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 * <<
122 struct buffer_head *bh_res <<
123 { <<
124 int ret = 0; <<
125 struct ocfs2_inode_info *oi = OCFS2_I( <<
126 <<
127 down_read(&oi->ip_alloc_sem); <<
128 ret = ocfs2_get_block(inode, iblock, b <<
129 up_read(&oi->ip_alloc_sem); <<
130 <<
131 return ret; <<
132 } <<
133 <<
134 int ocfs2_get_block(struct inode *inode, secto 137 int ocfs2_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh_res 138 struct buffer_head *bh_result, int create)
136 { 139 {
137 int err = 0; 140 int err = 0;
138 unsigned int ext_flags; 141 unsigned int ext_flags;
139 u64 max_blocks = bh_result->b_size >> 142 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
140 u64 p_blkno, count, past_eof; 143 u64 p_blkno, count, past_eof;
141 struct ocfs2_super *osb = OCFS2_SB(ino 144 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
142 145
143 trace_ocfs2_get_block((unsigned long l 146 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
144 (unsigned long l 147 (unsigned long long)iblock, bh_result, create);
145 148
146 if (OCFS2_I(inode)->ip_flags & OCFS2_I 149 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
147 mlog(ML_NOTICE, "get_block on 150 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
148 inode, inode->i_ino); 151 inode, inode->i_ino);
149 152
150 if (S_ISLNK(inode->i_mode)) { 153 if (S_ISLNK(inode->i_mode)) {
151 /* this always does I/O for so 154 /* this always does I/O for some reason. */
152 err = ocfs2_symlink_get_block( 155 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
153 goto bail; 156 goto bail;
154 } 157 }
155 158
156 err = ocfs2_extent_map_get_blocks(inod 159 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
157 &ext 160 &ext_flags);
158 if (err) { 161 if (err) {
159 mlog(ML_ERROR, "get_blocks() f !! 162 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
160 "block: %llu\n", inode, ( !! 163 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
>> 164 (unsigned long long)p_blkno);
161 goto bail; 165 goto bail;
162 } 166 }
163 167
164 if (max_blocks < count) 168 if (max_blocks < count)
165 count = max_blocks; 169 count = max_blocks;
166 170
167 /* 171 /*
168 * ocfs2 never allocates in this funct 172 * ocfs2 never allocates in this function - the only time we
169 * need to use BH_New is when we're ex 173 * need to use BH_New is when we're extending i_size on a file
170 * system which doesn't support holes, 174 * system which doesn't support holes, in which case BH_New
171 * allows __block_write_begin() to zer 175 * allows __block_write_begin() to zero.
172 * 176 *
173 * If we see this on a sparse file sys 177 * If we see this on a sparse file system, then a truncate has
174 * raced us and removed the cluster. I 178 * raced us and removed the cluster. In this case, we clear
175 * the buffers dirty and uptodate bits 179 * the buffers dirty and uptodate bits and let the buffer code
176 * ignore it as a hole. 180 * ignore it as a hole.
177 */ 181 */
178 if (create && p_blkno == 0 && ocfs2_sp 182 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
179 clear_buffer_dirty(bh_result); 183 clear_buffer_dirty(bh_result);
180 clear_buffer_uptodate(bh_resul 184 clear_buffer_uptodate(bh_result);
181 goto bail; 185 goto bail;
182 } 186 }
183 187
184 /* Treat the unwritten extent as a hol 188 /* Treat the unwritten extent as a hole for zeroing purposes. */
185 if (p_blkno && !(ext_flags & OCFS2_EXT 189 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
186 map_bh(bh_result, inode->i_sb, 190 map_bh(bh_result, inode->i_sb, p_blkno);
187 191
188 bh_result->b_size = count << inode->i_ 192 bh_result->b_size = count << inode->i_blkbits;
189 193
190 if (!ocfs2_sparse_alloc(osb)) { 194 if (!ocfs2_sparse_alloc(osb)) {
191 if (p_blkno == 0) { 195 if (p_blkno == 0) {
192 err = -EIO; 196 err = -EIO;
193 mlog(ML_ERROR, 197 mlog(ML_ERROR,
194 "iblock = %llu p_ 198 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
195 (unsigned long lo 199 (unsigned long long)iblock,
196 (unsigned long lo 200 (unsigned long long)p_blkno,
197 (unsigned long lo 201 (unsigned long long)OCFS2_I(inode)->ip_blkno);
198 mlog(ML_ERROR, "Size % 202 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
199 dump_stack(); 203 dump_stack();
200 goto bail; 204 goto bail;
201 } 205 }
202 } 206 }
203 207
204 past_eof = ocfs2_blocks_for_bytes(inod 208 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
205 209
206 trace_ocfs2_get_block_end((unsigned lo 210 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
207 (unsigned lo 211 (unsigned long long)past_eof);
208 if (create && (iblock >= past_eof)) 212 if (create && (iblock >= past_eof))
209 set_buffer_new(bh_result); 213 set_buffer_new(bh_result);
210 214
211 bail: 215 bail:
212 if (err < 0) 216 if (err < 0)
213 err = -EIO; 217 err = -EIO;
214 218
215 return err; 219 return err;
216 } 220 }
217 221
218 int ocfs2_read_inline_data(struct inode *inode 222 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
219 struct buffer_head 223 struct buffer_head *di_bh)
220 { 224 {
221 void *kaddr; 225 void *kaddr;
222 loff_t size; 226 loff_t size;
223 struct ocfs2_dinode *di = (struct ocfs 227 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
224 228
225 if (!(le16_to_cpu(di->i_dyn_features) 229 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
226 ocfs2_error(inode->i_sb, "Inod 230 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
227 (unsigned long lon 231 (unsigned long long)OCFS2_I(inode)->ip_blkno);
228 return -EROFS; 232 return -EROFS;
229 } 233 }
230 234
231 size = i_size_read(inode); 235 size = i_size_read(inode);
232 236
233 if (size > PAGE_SIZE || 237 if (size > PAGE_SIZE ||
234 size > ocfs2_max_inline_data_with_ 238 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
235 ocfs2_error(inode->i_sb, 239 ocfs2_error(inode->i_sb,
236 "Inode %llu has wi 240 "Inode %llu has with inline data has bad size: %Lu\n",
237 (unsigned long lon 241 (unsigned long long)OCFS2_I(inode)->ip_blkno,
238 (unsigned long lon 242 (unsigned long long)size);
239 return -EROFS; 243 return -EROFS;
240 } 244 }
241 245
242 kaddr = kmap_atomic(page); 246 kaddr = kmap_atomic(page);
243 if (size) 247 if (size)
244 memcpy(kaddr, di->id2.i_data.i 248 memcpy(kaddr, di->id2.i_data.id_data, size);
245 /* Clear the remaining part of the pag 249 /* Clear the remaining part of the page */
246 memset(kaddr + size, 0, PAGE_SIZE - si 250 memset(kaddr + size, 0, PAGE_SIZE - size);
247 flush_dcache_page(page); 251 flush_dcache_page(page);
248 kunmap_atomic(kaddr); 252 kunmap_atomic(kaddr);
249 253
250 SetPageUptodate(page); 254 SetPageUptodate(page);
251 255
252 return 0; 256 return 0;
253 } 257 }
254 258
255 static int ocfs2_readpage_inline(struct inode 259 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
256 { 260 {
257 int ret; 261 int ret;
258 struct buffer_head *di_bh = NULL; 262 struct buffer_head *di_bh = NULL;
259 263
260 BUG_ON(!PageLocked(page)); 264 BUG_ON(!PageLocked(page));
261 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 265 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
262 266
263 ret = ocfs2_read_inode_block(inode, &d 267 ret = ocfs2_read_inode_block(inode, &di_bh);
264 if (ret) { 268 if (ret) {
265 mlog_errno(ret); 269 mlog_errno(ret);
266 goto out; 270 goto out;
267 } 271 }
268 272
269 ret = ocfs2_read_inline_data(inode, pa 273 ret = ocfs2_read_inline_data(inode, page, di_bh);
270 out: 274 out:
271 unlock_page(page); 275 unlock_page(page);
272 276
273 brelse(di_bh); 277 brelse(di_bh);
274 return ret; 278 return ret;
275 } 279 }
276 280
277 static int ocfs2_read_folio(struct file *file, !! 281 static int ocfs2_readpage(struct file *file, struct page *page)
278 { 282 {
279 struct inode *inode = folio->mapping-> !! 283 struct inode *inode = page->mapping->host;
280 struct ocfs2_inode_info *oi = OCFS2_I( 284 struct ocfs2_inode_info *oi = OCFS2_I(inode);
281 loff_t start = folio_pos(folio); !! 285 loff_t start = (loff_t)page->index << PAGE_SHIFT;
282 int ret, unlock = 1; 286 int ret, unlock = 1;
283 287
284 trace_ocfs2_readpage((unsigned long lo !! 288 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
>> 289 (page ? page->index : 0));
285 290
286 ret = ocfs2_inode_lock_with_page(inode !! 291 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
287 if (ret != 0) { 292 if (ret != 0) {
288 if (ret == AOP_TRUNCATED_PAGE) 293 if (ret == AOP_TRUNCATED_PAGE)
289 unlock = 0; 294 unlock = 0;
290 mlog_errno(ret); 295 mlog_errno(ret);
291 goto out; 296 goto out;
292 } 297 }
293 298
294 if (down_read_trylock(&oi->ip_alloc_se 299 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
295 /* 300 /*
296 * Unlock the folio and cycle !! 301 * Unlock the page and cycle ip_alloc_sem so that we don't
297 * busyloop waiting for ip_all 302 * busyloop waiting for ip_alloc_sem to unlock
298 */ 303 */
299 ret = AOP_TRUNCATED_PAGE; 304 ret = AOP_TRUNCATED_PAGE;
300 folio_unlock(folio); !! 305 unlock_page(page);
301 unlock = 0; 306 unlock = 0;
302 down_read(&oi->ip_alloc_sem); 307 down_read(&oi->ip_alloc_sem);
303 up_read(&oi->ip_alloc_sem); 308 up_read(&oi->ip_alloc_sem);
304 goto out_inode_unlock; 309 goto out_inode_unlock;
305 } 310 }
306 311
307 /* 312 /*
308 * i_size might have just been updated 313 * i_size might have just been updated as we grabed the meta lock. We
309 * might now be discovering a truncate 314 * might now be discovering a truncate that hit on another node.
310 * block_read_full_folio->get_block fr !! 315 * block_read_full_page->get_block freaks out if it is asked to read
311 * beyond the end of a file, so we che 316 * beyond the end of a file, so we check here. Callers
312 * (generic_file_read, vm_ops->fault) 317 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
313 * and notice that the folio they just !! 318 * and notice that the page they just read isn't needed.
314 * 319 *
315 * XXX sys_readahead() seems to get th 320 * XXX sys_readahead() seems to get that wrong?
316 */ 321 */
317 if (start >= i_size_read(inode)) { 322 if (start >= i_size_read(inode)) {
318 folio_zero_segment(folio, 0, f !! 323 zero_user(page, 0, PAGE_SIZE);
319 folio_mark_uptodate(folio); !! 324 SetPageUptodate(page);
320 ret = 0; 325 ret = 0;
321 goto out_alloc; 326 goto out_alloc;
322 } 327 }
323 328
324 if (oi->ip_dyn_features & OCFS2_INLINE 329 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
325 ret = ocfs2_readpage_inline(in !! 330 ret = ocfs2_readpage_inline(inode, page);
326 else 331 else
327 ret = block_read_full_folio(fo !! 332 ret = block_read_full_page(page, ocfs2_get_block);
328 unlock = 0; 333 unlock = 0;
329 334
330 out_alloc: 335 out_alloc:
331 up_read(&oi->ip_alloc_sem); !! 336 up_read(&OCFS2_I(inode)->ip_alloc_sem);
332 out_inode_unlock: 337 out_inode_unlock:
333 ocfs2_inode_unlock(inode, 0); 338 ocfs2_inode_unlock(inode, 0);
334 out: 339 out:
335 if (unlock) 340 if (unlock)
336 folio_unlock(folio); !! 341 unlock_page(page);
337 return ret; 342 return ret;
338 } 343 }
339 344
340 /* 345 /*
341 * This is used only for read-ahead. Failures 346 * This is used only for read-ahead. Failures or difficult to handle
342 * situations are safe to ignore. 347 * situations are safe to ignore.
343 * 348 *
344 * Right now, we don't bother with BH_Boundary 349 * Right now, we don't bother with BH_Boundary - in-inode extent lists
345 * are quite large (243 extents on 4k blocks), 350 * are quite large (243 extents on 4k blocks), so most inodes don't
346 * grow out to a tree. If need be, detecting b 351 * grow out to a tree. If need be, detecting boundary extents could
347 * trivially be added in a future version of o 352 * trivially be added in a future version of ocfs2_get_block().
348 */ 353 */
349 static void ocfs2_readahead(struct readahead_c !! 354 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
>> 355 struct list_head *pages, unsigned nr_pages)
350 { 356 {
351 int ret; !! 357 int ret, err = -EIO;
352 struct inode *inode = rac->mapping->ho !! 358 struct inode *inode = mapping->host;
353 struct ocfs2_inode_info *oi = OCFS2_I( 359 struct ocfs2_inode_info *oi = OCFS2_I(inode);
>> 360 loff_t start;
>> 361 struct page *last;
354 362
355 /* 363 /*
356 * Use the nonblocking flag for the dl 364 * Use the nonblocking flag for the dlm code to avoid page
357 * lock inversion, but don't bother wi 365 * lock inversion, but don't bother with retrying.
358 */ 366 */
359 ret = ocfs2_inode_lock_full(inode, NUL 367 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
360 if (ret) 368 if (ret)
361 return; !! 369 return err;
362 370
363 if (down_read_trylock(&oi->ip_alloc_se !! 371 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
364 goto out_unlock; !! 372 ocfs2_inode_unlock(inode, 0);
>> 373 return err;
>> 374 }
365 375
366 /* 376 /*
367 * Don't bother with inline-data. Ther 377 * Don't bother with inline-data. There isn't anything
368 * to read-ahead in that case anyway.. 378 * to read-ahead in that case anyway...
369 */ 379 */
370 if (oi->ip_dyn_features & OCFS2_INLINE 380 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
371 goto out_up; !! 381 goto out_unlock;
372 382
373 /* 383 /*
374 * Check whether a remote node truncat 384 * Check whether a remote node truncated this file - we just
375 * drop out in that case as it's not w 385 * drop out in that case as it's not worth handling here.
376 */ 386 */
377 if (readahead_pos(rac) >= i_size_read( !! 387 last = list_entry(pages->prev, struct page, lru);
378 goto out_up; !! 388 start = (loff_t)last->index << PAGE_SHIFT;
>> 389 if (start >= i_size_read(inode))
>> 390 goto out_unlock;
379 391
380 mpage_readahead(rac, ocfs2_get_block); !! 392 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
381 393
382 out_up: <<
383 up_read(&oi->ip_alloc_sem); <<
384 out_unlock: 394 out_unlock:
>> 395 up_read(&oi->ip_alloc_sem);
385 ocfs2_inode_unlock(inode, 0); 396 ocfs2_inode_unlock(inode, 0);
>> 397
>> 398 return err;
386 } 399 }
387 400
388 /* Note: Because we don't support holes, our a 401 /* Note: Because we don't support holes, our allocation has
389 * already happened (allocation writes zeros t 402 * already happened (allocation writes zeros to the file data)
390 * so we don't have to worry about ordered wri 403 * so we don't have to worry about ordered writes in
391 * ocfs2_writepages. !! 404 * ocfs2_writepage.
392 * 405 *
393 * ->writepages is called during the process o !! 406 * ->writepage is called during the process of invalidating the page cache
394 * during blocked lock processing. It can't b 407 * during blocked lock processing. It can't block on any cluster locks
395 * to during block mapping. It's relying on t 408 * to during block mapping. It's relying on the fact that the block
396 * mapping can't have disappeared under the di 409 * mapping can't have disappeared under the dirty pages that it is
397 * being asked to write back. 410 * being asked to write back.
