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, "Error %d from 159 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
160 "%llu, NULL)\n", err, ino 160 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
161 (unsigned long long)p_blk 161 (unsigned long long)p_blkno);
162 goto bail; 162 goto bail;
163 } 163 }
164 164
165 if (max_blocks < count) 165 if (max_blocks < count)
166 count = max_blocks; 166 count = max_blocks;
167 167
168 /* 168 /*
169 * ocfs2 never allocates in this funct 169 * ocfs2 never allocates in this function - the only time we
170 * 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
171 * system which doesn't support holes, 171 * system which doesn't support holes, in which case BH_New
172 * allows __block_write_begin() to zer 172 * allows __block_write_begin() to zero.
173 * 173 *
174 * If we see this on a sparse file sys 174 * If we see this on a sparse file system, then a truncate has
175 * raced us and removed the cluster. I 175 * raced us and removed the cluster. In this case, we clear
176 * the buffers dirty and uptodate bits 176 * the buffers dirty and uptodate bits and let the buffer code
177 * ignore it as a hole. 177 * ignore it as a hole.
178 */ 178 */
179 if (create && p_blkno == 0 && ocfs2_sp 179 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
180 clear_buffer_dirty(bh_result); 180 clear_buffer_dirty(bh_result);
181 clear_buffer_uptodate(bh_resul 181 clear_buffer_uptodate(bh_result);
182 goto bail; 182 goto bail;
183 } 183 }
184 184
185 /* Treat the unwritten extent as a hol 185 /* Treat the unwritten extent as a hole for zeroing purposes. */
186 if (p_blkno && !(ext_flags & OCFS2_EXT 186 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
187 map_bh(bh_result, inode->i_sb, 187 map_bh(bh_result, inode->i_sb, p_blkno);
188 188
189 bh_result->b_size = count << inode->i_ 189 bh_result->b_size = count << inode->i_blkbits;
190 190
191 if (!ocfs2_sparse_alloc(osb)) { 191 if (!ocfs2_sparse_alloc(osb)) {
192 if (p_blkno == 0) { 192 if (p_blkno == 0) {
193 err = -EIO; 193 err = -EIO;
194 mlog(ML_ERROR, 194 mlog(ML_ERROR,
195 "iblock = %llu p_ 195 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
196 (unsigned long lo 196 (unsigned long long)iblock,
197 (unsigned long lo 197 (unsigned long long)p_blkno,
198 (unsigned long lo 198 (unsigned long long)OCFS2_I(inode)->ip_blkno);
199 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);
200 dump_stack(); 200 dump_stack();
201 goto bail; 201 goto bail;
202 } 202 }
203 } 203 }
204 204
205 past_eof = ocfs2_blocks_for_bytes(inod 205 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
206 206
207 trace_ocfs2_get_block_end((unsigned lo 207 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
208 (unsigned lo 208 (unsigned long long)past_eof);
209 if (create && (iblock >= past_eof)) 209 if (create && (iblock >= past_eof))
210 set_buffer_new(bh_result); 210 set_buffer_new(bh_result);
211 211
212 bail: 212 bail:
213 if (err < 0) 213 if (err < 0)
214 err = -EIO; 214 err = -EIO;
215 215
216 return err; 216 return err;
217 } 217 }
218 218
219 int ocfs2_read_inline_data(struct inode *inode 219 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
220 struct buffer_head 220 struct buffer_head *di_bh)
221 { 221 {
222 void *kaddr; 222 void *kaddr;
223 loff_t size; 223 loff_t size;
224 struct ocfs2_dinode *di = (struct ocfs 224 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
225 225
226 if (!(le16_to_cpu(di->i_dyn_features) 226 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
227 ocfs2_error(inode->i_sb, "Inod 227 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
228 (unsigned long lon 228 (unsigned long long)OCFS2_I(inode)->ip_blkno);
229 return -EROFS; 229 return -EROFS;
230 } 230 }
231 231
232 size = i_size_read(inode); 232 size = i_size_read(inode);
233 233
234 if (size > PAGE_SIZE || 234 if (size > PAGE_SIZE ||
235 size > ocfs2_max_inline_data_with_ 235 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
236 ocfs2_error(inode->i_sb, 236 ocfs2_error(inode->i_sb,
237 "Inode %llu has wi 237 "Inode %llu has with inline data has bad size: %Lu\n",
238 (unsigned long lon 238 (unsigned long long)OCFS2_I(inode)->ip_blkno,
239 (unsigned long lon 239 (unsigned long long)size);
240 return -EROFS; 240 return -EROFS;
241 } 241 }
242 242
243 kaddr = kmap_atomic(page); 243 kaddr = kmap_atomic(page);
244 if (size) 244 if (size)
245 memcpy(kaddr, di->id2.i_data.i 245 memcpy(kaddr, di->id2.i_data.id_data, size);
246 /* Clear the remaining part of the pag 246 /* Clear the remaining part of the page */
247 memset(kaddr + size, 0, PAGE_SIZE - si 247 memset(kaddr + size, 0, PAGE_SIZE - size);
248 flush_dcache_page(page); 248 flush_dcache_page(page);
249 kunmap_atomic(kaddr); 249 kunmap_atomic(kaddr);
250 250
251 SetPageUptodate(page); 251 SetPageUptodate(page);
252 252
253 return 0; 253 return 0;
254 } 254 }
255 255
256 static int ocfs2_readpage_inline(struct inode 256 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
257 { 257 {
258 int ret; 258 int ret;
259 struct buffer_head *di_bh = NULL; 259 struct buffer_head *di_bh = NULL;
260 260
261 BUG_ON(!PageLocked(page)); 261 BUG_ON(!PageLocked(page));
262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
263 263
264 ret = ocfs2_read_inode_block(inode, &d 264 ret = ocfs2_read_inode_block(inode, &di_bh);
265 if (ret) { 265 if (ret) {
266 mlog_errno(ret); 266 mlog_errno(ret);
267 goto out; 267 goto out;
268 } 268 }
269 269
270 ret = ocfs2_read_inline_data(inode, pa 270 ret = ocfs2_read_inline_data(inode, page, di_bh);
271 out: 271 out:
272 unlock_page(page); 272 unlock_page(page);
273 273
274 brelse(di_bh); 274 brelse(di_bh);
275 return ret; 275 return ret;
276 } 276 }
277 277
278 static int ocfs2_read_folio(struct file *file, !! 278 static int ocfs2_readpage(struct file *file, struct page *page)
279 { 279 {
280 struct inode *inode = folio->mapping-> !! 280 struct inode *inode = page->mapping->host;
281 struct ocfs2_inode_info *oi = OCFS2_I( 281 struct ocfs2_inode_info *oi = OCFS2_I(inode);
282 loff_t start = folio_pos(folio); !! 282 loff_t start = (loff_t)page->index << PAGE_SHIFT;
283 int ret, unlock = 1; 283 int ret, unlock = 1;
284 284
285 trace_ocfs2_readpage((unsigned long lo !! 285 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
>> 286 (page ? page->index : 0));
286 287
287 ret = ocfs2_inode_lock_with_page(inode !! 288 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
288 if (ret != 0) { 289 if (ret != 0) {
289 if (ret == AOP_TRUNCATED_PAGE) 290 if (ret == AOP_TRUNCATED_PAGE)
290 unlock = 0; 291 unlock = 0;
291 mlog_errno(ret); 292 mlog_errno(ret);
292 goto out; 293 goto out;
293 } 294 }
294 295
295 if (down_read_trylock(&oi->ip_alloc_se 296 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
296 /* 297 /*
297 * Unlock the folio and cycle !! 298 * Unlock the page and cycle ip_alloc_sem so that we don't
298 * busyloop waiting for ip_all 299 * busyloop waiting for ip_alloc_sem to unlock
299 */ 300 */
300 ret = AOP_TRUNCATED_PAGE; 301 ret = AOP_TRUNCATED_PAGE;
301 folio_unlock(folio); !! 302 unlock_page(page);
302 unlock = 0; 303 unlock = 0;
303 down_read(&oi->ip_alloc_sem); 304 down_read(&oi->ip_alloc_sem);
304 up_read(&oi->ip_alloc_sem); 305 up_read(&oi->ip_alloc_sem);
305 goto out_inode_unlock; 306 goto out_inode_unlock;
306 } 307 }
307 308
308 /* 309 /*
309 * i_size might have just been updated 310 * i_size might have just been updated as we grabed the meta lock. We
310 * might now be discovering a truncate 311 * might now be discovering a truncate that hit on another node.
311 * block_read_full_folio->get_block fr !! 312 * block_read_full_page->get_block freaks out if it is asked to read
312 * beyond the end of a file, so we che 313 * beyond the end of a file, so we check here. Callers
313 * (generic_file_read, vm_ops->fault) 314 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
314 * and notice that the folio they just !! 315 * and notice that the page they just read isn't needed.
315 * 316 *
316 * XXX sys_readahead() seems to get th 317 * XXX sys_readahead() seems to get that wrong?
317 */ 318 */
318 if (start >= i_size_read(inode)) { 319 if (start >= i_size_read(inode)) {
319 folio_zero_segment(folio, 0, f !! 320 zero_user(page, 0, PAGE_SIZE);
320 folio_mark_uptodate(folio); !! 321 SetPageUptodate(page);
321 ret = 0; 322 ret = 0;
322 goto out_alloc; 323 goto out_alloc;
323 } 324 }
324 325
325 if (oi->ip_dyn_features & OCFS2_INLINE 326 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
326 ret = ocfs2_readpage_inline(in !! 327 ret = ocfs2_readpage_inline(inode, page);
327 else 328 else
328 ret = block_read_full_folio(fo !! 329 ret = block_read_full_page(page, ocfs2_get_block);
329 unlock = 0; 330 unlock = 0;
330 331
331 out_alloc: 332 out_alloc:
332 up_read(&oi->ip_alloc_sem); 333 up_read(&oi->ip_alloc_sem);
333 out_inode_unlock: 334 out_inode_unlock:
334 ocfs2_inode_unlock(inode, 0); 335 ocfs2_inode_unlock(inode, 0);
335 out: 336 out:
336 if (unlock) 337 if (unlock)
337 folio_unlock(folio); !! 338 unlock_page(page);
338 return ret; 339 return ret;
339 } 340 }
340 341
341 /* 342 /*
342 * This is used only for read-ahead. Failures 343 * This is used only for read-ahead. Failures or difficult to handle
343 * situations are safe to ignore. 344 * situations are safe to ignore.
344 * 345 *
345 * Right now, we don't bother with BH_Boundary 346 * Right now, we don't bother with BH_Boundary - in-inode extent lists
346 * are quite large (243 extents on 4k blocks), 347 * are quite large (243 extents on 4k blocks), so most inodes don't
347 * grow out to a tree. If need be, detecting b 348 * grow out to a tree. If need be, detecting boundary extents could
348 * trivially be added in a future version of o 349 * trivially be added in a future version of ocfs2_get_block().
349 */ 350 */
350 static void ocfs2_readahead(struct readahead_c 351 static void ocfs2_readahead(struct readahead_control *rac)
351 { 352 {
352 int ret; 353 int ret;
353 struct inode *inode = rac->mapping->ho 354 struct inode *inode = rac->mapping->host;
354 struct ocfs2_inode_info *oi = OCFS2_I( 355 struct ocfs2_inode_info *oi = OCFS2_I(inode);
355 356
356 /* 357 /*
357 * Use the nonblocking flag for the dl 358 * Use the nonblocking flag for the dlm code to avoid page
358 * lock inversion, but don't bother wi 359 * lock inversion, but don't bother with retrying.
359 */ 360 */
360 ret = ocfs2_inode_lock_full(inode, NUL 361 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
361 if (ret) 362 if (ret)
362 return; 363 return;
363 364
364 if (down_read_trylock(&oi->ip_alloc_se 365 if (down_read_trylock(&oi->ip_alloc_sem) == 0)
365 goto out_unlock; 366 goto out_unlock;
366 367
367 /* 368 /*
368 * Don't bother with inline-data. Ther 369 * Don't bother with inline-data. There isn't anything
369 * to read-ahead in that case anyway.. 370 * to read-ahead in that case anyway...
370 */ 371 */
371 if (oi->ip_dyn_features & OCFS2_INLINE 372 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
372 goto out_up; 373 goto out_up;
373 374
374 /* 375 /*
375 * Check whether a remote node truncat 376 * Check whether a remote node truncated this file - we just
376 * drop out in that case as it's not w 377 * drop out in that case as it's not worth handling here.
377 */ 378 */
378 if (readahead_pos(rac) >= i_size_read( 379 if (readahead_pos(rac) >= i_size_read(inode))
379 goto out_up; 380 goto out_up;
380 381
381 mpage_readahead(rac, ocfs2_get_block); 382 mpage_readahead(rac, ocfs2_get_block);
382 383
383 out_up: 384 out_up:
384 up_read(&oi->ip_alloc_sem); 385 up_read(&oi->ip_alloc_sem);
385 out_unlock: 386 out_unlock:
386 ocfs2_inode_unlock(inode, 0); 387 ocfs2_inode_unlock(inode, 0);
387 } 388 }
388 389
389 /* Note: Because we don't support holes, our a 390 /* Note: Because we don't support holes, our allocation has
390 * already happened (allocation writes zeros t 391 * already happened (allocation writes zeros to the file data)
391 * so we don't have to worry about ordered wri 392 * so we don't have to worry about ordered writes in
392 * ocfs2_writepages. !! 393 * ocfs2_writepage.
393 * 394 *
394 * ->writepages is called during the process o !! 395 * ->writepage is called during the process of invalidating the page cache
395 * during blocked lock processing. It can't b 396 * during blocked lock processing. It can't block on any cluster locks
396 * to during block mapping. It's relying on t 397 * to during block mapping. It's relying on the fact that the block
397 * mapping can't have disappeared under the di 398 * mapping can't have disappeared under the dirty pages that it is
398 * being asked to write back. 399 * being asked to write back.
399 */ 400 */
400 static int ocfs2_writepages(struct address_spa !! 401 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
401 struct writeback_control *wbc) <<
402 { 402 {
403 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);
404 } 408 }
405 409
406 /* Taken from ext3. We don't necessarily need 410 /* Taken from ext3. We don't necessarily need the full blown
407 * functionality yet, but IMHO it's better to 411 * functionality yet, but IMHO it's better to cut and paste the whole
408 * thing so we can avoid introducing our own b 412 * thing so we can avoid introducing our own bugs (and easily pick up
409 * their fixes when they happen) --Mark */ 413 * their fixes when they happen) --Mark */
410 int walk_page_buffers( handle_t *handle, 414 int walk_page_buffers( handle_t *handle,
411 struct buffer_head *he 415 struct buffer_head *head,
412 unsigned from, 416 unsigned from,
413 unsigned to, 417 unsigned to,
414 int *partial, 418 int *partial,
415 int (*fn)( handle 419 int (*fn)( handle_t *handle,
416 struct 420 struct buffer_head *bh))
417 { 421 {
418 struct buffer_head *bh; 422 struct buffer_head *bh;
419 unsigned block_start, block_end; 423 unsigned block_start, block_end;
420 unsigned blocksize = head->b_size; 424 unsigned blocksize = head->b_size;
421 int err, ret = 0; 425 int err, ret = 0;
422 struct buffer_head *next; 426 struct buffer_head *next;
423 427
424 for ( bh = head, block_start = 0; 428 for ( bh = head, block_start = 0;
425 ret == 0 && (bh != head || !bl 429 ret == 0 && (bh != head || !block_start);
426 block_start = block_end, bh = 430 block_start = block_end, bh = next)
427 { 431 {
428 next = bh->b_this_page; 432 next = bh->b_this_page;
429 block_end = block_start + bloc 433 block_end = block_start + blocksize;
430 if (block_end <= from || block 434 if (block_end <= from || block_start >= to) {
431 if (partial && !buffer 435 if (partial && !buffer_uptodate(bh))
432 *partial = 1; 436 *partial = 1;
433 continue; 437 continue;
434 } 438 }
435 err = (*fn)(handle, bh); 439 err = (*fn)(handle, bh);
436 if (!ret) 440 if (!ret)
437 ret = err; 441 ret = err;
438 } 442 }
439 return ret; 443 return ret;
440 } 444 }
441 445
442 static sector_t ocfs2_bmap(struct address_spac 446 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
443 { 447 {
444 sector_t status; 448 sector_t status;
445 u64 p_blkno = 0; 449 u64 p_blkno = 0;
446 int err = 0; 450 int err = 0;
447 struct inode *inode = mapping->host; 451 struct inode *inode = mapping->host;
448 452
449 trace_ocfs2_bmap((unsigned long long)O 453 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
450 (unsigned long long)b 454 (unsigned long long)block);
451 455
452 /* 456 /*
453 * The swap code (ab-)uses ->bmap to g 457 * The swap code (ab-)uses ->bmap to get a block mapping and then
454 * bypasseѕ the file system for actua 458 * bypasseѕ the file system for actual I/O. We really can't allow
455 * that on refcounted inodes, so we ha 459 * that on refcounted inodes, so we have to skip out here. And yes,
456 * 0 is the magic code for a bmap erro 460 * 0 is the magic code for a bmap error..
