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, "Error %d from 162 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
160 "%llu, NULL)\n", err, ino 163 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
161 (unsigned long long)p_blk 164 (unsigned long long)p_blkno);
162 goto bail; 165 goto bail;
163 } 166 }
164 167
165 if (max_blocks < count) 168 if (max_blocks < count)
166 count = max_blocks; 169 count = max_blocks;
167 170
168 /* 171 /*
169 * ocfs2 never allocates in this funct 172 * ocfs2 never allocates in this function - the only time we
170 * 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
171 * system which doesn't support holes, 174 * system which doesn't support holes, in which case BH_New
172 * allows __block_write_begin() to zer 175 * allows __block_write_begin() to zero.
173 * 176 *
174 * If we see this on a sparse file sys 177 * If we see this on a sparse file system, then a truncate has
175 * raced us and removed the cluster. I 178 * raced us and removed the cluster. In this case, we clear
176 * the buffers dirty and uptodate bits 179 * the buffers dirty and uptodate bits and let the buffer code
177 * ignore it as a hole. 180 * ignore it as a hole.
178 */ 181 */
179 if (create && p_blkno == 0 && ocfs2_sp 182 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
180 clear_buffer_dirty(bh_result); 183 clear_buffer_dirty(bh_result);
181 clear_buffer_uptodate(bh_resul 184 clear_buffer_uptodate(bh_result);
182 goto bail; 185 goto bail;
183 } 186 }
184 187
185 /* Treat the unwritten extent as a hol 188 /* Treat the unwritten extent as a hole for zeroing purposes. */
186 if (p_blkno && !(ext_flags & OCFS2_EXT 189 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
187 map_bh(bh_result, inode->i_sb, 190 map_bh(bh_result, inode->i_sb, p_blkno);
188 191
189 bh_result->b_size = count << inode->i_ 192 bh_result->b_size = count << inode->i_blkbits;
190 193
191 if (!ocfs2_sparse_alloc(osb)) { 194 if (!ocfs2_sparse_alloc(osb)) {
192 if (p_blkno == 0) { 195 if (p_blkno == 0) {
193 err = -EIO; 196 err = -EIO;
194 mlog(ML_ERROR, 197 mlog(ML_ERROR,
195 "iblock = %llu p_ 198 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
196 (unsigned long lo 199 (unsigned long long)iblock,
197 (unsigned long lo 200 (unsigned long long)p_blkno,
198 (unsigned long lo 201 (unsigned long long)OCFS2_I(inode)->ip_blkno);
199 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);
200 dump_stack(); 203 dump_stack();
201 goto bail; 204 goto bail;
202 } 205 }
203 } 206 }
204 207
205 past_eof = ocfs2_blocks_for_bytes(inod 208 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
206 209
207 trace_ocfs2_get_block_end((unsigned lo 210 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
208 (unsigned lo 211 (unsigned long long)past_eof);
209 if (create && (iblock >= past_eof)) 212 if (create && (iblock >= past_eof))
210 set_buffer_new(bh_result); 213 set_buffer_new(bh_result);
211 214
212 bail: 215 bail:
213 if (err < 0) 216 if (err < 0)
214 err = -EIO; 217 err = -EIO;
215 218
216 return err; 219 return err;
217 } 220 }
218 221
219 int ocfs2_read_inline_data(struct inode *inode 222 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
220 struct buffer_head 223 struct buffer_head *di_bh)
221 { 224 {
222 void *kaddr; 225 void *kaddr;
223 loff_t size; 226 loff_t size;
224 struct ocfs2_dinode *di = (struct ocfs 227 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
225 228
226 if (!(le16_to_cpu(di->i_dyn_features) 229 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
227 ocfs2_error(inode->i_sb, "Inod 230 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
228 (unsigned long lon 231 (unsigned long long)OCFS2_I(inode)->ip_blkno);
229 return -EROFS; 232 return -EROFS;
230 } 233 }
231 234
232 size = i_size_read(inode); 235 size = i_size_read(inode);
233 236
234 if (size > PAGE_SIZE || 237 if (size > PAGE_SIZE ||
235 size > ocfs2_max_inline_data_with_ 238 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
236 ocfs2_error(inode->i_sb, 239 ocfs2_error(inode->i_sb,
237 "Inode %llu has wi 240 "Inode %llu has with inline data has bad size: %Lu\n",
238 (unsigned long lon 241 (unsigned long long)OCFS2_I(inode)->ip_blkno,
239 (unsigned long lon 242 (unsigned long long)size);
240 return -EROFS; 243 return -EROFS;
241 } 244 }
242 245
243 kaddr = kmap_atomic(page); 246 kaddr = kmap_atomic(page);
244 if (size) 247 if (size)
245 memcpy(kaddr, di->id2.i_data.i 248 memcpy(kaddr, di->id2.i_data.id_data, size);
246 /* Clear the remaining part of the pag 249 /* Clear the remaining part of the page */
247 memset(kaddr + size, 0, PAGE_SIZE - si 250 memset(kaddr + size, 0, PAGE_SIZE - size);
248 flush_dcache_page(page); 251 flush_dcache_page(page);
249 kunmap_atomic(kaddr); 252 kunmap_atomic(kaddr);
250 253
251 SetPageUptodate(page); 254 SetPageUptodate(page);
252 255
253 return 0; 256 return 0;
254 } 257 }
255 258
256 static int ocfs2_readpage_inline(struct inode 259 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
257 { 260 {
258 int ret; 261 int ret;
259 struct buffer_head *di_bh = NULL; 262 struct buffer_head *di_bh = NULL;
260 263
261 BUG_ON(!PageLocked(page)); 264 BUG_ON(!PageLocked(page));
262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 265 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
263 266
264 ret = ocfs2_read_inode_block(inode, &d 267 ret = ocfs2_read_inode_block(inode, &di_bh);
265 if (ret) { 268 if (ret) {
266 mlog_errno(ret); 269 mlog_errno(ret);
267 goto out; 270 goto out;
268 } 271 }
269 272
270 ret = ocfs2_read_inline_data(inode, pa 273 ret = ocfs2_read_inline_data(inode, page, di_bh);
271 out: 274 out:
272 unlock_page(page); 275 unlock_page(page);
273 276
274 brelse(di_bh); 277 brelse(di_bh);
275 return ret; 278 return ret;
276 } 279 }
277 280
278 static int ocfs2_read_folio(struct file *file, !! 281 static int ocfs2_readpage(struct file *file, struct page *page)
279 { 282 {
280 struct inode *inode = folio->mapping-> !! 283 struct inode *inode = page->mapping->host;
281 struct ocfs2_inode_info *oi = OCFS2_I( 284 struct ocfs2_inode_info *oi = OCFS2_I(inode);
282 loff_t start = folio_pos(folio); !! 285 loff_t start = (loff_t)page->index << PAGE_SHIFT;
283 int ret, unlock = 1; 286 int ret, unlock = 1;
284 287
285 trace_ocfs2_readpage((unsigned long lo !! 288 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
>> 289 (page ? page->index : 0));
286 290
287 ret = ocfs2_inode_lock_with_page(inode !! 291 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
288 if (ret != 0) { 292 if (ret != 0) {
289 if (ret == AOP_TRUNCATED_PAGE) 293 if (ret == AOP_TRUNCATED_PAGE)
290 unlock = 0; 294 unlock = 0;
291 mlog_errno(ret); 295 mlog_errno(ret);
292 goto out; 296 goto out;
293 } 297 }
294 298
295 if (down_read_trylock(&oi->ip_alloc_se 299 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
296 /* 300 /*
297 * Unlock the folio and cycle !! 301 * Unlock the page and cycle ip_alloc_sem so that we don't
298 * busyloop waiting for ip_all 302 * busyloop waiting for ip_alloc_sem to unlock
299 */ 303 */
300 ret = AOP_TRUNCATED_PAGE; 304 ret = AOP_TRUNCATED_PAGE;
301 folio_unlock(folio); !! 305 unlock_page(page);
302 unlock = 0; 306 unlock = 0;
303 down_read(&oi->ip_alloc_sem); 307 down_read(&oi->ip_alloc_sem);
304 up_read(&oi->ip_alloc_sem); 308 up_read(&oi->ip_alloc_sem);
305 goto out_inode_unlock; 309 goto out_inode_unlock;
306 } 310 }
307 311
308 /* 312 /*
309 * i_size might have just been updated 313 * i_size might have just been updated as we grabed the meta lock. We
310 * might now be discovering a truncate 314 * might now be discovering a truncate that hit on another node.
311 * block_read_full_folio->get_block fr !! 315 * block_read_full_page->get_block freaks out if it is asked to read
312 * beyond the end of a file, so we che 316 * beyond the end of a file, so we check here. Callers
313 * (generic_file_read, vm_ops->fault) 317 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
314 * and notice that the folio they just !! 318 * and notice that the page they just read isn't needed.
315 * 319 *
316 * XXX sys_readahead() seems to get th 320 * XXX sys_readahead() seems to get that wrong?
317 */ 321 */
318 if (start >= i_size_read(inode)) { 322 if (start >= i_size_read(inode)) {
319 folio_zero_segment(folio, 0, f !! 323 zero_user(page, 0, PAGE_SIZE);
320 folio_mark_uptodate(folio); !! 324 SetPageUptodate(page);
321 ret = 0; 325 ret = 0;
322 goto out_alloc; 326 goto out_alloc;
323 } 327 }
324 328
325 if (oi->ip_dyn_features & OCFS2_INLINE 329 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
326 ret = ocfs2_readpage_inline(in !! 330 ret = ocfs2_readpage_inline(inode, page);
327 else 331 else
328 ret = block_read_full_folio(fo !! 332 ret = block_read_full_page(page, ocfs2_get_block);
329 unlock = 0; 333 unlock = 0;
330 334
331 out_alloc: 335 out_alloc:
332 up_read(&oi->ip_alloc_sem); !! 336 up_read(&OCFS2_I(inode)->ip_alloc_sem);
333 out_inode_unlock: 337 out_inode_unlock:
334 ocfs2_inode_unlock(inode, 0); 338 ocfs2_inode_unlock(inode, 0);
335 out: 339 out:
336 if (unlock) 340 if (unlock)
337 folio_unlock(folio); !! 341 unlock_page(page);
338 return ret; 342 return ret;
339 } 343 }
340 344
341 /* 345 /*
342 * This is used only for read-ahead. Failures 346 * This is used only for read-ahead. Failures or difficult to handle
343 * situations are safe to ignore. 347 * situations are safe to ignore.
344 * 348 *
345 * Right now, we don't bother with BH_Boundary 349 * Right now, we don't bother with BH_Boundary - in-inode extent lists
346 * are quite large (243 extents on 4k blocks), 350 * are quite large (243 extents on 4k blocks), so most inodes don't
347 * grow out to a tree. If need be, detecting b 351 * grow out to a tree. If need be, detecting boundary extents could
348 * trivially be added in a future version of o 352 * trivially be added in a future version of ocfs2_get_block().
349 */ 353 */
350 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)
351 { 356 {
352 int ret; !! 357 int ret, err = -EIO;
353 struct inode *inode = rac->mapping->ho !! 358 struct inode *inode = mapping->host;
354 struct ocfs2_inode_info *oi = OCFS2_I( 359 struct ocfs2_inode_info *oi = OCFS2_I(inode);
>> 360 loff_t start;
>> 361 struct page *last;
355 362
356 /* 363 /*
357 * Use the nonblocking flag for the dl 364 * Use the nonblocking flag for the dlm code to avoid page
358 * lock inversion, but don't bother wi 365 * lock inversion, but don't bother with retrying.
359 */ 366 */
360 ret = ocfs2_inode_lock_full(inode, NUL 367 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
361 if (ret) 368 if (ret)
362 return; !! 369 return err;
363 370
364 if (down_read_trylock(&oi->ip_alloc_se !! 371 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
365 goto out_unlock; !! 372 ocfs2_inode_unlock(inode, 0);
>> 373 return err;
>> 374 }
366 375
367 /* 376 /*
368 * Don't bother with inline-data. Ther 377 * Don't bother with inline-data. There isn't anything
369 * to read-ahead in that case anyway.. 378 * to read-ahead in that case anyway...
370 */ 379 */
371 if (oi->ip_dyn_features & OCFS2_INLINE 380 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
372 goto out_up; !! 381 goto out_unlock;
373 382
374 /* 383 /*
375 * Check whether a remote node truncat 384 * Check whether a remote node truncated this file - we just
376 * drop out in that case as it's not w 385 * drop out in that case as it's not worth handling here.
377 */ 386 */
378 if (readahead_pos(rac) >= i_size_read( !! 387 last = list_entry(pages->prev, struct page, lru);
379 goto out_up; !! 388 start = (loff_t)last->index << PAGE_SHIFT;
>> 389 if (start >= i_size_read(inode))
>> 390 goto out_unlock;
380 391
381 mpage_readahead(rac, ocfs2_get_block); !! 392 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
382 393
383 out_up: <<
384 up_read(&oi->ip_alloc_sem); <<
385 out_unlock: 394 out_unlock:
>> 395 up_read(&oi->ip_alloc_sem);
386 ocfs2_inode_unlock(inode, 0); 396 ocfs2_inode_unlock(inode, 0);
>> 397
>> 398 return err;
387 } 399 }
388 400
389 /* Note: Because we don't support holes, our a 401 /* Note: Because we don't support holes, our allocation has
390 * already happened (allocation writes zeros t 402 * already happened (allocation writes zeros to the file data)
391 * so we don't have to worry about ordered wri 403 * so we don't have to worry about ordered writes in
392 * ocfs2_writepages. !! 404 * ocfs2_writepage.
393 * 405 *
394 * ->writepages is called during the process o !! 406 * ->writepage is called during the process of invalidating the page cache
395 * during blocked lock processing. It can't b 407 * during blocked lock processing. It can't block on any cluster locks
396 * to during block mapping. It's relying on t 408 * to during block mapping. It's relying on the fact that the block
397 * mapping can't have disappeared under the di 409 * mapping can't have disappeared under the dirty pages that it is
398 * being asked to write back. 410 * being asked to write back.
399 */ 411 */
400 static int ocfs2_writepages(struct address_spa !! 412 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
401 struct writeback_control *wbc) <<
402 { 413 {
403 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);
404 } 419 }
405 420
406 /* Taken from ext3. We don't necessarily need 421 /* Taken from ext3. We don't necessarily need the full blown
407 * functionality yet, but IMHO it's better to 422 * functionality yet, but IMHO it's better to cut and paste the whole
408 * thing so we can avoid introducing our own b 423 * thing so we can avoid introducing our own bugs (and easily pick up
409 * their fixes when they happen) --Mark */ 424 * their fixes when they happen) --Mark */
410 int walk_page_buffers( handle_t *handle, 425 int walk_page_buffers( handle_t *handle,
411 struct buffer_head *he 426 struct buffer_head *head,
412 unsigned from, 427 unsigned from,
413 unsigned to, 428 unsigned to,
414 int *partial, 429 int *partial,
415 int (*fn)( handle 430 int (*fn)( handle_t *handle,
416 struct 431 struct buffer_head *bh))
417 { 432 {
418 struct buffer_head *bh; 433 struct buffer_head *bh;
419 unsigned block_start, block_end; 434 unsigned block_start, block_end;
420 unsigned blocksize = head->b_size; 435 unsigned blocksize = head->b_size;
421 int err, ret = 0; 436 int err, ret = 0;
422 struct buffer_head *next; 437 struct buffer_head *next;
423 438
424 for ( bh = head, block_start = 0; 439 for ( bh = head, block_start = 0;
425 ret == 0 && (bh != head || !bl 440 ret == 0 && (bh != head || !block_start);
426 block_start = block_end, bh = 441 block_start = block_end, bh = next)
427 { 442 {
428 next = bh->b_this_page; 443 next = bh->b_this_page;
429 block_end = block_start + bloc 444 block_end = block_start + blocksize;
430 if (block_end <= from || block 445 if (block_end <= from || block_start >= to) {
431 if (partial && !buffer 446 if (partial && !buffer_uptodate(bh))
432 *partial = 1; 447 *partial = 1;
433 continue; 448 continue;
434 } 449 }
435 err = (*fn)(handle, bh); 450 err = (*fn)(handle, bh);
436 if (!ret) 451 if (!ret)
437 ret = err; 452 ret = err;
438 } 453 }
439 return ret; 454 return ret;
440 } 455 }
441 456
442 static sector_t ocfs2_bmap(struct address_spac 457 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
443 { 458 {
444 sector_t status; 459 sector_t status;
445 u64 p_blkno = 0; 460 u64 p_blkno = 0;
446 int err = 0; 461 int err = 0;
447 struct inode *inode = mapping->host; 462 struct inode *inode = mapping->host;
448 463
449 trace_ocfs2_bmap((unsigned long long)O 464 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
450 (unsigned long long)b 465 (unsigned long long)block);
451 466
452 /* 467 /*
453 * The swap code (ab-)uses ->bmap to g 468 * The swap code (ab-)uses ->bmap to get a block mapping and then
454 * bypasseѕ the file system for actua 469 * bypasseѕ the file system for actual I/O. We really can't allow
455 * that on refcounted inodes, so we ha 470 * that on refcounted inodes, so we have to skip out here. And yes,
456 * 0 is the magic code for a bmap erro 471 * 0 is the magic code for a bmap error..
