buffer_head *di_bh = NULL;
2299 struct ocfs2_dinode *di; 2286 struct ocfs2_dinode *di;
2300 struct ocfs2_alloc_context *data_ac = 2287 struct ocfs2_alloc_context *data_ac = NULL;
2301 struct ocfs2_alloc_context *meta_ac = 2288 struct ocfs2_alloc_context *meta_ac = NULL;
2302 handle_t *handle = NULL; 2289 handle_t *handle = NULL;
2303 loff_t end = offset + bytes; 2290 loff_t end = offset + bytes;
2304 int ret = 0, credits = 0; !! 2291 int ret = 0, credits = 0, locked = 0;
2305 2292
2306 ocfs2_init_dealloc_ctxt(&dealloc); 2293 ocfs2_init_dealloc_ctxt(&dealloc);
2307 2294
2308 /* We do clear unwritten, delete orph 2295 /* We do clear unwritten, delete orphan, change i_size here. If neither
2309 * of these happen, we can skip all t 2296 * of these happen, we can skip all this. */
2310 if (list_empty(&dwc->dw_zero_list) && 2297 if (list_empty(&dwc->dw_zero_list) &&
2311 end <= i_size_read(inode) && 2298 end <= i_size_read(inode) &&
2312 !dwc->dw_orphaned) 2299 !dwc->dw_orphaned)
2313 goto out; 2300 goto out;
2314 2301
>> 2302 /* ocfs2_file_write_iter will get i_mutex, so we need not lock if we
>> 2303 * are in that context. */
>> 2304 if (dwc->dw_writer_pid != task_pid_nr(current)) {
>> 2305 inode_lock(inode);
>> 2306 locked = 1;
>> 2307 }
>> 2308
2315 ret = ocfs2_inode_lock(inode, &di_bh, 2309 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2316 if (ret < 0) { 2310 if (ret < 0) {
2317 mlog_errno(ret); 2311 mlog_errno(ret);
2318 goto out; 2312 goto out;
2319 } 2313 }
2320 2314
2321 down_write(&oi->ip_alloc_sem); 2315 down_write(&oi->ip_alloc_sem);
2322 2316
2323 /* Delete orphan before acquire i_rws !! 2317 /* Delete orphan before acquire i_mutex. */
2324 if (dwc->dw_orphaned) { 2318 if (dwc->dw_orphaned) {
2325 BUG_ON(dwc->dw_writer_pid != 2319 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2326 2320
2327 end = end > i_size_read(inode 2321 end = end > i_size_read(inode) ? end : 0;
2328 2322
2329 ret = ocfs2_del_inode_from_or 2323 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2330 !!end, end); 2324 !!end, end);
2331 if (ret < 0) 2325 if (ret < 0)
2332 mlog_errno(ret); 2326 mlog_errno(ret);
2333 } 2327 }
2334 2328
2335 di = (struct ocfs2_dinode *)di_bh->b_ 2329 di = (struct ocfs2_dinode *)di_bh->b_data;
2336 2330
2337 ocfs2_init_dinode_extent_tree(&et, IN 2331 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2338 2332
2339 /* Attach dealloc with extent tree in <<
2340 * which are already unlinked from cu <<
2341 * rotation and merging. <<
2342 */ <<
2343 et.et_dealloc = &dealloc; <<
2344 <<
2345 ret = ocfs2_lock_allocators(inode, &e 2333 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2346 &data_ac, 2334 &data_ac, &meta_ac);
2347 if (ret) { 2335 if (ret) {
