1 // SPDX-License-Identifier: GPL-2.0-or-later 1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* !! 2 /* -*- mode: c; c-basic-offset: 8; -*-
>> 3 * vim: noexpandtab sw=8 ts=8 sts=0:
>> 4 *
3 * Copyright (C) 2002, 2004 Oracle. All right 5 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
4 */ 6 */
5 7
6 #include <linux/fs.h> 8 #include <linux/fs.h>
7 #include <linux/slab.h> 9 #include <linux/slab.h>
8 #include <linux/highmem.h> 10 #include <linux/highmem.h>
9 #include <linux/pagemap.h> 11 #include <linux/pagemap.h>
10 #include <asm/byteorder.h> 12 #include <asm/byteorder.h>
11 #include <linux/swap.h> 13 #include <linux/swap.h>
12 #include <linux/mpage.h> 14 #include <linux/mpage.h>
13 #include <linux/quotaops.h> 15 #include <linux/quotaops.h>
14 #include <linux/blkdev.h> 16 #include <linux/blkdev.h>
15 #include <linux/uio.h> 17 #include <linux/uio.h>
16 #include <linux/mm.h> 18 #include <linux/mm.h>
17 19
18 #include <cluster/masklog.h> 20 #include <cluster/masklog.h>
19 21
20 #include "ocfs2.h" 22 #include "ocfs2.h"
21 23
22 #include "alloc.h" 24 #include "alloc.h"
23 #include "aops.h" 25 #include "aops.h"
24 #include "dlmglue.h" 26 #include "dlmglue.h"
25 #include "extent_map.h" 27 #include "extent_map.h"
26 #include "file.h" 28 #include "file.h"
27 #include "inode.h" 29 #include "inode.h"
28 #include "journal.h" 30 #include "journal.h"
29 #include "suballoc.h" 31 #include "suballoc.h"
30 #include "super.h" 32 #include "super.h"
31 #include "symlink.h" 33 #include "symlink.h"
32 #include "refcounttree.h" 34 #include "refcounttree.h"
33 #include "ocfs2_trace.h" 35 #include "ocfs2_trace.h"
34 36
35 #include "buffer_head_io.h" 37 #include "buffer_head_io.h"
36 #include "dir.h" 38 #include "dir.h"
37 #include "namei.h" 39 #include "namei.h"
38 #include "sysfile.h" 40 #include "sysfile.h"
39 41
40 static int ocfs2_symlink_get_block(struct inod 42 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
41 struct buff 43 struct buffer_head *bh_result, int create)
42 { 44 {
43 int err = -EIO; 45 int err = -EIO;
44 int status; 46 int status;
45 struct ocfs2_dinode *fe = NULL; 47 struct ocfs2_dinode *fe = NULL;
46 struct buffer_head *bh = NULL; 48 struct buffer_head *bh = NULL;
47 struct buffer_head *buffer_cache_bh = 49 struct buffer_head *buffer_cache_bh = NULL;
48 struct ocfs2_super *osb = OCFS2_SB(ino 50 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
49 void *kaddr; 51 void *kaddr;
50 52
51 trace_ocfs2_symlink_get_block( 53 trace_ocfs2_symlink_get_block(
52 (unsigned long long)OC 54 (unsigned long long)OCFS2_I(inode)->ip_blkno,
53 (unsigned long long)ib 55 (unsigned long long)iblock, bh_result, create);
54 56
55 BUG_ON(ocfs2_inode_is_fast_symlink(ino 57 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
56 58
57 if ((iblock << inode->i_sb->s_blocksiz 59 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
58 mlog(ML_ERROR, "block offset > 60 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
59 (unsigned long long)ibloc 61 (unsigned long long)iblock);
60 goto bail; 62 goto bail;
61 } 63 }
62 64
63 status = ocfs2_read_inode_block(inode, 65 status = ocfs2_read_inode_block(inode, &bh);
64 if (status < 0) { 66 if (status < 0) {
65 mlog_errno(status); 67 mlog_errno(status);
66 goto bail; 68 goto bail;
67 } 69 }
68 fe = (struct ocfs2_dinode *) bh->b_dat 70 fe = (struct ocfs2_dinode *) bh->b_data;
69 71
70 if ((u64)iblock >= ocfs2_clusters_to_b 72 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
71 73 le32_to_cpu(fe->i_clusters))) {
72 err = -ENOMEM; 74 err = -ENOMEM;
73 mlog(ML_ERROR, "block offset i 75 mlog(ML_ERROR, "block offset is outside the allocated size: "
74 "%llu\n", (unsigned long 76 "%llu\n", (unsigned long long)iblock);
75 goto bail; 77 goto bail;
76 } 78 }
77 79
78 /* We don't use the page cache to crea 80 /* We don't use the page cache to create symlink data, so if
79 * need be, copy it over from the buff 81 * need be, copy it over from the buffer cache. */
80 if (!buffer_uptodate(bh_result) && ocf 82 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
81 u64 blkno = le64_to_cpu(fe->id 83 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
82 iblock; 84 iblock;
83 buffer_cache_bh = sb_getblk(os 85 buffer_cache_bh = sb_getblk(osb->sb, blkno);
84 if (!buffer_cache_bh) { 86 if (!buffer_cache_bh) {
85 err = -ENOMEM; 87 err = -ENOMEM;
86 mlog(ML_ERROR, "couldn 88 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
87 goto bail; 89 goto bail;
88 } 90 }
89 91
90 /* we haven't locked out trans 92 /* we haven't locked out transactions, so a commit
91 * could've happened. Since we 93 * could've happened. Since we've got a reference on
92 * the bh, even if it commits 94 * the bh, even if it commits while we're doing the
93 * copy, the data is still goo 95 * copy, the data is still good. */
94 if (buffer_jbd(buffer_cache_bh 96 if (buffer_jbd(buffer_cache_bh)
95 && ocfs2_inode_is_new(inod 97 && ocfs2_inode_is_new(inode)) {
96 kaddr = kmap_atomic(bh 98 kaddr = kmap_atomic(bh_result->b_page);
97 if (!kaddr) { 99 if (!kaddr) {
98 mlog(ML_ERROR, 100 mlog(ML_ERROR, "couldn't kmap!\n");
99 goto bail; 101 goto bail;
100 } 102 }
101 memcpy(kaddr + (bh_res 103 memcpy(kaddr + (bh_result->b_size * iblock),
102 buffer_cache_bh 104 buffer_cache_bh->b_data,
103 bh_result->b_si 105 bh_result->b_size);
104 kunmap_atomic(kaddr); 106 kunmap_atomic(kaddr);
105 set_buffer_uptodate(bh 107 set_buffer_uptodate(bh_result);
106 } 108 }
107 brelse(buffer_cache_bh); 109 brelse(buffer_cache_bh);
108 } 110 }
109 111
110 map_bh(bh_result, inode->i_sb, 112 map_bh(bh_result, inode->i_sb,
111 le64_to_cpu(fe->id2.i_list.l_re 113 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
112 114
113 err = 0; 115 err = 0;
114 116
115 bail: 117 bail:
116 brelse(bh); 118 brelse(bh);
117 119
118 return err; 120 return err;
119 } 121 }
120 122
121 static int ocfs2_lock_get_block(struct inode * 123 static int ocfs2_lock_get_block(struct inode *inode, sector_t iblock,
122 struct buffer_head *bh_res 124 struct buffer_head *bh_result, int create)
123 { 125 {
124 int ret = 0; 126 int ret = 0;
125 struct ocfs2_inode_info *oi = OCFS2_I( 127 struct ocfs2_inode_info *oi = OCFS2_I(inode);
126 128
127 down_read(&oi->ip_alloc_sem); 129 down_read(&oi->ip_alloc_sem);
128 ret = ocfs2_get_block(inode, iblock, b 130 ret = ocfs2_get_block(inode, iblock, bh_result, create);
129 up_read(&oi->ip_alloc_sem); 131 up_read(&oi->ip_alloc_sem);
130 132
131 return ret; 133 return ret;
132 } 134 }
133 135
134 int ocfs2_get_block(struct inode *inode, secto 136 int ocfs2_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh_res 137 struct buffer_head *bh_result, int create)
136 { 138 {
137 int err = 0; 139 int err = 0;
138 unsigned int ext_flags; 140 unsigned int ext_flags;
139 u64 max_blocks = bh_result->b_size >> 141 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
140 u64 p_blkno, count, past_eof; 142 u64 p_blkno, count, past_eof;
141 struct ocfs2_super *osb = OCFS2_SB(ino 143 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
142 144
143 trace_ocfs2_get_block((unsigned long l 145 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
144 (unsigned long l 146 (unsigned long long)iblock, bh_result, create);
145 147
146 if (OCFS2_I(inode)->ip_flags & OCFS2_I 148 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
147 mlog(ML_NOTICE, "get_block on 149 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
148 inode, inode->i_ino); 150 inode, inode->i_ino);
149 151
150 if (S_ISLNK(inode->i_mode)) { 152 if (S_ISLNK(inode->i_mode)) {
151 /* this always does I/O for so 153 /* this always does I/O for some reason. */
152 err = ocfs2_symlink_get_block( 154 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
153 goto bail; 155 goto bail;
154 } 156 }
155 157
156 err = ocfs2_extent_map_get_blocks(inod 158 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
157 &ext 159 &ext_flags);
158 if (err) { 160 if (err) {
159 mlog(ML_ERROR, "get_blocks() f !! 161 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
160 "block: %llu\n", inode, ( !! 162 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
>> 163 (unsigned long long)p_blkno);
161 goto bail; 164 goto bail;
162 } 165 }
163 166
164 if (max_blocks < count) 167 if (max_blocks < count)
165 count = max_blocks; 168 count = max_blocks;
166 169
167 /* 170 /*
168 * ocfs2 never allocates in this funct 171 * ocfs2 never allocates in this function - the only time we
169 * need to use BH_New is when we're ex 172 * need to use BH_New is when we're extending i_size on a file
170 * system which doesn't support holes, 173 * system which doesn't support holes, in which case BH_New
171 * allows __block_write_begin() to zer 174 * allows __block_write_begin() to zero.
172 * 175 *
173 * If we see this on a sparse file sys 176 * If we see this on a sparse file system, then a truncate has
174 * raced us and removed the cluster. I 177 * raced us and removed the cluster. In this case, we clear
175 * the buffers dirty and uptodate bits 178 * the buffers dirty and uptodate bits and let the buffer code
176 * ignore it as a hole. 179 * ignore it as a hole.
177 */ 180 */
178 if (create && p_blkno == 0 && ocfs2_sp 181 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
179 clear_buffer_dirty(bh_result); 182 clear_buffer_dirty(bh_result);
180 clear_buffer_uptodate(bh_resul 183 clear_buffer_uptodate(bh_result);
181 goto bail; 184 goto bail;
182 } 185 }
183 186
184 /* Treat the unwritten extent as a hol 187 /* Treat the unwritten extent as a hole for zeroing purposes. */
185 if (p_blkno && !(ext_flags & OCFS2_EXT 188 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
186 map_bh(bh_result, inode->i_sb, 189 map_bh(bh_result, inode->i_sb, p_blkno);
187 190
188 bh_result->b_size = count << inode->i_ 191 bh_result->b_size = count << inode->i_blkbits;
189 192
190 if (!ocfs2_sparse_alloc(osb)) { 193 if (!ocfs2_sparse_alloc(osb)) {
191 if (p_blkno == 0) { 194 if (p_blkno == 0) {
192 err = -EIO; 195 err = -EIO;
193 mlog(ML_ERROR, 196 mlog(ML_ERROR,
194 "iblock = %llu p_ 197 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
195 (unsigned long lo 198 (unsigned long long)iblock,
196 (unsigned long lo 199 (unsigned long long)p_blkno,
197 (unsigned long lo 200 (unsigned long long)OCFS2_I(inode)->ip_blkno);
198 mlog(ML_ERROR, "Size % 201 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
199 dump_stack(); 202 dump_stack();
200 goto bail; 203 goto bail;
201 } 204 }
202 } 205 }
203 206
204 past_eof = ocfs2_blocks_for_bytes(inod 207 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
205 208
206 trace_ocfs2_get_block_end((unsigned lo 209 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
207 (unsigned lo 210 (unsigned long long)past_eof);
208 if (create && (iblock >= past_eof)) 211 if (create && (iblock >= past_eof))
209 set_buffer_new(bh_result); 212 set_buffer_new(bh_result);
210 213
211 bail: 214 bail:
212 if (err < 0) 215 if (err < 0)
213 err = -EIO; 216 err = -EIO;
214 217
215 return err; 218 return err;
216 } 219 }
217 220
218 int ocfs2_read_inline_data(struct inode *inode 221 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
219 struct buffer_head 222 struct buffer_head *di_bh)
220 { 223 {
221 void *kaddr; 224 void *kaddr;
222 loff_t size; 225 loff_t size;
223 struct ocfs2_dinode *di = (struct ocfs 226 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
224 227
225 if (!(le16_to_cpu(di->i_dyn_features) 228 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
226 ocfs2_error(inode->i_sb, "Inod 229 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
227 (unsigned long lon 230 (unsigned long long)OCFS2_I(inode)->ip_blkno);
228 return -EROFS; 231 return -EROFS;
229 } 232 }
230 233
231 size = i_size_read(inode); 234 size = i_size_read(inode);
232 235
233 if (size > PAGE_SIZE || 236 if (size > PAGE_SIZE ||
234 size > ocfs2_max_inline_data_with_ 237 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
235 ocfs2_error(inode->i_sb, 238 ocfs2_error(inode->i_sb,
236 "Inode %llu has wi 239 "Inode %llu has with inline data has bad size: %Lu\n",
237 (unsigned long lon 240 (unsigned long long)OCFS2_I(inode)->ip_blkno,
238 (unsigned long lon 241 (unsigned long long)size);
239 return -EROFS; 242 return -EROFS;
240 } 243 }
241 244
242 kaddr = kmap_atomic(page); 245 kaddr = kmap_atomic(page);
243 if (size) 246 if (size)
244 memcpy(kaddr, di->id2.i_data.i 247 memcpy(kaddr, di->id2.i_data.id_data, size);
245 /* Clear the remaining part of the pag 248 /* Clear the remaining part of the page */
246 memset(kaddr + size, 0, PAGE_SIZE - si 249 memset(kaddr + size, 0, PAGE_SIZE - size);
247 flush_dcache_page(page); 250 flush_dcache_page(page);
248 kunmap_atomic(kaddr); 251 kunmap_atomic(kaddr);
249 252
250 SetPageUptodate(page); 253 SetPageUptodate(page);
251 254
252 return 0; 255 return 0;
253 } 256 }
254 257
255 static int ocfs2_readpage_inline(struct inode 258 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
256 { 259 {
257 int ret; 260 int ret;
258 struct buffer_head *di_bh = NULL; 261 struct buffer_head *di_bh = NULL;
259 262
260 BUG_ON(!PageLocked(page)); 263 BUG_ON(!PageLocked(page));
261 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 264 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
262 265
263 ret = ocfs2_read_inode_block(inode, &d 266 ret = ocfs2_read_inode_block(inode, &di_bh);
264 if (ret) { 267 if (ret) {
265 mlog_errno(ret); 268 mlog_errno(ret);
266 goto out; 269 goto out;
267 } 270 }
268 271
269 ret = ocfs2_read_inline_data(inode, pa 272 ret = ocfs2_read_inline_data(inode, page, di_bh);
270 out: 273 out:
271 unlock_page(page); 274 unlock_page(page);
272 275
273 brelse(di_bh); 276 brelse(di_bh);
274 return ret; 277 return ret;
275 } 278 }
276 279
277 static int ocfs2_read_folio(struct file *file, !! 280 static int ocfs2_readpage(struct file *file, struct page *page)
278 { 281 {
279 struct inode *inode = folio->mapping-> !! 282 struct inode *inode = page->mapping->host;
280 struct ocfs2_inode_info *oi = OCFS2_I( 283 struct ocfs2_inode_info *oi = OCFS2_I(inode);
281 loff_t start = folio_pos(folio); !! 284 loff_t start = (loff_t)page->index << PAGE_SHIFT;
282 int ret, unlock = 1; 285 int ret, unlock = 1;
283 286
284 trace_ocfs2_readpage((unsigned long lo !! 287 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
>> 288 (page ? page->index : 0));
285 289
286 ret = ocfs2_inode_lock_with_page(inode !! 290 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
287 if (ret != 0) { 291 if (ret != 0) {
288 if (ret == AOP_TRUNCATED_PAGE) 292 if (ret == AOP_TRUNCATED_PAGE)
289 unlock = 0; 293 unlock = 0;
290 mlog_errno(ret); 294 mlog_errno(ret);
291 goto out; 295 goto out;
292 } 296 }
293 297
294 if (down_read_trylock(&oi->ip_alloc_se 298 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
295 /* 299 /*
296 * Unlock the folio and cycle !! 300 * Unlock the page and cycle ip_alloc_sem so that we don't
297 * busyloop waiting for ip_all 301 * busyloop waiting for ip_alloc_sem to unlock
298 */ 302 */
299 ret = AOP_TRUNCATED_PAGE; 303 ret = AOP_TRUNCATED_PAGE;
300 folio_unlock(folio); !! 304 unlock_page(page);
301 unlock = 0; 305 unlock = 0;
302 down_read(&oi->ip_alloc_sem); 306 down_read(&oi->ip_alloc_sem);
303 up_read(&oi->ip_alloc_sem); 307 up_read(&oi->ip_alloc_sem);
304 goto out_inode_unlock; 308 goto out_inode_unlock;
305 } 309 }
306 310
307 /* 311 /*
308 * i_size might have just been updated 312 * i_size might have just been updated as we grabed the meta lock. We
309 * might now be discovering a truncate 313 * might now be discovering a truncate that hit on another node.