457 */ 472 */
458 if (ocfs2_is_refcount_inode(inode)) 473 if (ocfs2_is_refcount_inode(inode))
459 return 0; 474 return 0;
460 475
461 /* We don't need to lock journal syste 476 /* We don't need to lock journal system files, since they aren't
462 * accessed concurrently from multiple 477 * accessed concurrently from multiple nodes.
463 */ 478 */
464 if (!INODE_JOURNAL(inode)) { 479 if (!INODE_JOURNAL(inode)) {
465 err = ocfs2_inode_lock(inode, 480 err = ocfs2_inode_lock(inode, NULL, 0);
466 if (err) { 481 if (err) {
467 if (err != -ENOENT) 482 if (err != -ENOENT)
468 mlog_errno(err 483 mlog_errno(err);
469 goto bail; 484 goto bail;
470 } 485 }
471 down_read(&OCFS2_I(inode)->ip_ 486 down_read(&OCFS2_I(inode)->ip_alloc_sem);
472 } 487 }
473 488
474 if (!(OCFS2_I(inode)->ip_dyn_features 489 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
475 err = ocfs2_extent_map_get_blo 490 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
476 491 NULL);
477 492
478 if (!INODE_JOURNAL(inode)) { 493 if (!INODE_JOURNAL(inode)) {
479 up_read(&OCFS2_I(inode)->ip_al 494 up_read(&OCFS2_I(inode)->ip_alloc_sem);
480 ocfs2_inode_unlock(inode, 0); 495 ocfs2_inode_unlock(inode, 0);
481 } 496 }
482 497
483 if (err) { 498 if (err) {
484 mlog(ML_ERROR, "get_blocks() f 499 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
485 (unsigned long long)block 500 (unsigned long long)block);
486 mlog_errno(err); 501 mlog_errno(err);
487 goto bail; 502 goto bail;
488 } 503 }
489 504
490 bail: 505 bail:
491 status = err ? 0 : p_blkno; 506 status = err ? 0 : p_blkno;
492 507
493 return status; 508 return status;
494 } 509 }
495 510
496 static bool ocfs2_release_folio(struct folio * !! 511 static int ocfs2_releasepage(struct page *page, gfp_t wait)
497 { 512 {
498 if (!folio_buffers(folio)) !! 513 if (!page_has_buffers(page))
499 return false; !! 514 return 0;
500 return try_to_free_buffers(folio); !! 515 return try_to_free_buffers(page);
501 } 516 }
502 517
503 static void ocfs2_figure_cluster_boundaries(st 518 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
504 u3 519 u32 cpos,
505 un 520 unsigned int *start,
506 un 521 unsigned int *end)
507 { 522 {
508 unsigned int cluster_start = 0, cluste 523 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
509 524
510 if (unlikely(PAGE_SHIFT > osb->s_clust 525 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
511 unsigned int cpp; 526 unsigned int cpp;
512 527
513 cpp = 1 << (PAGE_SHIFT - osb-> 528 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
514 529
515 cluster_start = cpos % cpp; 530 cluster_start = cpos % cpp;
516 cluster_start = cluster_start 531 cluster_start = cluster_start << osb->s_clustersize_bits;
517 532
518 cluster_end = cluster_start + 533 cluster_end = cluster_start + osb->s_clustersize;
519 } 534 }
520 535
521 BUG_ON(cluster_start > PAGE_SIZE); 536 BUG_ON(cluster_start > PAGE_SIZE);
522 BUG_ON(cluster_end > PAGE_SIZE); 537 BUG_ON(cluster_end > PAGE_SIZE);
523 538
524 if (start) 539 if (start)
525 *start = cluster_start; 540 *start = cluster_start;
526 if (end) 541 if (end)
527 *end = cluster_end; 542 *end = cluster_end;
528 } 543 }
529 544
530 /* 545 /*
531 * 'from' and 'to' are the region in the page 546 * 'from' and 'to' are the region in the page to avoid zeroing.
532 * 547 *
533 * If pagesize > clustersize, this function wi 548 * If pagesize > clustersize, this function will avoid zeroing outside
534 * of the cluster boundary. 549 * of the cluster boundary.
535 * 550 *
536 * from == to == 0 is code for "zero the entir 551 * from == to == 0 is code for "zero the entire cluster region"
537 */ 552 */
538 static void ocfs2_clear_page_regions(struct pa 553 static void ocfs2_clear_page_regions(struct page *page,
539 struct oc 554 struct ocfs2_super *osb, u32 cpos,
540 unsigned 555 unsigned from, unsigned to)
541 { 556 {
542 void *kaddr; 557 void *kaddr;
543 unsigned int cluster_start, cluster_en 558 unsigned int cluster_start, cluster_end;
544 559
545 ocfs2_figure_cluster_boundaries(osb, c 560 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
546 561
547 kaddr = kmap_atomic(page); 562 kaddr = kmap_atomic(page);
548 563
549 if (from || to) { 564 if (from || to) {
550 if (from > cluster_start) 565 if (from > cluster_start)
551 memset(kaddr + cluster 566 memset(kaddr + cluster_start, 0, from - cluster_start);
552 if (to < cluster_end) 567 if (to < cluster_end)
553 memset(kaddr + to, 0, 568 memset(kaddr + to, 0, cluster_end - to);
554 } else { 569 } else {
555 memset(kaddr + cluster_start, 570 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
556 } 571 }
557 572
558 kunmap_atomic(kaddr); 573 kunmap_atomic(kaddr);
559 } 574 }
560 575
561 /* 576 /*
562 * Nonsparse file systems fully allocate befor 577 * Nonsparse file systems fully allocate before we get to the write
563 * code. This prevents ocfs2_write() from tagg 578 * code. This prevents ocfs2_write() from tagging the write as an
564 * allocating one, which means ocfs2_map_page_ 579 * allocating one, which means ocfs2_map_page_blocks() might try to
565 * 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
566 * testing i_size against each block offset. 581 * testing i_size against each block offset.
567 */ 582 */
568 static int ocfs2_should_read_blk(struct inode !! 583 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
569 unsigned int 584 unsigned int block_start)
570 { 585 {
571 u64 offset = folio_pos(folio) + block_ !! 586 u64 offset = page_offset(page) + block_start;
572 587
573 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 588 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
574 return 1; 589 return 1;
575 590
576 if (i_size_read(inode) > offset) 591 if (i_size_read(inode) > offset)
577 return 1; 592 return 1;
578 593
579 return 0; 594 return 0;
580 } 595 }
581 596
582 /* 597 /*
583 * Some of this taken from __block_write_begin 598 * Some of this taken from __block_write_begin(). We already have our
584 * mapping by now though, and the entire write 599 * mapping by now though, and the entire write will be allocating or
585 * it won't, so not much need to use BH_New. 600 * it won't, so not much need to use BH_New.
586 * 601 *
587 * This will also skip zeroing, which is handl 602 * This will also skip zeroing, which is handled externally.
588 */ 603 */
589 int ocfs2_map_page_blocks(struct page *page, u 604 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
590 struct inode *inode, 605 struct inode *inode, unsigned int from,
591 unsigned int to, int 606 unsigned int to, int new)
592 { 607 {
593 struct folio *folio = page_folio(page) <<
594 int ret = 0; 608 int ret = 0;
595 struct buffer_head *head, *bh, *wait[2 609 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
596 unsigned int block_end, block_start; 610 unsigned int block_end, block_start;
597 unsigned int bsize = i_blocksize(inode 611 unsigned int bsize = i_blocksize(inode);
598 612
599 head = folio_buffers(folio); !! 613 if (!page_has_buffers(page))
600 if (!head) !! 614 create_empty_buffers(page, bsize, 0);
601 head = create_empty_buffers(fo <<
602 615
>> 616 head = page_buffers(page);
603 for (bh = head, block_start = 0; bh != 617 for (bh = head, block_start = 0; bh != head || !block_start;
604 bh = bh->b_this_page, block_start 618 bh = bh->b_this_page, block_start += bsize) {
605 block_end = block_start + bsiz 619 block_end = block_start + bsize;
606 620
607 clear_buffer_new(bh); 621 clear_buffer_new(bh);
608 622
609 /* 623 /*
610 * Ignore blocks outside of ou 624 * Ignore blocks outside of our i/o range -
611 * they may belong to unalloca 625 * they may belong to unallocated clusters.
612 */ 626 */
613 if (block_start >= to || block 627 if (block_start >= to || block_end <= from) {
614 if (folio_test_uptodat !! 628 if (PageUptodate(page))
615 set_buffer_upt 629 set_buffer_uptodate(bh);
616 continue; 630 continue;
617 } 631 }
618 632
619 /* 633 /*
620 * For an allocating write wit 634 * For an allocating write with cluster size >= page
621 * size, we always write the e 635 * size, we always write the entire page.
622 */ 636 */
623 if (new) 637 if (new)
624 set_buffer_new(bh); 638 set_buffer_new(bh);
625 639
626 if (!buffer_mapped(bh)) { 640 if (!buffer_mapped(bh)) {
627 map_bh(bh, inode->i_sb 641 map_bh(bh, inode->i_sb, *p_blkno);
628 clean_bdev_bh_alias(bh 642 clean_bdev_bh_alias(bh);
629 } 643 }
630 644
631 if (folio_test_uptodate(folio) !! 645 if (PageUptodate(page)) {
632 set_buffer_uptodate(bh !! 646 if (!buffer_uptodate(bh))
>> 647 set_buffer_uptodate(bh);
633 } else if (!buffer_uptodate(bh 648 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
634 !buffer_new(bh) && 649 !buffer_new(bh) &&
635 ocfs2_should_read_b !! 650 ocfs2_should_read_blk(inode, page, block_start) &&
636 (block_start < from 651 (block_start < from || block_end > to)) {
637 bh_read_nowait(bh, 0); !! 652 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
638 *wait_bh++=bh; 653 *wait_bh++=bh;
639 } 654 }
640 655
641 *p_blkno = *p_blkno + 1; 656 *p_blkno = *p_blkno + 1;
642 } 657 }
643 658
644 /* 659 /*
645 * If we issued read requests - let th 660 * If we issued read requests - let them complete.
646 */ 661 */
647 while(wait_bh > wait) { 662 while(wait_bh > wait) {
648 wait_on_buffer(*--wait_bh); 663 wait_on_buffer(*--wait_bh);
649 if (!buffer_uptodate(*wait_bh) 664 if (!buffer_uptodate(*wait_bh))
650 ret = -EIO; 665 ret = -EIO;
651 } 666 }
652 667
653 if (ret == 0 || !new) 668 if (ret == 0 || !new)
654 return ret; 669 return ret;
655 670
656 /* 671 /*
657 * If we get -EIO above, zero out any 672 * If we get -EIO above, zero out any newly allocated blocks
658 * to avoid exposing stale data. 673 * to avoid exposing stale data.
659 */ 674 */
660 bh = head; 675 bh = head;
661 block_start = 0; 676 block_start = 0;
662 do { 677 do {
663 block_end = block_start + bsiz 678 block_end = block_start + bsize;
664 if (block_end <= from) 679 if (block_end <= from)
665 goto next_bh; 680 goto next_bh;
666 if (block_start >= to) 681 if (block_start >= to)
667 break; 682 break;
668 683
669 folio_zero_range(folio, block_ !! 684 zero_user(page, block_start, bh->b_size);
670 set_buffer_uptodate(bh); 685 set_buffer_uptodate(bh);
671 mark_buffer_dirty(bh); 686 mark_buffer_dirty(bh);
672 687
673 next_bh: 688 next_bh:
674 block_start = block_end; 689 block_start = block_end;
675 bh = bh->b_this_page; 690 bh = bh->b_this_page;
676 } while (bh != head); 691 } while (bh != head);
677 692
678 return ret; 693 return ret;
679 } 694 }
680 695
681 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 696 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
682 #define OCFS2_MAX_CTXT_PAGES 1 697 #define OCFS2_MAX_CTXT_PAGES 1
683 #else 698 #else
684 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 699 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
685 #endif 700 #endif
686 701
687 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 702 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
688 703
689 struct ocfs2_unwritten_extent { 704 struct ocfs2_unwritten_extent {
690 struct list_head ue_node; 705 struct list_head ue_node;
691 struct list_head ue_ip_node; 706 struct list_head ue_ip_node;
692 u32 ue_cpos; 707 u32 ue_cpos;
693 u32 ue_phys; 708 u32 ue_phys;
694 }; 709 };
695 710
696 /* 711 /*
697 * Describe the state of a single cluster to b 712 * Describe the state of a single cluster to be written to.
698 */ 713 */
699 struct ocfs2_write_cluster_desc { 714 struct ocfs2_write_cluster_desc {
700 u32 c_cpos; 715 u32 c_cpos;
701 u32 c_phys; 716 u32 c_phys;
702 /* 717 /*
703 * Give this a unique field because c_ 718 * Give this a unique field because c_phys eventually gets
704 * filled. 719 * filled.
705 */ 720 */
706 unsigned c_new; 721 unsigned c_new;
707 unsigned c_clear_unwritten; 722 unsigned c_clear_unwritten;
708 unsigned c_needs_zero; 723 unsigned c_needs_zero;
709 }; 724 };
710 725
711 struct ocfs2_write_ctxt { 726 struct ocfs2_write_ctxt {
712 /* Logical cluster position / len of w 727 /* Logical cluster position / len of write */
713 u32 w_cpos 728 u32 w_cpos;
714 u32 w_clen 729 u32 w_clen;
715 730
716 /* First cluster allocated in a nonspa 731 /* First cluster allocated in a nonsparse extend */
717 u32 w_firs 732 u32 w_first_new_cpos;
718 733
719 /* Type of caller. Must be one of buff 734 /* Type of caller. Must be one of buffer, mmap, direct. */
720 ocfs2_write_type_t w_type 735 ocfs2_write_type_t w_type;
721 736
722 struct ocfs2_write_cluster_desc w_desc 737 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
723 738
724 /* 739 /*
725 * This is true if page_size > cluster 740 * This is true if page_size > cluster_size.
726 * 741 *
727 * It triggers a set of special cases 742 * It triggers a set of special cases during write which might
728 * have to deal with allocating writes 743 * have to deal with allocating writes to partial pages.
729 */ 744 */
730 unsigned int w_larg 745 unsigned int w_large_pages;
731 746
732 /* 747 /*
733 * Pages involved in this write. 748 * Pages involved in this write.
734 * 749 *
735 * w_target_page is the page being wri 750 * w_target_page is the page being written to by the user.
736 * 751 *
737 * w_pages is an array of pages which 752 * w_pages is an array of pages which always contains
738 * w_target_page, and in the case of a 753 * w_target_page, and in the case of an allocating write with
739 * page_size < cluster size, it will c 754 * page_size < cluster size, it will contain zero'd and mapped
740 * pages adjacent to w_target_page whi 755 * pages adjacent to w_target_page which need to be written
741 * out in so that future reads from th 756 * out in so that future reads from that region will get
742 * zero's. 757 * zero's.
743 */ 758 */
744 unsigned int w_num_ 759 unsigned int w_num_pages;
745 struct page *w_pag 760 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
746 struct page *w_tar 761 struct page *w_target_page;
747 762
748 /* 763 /*
749 * w_target_locked is used for page_mk 764 * w_target_locked is used for page_mkwrite path indicating no unlocking
750 * against w_target_page in ocfs2_writ 765 * against w_target_page in ocfs2_write_end_nolock.