457 */ 461 */
458 if (ocfs2_is_refcount_inode(inode)) 462 if (ocfs2_is_refcount_inode(inode))
459 return 0; 463 return 0;
460 464
461 /* We don't need to lock journal syste 465 /* We don't need to lock journal system files, since they aren't
462 * accessed concurrently from multiple 466 * accessed concurrently from multiple nodes.
463 */ 467 */
464 if (!INODE_JOURNAL(inode)) { 468 if (!INODE_JOURNAL(inode)) {
465 err = ocfs2_inode_lock(inode, 469 err = ocfs2_inode_lock(inode, NULL, 0);
466 if (err) { 470 if (err) {
467 if (err != -ENOENT) 471 if (err != -ENOENT)
468 mlog_errno(err 472 mlog_errno(err);
469 goto bail; 473 goto bail;
470 } 474 }
471 down_read(&OCFS2_I(inode)->ip_ 475 down_read(&OCFS2_I(inode)->ip_alloc_sem);
472 } 476 }
473 477
474 if (!(OCFS2_I(inode)->ip_dyn_features 478 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
475 err = ocfs2_extent_map_get_blo 479 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
476 480 NULL);
477 481
478 if (!INODE_JOURNAL(inode)) { 482 if (!INODE_JOURNAL(inode)) {
479 up_read(&OCFS2_I(inode)->ip_al 483 up_read(&OCFS2_I(inode)->ip_alloc_sem);
480 ocfs2_inode_unlock(inode, 0); 484 ocfs2_inode_unlock(inode, 0);
481 } 485 }
482 486
483 if (err) { 487 if (err) {
484 mlog(ML_ERROR, "get_blocks() f 488 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
485 (unsigned long long)block 489 (unsigned long long)block);
486 mlog_errno(err); 490 mlog_errno(err);
487 goto bail; 491 goto bail;
488 } 492 }
489 493
490 bail: 494 bail:
491 status = err ? 0 : p_blkno; 495 status = err ? 0 : p_blkno;
492 496
493 return status; 497 return status;
494 } 498 }
495 499
496 static bool ocfs2_release_folio(struct folio * !! 500 static int ocfs2_releasepage(struct page *page, gfp_t wait)
497 { 501 {
498 if (!folio_buffers(folio)) !! 502 if (!page_has_buffers(page))
499 return false; !! 503 return 0;
500 return try_to_free_buffers(folio); !! 504 return try_to_free_buffers(page);
501 } 505 }
502 506
503 static void ocfs2_figure_cluster_boundaries(st 507 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
504 u3 508 u32 cpos,
505 un 509 unsigned int *start,
506 un 510 unsigned int *end)
507 { 511 {
508 unsigned int cluster_start = 0, cluste 512 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
509 513
510 if (unlikely(PAGE_SHIFT > osb->s_clust 514 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
511 unsigned int cpp; 515 unsigned int cpp;
512 516
513 cpp = 1 << (PAGE_SHIFT - osb-> 517 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
514 518
515 cluster_start = cpos % cpp; 519 cluster_start = cpos % cpp;
516 cluster_start = cluster_start 520 cluster_start = cluster_start << osb->s_clustersize_bits;
517 521
518 cluster_end = cluster_start + 522 cluster_end = cluster_start + osb->s_clustersize;
519 } 523 }
520 524
521 BUG_ON(cluster_start > PAGE_SIZE); 525 BUG_ON(cluster_start > PAGE_SIZE);
522 BUG_ON(cluster_end > PAGE_SIZE); 526 BUG_ON(cluster_end > PAGE_SIZE);
523 527
524 if (start) 528 if (start)
525 *start = cluster_start; 529 *start = cluster_start;
526 if (end) 530 if (end)
527 *end = cluster_end; 531 *end = cluster_end;
528 } 532 }
529 533
530 /* 534 /*
531 * 'from' and 'to' are the region in the page 535 * 'from' and 'to' are the region in the page to avoid zeroing.
532 * 536 *
533 * If pagesize > clustersize, this function wi 537 * If pagesize > clustersize, this function will avoid zeroing outside
534 * of the cluster boundary. 538 * of the cluster boundary.
535 * 539 *
536 * from == to == 0 is code for "zero the entir 540 * from == to == 0 is code for "zero the entire cluster region"
537 */ 541 */
538 static void ocfs2_clear_page_regions(struct pa 542 static void ocfs2_clear_page_regions(struct page *page,
539 struct oc 543 struct ocfs2_super *osb, u32 cpos,
540 unsigned 544 unsigned from, unsigned to)
541 { 545 {
542 void *kaddr; 546 void *kaddr;
543 unsigned int cluster_start, cluster_en 547 unsigned int cluster_start, cluster_end;
544 548
545 ocfs2_figure_cluster_boundaries(osb, c 549 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
546 550
547 kaddr = kmap_atomic(page); 551 kaddr = kmap_atomic(page);
548 552
549 if (from || to) { 553 if (from || to) {
550 if (from > cluster_start) 554 if (from > cluster_start)
551 memset(kaddr + cluster 555 memset(kaddr + cluster_start, 0, from - cluster_start);
552 if (to < cluster_end) 556 if (to < cluster_end)
553 memset(kaddr + to, 0, 557 memset(kaddr + to, 0, cluster_end - to);
554 } else { 558 } else {
555 memset(kaddr + cluster_start, 559 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
556 } 560 }
557 561
558 kunmap_atomic(kaddr); 562 kunmap_atomic(kaddr);
559 } 563 }
560 564
561 /* 565 /*
562 * Nonsparse file systems fully allocate befor 566 * Nonsparse file systems fully allocate before we get to the write
563 * code. This prevents ocfs2_write() from tagg 567 * code. This prevents ocfs2_write() from tagging the write as an
564 * allocating one, which means ocfs2_map_page_ 568 * allocating one, which means ocfs2_map_page_blocks() might try to
565 * 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
566 * testing i_size against each block offset. 570 * testing i_size against each block offset.
567 */ 571 */
568 static int ocfs2_should_read_blk(struct inode !! 572 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
569 unsigned int 573 unsigned int block_start)
570 { 574 {
571 u64 offset = folio_pos(folio) + block_ !! 575 u64 offset = page_offset(page) + block_start;
572 576
573 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 577 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
574 return 1; 578 return 1;
575 579
576 if (i_size_read(inode) > offset) 580 if (i_size_read(inode) > offset)
577 return 1; 581 return 1;
578 582
579 return 0; 583 return 0;
580 } 584 }
581 585
582 /* 586 /*
583 * Some of this taken from __block_write_begin 587 * Some of this taken from __block_write_begin(). We already have our
584 * mapping by now though, and the entire write 588 * mapping by now though, and the entire write will be allocating or
585 * it won't, so not much need to use BH_New. 589 * it won't, so not much need to use BH_New.
586 * 590 *
587 * This will also skip zeroing, which is handl 591 * This will also skip zeroing, which is handled externally.
588 */ 592 */
589 int ocfs2_map_page_blocks(struct page *page, u 593 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
590 struct inode *inode, 594 struct inode *inode, unsigned int from,
591 unsigned int to, int 595 unsigned int to, int new)
592 { 596 {
593 struct folio *folio = page_folio(page) <<
594 int ret = 0; 597 int ret = 0;
595 struct buffer_head *head, *bh, *wait[2 598 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
596 unsigned int block_end, block_start; 599 unsigned int block_end, block_start;
597 unsigned int bsize = i_blocksize(inode 600 unsigned int bsize = i_blocksize(inode);
598 601
599 head = folio_buffers(folio); !! 602 if (!page_has_buffers(page))
600 if (!head) !! 603 create_empty_buffers(page, bsize, 0);
601 head = create_empty_buffers(fo <<
602 604
>> 605 head = page_buffers(page);
603 for (bh = head, block_start = 0; bh != 606 for (bh = head, block_start = 0; bh != head || !block_start;
604 bh = bh->b_this_page, block_start 607 bh = bh->b_this_page, block_start += bsize) {
605 block_end = block_start + bsiz 608 block_end = block_start + bsize;
606 609
607 clear_buffer_new(bh); 610 clear_buffer_new(bh);
608 611
609 /* 612 /*
610 * Ignore blocks outside of ou 613 * Ignore blocks outside of our i/o range -
611 * they may belong to unalloca 614 * they may belong to unallocated clusters.
612 */ 615 */
613 if (block_start >= to || block 616 if (block_start >= to || block_end <= from) {
614 if (folio_test_uptodat !! 617 if (PageUptodate(page))
615 set_buffer_upt 618 set_buffer_uptodate(bh);
616 continue; 619 continue;
617 } 620 }
618 621
619 /* 622 /*
620 * For an allocating write wit 623 * For an allocating write with cluster size >= page
621 * size, we always write the e 624 * size, we always write the entire page.
622 */ 625 */
623 if (new) 626 if (new)
624 set_buffer_new(bh); 627 set_buffer_new(bh);
625 628
626 if (!buffer_mapped(bh)) { 629 if (!buffer_mapped(bh)) {
627 map_bh(bh, inode->i_sb 630 map_bh(bh, inode->i_sb, *p_blkno);
628 clean_bdev_bh_alias(bh 631 clean_bdev_bh_alias(bh);
629 } 632 }
630 633
631 if (folio_test_uptodate(folio) !! 634 if (PageUptodate(page)) {
632 set_buffer_uptodate(bh !! 635 if (!buffer_uptodate(bh))
>> 636 set_buffer_uptodate(bh);
633 } else if (!buffer_uptodate(bh 637 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
634 !buffer_new(bh) && 638 !buffer_new(bh) &&
635 ocfs2_should_read_b !! 639 ocfs2_should_read_blk(inode, page, block_start) &&
636 (block_start < from 640 (block_start < from || block_end > to)) {
637 bh_read_nowait(bh, 0); !! 641 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
638 *wait_bh++=bh; 642 *wait_bh++=bh;
639 } 643 }
640 644
641 *p_blkno = *p_blkno + 1; 645 *p_blkno = *p_blkno + 1;
642 } 646 }
643 647
644 /* 648 /*
645 * If we issued read requests - let th 649 * If we issued read requests - let them complete.
646 */ 650 */
647 while(wait_bh > wait) { 651 while(wait_bh > wait) {
648 wait_on_buffer(*--wait_bh); 652 wait_on_buffer(*--wait_bh);
649 if (!buffer_uptodate(*wait_bh) 653 if (!buffer_uptodate(*wait_bh))
650 ret = -EIO; 654 ret = -EIO;
651 } 655 }
652 656
653 if (ret == 0 || !new) 657 if (ret == 0 || !new)
654 return ret; 658 return ret;
655 659
656 /* 660 /*
657 * If we get -EIO above, zero out any 661 * If we get -EIO above, zero out any newly allocated blocks
658 * to avoid exposing stale data. 662 * to avoid exposing stale data.
659 */ 663 */
660 bh = head; 664 bh = head;
661 block_start = 0; 665 block_start = 0;
662 do { 666 do {
663 block_end = block_start + bsiz 667 block_end = block_start + bsize;
664 if (block_end <= from) 668 if (block_end <= from)
665 goto next_bh; 669 goto next_bh;
666 if (block_start >= to) 670 if (block_start >= to)
667 break; 671 break;
668 672
669 folio_zero_range(folio, block_ !! 673 zero_user(page, block_start, bh->b_size);
670 set_buffer_uptodate(bh); 674 set_buffer_uptodate(bh);
671 mark_buffer_dirty(bh); 675 mark_buffer_dirty(bh);
672 676
673 next_bh: 677 next_bh:
674 block_start = block_end; 678 block_start = block_end;
675 bh = bh->b_this_page; 679 bh = bh->b_this_page;
676 } while (bh != head); 680 } while (bh != head);
677 681
678 return ret; 682 return ret;
679 } 683 }
680 684
681 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 685 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
682 #define OCFS2_MAX_CTXT_PAGES 1 686 #define OCFS2_MAX_CTXT_PAGES 1
683 #else 687 #else
684 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 688 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
685 #endif 689 #endif
686 690
687 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 691 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
688 692
689 struct ocfs2_unwritten_extent { 693 struct ocfs2_unwritten_extent {
690 struct list_head ue_node; 694 struct list_head ue_node;
691 struct list_head ue_ip_node; 695 struct list_head ue_ip_node;
692 u32 ue_cpos; 696 u32 ue_cpos;
693 u32 ue_phys; 697 u32 ue_phys;
694 }; 698 };
695 699
696 /* 700 /*
697 * Describe the state of a single cluster to b 701 * Describe the state of a single cluster to be written to.
698 */ 702 */
699 struct ocfs2_write_cluster_desc { 703 struct ocfs2_write_cluster_desc {
700 u32 c_cpos; 704 u32 c_cpos;
701 u32 c_phys; 705 u32 c_phys;
702 /* 706 /*
703 * Give this a unique field because c_ 707 * Give this a unique field because c_phys eventually gets
704 * filled. 708 * filled.
705 */ 709 */
706 unsigned c_new; 710 unsigned c_new;
707 unsigned c_clear_unwritten; 711 unsigned c_clear_unwritten;
708 unsigned c_needs_zero; 712 unsigned c_needs_zero;
709 }; 713 };
710 714
711 struct ocfs2_write_ctxt { 715 struct ocfs2_write_ctxt {
712 /* Logical cluster position / len of w 716 /* Logical cluster position / len of write */
713 u32 w_cpos 717 u32 w_cpos;
714 u32 w_clen 718 u32 w_clen;
715 719
716 /* First cluster allocated in a nonspa 720 /* First cluster allocated in a nonsparse extend */
717 u32 w_firs 721 u32 w_first_new_cpos;
718 722
719 /* Type of caller. Must be one of buff 723 /* Type of caller. Must be one of buffer, mmap, direct. */
720 ocfs2_write_type_t w_type 724 ocfs2_write_type_t w_type;
721 725
722 struct ocfs2_write_cluster_desc w_desc 726 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
723 727
724 /* 728 /*
725 * This is true if page_size > cluster 729 * This is true if page_size > cluster_size.
726 * 730 *
727 * It triggers a set of special cases 731 * It triggers a set of special cases during write which might
728 * have to deal with allocating writes 732 * have to deal with allocating writes to partial pages.
729 */ 733 */
730 unsigned int w_larg 734 unsigned int w_large_pages;
731 735
732 /* 736 /*
733 * Pages involved in this write. 737 * Pages involved in this write.
734 * 738 *
735 * w_target_page is the page being wri 739 * w_target_page is the page being written to by the user.
736 * 740 *
737 * w_pages is an array of pages which 741 * w_pages is an array of pages which always contains
738 * w_target_page, and in the case of a 742 * w_target_page, and in the case of an allocating write with
739 * page_size < cluster size, it will c 743 * page_size < cluster size, it will contain zero'd and mapped
740 * pages adjacent to w_target_page whi 744 * pages adjacent to w_target_page which need to be written
741 * out in so that future reads from th 745 * out in so that future reads from that region will get
742 * zero's. 746 * zero's.
743 */ 747 */
744 unsigned int w_num_ 748 unsigned int w_num_pages;
745 struct page *w_pag 749 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
746 struct page *w_tar 750 struct page *w_target_page;
747 751
748 /* 752 /*
749 * w_target_locked is used for page_mk 753 * w_target_locked is used for page_mkwrite path indicating no unlocking
750 * against w_target_page in ocfs2_writ 754 * against w_target_page in ocfs2_write_end_nolock.
751 */ 755 */
752 unsigned int w_targ 756 unsigned int w_target_locked:1;
753 757
754 /* 758 /*
755 * ocfs2_write_end() uses this to know 759 * ocfs2_write_end() uses this to know what the real range to
756 * write in the target should be. 760 * write in the target should be.