398 */ 411 */
399 static int ocfs2_writepages(struct address_spa !! 412 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
400 struct writeback_control *wbc) <<
401 { 413 {
402 return mpage_writepages(mapping, wbc, !! 414 trace_ocfs2_writepage(
>> 415 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
>> 416 page->index);
>> 417
>> 418 return block_write_full_page(page, ocfs2_get_block, wbc);
403 } 419 }
404 420
405 /* Taken from ext3. We don't necessarily need 421 /* Taken from ext3. We don't necessarily need the full blown
406 * functionality yet, but IMHO it's better to 422 * functionality yet, but IMHO it's better to cut and paste the whole
407 * thing so we can avoid introducing our own b 423 * thing so we can avoid introducing our own bugs (and easily pick up
408 * their fixes when they happen) --Mark */ 424 * their fixes when they happen) --Mark */
409 int walk_page_buffers( handle_t *handle, 425 int walk_page_buffers( handle_t *handle,
410 struct buffer_head *he 426 struct buffer_head *head,
411 unsigned from, 427 unsigned from,
412 unsigned to, 428 unsigned to,
413 int *partial, 429 int *partial,
414 int (*fn)( handle 430 int (*fn)( handle_t *handle,
415 struct 431 struct buffer_head *bh))
416 { 432 {
417 struct buffer_head *bh; 433 struct buffer_head *bh;
418 unsigned block_start, block_end; 434 unsigned block_start, block_end;
419 unsigned blocksize = head->b_size; 435 unsigned blocksize = head->b_size;
420 int err, ret = 0; 436 int err, ret = 0;
421 struct buffer_head *next; 437 struct buffer_head *next;
422 438
423 for ( bh = head, block_start = 0; 439 for ( bh = head, block_start = 0;
424 ret == 0 && (bh != head || !bl 440 ret == 0 && (bh != head || !block_start);
425 block_start = block_end, bh = 441 block_start = block_end, bh = next)
426 { 442 {
427 next = bh->b_this_page; 443 next = bh->b_this_page;
428 block_end = block_start + bloc 444 block_end = block_start + blocksize;
429 if (block_end <= from || block 445 if (block_end <= from || block_start >= to) {
430 if (partial && !buffer 446 if (partial && !buffer_uptodate(bh))
431 *partial = 1; 447 *partial = 1;
432 continue; 448 continue;
433 } 449 }
434 err = (*fn)(handle, bh); 450 err = (*fn)(handle, bh);
435 if (!ret) 451 if (!ret)
436 ret = err; 452 ret = err;
437 } 453 }
438 return ret; 454 return ret;
439 } 455 }
440 456
441 static sector_t ocfs2_bmap(struct address_spac 457 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
442 { 458 {
443 sector_t status; 459 sector_t status;
444 u64 p_blkno = 0; 460 u64 p_blkno = 0;
445 int err = 0; 461 int err = 0;
446 struct inode *inode = mapping->host; 462 struct inode *inode = mapping->host;
447 463
448 trace_ocfs2_bmap((unsigned long long)O 464 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
449 (unsigned long long)b 465 (unsigned long long)block);
450 466
451 /* 467 /*
452 * The swap code (ab-)uses ->bmap to g 468 * The swap code (ab-)uses ->bmap to get a block mapping and then
453 * bypasseѕ the file system for actua 469 * bypasseѕ the file system for actual I/O. We really can't allow
454 * that on refcounted inodes, so we ha 470 * that on refcounted inodes, so we have to skip out here. And yes,
455 * 0 is the magic code for a bmap erro 471 * 0 is the magic code for a bmap error..
456 */ 472 */
457 if (ocfs2_is_refcount_inode(inode)) 473 if (ocfs2_is_refcount_inode(inode))
458 return 0; 474 return 0;
459 475
460 /* We don't need to lock journal syste 476 /* We don't need to lock journal system files, since they aren't
461 * accessed concurrently from multiple 477 * accessed concurrently from multiple nodes.
462 */ 478 */
463 if (!INODE_JOURNAL(inode)) { 479 if (!INODE_JOURNAL(inode)) {
464 err = ocfs2_inode_lock(inode, 480 err = ocfs2_inode_lock(inode, NULL, 0);
465 if (err) { 481 if (err) {
466 if (err != -ENOENT) 482 if (err != -ENOENT)
467 mlog_errno(err 483 mlog_errno(err);
468 goto bail; 484 goto bail;
469 } 485 }
470 down_read(&OCFS2_I(inode)->ip_ 486 down_read(&OCFS2_I(inode)->ip_alloc_sem);
471 } 487 }
472 488
473 if (!(OCFS2_I(inode)->ip_dyn_features 489 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
474 err = ocfs2_extent_map_get_blo 490 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
475 491 NULL);
476 492
477 if (!INODE_JOURNAL(inode)) { 493 if (!INODE_JOURNAL(inode)) {
478 up_read(&OCFS2_I(inode)->ip_al 494 up_read(&OCFS2_I(inode)->ip_alloc_sem);
479 ocfs2_inode_unlock(inode, 0); 495 ocfs2_inode_unlock(inode, 0);
480 } 496 }
481 497
482 if (err) { 498 if (err) {
483 mlog(ML_ERROR, "get_blocks() f 499 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
484 (unsigned long long)block 500 (unsigned long long)block);
485 mlog_errno(err); 501 mlog_errno(err);
486 goto bail; 502 goto bail;
487 } 503 }
488 504
489 bail: 505 bail:
490 status = err ? 0 : p_blkno; 506 status = err ? 0 : p_blkno;
491 507
492 return status; 508 return status;
493 } 509 }
494 510
495 static bool ocfs2_release_folio(struct folio * !! 511 static int ocfs2_releasepage(struct page *page, gfp_t wait)
496 { 512 {
497 if (!folio_buffers(folio)) !! 513 if (!page_has_buffers(page))
498 return false; !! 514 return 0;
499 return try_to_free_buffers(folio); !! 515 return try_to_free_buffers(page);
500 } 516 }
501 517
502 static void ocfs2_figure_cluster_boundaries(st 518 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
503 u3 519 u32 cpos,
504 un 520 unsigned int *start,
505 un 521 unsigned int *end)
506 { 522 {
507 unsigned int cluster_start = 0, cluste 523 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
508 524
509 if (unlikely(PAGE_SHIFT > osb->s_clust 525 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
510 unsigned int cpp; 526 unsigned int cpp;
511 527
512 cpp = 1 << (PAGE_SHIFT - osb-> 528 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
513 529
514 cluster_start = cpos % cpp; 530 cluster_start = cpos % cpp;
515 cluster_start = cluster_start 531 cluster_start = cluster_start << osb->s_clustersize_bits;
516 532
517 cluster_end = cluster_start + 533 cluster_end = cluster_start + osb->s_clustersize;
518 } 534 }
519 535
520 BUG_ON(cluster_start > PAGE_SIZE); 536 BUG_ON(cluster_start > PAGE_SIZE);
521 BUG_ON(cluster_end > PAGE_SIZE); 537 BUG_ON(cluster_end > PAGE_SIZE);
522 538
523 if (start) 539 if (start)
524 *start = cluster_start; 540 *start = cluster_start;
525 if (end) 541 if (end)
526 *end = cluster_end; 542 *end = cluster_end;
527 } 543 }
528 544
529 /* 545 /*
530 * 'from' and 'to' are the region in the page 546 * 'from' and 'to' are the region in the page to avoid zeroing.
531 * 547 *
532 * If pagesize > clustersize, this function wi 548 * If pagesize > clustersize, this function will avoid zeroing outside
533 * of the cluster boundary. 549 * of the cluster boundary.
534 * 550 *
535 * from == to == 0 is code for "zero the entir 551 * from == to == 0 is code for "zero the entire cluster region"
536 */ 552 */
537 static void ocfs2_clear_page_regions(struct pa 553 static void ocfs2_clear_page_regions(struct page *page,
538 struct oc 554 struct ocfs2_super *osb, u32 cpos,
539 unsigned 555 unsigned from, unsigned to)
540 { 556 {
541 void *kaddr; 557 void *kaddr;
542 unsigned int cluster_start, cluster_en 558 unsigned int cluster_start, cluster_end;
543 559
544 ocfs2_figure_cluster_boundaries(osb, c 560 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
545 561
546 kaddr = kmap_atomic(page); 562 kaddr = kmap_atomic(page);
547 563
548 if (from || to) { 564 if (from || to) {
549 if (from > cluster_start) 565 if (from > cluster_start)
550 memset(kaddr + cluster 566 memset(kaddr + cluster_start, 0, from - cluster_start);
551 if (to < cluster_end) 567 if (to < cluster_end)
552 memset(kaddr + to, 0, 568 memset(kaddr + to, 0, cluster_end - to);
553 } else { 569 } else {
554 memset(kaddr + cluster_start, 570 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
555 } 571 }
556 572
557 kunmap_atomic(kaddr); 573 kunmap_atomic(kaddr);
558 } 574 }
559 575
560 /* 576 /*
561 * Nonsparse file systems fully allocate befor 577 * Nonsparse file systems fully allocate before we get to the write
562 * code. This prevents ocfs2_write() from tagg 578 * code. This prevents ocfs2_write() from tagging the write as an
563 * allocating one, which means ocfs2_map_page_ 579 * allocating one, which means ocfs2_map_page_blocks() might try to
564 * read-in the blocks at the tail of our file. 580 * read-in the blocks at the tail of our file. Avoid reading them by
565 * testing i_size against each block offset. 581 * testing i_size against each block offset.
566 */ 582 */
567 static int ocfs2_should_read_blk(struct inode !! 583 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
568 unsigned int 584 unsigned int block_start)
569 { 585 {
570 u64 offset = folio_pos(folio) + block_ !! 586 u64 offset = page_offset(page) + block_start;
571 587
572 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 588 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
573 return 1; 589 return 1;
574 590
575 if (i_size_read(inode) > offset) 591 if (i_size_read(inode) > offset)
576 return 1; 592 return 1;
577 593
578 return 0; 594 return 0;
579 } 595 }
580 596
581 /* 597 /*
582 * Some of this taken from __block_write_begin 598 * Some of this taken from __block_write_begin(). We already have our
583 * mapping by now though, and the entire write 599 * mapping by now though, and the entire write will be allocating or
584 * it won't, so not much need to use BH_New. 600 * it won't, so not much need to use BH_New.
585 * 601 *
586 * This will also skip zeroing, which is handl 602 * This will also skip zeroing, which is handled externally.
587 */ 603 */
588 int ocfs2_map_page_blocks(struct page *page, u 604 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
589 struct inode *inode, 605 struct inode *inode, unsigned int from,
590 unsigned int to, int 606 unsigned int to, int new)
591 { 607 {
592 struct folio *folio = page_folio(page) <<
593 int ret = 0; 608 int ret = 0;
594 struct buffer_head *head, *bh, *wait[2 609 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
595 unsigned int block_end, block_start; 610 unsigned int block_end, block_start;
596 unsigned int bsize = i_blocksize(inode 611 unsigned int bsize = i_blocksize(inode);
597 612
598 head = folio_buffers(folio); !! 613 if (!page_has_buffers(page))
599 if (!head) !! 614 create_empty_buffers(page, bsize, 0);
600 head = create_empty_buffers(fo <<
601 615
>> 616 head = page_buffers(page);
602 for (bh = head, block_start = 0; bh != 617 for (bh = head, block_start = 0; bh != head || !block_start;
603 bh = bh->b_this_page, block_start 618 bh = bh->b_this_page, block_start += bsize) {
604 block_end = block_start + bsiz 619 block_end = block_start + bsize;
605 620
606 clear_buffer_new(bh); 621 clear_buffer_new(bh);
607 622
608 /* 623 /*
609 * Ignore blocks outside of ou 624 * Ignore blocks outside of our i/o range -
610 * they may belong to unalloca 625 * they may belong to unallocated clusters.
611 */ 626 */
612 if (block_start >= to || block 627 if (block_start >= to || block_end <= from) {
613 if (folio_test_uptodat !! 628 if (PageUptodate(page))
614 set_buffer_upt 629 set_buffer_uptodate(bh);
615 continue; 630 continue;
616 } 631 }
617 632
618 /* 633 /*
619 * For an allocating write wit 634 * For an allocating write with cluster size >= page
620 * size, we always write the e 635 * size, we always write the entire page.
621 */ 636 */
622 if (new) 637 if (new)
623 set_buffer_new(bh); 638 set_buffer_new(bh);
624 639
625 if (!buffer_mapped(bh)) { 640 if (!buffer_mapped(bh)) {
626 map_bh(bh, inode->i_sb 641 map_bh(bh, inode->i_sb, *p_blkno);
627 clean_bdev_bh_alias(bh 642 clean_bdev_bh_alias(bh);
628 } 643 }
629 644
630 if (folio_test_uptodate(folio) !! 645 if (PageUptodate(page)) {
631 set_buffer_uptodate(bh !! 646 if (!buffer_uptodate(bh))
>> 647 set_buffer_uptodate(bh);
632 } else if (!buffer_uptodate(bh 648 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
633 !buffer_new(bh) && 649 !buffer_new(bh) &&
634 ocfs2_should_read_b !! 650 ocfs2_should_read_blk(inode, page, block_start) &&
635 (block_start < from 651 (block_start < from || block_end > to)) {
636 bh_read_nowait(bh, 0); !! 652 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
637 *wait_bh++=bh; 653 *wait_bh++=bh;
638 } 654 }
639 655
640 *p_blkno = *p_blkno + 1; 656 *p_blkno = *p_blkno + 1;
641 } 657 }
642 658
643 /* 659 /*
644 * If we issued read requests - let th 660 * If we issued read requests - let them complete.
645 */ 661 */
646 while(wait_bh > wait) { 662 while(wait_bh > wait) {
647 wait_on_buffer(*--wait_bh); 663 wait_on_buffer(*--wait_bh);
648 if (!buffer_uptodate(*wait_bh) 664 if (!buffer_uptodate(*wait_bh))
649 ret = -EIO; 665 ret = -EIO;
650 } 666 }
651 667
652 if (ret == 0 || !new) 668 if (ret == 0 || !new)
653 return ret; 669 return ret;
654 670
655 /* 671 /*
656 * If we get -EIO above, zero out any 672 * If we get -EIO above, zero out any newly allocated blocks
657 * to avoid exposing stale data. 673 * to avoid exposing stale data.
658 */ 674 */
659 bh = head; 675 bh = head;
660 block_start = 0; 676 block_start = 0;
661 do { 677 do {
662 block_end = block_start + bsiz 678 block_end = block_start + bsize;
663 if (block_end <= from) 679 if (block_end <= from)
664 goto next_bh; 680 goto next_bh;
665 if (block_start >= to) 681 if (block_start >= to)
666 break; 682 break;
667 683
668 folio_zero_range(folio, block_ !! 684 zero_user(page, block_start, bh->b_size);
669 set_buffer_uptodate(bh); 685 set_buffer_uptodate(bh);
670 mark_buffer_dirty(bh); 686 mark_buffer_dirty(bh);
671 687
672 next_bh: 688 next_bh:
673 block_start = block_end; 689 block_start = block_end;
674 bh = bh->b_this_page; 690 bh = bh->b_this_page;
675 } while (bh != head); 691 } while (bh != head);
676 692
677 return ret; 693 return ret;
678 } 694 }
679 695
680 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 696 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
681 #define OCFS2_MAX_CTXT_PAGES 1 697 #define OCFS2_MAX_CTXT_PAGES 1
682 #else 698 #else
683 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 699 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
684 #endif 700 #endif
685 701
686 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 702 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
687 703
688 struct ocfs2_unwritten_extent { 704 struct ocfs2_unwritten_extent {
689 struct list_head ue_node; 705 struct list_head ue_node;
690 struct list_head ue_ip_node; 706 struct list_head ue_ip_node;
691 u32 ue_cpos; 707 u32 ue_cpos;
692 u32 ue_phys; 708 u32 ue_phys;
693 }; 709 };
694 710
695 /* 711 /*
696 * Describe the state of a single cluster to b 712 * Describe the state of a single cluster to be written to.
697 */ 713 */
698 struct ocfs2_write_cluster_desc { 714 struct ocfs2_write_cluster_desc {
699 u32 c_cpos; 715 u32 c_cpos;
700 u32 c_phys; 716 u32 c_phys;
701 /* 717 /*
702 * Give this a unique field because c_ 718 * Give this a unique field because c_phys eventually gets
703 * filled. 719 * filled.