751 */ 766 */
752 unsigned int w_targ 767 unsigned int w_target_locked:1;
753 768
754 /* 769 /*
755 * ocfs2_write_end() uses this to know 770 * ocfs2_write_end() uses this to know what the real range to
756 * write in the target should be. 771 * write in the target should be.
757 */ 772 */
758 unsigned int w_targ 773 unsigned int w_target_from;
759 unsigned int w_targ 774 unsigned int w_target_to;
760 775
761 /* 776 /*
762 * We could use journal_current_handle 777 * We could use journal_current_handle() but this is cleaner,
763 * IMHO -Mark 778 * IMHO -Mark
764 */ 779 */
765 handle_t *w_han 780 handle_t *w_handle;
766 781
767 struct buffer_head *w_di_ 782 struct buffer_head *w_di_bh;
768 783
769 struct ocfs2_cached_dealloc_ctxt w_dea 784 struct ocfs2_cached_dealloc_ctxt w_dealloc;
770 785
771 struct list_head w_unwr 786 struct list_head w_unwritten_list;
772 unsigned int w_unwr <<
773 }; 787 };
774 788
775 void ocfs2_unlock_and_free_pages(struct page * 789 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
776 { 790 {
777 int i; 791 int i;
778 792
779 for(i = 0; i < num_pages; i++) { 793 for(i = 0; i < num_pages; i++) {
780 if (pages[i]) { 794 if (pages[i]) {
781 unlock_page(pages[i]); 795 unlock_page(pages[i]);
782 mark_page_accessed(pag 796 mark_page_accessed(pages[i]);
783 put_page(pages[i]); 797 put_page(pages[i]);
784 } 798 }
785 } 799 }
786 } 800 }
787 801
788 static void ocfs2_unlock_pages(struct ocfs2_wr 802 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
789 { 803 {
790 int i; 804 int i;
791 805
792 /* 806 /*
793 * w_target_locked is only set to true 807 * w_target_locked is only set to true in the page_mkwrite() case.
794 * The intent is to allow us to lock t 808 * The intent is to allow us to lock the target page from write_begin()
795 * to write_end(). The caller must hol 809 * to write_end(). The caller must hold a ref on w_target_page.
796 */ 810 */
797 if (wc->w_target_locked) { 811 if (wc->w_target_locked) {
798 BUG_ON(!wc->w_target_page); 812 BUG_ON(!wc->w_target_page);
799 for (i = 0; i < wc->w_num_page 813 for (i = 0; i < wc->w_num_pages; i++) {
800 if (wc->w_target_page 814 if (wc->w_target_page == wc->w_pages[i]) {
801 wc->w_pages[i] 815 wc->w_pages[i] = NULL;
802 break; 816 break;
803 } 817 }
804 } 818 }
805 mark_page_accessed(wc->w_targe 819 mark_page_accessed(wc->w_target_page);
806 put_page(wc->w_target_page); 820 put_page(wc->w_target_page);
807 } 821 }
808 ocfs2_unlock_and_free_pages(wc->w_page 822 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
809 } 823 }
810 824
811 static void ocfs2_free_unwritten_list(struct i 825 static void ocfs2_free_unwritten_list(struct inode *inode,
812 struct list_h 826 struct list_head *head)
813 { 827 {
814 struct ocfs2_inode_info *oi = OCFS2_I( 828 struct ocfs2_inode_info *oi = OCFS2_I(inode);
815 struct ocfs2_unwritten_extent *ue = NU 829 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
816 830
817 list_for_each_entry_safe(ue, tmp, head 831 list_for_each_entry_safe(ue, tmp, head, ue_node) {
818 list_del(&ue->ue_node); 832 list_del(&ue->ue_node);
819 spin_lock(&oi->ip_lock); 833 spin_lock(&oi->ip_lock);
820 list_del(&ue->ue_ip_node); 834 list_del(&ue->ue_ip_node);
821 spin_unlock(&oi->ip_lock); 835 spin_unlock(&oi->ip_lock);
822 kfree(ue); 836 kfree(ue);
823 } 837 }
824 } 838 }
825 839
826 static void ocfs2_free_write_ctxt(struct inode 840 static void ocfs2_free_write_ctxt(struct inode *inode,
827 struct ocfs2 841 struct ocfs2_write_ctxt *wc)
828 { 842 {
829 ocfs2_free_unwritten_list(inode, &wc-> 843 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
830 ocfs2_unlock_pages(wc); 844 ocfs2_unlock_pages(wc);
831 brelse(wc->w_di_bh); 845 brelse(wc->w_di_bh);
832 kfree(wc); 846 kfree(wc);
833 } 847 }
834 848
835 static int ocfs2_alloc_write_ctxt(struct ocfs2 849 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
836 struct ocfs2 850 struct ocfs2_super *osb, loff_t pos,
837 unsigned len 851 unsigned len, ocfs2_write_type_t type,
838 struct buffe 852 struct buffer_head *di_bh)
839 { 853 {
840 u32 cend; 854 u32 cend;
841 struct ocfs2_write_ctxt *wc; 855 struct ocfs2_write_ctxt *wc;
842 856
843 wc = kzalloc(sizeof(struct ocfs2_write 857 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
844 if (!wc) 858 if (!wc)
845 return -ENOMEM; 859 return -ENOMEM;
846 860
847 wc->w_cpos = pos >> osb->s_clustersize 861 wc->w_cpos = pos >> osb->s_clustersize_bits;
848 wc->w_first_new_cpos = UINT_MAX; 862 wc->w_first_new_cpos = UINT_MAX;
849 cend = (pos + len - 1) >> osb->s_clust 863 cend = (pos + len - 1) >> osb->s_clustersize_bits;
850 wc->w_clen = cend - wc->w_cpos + 1; 864 wc->w_clen = cend - wc->w_cpos + 1;
851 get_bh(di_bh); 865 get_bh(di_bh);
852 wc->w_di_bh = di_bh; 866 wc->w_di_bh = di_bh;
853 wc->w_type = type; 867 wc->w_type = type;
854 868
855 if (unlikely(PAGE_SHIFT > osb->s_clust 869 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
856 wc->w_large_pages = 1; 870 wc->w_large_pages = 1;
857 else 871 else
858 wc->w_large_pages = 0; 872 wc->w_large_pages = 0;
859 873
860 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 874 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
861 INIT_LIST_HEAD(&wc->w_unwritten_list); 875 INIT_LIST_HEAD(&wc->w_unwritten_list);
862 876
863 *wcp = wc; 877 *wcp = wc;
864 878
865 return 0; 879 return 0;
866 } 880 }
867 881
868 /* 882 /*
869 * 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
870 * and dirty so they'll be written out (in ord 884 * and dirty so they'll be written out (in order to prevent uninitialised
871 * block data from leaking). And clear the new 885 * block data from leaking). And clear the new bit.
872 */ 886 */
873 static void ocfs2_zero_new_buffers(struct page 887 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
874 { 888 {
875 unsigned int block_start, block_end; 889 unsigned int block_start, block_end;
876 struct buffer_head *head, *bh; 890 struct buffer_head *head, *bh;
877 891
878 BUG_ON(!PageLocked(page)); 892 BUG_ON(!PageLocked(page));
879 if (!page_has_buffers(page)) 893 if (!page_has_buffers(page))
880 return; 894 return;
881 895
882 bh = head = page_buffers(page); 896 bh = head = page_buffers(page);
883 block_start = 0; 897 block_start = 0;
884 do { 898 do {
885 block_end = block_start + bh-> 899 block_end = block_start + bh->b_size;
886 900
887 if (buffer_new(bh)) { 901 if (buffer_new(bh)) {
888 if (block_end > from & 902 if (block_end > from && block_start < to) {
889 if (!PageUptod 903 if (!PageUptodate(page)) {
890 unsign 904 unsigned start, end;
891 905
892 start 906 start = max(from, block_start);
893 end = 907 end = min(to, block_end);
894 908
895 zero_u 909 zero_user_segment(page, start, end);
896 set_bu 910 set_buffer_uptodate(bh);
897 } 911 }
898 912
899 clear_buffer_n 913 clear_buffer_new(bh);
900 mark_buffer_di 914 mark_buffer_dirty(bh);
901 } 915 }
902 } 916 }
903 917
904 block_start = block_end; 918 block_start = block_end;
905 bh = bh->b_this_page; 919 bh = bh->b_this_page;
906 } while (bh != head); 920 } while (bh != head);
907 } 921 }
908 922
909 /* 923 /*
910 * Only called when we have a failure during a 924 * Only called when we have a failure during allocating write to write
911 * zero's to the newly allocated region. 925 * zero's to the newly allocated region.
912 */ 926 */
913 static void ocfs2_write_failure(struct inode * 927 static void ocfs2_write_failure(struct inode *inode,
914 struct ocfs2_w 928 struct ocfs2_write_ctxt *wc,
915 loff_t user_po 929 loff_t user_pos, unsigned user_len)
916 { 930 {
917 int i; 931 int i;
918 unsigned from = user_pos & (PAGE_SIZE 932 unsigned from = user_pos & (PAGE_SIZE - 1),
919 to = user_pos + user_len; 933 to = user_pos + user_len;
920 struct page *tmppage; 934 struct page *tmppage;
921 935
922 if (wc->w_target_page) 936 if (wc->w_target_page)
923 ocfs2_zero_new_buffers(wc->w_t 937 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
924 938
925 for(i = 0; i < wc->w_num_pages; i++) { 939 for(i = 0; i < wc->w_num_pages; i++) {
926 tmppage = wc->w_pages[i]; 940 tmppage = wc->w_pages[i];
927 941
928 if (tmppage && page_has_buffer 942 if (tmppage && page_has_buffers(tmppage)) {
929 if (ocfs2_should_order 943 if (ocfs2_should_order_data(inode))
930 ocfs2_jbd2_ino !! 944 ocfs2_jbd2_file_inode(wc->w_handle, inode);
931 <<
932 945
933 block_commit_write(tmp 946 block_commit_write(tmppage, from, to);
934 } 947 }
935 } 948 }
936 } 949 }
937 950
938 static int ocfs2_prepare_page_for_write(struct 951 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
939 struct 952 struct ocfs2_write_ctxt *wc,
940 struct 953 struct page *page, u32 cpos,
941 loff_t 954 loff_t user_pos, unsigned user_len,
942 int ne 955 int new)
943 { 956 {
944 int ret; 957 int ret;
945 unsigned int map_from = 0, map_to = 0; 958 unsigned int map_from = 0, map_to = 0;
946 unsigned int cluster_start, cluster_en 959 unsigned int cluster_start, cluster_end;
947 unsigned int user_data_from = 0, user_ 960 unsigned int user_data_from = 0, user_data_to = 0;
948 961
949 ocfs2_figure_cluster_boundaries(OCFS2_ 962 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
950 &clust 963 &cluster_start, &cluster_end);
951 964
952 /* treat the write as new if the a hol 965 /* treat the write as new if the a hole/lseek spanned across
953 * the page boundary. 966 * the page boundary.
954 */ 967 */
955 new = new | ((i_size_read(inode) <= pa 968 new = new | ((i_size_read(inode) <= page_offset(page)) &&
956 (page_offset(page) <= 969 (page_offset(page) <= user_pos));
957 970
958 if (page == wc->w_target_page) { 971 if (page == wc->w_target_page) {
959 map_from = user_pos & (PAGE_SI 972 map_from = user_pos & (PAGE_SIZE - 1);
960 map_to = map_from + user_len; 973 map_to = map_from + user_len;
961 974
962 if (new) 975 if (new)
963 ret = ocfs2_map_page_b 976 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
964 977 cluster_start, cluster_end,
965 978 new);
966 else 979 else
967 ret = ocfs2_map_page_b 980 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
968 981 map_from, map_to, new);
969 if (ret) { 982 if (ret) {
970 mlog_errno(ret); 983 mlog_errno(ret);
971 goto out; 984 goto out;
972 } 985 }
973 986
974 user_data_from = map_from; 987 user_data_from = map_from;
975 user_data_to = map_to; 988 user_data_to = map_to;
976 if (new) { 989 if (new) {
977 map_from = cluster_sta 990 map_from = cluster_start;
978 map_to = cluster_end; 991 map_to = cluster_end;
979 } 992 }
980 } else { 993 } else {
981 /* 994 /*
982 * If we haven't allocated the 995 * If we haven't allocated the new page yet, we
983 * shouldn't be writing it out 996 * shouldn't be writing it out without copying user
984 * data. This is likely a math 997 * data. This is likely a math error from the caller.
985 */ 998 */
986 BUG_ON(!new); 999 BUG_ON(!new);
987 1000
988 map_from = cluster_start; 1001 map_from = cluster_start;
989 map_to = cluster_end; 1002 map_to = cluster_end;
990 1003
991 ret = ocfs2_map_page_blocks(pa 1004 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
992 cl 1005 cluster_start, cluster_end, new);
993 if (ret) { 1006 if (ret) {
994 mlog_errno(ret); 1007 mlog_errno(ret);
995 goto out; 1008 goto out;
996 } 1009 }
997 } 1010 }
998 1011
999 /* 1012 /*
1000 * Parts of newly allocated pages nee 1013 * Parts of newly allocated pages need to be zero'd.
1001 * 1014 *
1002 * Above, we have also rewritten 'to' 1015 * Above, we have also rewritten 'to' and 'from' - as far as
1003 * the rest of the function is concer 1016 * the rest of the function is concerned, the entire cluster
1004 * range inside of a page needs to be 1017 * range inside of a page needs to be written.
1005 * 1018 *
1006 * We can skip this if the page is up 1019 * We can skip this if the page is up to date - it's already
1007 * been zero'd from being read in as 1020 * been zero'd from being read in as a hole.
1008 */ 1021 */
1009 if (new && !PageUptodate(page)) 1022 if (new && !PageUptodate(page))
1010 ocfs2_clear_page_regions(page 1023 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1011 cpos 1024 cpos, user_data_from, user_data_to);
1012 1025
1013 flush_dcache_page(page); 1026 flush_dcache_page(page);
1014 1027
1015 out: 1028 out:
1016 return ret; 1029 return ret;
1017 } 1030 }
1018 1031
1019 /* 1032 /*
1020 * This function will only grab one clusters 1033 * This function will only grab one clusters worth of pages.
1021 */ 1034 */
1022 static int ocfs2_grab_pages_for_write(struct 1035 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1023 struct 1036 struct ocfs2_write_ctxt *wc,
1024 u32 cpo 1037 u32 cpos, loff_t user_pos,
1025 unsigne 1038 unsigned user_len, int new,
1026 struct 1039 struct page *mmap_page)
1027 { 1040 {
1028 int ret = 0, i; 1041 int ret = 0, i;
1029 unsigned long start, target_index, en 1042 unsigned long start, target_index, end_index, index;
1030 struct inode *inode = mapping->host; 1043 struct inode *inode = mapping->host;
1031 loff_t last_byte; 1044 loff_t last_byte;
1032 1045
1033 target_index = user_pos >> PAGE_SHIFT 1046 target_index = user_pos >> PAGE_SHIFT;
1034 1047
1035 /* 1048 /*
1036 * Figure out how many pages we'll be 1049 * Figure out how many pages we'll be manipulating here. For
1037 * non allocating write, we just chan 1050 * non allocating write, we just change the one
1038 * page. Otherwise, we'll need a whol 1051 * page. Otherwise, we'll need a whole clusters worth. If we're
1039 * writing past i_size, we only need 1052 * writing past i_size, we only need enough pages to cover the
1040 * last page of the write. 1053 * last page of the write.
1041 */ 1054 */
1042 if (new) { 1055 if (new) {
1043 wc->w_num_pages = ocfs2_pages 1056 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1044 start = ocfs2_align_clusters_ 1057 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1045 /* 1058 /*
1046 * We need the index *past* t 1059 * We need the index *past* the last page we could possibly
1047 * touch. This is the page p 1060 * touch. This is the page past the end of the write or
1048 * i_size, whichever is great 1061 * i_size, whichever is greater.