310 * block_read_full_folio->get_block fr !! 314 * block_read_full_page->get_block freaks out if it is asked to read
311 * beyond the end of a file, so we che 315 * beyond the end of a file, so we check here. Callers
312 * (generic_file_read, vm_ops->fault) 316 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
313 * and notice that the folio they just !! 317 * and notice that the page they just read isn't needed.
314 * 318 *
315 * XXX sys_readahead() seems to get th 319 * XXX sys_readahead() seems to get that wrong?
316 */ 320 */
317 if (start >= i_size_read(inode)) { 321 if (start >= i_size_read(inode)) {
318 folio_zero_segment(folio, 0, f !! 322 zero_user(page, 0, PAGE_SIZE);
319 folio_mark_uptodate(folio); !! 323 SetPageUptodate(page);
320 ret = 0; 324 ret = 0;
321 goto out_alloc; 325 goto out_alloc;
322 } 326 }
323 327
324 if (oi->ip_dyn_features & OCFS2_INLINE 328 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
325 ret = ocfs2_readpage_inline(in !! 329 ret = ocfs2_readpage_inline(inode, page);
326 else 330 else
327 ret = block_read_full_folio(fo !! 331 ret = block_read_full_page(page, ocfs2_get_block);
328 unlock = 0; 332 unlock = 0;
329 333
330 out_alloc: 334 out_alloc:
331 up_read(&oi->ip_alloc_sem); 335 up_read(&oi->ip_alloc_sem);
332 out_inode_unlock: 336 out_inode_unlock:
333 ocfs2_inode_unlock(inode, 0); 337 ocfs2_inode_unlock(inode, 0);
334 out: 338 out:
335 if (unlock) 339 if (unlock)
336 folio_unlock(folio); !! 340 unlock_page(page);
337 return ret; 341 return ret;
338 } 342 }
339 343
340 /* 344 /*
341 * This is used only for read-ahead. Failures 345 * This is used only for read-ahead. Failures or difficult to handle
342 * situations are safe to ignore. 346 * situations are safe to ignore.
343 * 347 *
344 * Right now, we don't bother with BH_Boundary 348 * Right now, we don't bother with BH_Boundary - in-inode extent lists
345 * are quite large (243 extents on 4k blocks), 349 * are quite large (243 extents on 4k blocks), so most inodes don't
346 * grow out to a tree. If need be, detecting b 350 * grow out to a tree. If need be, detecting boundary extents could
347 * trivially be added in a future version of o 351 * trivially be added in a future version of ocfs2_get_block().
348 */ 352 */
349 static void ocfs2_readahead(struct readahead_c 353 static void ocfs2_readahead(struct readahead_control *rac)
350 { 354 {
351 int ret; 355 int ret;
352 struct inode *inode = rac->mapping->ho 356 struct inode *inode = rac->mapping->host;
353 struct ocfs2_inode_info *oi = OCFS2_I( 357 struct ocfs2_inode_info *oi = OCFS2_I(inode);
354 358
355 /* 359 /*
356 * Use the nonblocking flag for the dl 360 * Use the nonblocking flag for the dlm code to avoid page
357 * lock inversion, but don't bother wi 361 * lock inversion, but don't bother with retrying.
358 */ 362 */
359 ret = ocfs2_inode_lock_full(inode, NUL 363 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
360 if (ret) 364 if (ret)
361 return; 365 return;
362 366
363 if (down_read_trylock(&oi->ip_alloc_se 367 if (down_read_trylock(&oi->ip_alloc_sem) == 0)
364 goto out_unlock; 368 goto out_unlock;
365 369
366 /* 370 /*
367 * Don't bother with inline-data. Ther 371 * Don't bother with inline-data. There isn't anything
368 * to read-ahead in that case anyway.. 372 * to read-ahead in that case anyway...
369 */ 373 */
370 if (oi->ip_dyn_features & OCFS2_INLINE 374 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
371 goto out_up; 375 goto out_up;
372 376
373 /* 377 /*
374 * Check whether a remote node truncat 378 * Check whether a remote node truncated this file - we just
375 * drop out in that case as it's not w 379 * drop out in that case as it's not worth handling here.
376 */ 380 */
377 if (readahead_pos(rac) >= i_size_read( 381 if (readahead_pos(rac) >= i_size_read(inode))
378 goto out_up; 382 goto out_up;
379 383
380 mpage_readahead(rac, ocfs2_get_block); 384 mpage_readahead(rac, ocfs2_get_block);
381 385
382 out_up: 386 out_up:
383 up_read(&oi->ip_alloc_sem); 387 up_read(&oi->ip_alloc_sem);
384 out_unlock: 388 out_unlock:
385 ocfs2_inode_unlock(inode, 0); 389 ocfs2_inode_unlock(inode, 0);
386 } 390 }
387 391
388 /* Note: Because we don't support holes, our a 392 /* Note: Because we don't support holes, our allocation has
389 * already happened (allocation writes zeros t 393 * already happened (allocation writes zeros to the file data)
390 * so we don't have to worry about ordered wri 394 * so we don't have to worry about ordered writes in
391 * ocfs2_writepages. !! 395 * ocfs2_writepage.
392 * 396 *
393 * ->writepages is called during the process o !! 397 * ->writepage is called during the process of invalidating the page cache
394 * during blocked lock processing. It can't b 398 * during blocked lock processing. It can't block on any cluster locks
395 * to during block mapping. It's relying on t 399 * to during block mapping. It's relying on the fact that the block
396 * mapping can't have disappeared under the di 400 * mapping can't have disappeared under the dirty pages that it is
397 * being asked to write back. 401 * being asked to write back.
398 */ 402 */
399 static int ocfs2_writepages(struct address_spa !! 403 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
400 struct writeback_control *wbc) <<
401 { 404 {
402 return mpage_writepages(mapping, wbc, !! 405 trace_ocfs2_writepage(
>> 406 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
>> 407 page->index);
>> 408
>> 409 return block_write_full_page(page, ocfs2_get_block, wbc);
403 } 410 }
404 411
405 /* Taken from ext3. We don't necessarily need 412 /* Taken from ext3. We don't necessarily need the full blown
406 * functionality yet, but IMHO it's better to 413 * functionality yet, but IMHO it's better to cut and paste the whole
407 * thing so we can avoid introducing our own b 414 * thing so we can avoid introducing our own bugs (and easily pick up
408 * their fixes when they happen) --Mark */ 415 * their fixes when they happen) --Mark */
409 int walk_page_buffers( handle_t *handle, 416 int walk_page_buffers( handle_t *handle,
410 struct buffer_head *he 417 struct buffer_head *head,
411 unsigned from, 418 unsigned from,
412 unsigned to, 419 unsigned to,
413 int *partial, 420 int *partial,
414 int (*fn)( handle 421 int (*fn)( handle_t *handle,
415 struct 422 struct buffer_head *bh))
416 { 423 {
417 struct buffer_head *bh; 424 struct buffer_head *bh;
418 unsigned block_start, block_end; 425 unsigned block_start, block_end;
419 unsigned blocksize = head->b_size; 426 unsigned blocksize = head->b_size;
420 int err, ret = 0; 427 int err, ret = 0;
421 struct buffer_head *next; 428 struct buffer_head *next;
422 429
423 for ( bh = head, block_start = 0; 430 for ( bh = head, block_start = 0;
424 ret == 0 && (bh != head || !bl 431 ret == 0 && (bh != head || !block_start);
425 block_start = block_end, bh = 432 block_start = block_end, bh = next)
426 { 433 {
427 next = bh->b_this_page; 434 next = bh->b_this_page;
428 block_end = block_start + bloc 435 block_end = block_start + blocksize;
429 if (block_end <= from || block 436 if (block_end <= from || block_start >= to) {
430 if (partial && !buffer 437 if (partial && !buffer_uptodate(bh))
431 *partial = 1; 438 *partial = 1;
432 continue; 439 continue;
433 } 440 }
434 err = (*fn)(handle, bh); 441 err = (*fn)(handle, bh);
435 if (!ret) 442 if (!ret)
436 ret = err; 443 ret = err;
437 } 444 }
438 return ret; 445 return ret;
439 } 446 }
440 447
441 static sector_t ocfs2_bmap(struct address_spac 448 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
442 { 449 {
443 sector_t status; 450 sector_t status;
444 u64 p_blkno = 0; 451 u64 p_blkno = 0;
445 int err = 0; 452 int err = 0;
446 struct inode *inode = mapping->host; 453 struct inode *inode = mapping->host;
447 454
448 trace_ocfs2_bmap((unsigned long long)O 455 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
449 (unsigned long long)b 456 (unsigned long long)block);
450 457
451 /* 458 /*
452 * The swap code (ab-)uses ->bmap to g 459 * The swap code (ab-)uses ->bmap to get a block mapping and then
453 * bypasseѕ the file system for actua 460 * bypasseѕ the file system for actual I/O. We really can't allow
454 * that on refcounted inodes, so we ha 461 * that on refcounted inodes, so we have to skip out here. And yes,
455 * 0 is the magic code for a bmap erro 462 * 0 is the magic code for a bmap error..
456 */ 463 */
457 if (ocfs2_is_refcount_inode(inode)) 464 if (ocfs2_is_refcount_inode(inode))
458 return 0; 465 return 0;
459 466
460 /* We don't need to lock journal syste 467 /* We don't need to lock journal system files, since they aren't
461 * accessed concurrently from multiple 468 * accessed concurrently from multiple nodes.
462 */ 469 */
463 if (!INODE_JOURNAL(inode)) { 470 if (!INODE_JOURNAL(inode)) {
464 err = ocfs2_inode_lock(inode, 471 err = ocfs2_inode_lock(inode, NULL, 0);
465 if (err) { 472 if (err) {
466 if (err != -ENOENT) 473 if (err != -ENOENT)
467 mlog_errno(err 474 mlog_errno(err);
468 goto bail; 475 goto bail;
469 } 476 }
470 down_read(&OCFS2_I(inode)->ip_ 477 down_read(&OCFS2_I(inode)->ip_alloc_sem);
471 } 478 }
472 479
473 if (!(OCFS2_I(inode)->ip_dyn_features 480 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
474 err = ocfs2_extent_map_get_blo 481 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
475 482 NULL);
476 483
477 if (!INODE_JOURNAL(inode)) { 484 if (!INODE_JOURNAL(inode)) {
478 up_read(&OCFS2_I(inode)->ip_al 485 up_read(&OCFS2_I(inode)->ip_alloc_sem);
479 ocfs2_inode_unlock(inode, 0); 486 ocfs2_inode_unlock(inode, 0);
480 } 487 }
481 488
482 if (err) { 489 if (err) {
483 mlog(ML_ERROR, "get_blocks() f 490 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
484 (unsigned long long)block 491 (unsigned long long)block);
485 mlog_errno(err); 492 mlog_errno(err);
486 goto bail; 493 goto bail;
487 } 494 }
488 495
489 bail: 496 bail:
490 status = err ? 0 : p_blkno; 497 status = err ? 0 : p_blkno;
491 498
492 return status; 499 return status;
493 } 500 }
494 501
495 static bool ocfs2_release_folio(struct folio * !! 502 static int ocfs2_releasepage(struct page *page, gfp_t wait)
496 { 503 {
497 if (!folio_buffers(folio)) !! 504 if (!page_has_buffers(page))
498 return false; !! 505 return 0;
499 return try_to_free_buffers(folio); !! 506 return try_to_free_buffers(page);
500 } 507 }
501 508
502 static void ocfs2_figure_cluster_boundaries(st 509 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
503 u3 510 u32 cpos,
504 un 511 unsigned int *start,
505 un 512 unsigned int *end)
506 { 513 {
507 unsigned int cluster_start = 0, cluste 514 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
508 515
509 if (unlikely(PAGE_SHIFT > osb->s_clust 516 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
510 unsigned int cpp; 517 unsigned int cpp;
511 518
512 cpp = 1 << (PAGE_SHIFT - osb-> 519 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
513 520
514 cluster_start = cpos % cpp; 521 cluster_start = cpos % cpp;
515 cluster_start = cluster_start 522 cluster_start = cluster_start << osb->s_clustersize_bits;
516 523
517 cluster_end = cluster_start + 524 cluster_end = cluster_start + osb->s_clustersize;
518 } 525 }
519 526
520 BUG_ON(cluster_start > PAGE_SIZE); 527 BUG_ON(cluster_start > PAGE_SIZE);
521 BUG_ON(cluster_end > PAGE_SIZE); 528 BUG_ON(cluster_end > PAGE_SIZE);
522 529
523 if (start) 530 if (start)
524 *start = cluster_start; 531 *start = cluster_start;
525 if (end) 532 if (end)
526 *end = cluster_end; 533 *end = cluster_end;
527 } 534 }
528 535
529 /* 536 /*
530 * 'from' and 'to' are the region in the page 537 * 'from' and 'to' are the region in the page to avoid zeroing.
531 * 538 *
532 * If pagesize > clustersize, this function wi 539 * If pagesize > clustersize, this function will avoid zeroing outside
533 * of the cluster boundary. 540 * of the cluster boundary.
534 * 541 *
535 * from == to == 0 is code for "zero the entir 542 * from == to == 0 is code for "zero the entire cluster region"
536 */ 543 */
537 static void ocfs2_clear_page_regions(struct pa 544 static void ocfs2_clear_page_regions(struct page *page,
538 struct oc 545 struct ocfs2_super *osb, u32 cpos,
539 unsigned 546 unsigned from, unsigned to)
540 { 547 {
541 void *kaddr; 548 void *kaddr;
542 unsigned int cluster_start, cluster_en 549 unsigned int cluster_start, cluster_end;
543 550
544 ocfs2_figure_cluster_boundaries(osb, c 551 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
545 552
546 kaddr = kmap_atomic(page); 553 kaddr = kmap_atomic(page);
547 554
548 if (from || to) { 555 if (from || to) {
549 if (from > cluster_start) 556 if (from > cluster_start)
550 memset(kaddr + cluster 557 memset(kaddr + cluster_start, 0, from - cluster_start);
551 if (to < cluster_end) 558 if (to < cluster_end)
552 memset(kaddr + to, 0, 559 memset(kaddr + to, 0, cluster_end - to);
553 } else { 560 } else {
554 memset(kaddr + cluster_start, 561 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
555 } 562 }
556 563
557 kunmap_atomic(kaddr); 564 kunmap_atomic(kaddr);
558 } 565 }
559 566
560 /* 567 /*
561 * Nonsparse file systems fully allocate befor 568 * Nonsparse file systems fully allocate before we get to the write
562 * code. This prevents ocfs2_write() from tagg 569 * code. This prevents ocfs2_write() from tagging the write as an
563 * allocating one, which means ocfs2_map_page_ 570 * allocating one, which means ocfs2_map_page_blocks() might try to
564 * read-in the blocks at the tail of our file. 571 * read-in the blocks at the tail of our file. Avoid reading them by
565 * testing i_size against each block offset. 572 * testing i_size against each block offset.
566 */ 573 */
567 static int ocfs2_should_read_blk(struct inode !! 574 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
568 unsigned int 575 unsigned int block_start)
569 { 576 {
570 u64 offset = folio_pos(folio) + block_ !! 577 u64 offset = page_offset(page) + block_start;
571 578
572 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 579 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
573 return 1; 580 return 1;
574 581
575 if (i_size_read(inode) > offset) 582 if (i_size_read(inode) > offset)
576 return 1; 583 return 1;
577 584
578 return 0; 585 return 0;
579 } 586 }
580 587
581 /* 588 /*
582 * Some of this taken from __block_write_begin 589 * Some of this taken from __block_write_begin(). We already have our
583 * mapping by now though, and the entire write 590 * mapping by now though, and the entire write will be allocating or
584 * it won't, so not much need to use BH_New. 591 * it won't, so not much need to use BH_New.
585 * 592 *
586 * This will also skip zeroing, which is handl 593 * This will also skip zeroing, which is handled externally.
587 */ 594 */
588 int ocfs2_map_page_blocks(struct page *page, u 595 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
589 struct inode *inode, 596 struct inode *inode, unsigned int from,
590 unsigned int to, int 597 unsigned int to, int new)
591 { 598 {
592 struct folio *folio = page_folio(page) <<
593 int ret = 0; 599 int ret = 0;
594 struct buffer_head *head, *bh, *wait[2 600 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
595 unsigned int block_end, block_start; 601 unsigned int block_end, block_start;
596 unsigned int bsize = i_blocksize(inode 602 unsigned int bsize = i_blocksize(inode);
597 603
598 head = folio_buffers(folio); !! 604 if (!page_has_buffers(page))
599 if (!head) !! 605 create_empty_buffers(page, bsize, 0);
600 head = create_empty_buffers(fo <<
601 606
>> 607 head = page_buffers(page);
602 for (bh = head, block_start = 0; bh != 608 for (bh = head, block_start = 0; bh != head || !block_start;
603 bh = bh->b_this_page, block_start 609 bh = bh->b_this_page, block_start += bsize) {
604 block_end = block_start + bsiz 610 block_end = block_start + bsize;
605 611
606 clear_buffer_new(bh); 612 clear_buffer_new(bh);
607 613
608 /* 614 /*
609 * Ignore blocks outside of ou 615 * Ignore blocks outside of our i/o range -
610 * they may belong to unalloca 616 * they may belong to unallocated clusters.
611 */ 617 */
612 if (block_start >= to || block 618 if (block_start >= to || block_end <= from) {
613 if (folio_test_uptodat !! 619 if (PageUptodate(page))
614 set_buffer_upt 620 set_buffer_uptodate(bh);
615 continue; 621 continue;
616 } 622 }
617 623
618 /* 624 /*
619 * For an allocating write wit 625 * For an allocating write with cluster size >= page
620 * size, we always write the e 626 * size, we always write the entire page.