704 */ 720 */
705 unsigned c_new; 721 unsigned c_new;
706 unsigned c_clear_unwritten; 722 unsigned c_clear_unwritten;
707 unsigned c_needs_zero; 723 unsigned c_needs_zero;
708 }; 724 };
709 725
710 struct ocfs2_write_ctxt { 726 struct ocfs2_write_ctxt {
711 /* Logical cluster position / len of w 727 /* Logical cluster position / len of write */
712 u32 w_cpos 728 u32 w_cpos;
713 u32 w_clen 729 u32 w_clen;
714 730
715 /* First cluster allocated in a nonspa 731 /* First cluster allocated in a nonsparse extend */
716 u32 w_firs 732 u32 w_first_new_cpos;
717 733
718 /* Type of caller. Must be one of buff 734 /* Type of caller. Must be one of buffer, mmap, direct. */
719 ocfs2_write_type_t w_type 735 ocfs2_write_type_t w_type;
720 736
721 struct ocfs2_write_cluster_desc w_desc 737 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
722 738
723 /* 739 /*
724 * This is true if page_size > cluster 740 * This is true if page_size > cluster_size.
725 * 741 *
726 * It triggers a set of special cases 742 * It triggers a set of special cases during write which might
727 * have to deal with allocating writes 743 * have to deal with allocating writes to partial pages.
728 */ 744 */
729 unsigned int w_larg 745 unsigned int w_large_pages;
730 746
731 /* 747 /*
732 * Pages involved in this write. 748 * Pages involved in this write.
733 * 749 *
734 * w_target_page is the page being wri 750 * w_target_page is the page being written to by the user.
735 * 751 *
736 * w_pages is an array of pages which 752 * w_pages is an array of pages which always contains
737 * w_target_page, and in the case of a 753 * w_target_page, and in the case of an allocating write with
738 * page_size < cluster size, it will c 754 * page_size < cluster size, it will contain zero'd and mapped
739 * pages adjacent to w_target_page whi 755 * pages adjacent to w_target_page which need to be written
740 * out in so that future reads from th 756 * out in so that future reads from that region will get
741 * zero's. 757 * zero's.
742 */ 758 */
743 unsigned int w_num_ 759 unsigned int w_num_pages;
744 struct page *w_pag 760 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
745 struct page *w_tar 761 struct page *w_target_page;
746 762
747 /* 763 /*
748 * w_target_locked is used for page_mk 764 * w_target_locked is used for page_mkwrite path indicating no unlocking
749 * against w_target_page in ocfs2_writ 765 * against w_target_page in ocfs2_write_end_nolock.
750 */ 766 */
751 unsigned int w_targ 767 unsigned int w_target_locked:1;
752 768
753 /* 769 /*
754 * ocfs2_write_end() uses this to know 770 * ocfs2_write_end() uses this to know what the real range to
755 * write in the target should be. 771 * write in the target should be.
756 */ 772 */
757 unsigned int w_targ 773 unsigned int w_target_from;
758 unsigned int w_targ 774 unsigned int w_target_to;
759 775
760 /* 776 /*
761 * We could use journal_current_handle 777 * We could use journal_current_handle() but this is cleaner,
762 * IMHO -Mark 778 * IMHO -Mark
763 */ 779 */
764 handle_t *w_han 780 handle_t *w_handle;
765 781
766 struct buffer_head *w_di_ 782 struct buffer_head *w_di_bh;
767 783
768 struct ocfs2_cached_dealloc_ctxt w_dea 784 struct ocfs2_cached_dealloc_ctxt w_dealloc;
769 785
770 struct list_head w_unwr 786 struct list_head w_unwritten_list;
771 unsigned int w_unwr <<
772 }; 787 };
773 788
774 void ocfs2_unlock_and_free_pages(struct page * 789 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
775 { 790 {
776 int i; 791 int i;
777 792
778 for(i = 0; i < num_pages; i++) { 793 for(i = 0; i < num_pages; i++) {
779 if (pages[i]) { 794 if (pages[i]) {
780 unlock_page(pages[i]); 795 unlock_page(pages[i]);
781 mark_page_accessed(pag 796 mark_page_accessed(pages[i]);
782 put_page(pages[i]); 797 put_page(pages[i]);
783 } 798 }
784 } 799 }
785 } 800 }
786 801
787 static void ocfs2_unlock_pages(struct ocfs2_wr 802 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
788 { 803 {
789 int i; 804 int i;
790 805
791 /* 806 /*
792 * w_target_locked is only set to true 807 * w_target_locked is only set to true in the page_mkwrite() case.
793 * The intent is to allow us to lock t 808 * The intent is to allow us to lock the target page from write_begin()
794 * to write_end(). The caller must hol 809 * to write_end(). The caller must hold a ref on w_target_page.
795 */ 810 */
796 if (wc->w_target_locked) { 811 if (wc->w_target_locked) {
797 BUG_ON(!wc->w_target_page); 812 BUG_ON(!wc->w_target_page);
798 for (i = 0; i < wc->w_num_page 813 for (i = 0; i < wc->w_num_pages; i++) {
799 if (wc->w_target_page 814 if (wc->w_target_page == wc->w_pages[i]) {
800 wc->w_pages[i] 815 wc->w_pages[i] = NULL;
801 break; 816 break;
802 } 817 }
803 } 818 }
804 mark_page_accessed(wc->w_targe 819 mark_page_accessed(wc->w_target_page);
805 put_page(wc->w_target_page); 820 put_page(wc->w_target_page);
806 } 821 }
807 ocfs2_unlock_and_free_pages(wc->w_page 822 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
808 } 823 }
809 824
810 static void ocfs2_free_unwritten_list(struct i 825 static void ocfs2_free_unwritten_list(struct inode *inode,
811 struct list_h 826 struct list_head *head)
812 { 827 {
813 struct ocfs2_inode_info *oi = OCFS2_I( 828 struct ocfs2_inode_info *oi = OCFS2_I(inode);
814 struct ocfs2_unwritten_extent *ue = NU 829 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
815 830
816 list_for_each_entry_safe(ue, tmp, head 831 list_for_each_entry_safe(ue, tmp, head, ue_node) {
817 list_del(&ue->ue_node); 832 list_del(&ue->ue_node);
818 spin_lock(&oi->ip_lock); 833 spin_lock(&oi->ip_lock);
819 list_del(&ue->ue_ip_node); 834 list_del(&ue->ue_ip_node);
820 spin_unlock(&oi->ip_lock); 835 spin_unlock(&oi->ip_lock);
821 kfree(ue); 836 kfree(ue);
822 } 837 }
823 } 838 }
824 839
825 static void ocfs2_free_write_ctxt(struct inode 840 static void ocfs2_free_write_ctxt(struct inode *inode,
826 struct ocfs2 841 struct ocfs2_write_ctxt *wc)
827 { 842 {
828 ocfs2_free_unwritten_list(inode, &wc-> 843 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
829 ocfs2_unlock_pages(wc); 844 ocfs2_unlock_pages(wc);
830 brelse(wc->w_di_bh); 845 brelse(wc->w_di_bh);
831 kfree(wc); 846 kfree(wc);
832 } 847 }
833 848
834 static int ocfs2_alloc_write_ctxt(struct ocfs2 849 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
835 struct ocfs2 850 struct ocfs2_super *osb, loff_t pos,
836 unsigned len 851 unsigned len, ocfs2_write_type_t type,
837 struct buffe 852 struct buffer_head *di_bh)
838 { 853 {
839 u32 cend; 854 u32 cend;
840 struct ocfs2_write_ctxt *wc; 855 struct ocfs2_write_ctxt *wc;
841 856
842 wc = kzalloc(sizeof(struct ocfs2_write 857 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
843 if (!wc) 858 if (!wc)
844 return -ENOMEM; 859 return -ENOMEM;
845 860
846 wc->w_cpos = pos >> osb->s_clustersize 861 wc->w_cpos = pos >> osb->s_clustersize_bits;
847 wc->w_first_new_cpos = UINT_MAX; 862 wc->w_first_new_cpos = UINT_MAX;
848 cend = (pos + len - 1) >> osb->s_clust 863 cend = (pos + len - 1) >> osb->s_clustersize_bits;
849 wc->w_clen = cend - wc->w_cpos + 1; 864 wc->w_clen = cend - wc->w_cpos + 1;
850 get_bh(di_bh); 865 get_bh(di_bh);
851 wc->w_di_bh = di_bh; 866 wc->w_di_bh = di_bh;
852 wc->w_type = type; 867 wc->w_type = type;
853 868
854 if (unlikely(PAGE_SHIFT > osb->s_clust 869 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
855 wc->w_large_pages = 1; 870 wc->w_large_pages = 1;
856 else 871 else
857 wc->w_large_pages = 0; 872 wc->w_large_pages = 0;
858 873
859 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 874 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
860 INIT_LIST_HEAD(&wc->w_unwritten_list); 875 INIT_LIST_HEAD(&wc->w_unwritten_list);
861 876
862 *wcp = wc; 877 *wcp = wc;
863 878
864 return 0; 879 return 0;
865 } 880 }
866 881
867 /* 882 /*
868 * If a page has any new buffers, zero them ou 883 * 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 884 * and dirty so they'll be written out (in order to prevent uninitialised
870 * block data from leaking). And clear the new 885 * block data from leaking). And clear the new bit.
871 */ 886 */
872 static void ocfs2_zero_new_buffers(struct page 887 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
873 { 888 {
874 unsigned int block_start, block_end; 889 unsigned int block_start, block_end;
875 struct buffer_head *head, *bh; 890 struct buffer_head *head, *bh;
876 891
877 BUG_ON(!PageLocked(page)); 892 BUG_ON(!PageLocked(page));
878 if (!page_has_buffers(page)) 893 if (!page_has_buffers(page))
879 return; 894 return;
880 895
881 bh = head = page_buffers(page); 896 bh = head = page_buffers(page);
882 block_start = 0; 897 block_start = 0;
883 do { 898 do {
884 block_end = block_start + bh-> 899 block_end = block_start + bh->b_size;
885 900
886 if (buffer_new(bh)) { 901 if (buffer_new(bh)) {
887 if (block_end > from & 902 if (block_end > from && block_start < to) {
888 if (!PageUptod 903 if (!PageUptodate(page)) {
889 unsign 904 unsigned start, end;
890 905
891 start 906 start = max(from, block_start);
892 end = 907 end = min(to, block_end);
893 908
894 zero_u 909 zero_user_segment(page, start, end);
895 set_bu 910 set_buffer_uptodate(bh);
896 } 911 }
897 912
898 clear_buffer_n 913 clear_buffer_new(bh);
899 mark_buffer_di 914 mark_buffer_dirty(bh);
900 } 915 }
901 } 916 }
902 917
903 block_start = block_end; 918 block_start = block_end;
904 bh = bh->b_this_page; 919 bh = bh->b_this_page;
905 } while (bh != head); 920 } while (bh != head);
906 } 921 }
907 922
908 /* 923 /*
909 * Only called when we have a failure during a 924 * Only called when we have a failure during allocating write to write
910 * zero's to the newly allocated region. 925 * zero's to the newly allocated region.
911 */ 926 */
912 static void ocfs2_write_failure(struct inode * 927 static void ocfs2_write_failure(struct inode *inode,
913 struct ocfs2_w 928 struct ocfs2_write_ctxt *wc,
914 loff_t user_po 929 loff_t user_pos, unsigned user_len)
915 { 930 {
916 int i; 931 int i;
917 unsigned from = user_pos & (PAGE_SIZE 932 unsigned from = user_pos & (PAGE_SIZE - 1),
918 to = user_pos + user_len; 933 to = user_pos + user_len;
919 struct page *tmppage; 934 struct page *tmppage;
920 935
921 if (wc->w_target_page) 936 if (wc->w_target_page)
922 ocfs2_zero_new_buffers(wc->w_t 937 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
923 938
924 for(i = 0; i < wc->w_num_pages; i++) { 939 for(i = 0; i < wc->w_num_pages; i++) {
925 tmppage = wc->w_pages[i]; 940 tmppage = wc->w_pages[i];
926 941
927 if (tmppage && page_has_buffer 942 if (tmppage && page_has_buffers(tmppage)) {
928 if (ocfs2_should_order 943 if (ocfs2_should_order_data(inode))
929 ocfs2_jbd2_ino !! 944 ocfs2_jbd2_file_inode(wc->w_handle, inode);
930 <<
931 945
932 block_commit_write(tmp 946 block_commit_write(tmppage, from, to);
933 } 947 }
934 } 948 }
935 } 949 }
936 950
937 static int ocfs2_prepare_page_for_write(struct 951 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
938 struct 952 struct ocfs2_write_ctxt *wc,
939 struct 953 struct page *page, u32 cpos,
940 loff_t 954 loff_t user_pos, unsigned user_len,
941 int ne 955 int new)
942 { 956 {
943 int ret; 957 int ret;
944 unsigned int map_from = 0, map_to = 0; 958 unsigned int map_from = 0, map_to = 0;
945 unsigned int cluster_start, cluster_en 959 unsigned int cluster_start, cluster_end;
946 unsigned int user_data_from = 0, user_ 960 unsigned int user_data_from = 0, user_data_to = 0;
947 961
948 ocfs2_figure_cluster_boundaries(OCFS2_ 962 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
949 &clust 963 &cluster_start, &cluster_end);
950 964
951 /* treat the write as new if the a hol 965 /* treat the write as new if the a hole/lseek spanned across
952 * the page boundary. 966 * the page boundary.
953 */ 967 */
954 new = new | ((i_size_read(inode) <= pa 968 new = new | ((i_size_read(inode) <= page_offset(page)) &&
955 (page_offset(page) <= 969 (page_offset(page) <= user_pos));
956 970
957 if (page == wc->w_target_page) { 971 if (page == wc->w_target_page) {
958 map_from = user_pos & (PAGE_SI 972 map_from = user_pos & (PAGE_SIZE - 1);
959 map_to = map_from + user_len; 973 map_to = map_from + user_len;
960 974
961 if (new) 975 if (new)
962 ret = ocfs2_map_page_b 976 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
963 977 cluster_start, cluster_end,
964 978 new);
965 else 979 else
966 ret = ocfs2_map_page_b 980 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
967 981 map_from, map_to, new);
968 if (ret) { 982 if (ret) {
969 mlog_errno(ret); 983 mlog_errno(ret);
970 goto out; 984 goto out;
971 } 985 }
972 986
973 user_data_from = map_from; 987 user_data_from = map_from;
974 user_data_to = map_to; 988 user_data_to = map_to;
975 if (new) { 989 if (new) {
976 map_from = cluster_sta 990 map_from = cluster_start;
977 map_to = cluster_end; 991 map_to = cluster_end;
978 } 992 }
979 } else { 993 } else {
980 /* 994 /*
981 * If we haven't allocated the 995 * If we haven't allocated the new page yet, we
982 * shouldn't be writing it out 996 * shouldn't be writing it out without copying user
983 * data. This is likely a math 997 * data. This is likely a math error from the caller.
984 */ 998 */
985 BUG_ON(!new); 999 BUG_ON(!new);
986 1000
987 map_from = cluster_start; 1001 map_from = cluster_start;
988 map_to = cluster_end; 1002 map_to = cluster_end;
989 1003
990 ret = ocfs2_map_page_blocks(pa 1004 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
991 cl 1005 cluster_start, cluster_end, new);
992 if (ret) { 1006 if (ret) {
993 mlog_errno(ret); 1007 mlog_errno(ret);
994 goto out; 1008 goto out;
995 } 1009 }
996 } 1010 }
997 1011
998 /* 1012 /*
999 * Parts of newly allocated pages need 1013 * Parts of newly allocated pages need to be zero'd.
1000 * 1014 *
1001 * Above, we have also rewritten 'to' 1015 * Above, we have also rewritten 'to' and 'from' - as far as
1002 * the rest of the function is concer 1016 * the rest of the function is concerned, the entire cluster
1003 * range inside of a page needs to be 1017 * range inside of a page needs to be written.
1004 * 1018 *
1005 * We can skip this if the page is up 1019 * We can skip this if the page is up to date - it's already
1006 * been zero'd from being read in as 1020 * been zero'd from being read in as a hole.
1007 */ 1021 */
1008 if (new && !PageUptodate(page)) 1022 if (new && !PageUptodate(page))
1009 ocfs2_clear_page_regions(page 1023 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1010 cpos 1024 cpos, user_data_from, user_data_to);
1011 1025
1012 flush_dcache_page(page); 1026 flush_dcache_page(page);
1013 1027
1014 out: 1028 out:
1015 return ret; 1029 return ret;
1016 } 1030 }
1017 1031
1018 /* 1032 /*
1019 * This function will only grab one clusters 1033 * This function will only grab one clusters worth of pages.