1049 */ 1062 */
1050 last_byte = max(user_pos + us 1063 last_byte = max(user_pos + user_len, i_size_read(inode));
1051 BUG_ON(last_byte < 1); 1064 BUG_ON(last_byte < 1);
1052 end_index = ((last_byte - 1) 1065 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
1053 if ((start + wc->w_num_pages) 1066 if ((start + wc->w_num_pages) > end_index)
1054 wc->w_num_pages = end 1067 wc->w_num_pages = end_index - start;
1055 } else { 1068 } else {
1056 wc->w_num_pages = 1; 1069 wc->w_num_pages = 1;
1057 start = target_index; 1070 start = target_index;
1058 } 1071 }
1059 end_index = (user_pos + user_len - 1) 1072 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
1060 1073
1061 for(i = 0; i < wc->w_num_pages; i++) 1074 for(i = 0; i < wc->w_num_pages; i++) {
1062 index = start + i; 1075 index = start + i;
1063 1076
1064 if (index >= target_index && 1077 if (index >= target_index && index <= end_index &&
1065 wc->w_type == OCFS2_WRITE 1078 wc->w_type == OCFS2_WRITE_MMAP) {
1066 /* 1079 /*
1067 * ocfs2_pagemkwrite( 1080 * ocfs2_pagemkwrite() is a little different
1068 * and wants us to di 1081 * and wants us to directly use the page
1069 * passed in. 1082 * passed in.
1070 */ 1083 */
1071 lock_page(mmap_page); 1084 lock_page(mmap_page);
1072 1085
1073 /* Exit and let the c 1086 /* Exit and let the caller retry */
1074 if (mmap_page->mappin 1087 if (mmap_page->mapping != mapping) {
1075 WARN_ON(mmap_ 1088 WARN_ON(mmap_page->mapping);
1076 unlock_page(m 1089 unlock_page(mmap_page);
1077 ret = -EAGAIN 1090 ret = -EAGAIN;
1078 goto out; 1091 goto out;
1079 } 1092 }
1080 1093
1081 get_page(mmap_page); 1094 get_page(mmap_page);
1082 wc->w_pages[i] = mmap 1095 wc->w_pages[i] = mmap_page;
1083 wc->w_target_locked = 1096 wc->w_target_locked = true;
1084 } else if (index >= target_in 1097 } else if (index >= target_index && index <= end_index &&
1085 wc->w_type == OCFS 1098 wc->w_type == OCFS2_WRITE_DIRECT) {
1086 /* Direct write has n 1099 /* Direct write has no mapping page. */
1087 wc->w_pages[i] = NULL 1100 wc->w_pages[i] = NULL;
1088 continue; 1101 continue;
1089 } else { 1102 } else {
1090 wc->w_pages[i] = find 1103 wc->w_pages[i] = find_or_create_page(mapping, index,
1091 1104 GFP_NOFS);
1092 if (!wc->w_pages[i]) 1105 if (!wc->w_pages[i]) {
1093 ret = -ENOMEM 1106 ret = -ENOMEM;
1094 mlog_errno(re 1107 mlog_errno(ret);
1095 goto out; 1108 goto out;
1096 } 1109 }
1097 } 1110 }
1098 wait_for_stable_page(wc->w_pa 1111 wait_for_stable_page(wc->w_pages[i]);
1099 1112
1100 if (index == target_index) 1113 if (index == target_index)
1101 wc->w_target_page = w 1114 wc->w_target_page = wc->w_pages[i];
1102 } 1115 }
1103 out: 1116 out:
1104 if (ret) 1117 if (ret)
1105 wc->w_target_locked = false; 1118 wc->w_target_locked = false;
1106 return ret; 1119 return ret;
1107 } 1120 }
1108 1121
1109 /* 1122 /*
1110 * Prepare a single cluster for write one clu 1123 * Prepare a single cluster for write one cluster into the file.
1111 */ 1124 */
1112 static int ocfs2_write_cluster(struct address 1125 static int ocfs2_write_cluster(struct address_space *mapping,
1113 u32 *phys, uns 1126 u32 *phys, unsigned int new,
1114 unsigned int c 1127 unsigned int clear_unwritten,
1115 unsigned int s 1128 unsigned int should_zero,
1116 struct ocfs2_a 1129 struct ocfs2_alloc_context *data_ac,
1117 struct ocfs2_a 1130 struct ocfs2_alloc_context *meta_ac,
1118 struct ocfs2_w 1131 struct ocfs2_write_ctxt *wc, u32 cpos,
1119 loff_t user_po 1132 loff_t user_pos, unsigned user_len)
1120 { 1133 {
1121 int ret, i; 1134 int ret, i;
1122 u64 p_blkno; 1135 u64 p_blkno;
1123 struct inode *inode = mapping->host; 1136 struct inode *inode = mapping->host;
1124 struct ocfs2_extent_tree et; 1137 struct ocfs2_extent_tree et;
1125 int bpc = ocfs2_clusters_to_blocks(in 1138 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1126 1139
1127 if (new) { 1140 if (new) {
1128 u32 tmp_pos; 1141 u32 tmp_pos;
1129 1142
1130 /* 1143 /*
1131 * This is safe to call with 1144 * This is safe to call with the page locks - it won't take
1132 * any additional semaphores 1145 * any additional semaphores or cluster locks.
1133 */ 1146 */
1134 tmp_pos = cpos; 1147 tmp_pos = cpos;
1135 ret = ocfs2_add_inode_data(OC 1148 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1136 &t 1149 &tmp_pos, 1, !clear_unwritten,
1137 wc 1150 wc->w_di_bh, wc->w_handle,
1138 da 1151 data_ac, meta_ac, NULL);
1139 /* 1152 /*
1140 * This shouldn't happen beca 1153 * This shouldn't happen because we must have already
1141 * calculated the correct met 1154 * calculated the correct meta data allocation required. The
1142 * internal tree allocation c 1155 * internal tree allocation code should know how to increase
1143 * transaction credits itself 1156 * transaction credits itself.
1144 * 1157 *
1145 * If need be, we could handl 1158 * If need be, we could handle -EAGAIN for a
1146 * RESTART_TRANS here. 1159 * RESTART_TRANS here.
1147 */ 1160 */
1148 mlog_bug_on_msg(ret == -EAGAI 1161 mlog_bug_on_msg(ret == -EAGAIN,
1149 "Inode %llu: 1162 "Inode %llu: EAGAIN return during allocation.\n",
1150 (unsigned lon 1163 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1151 if (ret < 0) { 1164 if (ret < 0) {
1152 mlog_errno(ret); 1165 mlog_errno(ret);
1153 goto out; 1166 goto out;
1154 } 1167 }
1155 } else if (clear_unwritten) { 1168 } else if (clear_unwritten) {
1156 ocfs2_init_dinode_extent_tree 1169 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1157 1170 wc->w_di_bh);
1158 ret = ocfs2_mark_extent_writt 1171 ret = ocfs2_mark_extent_written(inode, &et,
1159 1172 wc->w_handle, cpos, 1, *phys,
1160 1173 meta_ac, &wc->w_dealloc);
1161 if (ret < 0) { 1174 if (ret < 0) {
1162 mlog_errno(ret); 1175 mlog_errno(ret);
1163 goto out; 1176 goto out;
1164 } 1177 }
1165 } 1178 }
1166 1179
1167 /* 1180 /*
1168 * The only reason this should fail i 1181 * The only reason this should fail is due to an inability to
1169 * find the extent added. 1182 * find the extent added.
1170 */ 1183 */
1171 ret = ocfs2_get_clusters(inode, cpos, 1184 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
1172 if (ret < 0) { 1185 if (ret < 0) {
1173 mlog(ML_ERROR, "Get physical 1186 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1174 "at logical clust 1187 "at logical cluster %u",
1175 (unsigned long lo 1188 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
1176 goto out; 1189 goto out;
1177 } 1190 }
1178 1191
1179 BUG_ON(*phys == 0); 1192 BUG_ON(*phys == 0);
1180 1193
1181 p_blkno = ocfs2_clusters_to_blocks(in 1194 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1182 if (!should_zero) 1195 if (!should_zero)
1183 p_blkno += (user_pos >> inode 1196 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
1184 1197
1185 for(i = 0; i < wc->w_num_pages; i++) 1198 for(i = 0; i < wc->w_num_pages; i++) {
1186 int tmpret; 1199 int tmpret;
1187 1200
1188 /* This is the direct io targ 1201 /* This is the direct io target page. */
1189 if (wc->w_pages[i] == NULL) { 1202 if (wc->w_pages[i] == NULL) {
1190 p_blkno++; 1203 p_blkno++;
1191 continue; 1204 continue;
1192 } 1205 }
1193 1206
1194 tmpret = ocfs2_prepare_page_f 1207 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1195 1208 wc->w_pages[i], cpos,
1196 1209 user_pos, user_len,
1197 1210 should_zero);
1198 if (tmpret) { 1211 if (tmpret) {
1199 mlog_errno(tmpret); 1212 mlog_errno(tmpret);
1200 if (ret == 0) 1213 if (ret == 0)
1201 ret = tmpret; 1214 ret = tmpret;
1202 } 1215 }
1203 } 1216 }
1204 1217
1205 /* 1218 /*
1206 * We only have cleanup to do in case 1219 * We only have cleanup to do in case of allocating write.
1207 */ 1220 */
1208 if (ret && new) 1221 if (ret && new)
1209 ocfs2_write_failure(inode, wc 1222 ocfs2_write_failure(inode, wc, user_pos, user_len);
1210 1223
1211 out: 1224 out:
1212 1225
1213 return ret; 1226 return ret;
1214 } 1227 }
1215 1228
1216 static int ocfs2_write_cluster_by_desc(struct 1229 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1217 struct 1230 struct ocfs2_alloc_context *data_ac,
1218 struct 1231 struct ocfs2_alloc_context *meta_ac,
1219 struct 1232 struct ocfs2_write_ctxt *wc,
1220 loff_t 1233 loff_t pos, unsigned len)
1221 { 1234 {
1222 int ret, i; 1235 int ret, i;
1223 loff_t cluster_off; 1236 loff_t cluster_off;
1224 unsigned int local_len = len; 1237 unsigned int local_len = len;
1225 struct ocfs2_write_cluster_desc *desc 1238 struct ocfs2_write_cluster_desc *desc;
1226 struct ocfs2_super *osb = OCFS2_SB(ma 1239 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1227 1240
1228 for (i = 0; i < wc->w_clen; i++) { 1241 for (i = 0; i < wc->w_clen; i++) {
1229 desc = &wc->w_desc[i]; 1242 desc = &wc->w_desc[i];
1230 1243
1231 /* 1244 /*
1232 * We have to make sure that 1245 * We have to make sure that the total write passed in
1233 * doesn't extend past a sing 1246 * doesn't extend past a single cluster.
1234 */ 1247 */
1235 local_len = len; 1248 local_len = len;
1236 cluster_off = pos & (osb->s_c 1249 cluster_off = pos & (osb->s_clustersize - 1);
1237 if ((cluster_off + local_len) 1250 if ((cluster_off + local_len) > osb->s_clustersize)
1238 local_len = osb->s_cl 1251 local_len = osb->s_clustersize - cluster_off;
1239 1252
1240 ret = ocfs2_write_cluster(map 1253 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
1241 des 1254 desc->c_new,
1242 des 1255 desc->c_clear_unwritten,
1243 des 1256 desc->c_needs_zero,
1244 dat 1257 data_ac, meta_ac,
1245 wc, 1258 wc, desc->c_cpos, pos, local_len);
1246 if (ret) { 1259 if (ret) {
1247 mlog_errno(ret); 1260 mlog_errno(ret);
1248 goto out; 1261 goto out;
1249 } 1262 }
1250 1263
1251 len -= local_len; 1264 len -= local_len;
1252 pos += local_len; 1265 pos += local_len;
1253 } 1266 }
1254 1267
1255 ret = 0; 1268 ret = 0;
1256 out: 1269 out:
1257 return ret; 1270 return ret;
1258 } 1271 }
1259 1272
1260 /* 1273 /*
1261 * ocfs2_write_end() wants to know which part 1274 * ocfs2_write_end() wants to know which parts of the target page it
1262 * should complete the write on. It's easiest 1275 * should complete the write on. It's easiest to compute them ahead of
1263 * time when a more complete view of the writ 1276 * time when a more complete view of the write is available.
1264 */ 1277 */
1265 static void ocfs2_set_target_boundaries(struc 1278 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1266 struc 1279 struct ocfs2_write_ctxt *wc,
1267 loff_ 1280 loff_t pos, unsigned len, int alloc)
1268 { 1281 {
1269 struct ocfs2_write_cluster_desc *desc 1282 struct ocfs2_write_cluster_desc *desc;
1270 1283
1271 wc->w_target_from = pos & (PAGE_SIZE 1284 wc->w_target_from = pos & (PAGE_SIZE - 1);
1272 wc->w_target_to = wc->w_target_from + 1285 wc->w_target_to = wc->w_target_from + len;
1273 1286
1274 if (alloc == 0) 1287 if (alloc == 0)
1275 return; 1288 return;
1276 1289
1277 /* 1290 /*
1278 * Allocating write - we may have dif 1291 * Allocating write - we may have different boundaries based
1279 * on page size and cluster size. 1292 * on page size and cluster size.
1280 * 1293 *
1281 * NOTE: We can no longer compute one 1294 * NOTE: We can no longer compute one value from the other as
1282 * the actual write length and user p 1295 * the actual write length and user provided length may be
1283 * different. 1296 * different.
1284 */ 1297 */
1285 1298
1286 if (wc->w_large_pages) { 1299 if (wc->w_large_pages) {
1287 /* 1300 /*
1288 * We only care about the 1st 1301 * We only care about the 1st and last cluster within
1289 * our range and whether they 1302 * our range and whether they should be zero'd or not. Either
1290 * value may be extended out 1303 * value may be extended out to the start/end of a
1291 * newly allocated cluster. 1304 * newly allocated cluster.
1292 */ 1305 */
1293 desc = &wc->w_desc[0]; 1306 desc = &wc->w_desc[0];
1294 if (desc->c_needs_zero) 1307 if (desc->c_needs_zero)
1295 ocfs2_figure_cluster_ 1308 ocfs2_figure_cluster_boundaries(osb,
1296 1309 desc->c_cpos,
1297 1310 &wc->w_target_from,
1298 1311 NULL);
1299 1312
1300 desc = &wc->w_desc[wc->w_clen 1313 desc = &wc->w_desc[wc->w_clen - 1];
1301 if (desc->c_needs_zero) 1314 if (desc->c_needs_zero)
1302 ocfs2_figure_cluster_ 1315 ocfs2_figure_cluster_boundaries(osb,
1303 1316 desc->c_cpos,
1304 1317 NULL,
1305 1318 &wc->w_target_to);
1306 } else { 1319 } else {
1307 wc->w_target_from = 0; 1320 wc->w_target_from = 0;
1308 wc->w_target_to = PAGE_SIZE; 1321 wc->w_target_to = PAGE_SIZE;
1309 } 1322 }
1310 } 1323 }
1311 1324
1312 /* 1325 /*
1313 * 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
1314 * 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
1315 * by the direct io procedure. 1328 * by the direct io procedure.
1316 * 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
1317 * the ip_unwritten_list. 1330 * the ip_unwritten_list.