2348 mlog_errno(ret); 2336 mlog_errno(ret);
2349 goto unlock; 2337 goto unlock;
2350 } 2338 }
2351 2339
2352 credits = ocfs2_calc_extend_credits(i 2340 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2353 2341
2354 handle = ocfs2_start_trans(osb, credi 2342 handle = ocfs2_start_trans(osb, credits);
2355 if (IS_ERR(handle)) { 2343 if (IS_ERR(handle)) {
2356 ret = PTR_ERR(handle); 2344 ret = PTR_ERR(handle);
2357 mlog_errno(ret); 2345 mlog_errno(ret);
2358 goto unlock; 2346 goto unlock;
2359 } 2347 }
2360 ret = ocfs2_journal_access_di(handle, 2348 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2361 OCFS2_J 2349 OCFS2_JOURNAL_ACCESS_WRITE);
2362 if (ret) { 2350 if (ret) {
2363 mlog_errno(ret); 2351 mlog_errno(ret);
2364 goto commit; 2352 goto commit;
2365 } 2353 }
2366 2354
2367 list_for_each_entry(ue, &dwc->dw_zero 2355 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) {
2368 ret = ocfs2_assure_trans_cred <<
2369 if (ret < 0) { <<
2370 mlog_errno(ret); <<
2371 break; <<
2372 } <<
2373 ret = ocfs2_mark_extent_writt 2356 ret = ocfs2_mark_extent_written(inode, &et, handle,
2374 2357 ue->ue_cpos, 1,
2375 2358 ue->ue_phys,
2376 2359 meta_ac, &dealloc);
2377 if (ret < 0) { 2360 if (ret < 0) {
2378 mlog_errno(ret); 2361 mlog_errno(ret);
2379 break; 2362 break;
2380 } 2363 }
2381 } 2364 }
2382 2365
2383 if (end > i_size_read(inode)) { 2366 if (end > i_size_read(inode)) {
2384 ret = ocfs2_set_inode_size(ha 2367 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2385 if (ret < 0) 2368 if (ret < 0)
2386 mlog_errno(ret); 2369 mlog_errno(ret);
2387 } 2370 }
2388 commit: 2371 commit:
2389 ocfs2_commit_trans(osb, handle); 2372 ocfs2_commit_trans(osb, handle);
2390 unlock: 2373 unlock:
2391 up_write(&oi->ip_alloc_sem); 2374 up_write(&oi->ip_alloc_sem);
2392 ocfs2_inode_unlock(inode, 1); 2375 ocfs2_inode_unlock(inode, 1);
2393 brelse(di_bh); 2376 brelse(di_bh);
2394 out: 2377 out:
2395 if (data_ac) 2378 if (data_ac)
2396 ocfs2_free_alloc_context(data 2379 ocfs2_free_alloc_context(data_ac);
2397 if (meta_ac) 2380 if (meta_ac)
2398 ocfs2_free_alloc_context(meta 2381 ocfs2_free_alloc_context(meta_ac);
2399 ocfs2_run_deallocs(osb, &dealloc); 2382 ocfs2_run_deallocs(osb, &dealloc);
>> 2383 if (locked)
>> 2384 inode_unlock(inode);
2400 ocfs2_dio_free_write_ctx(inode, dwc); 2385 ocfs2_dio_free_write_ctx(inode, dwc);
2401 2386
2402 return ret; 2387 return ret;
2403 } 2388 }
2404 2389
2405 /* 2390 /*
2406 * ocfs2_dio_end_io is called by the dio core 2391 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
2407 * particularly interested in the aio/dio cas 2392 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2408 * to protect io on one node from truncation 2393 * to protect io on one node from truncation on another.