757 */ 761 */
758 unsigned int w_targ 762 unsigned int w_target_from;
759 unsigned int w_targ 763 unsigned int w_target_to;
760 764
761 /* 765 /*
762 * We could use journal_current_handle 766 * We could use journal_current_handle() but this is cleaner,
763 * IMHO -Mark 767 * IMHO -Mark
764 */ 768 */
765 handle_t *w_han 769 handle_t *w_handle;
766 770
767 struct buffer_head *w_di_ 771 struct buffer_head *w_di_bh;
768 772
769 struct ocfs2_cached_dealloc_ctxt w_dea 773 struct ocfs2_cached_dealloc_ctxt w_dealloc;
770 774
771 struct list_head w_unwr 775 struct list_head w_unwritten_list;
772 unsigned int w_unwr 776 unsigned int w_unwritten_count;
773 }; 777 };
774 778
775 void ocfs2_unlock_and_free_pages(struct page * 779 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
776 { 780 {
777 int i; 781 int i;
778 782
779 for(i = 0; i < num_pages; i++) { 783 for(i = 0; i < num_pages; i++) {
780 if (pages[i]) { 784 if (pages[i]) {
781 unlock_page(pages[i]); 785 unlock_page(pages[i]);
782 mark_page_accessed(pag 786 mark_page_accessed(pages[i]);
783 put_page(pages[i]); 787 put_page(pages[i]);
784 } 788 }
785 } 789 }
786 } 790 }
787 791
788 static void ocfs2_unlock_pages(struct ocfs2_wr 792 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
789 { 793 {
790 int i; 794 int i;
791 795
792 /* 796 /*
793 * w_target_locked is only set to true 797 * w_target_locked is only set to true in the page_mkwrite() case.
794 * The intent is to allow us to lock t 798 * The intent is to allow us to lock the target page from write_begin()
795 * to write_end(). The caller must hol 799 * to write_end(). The caller must hold a ref on w_target_page.
796 */ 800 */
797 if (wc->w_target_locked) { 801 if (wc->w_target_locked) {
798 BUG_ON(!wc->w_target_page); 802 BUG_ON(!wc->w_target_page);
799 for (i = 0; i < wc->w_num_page 803 for (i = 0; i < wc->w_num_pages; i++) {
800 if (wc->w_target_page 804 if (wc->w_target_page == wc->w_pages[i]) {
801 wc->w_pages[i] 805 wc->w_pages[i] = NULL;
802 break; 806 break;
803 } 807 }
804 } 808 }
805 mark_page_accessed(wc->w_targe 809 mark_page_accessed(wc->w_target_page);
806 put_page(wc->w_target_page); 810 put_page(wc->w_target_page);
807 } 811 }
808 ocfs2_unlock_and_free_pages(wc->w_page 812 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
809 } 813 }
810 814
811 static void ocfs2_free_unwritten_list(struct i 815 static void ocfs2_free_unwritten_list(struct inode *inode,
812 struct list_h 816 struct list_head *head)
813 { 817 {
814 struct ocfs2_inode_info *oi = OCFS2_I( 818 struct ocfs2_inode_info *oi = OCFS2_I(inode);
815 struct ocfs2_unwritten_extent *ue = NU 819 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
816 820
817 list_for_each_entry_safe(ue, tmp, head 821 list_for_each_entry_safe(ue, tmp, head, ue_node) {
818 list_del(&ue->ue_node); 822 list_del(&ue->ue_node);
819 spin_lock(&oi->ip_lock); 823 spin_lock(&oi->ip_lock);
820 list_del(&ue->ue_ip_node); 824 list_del(&ue->ue_ip_node);
821 spin_unlock(&oi->ip_lock); 825 spin_unlock(&oi->ip_lock);
822 kfree(ue); 826 kfree(ue);
823 } 827 }
824 } 828 }
825 829
826 static void ocfs2_free_write_ctxt(struct inode 830 static void ocfs2_free_write_ctxt(struct inode *inode,
827 struct ocfs2 831 struct ocfs2_write_ctxt *wc)
828 { 832 {
829 ocfs2_free_unwritten_list(inode, &wc-> 833 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
830 ocfs2_unlock_pages(wc); 834 ocfs2_unlock_pages(wc);
831 brelse(wc->w_di_bh); 835 brelse(wc->w_di_bh);
832 kfree(wc); 836 kfree(wc);
833 } 837 }
834 838
835 static int ocfs2_alloc_write_ctxt(struct ocfs2 839 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
836 struct ocfs2 840 struct ocfs2_super *osb, loff_t pos,
837 unsigned len 841 unsigned len, ocfs2_write_type_t type,
838 struct buffe 842 struct buffer_head *di_bh)
839 { 843 {
840 u32 cend; 844 u32 cend;
841 struct ocfs2_write_ctxt *wc; 845 struct ocfs2_write_ctxt *wc;
842 846
843 wc = kzalloc(sizeof(struct ocfs2_write 847 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
844 if (!wc) 848 if (!wc)
845 return -ENOMEM; 849 return -ENOMEM;
846 850
847 wc->w_cpos = pos >> osb->s_clustersize 851 wc->w_cpos = pos >> osb->s_clustersize_bits;
848 wc->w_first_new_cpos = UINT_MAX; 852 wc->w_first_new_cpos = UINT_MAX;
849 cend = (pos + len - 1) >> osb->s_clust 853 cend = (pos + len - 1) >> osb->s_clustersize_bits;
850 wc->w_clen = cend - wc->w_cpos + 1; 854 wc->w_clen = cend - wc->w_cpos + 1;
851 get_bh(di_bh); 855 get_bh(di_bh);
852 wc->w_di_bh = di_bh; 856 wc->w_di_bh = di_bh;
853 wc->w_type = type; 857 wc->w_type = type;
854 858
855 if (unlikely(PAGE_SHIFT > osb->s_clust 859 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
856 wc->w_large_pages = 1; 860 wc->w_large_pages = 1;
857 else 861 else
858 wc->w_large_pages = 0; 862 wc->w_large_pages = 0;
859 863
860 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 864 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
861 INIT_LIST_HEAD(&wc->w_unwritten_list); 865 INIT_LIST_HEAD(&wc->w_unwritten_list);
862 866
863 *wcp = wc; 867 *wcp = wc;
864 868
865 return 0; 869 return 0;
866 } 870 }
867 871
868 /* 872 /*
869 * If a page has any new buffers, zero them ou 873 * If a page has any new buffers, zero them out here, and mark them uptodate
870 * and dirty so they'll be written out (in ord 874 * and dirty so they'll be written out (in order to prevent uninitialised
871 * block data from leaking). And clear the new 875 * block data from leaking). And clear the new bit.
872 */ 876 */
873 static void ocfs2_zero_new_buffers(struct page 877 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
874 { 878 {
875 unsigned int block_start, block_end; 879 unsigned int block_start, block_end;
876 struct buffer_head *head, *bh; 880 struct buffer_head *head, *bh;
877 881
878 BUG_ON(!PageLocked(page)); 882 BUG_ON(!PageLocked(page));
879 if (!page_has_buffers(page)) 883 if (!page_has_buffers(page))
880 return; 884 return;
881 885
882 bh = head = page_buffers(page); 886 bh = head = page_buffers(page);
883 block_start = 0; 887 block_start = 0;
884 do { 888 do {
885 block_end = block_start + bh-> 889 block_end = block_start + bh->b_size;
886 890
887 if (buffer_new(bh)) { 891 if (buffer_new(bh)) {
888 if (block_end > from & 892 if (block_end > from && block_start < to) {
889 if (!PageUptod 893 if (!PageUptodate(page)) {
890 unsign 894 unsigned start, end;
891 895
892 start 896 start = max(from, block_start);
893 end = 897 end = min(to, block_end);
894 898
895 zero_u 899 zero_user_segment(page, start, end);
896 set_bu 900 set_buffer_uptodate(bh);
897 } 901 }
898 902
899 clear_buffer_n 903 clear_buffer_new(bh);
900 mark_buffer_di 904 mark_buffer_dirty(bh);
901 } 905 }
902 } 906 }
903 907
904 block_start = block_end; 908 block_start = block_end;
905 bh = bh->b_this_page; 909 bh = bh->b_this_page;
906 } while (bh != head); 910 } while (bh != head);
907 } 911 }
908 912
909 /* 913 /*
910 * Only called when we have a failure during a 914 * Only called when we have a failure during allocating write to write
911 * zero's to the newly allocated region. 915 * zero's to the newly allocated region.
912 */ 916 */
913 static void ocfs2_write_failure(struct inode * 917 static void ocfs2_write_failure(struct inode *inode,
914 struct ocfs2_w 918 struct ocfs2_write_ctxt *wc,
915 loff_t user_po 919 loff_t user_pos, unsigned user_len)
916 { 920 {
917 int i; 921 int i;
918 unsigned from = user_pos & (PAGE_SIZE 922 unsigned from = user_pos & (PAGE_SIZE - 1),
919 to = user_pos + user_len; 923 to = user_pos + user_len;
920 struct page *tmppage; 924 struct page *tmppage;
921 925
922 if (wc->w_target_page) 926 if (wc->w_target_page)
923 ocfs2_zero_new_buffers(wc->w_t 927 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
924 928
925 for(i = 0; i < wc->w_num_pages; i++) { 929 for(i = 0; i < wc->w_num_pages; i++) {
926 tmppage = wc->w_pages[i]; 930 tmppage = wc->w_pages[i];
927 931
928 if (tmppage && page_has_buffer 932 if (tmppage && page_has_buffers(tmppage)) {
929 if (ocfs2_should_order 933 if (ocfs2_should_order_data(inode))
930 ocfs2_jbd2_ino 934 ocfs2_jbd2_inode_add_write(wc->w_handle, inode,
931 935 user_pos, user_len);
932 936
933 block_commit_write(tmp 937 block_commit_write(tmppage, from, to);
934 } 938 }
935 } 939 }
936 } 940 }
937 941
938 static int ocfs2_prepare_page_for_write(struct 942 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
939 struct 943 struct ocfs2_write_ctxt *wc,
940 struct 944 struct page *page, u32 cpos,
941 loff_t 945 loff_t user_pos, unsigned user_len,
942 int ne 946 int new)
943 { 947 {
944 int ret; 948 int ret;
945 unsigned int map_from = 0, map_to = 0; 949 unsigned int map_from = 0, map_to = 0;
946 unsigned int cluster_start, cluster_en 950 unsigned int cluster_start, cluster_end;
947 unsigned int user_data_from = 0, user_ 951 unsigned int user_data_from = 0, user_data_to = 0;
948 952
949 ocfs2_figure_cluster_boundaries(OCFS2_ 953 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
950 &clust 954 &cluster_start, &cluster_end);
951 955
952 /* treat the write as new if the a hol 956 /* treat the write as new if the a hole/lseek spanned across
953 * the page boundary. 957 * the page boundary.
954 */ 958 */
955 new = new | ((i_size_read(inode) <= pa 959 new = new | ((i_size_read(inode) <= page_offset(page)) &&
956 (page_offset(page) <= 960 (page_offset(page) <= user_pos));
957 961
958 if (page == wc->w_target_page) { 962 if (page == wc->w_target_page) {
959 map_from = user_pos & (PAGE_SI 963 map_from = user_pos & (PAGE_SIZE - 1);
960 map_to = map_from + user_len; 964 map_to = map_from + user_len;
961 965
962 if (new) 966 if (new)
963 ret = ocfs2_map_page_b 967 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
964 968 cluster_start, cluster_end,
965 969 new);
966 else 970 else
967 ret = ocfs2_map_page_b 971 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
968 972 map_from, map_to, new);
969 if (ret) { 973 if (ret) {
970 mlog_errno(ret); 974 mlog_errno(ret);
971 goto out; 975 goto out;
972 } 976 }
973 977
974 user_data_from = map_from; 978 user_data_from = map_from;
975 user_data_to = map_to; 979 user_data_to = map_to;
976 if (new) { 980 if (new) {
977 map_from = cluster_sta 981 map_from = cluster_start;
978 map_to = cluster_end; 982 map_to = cluster_end;
979 } 983 }
980 } else { 984 } else {
981 /* 985 /*
982 * If we haven't allocated the 986 * If we haven't allocated the new page yet, we
983 * shouldn't be writing it out 987 * shouldn't be writing it out without copying user
984 * data. This is likely a math 988 * data. This is likely a math error from the caller.
985 */ 989 */
986 BUG_ON(!new); 990 BUG_ON(!new);
987 991
988 map_from = cluster_start; 992 map_from = cluster_start;
989 map_to = cluster_end; 993 map_to = cluster_end;
990 994
991 ret = ocfs2_map_page_blocks(pa 995 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
992 cl 996 cluster_start, cluster_end, new);
993 if (ret) { 997 if (ret) {
994 mlog_errno(ret); 998 mlog_errno(ret);
995 goto out; 999 goto out;
996 } 1000 }
997 } 1001 }
998 1002
999 /* 1003 /*
1000 * Parts of newly allocated pages nee 1004 * Parts of newly allocated pages need to be zero'd.
1001 * 1005 *
1002 * Above, we have also rewritten 'to' 1006 * Above, we have also rewritten 'to' and 'from' - as far as
1003 * the rest of the function is concer 1007 * the rest of the function is concerned, the entire cluster
1004 * range inside of a page needs to be 1008 * range inside of a page needs to be written.
1005 * 1009 *
1006 * We can skip this if the page is up 1010 * We can skip this if the page is up to date - it's already
1007 * been zero'd from being read in as 1011 * been zero'd from being read in as a hole.
1008 */ 1012 */
1009 if (new && !PageUptodate(page)) 1013 if (new && !PageUptodate(page))
1010 ocfs2_clear_page_regions(page 1014 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1011 cpos 1015 cpos, user_data_from, user_data_to);
1012 1016
1013 flush_dcache_page(page); 1017 flush_dcache_page(page);
1014 1018
1015 out: 1019 out:
1016 return ret; 1020 return ret;
1017 } 1021 }
1018 1022
1019 /* 1023 /*
1020 * This function will only grab one clusters 1024 * This function will only grab one clusters worth of pages.
1021 */ 1025 */
1022 static int ocfs2_grab_pages_for_write(struct 1026 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1023 struct 1027 struct ocfs2_write_ctxt *wc,
1024 u32 cpo 1028 u32 cpos, loff_t user_pos,
1025 unsigne 1029 unsigned user_len, int new,
1026 struct 1030 struct page *mmap_page)
1027 { 1031 {
1028 int ret = 0, i; 1032 int ret = 0, i;
1029 unsigned long start, target_index, en 1033 unsigned long start, target_index, end_index, index;
1030 struct inode *inode = mapping->host; 1034 struct inode *inode = mapping->host;
1031 loff_t last_byte; 1035 loff_t last_byte;
1032 1036
1033 target_index = user_pos >> PAGE_SHIFT 1037 target_index = user_pos >> PAGE_SHIFT;
1034 1038
1035 /* 1039 /*
1036 * Figure out how many pages we'll be 1040 * Figure out how many pages we'll be manipulating here. For
1037 * non allocating write, we just chan 1041 * non allocating write, we just change the one
1038 * page. Otherwise, we'll need a whol 1042 * page. Otherwise, we'll need a whole clusters worth. If we're
1039 * writing past i_size, we only need 1043 * writing past i_size, we only need enough pages to cover the
1040 * last page of the write. 1044 * last page of the write.
1041 */ 1045 */
1042 if (new) { 1046 if (new) {
1043 wc->w_num_pages = ocfs2_pages 1047 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1044 start = ocfs2_align_clusters_ 1048 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1045 /* 1049 /*
1046 * We need the index *past* t 1050 * We need the index *past* the last page we could possibly
1047 * touch. This is the page p 1051 * touch. This is the page past the end of the write or
1048 * i_size, whichever is great 1052 * i_size, whichever is greater.
1049 */ 1053 */
1050 last_byte = max(user_pos + us 1054 last_byte = max(user_pos + user_len, i_size_read(inode));
1051 BUG_ON(last_byte < 1); 1055 BUG_ON(last_byte < 1);
1052 end_index = ((last_byte - 1) 1056 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
1053 if ((start + wc->w_num_pages) 1057 if ((start + wc->w_num_pages) > end_index)
1054 wc->w_num_pages = end 1058 wc->w_num_pages = end_index - start;
1055 } else { 1059 } else {
1056 wc->w_num_pages = 1; 1060 wc->w_num_pages = 1;
1057 start = target_index; 1061 start = target_index;
1058 } 1062 }
1059 end_index = (user_pos + user_len - 1) 1063 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
1060 1064
1061 for(i = 0; i < wc->w_num_pages; i++) 1065 for(i = 0; i < wc->w_num_pages; i++) {
1062 index = start + i; 1066 index = start + i;
1063 1067
1064 if (index >= target_index && 1068 if (index >= target_index && index <= end_index &&
1065 wc->w_type == OCFS2_WRITE 1069 wc->w_type == OCFS2_WRITE_MMAP) {
1066 /* 1070 /*
1067 * ocfs2_pagemkwrite( 1071 * ocfs2_pagemkwrite() is a little different
1068 * and wants us to di 1072 * and wants us to directly use the page
1069 * passed in. 1073 * passed in.