1318 */ 1331 */
1319 static int ocfs2_unwritten_check(struct inode 1332 static int ocfs2_unwritten_check(struct inode *inode,
1320 struct ocfs2 1333 struct ocfs2_write_ctxt *wc,
1321 struct ocfs2 1334 struct ocfs2_write_cluster_desc *desc)
1322 { 1335 {
1323 struct ocfs2_inode_info *oi = OCFS2_I 1336 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1324 struct ocfs2_unwritten_extent *ue = N 1337 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
1325 int ret = 0; 1338 int ret = 0;
1326 1339
1327 if (!desc->c_needs_zero) 1340 if (!desc->c_needs_zero)
1328 return 0; 1341 return 0;
1329 1342
1330 retry: 1343 retry:
1331 spin_lock(&oi->ip_lock); 1344 spin_lock(&oi->ip_lock);
1332 /* Needs not to zero no metter buffer 1345 /* Needs not to zero no metter buffer or direct. The one who is zero
1333 * the cluster is doing zero. And he 1346 * the cluster is doing zero. And he will clear unwritten after all
1334 * cluster io finished. */ 1347 * cluster io finished. */
1335 list_for_each_entry(ue, &oi->ip_unwri 1348 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1336 if (desc->c_cpos == ue->ue_cp 1349 if (desc->c_cpos == ue->ue_cpos) {
1337 BUG_ON(desc->c_new); 1350 BUG_ON(desc->c_new);
1338 desc->c_needs_zero = 1351 desc->c_needs_zero = 0;
1339 desc->c_clear_unwritt 1352 desc->c_clear_unwritten = 0;
1340 goto unlock; 1353 goto unlock;
1341 } 1354 }
1342 } 1355 }
1343 1356
1344 if (wc->w_type != OCFS2_WRITE_DIRECT) 1357 if (wc->w_type != OCFS2_WRITE_DIRECT)
1345 goto unlock; 1358 goto unlock;
1346 1359
1347 if (new == NULL) { 1360 if (new == NULL) {
1348 spin_unlock(&oi->ip_lock); 1361 spin_unlock(&oi->ip_lock);
1349 new = kmalloc(sizeof(struct o 1362 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1350 GFP_NOFS); 1363 GFP_NOFS);
1351 if (new == NULL) { 1364 if (new == NULL) {
1352 ret = -ENOMEM; 1365 ret = -ENOMEM;
1353 goto out; 1366 goto out;
1354 } 1367 }
1355 goto retry; 1368 goto retry;
1356 } 1369 }
1357 /* This direct write will doing zero. 1370 /* This direct write will doing zero. */
1358 new->ue_cpos = desc->c_cpos; 1371 new->ue_cpos = desc->c_cpos;
1359 new->ue_phys = desc->c_phys; 1372 new->ue_phys = desc->c_phys;
1360 desc->c_clear_unwritten = 0; 1373 desc->c_clear_unwritten = 0;
1361 list_add_tail(&new->ue_ip_node, &oi-> 1374 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1362 list_add_tail(&new->ue_node, &wc->w_u 1375 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
1363 wc->w_unwritten_count++; <<
1364 new = NULL; 1376 new = NULL;
1365 unlock: 1377 unlock:
1366 spin_unlock(&oi->ip_lock); 1378 spin_unlock(&oi->ip_lock);
1367 out: 1379 out:
1368 kfree(new); !! 1380 if (new)
>> 1381 kfree(new);
1369 return ret; 1382 return ret;
1370 } 1383 }
1371 1384
1372 /* 1385 /*
1373 * Populate each single-cluster write descrip 1386 * Populate each single-cluster write descriptor in the write context
1374 * with information about the i/o to be done. 1387 * with information about the i/o to be done.
1375 * 1388 *
1376 * Returns the number of clusters that will h 1389 * Returns the number of clusters that will have to be allocated, as
1377 * well as a worst case estimate of the numbe 1390 * well as a worst case estimate of the number of extent records that
1378 * would have to be created during a write to 1391 * would have to be created during a write to an unwritten region.
1379 */ 1392 */
1380 static int ocfs2_populate_write_desc(struct i 1393 static int ocfs2_populate_write_desc(struct inode *inode,
1381 struct o 1394 struct ocfs2_write_ctxt *wc,
1382 unsigned 1395 unsigned int *clusters_to_alloc,
1383 unsigned 1396 unsigned int *extents_to_split)
1384 { 1397 {
1385 int ret; 1398 int ret;
1386 struct ocfs2_write_cluster_desc *desc 1399 struct ocfs2_write_cluster_desc *desc;
1387 unsigned int num_clusters = 0; 1400 unsigned int num_clusters = 0;
1388 unsigned int ext_flags = 0; 1401 unsigned int ext_flags = 0;
1389 u32 phys = 0; 1402 u32 phys = 0;
1390 int i; 1403 int i;
1391 1404
1392 *clusters_to_alloc = 0; 1405 *clusters_to_alloc = 0;
1393 *extents_to_split = 0; 1406 *extents_to_split = 0;
1394 1407
1395 for (i = 0; i < wc->w_clen; i++) { 1408 for (i = 0; i < wc->w_clen; i++) {
1396 desc = &wc->w_desc[i]; 1409 desc = &wc->w_desc[i];
1397 desc->c_cpos = wc->w_cpos + i 1410 desc->c_cpos = wc->w_cpos + i;
1398 1411
1399 if (num_clusters == 0) { 1412 if (num_clusters == 0) {
1400 /* 1413 /*
1401 * Need to look up th 1414 * Need to look up the next extent record.
1402 */ 1415 */
1403 ret = ocfs2_get_clust 1416 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1404 1417 &num_clusters, &ext_flags);
1405 if (ret) { 1418 if (ret) {
1406 mlog_errno(re 1419 mlog_errno(ret);
1407 goto out; 1420 goto out;
1408 } 1421 }
1409 1422
1410 /* We should already 1423 /* We should already CoW the refcountd extent. */
1411 BUG_ON(ext_flags & OC 1424 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1412 1425
1413 /* 1426 /*
1414 * Assume worst case 1427 * Assume worst case - that we're writing in
1415 * the middle of the 1428 * the middle of the extent.
1416 * 1429 *
1417 * We can assume that 1430 * We can assume that the write proceeds from
1418 * left to right, in 1431 * left to right, in which case the extent
1419 * insert code is sma 1432 * insert code is smart enough to coalesce the
1420 * next splits into t 1433 * next splits into the previous records created.
1421 */ 1434 */
1422 if (ext_flags & OCFS2 1435 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1423 *extents_to_s 1436 *extents_to_split = *extents_to_split + 2;
1424 } else if (phys) { 1437 } else if (phys) {
1425 /* 1438 /*
1426 * Only increment phy 1439 * Only increment phys if it doesn't describe
1427 * a hole. 1440 * a hole.
1428 */ 1441 */
1429 phys++; 1442 phys++;
1430 } 1443 }
1431 1444
1432 /* 1445 /*
1433 * If w_first_new_cpos is < U 1446 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1434 * file that got extended. w 1447 * file that got extended. w_first_new_cpos tells us
1435 * where the newly allocated 1448 * where the newly allocated clusters are so we can
1436 * zero them. 1449 * zero them.
1437 */ 1450 */
1438 if (desc->c_cpos >= wc->w_fir 1451 if (desc->c_cpos >= wc->w_first_new_cpos) {
1439 BUG_ON(phys == 0); 1452 BUG_ON(phys == 0);
1440 desc->c_needs_zero = 1453 desc->c_needs_zero = 1;
1441 } 1454 }
1442 1455
1443 desc->c_phys = phys; 1456 desc->c_phys = phys;
1444 if (phys == 0) { 1457 if (phys == 0) {
1445 desc->c_new = 1; 1458 desc->c_new = 1;
1446 desc->c_needs_zero = 1459 desc->c_needs_zero = 1;
1447 desc->c_clear_unwritt 1460 desc->c_clear_unwritten = 1;
1448 *clusters_to_alloc = 1461 *clusters_to_alloc = *clusters_to_alloc + 1;
1449 } 1462 }
1450 1463
1451 if (ext_flags & OCFS2_EXT_UNW 1464 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1452 desc->c_clear_unwritt 1465 desc->c_clear_unwritten = 1;
1453 desc->c_needs_zero = 1466 desc->c_needs_zero = 1;
1454 } 1467 }
1455 1468
1456 ret = ocfs2_unwritten_check(i 1469 ret = ocfs2_unwritten_check(inode, wc, desc);
1457 if (ret) { 1470 if (ret) {
1458 mlog_errno(ret); 1471 mlog_errno(ret);
1459 goto out; 1472 goto out;
1460 } 1473 }
1461 1474
1462 num_clusters--; 1475 num_clusters--;
1463 } 1476 }
1464 1477
1465 ret = 0; 1478 ret = 0;
1466 out: 1479 out:
1467 return ret; 1480 return ret;
1468 } 1481 }
1469 1482
1470 static int ocfs2_write_begin_inline(struct ad 1483 static int ocfs2_write_begin_inline(struct address_space *mapping,
1471 struct in 1484 struct inode *inode,
1472 struct oc 1485 struct ocfs2_write_ctxt *wc)
1473 { 1486 {
1474 int ret; 1487 int ret;
1475 struct ocfs2_super *osb = OCFS2_SB(in 1488 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1476 struct page *page; 1489 struct page *page;
1477 handle_t *handle; 1490 handle_t *handle;
1478 struct ocfs2_dinode *di = (struct ocf 1491 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1479 1492
1480 handle = ocfs2_start_trans(osb, OCFS2 1493 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1481 if (IS_ERR(handle)) { 1494 if (IS_ERR(handle)) {
1482 ret = PTR_ERR(handle); 1495 ret = PTR_ERR(handle);
1483 mlog_errno(ret); 1496 mlog_errno(ret);
1484 goto out; 1497 goto out;
1485 } 1498 }
1486 1499
1487 page = find_or_create_page(mapping, 0 1500 page = find_or_create_page(mapping, 0, GFP_NOFS);
1488 if (!page) { 1501 if (!page) {
1489 ocfs2_commit_trans(osb, handl 1502 ocfs2_commit_trans(osb, handle);
1490 ret = -ENOMEM; 1503 ret = -ENOMEM;
1491 mlog_errno(ret); 1504 mlog_errno(ret);
1492 goto out; 1505 goto out;
1493 } 1506 }
1494 /* 1507 /*
1495 * 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
1496 * and freed on cleanup of the write 1509 * and freed on cleanup of the write context.
1497 */ 1510 */
1498 wc->w_pages[0] = wc->w_target_page = 1511 wc->w_pages[0] = wc->w_target_page = page;
1499 wc->w_num_pages = 1; 1512 wc->w_num_pages = 1;
1500 1513
1501 ret = ocfs2_journal_access_di(handle, 1514 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1502 OCFS2_J 1515 OCFS2_JOURNAL_ACCESS_WRITE);
1503 if (ret) { 1516 if (ret) {
1504 ocfs2_commit_trans(osb, handl 1517 ocfs2_commit_trans(osb, handle);
1505 1518
1506 mlog_errno(ret); 1519 mlog_errno(ret);
1507 goto out; 1520 goto out;
1508 } 1521 }
1509 1522
1510 if (!(OCFS2_I(inode)->ip_dyn_features 1523 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1511 ocfs2_set_inode_data_inline(i 1524 ocfs2_set_inode_data_inline(inode, di);
1512 1525
1513 if (!PageUptodate(page)) { 1526 if (!PageUptodate(page)) {
1514 ret = ocfs2_read_inline_data( 1527 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1515 if (ret) { 1528 if (ret) {
1516 ocfs2_commit_trans(os 1529 ocfs2_commit_trans(osb, handle);
1517 1530
1518 goto out; 1531 goto out;
1519 } 1532 }
1520 } 1533 }
1521 1534
1522 wc->w_handle = handle; 1535 wc->w_handle = handle;
1523 out: 1536 out:
1524 return ret; 1537 return ret;
1525 } 1538 }
1526 1539
1527 int ocfs2_size_fits_inline_data(struct buffer 1540 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1528 { 1541 {
1529 struct ocfs2_dinode *di = (struct ocf 1542 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1530 1543
1531 if (new_size <= le16_to_cpu(di->id2.i 1544 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1532 return 1; 1545 return 1;
1533 return 0; 1546 return 0;
1534 } 1547 }
1535 1548
1536 static int ocfs2_try_to_write_inline_data(str 1549 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1537 str 1550 struct inode *inode, loff_t pos,
1538 uns 1551 unsigned len, struct page *mmap_page,
1539 str 1552 struct ocfs2_write_ctxt *wc)
1540 { 1553 {
1541 int ret, written = 0; 1554 int ret, written = 0;
1542 loff_t end = pos + len; 1555 loff_t end = pos + len;
1543 struct ocfs2_inode_info *oi = OCFS2_I 1556 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1544 struct ocfs2_dinode *di = NULL; 1557 struct ocfs2_dinode *di = NULL;
1545 1558
1546 trace_ocfs2_try_to_write_inline_data( 1559 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1547 1560 len, (unsigned long long)pos,
1548 1561 oi->ip_dyn_features);
1549 1562
1550 /* 1563 /*
1551 * Handle inodes which already have i 1564 * Handle inodes which already have inline data 1st.
1552 */ 1565 */
1553 if (oi->ip_dyn_features & OCFS2_INLIN 1566 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1554 if (mmap_page == NULL && 1567 if (mmap_page == NULL &&
1555 ocfs2_size_fits_inline_da 1568 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1556 goto do_inline_write; 1569 goto do_inline_write;
1557 1570
1558 /* 1571 /*
1559 * The write won't fit - we h 1572 * The write won't fit - we have to give this inode an
1560 * inline extent list now. 1573 * inline extent list now.
1561 */ 1574 */
1562 ret = ocfs2_convert_inline_da 1575 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1563 if (ret) 1576 if (ret)
1564 mlog_errno(ret); 1577 mlog_errno(ret);
1565 goto out; 1578 goto out;
1566 } 1579 }
1567 1580
1568 /* 1581 /*
1569 * Check whether the inode can accept 1582 * Check whether the inode can accept inline data.
1570 */ 1583 */
1571 if (oi->ip_clusters != 0 || i_size_re 1584 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1572 return 0; 1585 return 0;
1573 1586
1574 /* 1587 /*
1575 * Check whether the write can fit. 1588 * Check whether the write can fit.
1576 */ 1589 */
1577 di = (struct ocfs2_dinode *)wc->w_di_ 1590 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1578 if (mmap_page || 1591 if (mmap_page ||
1579 end > ocfs2_max_inline_data_with_ 1592 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1580 return 0; 1593 return 0;
1581 1594
1582 do_inline_write: 1595 do_inline_write:
1583 ret = ocfs2_write_begin_inline(mappin 1596 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1584 if (ret) { 1597 if (ret) {
1585 mlog_errno(ret); 1598 mlog_errno(ret);
1586 goto out; 1599 goto out;
1587 } 1600 }
1588 1601
1589 /* 1602 /*
1590 * This signals to the caller that th 1603 * This signals to the caller that the data can be written
1591 * inline. 1604 * inline.
1592 */ 1605 */
1593 written = 1; 1606 written = 1;
1594 out: 1607 out:
1595 return written ? written : ret; 1608 return written ? written : ret;
1596 } 1609 }
1597 1610
1598 /* 1611 /*
1599 * This function only does anything for file 1612 * This function only does anything for file systems which can't
1600 * handle sparse files. 1613 * handle sparse files.
1601 * 1614 *
1602 * 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
1603 * of allocation and the end of our write. Th 1616 * of allocation and the end of our write. That way the rest of the
1604 * write path can treat it as an non-allocati 1617 * write path can treat it as an non-allocating write, which has no
1605 * special case code for sparse/nonsparse fil 1618 * special case code for sparse/nonsparse files.