621 */ 627 */
622 if (new) 628 if (new)
623 set_buffer_new(bh); 629 set_buffer_new(bh);
624 630
625 if (!buffer_mapped(bh)) { 631 if (!buffer_mapped(bh)) {
626 map_bh(bh, inode->i_sb 632 map_bh(bh, inode->i_sb, *p_blkno);
627 clean_bdev_bh_alias(bh 633 clean_bdev_bh_alias(bh);
628 } 634 }
629 635
630 if (folio_test_uptodate(folio) !! 636 if (PageUptodate(page)) {
631 set_buffer_uptodate(bh !! 637 if (!buffer_uptodate(bh))
>> 638 set_buffer_uptodate(bh);
632 } else if (!buffer_uptodate(bh 639 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
633 !buffer_new(bh) && 640 !buffer_new(bh) &&
634 ocfs2_should_read_b !! 641 ocfs2_should_read_blk(inode, page, block_start) &&
635 (block_start < from 642 (block_start < from || block_end > to)) {
636 bh_read_nowait(bh, 0); !! 643 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
637 *wait_bh++=bh; 644 *wait_bh++=bh;
638 } 645 }
639 646
640 *p_blkno = *p_blkno + 1; 647 *p_blkno = *p_blkno + 1;
641 } 648 }
642 649
643 /* 650 /*
644 * If we issued read requests - let th 651 * If we issued read requests - let them complete.
645 */ 652 */
646 while(wait_bh > wait) { 653 while(wait_bh > wait) {
647 wait_on_buffer(*--wait_bh); 654 wait_on_buffer(*--wait_bh);
648 if (!buffer_uptodate(*wait_bh) 655 if (!buffer_uptodate(*wait_bh))
649 ret = -EIO; 656 ret = -EIO;
650 } 657 }
651 658
652 if (ret == 0 || !new) 659 if (ret == 0 || !new)
653 return ret; 660 return ret;
654 661
655 /* 662 /*
656 * If we get -EIO above, zero out any 663 * If we get -EIO above, zero out any newly allocated blocks
657 * to avoid exposing stale data. 664 * to avoid exposing stale data.
658 */ 665 */
659 bh = head; 666 bh = head;
660 block_start = 0; 667 block_start = 0;
661 do { 668 do {
662 block_end = block_start + bsiz 669 block_end = block_start + bsize;
663 if (block_end <= from) 670 if (block_end <= from)
664 goto next_bh; 671 goto next_bh;
665 if (block_start >= to) 672 if (block_start >= to)
666 break; 673 break;
667 674
668 folio_zero_range(folio, block_ !! 675 zero_user(page, block_start, bh->b_size);
669 set_buffer_uptodate(bh); 676 set_buffer_uptodate(bh);
670 mark_buffer_dirty(bh); 677 mark_buffer_dirty(bh);
671 678
672 next_bh: 679 next_bh:
673 block_start = block_end; 680 block_start = block_end;
674 bh = bh->b_this_page; 681 bh = bh->b_this_page;
675 } while (bh != head); 682 } while (bh != head);
676 683
677 return ret; 684 return ret;
678 } 685 }
679 686
680 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 687 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
681 #define OCFS2_MAX_CTXT_PAGES 1 688 #define OCFS2_MAX_CTXT_PAGES 1
682 #else 689 #else
683 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 690 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
684 #endif 691 #endif
685 692
686 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 693 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
687 694
688 struct ocfs2_unwritten_extent { 695 struct ocfs2_unwritten_extent {
689 struct list_head ue_node; 696 struct list_head ue_node;
690 struct list_head ue_ip_node; 697 struct list_head ue_ip_node;
691 u32 ue_cpos; 698 u32 ue_cpos;
692 u32 ue_phys; 699 u32 ue_phys;
693 }; 700 };
694 701
695 /* 702 /*
696 * Describe the state of a single cluster to b 703 * Describe the state of a single cluster to be written to.
697 */ 704 */
698 struct ocfs2_write_cluster_desc { 705 struct ocfs2_write_cluster_desc {
699 u32 c_cpos; 706 u32 c_cpos;
700 u32 c_phys; 707 u32 c_phys;
701 /* 708 /*
702 * Give this a unique field because c_ 709 * Give this a unique field because c_phys eventually gets
703 * filled. 710 * filled.
704 */ 711 */
705 unsigned c_new; 712 unsigned c_new;
706 unsigned c_clear_unwritten; 713 unsigned c_clear_unwritten;
707 unsigned c_needs_zero; 714 unsigned c_needs_zero;
708 }; 715 };
709 716
710 struct ocfs2_write_ctxt { 717 struct ocfs2_write_ctxt {
711 /* Logical cluster position / len of w 718 /* Logical cluster position / len of write */
712 u32 w_cpos 719 u32 w_cpos;
713 u32 w_clen 720 u32 w_clen;
714 721
715 /* First cluster allocated in a nonspa 722 /* First cluster allocated in a nonsparse extend */
716 u32 w_firs 723 u32 w_first_new_cpos;
717 724
718 /* Type of caller. Must be one of buff 725 /* Type of caller. Must be one of buffer, mmap, direct. */
719 ocfs2_write_type_t w_type 726 ocfs2_write_type_t w_type;
720 727
721 struct ocfs2_write_cluster_desc w_desc 728 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
722 729
723 /* 730 /*
724 * This is true if page_size > cluster 731 * This is true if page_size > cluster_size.
725 * 732 *
726 * It triggers a set of special cases 733 * It triggers a set of special cases during write which might
727 * have to deal with allocating writes 734 * have to deal with allocating writes to partial pages.
728 */ 735 */
729 unsigned int w_larg 736 unsigned int w_large_pages;
730 737
731 /* 738 /*
732 * Pages involved in this write. 739 * Pages involved in this write.
733 * 740 *
734 * w_target_page is the page being wri 741 * w_target_page is the page being written to by the user.
735 * 742 *
736 * w_pages is an array of pages which 743 * w_pages is an array of pages which always contains
737 * w_target_page, and in the case of a 744 * w_target_page, and in the case of an allocating write with
738 * page_size < cluster size, it will c 745 * page_size < cluster size, it will contain zero'd and mapped
739 * pages adjacent to w_target_page whi 746 * pages adjacent to w_target_page which need to be written
740 * out in so that future reads from th 747 * out in so that future reads from that region will get
741 * zero's. 748 * zero's.
742 */ 749 */
743 unsigned int w_num_ 750 unsigned int w_num_pages;
744 struct page *w_pag 751 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
745 struct page *w_tar 752 struct page *w_target_page;
746 753
747 /* 754 /*
748 * w_target_locked is used for page_mk 755 * w_target_locked is used for page_mkwrite path indicating no unlocking
749 * against w_target_page in ocfs2_writ 756 * against w_target_page in ocfs2_write_end_nolock.
750 */ 757 */
751 unsigned int w_targ 758 unsigned int w_target_locked:1;
752 759
753 /* 760 /*
754 * ocfs2_write_end() uses this to know 761 * ocfs2_write_end() uses this to know what the real range to
755 * write in the target should be. 762 * write in the target should be.
756 */ 763 */
757 unsigned int w_targ 764 unsigned int w_target_from;
758 unsigned int w_targ 765 unsigned int w_target_to;
759 766
760 /* 767 /*
761 * We could use journal_current_handle 768 * We could use journal_current_handle() but this is cleaner,
762 * IMHO -Mark 769 * IMHO -Mark
763 */ 770 */
764 handle_t *w_han 771 handle_t *w_handle;
765 772
766 struct buffer_head *w_di_ 773 struct buffer_head *w_di_bh;
767 774
768 struct ocfs2_cached_dealloc_ctxt w_dea 775 struct ocfs2_cached_dealloc_ctxt w_dealloc;
769 776
770 struct list_head w_unwr 777 struct list_head w_unwritten_list;
771 unsigned int w_unwr 778 unsigned int w_unwritten_count;
772 }; 779 };
773 780
774 void ocfs2_unlock_and_free_pages(struct page * 781 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
775 { 782 {
776 int i; 783 int i;
777 784
778 for(i = 0; i < num_pages; i++) { 785 for(i = 0; i < num_pages; i++) {
779 if (pages[i]) { 786 if (pages[i]) {
780 unlock_page(pages[i]); 787 unlock_page(pages[i]);
781 mark_page_accessed(pag 788 mark_page_accessed(pages[i]);
782 put_page(pages[i]); 789 put_page(pages[i]);
783 } 790 }
784 } 791 }
785 } 792 }
786 793
787 static void ocfs2_unlock_pages(struct ocfs2_wr 794 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
788 { 795 {
789 int i; 796 int i;
790 797
791 /* 798 /*
792 * w_target_locked is only set to true 799 * w_target_locked is only set to true in the page_mkwrite() case.
793 * The intent is to allow us to lock t 800 * The intent is to allow us to lock the target page from write_begin()
794 * to write_end(). The caller must hol 801 * to write_end(). The caller must hold a ref on w_target_page.
795 */ 802 */
796 if (wc->w_target_locked) { 803 if (wc->w_target_locked) {
797 BUG_ON(!wc->w_target_page); 804 BUG_ON(!wc->w_target_page);
798 for (i = 0; i < wc->w_num_page 805 for (i = 0; i < wc->w_num_pages; i++) {
799 if (wc->w_target_page 806 if (wc->w_target_page == wc->w_pages[i]) {
800 wc->w_pages[i] 807 wc->w_pages[i] = NULL;
801 break; 808 break;
802 } 809 }
803 } 810 }
804 mark_page_accessed(wc->w_targe 811 mark_page_accessed(wc->w_target_page);
805 put_page(wc->w_target_page); 812 put_page(wc->w_target_page);
806 } 813 }
807 ocfs2_unlock_and_free_pages(wc->w_page 814 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
808 } 815 }
809 816
810 static void ocfs2_free_unwritten_list(struct i 817 static void ocfs2_free_unwritten_list(struct inode *inode,
811 struct list_h 818 struct list_head *head)
812 { 819 {
813 struct ocfs2_inode_info *oi = OCFS2_I( 820 struct ocfs2_inode_info *oi = OCFS2_I(inode);
814 struct ocfs2_unwritten_extent *ue = NU 821 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
815 822
816 list_for_each_entry_safe(ue, tmp, head 823 list_for_each_entry_safe(ue, tmp, head, ue_node) {
817 list_del(&ue->ue_node); 824 list_del(&ue->ue_node);
818 spin_lock(&oi->ip_lock); 825 spin_lock(&oi->ip_lock);
819 list_del(&ue->ue_ip_node); 826 list_del(&ue->ue_ip_node);
820 spin_unlock(&oi->ip_lock); 827 spin_unlock(&oi->ip_lock);
821 kfree(ue); 828 kfree(ue);
822 } 829 }
823 } 830 }
824 831
825 static void ocfs2_free_write_ctxt(struct inode 832 static void ocfs2_free_write_ctxt(struct inode *inode,
826 struct ocfs2 833 struct ocfs2_write_ctxt *wc)
827 { 834 {
828 ocfs2_free_unwritten_list(inode, &wc-> 835 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
829 ocfs2_unlock_pages(wc); 836 ocfs2_unlock_pages(wc);
830 brelse(wc->w_di_bh); 837 brelse(wc->w_di_bh);
831 kfree(wc); 838 kfree(wc);
832 } 839 }
833 840
834 static int ocfs2_alloc_write_ctxt(struct ocfs2 841 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
835 struct ocfs2 842 struct ocfs2_super *osb, loff_t pos,
836 unsigned len 843 unsigned len, ocfs2_write_type_t type,
837 struct buffe 844 struct buffer_head *di_bh)
838 { 845 {
839 u32 cend; 846 u32 cend;
840 struct ocfs2_write_ctxt *wc; 847 struct ocfs2_write_ctxt *wc;
841 848
842 wc = kzalloc(sizeof(struct ocfs2_write 849 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
843 if (!wc) 850 if (!wc)
844 return -ENOMEM; 851 return -ENOMEM;
845 852
846 wc->w_cpos = pos >> osb->s_clustersize 853 wc->w_cpos = pos >> osb->s_clustersize_bits;
847 wc->w_first_new_cpos = UINT_MAX; 854 wc->w_first_new_cpos = UINT_MAX;
848 cend = (pos + len - 1) >> osb->s_clust 855 cend = (pos + len - 1) >> osb->s_clustersize_bits;
849 wc->w_clen = cend - wc->w_cpos + 1; 856 wc->w_clen = cend - wc->w_cpos + 1;
850 get_bh(di_bh); 857 get_bh(di_bh);
851 wc->w_di_bh = di_bh; 858 wc->w_di_bh = di_bh;
852 wc->w_type = type; 859 wc->w_type = type;
853 860
854 if (unlikely(PAGE_SHIFT > osb->s_clust 861 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
855 wc->w_large_pages = 1; 862 wc->w_large_pages = 1;
856 else 863 else
857 wc->w_large_pages = 0; 864 wc->w_large_pages = 0;
858 865
859 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 866 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
860 INIT_LIST_HEAD(&wc->w_unwritten_list); 867 INIT_LIST_HEAD(&wc->w_unwritten_list);
861 868
862 *wcp = wc; 869 *wcp = wc;
863 870
864 return 0; 871 return 0;
865 } 872 }
866 873
867 /* 874 /*
868 * If a page has any new buffers, zero them ou 875 * If a page has any new buffers, zero them out here, and mark them uptodate
869 * and dirty so they'll be written out (in ord 876 * and dirty so they'll be written out (in order to prevent uninitialised
870 * block data from leaking). And clear the new 877 * block data from leaking). And clear the new bit.
871 */ 878 */
872 static void ocfs2_zero_new_buffers(struct page 879 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
873 { 880 {
874 unsigned int block_start, block_end; 881 unsigned int block_start, block_end;
875 struct buffer_head *head, *bh; 882 struct buffer_head *head, *bh;
876 883
877 BUG_ON(!PageLocked(page)); 884 BUG_ON(!PageLocked(page));
878 if (!page_has_buffers(page)) 885 if (!page_has_buffers(page))
879 return; 886 return;
880 887
881 bh = head = page_buffers(page); 888 bh = head = page_buffers(page);
882 block_start = 0; 889 block_start = 0;
883 do { 890 do {
884 block_end = block_start + bh-> 891 block_end = block_start + bh->b_size;
885 892
886 if (buffer_new(bh)) { 893 if (buffer_new(bh)) {
887 if (block_end > from & 894 if (block_end > from && block_start < to) {
888 if (!PageUptod 895 if (!PageUptodate(page)) {
889 unsign 896 unsigned start, end;
890 897
891 start 898 start = max(from, block_start);
892 end = 899 end = min(to, block_end);
893 900
894 zero_u 901 zero_user_segment(page, start, end);
895 set_bu 902 set_buffer_uptodate(bh);
896 } 903 }
897 904
898 clear_buffer_n 905 clear_buffer_new(bh);
899 mark_buffer_di 906 mark_buffer_dirty(bh);
900 } 907 }
901 } 908 }
902 909
903 block_start = block_end; 910 block_start = block_end;
904 bh = bh->b_this_page; 911 bh = bh->b_this_page;
905 } while (bh != head); 912 } while (bh != head);
906 } 913 }
907 914
908 /* 915 /*
909 * Only called when we have a failure during a 916 * Only called when we have a failure during allocating write to write
910 * zero's to the newly allocated region. 917 * zero's to the newly allocated region.
911 */ 918 */
912 static void ocfs2_write_failure(struct inode * 919 static void ocfs2_write_failure(struct inode *inode,
913 struct ocfs2_w 920 struct ocfs2_write_ctxt *wc,
914 loff_t user_po 921 loff_t user_pos, unsigned user_len)
915 { 922 {
916 int i; 923 int i;
917 unsigned from = user_pos & (PAGE_SIZE 924 unsigned from = user_pos & (PAGE_SIZE - 1),
918 to = user_pos + user_len; 925 to = user_pos + user_len;
919 struct page *tmppage; 926 struct page *tmppage;
920 927
921 if (wc->w_target_page) 928 if (wc->w_target_page)
922 ocfs2_zero_new_buffers(wc->w_t 929 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
923 930
924 for(i = 0; i < wc->w_num_pages; i++) { 931 for(i = 0; i < wc->w_num_pages; i++) {
925 tmppage = wc->w_pages[i]; 932 tmppage = wc->w_pages[i];
926 933
927 if (tmppage && page_has_buffer 934 if (tmppage && page_has_buffers(tmppage)) {
928 if (ocfs2_should_order 935 if (ocfs2_should_order_data(inode))
929 ocfs2_jbd2_ino 936 ocfs2_jbd2_inode_add_write(wc->w_handle, inode,
930 937 user_pos, user_len);
931 938
932 block_commit_write(tmp 939 block_commit_write(tmppage, from, to);
933 } 940 }
934 } 941 }
935 } 942 }
936 943
937 static int ocfs2_prepare_page_for_write(struct 944 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
938 struct 945 struct ocfs2_write_ctxt *wc,
939 struct 946 struct page *page, u32 cpos,
940 loff_t 947 loff_t user_pos, unsigned user_len,
941 int ne 948 int new)
942 { 949 {
943 int ret; 950 int ret;
944 unsigned int map_from = 0, map_to = 0; 951 unsigned int map_from = 0, map_to = 0;
945 unsigned int cluster_start, cluster_en 952 unsigned int cluster_start, cluster_end;
946 unsigned int user_data_from = 0, user_ 953 unsigned int user_data_from = 0, user_data_to = 0;
947 954
948 ocfs2_figure_cluster_boundaries(OCFS2_ 955 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
949 &clust 956 &cluster_start, &cluster_end);
950 957
951 /* treat the write as new if the a hol 958 /* treat the write as new if the a hole/lseek spanned across
952 * the page boundary. 959 * the page boundary.