2409 */ 2394 */
2410 static int ocfs2_dio_end_io(struct kiocb *ioc 2395 static int ocfs2_dio_end_io(struct kiocb *iocb,
2411 loff_t offset, 2396 loff_t offset,
2412 ssize_t bytes, 2397 ssize_t bytes,
2413 void *private) 2398 void *private)
2414 { 2399 {
2415 struct inode *inode = file_inode(iocb 2400 struct inode *inode = file_inode(iocb->ki_filp);
2416 int level; 2401 int level;
2417 int ret = 0; 2402 int ret = 0;
2418 2403
2419 /* this io's submitter should not hav 2404 /* this io's submitter should not have unlocked this before we could */
2420 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2405 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2421 2406
2422 if (bytes <= 0) !! 2407 if (bytes > 0 && private)
2423 mlog_ratelimited(ML_ERROR, "D !! 2408 ret = ocfs2_dio_end_io_write(inode, private, offset, bytes);
2424 (long long)b <<
2425 if (private) { <<
2426 if (bytes > 0) <<
2427 ret = ocfs2_dio_end_i <<
2428 <<
2429 else <<
2430 ocfs2_dio_free_write_ <<
2431 } <<
2432 2409
2433 ocfs2_iocb_clear_rw_locked(iocb); 2410 ocfs2_iocb_clear_rw_locked(iocb);
2434 2411
2435 level = ocfs2_iocb_rw_locked_level(io 2412 level = ocfs2_iocb_rw_locked_level(iocb);
2436 ocfs2_rw_unlock(inode, level); 2413 ocfs2_rw_unlock(inode, level);
2437 return ret; 2414 return ret;
2438 } 2415 }
2439 2416
2440 static ssize_t ocfs2_direct_IO(struct kiocb * 2417 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2441 { 2418 {
2442 struct file *file = iocb->ki_filp; 2419 struct file *file = iocb->ki_filp;
2443 struct inode *inode = file->f_mapping 2420 struct inode *inode = file->f_mapping->host;
2444 struct ocfs2_super *osb = OCFS2_SB(in 2421 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2445 get_block_t *get_block; 2422 get_block_t *get_block;
2446 2423
2447 /* 2424 /*
2448 * Fallback to buffered I/O if we see 2425 * Fallback to buffered I/O if we see an inode without
2449 * extents. 2426 * extents.
2450 */ 2427 */
2451 if (OCFS2_I(inode)->ip_dyn_features & 2428 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2452 return 0; 2429 return 0;
2453 2430
2454 /* Fallback to buffered I/O if we do 2431 /* Fallback to buffered I/O if we do not support append dio. */
2455 if (iocb->ki_pos + iter->count > i_si 2432 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2456 !ocfs2_supports_append_dio(osb)) 2433 !ocfs2_supports_append_dio(osb))
2457 return 0; 2434 return 0;
2458 2435
2459 if (iov_iter_rw(iter) == READ) 2436 if (iov_iter_rw(iter) == READ)
2460 get_block = ocfs2_lock_get_bl 2437 get_block = ocfs2_lock_get_block;
2461 else 2438 else
2462 get_block = ocfs2_dio_wr_get_ 2439 get_block = ocfs2_dio_wr_get_block;
2463 2440
2464 return __blockdev_direct_IO(iocb, ino 2441 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2465 iter, get 2442 iter, get_block,
2466 ocfs2_dio !! 2443 ocfs2_dio_end_io, NULL, 0);
2467 } 2444 }
2468 2445
2469 const struct address_space_operations ocfs2_a 2446 const struct address_space_operations ocfs2_aops = {
2470 .dirty_folio = block_dirty !! 2447 .readpage = ocfs2_readpage,
2471 .read_folio = ocfs2_read_ !! 2448 .readpages = ocfs2_readpages,
2472 .readahead = ocfs2_reada !! 2449 .writepage = ocfs2_writepage,
2473 .writepages = ocfs2_write <<
2474 .write_begin = ocfs2_write 2450 .write_begin = ocfs2_write_begin,
2475 .write_end = ocfs2_write 2451 .write_end = ocfs2_write_end,
2476 .bmap = ocfs2_bmap, 2452 .bmap = ocfs2_bmap,
2477 .direct_IO = ocfs2_direc 2453 .direct_IO = ocfs2_direct_IO,
2478 .invalidate_folio = block_inval !! 2454 .invalidatepage = block_invalidatepage,
2479 .release_folio = ocfs2_relea !! 2455 .releasepage = ocfs2_releasepage,
2480 .migrate_folio = buffer_migr !! 2456 .migratepage = buffer_migrate_page,
2481 .is_partially_uptodate = block_is_pa 2457 .is_partially_uptodate = block_is_partially_uptodate,
2482 .error_remove_folio = generic_err !! 2458 .error_remove_page = generic_error_remove_page,
2483 }; 2459 };
2484 2460
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.