1606 */ 1619 */
1607 static int ocfs2_expand_nonsparse_inode(struc 1620 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1608 struc 1621 struct buffer_head *di_bh,
1609 loff_ 1622 loff_t pos, unsigned len,
1610 struc 1623 struct ocfs2_write_ctxt *wc)
1611 { 1624 {
1612 int ret; 1625 int ret;
1613 loff_t newsize = pos + len; 1626 loff_t newsize = pos + len;
1614 1627
1615 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1628 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1616 1629
1617 if (newsize <= i_size_read(inode)) 1630 if (newsize <= i_size_read(inode))
1618 return 0; 1631 return 0;
1619 1632
1620 ret = ocfs2_extend_no_holes(inode, di 1633 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1621 if (ret) 1634 if (ret)
1622 mlog_errno(ret); 1635 mlog_errno(ret);
1623 1636
1624 /* There is no wc if this is call fro 1637 /* There is no wc if this is call from direct. */
1625 if (wc) 1638 if (wc)
1626 wc->w_first_new_cpos = 1639 wc->w_first_new_cpos =
1627 ocfs2_clusters_for_by 1640 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1628 1641
1629 return ret; 1642 return ret;
1630 } 1643 }
1631 1644
1632 static int ocfs2_zero_tail(struct inode *inod 1645 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1633 loff_t pos) 1646 loff_t pos)
1634 { 1647 {
1635 int ret = 0; 1648 int ret = 0;
1636 1649
1637 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1650 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1638 if (pos > i_size_read(inode)) 1651 if (pos > i_size_read(inode))
1639 ret = ocfs2_zero_extend(inode 1652 ret = ocfs2_zero_extend(inode, di_bh, pos);
1640 1653
1641 return ret; 1654 return ret;
1642 } 1655 }
1643 1656
1644 int ocfs2_write_begin_nolock(struct address_s 1657 int ocfs2_write_begin_nolock(struct address_space *mapping,
1645 loff_t pos, unsi 1658 loff_t pos, unsigned len, ocfs2_write_type_t type,
1646 struct page **pa 1659 struct page **pagep, void **fsdata,
1647 struct buffer_he 1660 struct buffer_head *di_bh, struct page *mmap_page)
1648 { 1661 {
1649 int ret, cluster_of_pages, credits = 1662 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1650 unsigned int clusters_to_alloc, exten 1663 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1651 struct ocfs2_write_ctxt *wc; 1664 struct ocfs2_write_ctxt *wc;
1652 struct inode *inode = mapping->host; 1665 struct inode *inode = mapping->host;
1653 struct ocfs2_super *osb = OCFS2_SB(in 1666 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1654 struct ocfs2_dinode *di; 1667 struct ocfs2_dinode *di;
1655 struct ocfs2_alloc_context *data_ac = 1668 struct ocfs2_alloc_context *data_ac = NULL;
1656 struct ocfs2_alloc_context *meta_ac = 1669 struct ocfs2_alloc_context *meta_ac = NULL;
1657 handle_t *handle; 1670 handle_t *handle;
1658 struct ocfs2_extent_tree et; 1671 struct ocfs2_extent_tree et;
1659 int try_free = 1, ret1; 1672 int try_free = 1, ret1;
1660 1673
1661 try_again: 1674 try_again:
1662 ret = ocfs2_alloc_write_ctxt(&wc, osb 1675 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
1663 if (ret) { 1676 if (ret) {
1664 mlog_errno(ret); 1677 mlog_errno(ret);
1665 return ret; 1678 return ret;
1666 } 1679 }
1667 1680
1668 if (ocfs2_supports_inline_data(osb)) 1681 if (ocfs2_supports_inline_data(osb)) {
1669 ret = ocfs2_try_to_write_inli 1682 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1670 1683 mmap_page, wc);
1671 if (ret == 1) { 1684 if (ret == 1) {
1672 ret = 0; 1685 ret = 0;
1673 goto success; 1686 goto success;
1674 } 1687 }
1675 if (ret < 0) { 1688 if (ret < 0) {
1676 mlog_errno(ret); 1689 mlog_errno(ret);
1677 goto out; 1690 goto out;
1678 } 1691 }
1679 } 1692 }
1680 1693
1681 /* Direct io change i_size late, shou 1694 /* Direct io change i_size late, should not zero tail here. */
1682 if (type != OCFS2_WRITE_DIRECT) { 1695 if (type != OCFS2_WRITE_DIRECT) {
1683 if (ocfs2_sparse_alloc(osb)) 1696 if (ocfs2_sparse_alloc(osb))
1684 ret = ocfs2_zero_tail 1697 ret = ocfs2_zero_tail(inode, di_bh, pos);
1685 else 1698 else
1686 ret = ocfs2_expand_no 1699 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1687 1700 len, wc);
1688 if (ret) { 1701 if (ret) {
1689 mlog_errno(ret); 1702 mlog_errno(ret);
1690 goto out; 1703 goto out;
1691 } 1704 }
1692 } 1705 }
1693 1706
1694 ret = ocfs2_check_range_for_refcount( 1707 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1695 if (ret < 0) { 1708 if (ret < 0) {
1696 mlog_errno(ret); 1709 mlog_errno(ret);
1697 goto out; 1710 goto out;
1698 } else if (ret == 1) { 1711 } else if (ret == 1) {
1699 clusters_need = wc->w_clen; 1712 clusters_need = wc->w_clen;
1700 ret = ocfs2_refcount_cow(inod 1713 ret = ocfs2_refcount_cow(inode, di_bh,
1701 wc-> 1714 wc->w_cpos, wc->w_clen, UINT_MAX);
1702 if (ret) { 1715 if (ret) {
1703 mlog_errno(ret); 1716 mlog_errno(ret);
1704 goto out; 1717 goto out;
1705 } 1718 }
1706 } 1719 }
1707 1720
1708 ret = ocfs2_populate_write_desc(inode 1721 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1709 &exte 1722 &extents_to_split);
1710 if (ret) { 1723 if (ret) {
1711 mlog_errno(ret); 1724 mlog_errno(ret);
1712 goto out; 1725 goto out;
1713 } 1726 }
1714 clusters_need += clusters_to_alloc; 1727 clusters_need += clusters_to_alloc;
1715 1728
1716 di = (struct ocfs2_dinode *)wc->w_di_ 1729 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1717 1730
1718 trace_ocfs2_write_begin_nolock( 1731 trace_ocfs2_write_begin_nolock(
1719 (unsigned long long)O 1732 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1720 (long long)i_size_rea 1733 (long long)i_size_read(inode),
1721 le32_to_cpu(di->i_clu 1734 le32_to_cpu(di->i_clusters),
1722 pos, len, type, mmap_ 1735 pos, len, type, mmap_page,
1723 clusters_to_alloc, ex 1736 clusters_to_alloc, extents_to_split);
1724 1737
1725 /* 1738 /*
1726 * We set w_target_from, w_target_to 1739 * We set w_target_from, w_target_to here so that
1727 * ocfs2_write_end() knows which rang 1740 * ocfs2_write_end() knows which range in the target page to
1728 * write out. An allocation requires 1741 * write out. An allocation requires that we write the entire
1729 * cluster range. 1742 * cluster range.
1730 */ 1743 */
1731 if (clusters_to_alloc || extents_to_s 1744 if (clusters_to_alloc || extents_to_split) {
1732 /* 1745 /*
1733 * XXX: We are stretching the 1746 * XXX: We are stretching the limits of
1734 * ocfs2_lock_allocators(). I 1747 * ocfs2_lock_allocators(). It greatly over-estimates
1735 * the work to be done. 1748 * the work to be done.
1736 */ 1749 */
1737 ocfs2_init_dinode_extent_tree 1750 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1738 1751 wc->w_di_bh);
1739 ret = ocfs2_lock_allocators(i 1752 ret = ocfs2_lock_allocators(inode, &et,
1740 c 1753 clusters_to_alloc, extents_to_split,
1741 & 1754 &data_ac, &meta_ac);
1742 if (ret) { 1755 if (ret) {
1743 mlog_errno(ret); 1756 mlog_errno(ret);
1744 goto out; 1757 goto out;
1745 } 1758 }
1746 1759
1747 if (data_ac) 1760 if (data_ac)
1748 data_ac->ac_resv = &O 1761 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1749 1762
1750 credits = ocfs2_calc_extend_c 1763 credits = ocfs2_calc_extend_credits(inode->i_sb,
1751 1764 &di->id2.i_list);
1752 } else if (type == OCFS2_WRITE_DIRECT 1765 } else if (type == OCFS2_WRITE_DIRECT)
1753 /* direct write needs not to 1766 /* direct write needs not to start trans if no extents alloc. */
1754 goto success; 1767 goto success;
1755 1768
1756 /* 1769 /*
1757 * We have to zero sparse allocated c 1770 * We have to zero sparse allocated clusters, unwritten extent clusters,
1758 * and non-sparse clusters we just ex 1771 * and non-sparse clusters we just extended. For non-sparse writes,
1759 * we know zeros will only be needed 1772 * we know zeros will only be needed in the first and/or last cluster.
1760 */ 1773 */
1761 if (wc->w_clen && (wc->w_desc[0].c_ne 1774 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1762 wc->w_desc[wc->w_c 1775 wc->w_desc[wc->w_clen - 1].c_needs_zero))
1763 cluster_of_pages = 1; 1776 cluster_of_pages = 1;
1764 else 1777 else
1765 cluster_of_pages = 0; 1778 cluster_of_pages = 0;
1766 1779
1767 ocfs2_set_target_boundaries(osb, wc, 1780 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1768 1781
1769 handle = ocfs2_start_trans(osb, credi 1782 handle = ocfs2_start_trans(osb, credits);
1770 if (IS_ERR(handle)) { 1783 if (IS_ERR(handle)) {
1771 ret = PTR_ERR(handle); 1784 ret = PTR_ERR(handle);
1772 mlog_errno(ret); 1785 mlog_errno(ret);
1773 goto out; 1786 goto out;
1774 } 1787 }
1775 1788
1776 wc->w_handle = handle; 1789 wc->w_handle = handle;
1777 1790
1778 if (clusters_to_alloc) { 1791 if (clusters_to_alloc) {
1779 ret = dquot_alloc_space_nodir 1792 ret = dquot_alloc_space_nodirty(inode,
1780 ocfs2_clusters_to_byt 1793 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1781 if (ret) 1794 if (ret)
1782 goto out_commit; 1795 goto out_commit;
1783 } 1796 }
1784 1797
1785 ret = ocfs2_journal_access_di(handle, 1798 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1786 OCFS2_J 1799 OCFS2_JOURNAL_ACCESS_WRITE);
1787 if (ret) { 1800 if (ret) {
1788 mlog_errno(ret); 1801 mlog_errno(ret);
1789 goto out_quota; 1802 goto out_quota;
1790 } 1803 }
1791 1804
1792 /* 1805 /*
1793 * Fill our page array first. That wa 1806 * Fill our page array first. That way we've grabbed enough so
1794 * that we can zero and flush if we e 1807 * that we can zero and flush if we error after adding the
1795 * extent. 1808 * extent.
1796 */ 1809 */
1797 ret = ocfs2_grab_pages_for_write(mapp 1810 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1798 clus 1811 cluster_of_pages, mmap_page);
1799 if (ret) { !! 1812 if (ret && ret != -EAGAIN) {
1800 /* <<
1801 * ocfs2_grab_pages_for_write <<
1802 * the target page. In this c <<
1803 * page. This will trigger th <<
1804 * the operation. <<
1805 */ <<
1806 if (type == OCFS2_WRITE_MMAP <<
1807 BUG_ON(wc->w_target_p <<
1808 ret = 0; <<
1809 goto out_quota; <<
1810 } <<
1811 <<
1812 mlog_errno(ret); 1813 mlog_errno(ret);
1813 goto out_quota; 1814 goto out_quota;
1814 } 1815 }
1815 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
1816 ret = ocfs2_write_cluster_by_desc(map 1829 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1817 len 1830 len);
1818 if (ret) { 1831 if (ret) {
1819 mlog_errno(ret); 1832 mlog_errno(ret);
1820 goto out_quota; 1833 goto out_quota;
1821 } 1834 }
1822 1835
1823 if (data_ac) 1836 if (data_ac)
1824 ocfs2_free_alloc_context(data 1837 ocfs2_free_alloc_context(data_ac);
1825 if (meta_ac) 1838 if (meta_ac)
1826 ocfs2_free_alloc_context(meta 1839 ocfs2_free_alloc_context(meta_ac);
1827 1840
1828 success: 1841 success:
1829 if (pagep) 1842 if (pagep)
1830 *pagep = wc->w_target_page; 1843 *pagep = wc->w_target_page;
1831 *fsdata = wc; 1844 *fsdata = wc;
1832 return 0; 1845 return 0;
1833 out_quota: 1846 out_quota:
1834 if (clusters_to_alloc) 1847 if (clusters_to_alloc)
1835 dquot_free_space(inode, 1848 dquot_free_space(inode,
1836 ocfs2_clusters_to_b 1849 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1837 out_commit: 1850 out_commit:
1838 ocfs2_commit_trans(osb, handle); 1851 ocfs2_commit_trans(osb, handle);
1839 1852
1840 out: 1853 out:
1841 /* 1854 /*
1842 * The mmapped page won't be unlocked 1855 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1843 * even in case of error here like EN 1856 * even in case of error here like ENOSPC and ENOMEM. So, we need
1844 * to unlock the target page manually 1857 * to unlock the target page manually to prevent deadlocks when
1845 * retrying again on ENOSPC, or when 1858 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1846 * to VM code. 1859 * to VM code.
1847 */ 1860 */
1848 if (wc->w_target_locked) 1861 if (wc->w_target_locked)
1849 unlock_page(mmap_page); 1862 unlock_page(mmap_page);
1850 1863
1851 ocfs2_free_write_ctxt(inode, wc); 1864 ocfs2_free_write_ctxt(inode, wc);
1852 1865
1853 if (data_ac) { 1866 if (data_ac) {
1854 ocfs2_free_alloc_context(data 1867 ocfs2_free_alloc_context(data_ac);
1855 data_ac = NULL; 1868 data_ac = NULL;
1856 } 1869 }
1857 if (meta_ac) { 1870 if (meta_ac) {
1858 ocfs2_free_alloc_context(meta 1871 ocfs2_free_alloc_context(meta_ac);
1859 meta_ac = NULL; 1872 meta_ac = NULL;
1860 } 1873 }
1861 1874
1862 if (ret == -ENOSPC && try_free) { 1875 if (ret == -ENOSPC && try_free) {
1863 /* 1876 /*
1864 * Try to free some truncate 1877 * Try to free some truncate log so that we can have enough
1865 * clusters to allocate. 1878 * clusters to allocate.
1866 */ 1879 */
1867 try_free = 0; 1880 try_free = 0;
1868 1881
1869 ret1 = ocfs2_try_to_free_trun 1882 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1870 if (ret1 == 1) 1883 if (ret1 == 1)
1871 goto try_again; 1884 goto try_again;
1872 1885
1873 if (ret1 < 0) 1886 if (ret1 < 0)
1874 mlog_errno(ret1); 1887 mlog_errno(ret1);
1875 } 1888 }
1876 1889
1877 return ret; 1890 return ret;
1878 } 1891 }
1879 1892
1880 static int ocfs2_write_begin(struct file *fil 1893 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1881 loff_t pos, unsi !! 1894 loff_t pos, unsigned len, unsigned flags,
1882 struct page **pa 1895 struct page **pagep, void **fsdata)
1883 { 1896 {
1884 int ret; 1897 int ret;
1885 struct buffer_head *di_bh = NULL; 1898 struct buffer_head *di_bh = NULL;
1886 struct inode *inode = mapping->host; 1899 struct inode *inode = mapping->host;
1887 1900
1888 ret = ocfs2_inode_lock(inode, &di_bh, 1901 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1889 if (ret) { 1902 if (ret) {
1890 mlog_errno(ret); 1903 mlog_errno(ret);
1891 return ret; 1904 return ret;
1892 } 1905 }
1893 1906
1894 /* 1907 /*
1895 * Take alloc sem here to prevent con 1908 * Take alloc sem here to prevent concurrent lookups. That way
1896 * the mapping, zeroing and tree mani 1909 * the mapping, zeroing and tree manipulation within
1897 * ocfs2_write() will be safe against !! 1910 * ocfs2_write() will be safe against ->readpage(). This
1898 * should also serve to lock out allo 1911 * should also serve to lock out allocation from a shared
1899 * writeable region. 1912 * writeable region.