953 */ 960 */
954 new = new | ((i_size_read(inode) <= pa 961 new = new | ((i_size_read(inode) <= page_offset(page)) &&
955 (page_offset(page) <= 962 (page_offset(page) <= user_pos));
956 963
957 if (page == wc->w_target_page) { 964 if (page == wc->w_target_page) {
958 map_from = user_pos & (PAGE_SI 965 map_from = user_pos & (PAGE_SIZE - 1);
959 map_to = map_from + user_len; 966 map_to = map_from + user_len;
960 967
961 if (new) 968 if (new)
962 ret = ocfs2_map_page_b 969 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
963 970 cluster_start, cluster_end,
964 971 new);
965 else 972 else
966 ret = ocfs2_map_page_b 973 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
967 974 map_from, map_to, new);
968 if (ret) { 975 if (ret) {
969 mlog_errno(ret); 976 mlog_errno(ret);
970 goto out; 977 goto out;
971 } 978 }
972 979
973 user_data_from = map_from; 980 user_data_from = map_from;
974 user_data_to = map_to; 981 user_data_to = map_to;
975 if (new) { 982 if (new) {
976 map_from = cluster_sta 983 map_from = cluster_start;
977 map_to = cluster_end; 984 map_to = cluster_end;
978 } 985 }
979 } else { 986 } else {
980 /* 987 /*
981 * If we haven't allocated the 988 * If we haven't allocated the new page yet, we
982 * shouldn't be writing it out 989 * shouldn't be writing it out without copying user
983 * data. This is likely a math 990 * data. This is likely a math error from the caller.
984 */ 991 */
985 BUG_ON(!new); 992 BUG_ON(!new);
986 993
987 map_from = cluster_start; 994 map_from = cluster_start;
988 map_to = cluster_end; 995 map_to = cluster_end;
989 996
990 ret = ocfs2_map_page_blocks(pa 997 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
991 cl 998 cluster_start, cluster_end, new);
992 if (ret) { 999 if (ret) {
993 mlog_errno(ret); 1000 mlog_errno(ret);
994 goto out; 1001 goto out;
995 } 1002 }
996 } 1003 }
997 1004
998 /* 1005 /*
999 * Parts of newly allocated pages need 1006 * Parts of newly allocated pages need to be zero'd.
1000 * 1007 *
1001 * Above, we have also rewritten 'to' 1008 * Above, we have also rewritten 'to' and 'from' - as far as
1002 * the rest of the function is concer 1009 * the rest of the function is concerned, the entire cluster
1003 * range inside of a page needs to be 1010 * range inside of a page needs to be written.
1004 * 1011 *
1005 * We can skip this if the page is up 1012 * We can skip this if the page is up to date - it's already
1006 * been zero'd from being read in as 1013 * been zero'd from being read in as a hole.
1007 */ 1014 */
1008 if (new && !PageUptodate(page)) 1015 if (new && !PageUptodate(page))
1009 ocfs2_clear_page_regions(page 1016 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1010 cpos 1017 cpos, user_data_from, user_data_to);
1011 1018
1012 flush_dcache_page(page); 1019 flush_dcache_page(page);
1013 1020
1014 out: 1021 out:
1015 return ret; 1022 return ret;
1016 } 1023 }
1017 1024
1018 /* 1025 /*
1019 * This function will only grab one clusters 1026 * This function will only grab one clusters worth of pages.
1020 */ 1027 */
1021 static int ocfs2_grab_pages_for_write(struct 1028 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1022 struct 1029 struct ocfs2_write_ctxt *wc,
1023 u32 cpo 1030 u32 cpos, loff_t user_pos,
1024 unsigne 1031 unsigned user_len, int new,
1025 struct 1032 struct page *mmap_page)
1026 { 1033 {
1027 int ret = 0, i; 1034 int ret = 0, i;
1028 unsigned long start, target_index, en 1035 unsigned long start, target_index, end_index, index;
1029 struct inode *inode = mapping->host; 1036 struct inode *inode = mapping->host;
1030 loff_t last_byte; 1037 loff_t last_byte;
1031 1038
1032 target_index = user_pos >> PAGE_SHIFT 1039 target_index = user_pos >> PAGE_SHIFT;
1033 1040
1034 /* 1041 /*
1035 * Figure out how many pages we'll be 1042 * Figure out how many pages we'll be manipulating here. For
1036 * non allocating write, we just chan 1043 * non allocating write, we just change the one
1037 * page. Otherwise, we'll need a whol 1044 * page. Otherwise, we'll need a whole clusters worth. If we're
1038 * writing past i_size, we only need 1045 * writing past i_size, we only need enough pages to cover the
1039 * last page of the write. 1046 * last page of the write.
1040 */ 1047 */
1041 if (new) { 1048 if (new) {
1042 wc->w_num_pages = ocfs2_pages 1049 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1043 start = ocfs2_align_clusters_ 1050 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1044 /* 1051 /*
1045 * We need the index *past* t 1052 * We need the index *past* the last page we could possibly
1046 * touch. This is the page p 1053 * touch. This is the page past the end of the write or
1047 * i_size, whichever is great 1054 * i_size, whichever is greater.
1048 */ 1055 */
1049 last_byte = max(user_pos + us 1056 last_byte = max(user_pos + user_len, i_size_read(inode));
1050 BUG_ON(last_byte < 1); 1057 BUG_ON(last_byte < 1);
1051 end_index = ((last_byte - 1) 1058 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
1052 if ((start + wc->w_num_pages) 1059 if ((start + wc->w_num_pages) > end_index)
1053 wc->w_num_pages = end 1060 wc->w_num_pages = end_index - start;
1054 } else { 1061 } else {
1055 wc->w_num_pages = 1; 1062 wc->w_num_pages = 1;
1056 start = target_index; 1063 start = target_index;
1057 } 1064 }
1058 end_index = (user_pos + user_len - 1) 1065 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
1059 1066
1060 for(i = 0; i < wc->w_num_pages; i++) 1067 for(i = 0; i < wc->w_num_pages; i++) {
1061 index = start + i; 1068 index = start + i;
1062 1069
1063 if (index >= target_index && 1070 if (index >= target_index && index <= end_index &&
1064 wc->w_type == OCFS2_WRITE 1071 wc->w_type == OCFS2_WRITE_MMAP) {
1065 /* 1072 /*
1066 * ocfs2_pagemkwrite( 1073 * ocfs2_pagemkwrite() is a little different
1067 * and wants us to di 1074 * and wants us to directly use the page
1068 * passed in. 1075 * passed in.
1069 */ 1076 */
1070 lock_page(mmap_page); 1077 lock_page(mmap_page);
1071 1078
1072 /* Exit and let the c 1079 /* Exit and let the caller retry */
1073 if (mmap_page->mappin 1080 if (mmap_page->mapping != mapping) {
1074 WARN_ON(mmap_ 1081 WARN_ON(mmap_page->mapping);
1075 unlock_page(m 1082 unlock_page(mmap_page);
1076 ret = -EAGAIN 1083 ret = -EAGAIN;
1077 goto out; 1084 goto out;
1078 } 1085 }
1079 1086
1080 get_page(mmap_page); 1087 get_page(mmap_page);
1081 wc->w_pages[i] = mmap 1088 wc->w_pages[i] = mmap_page;
1082 wc->w_target_locked = 1089 wc->w_target_locked = true;
1083 } else if (index >= target_in 1090 } else if (index >= target_index && index <= end_index &&
1084 wc->w_type == OCFS 1091 wc->w_type == OCFS2_WRITE_DIRECT) {
1085 /* Direct write has n 1092 /* Direct write has no mapping page. */
1086 wc->w_pages[i] = NULL 1093 wc->w_pages[i] = NULL;
1087 continue; 1094 continue;
1088 } else { 1095 } else {
1089 wc->w_pages[i] = find 1096 wc->w_pages[i] = find_or_create_page(mapping, index,
1090 1097 GFP_NOFS);
1091 if (!wc->w_pages[i]) 1098 if (!wc->w_pages[i]) {
1092 ret = -ENOMEM 1099 ret = -ENOMEM;
1093 mlog_errno(re 1100 mlog_errno(ret);
1094 goto out; 1101 goto out;
1095 } 1102 }
1096 } 1103 }
1097 wait_for_stable_page(wc->w_pa 1104 wait_for_stable_page(wc->w_pages[i]);
1098 1105
1099 if (index == target_index) 1106 if (index == target_index)
1100 wc->w_target_page = w 1107 wc->w_target_page = wc->w_pages[i];
1101 } 1108 }
1102 out: 1109 out:
1103 if (ret) 1110 if (ret)
1104 wc->w_target_locked = false; 1111 wc->w_target_locked = false;
1105 return ret; 1112 return ret;
1106 } 1113 }
1107 1114
1108 /* 1115 /*
1109 * Prepare a single cluster for write one clu 1116 * Prepare a single cluster for write one cluster into the file.
1110 */ 1117 */
1111 static int ocfs2_write_cluster(struct address 1118 static int ocfs2_write_cluster(struct address_space *mapping,
1112 u32 *phys, uns 1119 u32 *phys, unsigned int new,
1113 unsigned int c 1120 unsigned int clear_unwritten,
1114 unsigned int s 1121 unsigned int should_zero,
1115 struct ocfs2_a 1122 struct ocfs2_alloc_context *data_ac,
1116 struct ocfs2_a 1123 struct ocfs2_alloc_context *meta_ac,
1117 struct ocfs2_w 1124 struct ocfs2_write_ctxt *wc, u32 cpos,
1118 loff_t user_po 1125 loff_t user_pos, unsigned user_len)
1119 { 1126 {
1120 int ret, i; 1127 int ret, i;
1121 u64 p_blkno; 1128 u64 p_blkno;
1122 struct inode *inode = mapping->host; 1129 struct inode *inode = mapping->host;
1123 struct ocfs2_extent_tree et; 1130 struct ocfs2_extent_tree et;
1124 int bpc = ocfs2_clusters_to_blocks(in 1131 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1125 1132
1126 if (new) { 1133 if (new) {
1127 u32 tmp_pos; 1134 u32 tmp_pos;
1128 1135
1129 /* 1136 /*
1130 * This is safe to call with 1137 * This is safe to call with the page locks - it won't take
1131 * any additional semaphores 1138 * any additional semaphores or cluster locks.
1132 */ 1139 */
1133 tmp_pos = cpos; 1140 tmp_pos = cpos;
1134 ret = ocfs2_add_inode_data(OC 1141 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1135 &t 1142 &tmp_pos, 1, !clear_unwritten,
1136 wc 1143 wc->w_di_bh, wc->w_handle,
1137 da 1144 data_ac, meta_ac, NULL);
1138 /* 1145 /*
1139 * This shouldn't happen beca 1146 * This shouldn't happen because we must have already
1140 * calculated the correct met 1147 * calculated the correct meta data allocation required. The
1141 * internal tree allocation c 1148 * internal tree allocation code should know how to increase
1142 * transaction credits itself 1149 * transaction credits itself.
1143 * 1150 *
1144 * If need be, we could handl 1151 * If need be, we could handle -EAGAIN for a
1145 * RESTART_TRANS here. 1152 * RESTART_TRANS here.
1146 */ 1153 */
1147 mlog_bug_on_msg(ret == -EAGAI 1154 mlog_bug_on_msg(ret == -EAGAIN,
1148 "Inode %llu: 1155 "Inode %llu: EAGAIN return during allocation.\n",
1149 (unsigned lon 1156 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1150 if (ret < 0) { 1157 if (ret < 0) {
1151 mlog_errno(ret); 1158 mlog_errno(ret);
1152 goto out; 1159 goto out;
1153 } 1160 }
1154 } else if (clear_unwritten) { 1161 } else if (clear_unwritten) {
1155 ocfs2_init_dinode_extent_tree 1162 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1156 1163 wc->w_di_bh);
1157 ret = ocfs2_mark_extent_writt 1164 ret = ocfs2_mark_extent_written(inode, &et,
1158 1165 wc->w_handle, cpos, 1, *phys,
1159 1166 meta_ac, &wc->w_dealloc);
1160 if (ret < 0) { 1167 if (ret < 0) {
1161 mlog_errno(ret); 1168 mlog_errno(ret);
1162 goto out; 1169 goto out;
1163 } 1170 }
1164 } 1171 }
1165 1172
1166 /* 1173 /*
1167 * The only reason this should fail i 1174 * The only reason this should fail is due to an inability to
1168 * find the extent added. 1175 * find the extent added.
1169 */ 1176 */
1170 ret = ocfs2_get_clusters(inode, cpos, 1177 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
1171 if (ret < 0) { 1178 if (ret < 0) {
1172 mlog(ML_ERROR, "Get physical 1179 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1173 "at logical clust 1180 "at logical cluster %u",
1174 (unsigned long lo 1181 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
1175 goto out; 1182 goto out;
1176 } 1183 }
1177 1184
1178 BUG_ON(*phys == 0); 1185 BUG_ON(*phys == 0);
1179 1186
1180 p_blkno = ocfs2_clusters_to_blocks(in 1187 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1181 if (!should_zero) 1188 if (!should_zero)
1182 p_blkno += (user_pos >> inode 1189 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
1183 1190
1184 for(i = 0; i < wc->w_num_pages; i++) 1191 for(i = 0; i < wc->w_num_pages; i++) {
1185 int tmpret; 1192 int tmpret;
1186 1193
1187 /* This is the direct io targ 1194 /* This is the direct io target page. */
1188 if (wc->w_pages[i] == NULL) { 1195 if (wc->w_pages[i] == NULL) {
1189 p_blkno++; 1196 p_blkno++;
1190 continue; 1197 continue;
1191 } 1198 }
1192 1199
1193 tmpret = ocfs2_prepare_page_f 1200 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1194 1201 wc->w_pages[i], cpos,
1195 1202 user_pos, user_len,
1196 1203 should_zero);
1197 if (tmpret) { 1204 if (tmpret) {
1198 mlog_errno(tmpret); 1205 mlog_errno(tmpret);
1199 if (ret == 0) 1206 if (ret == 0)
1200 ret = tmpret; 1207 ret = tmpret;
1201 } 1208 }
1202 } 1209 }
1203 1210
1204 /* 1211 /*
1205 * We only have cleanup to do in case 1212 * We only have cleanup to do in case of allocating write.
1206 */ 1213 */
1207 if (ret && new) 1214 if (ret && new)
1208 ocfs2_write_failure(inode, wc 1215 ocfs2_write_failure(inode, wc, user_pos, user_len);
1209 1216
1210 out: 1217 out:
1211 1218
1212 return ret; 1219 return ret;
1213 } 1220 }
1214 1221
1215 static int ocfs2_write_cluster_by_desc(struct 1222 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1216 struct 1223 struct ocfs2_alloc_context *data_ac,
1217 struct 1224 struct ocfs2_alloc_context *meta_ac,
1218 struct 1225 struct ocfs2_write_ctxt *wc,
1219 loff_t 1226 loff_t pos, unsigned len)
1220 { 1227 {
1221 int ret, i; 1228 int ret, i;
1222 loff_t cluster_off; 1229 loff_t cluster_off;
1223 unsigned int local_len = len; 1230 unsigned int local_len = len;
1224 struct ocfs2_write_cluster_desc *desc 1231 struct ocfs2_write_cluster_desc *desc;
1225 struct ocfs2_super *osb = OCFS2_SB(ma 1232 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1226 1233
1227 for (i = 0; i < wc->w_clen; i++) { 1234 for (i = 0; i < wc->w_clen; i++) {
1228 desc = &wc->w_desc[i]; 1235 desc = &wc->w_desc[i];
1229 1236
1230 /* 1237 /*
1231 * We have to make sure that 1238 * We have to make sure that the total write passed in
1232 * doesn't extend past a sing 1239 * doesn't extend past a single cluster.
1233 */ 1240 */
1234 local_len = len; 1241 local_len = len;
1235 cluster_off = pos & (osb->s_c 1242 cluster_off = pos & (osb->s_clustersize - 1);
1236 if ((cluster_off + local_len) 1243 if ((cluster_off + local_len) > osb->s_clustersize)
1237 local_len = osb->s_cl 1244 local_len = osb->s_clustersize - cluster_off;
1238 1245
1239 ret = ocfs2_write_cluster(map 1246 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
1240 des 1247 desc->c_new,
1241 des 1248 desc->c_clear_unwritten,
1242 des 1249 desc->c_needs_zero,
1243 dat 1250 data_ac, meta_ac,
1244 wc, 1251 wc, desc->c_cpos, pos, local_len);
1245 if (ret) { 1252 if (ret) {
1246 mlog_errno(ret); 1253 mlog_errno(ret);
1247 goto out; 1254 goto out;
1248 } 1255 }
1249 1256
1250 len -= local_len; 1257 len -= local_len;
1251 pos += local_len; 1258 pos += local_len;
1252 } 1259 }
1253 1260
1254 ret = 0; 1261 ret = 0;
1255 out: 1262 out:
1256 return ret; 1263 return ret;
1257 } 1264 }
1258 1265
1259 /* 1266 /*
1260 * ocfs2_write_end() wants to know which part 1267 * ocfs2_write_end() wants to know which parts of the target page it
1261 * should complete the write on. It's easiest 1268 * should complete the write on. It's easiest to compute them ahead of
1262 * time when a more complete view of the writ 1269 * time when a more complete view of the write is available.
1263 */ 1270 */
1264 static void ocfs2_set_target_boundaries(struc 1271 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1265 struc 1272 struct ocfs2_write_ctxt *wc,
1266 loff_ 1273 loff_t pos, unsigned len, int alloc)
1267 { 1274 {
1268 struct ocfs2_write_cluster_desc *desc 1275 struct ocfs2_write_cluster_desc *desc;
1269 1276
1270 wc->w_target_from = pos & (PAGE_SIZE 1277 wc->w_target_from = pos & (PAGE_SIZE - 1);
1271 wc->w_target_to = wc->w_target_from + 1278 wc->w_target_to = wc->w_target_from + len;
1272 1279
1273 if (alloc == 0) 1280 if (alloc == 0)
1274 return; 1281 return;
1275 1282
1276 /* 1283 /*
1277 * Allocating write - we may have dif 1284 * Allocating write - we may have different boundaries based
1278 * on page size and cluster size. 1285 * on page size and cluster size.