1070 */ 1074 */
1071 lock_page(mmap_page); 1075 lock_page(mmap_page);
1072 1076
1073 /* Exit and let the c 1077 /* Exit and let the caller retry */
1074 if (mmap_page->mappin 1078 if (mmap_page->mapping != mapping) {
1075 WARN_ON(mmap_ 1079 WARN_ON(mmap_page->mapping);
1076 unlock_page(m 1080 unlock_page(mmap_page);
1077 ret = -EAGAIN 1081 ret = -EAGAIN;
1078 goto out; 1082 goto out;
1079 } 1083 }
1080 1084
1081 get_page(mmap_page); 1085 get_page(mmap_page);
1082 wc->w_pages[i] = mmap 1086 wc->w_pages[i] = mmap_page;
1083 wc->w_target_locked = 1087 wc->w_target_locked = true;
1084 } else if (index >= target_in 1088 } else if (index >= target_index && index <= end_index &&
1085 wc->w_type == OCFS 1089 wc->w_type == OCFS2_WRITE_DIRECT) {
1086 /* Direct write has n 1090 /* Direct write has no mapping page. */
1087 wc->w_pages[i] = NULL 1091 wc->w_pages[i] = NULL;
1088 continue; 1092 continue;
1089 } else { 1093 } else {
1090 wc->w_pages[i] = find 1094 wc->w_pages[i] = find_or_create_page(mapping, index,
1091 1095 GFP_NOFS);
1092 if (!wc->w_pages[i]) 1096 if (!wc->w_pages[i]) {
1093 ret = -ENOMEM 1097 ret = -ENOMEM;
1094 mlog_errno(re 1098 mlog_errno(ret);
1095 goto out; 1099 goto out;
1096 } 1100 }
1097 } 1101 }
1098 wait_for_stable_page(wc->w_pa 1102 wait_for_stable_page(wc->w_pages[i]);
1099 1103
1100 if (index == target_index) 1104 if (index == target_index)
1101 wc->w_target_page = w 1105 wc->w_target_page = wc->w_pages[i];
1102 } 1106 }
1103 out: 1107 out:
1104 if (ret) 1108 if (ret)
1105 wc->w_target_locked = false; 1109 wc->w_target_locked = false;
1106 return ret; 1110 return ret;
1107 } 1111 }
1108 1112
1109 /* 1113 /*
1110 * Prepare a single cluster for write one clu 1114 * Prepare a single cluster for write one cluster into the file.
1111 */ 1115 */
1112 static int ocfs2_write_cluster(struct address 1116 static int ocfs2_write_cluster(struct address_space *mapping,
1113 u32 *phys, uns 1117 u32 *phys, unsigned int new,
1114 unsigned int c 1118 unsigned int clear_unwritten,
1115 unsigned int s 1119 unsigned int should_zero,
1116 struct ocfs2_a 1120 struct ocfs2_alloc_context *data_ac,
1117 struct ocfs2_a 1121 struct ocfs2_alloc_context *meta_ac,
1118 struct ocfs2_w 1122 struct ocfs2_write_ctxt *wc, u32 cpos,
1119 loff_t user_po 1123 loff_t user_pos, unsigned user_len)
1120 { 1124 {
1121 int ret, i; 1125 int ret, i;
1122 u64 p_blkno; 1126 u64 p_blkno;
1123 struct inode *inode = mapping->host; 1127 struct inode *inode = mapping->host;
1124 struct ocfs2_extent_tree et; 1128 struct ocfs2_extent_tree et;
1125 int bpc = ocfs2_clusters_to_blocks(in 1129 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1126 1130
1127 if (new) { 1131 if (new) {
1128 u32 tmp_pos; 1132 u32 tmp_pos;
1129 1133
1130 /* 1134 /*
1131 * This is safe to call with 1135 * This is safe to call with the page locks - it won't take
1132 * any additional semaphores 1136 * any additional semaphores or cluster locks.
1133 */ 1137 */
1134 tmp_pos = cpos; 1138 tmp_pos = cpos;
1135 ret = ocfs2_add_inode_data(OC 1139 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1136 &t 1140 &tmp_pos, 1, !clear_unwritten,
1137 wc 1141 wc->w_di_bh, wc->w_handle,
1138 da 1142 data_ac, meta_ac, NULL);
1139 /* 1143 /*
1140 * This shouldn't happen beca 1144 * This shouldn't happen because we must have already
1141 * calculated the correct met 1145 * calculated the correct meta data allocation required. The
1142 * internal tree allocation c 1146 * internal tree allocation code should know how to increase
1143 * transaction credits itself 1147 * transaction credits itself.
1144 * 1148 *
1145 * If need be, we could handl 1149 * If need be, we could handle -EAGAIN for a
1146 * RESTART_TRANS here. 1150 * RESTART_TRANS here.
1147 */ 1151 */
1148 mlog_bug_on_msg(ret == -EAGAI 1152 mlog_bug_on_msg(ret == -EAGAIN,
1149 "Inode %llu: 1153 "Inode %llu: EAGAIN return during allocation.\n",
1150 (unsigned lon 1154 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1151 if (ret < 0) { 1155 if (ret < 0) {
1152 mlog_errno(ret); 1156 mlog_errno(ret);
1153 goto out; 1157 goto out;
1154 } 1158 }
1155 } else if (clear_unwritten) { 1159 } else if (clear_unwritten) {
1156 ocfs2_init_dinode_extent_tree 1160 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1157 1161 wc->w_di_bh);
1158 ret = ocfs2_mark_extent_writt 1162 ret = ocfs2_mark_extent_written(inode, &et,
1159 1163 wc->w_handle, cpos, 1, *phys,
1160 1164 meta_ac, &wc->w_dealloc);
1161 if (ret < 0) { 1165 if (ret < 0) {
1162 mlog_errno(ret); 1166 mlog_errno(ret);
1163 goto out; 1167 goto out;
1164 } 1168 }
1165 } 1169 }
1166 1170
1167 /* 1171 /*
1168 * The only reason this should fail i 1172 * The only reason this should fail is due to an inability to
1169 * find the extent added. 1173 * find the extent added.
1170 */ 1174 */
1171 ret = ocfs2_get_clusters(inode, cpos, 1175 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
1172 if (ret < 0) { 1176 if (ret < 0) {
1173 mlog(ML_ERROR, "Get physical 1177 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1174 "at logical clust 1178 "at logical cluster %u",
1175 (unsigned long lo 1179 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
1176 goto out; 1180 goto out;
1177 } 1181 }
1178 1182
1179 BUG_ON(*phys == 0); 1183 BUG_ON(*phys == 0);
1180 1184
1181 p_blkno = ocfs2_clusters_to_blocks(in 1185 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1182 if (!should_zero) 1186 if (!should_zero)
1183 p_blkno += (user_pos >> inode 1187 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
1184 1188
1185 for(i = 0; i < wc->w_num_pages; i++) 1189 for(i = 0; i < wc->w_num_pages; i++) {
1186 int tmpret; 1190 int tmpret;
1187 1191
1188 /* This is the direct io targ 1192 /* This is the direct io target page. */
1189 if (wc->w_pages[i] == NULL) { 1193 if (wc->w_pages[i] == NULL) {
1190 p_blkno++; 1194 p_blkno++;
1191 continue; 1195 continue;
1192 } 1196 }
1193 1197
1194 tmpret = ocfs2_prepare_page_f 1198 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1195 1199 wc->w_pages[i], cpos,
1196 1200 user_pos, user_len,
1197 1201 should_zero);
1198 if (tmpret) { 1202 if (tmpret) {
1199 mlog_errno(tmpret); 1203 mlog_errno(tmpret);
1200 if (ret == 0) 1204 if (ret == 0)
1201 ret = tmpret; 1205 ret = tmpret;
1202 } 1206 }
1203 } 1207 }
1204 1208
1205 /* 1209 /*
1206 * We only have cleanup to do in case 1210 * We only have cleanup to do in case of allocating write.
1207 */ 1211 */
1208 if (ret && new) 1212 if (ret && new)
1209 ocfs2_write_failure(inode, wc 1213 ocfs2_write_failure(inode, wc, user_pos, user_len);
1210 1214
1211 out: 1215 out:
1212 1216
1213 return ret; 1217 return ret;
1214 } 1218 }
1215 1219
1216 static int ocfs2_write_cluster_by_desc(struct 1220 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1217 struct 1221 struct ocfs2_alloc_context *data_ac,
1218 struct 1222 struct ocfs2_alloc_context *meta_ac,
1219 struct 1223 struct ocfs2_write_ctxt *wc,
1220 loff_t 1224 loff_t pos, unsigned len)
1221 { 1225 {
1222 int ret, i; 1226 int ret, i;
1223 loff_t cluster_off; 1227 loff_t cluster_off;
1224 unsigned int local_len = len; 1228 unsigned int local_len = len;
1225 struct ocfs2_write_cluster_desc *desc 1229 struct ocfs2_write_cluster_desc *desc;
1226 struct ocfs2_super *osb = OCFS2_SB(ma 1230 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1227 1231
1228 for (i = 0; i < wc->w_clen; i++) { 1232 for (i = 0; i < wc->w_clen; i++) {
1229 desc = &wc->w_desc[i]; 1233 desc = &wc->w_desc[i];
1230 1234
1231 /* 1235 /*
1232 * We have to make sure that 1236 * We have to make sure that the total write passed in
1233 * doesn't extend past a sing 1237 * doesn't extend past a single cluster.
1234 */ 1238 */
1235 local_len = len; 1239 local_len = len;
1236 cluster_off = pos & (osb->s_c 1240 cluster_off = pos & (osb->s_clustersize - 1);
1237 if ((cluster_off + local_len) 1241 if ((cluster_off + local_len) > osb->s_clustersize)
1238 local_len = osb->s_cl 1242 local_len = osb->s_clustersize - cluster_off;
1239 1243
1240 ret = ocfs2_write_cluster(map 1244 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
1241 des 1245 desc->c_new,
1242 des 1246 desc->c_clear_unwritten,
1243 des 1247 desc->c_needs_zero,
1244 dat 1248 data_ac, meta_ac,
1245 wc, 1249 wc, desc->c_cpos, pos, local_len);
1246 if (ret) { 1250 if (ret) {
1247 mlog_errno(ret); 1251 mlog_errno(ret);
1248 goto out; 1252 goto out;
1249 } 1253 }
1250 1254
1251 len -= local_len; 1255 len -= local_len;
1252 pos += local_len; 1256 pos += local_len;
1253 } 1257 }
1254 1258
1255 ret = 0; 1259 ret = 0;
1256 out: 1260 out:
1257 return ret; 1261 return ret;
1258 } 1262 }
1259 1263
1260 /* 1264 /*
1261 * ocfs2_write_end() wants to know which part 1265 * ocfs2_write_end() wants to know which parts of the target page it
1262 * should complete the write on. It's easiest 1266 * should complete the write on. It's easiest to compute them ahead of
1263 * time when a more complete view of the writ 1267 * time when a more complete view of the write is available.
1264 */ 1268 */
1265 static void ocfs2_set_target_boundaries(struc 1269 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1266 struc 1270 struct ocfs2_write_ctxt *wc,
1267 loff_ 1271 loff_t pos, unsigned len, int alloc)
1268 { 1272 {
1269 struct ocfs2_write_cluster_desc *desc 1273 struct ocfs2_write_cluster_desc *desc;
1270 1274
1271 wc->w_target_from = pos & (PAGE_SIZE 1275 wc->w_target_from = pos & (PAGE_SIZE - 1);
1272 wc->w_target_to = wc->w_target_from + 1276 wc->w_target_to = wc->w_target_from + len;
1273 1277
1274 if (alloc == 0) 1278 if (alloc == 0)
1275 return; 1279 return;
1276 1280
1277 /* 1281 /*
1278 * Allocating write - we may have dif 1282 * Allocating write - we may have different boundaries based
1279 * on page size and cluster size. 1283 * on page size and cluster size.
1280 * 1284 *
1281 * NOTE: We can no longer compute one 1285 * NOTE: We can no longer compute one value from the other as
1282 * the actual write length and user p 1286 * the actual write length and user provided length may be
1283 * different. 1287 * different.
1284 */ 1288 */
1285 1289
1286 if (wc->w_large_pages) { 1290 if (wc->w_large_pages) {
1287 /* 1291 /*
1288 * We only care about the 1st 1292 * We only care about the 1st and last cluster within
1289 * our range and whether they 1293 * our range and whether they should be zero'd or not. Either
1290 * value may be extended out 1294 * value may be extended out to the start/end of a
1291 * newly allocated cluster. 1295 * newly allocated cluster.
1292 */ 1296 */
1293 desc = &wc->w_desc[0]; 1297 desc = &wc->w_desc[0];
1294 if (desc->c_needs_zero) 1298 if (desc->c_needs_zero)
1295 ocfs2_figure_cluster_ 1299 ocfs2_figure_cluster_boundaries(osb,
1296 1300 desc->c_cpos,
1297 1301 &wc->w_target_from,
1298 1302 NULL);
1299 1303
1300 desc = &wc->w_desc[wc->w_clen 1304 desc = &wc->w_desc[wc->w_clen - 1];
1301 if (desc->c_needs_zero) 1305 if (desc->c_needs_zero)
1302 ocfs2_figure_cluster_ 1306 ocfs2_figure_cluster_boundaries(osb,
1303 1307 desc->c_cpos,
1304 1308 NULL,
1305 1309 &wc->w_target_to);
1306 } else { 1310 } else {
1307 wc->w_target_from = 0; 1311 wc->w_target_from = 0;
1308 wc->w_target_to = PAGE_SIZE; 1312 wc->w_target_to = PAGE_SIZE;
1309 } 1313 }
1310 } 1314 }
1311 1315
1312 /* 1316 /*
1313 * Check if this extent is marked UNWRITTEN b 1317 * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to
1314 * do the zero work. And should not to clear 1318 * do the zero work. And should not to clear UNWRITTEN since it will be cleared
1315 * by the direct io procedure. 1319 * by the direct io procedure.
1316 * If this is a new extent that allocated by 1320 * If this is a new extent that allocated by direct io, we should mark it in
1317 * the ip_unwritten_list. 1321 * the ip_unwritten_list.
1318 */ 1322 */
1319 static int ocfs2_unwritten_check(struct inode 1323 static int ocfs2_unwritten_check(struct inode *inode,
1320 struct ocfs2 1324 struct ocfs2_write_ctxt *wc,
1321 struct ocfs2 1325 struct ocfs2_write_cluster_desc *desc)
1322 { 1326 {
1323 struct ocfs2_inode_info *oi = OCFS2_I 1327 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1324 struct ocfs2_unwritten_extent *ue = N 1328 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
1325 int ret = 0; 1329 int ret = 0;
1326 1330
1327 if (!desc->c_needs_zero) 1331 if (!desc->c_needs_zero)
1328 return 0; 1332 return 0;
1329 1333
1330 retry: 1334 retry:
1331 spin_lock(&oi->ip_lock); 1335 spin_lock(&oi->ip_lock);
1332 /* Needs not to zero no metter buffer 1336 /* Needs not to zero no metter buffer or direct. The one who is zero
1333 * the cluster is doing zero. And he 1337 * the cluster is doing zero. And he will clear unwritten after all
1334 * cluster io finished. */ 1338 * cluster io finished. */
1335 list_for_each_entry(ue, &oi->ip_unwri 1339 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1336 if (desc->c_cpos == ue->ue_cp 1340 if (desc->c_cpos == ue->ue_cpos) {
1337 BUG_ON(desc->c_new); 1341 BUG_ON(desc->c_new);
1338 desc->c_needs_zero = 1342 desc->c_needs_zero = 0;
1339 desc->c_clear_unwritt 1343 desc->c_clear_unwritten = 0;
1340 goto unlock; 1344 goto unlock;
1341 } 1345 }
1342 } 1346 }
1343 1347
1344 if (wc->w_type != OCFS2_WRITE_DIRECT) 1348 if (wc->w_type != OCFS2_WRITE_DIRECT)
1345 goto unlock; 1349 goto unlock;
1346 1350
1347 if (new == NULL) { 1351 if (new == NULL) {
1348 spin_unlock(&oi->ip_lock); 1352 spin_unlock(&oi->ip_lock);
1349 new = kmalloc(sizeof(struct o 1353 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1350 GFP_NOFS); 1354 GFP_NOFS);
1351 if (new == NULL) { 1355 if (new == NULL) {
1352 ret = -ENOMEM; 1356 ret = -ENOMEM;
1353 goto out; 1357 goto out;
1354 } 1358 }
1355 goto retry; 1359 goto retry;
1356 } 1360 }
1357 /* This direct write will doing zero. 1361 /* This direct write will doing zero. */
1358 new->ue_cpos = desc->c_cpos; 1362 new->ue_cpos = desc->c_cpos;
1359 new->ue_phys = desc->c_phys; 1363 new->ue_phys = desc->c_phys;
1360 desc->c_clear_unwritten = 0; 1364 desc->c_clear_unwritten = 0;
1361 list_add_tail(&new->ue_ip_node, &oi-> 1365 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1362 list_add_tail(&new->ue_node, &wc->w_u 1366 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
1363 wc->w_unwritten_count++; 1367 wc->w_unwritten_count++;
1364 new = NULL; 1368 new = NULL;
1365 unlock: 1369 unlock:
1366 spin_unlock(&oi->ip_lock); 1370 spin_unlock(&oi->ip_lock);
1367 out: 1371 out:
1368 kfree(new); 1372 kfree(new);
1369 return ret; 1373 return ret;
1370 } 1374 }
1371 1375
1372 /* 1376 /*
1373 * Populate each single-cluster write descrip 1377 * Populate each single-cluster write descriptor in the write context
1374 * with information about the i/o to be done. 1378 * with information about the i/o to be done.