1900 */ 1913 */
1901 down_write(&OCFS2_I(inode)->ip_alloc_ 1914 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1902 1915
1903 ret = ocfs2_write_begin_nolock(mappin 1916 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1904 pagep, 1917 pagep, fsdata, di_bh, NULL);
1905 if (ret) { 1918 if (ret) {
1906 mlog_errno(ret); 1919 mlog_errno(ret);
1907 goto out_fail; 1920 goto out_fail;
1908 } 1921 }
1909 1922
1910 brelse(di_bh); 1923 brelse(di_bh);
1911 1924
1912 return 0; 1925 return 0;
1913 1926
1914 out_fail: 1927 out_fail:
1915 up_write(&OCFS2_I(inode)->ip_alloc_se 1928 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1916 1929
1917 brelse(di_bh); 1930 brelse(di_bh);
1918 ocfs2_inode_unlock(inode, 1); 1931 ocfs2_inode_unlock(inode, 1);
1919 1932
1920 return ret; 1933 return ret;
1921 } 1934 }
1922 1935
1923 static void ocfs2_write_end_inline(struct ino 1936 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1924 unsigned l 1937 unsigned len, unsigned *copied,
1925 struct ocf 1938 struct ocfs2_dinode *di,
1926 struct ocf 1939 struct ocfs2_write_ctxt *wc)
1927 { 1940 {
1928 void *kaddr; 1941 void *kaddr;
1929 1942
1930 if (unlikely(*copied < len)) { 1943 if (unlikely(*copied < len)) {
1931 if (!PageUptodate(wc->w_targe 1944 if (!PageUptodate(wc->w_target_page)) {
1932 *copied = 0; 1945 *copied = 0;
1933 return; 1946 return;
1934 } 1947 }
1935 } 1948 }
1936 1949
1937 kaddr = kmap_atomic(wc->w_target_page 1950 kaddr = kmap_atomic(wc->w_target_page);
1938 memcpy(di->id2.i_data.id_data + pos, 1951 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1939 kunmap_atomic(kaddr); 1952 kunmap_atomic(kaddr);
1940 1953
1941 trace_ocfs2_write_end_inline( 1954 trace_ocfs2_write_end_inline(
1942 (unsigned long long)OCFS2_I(inod 1955 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1943 (unsigned long long)pos, *copied 1956 (unsigned long long)pos, *copied,
1944 le16_to_cpu(di->id2.i_data.id_co 1957 le16_to_cpu(di->id2.i_data.id_count),
1945 le16_to_cpu(di->i_dyn_features)) 1958 le16_to_cpu(di->i_dyn_features));
1946 } 1959 }
1947 1960
1948 int ocfs2_write_end_nolock(struct address_spa 1961 int ocfs2_write_end_nolock(struct address_space *mapping,
1949 loff_t pos, unsign 1962 loff_t pos, unsigned len, unsigned copied, void *fsdata)
1950 { 1963 {
1951 int i, ret; 1964 int i, ret;
1952 unsigned from, to, start = pos & (PAG 1965 unsigned from, to, start = pos & (PAGE_SIZE - 1);
1953 struct inode *inode = mapping->host; 1966 struct inode *inode = mapping->host;
1954 struct ocfs2_super *osb = OCFS2_SB(in 1967 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1955 struct ocfs2_write_ctxt *wc = fsdata; 1968 struct ocfs2_write_ctxt *wc = fsdata;
1956 struct ocfs2_dinode *di = (struct ocf 1969 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1957 handle_t *handle = wc->w_handle; 1970 handle_t *handle = wc->w_handle;
1958 struct page *tmppage; 1971 struct page *tmppage;
1959 1972
1960 BUG_ON(!list_empty(&wc->w_unwritten_l 1973 BUG_ON(!list_empty(&wc->w_unwritten_list));
1961 1974
1962 if (handle) { 1975 if (handle) {
1963 ret = ocfs2_journal_access_di 1976 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1964 wc->w_di_bh, 1977 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1965 if (ret) { 1978 if (ret) {
1966 copied = ret; 1979 copied = ret;
1967 mlog_errno(ret); 1980 mlog_errno(ret);
1968 goto out; 1981 goto out;
1969 } 1982 }
1970 } 1983 }
1971 1984
1972 if (OCFS2_I(inode)->ip_dyn_features & 1985 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1973 ocfs2_write_end_inline(inode, 1986 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1974 goto out_write_size; 1987 goto out_write_size;
1975 } 1988 }
1976 1989
1977 if (unlikely(copied < len) && wc->w_t 1990 if (unlikely(copied < len) && wc->w_target_page) {
1978 loff_t new_isize; <<
1979 <<
1980 if (!PageUptodate(wc->w_targe 1991 if (!PageUptodate(wc->w_target_page))
1981 copied = 0; 1992 copied = 0;
1982 1993
1983 new_isize = max_t(loff_t, i_s !! 1994 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1984 if (new_isize > page_offset(w !! 1995 start+len);
1985 ocfs2_zero_new_buffer <<
1986 <<
1987 else { <<
1988 /* <<
1989 * When page is fully <<
1990 * failed), do not bo <<
1991 * it instead so that <<
1992 * put page & buffer <<
1993 * state. <<
1994 */ <<
1995 block_invalidate_foli <<
1996 <<
1997 } <<
1998 } 1996 }
1999 if (wc->w_target_page) 1997 if (wc->w_target_page)
2000 flush_dcache_page(wc->w_targe 1998 flush_dcache_page(wc->w_target_page);
2001 1999
2002 for(i = 0; i < wc->w_num_pages; i++) 2000 for(i = 0; i < wc->w_num_pages; i++) {
2003 tmppage = wc->w_pages[i]; 2001 tmppage = wc->w_pages[i];
2004 2002
2005 /* This is the direct io targ 2003 /* This is the direct io target page. */
2006 if (tmppage == NULL) 2004 if (tmppage == NULL)
2007 continue; 2005 continue;
2008 2006
2009 if (tmppage == wc->w_target_p 2007 if (tmppage == wc->w_target_page) {
2010 from = wc->w_target_f 2008 from = wc->w_target_from;
2011 to = wc->w_target_to; 2009 to = wc->w_target_to;
2012 2010
2013 BUG_ON(from > PAGE_SI 2011 BUG_ON(from > PAGE_SIZE ||
2014 to > PAGE_SIZE 2012 to > PAGE_SIZE ||
2015 to < from); 2013 to < from);
2016 } else { 2014 } else {
2017 /* 2015 /*
2018 * Pages adjacent to 2016 * Pages adjacent to the target (if any) imply
2019 * a hole-filling wri 2017 * a hole-filling write in which case we want
2020 * to flush their ent 2018 * to flush their entire range.
2021 */ 2019 */
2022 from = 0; 2020 from = 0;
2023 to = PAGE_SIZE; 2021 to = PAGE_SIZE;
2024 } 2022 }
2025 2023
2026 if (page_has_buffers(tmppage) 2024 if (page_has_buffers(tmppage)) {
2027 if (handle && ocfs2_s !! 2025 if (handle && ocfs2_should_order_data(inode))
2028 loff_t start_ !! 2026 ocfs2_jbd2_file_inode(handle, inode);
2029 ((lof <<
2030 from; <<
2031 loff_t length <<
2032 ocfs2_jbd2_in <<
2033 <<
2034 } <<
2035 block_commit_write(tm 2027 block_commit_write(tmppage, from, to);
2036 } 2028 }
2037 } 2029 }
2038 2030
2039 out_write_size: 2031 out_write_size:
2040 /* Direct io do not update i_size her 2032 /* Direct io do not update i_size here. */
2041 if (wc->w_type != OCFS2_WRITE_DIRECT) 2033 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2042 pos += copied; 2034 pos += copied;
2043 if (pos > i_size_read(inode)) 2035 if (pos > i_size_read(inode)) {
2044 i_size_write(inode, p 2036 i_size_write(inode, pos);
2045 mark_inode_dirty(inod 2037 mark_inode_dirty(inode);
2046 } 2038 }
2047 inode->i_blocks = ocfs2_inode 2039 inode->i_blocks = ocfs2_inode_sector_count(inode);
2048 di->i_size = cpu_to_le64((u64 2040 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2049 inode_set_mtime_to_ts(inode, !! 2041 inode->i_mtime = inode->i_ctime = current_time(inode);
2050 di->i_mtime = di->i_ctime = c !! 2042 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2051 di->i_mtime_nsec = di->i_ctim !! 2043 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2052 if (handle) !! 2044 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2053 ocfs2_update_inode_fs <<
2054 } 2045 }
2055 if (handle) 2046 if (handle)
2056 ocfs2_journal_dirty(handle, w 2047 ocfs2_journal_dirty(handle, wc->w_di_bh);
2057 2048
2058 out: 2049 out:
2059 /* unlock pages before dealloc since 2050 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2060 * lock, or it will cause a deadlock 2051 * lock, or it will cause a deadlock since journal commit threads holds
2061 * this lock and will ask for the pag 2052 * this lock and will ask for the page lock when flushing the data.
2062 * put it here to preserve the unlock 2053 * put it here to preserve the unlock order.
2063 */ 2054 */
2064 ocfs2_unlock_pages(wc); 2055 ocfs2_unlock_pages(wc);
2065 2056
2066 if (handle) 2057 if (handle)
2067 ocfs2_commit_trans(osb, handl 2058 ocfs2_commit_trans(osb, handle);
2068 2059
2069 ocfs2_run_deallocs(osb, &wc->w_deallo 2060 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2070 2061
2071 brelse(wc->w_di_bh); 2062 brelse(wc->w_di_bh);
2072 kfree(wc); 2063 kfree(wc);
2073 2064
2074 return copied; 2065 return copied;
2075 } 2066 }
2076 2067
2077 static int ocfs2_write_end(struct file *file, 2068 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2078 loff_t pos, unsign 2069 loff_t pos, unsigned len, unsigned copied,
2079 struct page *page, 2070 struct page *page, void *fsdata)
2080 { 2071 {
2081 int ret; 2072 int ret;
2082 struct inode *inode = mapping->host; 2073 struct inode *inode = mapping->host;
2083 2074
2084 ret = ocfs2_write_end_nolock(mapping, 2075 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata);
2085 2076
2086 up_write(&OCFS2_I(inode)->ip_alloc_se 2077 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2087 ocfs2_inode_unlock(inode, 1); 2078 ocfs2_inode_unlock(inode, 1);
2088 2079
2089 return ret; 2080 return ret;
2090 } 2081 }
2091 2082
2092 struct ocfs2_dio_write_ctxt { 2083 struct ocfs2_dio_write_ctxt {
2093 struct list_head dw_zero_list; 2084 struct list_head dw_zero_list;
2094 unsigned dw_zero_count 2085 unsigned dw_zero_count;
2095 int dw_orphaned; 2086 int dw_orphaned;
2096 pid_t dw_writer_pid 2087 pid_t dw_writer_pid;
2097 }; 2088 };
2098 2089
2099 static struct ocfs2_dio_write_ctxt * 2090 static struct ocfs2_dio_write_ctxt *
2100 ocfs2_dio_alloc_write_ctx(struct buffer_head 2091 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2101 { 2092 {
2102 struct ocfs2_dio_write_ctxt *dwc = NU 2093 struct ocfs2_dio_write_ctxt *dwc = NULL;
2103 2094
2104 if (bh->b_private) 2095 if (bh->b_private)
2105 return bh->b_private; 2096 return bh->b_private;
2106 2097
2107 dwc = kmalloc(sizeof(struct ocfs2_dio 2098 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2108 if (dwc == NULL) 2099 if (dwc == NULL)
2109 return NULL; 2100 return NULL;
2110 INIT_LIST_HEAD(&dwc->dw_zero_list); 2101 INIT_LIST_HEAD(&dwc->dw_zero_list);
2111 dwc->dw_zero_count = 0; 2102 dwc->dw_zero_count = 0;
2112 dwc->dw_orphaned = 0; 2103 dwc->dw_orphaned = 0;
2113 dwc->dw_writer_pid = task_pid_nr(curr 2104 dwc->dw_writer_pid = task_pid_nr(current);
2114 bh->b_private = dwc; 2105 bh->b_private = dwc;
2115 *alloc = 1; 2106 *alloc = 1;
2116 2107
2117 return dwc; 2108 return dwc;
2118 } 2109 }
2119 2110
2120 static void ocfs2_dio_free_write_ctx(struct i 2111 static void ocfs2_dio_free_write_ctx(struct inode *inode,
2121 struct o 2112 struct ocfs2_dio_write_ctxt *dwc)
2122 { 2113 {
2123 ocfs2_free_unwritten_list(inode, &dwc 2114 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2124 kfree(dwc); 2115 kfree(dwc);
2125 } 2116 }
2126 2117
2127 /* 2118 /*
2128 * TODO: Make this into a generic get_blocks 2119 * TODO: Make this into a generic get_blocks function.
2129 * 2120 *
2130 * From do_direct_io in direct-io.c: 2121 * From do_direct_io in direct-io.c:
2131 * "So what we do is to permit the ->get_blo 2122 * "So what we do is to permit the ->get_blocks function to populate
2132 * bh.b_size with the size of IO which is p 2123 * bh.b_size with the size of IO which is permitted at this offset and
2133 * this i_blkbits." 2124 * this i_blkbits."
2134 * 2125 *
2135 * This function is called directly from get_ 2126 * This function is called directly from get_more_blocks in direct-io.c.
2136 * 2127 *
2137 * called like this: dio->get_blocks(dio->ino 2128 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2138 * fs_co 2129 * fs_count, map_bh, dio->rw == WRITE);
2139 */ 2130 */
2140 static int ocfs2_dio_wr_get_block(struct inod !! 2131 static int ocfs2_dio_get_block(struct inode *inode, sector_t iblock,
2141 struct buffer_ 2132 struct buffer_head *bh_result, int create)
2142 { 2133 {
2143 struct ocfs2_super *osb = OCFS2_SB(in 2134 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2144 struct ocfs2_inode_info *oi = OCFS2_I 2135 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2145 struct ocfs2_write_ctxt *wc; 2136 struct ocfs2_write_ctxt *wc;
2146 struct ocfs2_write_cluster_desc *desc 2137 struct ocfs2_write_cluster_desc *desc = NULL;
2147 struct ocfs2_dio_write_ctxt *dwc = NU 2138 struct ocfs2_dio_write_ctxt *dwc = NULL;
2148 struct buffer_head *di_bh = NULL; 2139 struct buffer_head *di_bh = NULL;
2149 u64 p_blkno; 2140 u64 p_blkno;
2150 unsigned int i_blkbits = inode->i_sb- !! 2141 loff_t pos = iblock << inode->i_sb->s_blocksize_bits;
2151 loff_t pos = iblock << i_blkbits; <<
2152 sector_t endblk = (i_size_read(inode) <<
2153 unsigned len, total_len = bh_result-> 2142 unsigned len, total_len = bh_result->b_size;
2154 int ret = 0, first_get_block = 0; 2143 int ret = 0, first_get_block = 0;
2155 2144
2156 len = osb->s_clustersize - (pos & (os 2145 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2157 len = min(total_len, len); 2146 len = min(total_len, len);
2158 2147
2159 /* <<
2160 * bh_result->b_size is count in get_ <<
2161 * "pos" and "end", we need map twice <<
2162 * 1. area in file size, not set NEW; <<
2163 * 2. area out file size, set NEW. <<
2164 * <<
2165 * iblock endblk <<
2166 * |--------|---------|---------|---- <<
2167 * |<-------area in file------->| <<
2168 */ <<
2169 <<
2170 if ((iblock <= endblk) && <<
2171 ((iblock + ((len - 1) >> i_blkbit <<
2172 len = (endblk - iblock + 1) < <<
2173 <<
2174 mlog(0, "get block of %lu at %llu:%u 2148 mlog(0, "get block of %lu at %llu:%u req %u\n",
2175 inode->i_ino, pos, le 2149 inode->i_ino, pos, len, total_len);
2176 2150
2177 /* 2151 /*
2178 * Because we need to change file siz 2152 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2179 * 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
2180 * while file size will be changed. 2154 * while file size will be changed.
2181 */ 2155 */
2182 if (pos + total_len <= i_size_read(in 2156 if (pos + total_len <= i_size_read(inode)) {
2183 !! 2157 down_read(&oi->ip_alloc_sem);
2184 /* This is the fast path for 2158 /* This is the fast path for re-write. */
2185 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
2186 if (buffer_mapped(bh_result) 2163 if (buffer_mapped(bh_result) &&
2187 !buffer_new(bh_result) && 2164 !buffer_new(bh_result) &&
2188 ret == 0) 2165 ret == 0)
2189 goto out; 2166 goto out;
2190 2167
2191 /* Clear state set by ocfs2_g 2168 /* Clear state set by ocfs2_get_block. */
2192 bh_result->b_state = 0; 2169 bh_result->b_state = 0;
2193 } 2170 }
2194 2171
2195 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2172 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2196 if (unlikely(dwc == NULL)) { 2173 if (unlikely(dwc == NULL)) {
2197 ret = -ENOMEM; 2174 ret = -ENOMEM;
2198 mlog_errno(ret); 2175 mlog_errno(ret);
2199 goto out; 2176 goto out;
2200 } 2177 }
2201 2178
2202 if (ocfs2_clusters_for_bytes(inode->i 2179 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2203 ocfs2_clusters_for_bytes(inode->i 2180 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2204 !dwc->dw_orphaned) { 2181 !dwc->dw_orphaned) {
2205 /* 2182 /*
2206 * when we are going to alloc 2183 * when we are going to alloc extents beyond file size, add the
2207 * inode to orphan dir, so we 2184 * inode to orphan dir, so we can recall those spaces when
2208 * system crashed during writ 2185 * system crashed during write.