1279 * 1286 *
1280 * NOTE: We can no longer compute one 1287 * NOTE: We can no longer compute one value from the other as
1281 * the actual write length and user p 1288 * the actual write length and user provided length may be
1282 * different. 1289 * different.
1283 */ 1290 */
1284 1291
1285 if (wc->w_large_pages) { 1292 if (wc->w_large_pages) {
1286 /* 1293 /*
1287 * We only care about the 1st 1294 * We only care about the 1st and last cluster within
1288 * our range and whether they 1295 * our range and whether they should be zero'd or not. Either
1289 * value may be extended out 1296 * value may be extended out to the start/end of a
1290 * newly allocated cluster. 1297 * newly allocated cluster.
1291 */ 1298 */
1292 desc = &wc->w_desc[0]; 1299 desc = &wc->w_desc[0];
1293 if (desc->c_needs_zero) 1300 if (desc->c_needs_zero)
1294 ocfs2_figure_cluster_ 1301 ocfs2_figure_cluster_boundaries(osb,
1295 1302 desc->c_cpos,
1296 1303 &wc->w_target_from,
1297 1304 NULL);
1298 1305
1299 desc = &wc->w_desc[wc->w_clen 1306 desc = &wc->w_desc[wc->w_clen - 1];
1300 if (desc->c_needs_zero) 1307 if (desc->c_needs_zero)
1301 ocfs2_figure_cluster_ 1308 ocfs2_figure_cluster_boundaries(osb,
1302 1309 desc->c_cpos,
1303 1310 NULL,
1304 1311 &wc->w_target_to);
1305 } else { 1312 } else {
1306 wc->w_target_from = 0; 1313 wc->w_target_from = 0;
1307 wc->w_target_to = PAGE_SIZE; 1314 wc->w_target_to = PAGE_SIZE;
1308 } 1315 }
1309 } 1316 }
1310 1317
1311 /* 1318 /*
1312 * Check if this extent is marked UNWRITTEN b 1319 * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to
1313 * do the zero work. And should not to clear 1320 * do the zero work. And should not to clear UNWRITTEN since it will be cleared
1314 * by the direct io procedure. 1321 * by the direct io procedure.
1315 * If this is a new extent that allocated by 1322 * If this is a new extent that allocated by direct io, we should mark it in
1316 * the ip_unwritten_list. 1323 * the ip_unwritten_list.
1317 */ 1324 */
1318 static int ocfs2_unwritten_check(struct inode 1325 static int ocfs2_unwritten_check(struct inode *inode,
1319 struct ocfs2 1326 struct ocfs2_write_ctxt *wc,
1320 struct ocfs2 1327 struct ocfs2_write_cluster_desc *desc)
1321 { 1328 {
1322 struct ocfs2_inode_info *oi = OCFS2_I 1329 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1323 struct ocfs2_unwritten_extent *ue = N 1330 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
1324 int ret = 0; 1331 int ret = 0;
1325 1332
1326 if (!desc->c_needs_zero) 1333 if (!desc->c_needs_zero)
1327 return 0; 1334 return 0;
1328 1335
1329 retry: 1336 retry:
1330 spin_lock(&oi->ip_lock); 1337 spin_lock(&oi->ip_lock);
1331 /* Needs not to zero no metter buffer 1338 /* Needs not to zero no metter buffer or direct. The one who is zero
1332 * the cluster is doing zero. And he 1339 * the cluster is doing zero. And he will clear unwritten after all
1333 * cluster io finished. */ 1340 * cluster io finished. */
1334 list_for_each_entry(ue, &oi->ip_unwri 1341 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1335 if (desc->c_cpos == ue->ue_cp 1342 if (desc->c_cpos == ue->ue_cpos) {
1336 BUG_ON(desc->c_new); 1343 BUG_ON(desc->c_new);
1337 desc->c_needs_zero = 1344 desc->c_needs_zero = 0;
1338 desc->c_clear_unwritt 1345 desc->c_clear_unwritten = 0;
1339 goto unlock; 1346 goto unlock;
1340 } 1347 }
1341 } 1348 }
1342 1349
1343 if (wc->w_type != OCFS2_WRITE_DIRECT) 1350 if (wc->w_type != OCFS2_WRITE_DIRECT)
1344 goto unlock; 1351 goto unlock;
1345 1352
1346 if (new == NULL) { 1353 if (new == NULL) {
1347 spin_unlock(&oi->ip_lock); 1354 spin_unlock(&oi->ip_lock);
1348 new = kmalloc(sizeof(struct o 1355 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1349 GFP_NOFS); 1356 GFP_NOFS);
1350 if (new == NULL) { 1357 if (new == NULL) {
1351 ret = -ENOMEM; 1358 ret = -ENOMEM;
1352 goto out; 1359 goto out;
1353 } 1360 }
1354 goto retry; 1361 goto retry;
1355 } 1362 }
1356 /* This direct write will doing zero. 1363 /* This direct write will doing zero. */
1357 new->ue_cpos = desc->c_cpos; 1364 new->ue_cpos = desc->c_cpos;
1358 new->ue_phys = desc->c_phys; 1365 new->ue_phys = desc->c_phys;
1359 desc->c_clear_unwritten = 0; 1366 desc->c_clear_unwritten = 0;
1360 list_add_tail(&new->ue_ip_node, &oi-> 1367 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1361 list_add_tail(&new->ue_node, &wc->w_u 1368 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
1362 wc->w_unwritten_count++; 1369 wc->w_unwritten_count++;
1363 new = NULL; 1370 new = NULL;
1364 unlock: 1371 unlock:
1365 spin_unlock(&oi->ip_lock); 1372 spin_unlock(&oi->ip_lock);
1366 out: 1373 out:
1367 kfree(new); 1374 kfree(new);
1368 return ret; 1375 return ret;
1369 } 1376 }
1370 1377
1371 /* 1378 /*
1372 * Populate each single-cluster write descrip 1379 * Populate each single-cluster write descriptor in the write context
1373 * with information about the i/o to be done. 1380 * with information about the i/o to be done.
1374 * 1381 *
1375 * Returns the number of clusters that will h 1382 * Returns the number of clusters that will have to be allocated, as
1376 * well as a worst case estimate of the numbe 1383 * well as a worst case estimate of the number of extent records that
1377 * would have to be created during a write to 1384 * would have to be created during a write to an unwritten region.
1378 */ 1385 */
1379 static int ocfs2_populate_write_desc(struct i 1386 static int ocfs2_populate_write_desc(struct inode *inode,
1380 struct o 1387 struct ocfs2_write_ctxt *wc,
1381 unsigned 1388 unsigned int *clusters_to_alloc,
1382 unsigned 1389 unsigned int *extents_to_split)
1383 { 1390 {
1384 int ret; 1391 int ret;
1385 struct ocfs2_write_cluster_desc *desc 1392 struct ocfs2_write_cluster_desc *desc;
1386 unsigned int num_clusters = 0; 1393 unsigned int num_clusters = 0;
1387 unsigned int ext_flags = 0; 1394 unsigned int ext_flags = 0;
1388 u32 phys = 0; 1395 u32 phys = 0;
1389 int i; 1396 int i;
1390 1397
1391 *clusters_to_alloc = 0; 1398 *clusters_to_alloc = 0;
1392 *extents_to_split = 0; 1399 *extents_to_split = 0;
1393 1400
1394 for (i = 0; i < wc->w_clen; i++) { 1401 for (i = 0; i < wc->w_clen; i++) {
1395 desc = &wc->w_desc[i]; 1402 desc = &wc->w_desc[i];
1396 desc->c_cpos = wc->w_cpos + i 1403 desc->c_cpos = wc->w_cpos + i;
1397 1404
1398 if (num_clusters == 0) { 1405 if (num_clusters == 0) {
1399 /* 1406 /*
1400 * Need to look up th 1407 * Need to look up the next extent record.
1401 */ 1408 */
1402 ret = ocfs2_get_clust 1409 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1403 1410 &num_clusters, &ext_flags);
1404 if (ret) { 1411 if (ret) {
1405 mlog_errno(re 1412 mlog_errno(ret);
1406 goto out; 1413 goto out;
1407 } 1414 }
1408 1415
1409 /* We should already 1416 /* We should already CoW the refcountd extent. */
1410 BUG_ON(ext_flags & OC 1417 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1411 1418
1412 /* 1419 /*
1413 * Assume worst case 1420 * Assume worst case - that we're writing in
1414 * the middle of the 1421 * the middle of the extent.
1415 * 1422 *
1416 * We can assume that 1423 * We can assume that the write proceeds from
1417 * left to right, in 1424 * left to right, in which case the extent
1418 * insert code is sma 1425 * insert code is smart enough to coalesce the
1419 * next splits into t 1426 * next splits into the previous records created.
1420 */ 1427 */
1421 if (ext_flags & OCFS2 1428 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1422 *extents_to_s 1429 *extents_to_split = *extents_to_split + 2;
1423 } else if (phys) { 1430 } else if (phys) {
1424 /* 1431 /*
1425 * Only increment phy 1432 * Only increment phys if it doesn't describe
1426 * a hole. 1433 * a hole.
1427 */ 1434 */
1428 phys++; 1435 phys++;
1429 } 1436 }
1430 1437
1431 /* 1438 /*
1432 * If w_first_new_cpos is < U 1439 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1433 * file that got extended. w 1440 * file that got extended. w_first_new_cpos tells us
1434 * where the newly allocated 1441 * where the newly allocated clusters are so we can
1435 * zero them. 1442 * zero them.
1436 */ 1443 */
1437 if (desc->c_cpos >= wc->w_fir 1444 if (desc->c_cpos >= wc->w_first_new_cpos) {
1438 BUG_ON(phys == 0); 1445 BUG_ON(phys == 0);
1439 desc->c_needs_zero = 1446 desc->c_needs_zero = 1;
1440 } 1447 }
1441 1448
1442 desc->c_phys = phys; 1449 desc->c_phys = phys;
1443 if (phys == 0) { 1450 if (phys == 0) {
1444 desc->c_new = 1; 1451 desc->c_new = 1;
1445 desc->c_needs_zero = 1452 desc->c_needs_zero = 1;
1446 desc->c_clear_unwritt 1453 desc->c_clear_unwritten = 1;
1447 *clusters_to_alloc = 1454 *clusters_to_alloc = *clusters_to_alloc + 1;
1448 } 1455 }
1449 1456
1450 if (ext_flags & OCFS2_EXT_UNW 1457 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1451 desc->c_clear_unwritt 1458 desc->c_clear_unwritten = 1;
1452 desc->c_needs_zero = 1459 desc->c_needs_zero = 1;
1453 } 1460 }
1454 1461
1455 ret = ocfs2_unwritten_check(i 1462 ret = ocfs2_unwritten_check(inode, wc, desc);
1456 if (ret) { 1463 if (ret) {
1457 mlog_errno(ret); 1464 mlog_errno(ret);
1458 goto out; 1465 goto out;
1459 } 1466 }
1460 1467
1461 num_clusters--; 1468 num_clusters--;
1462 } 1469 }
1463 1470
1464 ret = 0; 1471 ret = 0;
1465 out: 1472 out:
1466 return ret; 1473 return ret;
1467 } 1474 }
1468 1475
1469 static int ocfs2_write_begin_inline(struct ad 1476 static int ocfs2_write_begin_inline(struct address_space *mapping,
1470 struct in 1477 struct inode *inode,
1471 struct oc 1478 struct ocfs2_write_ctxt *wc)
1472 { 1479 {
1473 int ret; 1480 int ret;
1474 struct ocfs2_super *osb = OCFS2_SB(in 1481 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1475 struct page *page; 1482 struct page *page;
1476 handle_t *handle; 1483 handle_t *handle;
1477 struct ocfs2_dinode *di = (struct ocf 1484 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1478 1485
1479 handle = ocfs2_start_trans(osb, OCFS2 1486 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1480 if (IS_ERR(handle)) { 1487 if (IS_ERR(handle)) {
1481 ret = PTR_ERR(handle); 1488 ret = PTR_ERR(handle);
1482 mlog_errno(ret); 1489 mlog_errno(ret);
1483 goto out; 1490 goto out;
1484 } 1491 }
1485 1492
1486 page = find_or_create_page(mapping, 0 1493 page = find_or_create_page(mapping, 0, GFP_NOFS);
1487 if (!page) { 1494 if (!page) {
1488 ocfs2_commit_trans(osb, handl 1495 ocfs2_commit_trans(osb, handle);
1489 ret = -ENOMEM; 1496 ret = -ENOMEM;
1490 mlog_errno(ret); 1497 mlog_errno(ret);
1491 goto out; 1498 goto out;
1492 } 1499 }
1493 /* 1500 /*
1494 * If we don't set w_num_pages then t 1501 * If we don't set w_num_pages then this page won't get unlocked
1495 * and freed on cleanup of the write 1502 * and freed on cleanup of the write context.
1496 */ 1503 */
1497 wc->w_pages[0] = wc->w_target_page = 1504 wc->w_pages[0] = wc->w_target_page = page;
1498 wc->w_num_pages = 1; 1505 wc->w_num_pages = 1;
1499 1506
1500 ret = ocfs2_journal_access_di(handle, 1507 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1501 OCFS2_J 1508 OCFS2_JOURNAL_ACCESS_WRITE);
1502 if (ret) { 1509 if (ret) {
1503 ocfs2_commit_trans(osb, handl 1510 ocfs2_commit_trans(osb, handle);
1504 1511
1505 mlog_errno(ret); 1512 mlog_errno(ret);
1506 goto out; 1513 goto out;
1507 } 1514 }
1508 1515
1509 if (!(OCFS2_I(inode)->ip_dyn_features 1516 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1510 ocfs2_set_inode_data_inline(i 1517 ocfs2_set_inode_data_inline(inode, di);
1511 1518
1512 if (!PageUptodate(page)) { 1519 if (!PageUptodate(page)) {
1513 ret = ocfs2_read_inline_data( 1520 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1514 if (ret) { 1521 if (ret) {
1515 ocfs2_commit_trans(os 1522 ocfs2_commit_trans(osb, handle);
1516 1523
1517 goto out; 1524 goto out;
1518 } 1525 }
1519 } 1526 }
1520 1527
1521 wc->w_handle = handle; 1528 wc->w_handle = handle;
1522 out: 1529 out:
1523 return ret; 1530 return ret;
1524 } 1531 }
1525 1532
1526 int ocfs2_size_fits_inline_data(struct buffer 1533 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1527 { 1534 {
1528 struct ocfs2_dinode *di = (struct ocf 1535 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1529 1536
1530 if (new_size <= le16_to_cpu(di->id2.i 1537 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1531 return 1; 1538 return 1;
1532 return 0; 1539 return 0;
1533 } 1540 }
1534 1541
1535 static int ocfs2_try_to_write_inline_data(str 1542 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1536 str 1543 struct inode *inode, loff_t pos,
1537 uns 1544 unsigned len, struct page *mmap_page,
1538 str 1545 struct ocfs2_write_ctxt *wc)
1539 { 1546 {
1540 int ret, written = 0; 1547 int ret, written = 0;
1541 loff_t end = pos + len; 1548 loff_t end = pos + len;
1542 struct ocfs2_inode_info *oi = OCFS2_I 1549 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1543 struct ocfs2_dinode *di = NULL; 1550 struct ocfs2_dinode *di = NULL;
1544 1551
1545 trace_ocfs2_try_to_write_inline_data( 1552 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1546 1553 len, (unsigned long long)pos,
1547 1554 oi->ip_dyn_features);
1548 1555
1549 /* 1556 /*
1550 * Handle inodes which already have i 1557 * Handle inodes which already have inline data 1st.
1551 */ 1558 */
1552 if (oi->ip_dyn_features & OCFS2_INLIN 1559 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1553 if (mmap_page == NULL && 1560 if (mmap_page == NULL &&
1554 ocfs2_size_fits_inline_da 1561 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1555 goto do_inline_write; 1562 goto do_inline_write;
1556 1563
1557 /* 1564 /*
1558 * The write won't fit - we h 1565 * The write won't fit - we have to give this inode an
1559 * inline extent list now. 1566 * inline extent list now.
1560 */ 1567 */
1561 ret = ocfs2_convert_inline_da 1568 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1562 if (ret) 1569 if (ret)
1563 mlog_errno(ret); 1570 mlog_errno(ret);
1564 goto out; 1571 goto out;
1565 } 1572 }
1566 1573
1567 /* 1574 /*
1568 * Check whether the inode can accept 1575 * Check whether the inode can accept inline data.
1569 */ 1576 */
1570 if (oi->ip_clusters != 0 || i_size_re 1577 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1571 return 0; 1578 return 0;
1572 1579
1573 /* 1580 /*
1574 * Check whether the write can fit. 1581 * Check whether the write can fit.
1575 */ 1582 */
1576 di = (struct ocfs2_dinode *)wc->w_di_ 1583 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1577 if (mmap_page || 1584 if (mmap_page ||
1578 end > ocfs2_max_inline_data_with_ 1585 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1579 return 0; 1586 return 0;
1580 1587
1581 do_inline_write: 1588 do_inline_write:
1582 ret = ocfs2_write_begin_inline(mappin 1589 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1583 if (ret) { 1590 if (ret) {
1584 mlog_errno(ret); 1591 mlog_errno(ret);
1585 goto out; 1592 goto out;
1586 } 1593 }
1587 1594
1588 /* 1595 /*
1589 * This signals to the caller that th 1596 * This signals to the caller that the data can be written
1590 * inline. 1597 * inline.
1591 */ 1598 */
1592 written = 1; 1599 written = 1;
1593 out: 1600 out:
1594 return written ? written : ret; 1601 return written ? written : ret;
1595 } 1602 }
1596 1603
1597 /* 1604 /*
1598 * This function only does anything for file 1605 * This function only does anything for file systems which can't
1599 * handle sparse files. 1606 * handle sparse files.