1375 * 1379 *
1376 * Returns the number of clusters that will h 1380 * Returns the number of clusters that will have to be allocated, as
1377 * well as a worst case estimate of the numbe 1381 * well as a worst case estimate of the number of extent records that
1378 * would have to be created during a write to 1382 * would have to be created during a write to an unwritten region.
1379 */ 1383 */
1380 static int ocfs2_populate_write_desc(struct i 1384 static int ocfs2_populate_write_desc(struct inode *inode,
1381 struct o 1385 struct ocfs2_write_ctxt *wc,
1382 unsigned 1386 unsigned int *clusters_to_alloc,
1383 unsigned 1387 unsigned int *extents_to_split)
1384 { 1388 {
1385 int ret; 1389 int ret;
1386 struct ocfs2_write_cluster_desc *desc 1390 struct ocfs2_write_cluster_desc *desc;
1387 unsigned int num_clusters = 0; 1391 unsigned int num_clusters = 0;
1388 unsigned int ext_flags = 0; 1392 unsigned int ext_flags = 0;
1389 u32 phys = 0; 1393 u32 phys = 0;
1390 int i; 1394 int i;
1391 1395
1392 *clusters_to_alloc = 0; 1396 *clusters_to_alloc = 0;
1393 *extents_to_split = 0; 1397 *extents_to_split = 0;
1394 1398
1395 for (i = 0; i < wc->w_clen; i++) { 1399 for (i = 0; i < wc->w_clen; i++) {
1396 desc = &wc->w_desc[i]; 1400 desc = &wc->w_desc[i];
1397 desc->c_cpos = wc->w_cpos + i 1401 desc->c_cpos = wc->w_cpos + i;
1398 1402
1399 if (num_clusters == 0) { 1403 if (num_clusters == 0) {
1400 /* 1404 /*
1401 * Need to look up th 1405 * Need to look up the next extent record.
1402 */ 1406 */
1403 ret = ocfs2_get_clust 1407 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1404 1408 &num_clusters, &ext_flags);
1405 if (ret) { 1409 if (ret) {
1406 mlog_errno(re 1410 mlog_errno(ret);
1407 goto out; 1411 goto out;
1408 } 1412 }
1409 1413
1410 /* We should already 1414 /* We should already CoW the refcountd extent. */
1411 BUG_ON(ext_flags & OC 1415 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1412 1416
1413 /* 1417 /*
1414 * Assume worst case 1418 * Assume worst case - that we're writing in
1415 * the middle of the 1419 * the middle of the extent.
1416 * 1420 *
1417 * We can assume that 1421 * We can assume that the write proceeds from
1418 * left to right, in 1422 * left to right, in which case the extent
1419 * insert code is sma 1423 * insert code is smart enough to coalesce the
1420 * next splits into t 1424 * next splits into the previous records created.
1421 */ 1425 */
1422 if (ext_flags & OCFS2 1426 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1423 *extents_to_s 1427 *extents_to_split = *extents_to_split + 2;
1424 } else if (phys) { 1428 } else if (phys) {
1425 /* 1429 /*
1426 * Only increment phy 1430 * Only increment phys if it doesn't describe
1427 * a hole. 1431 * a hole.
1428 */ 1432 */
1429 phys++; 1433 phys++;
1430 } 1434 }
1431 1435
1432 /* 1436 /*
1433 * If w_first_new_cpos is < U 1437 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1434 * file that got extended. w 1438 * file that got extended. w_first_new_cpos tells us
1435 * where the newly allocated 1439 * where the newly allocated clusters are so we can
1436 * zero them. 1440 * zero them.
1437 */ 1441 */
1438 if (desc->c_cpos >= wc->w_fir 1442 if (desc->c_cpos >= wc->w_first_new_cpos) {
1439 BUG_ON(phys == 0); 1443 BUG_ON(phys == 0);
1440 desc->c_needs_zero = 1444 desc->c_needs_zero = 1;
1441 } 1445 }
1442 1446
1443 desc->c_phys = phys; 1447 desc->c_phys = phys;
1444 if (phys == 0) { 1448 if (phys == 0) {
1445 desc->c_new = 1; 1449 desc->c_new = 1;
1446 desc->c_needs_zero = 1450 desc->c_needs_zero = 1;
1447 desc->c_clear_unwritt 1451 desc->c_clear_unwritten = 1;
1448 *clusters_to_alloc = 1452 *clusters_to_alloc = *clusters_to_alloc + 1;
1449 } 1453 }
1450 1454
1451 if (ext_flags & OCFS2_EXT_UNW 1455 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1452 desc->c_clear_unwritt 1456 desc->c_clear_unwritten = 1;
1453 desc->c_needs_zero = 1457 desc->c_needs_zero = 1;
1454 } 1458 }
1455 1459
1456 ret = ocfs2_unwritten_check(i 1460 ret = ocfs2_unwritten_check(inode, wc, desc);
1457 if (ret) { 1461 if (ret) {
1458 mlog_errno(ret); 1462 mlog_errno(ret);
1459 goto out; 1463 goto out;
1460 } 1464 }
1461 1465
1462 num_clusters--; 1466 num_clusters--;
1463 } 1467 }
1464 1468
1465 ret = 0; 1469 ret = 0;
1466 out: 1470 out:
1467 return ret; 1471 return ret;
1468 } 1472 }
1469 1473
1470 static int ocfs2_write_begin_inline(struct ad 1474 static int ocfs2_write_begin_inline(struct address_space *mapping,
1471 struct in 1475 struct inode *inode,
1472 struct oc 1476 struct ocfs2_write_ctxt *wc)
1473 { 1477 {
1474 int ret; 1478 int ret;
1475 struct ocfs2_super *osb = OCFS2_SB(in 1479 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1476 struct page *page; 1480 struct page *page;
1477 handle_t *handle; 1481 handle_t *handle;
1478 struct ocfs2_dinode *di = (struct ocf 1482 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1479 1483
1480 handle = ocfs2_start_trans(osb, OCFS2 1484 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1481 if (IS_ERR(handle)) { 1485 if (IS_ERR(handle)) {
1482 ret = PTR_ERR(handle); 1486 ret = PTR_ERR(handle);
1483 mlog_errno(ret); 1487 mlog_errno(ret);
1484 goto out; 1488 goto out;
1485 } 1489 }
1486 1490
1487 page = find_or_create_page(mapping, 0 1491 page = find_or_create_page(mapping, 0, GFP_NOFS);
1488 if (!page) { 1492 if (!page) {
1489 ocfs2_commit_trans(osb, handl 1493 ocfs2_commit_trans(osb, handle);
1490 ret = -ENOMEM; 1494 ret = -ENOMEM;
1491 mlog_errno(ret); 1495 mlog_errno(ret);
1492 goto out; 1496 goto out;
1493 } 1497 }
1494 /* 1498 /*
1495 * If we don't set w_num_pages then t 1499 * If we don't set w_num_pages then this page won't get unlocked
1496 * and freed on cleanup of the write 1500 * and freed on cleanup of the write context.
1497 */ 1501 */
1498 wc->w_pages[0] = wc->w_target_page = 1502 wc->w_pages[0] = wc->w_target_page = page;
1499 wc->w_num_pages = 1; 1503 wc->w_num_pages = 1;
1500 1504
1501 ret = ocfs2_journal_access_di(handle, 1505 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1502 OCFS2_J 1506 OCFS2_JOURNAL_ACCESS_WRITE);
1503 if (ret) { 1507 if (ret) {
1504 ocfs2_commit_trans(osb, handl 1508 ocfs2_commit_trans(osb, handle);
1505 1509
1506 mlog_errno(ret); 1510 mlog_errno(ret);
1507 goto out; 1511 goto out;
1508 } 1512 }
1509 1513
1510 if (!(OCFS2_I(inode)->ip_dyn_features 1514 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1511 ocfs2_set_inode_data_inline(i 1515 ocfs2_set_inode_data_inline(inode, di);
1512 1516
1513 if (!PageUptodate(page)) { 1517 if (!PageUptodate(page)) {
1514 ret = ocfs2_read_inline_data( 1518 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1515 if (ret) { 1519 if (ret) {
1516 ocfs2_commit_trans(os 1520 ocfs2_commit_trans(osb, handle);
1517 1521
1518 goto out; 1522 goto out;
1519 } 1523 }
1520 } 1524 }
1521 1525
1522 wc->w_handle = handle; 1526 wc->w_handle = handle;
1523 out: 1527 out:
1524 return ret; 1528 return ret;
1525 } 1529 }
1526 1530
1527 int ocfs2_size_fits_inline_data(struct buffer 1531 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1528 { 1532 {
1529 struct ocfs2_dinode *di = (struct ocf 1533 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1530 1534
1531 if (new_size <= le16_to_cpu(di->id2.i 1535 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1532 return 1; 1536 return 1;
1533 return 0; 1537 return 0;
1534 } 1538 }
1535 1539
1536 static int ocfs2_try_to_write_inline_data(str 1540 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1537 str 1541 struct inode *inode, loff_t pos,
1538 uns 1542 unsigned len, struct page *mmap_page,
1539 str 1543 struct ocfs2_write_ctxt *wc)
1540 { 1544 {
1541 int ret, written = 0; 1545 int ret, written = 0;
1542 loff_t end = pos + len; 1546 loff_t end = pos + len;
1543 struct ocfs2_inode_info *oi = OCFS2_I 1547 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1544 struct ocfs2_dinode *di = NULL; 1548 struct ocfs2_dinode *di = NULL;
1545 1549
1546 trace_ocfs2_try_to_write_inline_data( 1550 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1547 1551 len, (unsigned long long)pos,
1548 1552 oi->ip_dyn_features);
1549 1553
1550 /* 1554 /*
1551 * Handle inodes which already have i 1555 * Handle inodes which already have inline data 1st.
1552 */ 1556 */
1553 if (oi->ip_dyn_features & OCFS2_INLIN 1557 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1554 if (mmap_page == NULL && 1558 if (mmap_page == NULL &&
1555 ocfs2_size_fits_inline_da 1559 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1556 goto do_inline_write; 1560 goto do_inline_write;
1557 1561
1558 /* 1562 /*
1559 * The write won't fit - we h 1563 * The write won't fit - we have to give this inode an
1560 * inline extent list now. 1564 * inline extent list now.
1561 */ 1565 */
1562 ret = ocfs2_convert_inline_da 1566 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1563 if (ret) 1567 if (ret)
1564 mlog_errno(ret); 1568 mlog_errno(ret);
1565 goto out; 1569 goto out;
1566 } 1570 }
1567 1571
1568 /* 1572 /*
1569 * Check whether the inode can accept 1573 * Check whether the inode can accept inline data.
1570 */ 1574 */
1571 if (oi->ip_clusters != 0 || i_size_re 1575 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1572 return 0; 1576 return 0;
1573 1577
1574 /* 1578 /*
1575 * Check whether the write can fit. 1579 * Check whether the write can fit.
1576 */ 1580 */
1577 di = (struct ocfs2_dinode *)wc->w_di_ 1581 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1578 if (mmap_page || 1582 if (mmap_page ||
1579 end > ocfs2_max_inline_data_with_ 1583 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1580 return 0; 1584 return 0;
1581 1585
1582 do_inline_write: 1586 do_inline_write:
1583 ret = ocfs2_write_begin_inline(mappin 1587 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1584 if (ret) { 1588 if (ret) {
1585 mlog_errno(ret); 1589 mlog_errno(ret);
1586 goto out; 1590 goto out;
1587 } 1591 }
1588 1592
1589 /* 1593 /*
1590 * This signals to the caller that th 1594 * This signals to the caller that the data can be written
1591 * inline. 1595 * inline.
1592 */ 1596 */
1593 written = 1; 1597 written = 1;
1594 out: 1598 out:
1595 return written ? written : ret; 1599 return written ? written : ret;
1596 } 1600 }
1597 1601
1598 /* 1602 /*
1599 * This function only does anything for file 1603 * This function only does anything for file systems which can't
1600 * handle sparse files. 1604 * handle sparse files.
1601 * 1605 *
1602 * What we want to do here is fill in any hol 1606 * What we want to do here is fill in any hole between the current end
1603 * of allocation and the end of our write. Th 1607 * of allocation and the end of our write. That way the rest of the
1604 * write path can treat it as an non-allocati 1608 * write path can treat it as an non-allocating write, which has no
1605 * special case code for sparse/nonsparse fil 1609 * special case code for sparse/nonsparse files.
