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