1600 * 1607 *
1601 * What we want to do here is fill in any hol 1608 * What we want to do here is fill in any hole between the current end
1602 * of allocation and the end of our write. Th 1609 * of allocation and the end of our write. That way the rest of the
1603 * write path can treat it as an non-allocati 1610 * write path can treat it as an non-allocating write, which has no
1604 * special case code for sparse/nonsparse fil 1611 * special case code for sparse/nonsparse files.
1605 */ 1612 */
1606 static int ocfs2_expand_nonsparse_inode(struc 1613 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1607 struc 1614 struct buffer_head *di_bh,
1608 loff_ 1615 loff_t pos, unsigned len,
1609 struc 1616 struct ocfs2_write_ctxt *wc)
1610 { 1617 {
1611 int ret; 1618 int ret;
1612 loff_t newsize = pos + len; 1619 loff_t newsize = pos + len;
1613 1620
1614 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1621 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1615 1622
1616 if (newsize <= i_size_read(inode)) 1623 if (newsize <= i_size_read(inode))
1617 return 0; 1624 return 0;
1618 1625
1619 ret = ocfs2_extend_no_holes(inode, di 1626 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1620 if (ret) 1627 if (ret)
1621 mlog_errno(ret); 1628 mlog_errno(ret);
1622 1629
1623 /* There is no wc if this is call fro 1630 /* There is no wc if this is call from direct. */
1624 if (wc) 1631 if (wc)
1625 wc->w_first_new_cpos = 1632 wc->w_first_new_cpos =
1626 ocfs2_clusters_for_by 1633 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1627 1634
1628 return ret; 1635 return ret;
1629 } 1636 }
1630 1637
1631 static int ocfs2_zero_tail(struct inode *inod 1638 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1632 loff_t pos) 1639 loff_t pos)
1633 { 1640 {
1634 int ret = 0; 1641 int ret = 0;
1635 1642
1636 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1643 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1637 if (pos > i_size_read(inode)) 1644 if (pos > i_size_read(inode))
1638 ret = ocfs2_zero_extend(inode 1645 ret = ocfs2_zero_extend(inode, di_bh, pos);
1639 1646
1640 return ret; 1647 return ret;
1641 } 1648 }
1642 1649
1643 int ocfs2_write_begin_nolock(struct address_s 1650 int ocfs2_write_begin_nolock(struct address_space *mapping,
1644 loff_t pos, unsi 1651 loff_t pos, unsigned len, ocfs2_write_type_t type,
1645 struct page **pa 1652 struct page **pagep, void **fsdata,
1646 struct buffer_he 1653 struct buffer_head *di_bh, struct page *mmap_page)
1647 { 1654 {
1648 int ret, cluster_of_pages, credits = 1655 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1649 unsigned int clusters_to_alloc, exten 1656 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1650 struct ocfs2_write_ctxt *wc; 1657 struct ocfs2_write_ctxt *wc;
1651 struct inode *inode = mapping->host; 1658 struct inode *inode = mapping->host;
1652 struct ocfs2_super *osb = OCFS2_SB(in 1659 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1653 struct ocfs2_dinode *di; 1660 struct ocfs2_dinode *di;
1654 struct ocfs2_alloc_context *data_ac = 1661 struct ocfs2_alloc_context *data_ac = NULL;
1655 struct ocfs2_alloc_context *meta_ac = 1662 struct ocfs2_alloc_context *meta_ac = NULL;
1656 handle_t *handle; 1663 handle_t *handle;
1657 struct ocfs2_extent_tree et; 1664 struct ocfs2_extent_tree et;
1658 int try_free = 1, ret1; 1665 int try_free = 1, ret1;
1659 1666
1660 try_again: 1667 try_again:
1661 ret = ocfs2_alloc_write_ctxt(&wc, osb 1668 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
1662 if (ret) { 1669 if (ret) {
1663 mlog_errno(ret); 1670 mlog_errno(ret);
1664 return ret; 1671 return ret;
1665 } 1672 }
1666 1673
1667 if (ocfs2_supports_inline_data(osb)) 1674 if (ocfs2_supports_inline_data(osb)) {
1668 ret = ocfs2_try_to_write_inli 1675 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1669 1676 mmap_page, wc);
1670 if (ret == 1) { 1677 if (ret == 1) {
1671 ret = 0; 1678 ret = 0;
1672 goto success; 1679 goto success;
1673 } 1680 }
1674 if (ret < 0) { 1681 if (ret < 0) {
1675 mlog_errno(ret); 1682 mlog_errno(ret);
1676 goto out; 1683 goto out;
1677 } 1684 }
1678 } 1685 }
1679 1686
1680 /* Direct io change i_size late, shou 1687 /* Direct io change i_size late, should not zero tail here. */
1681 if (type != OCFS2_WRITE_DIRECT) { 1688 if (type != OCFS2_WRITE_DIRECT) {
1682 if (ocfs2_sparse_alloc(osb)) 1689 if (ocfs2_sparse_alloc(osb))
1683 ret = ocfs2_zero_tail 1690 ret = ocfs2_zero_tail(inode, di_bh, pos);
1684 else 1691 else
1685 ret = ocfs2_expand_no 1692 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1686 1693 len, wc);
1687 if (ret) { 1694 if (ret) {
1688 mlog_errno(ret); 1695 mlog_errno(ret);
1689 goto out; 1696 goto out;
1690 } 1697 }
1691 } 1698 }
1692 1699
1693 ret = ocfs2_check_range_for_refcount( 1700 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1694 if (ret < 0) { 1701 if (ret < 0) {
1695 mlog_errno(ret); 1702 mlog_errno(ret);
1696 goto out; 1703 goto out;
1697 } else if (ret == 1) { 1704 } else if (ret == 1) {
1698 clusters_need = wc->w_clen; 1705 clusters_need = wc->w_clen;
1699 ret = ocfs2_refcount_cow(inod 1706 ret = ocfs2_refcount_cow(inode, di_bh,
1700 wc-> 1707 wc->w_cpos, wc->w_clen, UINT_MAX);
1701 if (ret) { 1708 if (ret) {
1702 mlog_errno(ret); 1709 mlog_errno(ret);
1703 goto out; 1710 goto out;
1704 } 1711 }
1705 } 1712 }
1706 1713
1707 ret = ocfs2_populate_write_desc(inode 1714 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1708 &exte 1715 &extents_to_split);
1709 if (ret) { 1716 if (ret) {
1710 mlog_errno(ret); 1717 mlog_errno(ret);
1711 goto out; 1718 goto out;
1712 } 1719 }
1713 clusters_need += clusters_to_alloc; 1720 clusters_need += clusters_to_alloc;
1714 1721
1715 di = (struct ocfs2_dinode *)wc->w_di_ 1722 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1716 1723
1717 trace_ocfs2_write_begin_nolock( 1724 trace_ocfs2_write_begin_nolock(
1718 (unsigned long long)O 1725 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1719 (long long)i_size_rea 1726 (long long)i_size_read(inode),
1720 le32_to_cpu(di->i_clu 1727 le32_to_cpu(di->i_clusters),
1721 pos, len, type, mmap_ 1728 pos, len, type, mmap_page,
1722 clusters_to_alloc, ex 1729 clusters_to_alloc, extents_to_split);
1723 1730
1724 /* 1731 /*
1725 * We set w_target_from, w_target_to 1732 * We set w_target_from, w_target_to here so that
1726 * ocfs2_write_end() knows which rang 1733 * ocfs2_write_end() knows which range in the target page to
1727 * write out. An allocation requires 1734 * write out. An allocation requires that we write the entire
1728 * cluster range. 1735 * cluster range.
1729 */ 1736 */
1730 if (clusters_to_alloc || extents_to_s 1737 if (clusters_to_alloc || extents_to_split) {
1731 /* 1738 /*
1732 * XXX: We are stretching the 1739 * XXX: We are stretching the limits of
1733 * ocfs2_lock_allocators(). I 1740 * ocfs2_lock_allocators(). It greatly over-estimates
1734 * the work to be done. 1741 * the work to be done.
1735 */ 1742 */
1736 ocfs2_init_dinode_extent_tree 1743 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1737 1744 wc->w_di_bh);
1738 ret = ocfs2_lock_allocators(i 1745 ret = ocfs2_lock_allocators(inode, &et,
1739 c 1746 clusters_to_alloc, extents_to_split,
1740 & 1747 &data_ac, &meta_ac);
1741 if (ret) { 1748 if (ret) {
1742 mlog_errno(ret); 1749 mlog_errno(ret);
1743 goto out; 1750 goto out;
1744 } 1751 }
1745 1752
1746 if (data_ac) 1753 if (data_ac)
1747 data_ac->ac_resv = &O 1754 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1748 1755
1749 credits = ocfs2_calc_extend_c 1756 credits = ocfs2_calc_extend_credits(inode->i_sb,
1750 1757 &di->id2.i_list);
1751 } else if (type == OCFS2_WRITE_DIRECT 1758 } else if (type == OCFS2_WRITE_DIRECT)
1752 /* direct write needs not to 1759 /* direct write needs not to start trans if no extents alloc. */
1753 goto success; 1760 goto success;
1754 1761
1755 /* 1762 /*
1756 * We have to zero sparse allocated c 1763 * We have to zero sparse allocated clusters, unwritten extent clusters,
1757 * and non-sparse clusters we just ex 1764 * and non-sparse clusters we just extended. For non-sparse writes,
1758 * we know zeros will only be needed 1765 * we know zeros will only be needed in the first and/or last cluster.
1759 */ 1766 */
1760 if (wc->w_clen && (wc->w_desc[0].c_ne 1767 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1761 wc->w_desc[wc->w_c 1768 wc->w_desc[wc->w_clen - 1].c_needs_zero))
1762 cluster_of_pages = 1; 1769 cluster_of_pages = 1;
1763 else 1770 else
1764 cluster_of_pages = 0; 1771 cluster_of_pages = 0;
1765 1772
1766 ocfs2_set_target_boundaries(osb, wc, 1773 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1767 1774
1768 handle = ocfs2_start_trans(osb, credi 1775 handle = ocfs2_start_trans(osb, credits);
1769 if (IS_ERR(handle)) { 1776 if (IS_ERR(handle)) {
1770 ret = PTR_ERR(handle); 1777 ret = PTR_ERR(handle);
1771 mlog_errno(ret); 1778 mlog_errno(ret);
1772 goto out; 1779 goto out;
1773 } 1780 }
1774 1781
1775 wc->w_handle = handle; 1782 wc->w_handle = handle;
1776 1783
1777 if (clusters_to_alloc) { 1784 if (clusters_to_alloc) {
1778 ret = dquot_alloc_space_nodir 1785 ret = dquot_alloc_space_nodirty(inode,
1779 ocfs2_clusters_to_byt 1786 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1780 if (ret) 1787 if (ret)
1781 goto out_commit; 1788 goto out_commit;
1782 } 1789 }
1783 1790
1784 ret = ocfs2_journal_access_di(handle, 1791 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1785 OCFS2_J 1792 OCFS2_JOURNAL_ACCESS_WRITE);
1786 if (ret) { 1793 if (ret) {
1787 mlog_errno(ret); 1794 mlog_errno(ret);
1788 goto out_quota; 1795 goto out_quota;
1789 } 1796 }
1790 1797
1791 /* 1798 /*
1792 * Fill our page array first. That wa 1799 * Fill our page array first. That way we've grabbed enough so
1793 * that we can zero and flush if we e 1800 * that we can zero and flush if we error after adding the
1794 * extent. 1801 * extent.
1795 */ 1802 */
1796 ret = ocfs2_grab_pages_for_write(mapp 1803 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1797 clus 1804 cluster_of_pages, mmap_page);
1798 if (ret) { !! 1805 if (ret && ret != -EAGAIN) {
1799 /* <<
1800 * ocfs2_grab_pages_for_write <<
1801 * the target page. In this c <<
1802 * page. This will trigger th <<
1803 * the operation. <<
1804 */ <<
1805 if (type == OCFS2_WRITE_MMAP <<
1806 BUG_ON(wc->w_target_p <<
1807 ret = 0; <<
1808 goto out_quota; <<
1809 } <<
1810 <<
1811 mlog_errno(ret); 1806 mlog_errno(ret);
1812 goto out_quota; 1807 goto out_quota;
1813 } 1808 }
1814 1809
>> 1810 /*
>> 1811 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
>> 1812 * the target page. In this case, we exit with no error and no target
>> 1813 * page. This will trigger the caller, page_mkwrite(), to re-try
>> 1814 * the operation.
>> 1815 */
>> 1816 if (ret == -EAGAIN) {
>> 1817 BUG_ON(wc->w_target_page);
>> 1818 ret = 0;
>> 1819 goto out_quota;
>> 1820 }
>> 1821
1815 ret = ocfs2_write_cluster_by_desc(map 1822 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1816 len 1823 len);
1817 if (ret) { 1824 if (ret) {
1818 mlog_errno(ret); 1825 mlog_errno(ret);
1819 goto out_quota; 1826 goto out_quota;
1820 } 1827 }
1821 1828
1822 if (data_ac) 1829 if (data_ac)
1823 ocfs2_free_alloc_context(data 1830 ocfs2_free_alloc_context(data_ac);
1824 if (meta_ac) 1831 if (meta_ac)
1825 ocfs2_free_alloc_context(meta 1832 ocfs2_free_alloc_context(meta_ac);
1826 1833
1827 success: 1834 success:
1828 if (pagep) 1835 if (pagep)
1829 *pagep = wc->w_target_page; 1836 *pagep = wc->w_target_page;
1830 *fsdata = wc; 1837 *fsdata = wc;
1831 return 0; 1838 return 0;
1832 out_quota: 1839 out_quota:
1833 if (clusters_to_alloc) 1840 if (clusters_to_alloc)
1834 dquot_free_space(inode, 1841 dquot_free_space(inode,
1835 ocfs2_clusters_to_b 1842 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1836 out_commit: 1843 out_commit:
1837 ocfs2_commit_trans(osb, handle); 1844 ocfs2_commit_trans(osb, handle);
1838 1845
1839 out: 1846 out:
1840 /* 1847 /*
1841 * The mmapped page won't be unlocked 1848 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1842 * even in case of error here like EN 1849 * even in case of error here like ENOSPC and ENOMEM. So, we need
1843 * to unlock the target page manually 1850 * to unlock the target page manually to prevent deadlocks when
1844 * retrying again on ENOSPC, or when 1851 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1845 * to VM code. 1852 * to VM code.
1846 */ 1853 */
1847 if (wc->w_target_locked) 1854 if (wc->w_target_locked)
1848 unlock_page(mmap_page); 1855 unlock_page(mmap_page);
1849 1856
1850 ocfs2_free_write_ctxt(inode, wc); 1857 ocfs2_free_write_ctxt(inode, wc);
1851 1858
1852 if (data_ac) { 1859 if (data_ac) {
1853 ocfs2_free_alloc_context(data 1860 ocfs2_free_alloc_context(data_ac);
1854 data_ac = NULL; 1861 data_ac = NULL;
1855 } 1862 }
1856 if (meta_ac) { 1863 if (meta_ac) {
1857 ocfs2_free_alloc_context(meta 1864 ocfs2_free_alloc_context(meta_ac);
1858 meta_ac = NULL; 1865 meta_ac = NULL;
1859 } 1866 }
1860 1867
1861 if (ret == -ENOSPC && try_free) { 1868 if (ret == -ENOSPC && try_free) {
1862 /* 1869 /*
1863 * Try to free some truncate 1870 * Try to free some truncate log so that we can have enough
1864 * clusters to allocate. 1871 * clusters to allocate.
1865 */ 1872 */
1866 try_free = 0; 1873 try_free = 0;
1867 1874
1868 ret1 = ocfs2_try_to_free_trun 1875 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1869 if (ret1 == 1) 1876 if (ret1 == 1)
1870 goto try_again; 1877 goto try_again;
1871 1878
1872 if (ret1 < 0) 1879 if (ret1 < 0)
1873 mlog_errno(ret1); 1880 mlog_errno(ret1);
1874 } 1881 }
1875 1882
1876 return ret; 1883 return ret;
1877 } 1884 }
1878 1885
1879 static int ocfs2_write_begin(struct file *fil 1886 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1880 loff_t pos, unsi !! 1887 loff_t pos, unsigned len, unsigned flags,
1881 struct page **pa 1888 struct page **pagep, void **fsdata)
1882 { 1889 {
1883 int ret; 1890 int ret;
1884 struct buffer_head *di_bh = NULL; 1891 struct buffer_head *di_bh = NULL;
1885 struct inode *inode = mapping->host; 1892 struct inode *inode = mapping->host;
1886 1893
1887 ret = ocfs2_inode_lock(inode, &di_bh, 1894 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1888 if (ret) { 1895 if (ret) {
1889 mlog_errno(ret); 1896 mlog_errno(ret);
1890 return ret; 1897 return ret;
1891 } 1898 }
1892 1899
1893 /* 1900 /*
1894 * Take alloc sem here to prevent con 1901 * Take alloc sem here to prevent concurrent lookups. That way
1895 * the mapping, zeroing and tree mani 1902 * the mapping, zeroing and tree manipulation within
1896 * ocfs2_write() will be safe against !! 1903 * ocfs2_write() will be safe against ->readpage(). This
1897 * should also serve to lock out allo 1904 * should also serve to lock out allocation from a shared
1898 * writeable region. 1905 * writeable region.