1606 */ 1610 */
1607 static int ocfs2_expand_nonsparse_inode(struc 1611 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1608 struc 1612 struct buffer_head *di_bh,
1609 loff_ 1613 loff_t pos, unsigned len,
1610 struc 1614 struct ocfs2_write_ctxt *wc)
1611 { 1615 {
1612 int ret; 1616 int ret;
1613 loff_t newsize = pos + len; 1617 loff_t newsize = pos + len;
1614 1618
1615 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1619 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1616 1620
1617 if (newsize <= i_size_read(inode)) 1621 if (newsize <= i_size_read(inode))
1618 return 0; 1622 return 0;
1619 1623
1620 ret = ocfs2_extend_no_holes(inode, di 1624 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1621 if (ret) 1625 if (ret)
1622 mlog_errno(ret); 1626 mlog_errno(ret);
1623 1627
1624 /* There is no wc if this is call fro 1628 /* There is no wc if this is call from direct. */
1625 if (wc) 1629 if (wc)
1626 wc->w_first_new_cpos = 1630 wc->w_first_new_cpos =
1627 ocfs2_clusters_for_by 1631 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1628 1632
1629 return ret; 1633 return ret;
1630 } 1634 }
1631 1635
1632 static int ocfs2_zero_tail(struct inode *inod 1636 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1633 loff_t pos) 1637 loff_t pos)
1634 { 1638 {
1635 int ret = 0; 1639 int ret = 0;
1636 1640
1637 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1641 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1638 if (pos > i_size_read(inode)) 1642 if (pos > i_size_read(inode))
1639 ret = ocfs2_zero_extend(inode 1643 ret = ocfs2_zero_extend(inode, di_bh, pos);
1640 1644
1641 return ret; 1645 return ret;
1642 } 1646 }
1643 1647
1644 int ocfs2_write_begin_nolock(struct address_s 1648 int ocfs2_write_begin_nolock(struct address_space *mapping,
1645 loff_t pos, unsi 1649 loff_t pos, unsigned len, ocfs2_write_type_t type,
1646 struct page **pa 1650 struct page **pagep, void **fsdata,
1647 struct buffer_he 1651 struct buffer_head *di_bh, struct page *mmap_page)
1648 { 1652 {
1649 int ret, cluster_of_pages, credits = 1653 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1650 unsigned int clusters_to_alloc, exten 1654 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1651 struct ocfs2_write_ctxt *wc; 1655 struct ocfs2_write_ctxt *wc;
1652 struct inode *inode = mapping->host; 1656 struct inode *inode = mapping->host;
1653 struct ocfs2_super *osb = OCFS2_SB(in 1657 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1654 struct ocfs2_dinode *di; 1658 struct ocfs2_dinode *di;
1655 struct ocfs2_alloc_context *data_ac = 1659 struct ocfs2_alloc_context *data_ac = NULL;
1656 struct ocfs2_alloc_context *meta_ac = 1660 struct ocfs2_alloc_context *meta_ac = NULL;
1657 handle_t *handle; 1661 handle_t *handle;
1658 struct ocfs2_extent_tree et; 1662 struct ocfs2_extent_tree et;
1659 int try_free = 1, ret1; 1663 int try_free = 1, ret1;
1660 1664
1661 try_again: 1665 try_again:
1662 ret = ocfs2_alloc_write_ctxt(&wc, osb 1666 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
1663 if (ret) { 1667 if (ret) {
1664 mlog_errno(ret); 1668 mlog_errno(ret);
1665 return ret; 1669 return ret;
1666 } 1670 }
1667 1671
1668 if (ocfs2_supports_inline_data(osb)) 1672 if (ocfs2_supports_inline_data(osb)) {
1669 ret = ocfs2_try_to_write_inli 1673 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1670 1674 mmap_page, wc);
1671 if (ret == 1) { 1675 if (ret == 1) {
1672 ret = 0; 1676 ret = 0;
1673 goto success; 1677 goto success;
1674 } 1678 }
1675 if (ret < 0) { 1679 if (ret < 0) {
1676 mlog_errno(ret); 1680 mlog_errno(ret);
1677 goto out; 1681 goto out;
1678 } 1682 }
1679 } 1683 }
1680 1684
1681 /* Direct io change i_size late, shou 1685 /* Direct io change i_size late, should not zero tail here. */
1682 if (type != OCFS2_WRITE_DIRECT) { 1686 if (type != OCFS2_WRITE_DIRECT) {
1683 if (ocfs2_sparse_alloc(osb)) 1687 if (ocfs2_sparse_alloc(osb))
1684 ret = ocfs2_zero_tail 1688 ret = ocfs2_zero_tail(inode, di_bh, pos);
1685 else 1689 else
1686 ret = ocfs2_expand_no 1690 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1687 1691 len, wc);
1688 if (ret) { 1692 if (ret) {
1689 mlog_errno(ret); 1693 mlog_errno(ret);
1690 goto out; 1694 goto out;
1691 } 1695 }
1692 } 1696 }
1693 1697
1694 ret = ocfs2_check_range_for_refcount( 1698 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1695 if (ret < 0) { 1699 if (ret < 0) {
1696 mlog_errno(ret); 1700 mlog_errno(ret);
1697 goto out; 1701 goto out;
1698 } else if (ret == 1) { 1702 } else if (ret == 1) {
1699 clusters_need = wc->w_clen; 1703 clusters_need = wc->w_clen;
1700 ret = ocfs2_refcount_cow(inod 1704 ret = ocfs2_refcount_cow(inode, di_bh,
1701 wc-> 1705 wc->w_cpos, wc->w_clen, UINT_MAX);
1702 if (ret) { 1706 if (ret) {
1703 mlog_errno(ret); 1707 mlog_errno(ret);
1704 goto out; 1708 goto out;
1705 } 1709 }
1706 } 1710 }
1707 1711
1708 ret = ocfs2_populate_write_desc(inode 1712 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1709 &exte 1713 &extents_to_split);
1710 if (ret) { 1714 if (ret) {
1711 mlog_errno(ret); 1715 mlog_errno(ret);
1712 goto out; 1716 goto out;
1713 } 1717 }
1714 clusters_need += clusters_to_alloc; 1718 clusters_need += clusters_to_alloc;
1715 1719
1716 di = (struct ocfs2_dinode *)wc->w_di_ 1720 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1717 1721
1718 trace_ocfs2_write_begin_nolock( 1722 trace_ocfs2_write_begin_nolock(
1719 (unsigned long long)O 1723 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1720 (long long)i_size_rea 1724 (long long)i_size_read(inode),
1721 le32_to_cpu(di->i_clu 1725 le32_to_cpu(di->i_clusters),
1722 pos, len, type, mmap_ 1726 pos, len, type, mmap_page,
1723 clusters_to_alloc, ex 1727 clusters_to_alloc, extents_to_split);
1724 1728
1725 /* 1729 /*
1726 * We set w_target_from, w_target_to 1730 * We set w_target_from, w_target_to here so that
1727 * ocfs2_write_end() knows which rang 1731 * ocfs2_write_end() knows which range in the target page to
1728 * write out. An allocation requires 1732 * write out. An allocation requires that we write the entire
1729 * cluster range. 1733 * cluster range.
1730 */ 1734 */
1731 if (clusters_to_alloc || extents_to_s 1735 if (clusters_to_alloc || extents_to_split) {
1732 /* 1736 /*
1733 * XXX: We are stretching the 1737 * XXX: We are stretching the limits of
1734 * ocfs2_lock_allocators(). I 1738 * ocfs2_lock_allocators(). It greatly over-estimates
1735 * the work to be done. 1739 * the work to be done.
1736 */ 1740 */
1737 ocfs2_init_dinode_extent_tree 1741 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1738 1742 wc->w_di_bh);
1739 ret = ocfs2_lock_allocators(i 1743 ret = ocfs2_lock_allocators(inode, &et,
1740 c 1744 clusters_to_alloc, extents_to_split,
1741 & 1745 &data_ac, &meta_ac);
1742 if (ret) { 1746 if (ret) {
1743 mlog_errno(ret); 1747 mlog_errno(ret);
1744 goto out; 1748 goto out;
1745 } 1749 }
1746 1750
1747 if (data_ac) 1751 if (data_ac)
1748 data_ac->ac_resv = &O 1752 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1749 1753
1750 credits = ocfs2_calc_extend_c 1754 credits = ocfs2_calc_extend_credits(inode->i_sb,
1751 1755 &di->id2.i_list);
1752 } else if (type == OCFS2_WRITE_DIRECT 1756 } else if (type == OCFS2_WRITE_DIRECT)
1753 /* direct write needs not to 1757 /* direct write needs not to start trans if no extents alloc. */
1754 goto success; 1758 goto success;
1755 1759
1756 /* 1760 /*
1757 * We have to zero sparse allocated c 1761 * We have to zero sparse allocated clusters, unwritten extent clusters,
1758 * and non-sparse clusters we just ex 1762 * and non-sparse clusters we just extended. For non-sparse writes,
1759 * we know zeros will only be needed 1763 * we know zeros will only be needed in the first and/or last cluster.
1760 */ 1764 */
1761 if (wc->w_clen && (wc->w_desc[0].c_ne 1765 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1762 wc->w_desc[wc->w_c 1766 wc->w_desc[wc->w_clen - 1].c_needs_zero))
1763 cluster_of_pages = 1; 1767 cluster_of_pages = 1;
1764 else 1768 else
1765 cluster_of_pages = 0; 1769 cluster_of_pages = 0;
1766 1770
1767 ocfs2_set_target_boundaries(osb, wc, 1771 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1768 1772
1769 handle = ocfs2_start_trans(osb, credi 1773 handle = ocfs2_start_trans(osb, credits);
1770 if (IS_ERR(handle)) { 1774 if (IS_ERR(handle)) {
1771 ret = PTR_ERR(handle); 1775 ret = PTR_ERR(handle);
1772 mlog_errno(ret); 1776 mlog_errno(ret);
1773 goto out; 1777 goto out;
1774 } 1778 }
1775 1779
1776 wc->w_handle = handle; 1780 wc->w_handle = handle;
1777 1781
1778 if (clusters_to_alloc) { 1782 if (clusters_to_alloc) {
1779 ret = dquot_alloc_space_nodir 1783 ret = dquot_alloc_space_nodirty(inode,
1780 ocfs2_clusters_to_byt 1784 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1781 if (ret) 1785 if (ret)
1782 goto out_commit; 1786 goto out_commit;
1783 } 1787 }
1784 1788
1785 ret = ocfs2_journal_access_di(handle, 1789 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1786 OCFS2_J 1790 OCFS2_JOURNAL_ACCESS_WRITE);
1787 if (ret) { 1791 if (ret) {
1788 mlog_errno(ret); 1792 mlog_errno(ret);
1789 goto out_quota; 1793 goto out_quota;
1790 } 1794 }
1791 1795
1792 /* 1796 /*
1793 * Fill our page array first. That wa 1797 * Fill our page array first. That way we've grabbed enough so
1794 * that we can zero and flush if we e 1798 * that we can zero and flush if we error after adding the
1795 * extent. 1799 * extent.
1796 */ 1800 */
1797 ret = ocfs2_grab_pages_for_write(mapp 1801 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1798 clus 1802 cluster_of_pages, mmap_page);
1799 if (ret) { !! 1803 if (ret && ret != -EAGAIN) {
1800 /* <<
1801 * ocfs2_grab_pages_for_write <<
1802 * the target page. In this c <<
1803 * page. This will trigger th <<
1804 * the operation. <<
1805 */ <<
1806 if (type == OCFS2_WRITE_MMAP <<
1807 BUG_ON(wc->w_target_p <<
1808 ret = 0; <<
1809 goto out_quota; <<
1810 } <<
1811 <<
1812 mlog_errno(ret); 1804 mlog_errno(ret);
1813 goto out_quota; 1805 goto out_quota;
1814 } 1806 }
1815 1807
>> 1808 /*
>> 1809 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
>> 1810 * the target page. In this case, we exit with no error and no target
>> 1811 * page. This will trigger the caller, page_mkwrite(), to re-try
>> 1812 * the operation.
>> 1813 */
>> 1814 if (ret == -EAGAIN) {
>> 1815 BUG_ON(wc->w_target_page);
>> 1816 ret = 0;
>> 1817 goto out_quota;
>> 1818 }
>> 1819
1816 ret = ocfs2_write_cluster_by_desc(map 1820 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1817 len 1821 len);
1818 if (ret) { 1822 if (ret) {
1819 mlog_errno(ret); 1823 mlog_errno(ret);
1820 goto out_quota; 1824 goto out_quota;
1821 } 1825 }
1822 1826
1823 if (data_ac) 1827 if (data_ac)
1824 ocfs2_free_alloc_context(data 1828 ocfs2_free_alloc_context(data_ac);
1825 if (meta_ac) 1829 if (meta_ac)
1826 ocfs2_free_alloc_context(meta 1830 ocfs2_free_alloc_context(meta_ac);
1827 1831
1828 success: 1832 success:
1829 if (pagep) 1833 if (pagep)
1830 *pagep = wc->w_target_page; 1834 *pagep = wc->w_target_page;
1831 *fsdata = wc; 1835 *fsdata = wc;
1832 return 0; 1836 return 0;
1833 out_quota: 1837 out_quota:
1834 if (clusters_to_alloc) 1838 if (clusters_to_alloc)
1835 dquot_free_space(inode, 1839 dquot_free_space(inode,
1836 ocfs2_clusters_to_b 1840 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1837 out_commit: 1841 out_commit:
1838 ocfs2_commit_trans(osb, handle); 1842 ocfs2_commit_trans(osb, handle);
1839 1843
1840 out: 1844 out:
1841 /* 1845 /*
1842 * The mmapped page won't be unlocked 1846 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1843 * even in case of error here like EN 1847 * even in case of error here like ENOSPC and ENOMEM. So, we need
1844 * to unlock the target page manually 1848 * to unlock the target page manually to prevent deadlocks when
1845 * retrying again on ENOSPC, or when 1849 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1846 * to VM code. 1850 * to VM code.
1847 */ 1851 */
1848 if (wc->w_target_locked) 1852 if (wc->w_target_locked)
1849 unlock_page(mmap_page); 1853 unlock_page(mmap_page);
1850 1854
1851 ocfs2_free_write_ctxt(inode, wc); 1855 ocfs2_free_write_ctxt(inode, wc);
1852 1856
1853 if (data_ac) { 1857 if (data_ac) {
1854 ocfs2_free_alloc_context(data 1858 ocfs2_free_alloc_context(data_ac);
1855 data_ac = NULL; 1859 data_ac = NULL;
1856 } 1860 }
1857 if (meta_ac) { 1861 if (meta_ac) {
1858 ocfs2_free_alloc_context(meta 1862 ocfs2_free_alloc_context(meta_ac);
1859 meta_ac = NULL; 1863 meta_ac = NULL;
1860 } 1864 }
1861 1865
1862 if (ret == -ENOSPC && try_free) { 1866 if (ret == -ENOSPC && try_free) {
1863 /* 1867 /*
1864 * Try to free some truncate 1868 * Try to free some truncate log so that we can have enough
1865 * clusters to allocate. 1869 * clusters to allocate.
1866 */ 1870 */
1867 try_free = 0; 1871 try_free = 0;
1868 1872
1869 ret1 = ocfs2_try_to_free_trun 1873 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1870 if (ret1 == 1) 1874 if (ret1 == 1)
1871 goto try_again; 1875 goto try_again;
1872 1876
1873 if (ret1 < 0) 1877 if (ret1 < 0)
1874 mlog_errno(ret1); 1878 mlog_errno(ret1);
1875 } 1879 }
1876 1880
1877 return ret; 1881 return ret;
1878 } 1882 }
1879 1883
1880 static int ocfs2_write_begin(struct file *fil 1884 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1881 loff_t pos, unsi !! 1885 loff_t pos, unsigned len, unsigned flags,
1882 struct page **pa 1886 struct page **pagep, void **fsdata)
1883 { 1887 {
1884 int ret; 1888 int ret;
1885 struct buffer_head *di_bh = NULL; 1889 struct buffer_head *di_bh = NULL;
1886 struct inode *inode = mapping->host; 1890 struct inode *inode = mapping->host;
1887 1891
1888 ret = ocfs2_inode_lock(inode, &di_bh, 1892 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1889 if (ret) { 1893 if (ret) {
1890 mlog_errno(ret); 1894 mlog_errno(ret);
1891 return ret; 1895 return ret;
1892 } 1896 }
1893 1897
1894 /* 1898 /*
1895 * Take alloc sem here to prevent con 1899 * Take alloc sem here to prevent concurrent lookups. That way
1896 * the mapping, zeroing and tree mani 1900 * the mapping, zeroing and tree manipulation within
1897 * ocfs2_write() will be safe against !! 1901 * ocfs2_write() will be safe against ->readpage(). This
1898 * should also serve to lock out allo 1902 * should also serve to lock out allocation from a shared
1899 * writeable region. 1903 * writeable region.