1020 */ 1034 */
1021 static int ocfs2_grab_pages_for_write(struct 1035 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1022 struct 1036 struct ocfs2_write_ctxt *wc,
1023 u32 cpo 1037 u32 cpos, loff_t user_pos,
1024 unsigne 1038 unsigned user_len, int new,
1025 struct 1039 struct page *mmap_page)
1026 { 1040 {
1027 int ret = 0, i; 1041 int ret = 0, i;
1028 unsigned long start, target_index, en 1042 unsigned long start, target_index, end_index, index;
1029 struct inode *inode = mapping->host; 1043 struct inode *inode = mapping->host;
1030 loff_t last_byte; 1044 loff_t last_byte;
1031 1045
1032 target_index = user_pos >> PAGE_SHIFT 1046 target_index = user_pos >> PAGE_SHIFT;
1033 1047
1034 /* 1048 /*
1035 * Figure out how many pages we'll be 1049 * Figure out how many pages we'll be manipulating here. For
1036 * non allocating write, we just chan 1050 * non allocating write, we just change the one
1037 * page. Otherwise, we'll need a whol 1051 * page. Otherwise, we'll need a whole clusters worth. If we're
1038 * writing past i_size, we only need 1052 * writing past i_size, we only need enough pages to cover the
1039 * last page of the write. 1053 * last page of the write.
1040 */ 1054 */
1041 if (new) { 1055 if (new) {
1042 wc->w_num_pages = ocfs2_pages 1056 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1043 start = ocfs2_align_clusters_ 1057 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1044 /* 1058 /*
1045 * We need the index *past* t 1059 * We need the index *past* the last page we could possibly
1046 * touch. This is the page p 1060 * touch. This is the page past the end of the write or
1047 * i_size, whichever is great 1061 * i_size, whichever is greater.
1048 */ 1062 */
1049 last_byte = max(user_pos + us 1063 last_byte = max(user_pos + user_len, i_size_read(inode));
1050 BUG_ON(last_byte < 1); 1064 BUG_ON(last_byte < 1);
1051 end_index = ((last_byte - 1) 1065 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
1052 if ((start + wc->w_num_pages) 1066 if ((start + wc->w_num_pages) > end_index)
1053 wc->w_num_pages = end 1067 wc->w_num_pages = end_index - start;
1054 } else { 1068 } else {
1055 wc->w_num_pages = 1; 1069 wc->w_num_pages = 1;
1056 start = target_index; 1070 start = target_index;
1057 } 1071 }
1058 end_index = (user_pos + user_len - 1) 1072 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
1059 1073
1060 for(i = 0; i < wc->w_num_pages; i++) 1074 for(i = 0; i < wc->w_num_pages; i++) {
1061 index = start + i; 1075 index = start + i;
1062 1076
1063 if (index >= target_index && 1077 if (index >= target_index && index <= end_index &&
1064 wc->w_type == OCFS2_WRITE 1078 wc->w_type == OCFS2_WRITE_MMAP) {
1065 /* 1079 /*
1066 * ocfs2_pagemkwrite( 1080 * ocfs2_pagemkwrite() is a little different
1067 * and wants us to di 1081 * and wants us to directly use the page
1068 * passed in. 1082 * passed in.
1069 */ 1083 */
1070 lock_page(mmap_page); 1084 lock_page(mmap_page);
1071 1085
1072 /* Exit and let the c 1086 /* Exit and let the caller retry */
1073 if (mmap_page->mappin 1087 if (mmap_page->mapping != mapping) {
1074 WARN_ON(mmap_ 1088 WARN_ON(mmap_page->mapping);
1075 unlock_page(m 1089 unlock_page(mmap_page);
1076 ret = -EAGAIN 1090 ret = -EAGAIN;
1077 goto out; 1091 goto out;
1078 } 1092 }
1079 1093
1080 get_page(mmap_page); 1094 get_page(mmap_page);
1081 wc->w_pages[i] = mmap 1095 wc->w_pages[i] = mmap_page;
1082 wc->w_target_locked = 1096 wc->w_target_locked = true;
1083 } else if (index >= target_in 1097 } else if (index >= target_index && index <= end_index &&
1084 wc->w_type == OCFS 1098 wc->w_type == OCFS2_WRITE_DIRECT) {
1085 /* Direct write has n 1099 /* Direct write has no mapping page. */
1086 wc->w_pages[i] = NULL 1100 wc->w_pages[i] = NULL;
1087 continue; 1101 continue;
1088 } else { 1102 } else {
1089 wc->w_pages[i] = find 1103 wc->w_pages[i] = find_or_create_page(mapping, index,
1090 1104 GFP_NOFS);
1091 if (!wc->w_pages[i]) 1105 if (!wc->w_pages[i]) {
1092 ret = -ENOMEM 1106 ret = -ENOMEM;
1093 mlog_errno(re 1107 mlog_errno(ret);
1094 goto out; 1108 goto out;
1095 } 1109 }
1096 } 1110 }
1097 wait_for_stable_page(wc->w_pa 1111 wait_for_stable_page(wc->w_pages[i]);
1098 1112
1099 if (index == target_index) 1113 if (index == target_index)
1100 wc->w_target_page = w 1114 wc->w_target_page = wc->w_pages[i];
1101 } 1115 }
1102 out: 1116 out:
1103 if (ret) 1117 if (ret)
1104 wc->w_target_locked = false; 1118 wc->w_target_locked = false;
1105 return ret; 1119 return ret;
1106 } 1120 }
1107 1121
1108 /* 1122 /*
1109 * Prepare a single cluster for write one clu 1123 * Prepare a single cluster for write one cluster into the file.
1110 */ 1124 */
1111 static int ocfs2_write_cluster(struct address 1125 static int ocfs2_write_cluster(struct address_space *mapping,
1112 u32 *phys, uns 1126 u32 *phys, unsigned int new,
1113 unsigned int c 1127 unsigned int clear_unwritten,
1114 unsigned int s 1128 unsigned int should_zero,
1115 struct ocfs2_a 1129 struct ocfs2_alloc_context *data_ac,
1116 struct ocfs2_a 1130 struct ocfs2_alloc_context *meta_ac,
1117 struct ocfs2_w 1131 struct ocfs2_write_ctxt *wc, u32 cpos,
1118 loff_t user_po 1132 loff_t user_pos, unsigned user_len)
1119 { 1133 {
1120 int ret, i; 1134 int ret, i;
1121 u64 p_blkno; 1135 u64 p_blkno;
1122 struct inode *inode = mapping->host; 1136 struct inode *inode = mapping->host;
1123 struct ocfs2_extent_tree et; 1137 struct ocfs2_extent_tree et;
1124 int bpc = ocfs2_clusters_to_blocks(in 1138 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1125 1139
1126 if (new) { 1140 if (new) {
1127 u32 tmp_pos; 1141 u32 tmp_pos;
1128 1142
1129 /* 1143 /*
1130 * This is safe to call with 1144 * This is safe to call with the page locks - it won't take
1131 * any additional semaphores 1145 * any additional semaphores or cluster locks.
1132 */ 1146 */
1133 tmp_pos = cpos; 1147 tmp_pos = cpos;
1134 ret = ocfs2_add_inode_data(OC 1148 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1135 &t 1149 &tmp_pos, 1, !clear_unwritten,
1136 wc 1150 wc->w_di_bh, wc->w_handle,
1137 da 1151 data_ac, meta_ac, NULL);
1138 /* 1152 /*
1139 * This shouldn't happen beca 1153 * This shouldn't happen because we must have already
1140 * calculated the correct met 1154 * calculated the correct meta data allocation required. The
1141 * internal tree allocation c 1155 * internal tree allocation code should know how to increase
1142 * transaction credits itself 1156 * transaction credits itself.
1143 * 1157 *
1144 * If need be, we could handl 1158 * If need be, we could handle -EAGAIN for a
1145 * RESTART_TRANS here. 1159 * RESTART_TRANS here.
1146 */ 1160 */
1147 mlog_bug_on_msg(ret == -EAGAI 1161 mlog_bug_on_msg(ret == -EAGAIN,
1148 "Inode %llu: 1162 "Inode %llu: EAGAIN return during allocation.\n",
1149 (unsigned lon 1163 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1150 if (ret < 0) { 1164 if (ret < 0) {
1151 mlog_errno(ret); 1165 mlog_errno(ret);
1152 goto out; 1166 goto out;
1153 } 1167 }
1154 } else if (clear_unwritten) { 1168 } else if (clear_unwritten) {
1155 ocfs2_init_dinode_extent_tree 1169 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1156 1170 wc->w_di_bh);
1157 ret = ocfs2_mark_extent_writt 1171 ret = ocfs2_mark_extent_written(inode, &et,
1158 1172 wc->w_handle, cpos, 1, *phys,
1159 1173 meta_ac, &wc->w_dealloc);
1160 if (ret < 0) { 1174 if (ret < 0) {
1161 mlog_errno(ret); 1175 mlog_errno(ret);
1162 goto out; 1176 goto out;
1163 } 1177 }
1164 } 1178 }
1165 1179
1166 /* 1180 /*
1167 * The only reason this should fail i 1181 * The only reason this should fail is due to an inability to
1168 * find the extent added. 1182 * find the extent added.
1169 */ 1183 */
1170 ret = ocfs2_get_clusters(inode, cpos, 1184 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
1171 if (ret < 0) { 1185 if (ret < 0) {
1172 mlog(ML_ERROR, "Get physical 1186 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1173 "at logical clust 1187 "at logical cluster %u",
1174 (unsigned long lo 1188 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
1175 goto out; 1189 goto out;
1176 } 1190 }
1177 1191
1178 BUG_ON(*phys == 0); 1192 BUG_ON(*phys == 0);
1179 1193
1180 p_blkno = ocfs2_clusters_to_blocks(in 1194 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1181 if (!should_zero) 1195 if (!should_zero)
1182 p_blkno += (user_pos >> inode 1196 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
1183 1197
1184 for(i = 0; i < wc->w_num_pages; i++) 1198 for(i = 0; i < wc->w_num_pages; i++) {
1185 int tmpret; 1199 int tmpret;
1186 1200
1187 /* This is the direct io targ 1201 /* This is the direct io target page. */
1188 if (wc->w_pages[i] == NULL) { 1202 if (wc->w_pages[i] == NULL) {
1189 p_blkno++; 1203 p_blkno++;
1190 continue; 1204 continue;
1191 } 1205 }
1192 1206
1193 tmpret = ocfs2_prepare_page_f 1207 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1194 1208 wc->w_pages[i], cpos,
1195 1209 user_pos, user_len,
1196 1210 should_zero);
1197 if (tmpret) { 1211 if (tmpret) {
1198 mlog_errno(tmpret); 1212 mlog_errno(tmpret);
1199 if (ret == 0) 1213 if (ret == 0)
1200 ret = tmpret; 1214 ret = tmpret;
1201 } 1215 }
1202 } 1216 }
1203 1217
1204 /* 1218 /*
1205 * We only have cleanup to do in case 1219 * We only have cleanup to do in case of allocating write.
1206 */ 1220 */
1207 if (ret && new) 1221 if (ret && new)
1208 ocfs2_write_failure(inode, wc 1222 ocfs2_write_failure(inode, wc, user_pos, user_len);
1209 1223
1210 out: 1224 out:
1211 1225
1212 return ret; 1226 return ret;
1213 } 1227 }
1214 1228
1215 static int ocfs2_write_cluster_by_desc(struct 1229 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1216 struct 1230 struct ocfs2_alloc_context *data_ac,
1217 struct 1231 struct ocfs2_alloc_context *meta_ac,
1218 struct 1232 struct ocfs2_write_ctxt *wc,
1219 loff_t 1233 loff_t pos, unsigned len)
1220 { 1234 {
1221 int ret, i; 1235 int ret, i;
1222 loff_t cluster_off; 1236 loff_t cluster_off;
1223 unsigned int local_len = len; 1237 unsigned int local_len = len;
1224 struct ocfs2_write_cluster_desc *desc 1238 struct ocfs2_write_cluster_desc *desc;
1225 struct ocfs2_super *osb = OCFS2_SB(ma 1239 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1226 1240
1227 for (i = 0; i < wc->w_clen; i++) { 1241 for (i = 0; i < wc->w_clen; i++) {
1228 desc = &wc->w_desc[i]; 1242 desc = &wc->w_desc[i];
1229 1243
1230 /* 1244 /*
1231 * We have to make sure that 1245 * We have to make sure that the total write passed in
1232 * doesn't extend past a sing 1246 * doesn't extend past a single cluster.
1233 */ 1247 */
1234 local_len = len; 1248 local_len = len;
1235 cluster_off = pos & (osb->s_c 1249 cluster_off = pos & (osb->s_clustersize - 1);
1236 if ((cluster_off + local_len) 1250 if ((cluster_off + local_len) > osb->s_clustersize)
1237 local_len = osb->s_cl 1251 local_len = osb->s_clustersize - cluster_off;
1238 1252
1239 ret = ocfs2_write_cluster(map 1253 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
1240 des 1254 desc->c_new,
1241 des 1255 desc->c_clear_unwritten,
1242 des 1256 desc->c_needs_zero,
1243 dat 1257 data_ac, meta_ac,
1244 wc, 1258 wc, desc->c_cpos, pos, local_len);
1245 if (ret) { 1259 if (ret) {
1246 mlog_errno(ret); 1260 mlog_errno(ret);
1247 goto out; 1261 goto out;
1248 } 1262 }
1249 1263
1250 len -= local_len; 1264 len -= local_len;
1251 pos += local_len; 1265 pos += local_len;
1252 } 1266 }
1253 1267
1254 ret = 0; 1268 ret = 0;
1255 out: 1269 out:
1256 return ret; 1270 return ret;
1257 } 1271 }
1258 1272
1259 /* 1273 /*
1260 * ocfs2_write_end() wants to know which part 1274 * ocfs2_write_end() wants to know which parts of the target page it
1261 * should complete the write on. It's easiest 1275 * should complete the write on. It's easiest to compute them ahead of
1262 * time when a more complete view of the writ 1276 * time when a more complete view of the write is available.
1263 */ 1277 */
1264 static void ocfs2_set_target_boundaries(struc 1278 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1265 struc 1279 struct ocfs2_write_ctxt *wc,
1266 loff_ 1280 loff_t pos, unsigned len, int alloc)
1267 { 1281 {
1268 struct ocfs2_write_cluster_desc *desc 1282 struct ocfs2_write_cluster_desc *desc;
1269 1283
1270 wc->w_target_from = pos & (PAGE_SIZE 1284 wc->w_target_from = pos & (PAGE_SIZE - 1);
1271 wc->w_target_to = wc->w_target_from + 1285 wc->w_target_to = wc->w_target_from + len;
1272 1286
1273 if (alloc == 0) 1287 if (alloc == 0)
1274 return; 1288 return;
1275 1289
1276 /* 1290 /*
1277 * Allocating write - we may have dif 1291 * Allocating write - we may have different boundaries based
1278 * on page size and cluster size. 1292 * on page size and cluster size.
1279 * 1293 *
1280 * NOTE: We can no longer compute one 1294 * NOTE: We can no longer compute one value from the other as
1281 * the actual write length and user p 1295 * the actual write length and user provided length may be
1282 * different. 1296 * different.
1283 */ 1297 */
1284 1298
1285 if (wc->w_large_pages) { 1299 if (wc->w_large_pages) {
1286 /* 1300 /*
1287 * We only care about the 1st 1301 * We only care about the 1st and last cluster within
1288 * our range and whether they 1302 * our range and whether they should be zero'd or not. Either
1289 * value may be extended out 1303 * value may be extended out to the start/end of a
1290 * newly allocated cluster. 1304 * newly allocated cluster.
1291 */ 1305 */
1292 desc = &wc->w_desc[0]; 1306 desc = &wc->w_desc[0];
1293 if (desc->c_needs_zero) 1307 if (desc->c_needs_zero)
1294 ocfs2_figure_cluster_ 1308 ocfs2_figure_cluster_boundaries(osb,
1295 1309 desc->c_cpos,
1296 1310 &wc->w_target_from,
1297 1311 NULL);
1298 1312
1299 desc = &wc->w_desc[wc->w_clen 1313 desc = &wc->w_desc[wc->w_clen - 1];
1300 if (desc->c_needs_zero) 1314 if (desc->c_needs_zero)
1301 ocfs2_figure_cluster_ 1315 ocfs2_figure_cluster_boundaries(osb,
1302 1316 desc->c_cpos,
1303 1317 NULL,
1304 1318 &wc->w_target_to);
1305 } else { 1319 } else {
1306 wc->w_target_from = 0; 1320 wc->w_target_from = 0;
1307 wc->w_target_to = PAGE_SIZE; 1321 wc->w_target_to = PAGE_SIZE;
1308 } 1322 }
1309 } 1323 }
1310 1324
1311 /* 1325 /*
1312 * Check if this extent is marked UNWRITTEN b 1326 * 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 1327 * do the zero work. And should not to clear UNWRITTEN since it will be cleared
1314 * by the direct io procedure. 1328 * by the direct io procedure.