1899 */ 1906 */
1900 down_write(&OCFS2_I(inode)->ip_alloc_ 1907 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1901 1908
1902 ret = ocfs2_write_begin_nolock(mappin 1909 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1903 pagep, 1910 pagep, fsdata, di_bh, NULL);
1904 if (ret) { 1911 if (ret) {
1905 mlog_errno(ret); 1912 mlog_errno(ret);
1906 goto out_fail; 1913 goto out_fail;
1907 } 1914 }
1908 1915
1909 brelse(di_bh); 1916 brelse(di_bh);
1910 1917
1911 return 0; 1918 return 0;
1912 1919
1913 out_fail: 1920 out_fail:
1914 up_write(&OCFS2_I(inode)->ip_alloc_se 1921 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1915 1922
1916 brelse(di_bh); 1923 brelse(di_bh);
1917 ocfs2_inode_unlock(inode, 1); 1924 ocfs2_inode_unlock(inode, 1);
1918 1925
1919 return ret; 1926 return ret;
1920 } 1927 }
1921 1928
1922 static void ocfs2_write_end_inline(struct ino 1929 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1923 unsigned l 1930 unsigned len, unsigned *copied,
1924 struct ocf 1931 struct ocfs2_dinode *di,
1925 struct ocf 1932 struct ocfs2_write_ctxt *wc)
1926 { 1933 {
1927 void *kaddr; 1934 void *kaddr;
1928 1935
1929 if (unlikely(*copied < len)) { 1936 if (unlikely(*copied < len)) {
1930 if (!PageUptodate(wc->w_targe 1937 if (!PageUptodate(wc->w_target_page)) {
1931 *copied = 0; 1938 *copied = 0;
1932 return; 1939 return;
1933 } 1940 }
1934 } 1941 }
1935 1942
1936 kaddr = kmap_atomic(wc->w_target_page 1943 kaddr = kmap_atomic(wc->w_target_page);
1937 memcpy(di->id2.i_data.id_data + pos, 1944 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1938 kunmap_atomic(kaddr); 1945 kunmap_atomic(kaddr);
1939 1946
1940 trace_ocfs2_write_end_inline( 1947 trace_ocfs2_write_end_inline(
1941 (unsigned long long)OCFS2_I(inod 1948 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1942 (unsigned long long)pos, *copied 1949 (unsigned long long)pos, *copied,
1943 le16_to_cpu(di->id2.i_data.id_co 1950 le16_to_cpu(di->id2.i_data.id_count),
1944 le16_to_cpu(di->i_dyn_features)) 1951 le16_to_cpu(di->i_dyn_features));
1945 } 1952 }
1946 1953
1947 int ocfs2_write_end_nolock(struct address_spa 1954 int ocfs2_write_end_nolock(struct address_space *mapping,
1948 loff_t pos, unsign 1955 loff_t pos, unsigned len, unsigned copied, void *fsdata)
1949 { 1956 {
1950 int i, ret; 1957 int i, ret;
1951 unsigned from, to, start = pos & (PAG 1958 unsigned from, to, start = pos & (PAGE_SIZE - 1);
1952 struct inode *inode = mapping->host; 1959 struct inode *inode = mapping->host;
1953 struct ocfs2_super *osb = OCFS2_SB(in 1960 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1954 struct ocfs2_write_ctxt *wc = fsdata; 1961 struct ocfs2_write_ctxt *wc = fsdata;
1955 struct ocfs2_dinode *di = (struct ocf 1962 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1956 handle_t *handle = wc->w_handle; 1963 handle_t *handle = wc->w_handle;
1957 struct page *tmppage; 1964 struct page *tmppage;
1958 1965
1959 BUG_ON(!list_empty(&wc->w_unwritten_l 1966 BUG_ON(!list_empty(&wc->w_unwritten_list));
1960 1967
1961 if (handle) { 1968 if (handle) {
1962 ret = ocfs2_journal_access_di 1969 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1963 wc->w_di_bh, 1970 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1964 if (ret) { 1971 if (ret) {
1965 copied = ret; 1972 copied = ret;
1966 mlog_errno(ret); 1973 mlog_errno(ret);
1967 goto out; 1974 goto out;
1968 } 1975 }
1969 } 1976 }
1970 1977
1971 if (OCFS2_I(inode)->ip_dyn_features & 1978 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1972 ocfs2_write_end_inline(inode, 1979 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1973 goto out_write_size; 1980 goto out_write_size;
1974 } 1981 }
1975 1982
1976 if (unlikely(copied < len) && wc->w_t 1983 if (unlikely(copied < len) && wc->w_target_page) {
1977 loff_t new_isize; <<
1978 <<
1979 if (!PageUptodate(wc->w_targe 1984 if (!PageUptodate(wc->w_target_page))
1980 copied = 0; 1985 copied = 0;
1981 1986
1982 new_isize = max_t(loff_t, i_s !! 1987 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1983 if (new_isize > page_offset(w !! 1988 start+len);
1984 ocfs2_zero_new_buffer <<
1985 <<
1986 else { <<
1987 /* <<
1988 * When page is fully <<
1989 * failed), do not bo <<
1990 * it instead so that <<
1991 * put page & buffer <<
1992 * state. <<
1993 */ <<
1994 block_invalidate_foli <<
1995 <<
1996 } <<
1997 } 1989 }
1998 if (wc->w_target_page) 1990 if (wc->w_target_page)
1999 flush_dcache_page(wc->w_targe 1991 flush_dcache_page(wc->w_target_page);
2000 1992
2001 for(i = 0; i < wc->w_num_pages; i++) 1993 for(i = 0; i < wc->w_num_pages; i++) {
2002 tmppage = wc->w_pages[i]; 1994 tmppage = wc->w_pages[i];
2003 1995
2004 /* This is the direct io targ 1996 /* This is the direct io target page. */
2005 if (tmppage == NULL) 1997 if (tmppage == NULL)
2006 continue; 1998 continue;
2007 1999
2008 if (tmppage == wc->w_target_p 2000 if (tmppage == wc->w_target_page) {
2009 from = wc->w_target_f 2001 from = wc->w_target_from;
2010 to = wc->w_target_to; 2002 to = wc->w_target_to;
2011 2003
2012 BUG_ON(from > PAGE_SI 2004 BUG_ON(from > PAGE_SIZE ||
2013 to > PAGE_SIZE 2005 to > PAGE_SIZE ||
2014 to < from); 2006 to < from);
2015 } else { 2007 } else {
2016 /* 2008 /*
2017 * Pages adjacent to 2009 * Pages adjacent to the target (if any) imply
2018 * a hole-filling wri 2010 * a hole-filling write in which case we want
2019 * to flush their ent 2011 * to flush their entire range.
2020 */ 2012 */
2021 from = 0; 2013 from = 0;
2022 to = PAGE_SIZE; 2014 to = PAGE_SIZE;
2023 } 2015 }
2024 2016
2025 if (page_has_buffers(tmppage) 2017 if (page_has_buffers(tmppage)) {
2026 if (handle && ocfs2_s 2018 if (handle && ocfs2_should_order_data(inode)) {
2027 loff_t start_ 2019 loff_t start_byte =
2028 ((lof 2020 ((loff_t)tmppage->index << PAGE_SHIFT) +
2029 from; 2021 from;
2030 loff_t length 2022 loff_t length = to - from;
2031 ocfs2_jbd2_in 2023 ocfs2_jbd2_inode_add_write(handle, inode,
2032 2024 start_byte, length);
2033 } 2025 }
2034 block_commit_write(tm 2026 block_commit_write(tmppage, from, to);
2035 } 2027 }
2036 } 2028 }
2037 2029
2038 out_write_size: 2030 out_write_size:
2039 /* Direct io do not update i_size her 2031 /* Direct io do not update i_size here. */
2040 if (wc->w_type != OCFS2_WRITE_DIRECT) 2032 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2041 pos += copied; 2033 pos += copied;
2042 if (pos > i_size_read(inode)) 2034 if (pos > i_size_read(inode)) {
2043 i_size_write(inode, p 2035 i_size_write(inode, pos);
2044 mark_inode_dirty(inod 2036 mark_inode_dirty(inode);
2045 } 2037 }
2046 inode->i_blocks = ocfs2_inode 2038 inode->i_blocks = ocfs2_inode_sector_count(inode);
2047 di->i_size = cpu_to_le64((u64 2039 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2048 inode_set_mtime_to_ts(inode, !! 2040 inode->i_mtime = inode->i_ctime = current_time(inode);
2049 di->i_mtime = di->i_ctime = c !! 2041 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2050 di->i_mtime_nsec = di->i_ctim !! 2042 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2051 if (handle) 2043 if (handle)
2052 ocfs2_update_inode_fs 2044 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2053 } 2045 }
2054 if (handle) 2046 if (handle)
2055 ocfs2_journal_dirty(handle, w 2047 ocfs2_journal_dirty(handle, wc->w_di_bh);
2056 2048
2057 out: 2049 out:
2058 /* unlock pages before dealloc since 2050 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2059 * lock, or it will cause a deadlock 2051 * lock, or it will cause a deadlock since journal commit threads holds
2060 * this lock and will ask for the pag 2052 * this lock and will ask for the page lock when flushing the data.
2061 * put it here to preserve the unlock 2053 * put it here to preserve the unlock order.
2062 */ 2054 */
2063 ocfs2_unlock_pages(wc); 2055 ocfs2_unlock_pages(wc);
2064 2056
2065 if (handle) 2057 if (handle)
2066 ocfs2_commit_trans(osb, handl 2058 ocfs2_commit_trans(osb, handle);
2067 2059
2068 ocfs2_run_deallocs(osb, &wc->w_deallo 2060 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2069 2061
2070 brelse(wc->w_di_bh); 2062 brelse(wc->w_di_bh);
2071 kfree(wc); 2063 kfree(wc);
2072 2064
2073 return copied; 2065 return copied;
2074 } 2066 }
2075 2067
2076 static int ocfs2_write_end(struct file *file, 2068 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2077 loff_t pos, unsign 2069 loff_t pos, unsigned len, unsigned copied,
2078 struct page *page, 2070 struct page *page, void *fsdata)
2079 { 2071 {
2080 int ret; 2072 int ret;
2081 struct inode *inode = mapping->host; 2073 struct inode *inode = mapping->host;
2082 2074
2083 ret = ocfs2_write_end_nolock(mapping, 2075 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata);
2084 2076
2085 up_write(&OCFS2_I(inode)->ip_alloc_se 2077 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2086 ocfs2_inode_unlock(inode, 1); 2078 ocfs2_inode_unlock(inode, 1);
2087 2079
2088 return ret; 2080 return ret;
2089 } 2081 }
2090 2082
2091 struct ocfs2_dio_write_ctxt { 2083 struct ocfs2_dio_write_ctxt {
2092 struct list_head dw_zero_list; 2084 struct list_head dw_zero_list;
2093 unsigned dw_zero_count 2085 unsigned dw_zero_count;
2094 int dw_orphaned; 2086 int dw_orphaned;
2095 pid_t dw_writer_pid 2087 pid_t dw_writer_pid;
2096 }; 2088 };
2097 2089
2098 static struct ocfs2_dio_write_ctxt * 2090 static struct ocfs2_dio_write_ctxt *
2099 ocfs2_dio_alloc_write_ctx(struct buffer_head 2091 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2100 { 2092 {
2101 struct ocfs2_dio_write_ctxt *dwc = NU 2093 struct ocfs2_dio_write_ctxt *dwc = NULL;
2102 2094
2103 if (bh->b_private) 2095 if (bh->b_private)
2104 return bh->b_private; 2096 return bh->b_private;
2105 2097
2106 dwc = kmalloc(sizeof(struct ocfs2_dio 2098 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2107 if (dwc == NULL) 2099 if (dwc == NULL)
2108 return NULL; 2100 return NULL;
2109 INIT_LIST_HEAD(&dwc->dw_zero_list); 2101 INIT_LIST_HEAD(&dwc->dw_zero_list);
2110 dwc->dw_zero_count = 0; 2102 dwc->dw_zero_count = 0;
2111 dwc->dw_orphaned = 0; 2103 dwc->dw_orphaned = 0;
2112 dwc->dw_writer_pid = task_pid_nr(curr 2104 dwc->dw_writer_pid = task_pid_nr(current);
2113 bh->b_private = dwc; 2105 bh->b_private = dwc;
2114 *alloc = 1; 2106 *alloc = 1;
2115 2107
2116 return dwc; 2108 return dwc;
2117 } 2109 }
2118 2110
2119 static void ocfs2_dio_free_write_ctx(struct i 2111 static void ocfs2_dio_free_write_ctx(struct inode *inode,
2120 struct o 2112 struct ocfs2_dio_write_ctxt *dwc)
2121 { 2113 {
2122 ocfs2_free_unwritten_list(inode, &dwc 2114 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2123 kfree(dwc); 2115 kfree(dwc);
2124 } 2116 }
2125 2117
2126 /* 2118 /*
2127 * TODO: Make this into a generic get_blocks 2119 * TODO: Make this into a generic get_blocks function.
2128 * 2120 *
2129 * From do_direct_io in direct-io.c: 2121 * From do_direct_io in direct-io.c:
2130 * "So what we do is to permit the ->get_blo 2122 * "So what we do is to permit the ->get_blocks function to populate
2131 * bh.b_size with the size of IO which is p 2123 * bh.b_size with the size of IO which is permitted at this offset and
2132 * this i_blkbits." 2124 * this i_blkbits."
2133 * 2125 *
2134 * This function is called directly from get_ 2126 * This function is called directly from get_more_blocks in direct-io.c.
2135 * 2127 *
2136 * called like this: dio->get_blocks(dio->ino 2128 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2137 * fs_co 2129 * fs_count, map_bh, dio->rw == WRITE);
2138 */ 2130 */
2139 static int ocfs2_dio_wr_get_block(struct inod 2131 static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock,
2140 struct buffer_ 2132 struct buffer_head *bh_result, int create)
2141 { 2133 {
2142 struct ocfs2_super *osb = OCFS2_SB(in 2134 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2143 struct ocfs2_inode_info *oi = OCFS2_I 2135 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2144 struct ocfs2_write_ctxt *wc; 2136 struct ocfs2_write_ctxt *wc;
2145 struct ocfs2_write_cluster_desc *desc 2137 struct ocfs2_write_cluster_desc *desc = NULL;
2146 struct ocfs2_dio_write_ctxt *dwc = NU 2138 struct ocfs2_dio_write_ctxt *dwc = NULL;
2147 struct buffer_head *di_bh = NULL; 2139 struct buffer_head *di_bh = NULL;
2148 u64 p_blkno; 2140 u64 p_blkno;
2149 unsigned int i_blkbits = inode->i_sb- 2141 unsigned int i_blkbits = inode->i_sb->s_blocksize_bits;
2150 loff_t pos = iblock << i_blkbits; 2142 loff_t pos = iblock << i_blkbits;
2151 sector_t endblk = (i_size_read(inode) 2143 sector_t endblk = (i_size_read(inode) - 1) >> i_blkbits;
2152 unsigned len, total_len = bh_result-> 2144 unsigned len, total_len = bh_result->b_size;
2153 int ret = 0, first_get_block = 0; 2145 int ret = 0, first_get_block = 0;
2154 2146
2155 len = osb->s_clustersize - (pos & (os 2147 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2156 len = min(total_len, len); 2148 len = min(total_len, len);
2157 2149
2158 /* 2150 /*
2159 * bh_result->b_size is count in get_ 2151 * bh_result->b_size is count in get_more_blocks according to write
2160 * "pos" and "end", we need map twice 2152 * "pos" and "end", we need map twice to return different buffer state:
2161 * 1. area in file size, not set NEW; 2153 * 1. area in file size, not set NEW;
2162 * 2. area out file size, set NEW. 2154 * 2. area out file size, set NEW.
2163 * 2155 *
2164 * iblock endblk 2156 * iblock endblk
2165 * |--------|---------|---------|---- 2157 * |--------|---------|---------|---------
2166 * |<-------area in file------->| 2158 * |<-------area in file------->|
2167 */ 2159 */
2168 2160
2169 if ((iblock <= endblk) && 2161 if ((iblock <= endblk) &&
2170 ((iblock + ((len - 1) >> i_blkbit 2162 ((iblock + ((len - 1) >> i_blkbits)) > endblk))
2171 len = (endblk - iblock + 1) < 2163 len = (endblk - iblock + 1) << i_blkbits;
2172 2164
2173 mlog(0, "get block of %lu at %llu:%u 2165 mlog(0, "get block of %lu at %llu:%u req %u\n",
2174 inode->i_ino, pos, le 2166 inode->i_ino, pos, len, total_len);
2175 2167
2176 /* 2168 /*
2177 * Because we need to change file siz 2169 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2178 * we may need to add it to orphan di 2170 * we may need to add it to orphan dir. So can not fall to fast path
2179 * while file size will be changed. 2171 * while file size will be changed.
2180 */ 2172 */
2181 if (pos + total_len <= i_size_read(in 2173 if (pos + total_len <= i_size_read(inode)) {
2182 2174
2183 /* This is the fast path for 2175 /* This is the fast path for re-write. */
2184 ret = ocfs2_lock_get_block(in 2176 ret = ocfs2_lock_get_block(inode, iblock, bh_result, create);
2185 if (buffer_mapped(bh_result) 2177 if (buffer_mapped(bh_result) &&
2186 !buffer_new(bh_result) && 2178 !buffer_new(bh_result) &&
2187 ret == 0) 2179 ret == 0)
2188 goto out; 2180 goto out;
2189 2181
2190 /* Clear state set by ocfs2_g 2182 /* Clear state set by ocfs2_get_block. */
2191 bh_result->b_state = 0; 2183 bh_result->b_state = 0;
2192 } 2184 }
2193 2185
2194 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2186 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2195 if (unlikely(dwc == NULL)) { 2187 if (unlikely(dwc == NULL)) {
2196 ret = -ENOMEM; 2188 ret = -ENOMEM;
2197 mlog_errno(ret); 2189 mlog_errno(ret);
2198 goto out; 2190 goto out;
2199 } 2191 }
2200 2192
2201 if (ocfs2_clusters_for_bytes(inode->i 2193 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2202 ocfs2_clusters_for_bytes(inode->i 2194 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2203 !dwc->dw_orphaned) { 2195 !dwc->dw_orphaned) {
2204 /* 2196 /*
2205 * when we are going to alloc 2197 * when we are going to alloc extents beyond file size, add the
2206 * inode to orphan dir, so we 2198 * inode to orphan dir, so we can recall those spaces when
2207 * system crashed during writ 2199 * system crashed during write.