1900 */ 1904 */
1901 down_write(&OCFS2_I(inode)->ip_alloc_ 1905 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1902 1906
1903 ret = ocfs2_write_begin_nolock(mappin 1907 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1904 pagep, 1908 pagep, fsdata, di_bh, NULL);
1905 if (ret) { 1909 if (ret) {
1906 mlog_errno(ret); 1910 mlog_errno(ret);
1907 goto out_fail; 1911 goto out_fail;
1908 } 1912 }
1909 1913
1910 brelse(di_bh); 1914 brelse(di_bh);
1911 1915
1912 return 0; 1916 return 0;
1913 1917
1914 out_fail: 1918 out_fail:
1915 up_write(&OCFS2_I(inode)->ip_alloc_se 1919 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1916 1920
1917 brelse(di_bh); 1921 brelse(di_bh);
1918 ocfs2_inode_unlock(inode, 1); 1922 ocfs2_inode_unlock(inode, 1);
1919 1923
1920 return ret; 1924 return ret;
1921 } 1925 }
1922 1926
1923 static void ocfs2_write_end_inline(struct ino 1927 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1924 unsigned l 1928 unsigned len, unsigned *copied,
1925 struct ocf 1929 struct ocfs2_dinode *di,
1926 struct ocf 1930 struct ocfs2_write_ctxt *wc)
1927 { 1931 {
1928 void *kaddr; 1932 void *kaddr;
1929 1933
1930 if (unlikely(*copied < len)) { 1934 if (unlikely(*copied < len)) {
1931 if (!PageUptodate(wc->w_targe 1935 if (!PageUptodate(wc->w_target_page)) {
1932 *copied = 0; 1936 *copied = 0;
1933 return; 1937 return;
1934 } 1938 }
1935 } 1939 }
1936 1940
1937 kaddr = kmap_atomic(wc->w_target_page 1941 kaddr = kmap_atomic(wc->w_target_page);
1938 memcpy(di->id2.i_data.id_data + pos, 1942 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1939 kunmap_atomic(kaddr); 1943 kunmap_atomic(kaddr);
1940 1944
1941 trace_ocfs2_write_end_inline( 1945 trace_ocfs2_write_end_inline(
1942 (unsigned long long)OCFS2_I(inod 1946 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1943 (unsigned long long)pos, *copied 1947 (unsigned long long)pos, *copied,
1944 le16_to_cpu(di->id2.i_data.id_co 1948 le16_to_cpu(di->id2.i_data.id_count),
1945 le16_to_cpu(di->i_dyn_features)) 1949 le16_to_cpu(di->i_dyn_features));
1946 } 1950 }
1947 1951
1948 int ocfs2_write_end_nolock(struct address_spa 1952 int ocfs2_write_end_nolock(struct address_space *mapping,
1949 loff_t pos, unsign 1953 loff_t pos, unsigned len, unsigned copied, void *fsdata)
1950 { 1954 {
1951 int i, ret; 1955 int i, ret;
1952 unsigned from, to, start = pos & (PAG 1956 unsigned from, to, start = pos & (PAGE_SIZE - 1);
1953 struct inode *inode = mapping->host; 1957 struct inode *inode = mapping->host;
1954 struct ocfs2_super *osb = OCFS2_SB(in 1958 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1955 struct ocfs2_write_ctxt *wc = fsdata; 1959 struct ocfs2_write_ctxt *wc = fsdata;
1956 struct ocfs2_dinode *di = (struct ocf 1960 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1957 handle_t *handle = wc->w_handle; 1961 handle_t *handle = wc->w_handle;
1958 struct page *tmppage; 1962 struct page *tmppage;
1959 1963
1960 BUG_ON(!list_empty(&wc->w_unwritten_l 1964 BUG_ON(!list_empty(&wc->w_unwritten_list));
1961 1965
1962 if (handle) { 1966 if (handle) {
1963 ret = ocfs2_journal_access_di 1967 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1964 wc->w_di_bh, 1968 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1965 if (ret) { 1969 if (ret) {
1966 copied = ret; 1970 copied = ret;
1967 mlog_errno(ret); 1971 mlog_errno(ret);
1968 goto out; 1972 goto out;
1969 } 1973 }
1970 } 1974 }
1971 1975
1972 if (OCFS2_I(inode)->ip_dyn_features & 1976 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1973 ocfs2_write_end_inline(inode, 1977 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1974 goto out_write_size; 1978 goto out_write_size;
1975 } 1979 }
1976 1980
1977 if (unlikely(copied < len) && wc->w_t 1981 if (unlikely(copied < len) && wc->w_target_page) {
1978 loff_t new_isize; <<
1979 <<
1980 if (!PageUptodate(wc->w_targe 1982 if (!PageUptodate(wc->w_target_page))
1981 copied = 0; 1983 copied = 0;
1982 1984
1983 new_isize = max_t(loff_t, i_s !! 1985 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1984 if (new_isize > page_offset(w !! 1986 start+len);
1985 ocfs2_zero_new_buffer <<
1986 <<
1987 else { <<
1988 /* <<
1989 * When page is fully <<
1990 * failed), do not bo <<
1991 * it instead so that <<
1992 * put page & buffer <<
1993 * state. <<
1994 */ <<
1995 block_invalidate_foli <<
1996 <<
1997 } <<
1998 } 1987 }
1999 if (wc->w_target_page) 1988 if (wc->w_target_page)
2000 flush_dcache_page(wc->w_targe 1989 flush_dcache_page(wc->w_target_page);
2001 1990
2002 for(i = 0; i < wc->w_num_pages; i++) 1991 for(i = 0; i < wc->w_num_pages; i++) {
2003 tmppage = wc->w_pages[i]; 1992 tmppage = wc->w_pages[i];
2004 1993
2005 /* This is the direct io targ 1994 /* This is the direct io target page. */
2006 if (tmppage == NULL) 1995 if (tmppage == NULL)
2007 continue; 1996 continue;
2008 1997
2009 if (tmppage == wc->w_target_p 1998 if (tmppage == wc->w_target_page) {
2010 from = wc->w_target_f 1999 from = wc->w_target_from;
2011 to = wc->w_target_to; 2000 to = wc->w_target_to;
2012 2001
2013 BUG_ON(from > PAGE_SI 2002 BUG_ON(from > PAGE_SIZE ||
2014 to > PAGE_SIZE 2003 to > PAGE_SIZE ||
2015 to < from); 2004 to < from);
2016 } else { 2005 } else {
2017 /* 2006 /*
2018 * Pages adjacent to 2007 * Pages adjacent to the target (if any) imply
2019 * a hole-filling wri 2008 * a hole-filling write in which case we want
2020 * to flush their ent 2009 * to flush their entire range.
2021 */ 2010 */
2022 from = 0; 2011 from = 0;
2023 to = PAGE_SIZE; 2012 to = PAGE_SIZE;
2024 } 2013 }
2025 2014
2026 if (page_has_buffers(tmppage) 2015 if (page_has_buffers(tmppage)) {
2027 if (handle && ocfs2_s 2016 if (handle && ocfs2_should_order_data(inode)) {
2028 loff_t start_ 2017 loff_t start_byte =
2029 ((lof 2018 ((loff_t)tmppage->index << PAGE_SHIFT) +
2030 from; 2019 from;
2031 loff_t length 2020 loff_t length = to - from;
2032 ocfs2_jbd2_in 2021 ocfs2_jbd2_inode_add_write(handle, inode,
2033 2022 start_byte, length);
2034 } 2023 }
2035 block_commit_write(tm 2024 block_commit_write(tmppage, from, to);
2036 } 2025 }
2037 } 2026 }
2038 2027
2039 out_write_size: 2028 out_write_size:
2040 /* Direct io do not update i_size her 2029 /* Direct io do not update i_size here. */
2041 if (wc->w_type != OCFS2_WRITE_DIRECT) 2030 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2042 pos += copied; 2031 pos += copied;
2043 if (pos > i_size_read(inode)) 2032 if (pos > i_size_read(inode)) {
2044 i_size_write(inode, p 2033 i_size_write(inode, pos);
2045 mark_inode_dirty(inod 2034 mark_inode_dirty(inode);
2046 } 2035 }
2047 inode->i_blocks = ocfs2_inode 2036 inode->i_blocks = ocfs2_inode_sector_count(inode);
2048 di->i_size = cpu_to_le64((u64 2037 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2049 inode_set_mtime_to_ts(inode, !! 2038 inode->i_mtime = inode->i_ctime = current_time(inode);
2050 di->i_mtime = di->i_ctime = c !! 2039 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2051 di->i_mtime_nsec = di->i_ctim !! 2040 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2052 if (handle) 2041 if (handle)
2053 ocfs2_update_inode_fs 2042 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2054 } 2043 }
2055 if (handle) 2044 if (handle)
2056 ocfs2_journal_dirty(handle, w 2045 ocfs2_journal_dirty(handle, wc->w_di_bh);
2057 2046
2058 out: 2047 out:
2059 /* unlock pages before dealloc since 2048 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2060 * lock, or it will cause a deadlock 2049 * lock, or it will cause a deadlock since journal commit threads holds
2061 * this lock and will ask for the pag 2050 * this lock and will ask for the page lock when flushing the data.
2062 * put it here to preserve the unlock 2051 * put it here to preserve the unlock order.
2063 */ 2052 */
2064 ocfs2_unlock_pages(wc); 2053 ocfs2_unlock_pages(wc);
2065 2054
2066 if (handle) 2055 if (handle)
2067 ocfs2_commit_trans(osb, handl 2056 ocfs2_commit_trans(osb, handle);
2068 2057
2069 ocfs2_run_deallocs(osb, &wc->w_deallo 2058 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2070 2059
2071 brelse(wc->w_di_bh); 2060 brelse(wc->w_di_bh);
2072 kfree(wc); 2061 kfree(wc);
2073 2062
2074 return copied; 2063 return copied;
2075 } 2064 }
2076 2065
2077 static int ocfs2_write_end(struct file *file, 2066 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2078 loff_t pos, unsign 2067 loff_t pos, unsigned len, unsigned copied,
2079 struct page *page, 2068 struct page *page, void *fsdata)
2080 { 2069 {
2081 int ret; 2070 int ret;
2082 struct inode *inode = mapping->host; 2071 struct inode *inode = mapping->host;
2083 2072
2084 ret = ocfs2_write_end_nolock(mapping, 2073 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata);
2085 2074
2086 up_write(&OCFS2_I(inode)->ip_alloc_se 2075 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2087 ocfs2_inode_unlock(inode, 1); 2076 ocfs2_inode_unlock(inode, 1);
2088 2077
2089 return ret; 2078 return ret;
2090 } 2079 }
2091 2080
2092 struct ocfs2_dio_write_ctxt { 2081 struct ocfs2_dio_write_ctxt {
2093 struct list_head dw_zero_list; 2082 struct list_head dw_zero_list;
2094 unsigned dw_zero_count 2083 unsigned dw_zero_count;
2095 int dw_orphaned; 2084 int dw_orphaned;
2096 pid_t dw_writer_pid 2085 pid_t dw_writer_pid;
2097 }; 2086 };
2098 2087
2099 static struct ocfs2_dio_write_ctxt * 2088 static struct ocfs2_dio_write_ctxt *
2100 ocfs2_dio_alloc_write_ctx(struct buffer_head 2089 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2101 { 2090 {
2102 struct ocfs2_dio_write_ctxt *dwc = NU 2091 struct ocfs2_dio_write_ctxt *dwc = NULL;
2103 2092
2104 if (bh->b_private) 2093 if (bh->b_private)
2105 return bh->b_private; 2094 return bh->b_private;
2106 2095
2107 dwc = kmalloc(sizeof(struct ocfs2_dio 2096 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2108 if (dwc == NULL) 2097 if (dwc == NULL)
2109 return NULL; 2098 return NULL;
2110 INIT_LIST_HEAD(&dwc->dw_zero_list); 2099 INIT_LIST_HEAD(&dwc->dw_zero_list);
2111 dwc->dw_zero_count = 0; 2100 dwc->dw_zero_count = 0;
2112 dwc->dw_orphaned = 0; 2101 dwc->dw_orphaned = 0;
2113 dwc->dw_writer_pid = task_pid_nr(curr 2102 dwc->dw_writer_pid = task_pid_nr(current);
2114 bh->b_private = dwc; 2103 bh->b_private = dwc;
2115 *alloc = 1; 2104 *alloc = 1;
2116 2105
2117 return dwc; 2106 return dwc;
2118 } 2107 }
2119 2108
2120 static void ocfs2_dio_free_write_ctx(struct i 2109 static void ocfs2_dio_free_write_ctx(struct inode *inode,
2121 struct o 2110 struct ocfs2_dio_write_ctxt *dwc)
2122 { 2111 {
2123 ocfs2_free_unwritten_list(inode, &dwc 2112 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2124 kfree(dwc); 2113 kfree(dwc);
2125 } 2114 }
2126 2115
2127 /* 2116 /*
2128 * TODO: Make this into a generic get_blocks 2117 * TODO: Make this into a generic get_blocks function.
2129 * 2118 *
2130 * From do_direct_io in direct-io.c: 2119 * From do_direct_io in direct-io.c:
2131 * "So what we do is to permit the ->get_blo 2120 * "So what we do is to permit the ->get_blocks function to populate
2132 * bh.b_size with the size of IO which is p 2121 * bh.b_size with the size of IO which is permitted at this offset and
2133 * this i_blkbits." 2122 * this i_blkbits."
2134 * 2123 *
2135 * This function is called directly from get_ 2124 * This function is called directly from get_more_blocks in direct-io.c.
2136 * 2125 *
2137 * called like this: dio->get_blocks(dio->ino 2126 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2138 * fs_co 2127 * fs_count, map_bh, dio->rw == WRITE);
2139 */ 2128 */
2140 static int ocfs2_dio_wr_get_block(struct inod 2129 static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock,
2141 struct buffer_ 2130 struct buffer_head *bh_result, int create)
2142 { 2131 {
2143 struct ocfs2_super *osb = OCFS2_SB(in 2132 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2144 struct ocfs2_inode_info *oi = OCFS2_I 2133 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2145 struct ocfs2_write_ctxt *wc; 2134 struct ocfs2_write_ctxt *wc;
2146 struct ocfs2_write_cluster_desc *desc 2135 struct ocfs2_write_cluster_desc *desc = NULL;
2147 struct ocfs2_dio_write_ctxt *dwc = NU 2136 struct ocfs2_dio_write_ctxt *dwc = NULL;
2148 struct buffer_head *di_bh = NULL; 2137 struct buffer_head *di_bh = NULL;
2149 u64 p_blkno; 2138 u64 p_blkno;
2150 unsigned int i_blkbits = inode->i_sb- 2139 unsigned int i_blkbits = inode->i_sb->s_blocksize_bits;
2151 loff_t pos = iblock << i_blkbits; 2140 loff_t pos = iblock << i_blkbits;
2152 sector_t endblk = (i_size_read(inode) 2141 sector_t endblk = (i_size_read(inode) - 1) >> i_blkbits;
2153 unsigned len, total_len = bh_result-> 2142 unsigned len, total_len = bh_result->b_size;
2154 int ret = 0, first_get_block = 0; 2143 int ret = 0, first_get_block = 0;
2155 2144
2156 len = osb->s_clustersize - (pos & (os 2145 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2157 len = min(total_len, len); 2146 len = min(total_len, len);
2158 2147
2159 /* 2148 /*
2160 * bh_result->b_size is count in get_ 2149 * bh_result->b_size is count in get_more_blocks according to write
2161 * "pos" and "end", we need map twice 2150 * "pos" and "end", we need map twice to return different buffer state:
2162 * 1. area in file size, not set NEW; 2151 * 1. area in file size, not set NEW;
2163 * 2. area out file size, set NEW. 2152 * 2. area out file size, set NEW.
2164 * 2153 *
2165 * iblock endblk 2154 * iblock endblk
2166 * |--------|---------|---------|---- 2155 * |--------|---------|---------|---------
2167 * |<-------area in file------->| 2156 * |<-------area in file------->|
2168 */ 2157 */
2169 2158
2170 if ((iblock <= endblk) && 2159 if ((iblock <= endblk) &&
2171 ((iblock + ((len - 1) >> i_blkbit 2160 ((iblock + ((len - 1) >> i_blkbits)) > endblk))
2172 len = (endblk - iblock + 1) < 2161 len = (endblk - iblock + 1) << i_blkbits;
2173 2162
2174 mlog(0, "get block of %lu at %llu:%u 2163 mlog(0, "get block of %lu at %llu:%u req %u\n",
2175 inode->i_ino, pos, le 2164 inode->i_ino, pos, len, total_len);
2176 2165
2177 /* 2166 /*
2178 * Because we need to change file siz 2167 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2179 * we may need to add it to orphan di 2168 * we may need to add it to orphan dir. So can not fall to fast path
2180 * while file size will be changed. 2169 * while file size will be changed.
2181 */ 2170 */
2182 if (pos + total_len <= i_size_read(in 2171 if (pos + total_len <= i_size_read(inode)) {
2183 2172
2184 /* This is the fast path for 2173 /* This is the fast path for re-write. */
2185 ret = ocfs2_lock_get_block(in 2174 ret = ocfs2_lock_get_block(inode, iblock, bh_result, create);
2186 if (buffer_mapped(bh_result) 2175 if (buffer_mapped(bh_result) &&
2187 !buffer_new(bh_result) && 2176 !buffer_new(bh_result) &&
2188 ret == 0) 2177 ret == 0)
2189 goto out; 2178 goto out;
2190 2179
2191 /* Clear state set by ocfs2_g 2180 /* Clear state set by ocfs2_get_block. */
2192 bh_result->b_state = 0; 2181 bh_result->b_state = 0;
2193 } 2182 }
2194 2183
2195 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2184 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2196 if (unlikely(dwc == NULL)) { 2185 if (unlikely(dwc == NULL)) {
2197 ret = -ENOMEM; 2186 ret = -ENOMEM;
2198 mlog_errno(ret); 2187 mlog_errno(ret);
2199 goto out; 2188 goto out;
2200 } 2189 }
2201 2190
2202 if (ocfs2_clusters_for_bytes(inode->i 2191 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2203 ocfs2_clusters_for_bytes(inode->i 2192 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2204 !dwc->dw_orphaned) { 2193 !dwc->dw_orphaned) {
2205 /* 2194 /*
2206 * when we are going to alloc 2195 * when we are going to alloc extents beyond file size, add the
2207 * inode to orphan dir, so we 2196 * inode to orphan dir, so we can recall those spaces when
2208 * system crashed during writ 2197 * system crashed during write.