1315 * If this is a new extent that allocated by 1329 * If this is a new extent that allocated by direct io, we should mark it in
1316 * the ip_unwritten_list. 1330 * the ip_unwritten_list.
1317 */ 1331 */
1318 static int ocfs2_unwritten_check(struct inode 1332 static int ocfs2_unwritten_check(struct inode *inode,
1319 struct ocfs2 1333 struct ocfs2_write_ctxt *wc,
1320 struct ocfs2 1334 struct ocfs2_write_cluster_desc *desc)
1321 { 1335 {
1322 struct ocfs2_inode_info *oi = OCFS2_I 1336 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1323 struct ocfs2_unwritten_extent *ue = N 1337 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
1324 int ret = 0; 1338 int ret = 0;
1325 1339
1326 if (!desc->c_needs_zero) 1340 if (!desc->c_needs_zero)
1327 return 0; 1341 return 0;
1328 1342
1329 retry: 1343 retry:
1330 spin_lock(&oi->ip_lock); 1344 spin_lock(&oi->ip_lock);
1331 /* Needs not to zero no metter buffer 1345 /* Needs not to zero no metter buffer or direct. The one who is zero
1332 * the cluster is doing zero. And he 1346 * the cluster is doing zero. And he will clear unwritten after all
1333 * cluster io finished. */ 1347 * cluster io finished. */
1334 list_for_each_entry(ue, &oi->ip_unwri 1348 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1335 if (desc->c_cpos == ue->ue_cp 1349 if (desc->c_cpos == ue->ue_cpos) {
1336 BUG_ON(desc->c_new); 1350 BUG_ON(desc->c_new);
1337 desc->c_needs_zero = 1351 desc->c_needs_zero = 0;
1338 desc->c_clear_unwritt 1352 desc->c_clear_unwritten = 0;
1339 goto unlock; 1353 goto unlock;
1340 } 1354 }
1341 } 1355 }
1342 1356
1343 if (wc->w_type != OCFS2_WRITE_DIRECT) 1357 if (wc->w_type != OCFS2_WRITE_DIRECT)
1344 goto unlock; 1358 goto unlock;
1345 1359
1346 if (new == NULL) { 1360 if (new == NULL) {
1347 spin_unlock(&oi->ip_lock); 1361 spin_unlock(&oi->ip_lock);
1348 new = kmalloc(sizeof(struct o 1362 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1349 GFP_NOFS); 1363 GFP_NOFS);
1350 if (new == NULL) { 1364 if (new == NULL) {
1351 ret = -ENOMEM; 1365 ret = -ENOMEM;
1352 goto out; 1366 goto out;
1353 } 1367 }
1354 goto retry; 1368 goto retry;
1355 } 1369 }
1356 /* This direct write will doing zero. 1370 /* This direct write will doing zero. */
1357 new->ue_cpos = desc->c_cpos; 1371 new->ue_cpos = desc->c_cpos;
1358 new->ue_phys = desc->c_phys; 1372 new->ue_phys = desc->c_phys;
1359 desc->c_clear_unwritten = 0; 1373 desc->c_clear_unwritten = 0;
1360 list_add_tail(&new->ue_ip_node, &oi-> 1374 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1361 list_add_tail(&new->ue_node, &wc->w_u 1375 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
1362 wc->w_unwritten_count++; <<
1363 new = NULL; 1376 new = NULL;
1364 unlock: 1377 unlock:
1365 spin_unlock(&oi->ip_lock); 1378 spin_unlock(&oi->ip_lock);
1366 out: 1379 out:
1367 kfree(new); !! 1380 if (new)
>> 1381 kfree(new);
1368 return ret; 1382 return ret;
1369 } 1383 }
1370 1384
1371 /* 1385 /*
1372 * Populate each single-cluster write descrip 1386 * Populate each single-cluster write descriptor in the write context
1373 * with information about the i/o to be done. 1387 * with information about the i/o to be done.
1374 * 1388 *
1375 * Returns the number of clusters that will h 1389 * Returns the number of clusters that will have to be allocated, as
1376 * well as a worst case estimate of the numbe 1390 * well as a worst case estimate of the number of extent records that
1377 * would have to be created during a write to 1391 * would have to be created during a write to an unwritten region.
1378 */ 1392 */
1379 static int ocfs2_populate_write_desc(struct i 1393 static int ocfs2_populate_write_desc(struct inode *inode,
1380 struct o 1394 struct ocfs2_write_ctxt *wc,
1381 unsigned 1395 unsigned int *clusters_to_alloc,
1382 unsigned 1396 unsigned int *extents_to_split)
1383 { 1397 {
1384 int ret; 1398 int ret;
1385 struct ocfs2_write_cluster_desc *desc 1399 struct ocfs2_write_cluster_desc *desc;
1386 unsigned int num_clusters = 0; 1400 unsigned int num_clusters = 0;
1387 unsigned int ext_flags = 0; 1401 unsigned int ext_flags = 0;
1388 u32 phys = 0; 1402 u32 phys = 0;
1389 int i; 1403 int i;
1390 1404
1391 *clusters_to_alloc = 0; 1405 *clusters_to_alloc = 0;
1392 *extents_to_split = 0; 1406 *extents_to_split = 0;
1393 1407
1394 for (i = 0; i < wc->w_clen; i++) { 1408 for (i = 0; i < wc->w_clen; i++) {
1395 desc = &wc->w_desc[i]; 1409 desc = &wc->w_desc[i];
1396 desc->c_cpos = wc->w_cpos + i 1410 desc->c_cpos = wc->w_cpos + i;
1397 1411
1398 if (num_clusters == 0) { 1412 if (num_clusters == 0) {
1399 /* 1413 /*
1400 * Need to look up th 1414 * Need to look up the next extent record.
1401 */ 1415 */
1402 ret = ocfs2_get_clust 1416 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1403 1417 &num_clusters, &ext_flags);
1404 if (ret) { 1418 if (ret) {
1405 mlog_errno(re 1419 mlog_errno(ret);
1406 goto out; 1420 goto out;
1407 } 1421 }
1408 1422
1409 /* We should already 1423 /* We should already CoW the refcountd extent. */
1410 BUG_ON(ext_flags & OC 1424 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1411 1425
1412 /* 1426 /*
1413 * Assume worst case 1427 * Assume worst case - that we're writing in
1414 * the middle of the 1428 * the middle of the extent.
1415 * 1429 *
1416 * We can assume that 1430 * We can assume that the write proceeds from
1417 * left to right, in 1431 * left to right, in which case the extent
1418 * insert code is sma 1432 * insert code is smart enough to coalesce the
1419 * next splits into t 1433 * next splits into the previous records created.
1420 */ 1434 */
1421 if (ext_flags & OCFS2 1435 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1422 *extents_to_s 1436 *extents_to_split = *extents_to_split + 2;
1423 } else if (phys) { 1437 } else if (phys) {
1424 /* 1438 /*
1425 * Only increment phy 1439 * Only increment phys if it doesn't describe
1426 * a hole. 1440 * a hole.
1427 */ 1441 */
1428 phys++; 1442 phys++;
1429 } 1443 }
1430 1444
1431 /* 1445 /*
1432 * If w_first_new_cpos is < U 1446 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1433 * file that got extended. w 1447 * file that got extended. w_first_new_cpos tells us
1434 * where the newly allocated 1448 * where the newly allocated clusters are so we can
1435 * zero them. 1449 * zero them.
1436 */ 1450 */
1437 if (desc->c_cpos >= wc->w_fir 1451 if (desc->c_cpos >= wc->w_first_new_cpos) {
1438 BUG_ON(phys == 0); 1452 BUG_ON(phys == 0);
1439 desc->c_needs_zero = 1453 desc->c_needs_zero = 1;
1440 } 1454 }
1441 1455
1442 desc->c_phys = phys; 1456 desc->c_phys = phys;
1443 if (phys == 0) { 1457 if (phys == 0) {
1444 desc->c_new = 1; 1458 desc->c_new = 1;
1445 desc->c_needs_zero = 1459 desc->c_needs_zero = 1;
1446 desc->c_clear_unwritt 1460 desc->c_clear_unwritten = 1;
1447 *clusters_to_alloc = 1461 *clusters_to_alloc = *clusters_to_alloc + 1;
1448 } 1462 }
1449 1463
1450 if (ext_flags & OCFS2_EXT_UNW 1464 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1451 desc->c_clear_unwritt 1465 desc->c_clear_unwritten = 1;
1452 desc->c_needs_zero = 1466 desc->c_needs_zero = 1;
1453 } 1467 }
1454 1468
1455 ret = ocfs2_unwritten_check(i 1469 ret = ocfs2_unwritten_check(inode, wc, desc);
1456 if (ret) { 1470 if (ret) {
1457 mlog_errno(ret); 1471 mlog_errno(ret);
1458 goto out; 1472 goto out;
1459 } 1473 }
1460 1474
1461 num_clusters--; 1475 num_clusters--;
1462 } 1476 }
1463 1477
1464 ret = 0; 1478 ret = 0;
1465 out: 1479 out:
1466 return ret; 1480 return ret;
1467 } 1481 }
1468 1482
1469 static int ocfs2_write_begin_inline(struct ad 1483 static int ocfs2_write_begin_inline(struct address_space *mapping,
1470 struct in 1484 struct inode *inode,
1471 struct oc 1485 struct ocfs2_write_ctxt *wc)
1472 { 1486 {
1473 int ret; 1487 int ret;
1474 struct ocfs2_super *osb = OCFS2_SB(in 1488 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1475 struct page *page; 1489 struct page *page;
1476 handle_t *handle; 1490 handle_t *handle;
1477 struct ocfs2_dinode *di = (struct ocf 1491 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1478 1492
1479 handle = ocfs2_start_trans(osb, OCFS2 1493 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1480 if (IS_ERR(handle)) { 1494 if (IS_ERR(handle)) {
1481 ret = PTR_ERR(handle); 1495 ret = PTR_ERR(handle);
1482 mlog_errno(ret); 1496 mlog_errno(ret);
1483 goto out; 1497 goto out;
1484 } 1498 }
1485 1499
1486 page = find_or_create_page(mapping, 0 1500 page = find_or_create_page(mapping, 0, GFP_NOFS);
1487 if (!page) { 1501 if (!page) {
1488 ocfs2_commit_trans(osb, handl 1502 ocfs2_commit_trans(osb, handle);
1489 ret = -ENOMEM; 1503 ret = -ENOMEM;
1490 mlog_errno(ret); 1504 mlog_errno(ret);
1491 goto out; 1505 goto out;
1492 } 1506 }
1493 /* 1507 /*
1494 * If we don't set w_num_pages then t 1508 * If we don't set w_num_pages then this page won't get unlocked
1495 * and freed on cleanup of the write 1509 * and freed on cleanup of the write context.
1496 */ 1510 */
1497 wc->w_pages[0] = wc->w_target_page = 1511 wc->w_pages[0] = wc->w_target_page = page;
1498 wc->w_num_pages = 1; 1512 wc->w_num_pages = 1;
1499 1513
1500 ret = ocfs2_journal_access_di(handle, 1514 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1501 OCFS2_J 1515 OCFS2_JOURNAL_ACCESS_WRITE);
1502 if (ret) { 1516 if (ret) {
1503 ocfs2_commit_trans(osb, handl 1517 ocfs2_commit_trans(osb, handle);
1504 1518
1505 mlog_errno(ret); 1519 mlog_errno(ret);
1506 goto out; 1520 goto out;
1507 } 1521 }
1508 1522
1509 if (!(OCFS2_I(inode)->ip_dyn_features 1523 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1510 ocfs2_set_inode_data_inline(i 1524 ocfs2_set_inode_data_inline(inode, di);
1511 1525
1512 if (!PageUptodate(page)) { 1526 if (!PageUptodate(page)) {
1513 ret = ocfs2_read_inline_data( 1527 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1514 if (ret) { 1528 if (ret) {
1515 ocfs2_commit_trans(os 1529 ocfs2_commit_trans(osb, handle);
1516 1530
1517 goto out; 1531 goto out;
1518 } 1532 }
1519 } 1533 }
1520 1534
1521 wc->w_handle = handle; 1535 wc->w_handle = handle;
1522 out: 1536 out:
1523 return ret; 1537 return ret;
1524 } 1538 }
1525 1539
1526 int ocfs2_size_fits_inline_data(struct buffer 1540 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1527 { 1541 {
1528 struct ocfs2_dinode *di = (struct ocf 1542 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1529 1543
1530 if (new_size <= le16_to_cpu(di->id2.i 1544 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1531 return 1; 1545 return 1;
1532 return 0; 1546 return 0;
1533 } 1547 }
1534 1548
1535 static int ocfs2_try_to_write_inline_data(str 1549 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1536 str 1550 struct inode *inode, loff_t pos,
1537 uns 1551 unsigned len, struct page *mmap_page,
1538 str 1552 struct ocfs2_write_ctxt *wc)
1539 { 1553 {
1540 int ret, written = 0; 1554 int ret, written = 0;
1541 loff_t end = pos + len; 1555 loff_t end = pos + len;
1542 struct ocfs2_inode_info *oi = OCFS2_I 1556 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1543 struct ocfs2_dinode *di = NULL; 1557 struct ocfs2_dinode *di = NULL;
1544 1558
1545 trace_ocfs2_try_to_write_inline_data( 1559 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1546 1560 len, (unsigned long long)pos,
1547 1561 oi->ip_dyn_features);
1548 1562
1549 /* 1563 /*
1550 * Handle inodes which already have i 1564 * Handle inodes which already have inline data 1st.
1551 */ 1565 */
1552 if (oi->ip_dyn_features & OCFS2_INLIN 1566 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1553 if (mmap_page == NULL && 1567 if (mmap_page == NULL &&
1554 ocfs2_size_fits_inline_da 1568 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1555 goto do_inline_write; 1569 goto do_inline_write;
1556 1570
1557 /* 1571 /*
1558 * The write won't fit - we h 1572 * The write won't fit - we have to give this inode an
1559 * inline extent list now. 1573 * inline extent list now.
1560 */ 1574 */
1561 ret = ocfs2_convert_inline_da 1575 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1562 if (ret) 1576 if (ret)
1563 mlog_errno(ret); 1577 mlog_errno(ret);
1564 goto out; 1578 goto out;
1565 } 1579 }
1566 1580
1567 /* 1581 /*
1568 * Check whether the inode can accept 1582 * Check whether the inode can accept inline data.
1569 */ 1583 */
1570 if (oi->ip_clusters != 0 || i_size_re 1584 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1571 return 0; 1585 return 0;
1572 1586
1573 /* 1587 /*
1574 * Check whether the write can fit. 1588 * Check whether the write can fit.
1575 */ 1589 */
1576 di = (struct ocfs2_dinode *)wc->w_di_ 1590 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1577 if (mmap_page || 1591 if (mmap_page ||
1578 end > ocfs2_max_inline_data_with_ 1592 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1579 return 0; 1593 return 0;
1580 1594
1581 do_inline_write: 1595 do_inline_write:
1582 ret = ocfs2_write_begin_inline(mappin 1596 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1583 if (ret) { 1597 if (ret) {
1584 mlog_errno(ret); 1598 mlog_errno(ret);
1585 goto out; 1599 goto out;
1586 } 1600 }
1587 1601
1588 /* 1602 /*
1589 * This signals to the caller that th 1603 * This signals to the caller that the data can be written
1590 * inline. 1604 * inline.
1591 */ 1605 */
1592 written = 1; 1606 written = 1;
1593 out: 1607 out:
1594 return written ? written : ret; 1608 return written ? written : ret;
1595 } 1609 }
1596 1610
1597 /* 1611 /*
1598 * This function only does anything for file 1612 * This function only does anything for file systems which can't
1599 * handle sparse files. 1613 * handle sparse files.
1600 * 1614 *
1601 * What we want to do here is fill in any hol 1615 * 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 1616 * of allocation and the end of our write. That way the rest of the
1603 * write path can treat it as an non-allocati 1617 * write path can treat it as an non-allocating write, which has no
1604 * special case code for sparse/nonsparse fil 1618 * special case code for sparse/nonsparse files.