2208 */ 2200 */
2209 ret = ocfs2_add_inode_to_orph 2201 ret = ocfs2_add_inode_to_orphan(osb, inode);
2210 if (ret < 0) { 2202 if (ret < 0) {
2211 mlog_errno(ret); 2203 mlog_errno(ret);
2212 goto out; 2204 goto out;
2213 } 2205 }
2214 dwc->dw_orphaned = 1; 2206 dwc->dw_orphaned = 1;
2215 } 2207 }
2216 2208
2217 ret = ocfs2_inode_lock(inode, &di_bh, 2209 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2218 if (ret) { 2210 if (ret) {
2219 mlog_errno(ret); 2211 mlog_errno(ret);
2220 goto out; 2212 goto out;
2221 } 2213 }
2222 2214
2223 down_write(&oi->ip_alloc_sem); 2215 down_write(&oi->ip_alloc_sem);
2224 2216
2225 if (first_get_block) { 2217 if (first_get_block) {
2226 if (ocfs2_sparse_alloc(osb)) 2218 if (ocfs2_sparse_alloc(osb))
2227 ret = ocfs2_zero_tail 2219 ret = ocfs2_zero_tail(inode, di_bh, pos);
2228 else 2220 else
2229 ret = ocfs2_expand_no 2221 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2230 2222 total_len, NULL);
2231 if (ret < 0) { 2223 if (ret < 0) {
2232 mlog_errno(ret); 2224 mlog_errno(ret);
2233 goto unlock; 2225 goto unlock;
2234 } 2226 }
2235 } 2227 }
2236 2228
2237 ret = ocfs2_write_begin_nolock(inode- 2229 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2238 OCFS2_ 2230 OCFS2_WRITE_DIRECT, NULL,
2239 (void 2231 (void **)&wc, di_bh, NULL);
2240 if (ret) { 2232 if (ret) {
2241 mlog_errno(ret); 2233 mlog_errno(ret);
2242 goto unlock; 2234 goto unlock;
2243 } 2235 }
2244 2236
2245 desc = &wc->w_desc[0]; 2237 desc = &wc->w_desc[0];
2246 2238
2247 p_blkno = ocfs2_clusters_to_blocks(in 2239 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2248 BUG_ON(p_blkno == 0); 2240 BUG_ON(p_blkno == 0);
2249 p_blkno += iblock & (u64)(ocfs2_clust 2241 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2250 2242
2251 map_bh(bh_result, inode->i_sb, p_blkn 2243 map_bh(bh_result, inode->i_sb, p_blkno);
2252 bh_result->b_size = len; 2244 bh_result->b_size = len;
2253 if (desc->c_needs_zero) 2245 if (desc->c_needs_zero)
2254 set_buffer_new(bh_result); 2246 set_buffer_new(bh_result);
2255 2247
2256 if (iblock > endblk) 2248 if (iblock > endblk)
2257 set_buffer_new(bh_result); 2249 set_buffer_new(bh_result);
2258 2250
2259 /* May sleep in end_io. It should not 2251 /* May sleep in end_io. It should not happen in a irq context. So defer
2260 * it to dio work queue. */ 2252 * it to dio work queue. */
2261 set_buffer_defer_completion(bh_result 2253 set_buffer_defer_completion(bh_result);
2262 2254
2263 if (!list_empty(&wc->w_unwritten_list 2255 if (!list_empty(&wc->w_unwritten_list)) {
2264 struct ocfs2_unwritten_extent 2256 struct ocfs2_unwritten_extent *ue = NULL;
2265 2257
2266 ue = list_first_entry(&wc->w_ 2258 ue = list_first_entry(&wc->w_unwritten_list,
2267 struct 2259 struct ocfs2_unwritten_extent,
2268 ue_node 2260 ue_node);
2269 BUG_ON(ue->ue_cpos != desc->c 2261 BUG_ON(ue->ue_cpos != desc->c_cpos);
2270 /* The physical address may b 2262 /* The physical address may be 0, fill it. */
2271 ue->ue_phys = desc->c_phys; 2263 ue->ue_phys = desc->c_phys;
2272 2264
2273 list_splice_tail_init(&wc->w_ 2265 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
2274 dwc->dw_zero_count += wc->w_u 2266 dwc->dw_zero_count += wc->w_unwritten_count;
2275 } 2267 }
2276 2268
2277 ret = ocfs2_write_end_nolock(inode->i 2269 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc);
2278 BUG_ON(ret != len); 2270 BUG_ON(ret != len);
2279 ret = 0; 2271 ret = 0;
2280 unlock: 2272 unlock:
2281 up_write(&oi->ip_alloc_sem); 2273 up_write(&oi->ip_alloc_sem);
2282 ocfs2_inode_unlock(inode, 1); 2274 ocfs2_inode_unlock(inode, 1);
2283 brelse(di_bh); 2275 brelse(di_bh);
2284 out: 2276 out:
>> 2277 if (ret < 0)
>> 2278 ret = -EIO;
2285 return ret; 2279 return ret;
2286 } 2280 }
2287 2281
2288 static int ocfs2_dio_end_io_write(struct inod 2282 static int ocfs2_dio_end_io_write(struct inode *inode,
2289 struct ocfs 2283 struct ocfs2_dio_write_ctxt *dwc,
2290 loff_t offs 2284 loff_t offset,
2291 ssize_t byt 2285 ssize_t bytes)
2292 { 2286 {
2293 struct ocfs2_cached_dealloc_ctxt deal 2287 struct ocfs2_cached_dealloc_ctxt dealloc;
2294 struct ocfs2_extent_tree et; 2288 struct ocfs2_extent_tree et;
2295 struct ocfs2_super *osb = OCFS2_SB(in 2289 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2296 struct ocfs2_inode_info *oi = OCFS2_I 2290 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2297 struct ocfs2_unwritten_extent *ue = N 2291 struct ocfs2_unwritten_extent *ue = NULL;
2298 struct buffer_head *di_bh = NULL; 2292 struct buffer_head *di_bh = NULL;
2299 struct ocfs2_dinode *di; 2293 struct ocfs2_dinode *di;
2300 struct ocfs2_alloc_context *data_ac = 2294 struct ocfs2_alloc_context *data_ac = NULL;
2301 struct ocfs2_alloc_context *meta_ac = 2295 struct ocfs2_alloc_context *meta_ac = NULL;
2302 handle_t *handle = NULL; 2296 handle_t *handle = NULL;
2303 loff_t end = offset + bytes; 2297 loff_t end = offset + bytes;
2304 int ret = 0, credits = 0; 2298 int ret = 0, credits = 0;
2305 2299
2306 ocfs2_init_dealloc_ctxt(&dealloc); 2300 ocfs2_init_dealloc_ctxt(&dealloc);
2307 2301
2308 /* We do clear unwritten, delete orph 2302 /* We do clear unwritten, delete orphan, change i_size here. If neither
2309 * of these happen, we can skip all t 2303 * of these happen, we can skip all this. */
2310 if (list_empty(&dwc->dw_zero_list) && 2304 if (list_empty(&dwc->dw_zero_list) &&
2311 end <= i_size_read(inode) && 2305 end <= i_size_read(inode) &&
2312 !dwc->dw_orphaned) 2306 !dwc->dw_orphaned)
2313 goto out; 2307 goto out;
2314 2308
2315 ret = ocfs2_inode_lock(inode, &di_bh, 2309 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2316 if (ret < 0) { 2310 if (ret < 0) {
2317 mlog_errno(ret); 2311 mlog_errno(ret);
2318 goto out; 2312 goto out;
2319 } 2313 }
2320 2314
2321 down_write(&oi->ip_alloc_sem); 2315 down_write(&oi->ip_alloc_sem);
2322 2316
2323 /* Delete orphan before acquire i_rws !! 2317 /* Delete orphan before acquire i_mutex. */
2324 if (dwc->dw_orphaned) { 2318 if (dwc->dw_orphaned) {
2325 BUG_ON(dwc->dw_writer_pid != 2319 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2326 2320
2327 end = end > i_size_read(inode 2321 end = end > i_size_read(inode) ? end : 0;
2328 2322
2329 ret = ocfs2_del_inode_from_or 2323 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2330 !!end, end); 2324 !!end, end);
2331 if (ret < 0) 2325 if (ret < 0)
2332 mlog_errno(ret); 2326 mlog_errno(ret);
2333 } 2327 }
2334 2328
2335 di = (struct ocfs2_dinode *)di_bh->b_ 2329 di = (struct ocfs2_dinode *)di_bh->b_data;
2336 2330
2337 ocfs2_init_dinode_extent_tree(&et, IN 2331 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2338 2332
2339 /* Attach dealloc with extent tree in 2333 /* Attach dealloc with extent tree in case that we may reuse extents
2340 * which are already unlinked from cu 2334 * which are already unlinked from current extent tree due to extent
2341 * rotation and merging. 2335 * rotation and merging.
2342 */ 2336 */
2343 et.et_dealloc = &dealloc; 2337 et.et_dealloc = &dealloc;
2344 2338
2345 ret = ocfs2_lock_allocators(inode, &e 2339 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2346 &data_ac, 2340 &data_ac, &meta_ac);
2347 if (ret) { 2341 if (ret) {
2348 mlog_errno(ret); 2342 mlog_errno(ret);
2349 goto unlock; 2343 goto unlock;
2350 } 2344 }
2351 2345
2352 credits = ocfs2_calc_extend_credits(i 2346 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2353 2347
2354 handle = ocfs2_start_trans(osb, credi 2348 handle = ocfs2_start_trans(osb, credits);
2355 if (IS_ERR(handle)) { 2349 if (IS_ERR(handle)) {
2356 ret = PTR_ERR(handle); 2350 ret = PTR_ERR(handle);
2357 mlog_errno(ret); 2351 mlog_errno(ret);
2358 goto unlock; 2352 goto unlock;
2359 } 2353 }
2360 ret = ocfs2_journal_access_di(handle, 2354 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2361 OCFS2_J 2355 OCFS2_JOURNAL_ACCESS_WRITE);
2362 if (ret) { 2356 if (ret) {
2363 mlog_errno(ret); 2357 mlog_errno(ret);
2364 goto commit; 2358 goto commit;
2365 } 2359 }
2366 2360
2367 list_for_each_entry(ue, &dwc->dw_zero 2361 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) {
2368 ret = ocfs2_assure_trans_cred <<
2369 if (ret < 0) { <<
2370 mlog_errno(ret); <<
2371 break; <<
2372 } <<
2373 ret = ocfs2_mark_extent_writt 2362 ret = ocfs2_mark_extent_written(inode, &et, handle,
2374 2363 ue->ue_cpos, 1,
2375 2364 ue->ue_phys,
2376 2365 meta_ac, &dealloc);
2377 if (ret < 0) { 2366 if (ret < 0) {
2378 mlog_errno(ret); 2367 mlog_errno(ret);
2379 break; 2368 break;
2380 } 2369 }
2381 } 2370 }
2382 2371
2383 if (end > i_size_read(inode)) { 2372 if (end > i_size_read(inode)) {
2384 ret = ocfs2_set_inode_size(ha 2373 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2385 if (ret < 0) 2374 if (ret < 0)
2386 mlog_errno(ret); 2375 mlog_errno(ret);
2387 } 2376 }
2388 commit: 2377 commit:
2389 ocfs2_commit_trans(osb, handle); 2378 ocfs2_commit_trans(osb, handle);
2390 unlock: 2379 unlock:
2391 up_write(&oi->ip_alloc_sem); 2380 up_write(&oi->ip_alloc_sem);
2392 ocfs2_inode_unlock(inode, 1); 2381 ocfs2_inode_unlock(inode, 1);
2393 brelse(di_bh); 2382 brelse(di_bh);
2394 out: 2383 out:
2395 if (data_ac) 2384 if (data_ac)
2396 ocfs2_free_alloc_context(data 2385 ocfs2_free_alloc_context(data_ac);
2397 if (meta_ac) 2386 if (meta_ac)
2398 ocfs2_free_alloc_context(meta 2387 ocfs2_free_alloc_context(meta_ac);
2399 ocfs2_run_deallocs(osb, &dealloc); 2388 ocfs2_run_deallocs(osb, &dealloc);
2400 ocfs2_dio_free_write_ctx(inode, dwc); 2389 ocfs2_dio_free_write_ctx(inode, dwc);
2401 2390
2402 return ret; 2391 return ret;
2403 } 2392 }
2404 2393
2405 /* 2394 /*
2406 * ocfs2_dio_end_io is called by the dio core 2395 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
2407 * particularly interested in the aio/dio cas 2396 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2408 * to protect io on one node from truncation 2397 * to protect io on one node from truncation on another.
2409 */ 2398 */
2410 static int ocfs2_dio_end_io(struct kiocb *ioc 2399 static int ocfs2_dio_end_io(struct kiocb *iocb,
2411 loff_t offset, 2400 loff_t offset,
2412 ssize_t bytes, 2401 ssize_t bytes,
2413 void *private) 2402 void *private)
2414 { 2403 {
2415 struct inode *inode = file_inode(iocb 2404 struct inode *inode = file_inode(iocb->ki_filp);
2416 int level; 2405 int level;
2417 int ret = 0; 2406 int ret = 0;
2418 2407
2419 /* this io's submitter should not hav 2408 /* this io's submitter should not have unlocked this before we could */
2420 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2409 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2421 2410
2422 if (bytes <= 0) 2411 if (bytes <= 0)
2423 mlog_ratelimited(ML_ERROR, "D 2412 mlog_ratelimited(ML_ERROR, "Direct IO failed, bytes = %lld",
2424 (long long)b 2413 (long long)bytes);
2425 if (private) { 2414 if (private) {
2426 if (bytes > 0) 2415 if (bytes > 0)
2427 ret = ocfs2_dio_end_i 2416 ret = ocfs2_dio_end_io_write(inode, private, offset,
2428 2417 bytes);
2429 else 2418 else
2430 ocfs2_dio_free_write_ 2419 ocfs2_dio_free_write_ctx(inode, private);
2431 } 2420 }
2432 2421
2433 ocfs2_iocb_clear_rw_locked(iocb); 2422 ocfs2_iocb_clear_rw_locked(iocb);
2434 2423
2435 level = ocfs2_iocb_rw_locked_level(io 2424 level = ocfs2_iocb_rw_locked_level(iocb);
2436 ocfs2_rw_unlock(inode, level); 2425 ocfs2_rw_unlock(inode, level);
2437 return ret; 2426 return ret;
2438 } 2427 }
2439 2428
2440 static ssize_t ocfs2_direct_IO(struct kiocb * 2429 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2441 { 2430 {
2442 struct file *file = iocb->ki_filp; 2431 struct file *file = iocb->ki_filp;
2443 struct inode *inode = file->f_mapping 2432 struct inode *inode = file->f_mapping->host;
2444 struct ocfs2_super *osb = OCFS2_SB(in 2433 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2445 get_block_t *get_block; 2434 get_block_t *get_block;
2446 2435
2447 /* 2436 /*
2448 * Fallback to buffered I/O if we see 2437 * Fallback to buffered I/O if we see an inode without
2449 * extents. 2438 * extents.
2450 */ 2439 */
2451 if (OCFS2_I(inode)->ip_dyn_features & 2440 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2452 return 0; 2441 return 0;
2453 2442
2454 /* Fallback to buffered I/O if we do 2443 /* Fallback to buffered I/O if we do not support append dio. */
2455 if (iocb->ki_pos + iter->count > i_si 2444 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2456 !ocfs2_supports_append_dio(osb)) 2445 !ocfs2_supports_append_dio(osb))
2457 return 0; 2446 return 0;
2458 2447
2459 if (iov_iter_rw(iter) == READ) 2448 if (iov_iter_rw(iter) == READ)
2460 get_block = ocfs2_lock_get_bl 2449 get_block = ocfs2_lock_get_block;
2461 else 2450 else
2462 get_block = ocfs2_dio_wr_get_ 2451 get_block = ocfs2_dio_wr_get_block;
2463 2452
2464 return __blockdev_direct_IO(iocb, ino 2453 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2465 iter, get 2454 iter, get_block,
2466 ocfs2_dio !! 2455 ocfs2_dio_end_io, NULL, 0);
2467 } 2456 }
2468 2457
2469 const struct address_space_operations ocfs2_a 2458 const struct address_space_operations ocfs2_aops = {
2470 .dirty_folio = block_dirty !! 2459 .readpage = ocfs2_readpage,
2471 .read_folio = ocfs2_read_ <<
2472 .readahead = ocfs2_reada 2460 .readahead = ocfs2_readahead,
2473 .writepages = ocfs2_write !! 2461 .writepage = ocfs2_writepage,
2474 .write_begin = ocfs2_write 2462 .write_begin = ocfs2_write_begin,
2475 .write_end = ocfs2_write 2463 .write_end = ocfs2_write_end,
2476 .bmap = ocfs2_bmap, 2464 .bmap = ocfs2_bmap,
2477 .direct_IO = ocfs2_direc 2465 .direct_IO = ocfs2_direct_IO,
2478 .invalidate_folio = block_inval !! 2466 .invalidatepage = block_invalidatepage,
2479 .release_folio = ocfs2_relea !! 2467 .releasepage = ocfs2_releasepage,
2480 .migrate_folio = buffer_migr !! 2468 .migratepage = buffer_migrate_page,
2481 .is_partially_uptodate = block_is_pa 2469 .is_partially_uptodate = block_is_partially_uptodate,
2482 .error_remove_folio = generic_err !! 2470 .error_remove_page = generic_error_remove_page,
2483 }; 2471 };
2484 2472
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.