2209 */ 2198 */
2210 ret = ocfs2_add_inode_to_orph 2199 ret = ocfs2_add_inode_to_orphan(osb, inode);
2211 if (ret < 0) { 2200 if (ret < 0) {
2212 mlog_errno(ret); 2201 mlog_errno(ret);
2213 goto out; 2202 goto out;
2214 } 2203 }
2215 dwc->dw_orphaned = 1; 2204 dwc->dw_orphaned = 1;
2216 } 2205 }
2217 2206
2218 ret = ocfs2_inode_lock(inode, &di_bh, 2207 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2219 if (ret) { 2208 if (ret) {
2220 mlog_errno(ret); 2209 mlog_errno(ret);
2221 goto out; 2210 goto out;
2222 } 2211 }
2223 2212
2224 down_write(&oi->ip_alloc_sem); 2213 down_write(&oi->ip_alloc_sem);
2225 2214
2226 if (first_get_block) { 2215 if (first_get_block) {
2227 if (ocfs2_sparse_alloc(osb)) 2216 if (ocfs2_sparse_alloc(osb))
2228 ret = ocfs2_zero_tail 2217 ret = ocfs2_zero_tail(inode, di_bh, pos);
2229 else 2218 else
2230 ret = ocfs2_expand_no 2219 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2231 2220 total_len, NULL);
2232 if (ret < 0) { 2221 if (ret < 0) {
2233 mlog_errno(ret); 2222 mlog_errno(ret);
2234 goto unlock; 2223 goto unlock;
2235 } 2224 }
2236 } 2225 }
2237 2226
2238 ret = ocfs2_write_begin_nolock(inode- 2227 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2239 OCFS2_ 2228 OCFS2_WRITE_DIRECT, NULL,
2240 (void 2229 (void **)&wc, di_bh, NULL);
2241 if (ret) { 2230 if (ret) {
2242 mlog_errno(ret); 2231 mlog_errno(ret);
2243 goto unlock; 2232 goto unlock;
2244 } 2233 }
2245 2234
2246 desc = &wc->w_desc[0]; 2235 desc = &wc->w_desc[0];
2247 2236
2248 p_blkno = ocfs2_clusters_to_blocks(in 2237 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2249 BUG_ON(p_blkno == 0); 2238 BUG_ON(p_blkno == 0);
2250 p_blkno += iblock & (u64)(ocfs2_clust 2239 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2251 2240
2252 map_bh(bh_result, inode->i_sb, p_blkn 2241 map_bh(bh_result, inode->i_sb, p_blkno);
2253 bh_result->b_size = len; 2242 bh_result->b_size = len;
2254 if (desc->c_needs_zero) 2243 if (desc->c_needs_zero)
2255 set_buffer_new(bh_result); 2244 set_buffer_new(bh_result);
2256 2245
2257 if (iblock > endblk) 2246 if (iblock > endblk)
2258 set_buffer_new(bh_result); 2247 set_buffer_new(bh_result);
2259 2248
2260 /* May sleep in end_io. It should not 2249 /* May sleep in end_io. It should not happen in a irq context. So defer
2261 * it to dio work queue. */ 2250 * it to dio work queue. */
2262 set_buffer_defer_completion(bh_result 2251 set_buffer_defer_completion(bh_result);
2263 2252
2264 if (!list_empty(&wc->w_unwritten_list 2253 if (!list_empty(&wc->w_unwritten_list)) {
2265 struct ocfs2_unwritten_extent 2254 struct ocfs2_unwritten_extent *ue = NULL;
2266 2255
2267 ue = list_first_entry(&wc->w_ 2256 ue = list_first_entry(&wc->w_unwritten_list,
2268 struct 2257 struct ocfs2_unwritten_extent,
2269 ue_node 2258 ue_node);
2270 BUG_ON(ue->ue_cpos != desc->c 2259 BUG_ON(ue->ue_cpos != desc->c_cpos);
2271 /* The physical address may b 2260 /* The physical address may be 0, fill it. */
2272 ue->ue_phys = desc->c_phys; 2261 ue->ue_phys = desc->c_phys;
2273 2262
2274 list_splice_tail_init(&wc->w_ 2263 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
2275 dwc->dw_zero_count += wc->w_u 2264 dwc->dw_zero_count += wc->w_unwritten_count;
2276 } 2265 }
2277 2266
2278 ret = ocfs2_write_end_nolock(inode->i 2267 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc);
2279 BUG_ON(ret != len); 2268 BUG_ON(ret != len);
2280 ret = 0; 2269 ret = 0;
2281 unlock: 2270 unlock:
2282 up_write(&oi->ip_alloc_sem); 2271 up_write(&oi->ip_alloc_sem);
2283 ocfs2_inode_unlock(inode, 1); 2272 ocfs2_inode_unlock(inode, 1);
2284 brelse(di_bh); 2273 brelse(di_bh);
2285 out: 2274 out:
>> 2275 if (ret < 0)
>> 2276 ret = -EIO;
2286 return ret; 2277 return ret;
2287 } 2278 }
2288 2279
2289 static int ocfs2_dio_end_io_write(struct inod 2280 static int ocfs2_dio_end_io_write(struct inode *inode,
2290 struct ocfs 2281 struct ocfs2_dio_write_ctxt *dwc,
2291 loff_t offs 2282 loff_t offset,
2292 ssize_t byt 2283 ssize_t bytes)
2293 { 2284 {
2294 struct ocfs2_cached_dealloc_ctxt deal 2285 struct ocfs2_cached_dealloc_ctxt dealloc;
2295 struct ocfs2_extent_tree et; 2286 struct ocfs2_extent_tree et;
2296 struct ocfs2_super *osb = OCFS2_SB(in 2287 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2297 struct ocfs2_inode_info *oi = OCFS2_I 2288 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2298 struct ocfs2_unwritten_extent *ue = N 2289 struct ocfs2_unwritten_extent *ue = NULL;
2299 struct buffer_head *di_bh = NULL; 2290 struct buffer_head *di_bh = NULL;
2300 struct ocfs2_dinode *di; 2291 struct ocfs2_dinode *di;
2301 struct ocfs2_alloc_context *data_ac = 2292 struct ocfs2_alloc_context *data_ac = NULL;
2302 struct ocfs2_alloc_context *meta_ac = 2293 struct ocfs2_alloc_context *meta_ac = NULL;
2303 handle_t *handle = NULL; 2294 handle_t *handle = NULL;
2304 loff_t end = offset + bytes; 2295 loff_t end = offset + bytes;
2305 int ret = 0, credits = 0; 2296 int ret = 0, credits = 0;
2306 2297
2307 ocfs2_init_dealloc_ctxt(&dealloc); 2298 ocfs2_init_dealloc_ctxt(&dealloc);
2308 2299
2309 /* We do clear unwritten, delete orph 2300 /* We do clear unwritten, delete orphan, change i_size here. If neither
2310 * of these happen, we can skip all t 2301 * of these happen, we can skip all this. */
2311 if (list_empty(&dwc->dw_zero_list) && 2302 if (list_empty(&dwc->dw_zero_list) &&
2312 end <= i_size_read(inode) && 2303 end <= i_size_read(inode) &&
2313 !dwc->dw_orphaned) 2304 !dwc->dw_orphaned)
2314 goto out; 2305 goto out;
2315 2306
2316 ret = ocfs2_inode_lock(inode, &di_bh, 2307 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2317 if (ret < 0) { 2308 if (ret < 0) {
2318 mlog_errno(ret); 2309 mlog_errno(ret);
2319 goto out; 2310 goto out;
2320 } 2311 }
2321 2312
2322 down_write(&oi->ip_alloc_sem); 2313 down_write(&oi->ip_alloc_sem);
2323 2314
2324 /* Delete orphan before acquire i_rws !! 2315 /* Delete orphan before acquire i_mutex. */
2325 if (dwc->dw_orphaned) { 2316 if (dwc->dw_orphaned) {
2326 BUG_ON(dwc->dw_writer_pid != 2317 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2327 2318
2328 end = end > i_size_read(inode 2319 end = end > i_size_read(inode) ? end : 0;
2329 2320
2330 ret = ocfs2_del_inode_from_or 2321 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2331 !!end, end); 2322 !!end, end);
2332 if (ret < 0) 2323 if (ret < 0)
2333 mlog_errno(ret); 2324 mlog_errno(ret);
2334 } 2325 }
2335 2326
2336 di = (struct ocfs2_dinode *)di_bh->b_ 2327 di = (struct ocfs2_dinode *)di_bh->b_data;
2337 2328
2338 ocfs2_init_dinode_extent_tree(&et, IN 2329 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2339 2330
2340 /* Attach dealloc with extent tree in 2331 /* Attach dealloc with extent tree in case that we may reuse extents
2341 * which are already unlinked from cu 2332 * which are already unlinked from current extent tree due to extent
2342 * rotation and merging. 2333 * rotation and merging.
2343 */ 2334 */
2344 et.et_dealloc = &dealloc; 2335 et.et_dealloc = &dealloc;
2345 2336
2346 ret = ocfs2_lock_allocators(inode, &e 2337 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2347 &data_ac, 2338 &data_ac, &meta_ac);
2348 if (ret) { 2339 if (ret) {
2349 mlog_errno(ret); 2340 mlog_errno(ret);
2350 goto unlock; 2341 goto unlock;
2351 } 2342 }
2352 2343
2353 credits = ocfs2_calc_extend_credits(i 2344 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2354 2345
2355 handle = ocfs2_start_trans(osb, credi 2346 handle = ocfs2_start_trans(osb, credits);
2356 if (IS_ERR(handle)) { 2347 if (IS_ERR(handle)) {
2357 ret = PTR_ERR(handle); 2348 ret = PTR_ERR(handle);
2358 mlog_errno(ret); 2349 mlog_errno(ret);
2359 goto unlock; 2350 goto unlock;
2360 } 2351 }
2361 ret = ocfs2_journal_access_di(handle, 2352 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2362 OCFS2_J 2353 OCFS2_JOURNAL_ACCESS_WRITE);
2363 if (ret) { 2354 if (ret) {
2364 mlog_errno(ret); 2355 mlog_errno(ret);
2365 goto commit; 2356 goto commit;
2366 } 2357 }
2367 2358
2368 list_for_each_entry(ue, &dwc->dw_zero 2359 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) {
2369 ret = ocfs2_assure_trans_cred <<
2370 if (ret < 0) { <<
2371 mlog_errno(ret); <<
2372 break; <<
2373 } <<
2374 ret = ocfs2_mark_extent_writt 2360 ret = ocfs2_mark_extent_written(inode, &et, handle,
2375 2361 ue->ue_cpos, 1,
2376 2362 ue->ue_phys,
2377 2363 meta_ac, &dealloc);
2378 if (ret < 0) { 2364 if (ret < 0) {
2379 mlog_errno(ret); 2365 mlog_errno(ret);
2380 break; 2366 break;
2381 } 2367 }
2382 } 2368 }
2383 2369
2384 if (end > i_size_read(inode)) { 2370 if (end > i_size_read(inode)) {
2385 ret = ocfs2_set_inode_size(ha 2371 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2386 if (ret < 0) 2372 if (ret < 0)
2387 mlog_errno(ret); 2373 mlog_errno(ret);
2388 } 2374 }
2389 commit: 2375 commit:
2390 ocfs2_commit_trans(osb, handle); 2376 ocfs2_commit_trans(osb, handle);
2391 unlock: 2377 unlock:
2392 up_write(&oi->ip_alloc_sem); 2378 up_write(&oi->ip_alloc_sem);
2393 ocfs2_inode_unlock(inode, 1); 2379 ocfs2_inode_unlock(inode, 1);
2394 brelse(di_bh); 2380 brelse(di_bh);
2395 out: 2381 out:
2396 if (data_ac) 2382 if (data_ac)
2397 ocfs2_free_alloc_context(data 2383 ocfs2_free_alloc_context(data_ac);
2398 if (meta_ac) 2384 if (meta_ac)
2399 ocfs2_free_alloc_context(meta 2385 ocfs2_free_alloc_context(meta_ac);
2400 ocfs2_run_deallocs(osb, &dealloc); 2386 ocfs2_run_deallocs(osb, &dealloc);
2401 ocfs2_dio_free_write_ctx(inode, dwc); 2387 ocfs2_dio_free_write_ctx(inode, dwc);
2402 2388
2403 return ret; 2389 return ret;
2404 } 2390 }
2405 2391
2406 /* 2392 /*
2407 * ocfs2_dio_end_io is called by the dio core 2393 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
2408 * particularly interested in the aio/dio cas 2394 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2409 * to protect io on one node from truncation 2395 * to protect io on one node from truncation on another.
2410 */ 2396 */
2411 static int ocfs2_dio_end_io(struct kiocb *ioc 2397 static int ocfs2_dio_end_io(struct kiocb *iocb,
2412 loff_t offset, 2398 loff_t offset,
2413 ssize_t bytes, 2399 ssize_t bytes,
2414 void *private) 2400 void *private)
2415 { 2401 {
2416 struct inode *inode = file_inode(iocb 2402 struct inode *inode = file_inode(iocb->ki_filp);
2417 int level; 2403 int level;
2418 int ret = 0; 2404 int ret = 0;
2419 2405
2420 /* this io's submitter should not hav 2406 /* this io's submitter should not have unlocked this before we could */
2421 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2407 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2422 2408
2423 if (bytes <= 0) 2409 if (bytes <= 0)
2424 mlog_ratelimited(ML_ERROR, "D 2410 mlog_ratelimited(ML_ERROR, "Direct IO failed, bytes = %lld",
2425 (long long)b 2411 (long long)bytes);
2426 if (private) { 2412 if (private) {
2427 if (bytes > 0) 2413 if (bytes > 0)
2428 ret = ocfs2_dio_end_i 2414 ret = ocfs2_dio_end_io_write(inode, private, offset,
2429 2415 bytes);
2430 else 2416 else
2431 ocfs2_dio_free_write_ 2417 ocfs2_dio_free_write_ctx(inode, private);
2432 } 2418 }
2433 2419
2434 ocfs2_iocb_clear_rw_locked(iocb); 2420 ocfs2_iocb_clear_rw_locked(iocb);
2435 2421
2436 level = ocfs2_iocb_rw_locked_level(io 2422 level = ocfs2_iocb_rw_locked_level(iocb);
2437 ocfs2_rw_unlock(inode, level); 2423 ocfs2_rw_unlock(inode, level);
2438 return ret; 2424 return ret;
2439 } 2425 }
2440 2426
2441 static ssize_t ocfs2_direct_IO(struct kiocb * 2427 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2442 { 2428 {
2443 struct file *file = iocb->ki_filp; 2429 struct file *file = iocb->ki_filp;
2444 struct inode *inode = file->f_mapping 2430 struct inode *inode = file->f_mapping->host;
2445 struct ocfs2_super *osb = OCFS2_SB(in 2431 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2446 get_block_t *get_block; 2432 get_block_t *get_block;
2447 2433
2448 /* 2434 /*
2449 * Fallback to buffered I/O if we see 2435 * Fallback to buffered I/O if we see an inode without
2450 * extents. 2436 * extents.
2451 */ 2437 */
2452 if (OCFS2_I(inode)->ip_dyn_features & 2438 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2453 return 0; 2439 return 0;
2454 2440
2455 /* Fallback to buffered I/O if we do 2441 /* Fallback to buffered I/O if we do not support append dio. */
2456 if (iocb->ki_pos + iter->count > i_si 2442 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2457 !ocfs2_supports_append_dio(osb)) 2443 !ocfs2_supports_append_dio(osb))
2458 return 0; 2444 return 0;
2459 2445
2460 if (iov_iter_rw(iter) == READ) 2446 if (iov_iter_rw(iter) == READ)
2461 get_block = ocfs2_lock_get_bl 2447 get_block = ocfs2_lock_get_block;
2462 else 2448 else
2463 get_block = ocfs2_dio_wr_get_ 2449 get_block = ocfs2_dio_wr_get_block;
2464 2450
2465 return __blockdev_direct_IO(iocb, ino 2451 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2466 iter, get 2452 iter, get_block,
2467 ocfs2_dio !! 2453 ocfs2_dio_end_io, NULL, 0);
2468 } 2454 }
2469 2455
2470 const struct address_space_operations ocfs2_a 2456 const struct address_space_operations ocfs2_aops = {
2471 .dirty_folio = block_dirty !! 2457 .readpage = ocfs2_readpage,
2472 .read_folio = ocfs2_read_ <<
2473 .readahead = ocfs2_reada 2458 .readahead = ocfs2_readahead,
2474 .writepages = ocfs2_write !! 2459 .writepage = ocfs2_writepage,
2475 .write_begin = ocfs2_write 2460 .write_begin = ocfs2_write_begin,
2476 .write_end = ocfs2_write 2461 .write_end = ocfs2_write_end,
2477 .bmap = ocfs2_bmap, 2462 .bmap = ocfs2_bmap,
2478 .direct_IO = ocfs2_direc 2463 .direct_IO = ocfs2_direct_IO,
2479 .invalidate_folio = block_inval !! 2464 .invalidatepage = block_invalidatepage,
2480 .release_folio = ocfs2_relea !! 2465 .releasepage = ocfs2_releasepage,
2481 .migrate_folio = buffer_migr !! 2466 .migratepage = buffer_migrate_page,
2482 .is_partially_uptodate = block_is_pa 2467 .is_partially_uptodate = block_is_partially_uptodate,
2483 .error_remove_folio = generic_err !! 2468 .error_remove_page = generic_error_remove_page,
2484 }; 2469 };
2485 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.