1605 */ 1619 */
1606 static int ocfs2_expand_nonsparse_inode(struc 1620 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1607 struc 1621 struct buffer_head *di_bh,
1608 loff_ 1622 loff_t pos, unsigned len,
1609 struc 1623 struct ocfs2_write_ctxt *wc)
1610 { 1624 {
1611 int ret; 1625 int ret;
1612 loff_t newsize = pos + len; 1626 loff_t newsize = pos + len;
1613 1627
1614 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1628 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1615 1629
1616 if (newsize <= i_size_read(inode)) 1630 if (newsize <= i_size_read(inode))
1617 return 0; 1631 return 0;
1618 1632
1619 ret = ocfs2_extend_no_holes(inode, di 1633 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1620 if (ret) 1634 if (ret)
1621 mlog_errno(ret); 1635 mlog_errno(ret);
1622 1636
1623 /* There is no wc if this is call fro 1637 /* There is no wc if this is call from direct. */
1624 if (wc) 1638 if (wc)
1625 wc->w_first_new_cpos = 1639 wc->w_first_new_cpos =
1626 ocfs2_clusters_for_by 1640 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1627 1641
1628 return ret; 1642 return ret;
1629 } 1643 }
1630 1644
1631 static int ocfs2_zero_tail(struct inode *inod 1645 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1632 loff_t pos) 1646 loff_t pos)
1633 { 1647 {
1634 int ret = 0; 1648 int ret = 0;
1635 1649
1636 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1650 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1637 if (pos > i_size_read(inode)) 1651 if (pos > i_size_read(inode))
1638 ret = ocfs2_zero_extend(inode 1652 ret = ocfs2_zero_extend(inode, di_bh, pos);
1639 1653
1640 return ret; 1654 return ret;
1641 } 1655 }
1642 1656
1643 int ocfs2_write_begin_nolock(struct address_s 1657 int ocfs2_write_begin_nolock(struct address_space *mapping,
1644 loff_t pos, unsi 1658 loff_t pos, unsigned len, ocfs2_write_type_t type,
1645 struct page **pa 1659 struct page **pagep, void **fsdata,
1646 struct buffer_he 1660 struct buffer_head *di_bh, struct page *mmap_page)
1647 { 1661 {
1648 int ret, cluster_of_pages, credits = 1662 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1649 unsigned int clusters_to_alloc, exten 1663 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1650 struct ocfs2_write_ctxt *wc; 1664 struct ocfs2_write_ctxt *wc;
1651 struct inode *inode = mapping->host; 1665 struct inode *inode = mapping->host;
1652 struct ocfs2_super *osb = OCFS2_SB(in 1666 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1653 struct ocfs2_dinode *di; 1667 struct ocfs2_dinode *di;
1654 struct ocfs2_alloc_context *data_ac = 1668 struct ocfs2_alloc_context *data_ac = NULL;
1655 struct ocfs2_alloc_context *meta_ac = 1669 struct ocfs2_alloc_context *meta_ac = NULL;
1656 handle_t *handle; 1670 handle_t *handle;
1657 struct ocfs2_extent_tree et; 1671 struct ocfs2_extent_tree et;
1658 int try_free = 1, ret1; 1672 int try_free = 1, ret1;
1659 1673
1660 try_again: 1674 try_again:
1661 ret = ocfs2_alloc_write_ctxt(&wc, osb 1675 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
1662 if (ret) { 1676 if (ret) {
1663 mlog_errno(ret); 1677 mlog_errno(ret);
1664 return ret; 1678 return ret;
1665 } 1679 }
1666 1680
1667 if (ocfs2_supports_inline_data(osb)) 1681 if (ocfs2_supports_inline_data(osb)) {
1668 ret = ocfs2_try_to_write_inli 1682 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1669 1683 mmap_page, wc);
1670 if (ret == 1) { 1684 if (ret == 1) {
1671 ret = 0; 1685 ret = 0;
1672 goto success; 1686 goto success;
1673 } 1687 }
1674 if (ret < 0) { 1688 if (ret < 0) {
1675 mlog_errno(ret); 1689 mlog_errno(ret);
1676 goto out; 1690 goto out;
1677 } 1691 }
1678 } 1692 }
1679 1693
1680 /* Direct io change i_size late, shou 1694 /* Direct io change i_size late, should not zero tail here. */
1681 if (type != OCFS2_WRITE_DIRECT) { 1695 if (type != OCFS2_WRITE_DIRECT) {
1682 if (ocfs2_sparse_alloc(osb)) 1696 if (ocfs2_sparse_alloc(osb))
1683 ret = ocfs2_zero_tail 1697 ret = ocfs2_zero_tail(inode, di_bh, pos);
1684 else 1698 else
1685 ret = ocfs2_expand_no 1699 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1686 1700 len, wc);
1687 if (ret) { 1701 if (ret) {
1688 mlog_errno(ret); 1702 mlog_errno(ret);
1689 goto out; 1703 goto out;
1690 } 1704 }
1691 } 1705 }
1692 1706
1693 ret = ocfs2_check_range_for_refcount( 1707 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1694 if (ret < 0) { 1708 if (ret < 0) {
1695 mlog_errno(ret); 1709 mlog_errno(ret);
1696 goto out; 1710 goto out;
1697 } else if (ret == 1) { 1711 } else if (ret == 1) {
1698 clusters_need = wc->w_clen; 1712 clusters_need = wc->w_clen;
1699 ret = ocfs2_refcount_cow(inod 1713 ret = ocfs2_refcount_cow(inode, di_bh,
1700 wc-> 1714 wc->w_cpos, wc->w_clen, UINT_MAX);
1701 if (ret) { 1715 if (ret) {
1702 mlog_errno(ret); 1716 mlog_errno(ret);
1703 goto out; 1717 goto out;
1704 } 1718 }
1705 } 1719 }
1706 1720
1707 ret = ocfs2_populate_write_desc(inode 1721 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1708 &exte 1722 &extents_to_split);
1709 if (ret) { 1723 if (ret) {
1710 mlog_errno(ret); 1724 mlog_errno(ret);
1711 goto out; 1725 goto out;
1712 } 1726 }
1713 clusters_need += clusters_to_alloc; 1727 clusters_need += clusters_to_alloc;
1714 1728
1715 di = (struct ocfs2_dinode *)wc->w_di_ 1729 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1716 1730
1717 trace_ocfs2_write_begin_nolock( 1731 trace_ocfs2_write_begin_nolock(
1718 (unsigned long long)O 1732 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1719 (long long)i_size_rea 1733 (long long)i_size_read(inode),
1720 le32_to_cpu(di->i_clu 1734 le32_to_cpu(di->i_clusters),
1721 pos, len, type, mmap_ 1735 pos, len, type, mmap_page,
1722 clusters_to_alloc, ex 1736 clusters_to_alloc, extents_to_split);
1723 1737
1724 /* 1738 /*
1725 * We set w_target_from, w_target_to 1739 * We set w_target_from, w_target_to here so that
1726 * ocfs2_write_end() knows which rang 1740 * ocfs2_write_end() knows which range in the target page to
1727 * write out. An allocation requires 1741 * write out. An allocation requires that we write the entire
1728 * cluster range. 1742 * cluster range.
1729 */ 1743 */
1730 if (clusters_to_alloc || extents_to_s 1744 if (clusters_to_alloc || extents_to_split) {
1731 /* 1745 /*
1732 * XXX: We are stretching the 1746 * XXX: We are stretching the limits of
1733 * ocfs2_lock_allocators(). I 1747 * ocfs2_lock_allocators(). It greatly over-estimates
1734 * the work to be done. 1748 * the work to be done.
1735 */ 1749 */
1736 ocfs2_init_dinode_extent_tree 1750 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1737 1751 wc->w_di_bh);
1738 ret = ocfs2_lock_allocators(i 1752 ret = ocfs2_lock_allocators(inode, &et,
1739 c 1753 clusters_to_alloc, extents_to_split,
1740 & 1754 &data_ac, &meta_ac);
1741 if (ret) { 1755 if (ret) {
1742 mlog_errno(ret); 1756 mlog_errno(ret);
1743 goto out; 1757 goto out;
1744 } 1758 }
1745 1759
1746 if (data_ac) 1760 if (data_ac)
1747 data_ac->ac_resv = &O 1761 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1748 1762
1749 credits = ocfs2_calc_extend_c 1763 credits = ocfs2_calc_extend_credits(inode->i_sb,
1750 1764 &di->id2.i_list);
1751 } else if (type == OCFS2_WRITE_DIRECT 1765 } else if (type == OCFS2_WRITE_DIRECT)
1752 /* direct write needs not to 1766 /* direct write needs not to start trans if no extents alloc. */
1753 goto success; 1767 goto success;
1754 1768
1755 /* 1769 /*
1756 * We have to zero sparse allocated c 1770 * We have to zero sparse allocated clusters, unwritten extent clusters,
1757 * and non-sparse clusters we just ex 1771 * and non-sparse clusters we just extended. For non-sparse writes,
1758 * we know zeros will only be needed 1772 * we know zeros will only be needed in the first and/or last cluster.
1759 */ 1773 */
1760 if (wc->w_clen && (wc->w_desc[0].c_ne 1774 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1761 wc->w_desc[wc->w_c 1775 wc->w_desc[wc->w_clen - 1].c_needs_zero))
1762 cluster_of_pages = 1; 1776 cluster_of_pages = 1;
1763 else 1777 else
1764 cluster_of_pages = 0; 1778 cluster_of_pages = 0;
1765 1779
1766 ocfs2_set_target_boundaries(osb, wc, 1780 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1767 1781
1768 handle = ocfs2_start_trans(osb, credi 1782 handle = ocfs2_start_trans(osb, credits);
1769 if (IS_ERR(handle)) { 1783 if (IS_ERR(handle)) {
1770 ret = PTR_ERR(handle); 1784 ret = PTR_ERR(handle);
1771 mlog_errno(ret); 1785 mlog_errno(ret);
1772 goto out; 1786 goto out;
1773 } 1787 }
1774 1788
1775 wc->w_handle = handle; 1789 wc->w_handle = handle;
1776 1790
1777 if (clusters_to_alloc) { 1791 if (clusters_to_alloc) {
1778 ret = dquot_alloc_space_nodir 1792 ret = dquot_alloc_space_nodirty(inode,
1779 ocfs2_clusters_to_byt 1793 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1780 if (ret) 1794 if (ret)
1781 goto out_commit; 1795 goto out_commit;
1782 } 1796 }
1783 1797
1784 ret = ocfs2_journal_access_di(handle, 1798 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1785 OCFS2_J 1799 OCFS2_JOURNAL_ACCESS_WRITE);
1786 if (ret) { 1800 if (ret) {
1787 mlog_errno(ret); 1801 mlog_errno(ret);
1788 goto out_quota; 1802 goto out_quota;
1789 } 1803 }
1790 1804
1791 /* 1805 /*
1792 * Fill our page array first. That wa 1806 * Fill our page array first. That way we've grabbed enough so
1793 * that we can zero and flush if we e 1807 * that we can zero and flush if we error after adding the
1794 * extent. 1808 * extent.
1795 */ 1809 */
1796 ret = ocfs2_grab_pages_for_write(mapp 1810 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1797 clus 1811 cluster_of_pages, mmap_page);
1798 if (ret) { !! 1812 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); 1813 mlog_errno(ret);
1812 goto out_quota; 1814 goto out_quota;
1813 } 1815 }
1814 1816
>> 1817 /*
>> 1818 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
>> 1819 * the target page. In this case, we exit with no error and no target
>> 1820 * page. This will trigger the caller, page_mkwrite(), to re-try
>> 1821 * the operation.
>> 1822 */
>> 1823 if (ret == -EAGAIN) {
>> 1824 BUG_ON(wc->w_target_page);
>> 1825 ret = 0;
>> 1826 goto out_quota;
>> 1827 }
>> 1828
1815 ret = ocfs2_write_cluster_by_desc(map 1829 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1816 len 1830 len);
1817 if (ret) { 1831 if (ret) {
1818 mlog_errno(ret); 1832 mlog_errno(ret);
1819 goto out_quota; 1833 goto out_quota;
1820 } 1834 }
1821 1835
1822 if (data_ac) 1836 if (data_ac)
1823 ocfs2_free_alloc_context(data 1837 ocfs2_free_alloc_context(data_ac);
1824 if (meta_ac) 1838 if (meta_ac)
1825 ocfs2_free_alloc_context(meta 1839 ocfs2_free_alloc_context(meta_ac);
1826 1840
1827 success: 1841 success:
1828 if (pagep) 1842 if (pagep)
1829 *pagep = wc->w_target_page; 1843 *pagep = wc->w_target_page;
1830 *fsdata = wc; 1844 *fsdata = wc;
1831 return 0; 1845 return 0;
1832 out_quota: 1846 out_quota:
1833 if (clusters_to_alloc) 1847 if (clusters_to_alloc)
1834 dquot_free_space(inode, 1848 dquot_free_space(inode,
1835 ocfs2_clusters_to_b 1849 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1836 out_commit: 1850 out_commit:
1837 ocfs2_commit_trans(osb, handle); 1851 ocfs2_commit_trans(osb, handle);
1838 1852
1839 out: 1853 out:
1840 /* 1854 /*
1841 * The mmapped page won't be unlocked 1855 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1842 * even in case of error here like EN 1856 * even in case of error here like ENOSPC and ENOMEM. So, we need
1843 * to unlock the target page manually 1857 * to unlock the target page manually to prevent deadlocks when
1844 * retrying again on ENOSPC, or when 1858 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1845 * to VM code. 1859 * to VM code.
1846 */ 1860 */
1847 if (wc->w_target_locked) 1861 if (wc->w_target_locked)
1848 unlock_page(mmap_page); 1862 unlock_page(mmap_page);
1849 1863
1850 ocfs2_free_write_ctxt(inode, wc); 1864 ocfs2_free_write_ctxt(inode, wc);
1851 1865
1852 if (data_ac) { 1866 if (data_ac) {
1853 ocfs2_free_alloc_context(data 1867 ocfs2_free_alloc_context(data_ac);
1854 data_ac = NULL; 1868 data_ac = NULL;
1855 } 1869 }
1856 if (meta_ac) { 1870 if (meta_ac) {
1857 ocfs2_free_alloc_context(meta 1871 ocfs2_free_alloc_context(meta_ac);
1858 meta_ac = NULL; 1872 meta_ac = NULL;
1859 } 1873 }
1860 1874
1861 if (ret == -ENOSPC && try_free) { 1875 if (ret == -ENOSPC && try_free) {
1862 /* 1876 /*
1863 * Try to free some truncate 1877 * Try to free some truncate log so that we can have enough
1864 * clusters to allocate. 1878 * clusters to allocate.
1865 */ 1879 */
1866 try_free = 0; 1880 try_free = 0;
1867 1881
1868 ret1 = ocfs2_try_to_free_trun 1882 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1869 if (ret1 == 1) 1883 if (ret1 == 1)
1870 goto try_again; 1884 goto try_again;
1871 1885
1872 if (ret1 < 0) 1886 if (ret1 < 0)
1873 mlog_errno(ret1); 1887 mlog_errno(ret1);
1874 } 1888 }
1875 1889
1876 return ret; 1890 return ret;
1877 } 1891 }
1878 1892
1879 static int ocfs2_write_begin(struct file *fil 1893 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1880 loff_t pos, unsi !! 1894 loff_t pos, unsigned len, unsigned flags,
1881 struct page **pa 1895 struct page **pagep, void **fsdata)
1882 { 1896 {
1883 int ret; 1897 int ret;
1884 struct buffer_head *di_bh = NULL; 1898 struct buffer_head *di_bh = NULL;
1885 struct inode *inode = mapping->host; 1899 struct inode *inode = mapping->host;
1886 1900
1887 ret = ocfs2_inode_lock(inode, &di_bh, 1901 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1888 if (ret) { 1902 if (ret) {
1889 mlog_errno(ret); 1903 mlog_errno(ret);
1890 return ret; 1904 return ret;
1891 } 1905 }
1892 1906
1893 /* 1907 /*
1894 * Take alloc sem here to prevent con 1908 * Take alloc sem here to prevent concurrent lookups. That way
1895 * the mapping, zeroing and tree mani 1909 * the mapping, zeroing and tree manipulation within
1896 * ocfs2_write() will be safe against !! 1910 * ocfs2_write() will be safe against ->readpage(). This
1897 * should also serve to lock out allo 1911 * should also serve to lock out allocation from a shared
1898 * writeable region. 1912 * writeable region.