2209 */ 2186 */
2210 ret = ocfs2_add_inode_to_orph 2187 ret = ocfs2_add_inode_to_orphan(osb, inode);
2211 if (ret < 0) { 2188 if (ret < 0) {
2212 mlog_errno(ret); 2189 mlog_errno(ret);
2213 goto out; 2190 goto out;
2214 } 2191 }
2215 dwc->dw_orphaned = 1; 2192 dwc->dw_orphaned = 1;
2216 } 2193 }
2217 2194
2218 ret = ocfs2_inode_lock(inode, &di_bh, 2195 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2219 if (ret) { 2196 if (ret) {
2220 mlog_errno(ret); 2197 mlog_errno(ret);
2221 goto out; 2198 goto out;
2222 } 2199 }
2223 2200
2224 down_write(&oi->ip_alloc_sem); 2201 down_write(&oi->ip_alloc_sem);
2225 2202
2226 if (first_get_block) { 2203 if (first_get_block) {
2227 if (ocfs2_sparse_alloc(osb)) !! 2204 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
2228 ret = ocfs2_zero_tail 2205 ret = ocfs2_zero_tail(inode, di_bh, pos);
2229 else 2206 else
2230 ret = ocfs2_expand_no 2207 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2231 2208 total_len, NULL);
2232 if (ret < 0) { 2209 if (ret < 0) {
2233 mlog_errno(ret); 2210 mlog_errno(ret);
2234 goto unlock; 2211 goto unlock;
2235 } 2212 }
2236 } 2213 }
2237 2214
2238 ret = ocfs2_write_begin_nolock(inode- 2215 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2239 OCFS2_ 2216 OCFS2_WRITE_DIRECT, NULL,
2240 (void 2217 (void **)&wc, di_bh, NULL);
2241 if (ret) { 2218 if (ret) {
2242 mlog_errno(ret); 2219 mlog_errno(ret);
2243 goto unlock; 2220 goto unlock;
2244 } 2221 }
2245 2222
2246 desc = &wc->w_desc[0]; 2223 desc = &wc->w_desc[0];
2247 2224
2248 p_blkno = ocfs2_clusters_to_blocks(in 2225 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2249 BUG_ON(p_blkno == 0); 2226 BUG_ON(p_blkno == 0);
2250 p_blkno += iblock & (u64)(ocfs2_clust 2227 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2251 2228
2252 map_bh(bh_result, inode->i_sb, p_blkn 2229 map_bh(bh_result, inode->i_sb, p_blkno);
2253 bh_result->b_size = len; 2230 bh_result->b_size = len;
2254 if (desc->c_needs_zero) 2231 if (desc->c_needs_zero)
2255 set_buffer_new(bh_result); 2232 set_buffer_new(bh_result);
2256 2233
2257 if (iblock > endblk) <<
2258 set_buffer_new(bh_result); <<
2259 <<
2260 /* May sleep in end_io. It should not 2234 /* May sleep in end_io. It should not happen in a irq context. So defer
2261 * it to dio work queue. */ 2235 * it to dio work queue. */
2262 set_buffer_defer_completion(bh_result 2236 set_buffer_defer_completion(bh_result);
2263 2237
2264 if (!list_empty(&wc->w_unwritten_list 2238 if (!list_empty(&wc->w_unwritten_list)) {
2265 struct ocfs2_unwritten_extent 2239 struct ocfs2_unwritten_extent *ue = NULL;
2266 2240
2267 ue = list_first_entry(&wc->w_ 2241 ue = list_first_entry(&wc->w_unwritten_list,
2268 struct 2242 struct ocfs2_unwritten_extent,
2269 ue_node 2243 ue_node);
2270 BUG_ON(ue->ue_cpos != desc->c 2244 BUG_ON(ue->ue_cpos != desc->c_cpos);
2271 /* The physical address may b 2245 /* The physical address may be 0, fill it. */
2272 ue->ue_phys = desc->c_phys; 2246 ue->ue_phys = desc->c_phys;
2273 2247
2274 list_splice_tail_init(&wc->w_ 2248 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
2275 dwc->dw_zero_count += wc->w_u !! 2249 dwc->dw_zero_count++;
2276 } 2250 }
2277 2251
2278 ret = ocfs2_write_end_nolock(inode->i 2252 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc);
2279 BUG_ON(ret != len); 2253 BUG_ON(ret != len);
2280 ret = 0; 2254 ret = 0;
2281 unlock: 2255 unlock:
2282 up_write(&oi->ip_alloc_sem); 2256 up_write(&oi->ip_alloc_sem);
2283 ocfs2_inode_unlock(inode, 1); 2257 ocfs2_inode_unlock(inode, 1);
2284 brelse(di_bh); 2258 brelse(di_bh);
2285 out: 2259 out:
>> 2260 if (ret < 0)
>> 2261 ret = -EIO;
2286 return ret; 2262 return ret;
2287 } 2263 }
2288 2264
2289 static int ocfs2_dio_end_io_write(struct inod 2265 static int ocfs2_dio_end_io_write(struct inode *inode,
2290 struct ocfs 2266 struct ocfs2_dio_write_ctxt *dwc,
2291 loff_t offs 2267 loff_t offset,
2292 ssize_t byt 2268 ssize_t bytes)
2293 { 2269 {
2294 struct ocfs2_cached_dealloc_ctxt deal 2270 struct ocfs2_cached_dealloc_ctxt dealloc;
2295 struct ocfs2_extent_tree et; 2271 struct ocfs2_extent_tree et;
2296 struct ocfs2_super *osb = OCFS2_SB(in 2272 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2297 struct ocfs2_inode_info *oi = OCFS2_I 2273 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2298 struct ocfs2_unwritten_extent *ue = N 2274 struct ocfs2_unwritten_extent *ue = NULL;
2299 struct buffer_head *di_bh = NULL; 2275 struct buffer_head *di_bh = NULL;
2300 struct ocfs2_dinode *di; 2276 struct ocfs2_dinode *di;
2301 struct ocfs2_alloc_context *data_ac = 2277 struct ocfs2_alloc_context *data_ac = NULL;
2302 struct ocfs2_alloc_context *meta_ac = 2278 struct ocfs2_alloc_context *meta_ac = NULL;
2303 handle_t *handle = NULL; 2279 handle_t *handle = NULL;
2304 loff_t end = offset + bytes; 2280 loff_t end = offset + bytes;
2305 int ret = 0, credits = 0; !! 2281 int ret = 0, credits = 0, locked = 0;
2306 2282
2307 ocfs2_init_dealloc_ctxt(&dealloc); 2283 ocfs2_init_dealloc_ctxt(&dealloc);
2308 2284
2309 /* We do clear unwritten, delete orph 2285 /* We do clear unwritten, delete orphan, change i_size here. If neither
2310 * of these happen, we can skip all t 2286 * of these happen, we can skip all this. */
2311 if (list_empty(&dwc->dw_zero_list) && 2287 if (list_empty(&dwc->dw_zero_list) &&
2312 end <= i_size_read(inode) && 2288 end <= i_size_read(inode) &&
2313 !dwc->dw_orphaned) 2289 !dwc->dw_orphaned)
2314 goto out; 2290 goto out;
2315 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
2316 ret = ocfs2_inode_lock(inode, &di_bh, 2299 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2317 if (ret < 0) { 2300 if (ret < 0) {
2318 mlog_errno(ret); 2301 mlog_errno(ret);
2319 goto out; 2302 goto out;
2320 } 2303 }
2321 2304
2322 down_write(&oi->ip_alloc_sem); 2305 down_write(&oi->ip_alloc_sem);
2323 2306
2324 /* Delete orphan before acquire i_rws !! 2307 /* Delete orphan before acquire i_mutex. */
2325 if (dwc->dw_orphaned) { 2308 if (dwc->dw_orphaned) {
2326 BUG_ON(dwc->dw_writer_pid != 2309 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2327 2310
2328 end = end > i_size_read(inode 2311 end = end > i_size_read(inode) ? end : 0;
2329 2312
2330 ret = ocfs2_del_inode_from_or 2313 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2331 !!end, end); 2314 !!end, end);
2332 if (ret < 0) 2315 if (ret < 0)
2333 mlog_errno(ret); 2316 mlog_errno(ret);
2334 } 2317 }
2335 2318
2336 di = (struct ocfs2_dinode *)di_bh->b_ 2319 di = (struct ocfs2_dinode *)di_bh->b_data;
2337 2320
2338 ocfs2_init_dinode_extent_tree(&et, IN 2321 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2339 2322
2340 /* Attach dealloc with extent tree in <<
2341 * which are already unlinked from cu <<
2342 * rotation and merging. <<
2343 */ <<
2344 et.et_dealloc = &dealloc; <<
2345 <<
2346 ret = ocfs2_lock_allocators(inode, &e 2323 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2347 &data_ac, 2324 &data_ac, &meta_ac);
2348 if (ret) { 2325 if (ret) {
2349 mlog_errno(ret); 2326 mlog_errno(ret);
2350 goto unlock; 2327 goto unlock;
2351 } 2328 }
2352 2329
2353 credits = ocfs2_calc_extend_credits(i 2330 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2354 2331
2355 handle = ocfs2_start_trans(osb, credi 2332 handle = ocfs2_start_trans(osb, credits);
2356 if (IS_ERR(handle)) { 2333 if (IS_ERR(handle)) {
2357 ret = PTR_ERR(handle); 2334 ret = PTR_ERR(handle);
2358 mlog_errno(ret); 2335 mlog_errno(ret);
2359 goto unlock; 2336 goto unlock;
2360 } 2337 }
2361 ret = ocfs2_journal_access_di(handle, 2338 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2362 OCFS2_J 2339 OCFS2_JOURNAL_ACCESS_WRITE);
2363 if (ret) { 2340 if (ret) {
2364 mlog_errno(ret); 2341 mlog_errno(ret);
2365 goto commit; 2342 goto commit;
2366 } 2343 }
2367 2344
2368 list_for_each_entry(ue, &dwc->dw_zero 2345 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) {
2369 ret = ocfs2_assure_trans_cred <<
2370 if (ret < 0) { <<
2371 mlog_errno(ret); <<
2372 break; <<
2373 } <<
2374 ret = ocfs2_mark_extent_writt 2346 ret = ocfs2_mark_extent_written(inode, &et, handle,
2375 2347 ue->ue_cpos, 1,
2376 2348 ue->ue_phys,
2377 2349 meta_ac, &dealloc);
2378 if (ret < 0) { 2350 if (ret < 0) {
2379 mlog_errno(ret); 2351 mlog_errno(ret);
2380 break; 2352 break;
2381 } 2353 }
2382 } 2354 }
2383 2355
2384 if (end > i_size_read(inode)) { 2356 if (end > i_size_read(inode)) {
2385 ret = ocfs2_set_inode_size(ha 2357 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2386 if (ret < 0) 2358 if (ret < 0)
2387 mlog_errno(ret); 2359 mlog_errno(ret);
2388 } 2360 }
2389 commit: 2361 commit:
2390 ocfs2_commit_trans(osb, handle); 2362 ocfs2_commit_trans(osb, handle);
2391 unlock: 2363 unlock:
2392 up_write(&oi->ip_alloc_sem); 2364 up_write(&oi->ip_alloc_sem);
2393 ocfs2_inode_unlock(inode, 1); 2365 ocfs2_inode_unlock(inode, 1);
2394 brelse(di_bh); 2366 brelse(di_bh);
2395 out: 2367 out:
2396 if (data_ac) 2368 if (data_ac)
2397 ocfs2_free_alloc_context(data 2369 ocfs2_free_alloc_context(data_ac);
2398 if (meta_ac) 2370 if (meta_ac)
2399 ocfs2_free_alloc_context(meta 2371 ocfs2_free_alloc_context(meta_ac);
2400 ocfs2_run_deallocs(osb, &dealloc); 2372 ocfs2_run_deallocs(osb, &dealloc);
>> 2373 if (locked)
>> 2374 inode_unlock(inode);
2401 ocfs2_dio_free_write_ctx(inode, dwc); 2375 ocfs2_dio_free_write_ctx(inode, dwc);
2402 2376
2403 return ret; 2377 return ret;
2404 } 2378 }
2405 2379
2406 /* 2380 /*
2407 * 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
2408 * particularly interested in the aio/dio cas 2382 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2409 * to protect io on one node from truncation 2383 * to protect io on one node from truncation on another.
2410 */ 2384 */
2411 static int ocfs2_dio_end_io(struct kiocb *ioc 2385 static int ocfs2_dio_end_io(struct kiocb *iocb,
2412 loff_t offset, 2386 loff_t offset,
2413 ssize_t bytes, 2387 ssize_t bytes,
2414 void *private) 2388 void *private)
2415 { 2389 {
2416 struct inode *inode = file_inode(iocb 2390 struct inode *inode = file_inode(iocb->ki_filp);
2417 int level; 2391 int level;
2418 int ret = 0; 2392 int ret = 0;
2419 2393
2420 /* this io's submitter should not hav 2394 /* this io's submitter should not have unlocked this before we could */
2421 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2395 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2422 2396
2423 if (bytes <= 0) !! 2397 if (bytes > 0 && private)
2424 mlog_ratelimited(ML_ERROR, "D !! 2398 ret = ocfs2_dio_end_io_write(inode, private, offset, bytes);
2425 (long long)b <<
2426 if (private) { <<
2427 if (bytes > 0) <<
2428 ret = ocfs2_dio_end_i <<
2429 <<
2430 else <<
2431 ocfs2_dio_free_write_ <<
2432 } <<
2433 2399
2434 ocfs2_iocb_clear_rw_locked(iocb); 2400 ocfs2_iocb_clear_rw_locked(iocb);
2435 2401
2436 level = ocfs2_iocb_rw_locked_level(io 2402 level = ocfs2_iocb_rw_locked_level(iocb);
2437 ocfs2_rw_unlock(inode, level); 2403 ocfs2_rw_unlock(inode, level);
2438 return ret; 2404 return ret;
2439 } 2405 }
2440 2406
2441 static ssize_t ocfs2_direct_IO(struct kiocb * 2407 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2442 { 2408 {
2443 struct file *file = iocb->ki_filp; 2409 struct file *file = iocb->ki_filp;
2444 struct inode *inode = file->f_mapping 2410 struct inode *inode = file->f_mapping->host;
2445 struct ocfs2_super *osb = OCFS2_SB(in 2411 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2446 get_block_t *get_block; 2412 get_block_t *get_block;
2447 2413
2448 /* 2414 /*
2449 * Fallback to buffered I/O if we see 2415 * Fallback to buffered I/O if we see an inode without
2450 * extents. 2416 * extents.
2451 */ 2417 */
2452 if (OCFS2_I(inode)->ip_dyn_features & 2418 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2453 return 0; 2419 return 0;
2454 2420
2455 /* Fallback to buffered I/O if we do 2421 /* Fallback to buffered I/O if we do not support append dio. */
2456 if (iocb->ki_pos + iter->count > i_si 2422 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2457 !ocfs2_supports_append_dio(osb)) 2423 !ocfs2_supports_append_dio(osb))
2458 return 0; 2424 return 0;
2459 2425
2460 if (iov_iter_rw(iter) == READ) 2426 if (iov_iter_rw(iter) == READ)
2461 get_block = ocfs2_lock_get_bl !! 2427 get_block = ocfs2_get_block;
2462 else 2428 else
2463 get_block = ocfs2_dio_wr_get_ !! 2429 get_block = ocfs2_dio_get_block;
2464 2430
2465 return __blockdev_direct_IO(iocb, ino 2431 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2466 iter, get 2432 iter, get_block,
2467 ocfs2_dio !! 2433 ocfs2_dio_end_io, NULL, 0);
2468 } 2434 }
2469 2435
2470 const struct address_space_operations ocfs2_a 2436 const struct address_space_operations ocfs2_aops = {
2471 .dirty_folio = block_dirty !! 2437 .readpage = ocfs2_readpage,
2472 .read_folio = ocfs2_read_ !! 2438 .readpages = ocfs2_readpages,
2473 .readahead = ocfs2_reada !! 2439 .writepage = ocfs2_writepage,
2474 .writepages = ocfs2_write <<
2475 .write_begin = ocfs2_write 2440 .write_begin = ocfs2_write_begin,
2476 .write_end = ocfs2_write 2441 .write_end = ocfs2_write_end,
2477 .bmap = ocfs2_bmap, 2442 .bmap = ocfs2_bmap,
2478 .direct_IO = ocfs2_direc 2443 .direct_IO = ocfs2_direct_IO,
2479 .invalidate_folio = block_inval !! 2444 .invalidatepage = block_invalidatepage,
2480 .release_folio = ocfs2_relea !! 2445 .releasepage = ocfs2_releasepage,
2481 .migrate_folio = buffer_migr !! 2446 .migratepage = buffer_migrate_page,
2482 .is_partially_uptodate = block_is_pa 2447 .is_partially_uptodate = block_is_partially_uptodate,
2483 .error_remove_folio = generic_err !! 2448 .error_remove_page = generic_error_remove_page,
2484 }; 2449 };
2485 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.