1899 */ 1913 */
1900 down_write(&OCFS2_I(inode)->ip_alloc_ 1914 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1901 1915
1902 ret = ocfs2_write_begin_nolock(mappin 1916 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1903 pagep, 1917 pagep, fsdata, di_bh, NULL);
1904 if (ret) { 1918 if (ret) {
1905 mlog_errno(ret); 1919 mlog_errno(ret);
1906 goto out_fail; 1920 goto out_fail;
1907 } 1921 }
1908 1922
1909 brelse(di_bh); 1923 brelse(di_bh);
1910 1924
1911 return 0; 1925 return 0;
1912 1926
1913 out_fail: 1927 out_fail:
1914 up_write(&OCFS2_I(inode)->ip_alloc_se 1928 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1915 1929
1916 brelse(di_bh); 1930 brelse(di_bh);
1917 ocfs2_inode_unlock(inode, 1); 1931 ocfs2_inode_unlock(inode, 1);
1918 1932
1919 return ret; 1933 return ret;
1920 } 1934 }
1921 1935
1922 static void ocfs2_write_end_inline(struct ino 1936 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1923 unsigned l 1937 unsigned len, unsigned *copied,
1924 struct ocf 1938 struct ocfs2_dinode *di,
1925 struct ocf 1939 struct ocfs2_write_ctxt *wc)
1926 { 1940 {
1927 void *kaddr; 1941 void *kaddr;
1928 1942
1929 if (unlikely(*copied < len)) { 1943 if (unlikely(*copied < len)) {
1930 if (!PageUptodate(wc->w_targe 1944 if (!PageUptodate(wc->w_target_page)) {
1931 *copied = 0; 1945 *copied = 0;
1932 return; 1946 return;
1933 } 1947 }
1934 } 1948 }
1935 1949
1936 kaddr = kmap_atomic(wc->w_target_page 1950 kaddr = kmap_atomic(wc->w_target_page);
1937 memcpy(di->id2.i_data.id_data + pos, 1951 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1938 kunmap_atomic(kaddr); 1952 kunmap_atomic(kaddr);
1939 1953
1940 trace_ocfs2_write_end_inline( 1954 trace_ocfs2_write_end_inline(
1941 (unsigned long long)OCFS2_I(inod 1955 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1942 (unsigned long long)pos, *copied 1956 (unsigned long long)pos, *copied,
1943 le16_to_cpu(di->id2.i_data.id_co 1957 le16_to_cpu(di->id2.i_data.id_count),
1944 le16_to_cpu(di->i_dyn_features)) 1958 le16_to_cpu(di->i_dyn_features));
1945 } 1959 }
1946 1960
1947 int ocfs2_write_end_nolock(struct address_spa 1961 int ocfs2_write_end_nolock(struct address_space *mapping,
1948 loff_t pos, unsign 1962 loff_t pos, unsigned len, unsigned copied, void *fsdata)
1949 { 1963 {
1950 int i, ret; 1964 int i, ret;
1951 unsigned from, to, start = pos & (PAG 1965 unsigned from, to, start = pos & (PAGE_SIZE - 1);
1952 struct inode *inode = mapping->host; 1966 struct inode *inode = mapping->host;
1953 struct ocfs2_super *osb = OCFS2_SB(in 1967 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1954 struct ocfs2_write_ctxt *wc = fsdata; 1968 struct ocfs2_write_ctxt *wc = fsdata;
1955 struct ocfs2_dinode *di = (struct ocf 1969 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1956 handle_t *handle = wc->w_handle; 1970 handle_t *handle = wc->w_handle;
1957 struct page *tmppage; 1971 struct page *tmppage;
1958 1972
1959 BUG_ON(!list_empty(&wc->w_unwritten_l 1973 BUG_ON(!list_empty(&wc->w_unwritten_list));
1960 1974
1961 if (handle) { 1975 if (handle) {
1962 ret = ocfs2_journal_access_di 1976 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1963 wc->w_di_bh, 1977 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1964 if (ret) { 1978 if (ret) {
1965 copied = ret; 1979 copied = ret;
1966 mlog_errno(ret); 1980 mlog_errno(ret);
1967 goto out; 1981 goto out;
1968 } 1982 }
1969 } 1983 }
1970 1984
1971 if (OCFS2_I(inode)->ip_dyn_features & 1985 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1972 ocfs2_write_end_inline(inode, 1986 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1973 goto out_write_size; 1987 goto out_write_size;
1974 } 1988 }
1975 1989
1976 if (unlikely(copied < len) && wc->w_t 1990 if (unlikely(copied < len) && wc->w_target_page) {
1977 loff_t new_isize; <<
1978 <<
1979 if (!PageUptodate(wc->w_targe 1991 if (!PageUptodate(wc->w_target_page))
1980 copied = 0; 1992 copied = 0;
1981 1993
1982 new_isize = max_t(loff_t, i_s !! 1994 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1983 if (new_isize > page_offset(w !! 1995 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 } 1996 }
1998 if (wc->w_target_page) 1997 if (wc->w_target_page)
1999 flush_dcache_page(wc->w_targe 1998 flush_dcache_page(wc->w_target_page);
2000 1999
2001 for(i = 0; i < wc->w_num_pages; i++) 2000 for(i = 0; i < wc->w_num_pages; i++) {
2002 tmppage = wc->w_pages[i]; 2001 tmppage = wc->w_pages[i];
2003 2002
2004 /* This is the direct io targ 2003 /* This is the direct io target page. */
2005 if (tmppage == NULL) 2004 if (tmppage == NULL)
2006 continue; 2005 continue;
2007 2006
2008 if (tmppage == wc->w_target_p 2007 if (tmppage == wc->w_target_page) {
2009 from = wc->w_target_f 2008 from = wc->w_target_from;
2010 to = wc->w_target_to; 2009 to = wc->w_target_to;
2011 2010
2012 BUG_ON(from > PAGE_SI 2011 BUG_ON(from > PAGE_SIZE ||
2013 to > PAGE_SIZE 2012 to > PAGE_SIZE ||
2014 to < from); 2013 to < from);
2015 } else { 2014 } else {
2016 /* 2015 /*
2017 * Pages adjacent to 2016 * Pages adjacent to the target (if any) imply
2018 * a hole-filling wri 2017 * a hole-filling write in which case we want
2019 * to flush their ent 2018 * to flush their entire range.
2020 */ 2019 */
2021 from = 0; 2020 from = 0;
2022 to = PAGE_SIZE; 2021 to = PAGE_SIZE;
2023 } 2022 }
2024 2023
2025 if (page_has_buffers(tmppage) 2024 if (page_has_buffers(tmppage)) {
2026 if (handle && ocfs2_s !! 2025 if (handle && ocfs2_should_order_data(inode))
2027 loff_t start_ !! 2026 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 2027 block_commit_write(tmppage, from, to);
2035 } 2028 }
2036 } 2029 }
2037 2030
2038 out_write_size: 2031 out_write_size:
2039 /* Direct io do not update i_size her 2032 /* Direct io do not update i_size here. */
2040 if (wc->w_type != OCFS2_WRITE_DIRECT) 2033 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2041 pos += copied; 2034 pos += copied;
2042 if (pos > i_size_read(inode)) 2035 if (pos > i_size_read(inode)) {
2043 i_size_write(inode, p 2036 i_size_write(inode, pos);
2044 mark_inode_dirty(inod 2037 mark_inode_dirty(inode);
2045 } 2038 }
2046 inode->i_blocks = ocfs2_inode 2039 inode->i_blocks = ocfs2_inode_sector_count(inode);
2047 di->i_size = cpu_to_le64((u64 2040 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2048 inode_set_mtime_to_ts(inode, !! 2041 inode->i_mtime = inode->i_ctime = current_time(inode);
2049 di->i_mtime = di->i_ctime = c !! 2042 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2050 di->i_mtime_nsec = di->i_ctim !! 2043 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2051 if (handle) !! 2044 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2052 ocfs2_update_inode_fs <<
2053 } 2045 }
2054 if (handle) 2046 if (handle)
2055 ocfs2_journal_dirty(handle, w 2047 ocfs2_journal_dirty(handle, wc->w_di_bh);
2056 2048
2057 out: 2049 out:
2058 /* unlock pages before dealloc since 2050 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2059 * lock, or it will cause a deadlock 2051 * lock, or it will cause a deadlock since journal commit threads holds
2060 * this lock and will ask for the pag 2052 * this lock and will ask for the page lock when flushing the data.
2061 * put it here to preserve the unlock 2053 * put it here to preserve the unlock order.
2062 */ 2054 */
2063 ocfs2_unlock_pages(wc); 2055 ocfs2_unlock_pages(wc);
2064 2056
2065 if (handle) 2057 if (handle)
2066 ocfs2_commit_trans(osb, handl 2058 ocfs2_commit_trans(osb, handle);
2067 2059
2068 ocfs2_run_deallocs(osb, &wc->w_deallo 2060 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2069 2061
2070 brelse(wc->w_di_bh); 2062 brelse(wc->w_di_bh);
2071 kfree(wc); 2063 kfree(wc);
2072 2064
2073 return copied; 2065 return copied;
2074 } 2066 }
2075 2067
2076 static int ocfs2_write_end(struct file *file, 2068 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2077 loff_t pos, unsign 2069 loff_t pos, unsigned len, unsigned copied,
2078 struct page *page, 2070 struct page *page, void *fsdata)
2079 { 2071 {
2080 int ret; 2072 int ret;
2081 struct inode *inode = mapping->host; 2073 struct inode *inode = mapping->host;
2082 2074
2083 ret = ocfs2_write_end_nolock(mapping, 2075 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata);
2084 2076
2085 up_write(&OCFS2_I(inode)->ip_alloc_se 2077 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2086 ocfs2_inode_unlock(inode, 1); 2078 ocfs2_inode_unlock(inode, 1);
2087 2079
2088 return ret; 2080 return ret;
2089 } 2081 }
2090 2082
2091 struct ocfs2_dio_write_ctxt { 2083 struct ocfs2_dio_write_ctxt {
2092 struct list_head dw_zero_list; 2084 struct list_head dw_zero_list;
2093 unsigned dw_zero_count 2085 unsigned dw_zero_count;
2094 int dw_orphaned; 2086 int dw_orphaned;
2095 pid_t dw_writer_pid 2087 pid_t dw_writer_pid;
2096 }; 2088 };
2097 2089
2098 static struct ocfs2_dio_write_ctxt * 2090 static struct ocfs2_dio_write_ctxt *
2099 ocfs2_dio_alloc_write_ctx(struct buffer_head 2091 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2100 { 2092 {
2101 struct ocfs2_dio_write_ctxt *dwc = NU 2093 struct ocfs2_dio_write_ctxt *dwc = NULL;
2102 2094
2103 if (bh->b_private) 2095 if (bh->b_private)
2104 return bh->b_private; 2096 return bh->b_private;
2105 2097
2106 dwc = kmalloc(sizeof(struct ocfs2_dio 2098 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2107 if (dwc == NULL) 2099 if (dwc == NULL)
2108 return NULL; 2100 return NULL;
2109 INIT_LIST_HEAD(&dwc->dw_zero_list); 2101 INIT_LIST_HEAD(&dwc->dw_zero_list);
2110 dwc->dw_zero_count = 0; 2102 dwc->dw_zero_count = 0;
2111 dwc->dw_orphaned = 0; 2103 dwc->dw_orphaned = 0;
2112 dwc->dw_writer_pid = task_pid_nr(curr 2104 dwc->dw_writer_pid = task_pid_nr(current);
2113 bh->b_private = dwc; 2105 bh->b_private = dwc;
2114 *alloc = 1; 2106 *alloc = 1;
2115 2107
2116 return dwc; 2108 return dwc;
2117 } 2109 }
2118 2110
2119 static void ocfs2_dio_free_write_ctx(struct i 2111 static void ocfs2_dio_free_write_ctx(struct inode *inode,
2120 struct o 2112 struct ocfs2_dio_write_ctxt *dwc)
2121 { 2113 {
2122 ocfs2_free_unwritten_list(inode, &dwc 2114 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2123 kfree(dwc); 2115 kfree(dwc);
2124 } 2116 }
2125 2117
2126 /* 2118 /*
2127 * TODO: Make this into a generic get_blocks 2119 * TODO: Make this into a generic get_blocks function.
2128 * 2120 *
2129 * From do_direct_io in direct-io.c: 2121 * From do_direct_io in direct-io.c:
2130 * "So what we do is to permit the ->get_blo 2122 * "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 2123 * bh.b_size with the size of IO which is permitted at this offset and
2132 * this i_blkbits." 2124 * this i_blkbits."
2133 * 2125 *
2134 * This function is called directly from get_ 2126 * This function is called directly from get_more_blocks in direct-io.c.
2135 * 2127 *
2136 * called like this: dio->get_blocks(dio->ino 2128 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2137 * fs_co 2129 * fs_count, map_bh, dio->rw == WRITE);
2138 */ 2130 */
2139 static int ocfs2_dio_wr_get_block(struct inod !! 2131 static int ocfs2_dio_get_block(struct inode *inode, sector_t iblock,
2140 struct buffer_ 2132 struct buffer_head *bh_result, int create)
2141 { 2133 {
2142 struct ocfs2_super *osb = OCFS2_SB(in 2134 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2143 struct ocfs2_inode_info *oi = OCFS2_I 2135 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2144 struct ocfs2_write_ctxt *wc; 2136 struct ocfs2_write_ctxt *wc;
2145 struct ocfs2_write_cluster_desc *desc 2137 struct ocfs2_write_cluster_desc *desc = NULL;
2146 struct ocfs2_dio_write_ctxt *dwc = NU 2138 struct ocfs2_dio_write_ctxt *dwc = NULL;
2147 struct buffer_head *di_bh = NULL; 2139 struct buffer_head *di_bh = NULL;
2148 u64 p_blkno; 2140 u64 p_blkno;
2149 unsigned int i_blkbits = inode->i_sb- !! 2141 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-> 2142 unsigned len, total_len = bh_result->b_size;
2153 int ret = 0, first_get_block = 0; 2143 int ret = 0, first_get_block = 0;
2154 2144
2155 len = osb->s_clustersize - (pos & (os 2145 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2156 len = min(total_len, len); 2146 len = min(total_len, len);
2157 2147
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 2148 mlog(0, "get block of %lu at %llu:%u req %u\n",
2174 inode->i_ino, pos, le 2149 inode->i_ino, pos, len, total_len);
2175 2150
2176 /* 2151 /*
2177 * Because we need to change file siz 2152 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2178 * we may need to add it to orphan di 2153 * we may need to add it to orphan dir. So can not fall to fast path
2179 * while file size will be changed. 2154 * while file size will be changed.
2180 */ 2155 */
2181 if (pos + total_len <= i_size_read(in 2156 if (pos + total_len <= i_size_read(inode)) {
2182 !! 2157 down_read(&oi->ip_alloc_sem);
2183 /* This is the fast path for 2158 /* This is the fast path for re-write. */
2184 ret = ocfs2_lock_get_block(in !! 2159 ret = ocfs2_get_block(inode, iblock, bh_result, create);
>> 2160
>> 2161 up_read(&oi->ip_alloc_sem);
>> 2162
2185 if (buffer_mapped(bh_result) 2163 if (buffer_mapped(bh_result) &&
2186 !buffer_new(bh_result) && 2164 !buffer_new(bh_result) &&
2187 ret == 0) 2165 ret == 0)
2188 goto out; 2166 goto out;
2189 2167
2190 /* Clear state set by ocfs2_g 2168 /* Clear state set by ocfs2_get_block. */
2191 bh_result->b_state = 0; 2169 bh_result->b_state = 0;
2192 } 2170 }
2193 2171
2194 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2172 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2195 if (unlikely(dwc == NULL)) { 2173 if (unlikely(dwc == NULL)) {
2196 ret = -ENOMEM; 2174 ret = -ENOMEM;
2197 mlog_errno(ret); 2175 mlog_errno(ret);
2198 goto out; 2176 goto out;
2199 } 2177 }
2200 2178
2201 if (ocfs2_clusters_for_bytes(inode->i 2179 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2202 ocfs2_clusters_for_bytes(inode->i 2180 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2203 !dwc->dw_orphaned) { 2181 !dwc->dw_orphaned) {
2204 /* 2182 /*
2205 * when we are going to alloc 2183 * when we are going to alloc extents beyond file size, add the
2206 * inode to orphan dir, so we 2184 * inode to orphan dir, so we can recall those spaces when
2207 * system crashed during writ 2185 * system crashed during write.
2208 */ 2186 */
2209 ret = ocfs2_add_inode_to_orph 2187 ret = ocfs2_add_inode_to_orphan(osb, inode);
2210 if (ret < 0) { 2188 if (ret < 0) {
2211 mlog_errno(ret); 2189 mlog_errno(ret);
2212 goto out; 2190 goto out;
2213 } 2191 }
2214 dwc->dw_orphaned = 1; 2192 dwc->dw_orphaned = 1;
2215 } 2193 }
2216 2194
2217 ret = ocfs2_inode_lock(inode, &di_bh, 2195 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2218 if (ret) { 2196 if (ret) {
2219 mlog_errno(ret); 2197 mlog_errno(ret);
2220 goto out; 2198 goto out;
2221 } 2199 }
2222 2200
2223 down_write(&oi->ip_alloc_sem); 2201 down_write(&oi->ip_alloc_sem);
2224 2202
2225 if (first_get_block) { 2203 if (first_get_block) {
2226 if (ocfs2_sparse_alloc(osb)) !! 2204 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
2227 ret = ocfs2_zero_tail 2205 ret = ocfs2_zero_tail(inode, di_bh, pos);
2228 else 2206 else
2229 ret = ocfs2_expand_no 2207 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2230 2208 total_len, NULL);
2231 if (ret < 0) { 2209 if (ret < 0) {
2232 mlog_errno(ret); 2210 mlog_errno(ret);
2233 goto unlock; 2211 goto unlock;
2234 } 2212 }
2235 } 2213 }
2236 2214
2237 ret = ocfs2_write_begin_nolock(inode- 2215 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2238 OCFS2_ 2216 OCFS2_WRITE_DIRECT, NULL,
2239 (void 2217 (void **)&wc, di_bh, NULL);
2240 if (ret) { 2218 if (ret) {
2241 mlog_errno(ret); 2219 mlog_errno(ret);
2242 goto unlock; 2220 goto unlock;
2243 } 2221 }
2244 2222
2245 desc = &wc->w_desc[0]; 2223 desc = &wc->w_desc[0];
2246 2224
2247 p_blkno = ocfs2_clusters_to_blocks(in 2225 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2248 BUG_ON(p_blkno == 0); 2226 BUG_ON(p_blkno == 0);
2249 p_blkno += iblock & (u64)(ocfs2_clust 2227 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2250 2228
2251 map_bh(bh_result, inode->i_sb, p_blkn 2229 map_bh(bh_result, inode->i_sb, p_blkno);
2252 bh_result->b_size = len; 2230 bh_result->b_size = len;
2253 if (desc->c_needs_zero) 2231 if (desc->c_needs_zero)
2254 set_buffer_new(bh_result); 2232 set_buffer_new(bh_result);
2255 2233
2256 if (iblock > endblk) <<
2257 set_buffer_new(bh_result); <<
2258 <<
2259 /* May sleep in end_io. It should not 2234 /* May sleep in end_io. It should not happen in a irq context. So defer
2260 * it to dio work queue. */ 2235 * it to dio work queue. */
2261 set_buffer_defer_completion(bh_result 2236 set_buffer_defer_completion(bh_result);
2262 2237
2263 if (!list_empty(&wc->w_unwritten_list 2238 if (!list_empty(&wc->w_unwritten_list)) {
2264 struct ocfs2_unwritten_extent 2239 struct ocfs2_unwritten_extent *ue = NULL;
2265 2240
2266 ue = list_first_entry(&wc->w_ 2241 ue = list_first_entry(&wc->w_unwritten_list,
2267 struct 2242 struct ocfs2_unwritten_extent,
2268 ue_node 2243 ue_node);
2269 BUG_ON(ue->ue_cpos != desc->c 2244 BUG_ON(ue->ue_cpos != desc->c_cpos);
2270 /* The physical address may b 2245 /* The physical address may be 0, fill it. */
2271 ue->ue_phys = desc->c_phys; 2246 ue->ue_phys = desc->c_phys;
2272 2247
2273 list_splice_tail_init(&wc->w_ 2248 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
2274 dwc->dw_zero_count += wc->w_u !! 2249 dwc->dw_zero_count++;
2275 } 2250 }
2276 2251
2277 ret = ocfs2_write_end_nolock(inode->i 2252 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc);
2278 BUG_ON(ret != len); 2253 BUG_ON(ret != len);
2279 ret = 0; 2254 ret = 0;
2280 unlock: 2255 unlock:
2281 up_write(&oi->ip_alloc_sem); 2256 up_write(&oi->ip_alloc_sem);
2282 ocfs2_inode_unlock(inode, 1); 2257 ocfs2_inode_unlock(inode, 1);
2283 brelse(di_bh); 2258 brelse(di_bh);
2284 out: 2259 out:
>> 2260 if (ret < 0)
>> 2261 ret = -EIO;
2285 return ret; 2262 return ret;
2286 } 2263 }
2287 2264
2288 static int ocfs2_dio_end_io_write(struct inod 2265 static int ocfs2_dio_end_io_write(struct inode *inode,
2289 struct ocfs 2266 struct ocfs2_dio_write_ctxt *dwc,
2290 loff_t offs 2267 loff_t offset,
2291 ssize_t byt 2268 ssize_t bytes)
2292 { 2269 {
2293 struct ocfs2_cached_dealloc_ctxt deal 2270 struct ocfs2_cached_dealloc_ctxt dealloc;
2294 struct ocfs2_extent_tree et; 2271 struct ocfs2_extent_tree et;
2295 struct ocfs2_super *osb = OCFS2_SB(in 2272 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2296 struct ocfs2_inode_info *oi = OCFS2_I 2273 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2297 struct ocfs2_unwritten_extent *ue = N 2274 struct ocfs2_unwritten_extent *ue = NULL;
2298 struct buffer_head *di_bh = NULL; 2275 struct buffer_head *di_bh = NULL;
2299 struct ocfs2_dinode *di; 2276 struct ocfs2_dinode *di;
2300 struct ocfs2_alloc_context *data_ac = 2277 struct ocfs2_alloc_context *data_ac = NULL;
2301 struct ocfs2_alloc_context *meta_ac = 2278 struct ocfs2_alloc_context *meta_ac = NULL;
2302 handle_t *handle = NULL; 2279 handle_t *handle = NULL;
2303 loff_t end = offset + bytes; 2280 loff_t end = offset + bytes;
2304 int ret = 0, credits = 0; !! 2281 int ret = 0, credits = 0, locked = 0;
2305 2282
2306 ocfs2_init_dealloc_ctxt(&dealloc); 2283 ocfs2_init_dealloc_ctxt(&dealloc);
2307 2284
2308 /* We do clear unwritten, delete orph 2285 /* We do clear unwritten, delete orphan, change i_size here. If neither
2309 * of these happen, we can skip all t 2286 * of these happen, we can skip all this. */
2310 if (list_empty(&dwc->dw_zero_list) && 2287 if (list_empty(&dwc->dw_zero_list) &&
2311 end <= i_size_read(inode) && 2288 end <= i_size_read(inode) &&
2312 !dwc->dw_orphaned) 2289 !dwc->dw_orphaned)
2313 goto out; 2290 goto out;
2314 2291
>> 2292 /* ocfs2_file_write_iter will get i_mutex, so we need not lock if we
>> 2293 * are in that context. */
>> 2294 if (dwc->dw_writer_pid != task_pid_nr(current)) {
>> 2295 inode_lock(inode);
>> 2296 locked = 1;
>> 2297 }
>> 2298
2315 ret = ocfs2_inode_lock(inode, &di_bh, 2299 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2316 if (ret < 0) { 2300 if (ret < 0) {
2317 mlog_errno(ret); 2301 mlog_errno(ret);
2318 goto out; 2302 goto out;
2319 } 2303 }
2320 2304
2321 down_write(&oi->ip_alloc_sem); 2305 down_write(&oi->ip_alloc_sem);
2322 2306
2323 /* Delete orphan before acquire i_rws !! 2307 /* Delete orphan before acquire i_mutex. */
2324 if (dwc->dw_orphaned) { 2308 if (dwc->dw_orphaned) {
2325 BUG_ON(dwc->dw_writer_pid != 2309 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2326 2310
2327 end = end > i_size_read(inode 2311 end = end > i_size_read(inode) ? end : 0;
2328 2312
2329 ret = ocfs2_del_inode_from_or 2313 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2330 !!end, end); 2314 !!end, end);
2331 if (ret < 0) 2315 if (ret < 0)
2332 mlog_errno(ret); 2316 mlog_errno(ret);
2333 } 2317 }
2334 2318
2335 di = (struct ocfs2_dinode *)di_bh->b_ 2319 di = (struct ocfs2_dinode *)di_bh->b_data;
2336 2320
2337 ocfs2_init_dinode_extent_tree(&et, IN 2321 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2338 2322
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 2323 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2346 &data_ac, 2324 &data_ac, &meta_ac);
2347 if (ret) { 2325 if (ret) {
2348 mlog_errno(ret); 2326 mlog_errno(ret);
2349 goto unlock; 2327 goto unlock;
2350 } 2328 }
2351 2329
2352 credits = ocfs2_calc_extend_credits(i 2330 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2353 2331
2354 handle = ocfs2_start_trans(osb, credi 2332 handle = ocfs2_start_trans(osb, credits);
2355 if (IS_ERR(handle)) { 2333 if (IS_ERR(handle)) {
2356 ret = PTR_ERR(handle); 2334 ret = PTR_ERR(handle);
2357 mlog_errno(ret); 2335 mlog_errno(ret);
2358 goto unlock; 2336 goto unlock;
2359 } 2337 }
2360 ret = ocfs2_journal_access_di(handle, 2338 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2361 OCFS2_J 2339 OCFS2_JOURNAL_ACCESS_WRITE);
2362 if (ret) { 2340 if (ret) {
2363 mlog_errno(ret); 2341 mlog_errno(ret);
2364 goto commit; 2342 goto commit;
2365 } 2343 }
2366 2344
2367 list_for_each_entry(ue, &dwc->dw_zero 2345 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 2346 ret = ocfs2_mark_extent_written(inode, &et, handle,
2374 2347 ue->ue_cpos, 1,
2375 2348 ue->ue_phys,
2376 2349 meta_ac, &dealloc);
2377 if (ret < 0) { 2350 if (ret < 0) {
2378 mlog_errno(ret); 2351 mlog_errno(ret);
2379 break; 2352 break;
2380 } 2353 }
2381 } 2354 }
2382 2355
2383 if (end > i_size_read(inode)) { 2356 if (end > i_size_read(inode)) {
2384 ret = ocfs2_set_inode_size(ha 2357 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2385 if (ret < 0) 2358 if (ret < 0)
2386 mlog_errno(ret); 2359 mlog_errno(ret);
2387 } 2360 }
2388 commit: 2361 commit:
2389 ocfs2_commit_trans(osb, handle); 2362 ocfs2_commit_trans(osb, handle);
2390 unlock: 2363 unlock:
2391 up_write(&oi->ip_alloc_sem); 2364 up_write(&oi->ip_alloc_sem);
2392 ocfs2_inode_unlock(inode, 1); 2365 ocfs2_inode_unlock(inode, 1);
2393 brelse(di_bh); 2366 brelse(di_bh);
2394 out: 2367 out:
2395 if (data_ac) 2368 if (data_ac)
2396 ocfs2_free_alloc_context(data 2369 ocfs2_free_alloc_context(data_ac);
2397 if (meta_ac) 2370 if (meta_ac)
2398 ocfs2_free_alloc_context(meta 2371 ocfs2_free_alloc_context(meta_ac);
2399 ocfs2_run_deallocs(osb, &dealloc); 2372 ocfs2_run_deallocs(osb, &dealloc);
>> 2373 if (locked)
>> 2374 inode_unlock(inode);
2400 ocfs2_dio_free_write_ctx(inode, dwc); 2375 ocfs2_dio_free_write_ctx(inode, dwc);
2401 2376
2402 return ret; 2377 return ret;
2403 } 2378 }
2404 2379
2405 /* 2380 /*
2406 * ocfs2_dio_end_io is called by the dio core 2381 * 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 2382 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2408 * to protect io on one node from truncation 2383 * to protect io on one node from truncation on another.
2409 */ 2384 */
2410 static int ocfs2_dio_end_io(struct kiocb *ioc 2385 static int ocfs2_dio_end_io(struct kiocb *iocb,
2411 loff_t offset, 2386 loff_t offset,
2412 ssize_t bytes, 2387 ssize_t bytes,
2413 void *private) 2388 void *private)
2414 { 2389 {
2415 struct inode *inode = file_inode(iocb 2390 struct inode *inode = file_inode(iocb->ki_filp);
2416 int level; 2391 int level;
2417 int ret = 0; 2392 int ret = 0;
2418 2393
2419 /* this io's submitter should not hav 2394 /* this io's submitter should not have unlocked this before we could */
2420 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2395 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2421 2396
2422 if (bytes <= 0) !! 2397 if (bytes > 0 && private)
2423 mlog_ratelimited(ML_ERROR, "D !! 2398 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 2399
2433 ocfs2_iocb_clear_rw_locked(iocb); 2400 ocfs2_iocb_clear_rw_locked(iocb);
2434 2401
2435 level = ocfs2_iocb_rw_locked_level(io 2402 level = ocfs2_iocb_rw_locked_level(iocb);
2436 ocfs2_rw_unlock(inode, level); 2403 ocfs2_rw_unlock(inode, level);
2437 return ret; 2404 return ret;
2438 } 2405 }
2439 2406
2440 static ssize_t ocfs2_direct_IO(struct kiocb * 2407 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2441 { 2408 {
2442 struct file *file = iocb->ki_filp; 2409 struct file *file = iocb->ki_filp;
2443 struct inode *inode = file->f_mapping 2410 struct inode *inode = file->f_mapping->host;
2444 struct ocfs2_super *osb = OCFS2_SB(in 2411 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2445 get_block_t *get_block; 2412 get_block_t *get_block;
2446 2413
2447 /* 2414 /*
2448 * Fallback to buffered I/O if we see 2415 * Fallback to buffered I/O if we see an inode without
2449 * extents. 2416 * extents.
2450 */ 2417 */
2451 if (OCFS2_I(inode)->ip_dyn_features & 2418 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2452 return 0; 2419 return 0;
2453 2420
2454 /* Fallback to buffered I/O if we do 2421 /* Fallback to buffered I/O if we do not support append dio. */
2455 if (iocb->ki_pos + iter->count > i_si 2422 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2456 !ocfs2_supports_append_dio(osb)) 2423 !ocfs2_supports_append_dio(osb))
2457 return 0; 2424 return 0;
2458 2425
2459 if (iov_iter_rw(iter) == READ) 2426 if (iov_iter_rw(iter) == READ)
2460 get_block = ocfs2_lock_get_bl !! 2427 get_block = ocfs2_get_block;
2461 else 2428 else
2462 get_block = ocfs2_dio_wr_get_ !! 2429 get_block = ocfs2_dio_get_block;
2463 2430
2464 return __blockdev_direct_IO(iocb, ino 2431 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2465 iter, get 2432 iter, get_block,
2466 ocfs2_dio !! 2433 ocfs2_dio_end_io, NULL, 0);
2467 } 2434 }
2468 2435
2469 const struct address_space_operations ocfs2_a 2436 const struct address_space_operations ocfs2_aops = {
2470 .dirty_folio = block_dirty !! 2437 .readpage = ocfs2_readpage,
2471 .read_folio = ocfs2_read_ !! 2438 .readpages = ocfs2_readpages,
2472 .readahead = ocfs2_reada !! 2439 .writepage = ocfs2_writepage,
2473 .writepages = ocfs2_write <<
2474 .write_begin = ocfs2_write 2440 .write_begin = ocfs2_write_begin,
2475 .write_end = ocfs2_write 2441 .write_end = ocfs2_write_end,
2476 .bmap = ocfs2_bmap, 2442 .bmap = ocfs2_bmap,
2477 .direct_IO = ocfs2_direc 2443 .direct_IO = ocfs2_direct_IO,
2478 .invalidate_folio = block_inval !! 2444 .invalidatepage = block_invalidatepage,
2479 .release_folio = ocfs2_relea !! 2445 .releasepage = ocfs2_releasepage,
2480 .migrate_folio = buffer_migr !! 2446 .migratepage = buffer_migrate_page,
2481 .is_partially_uptodate = block_is_pa 2447 .is_partially_uptodate = block_is_partially_uptodate,
2482 .error_remove_folio = generic_err !! 2448 .error_remove_page = generic_error_remove_page,
2483 }; 2449 };
2484 2450
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.