1 // SPDX-License-Identifier: GPL-2.0-or-later 1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* !! 2 /* -*- mode: c; c-basic-offset: 8; -*-
>> 3 * vim: noexpandtab sw=8 ts=8 sts=0:
>> 4 *
3 * Copyright (C) 2002, 2004 Oracle. All right 5 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
4 */ 6 */
5 7
6 #include <linux/fs.h> 8 #include <linux/fs.h>
7 #include <linux/slab.h> 9 #include <linux/slab.h>
8 #include <linux/highmem.h> 10 #include <linux/highmem.h>
9 #include <linux/pagemap.h> 11 #include <linux/pagemap.h>
10 #include <asm/byteorder.h> 12 #include <asm/byteorder.h>
11 #include <linux/swap.h> 13 #include <linux/swap.h>
12 #include <linux/mpage.h> 14 #include <linux/mpage.h>
13 #include <linux/quotaops.h> 15 #include <linux/quotaops.h>
14 #include <linux/blkdev.h> 16 #include <linux/blkdev.h>
15 #include <linux/uio.h> 17 #include <linux/uio.h>
16 #include <linux/mm.h> 18 #include <linux/mm.h>
17 19
18 #include <cluster/masklog.h> 20 #include <cluster/masklog.h>
19 21
20 #include "ocfs2.h" 22 #include "ocfs2.h"
21 23
22 #include "alloc.h" 24 #include "alloc.h"
23 #include "aops.h" 25 #include "aops.h"
24 #include "dlmglue.h" 26 #include "dlmglue.h"
25 #include "extent_map.h" 27 #include "extent_map.h"
26 #include "file.h" 28 #include "file.h"
27 #include "inode.h" 29 #include "inode.h"
28 #include "journal.h" 30 #include "journal.h"
29 #include "suballoc.h" 31 #include "suballoc.h"
30 #include "super.h" 32 #include "super.h"
31 #include "symlink.h" 33 #include "symlink.h"
32 #include "refcounttree.h" 34 #include "refcounttree.h"
33 #include "ocfs2_trace.h" 35 #include "ocfs2_trace.h"
34 36
35 #include "buffer_head_io.h" 37 #include "buffer_head_io.h"
36 #include "dir.h" 38 #include "dir.h"
37 #include "namei.h" 39 #include "namei.h"
38 #include "sysfile.h" 40 #include "sysfile.h"
39 41
40 static int ocfs2_symlink_get_block(struct inod 42 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
41 struct buff 43 struct buffer_head *bh_result, int create)
42 { 44 {
43 int err = -EIO; 45 int err = -EIO;
44 int status; 46 int status;
45 struct ocfs2_dinode *fe = NULL; 47 struct ocfs2_dinode *fe = NULL;
46 struct buffer_head *bh = NULL; 48 struct buffer_head *bh = NULL;
47 struct buffer_head *buffer_cache_bh = 49 struct buffer_head *buffer_cache_bh = NULL;
48 struct ocfs2_super *osb = OCFS2_SB(ino 50 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
49 void *kaddr; 51 void *kaddr;
50 52
51 trace_ocfs2_symlink_get_block( 53 trace_ocfs2_symlink_get_block(
52 (unsigned long long)OC 54 (unsigned long long)OCFS2_I(inode)->ip_blkno,
53 (unsigned long long)ib 55 (unsigned long long)iblock, bh_result, create);
54 56
55 BUG_ON(ocfs2_inode_is_fast_symlink(ino 57 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
56 58
57 if ((iblock << inode->i_sb->s_blocksiz 59 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
58 mlog(ML_ERROR, "block offset > 60 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
59 (unsigned long long)ibloc 61 (unsigned long long)iblock);
60 goto bail; 62 goto bail;
61 } 63 }
62 64
63 status = ocfs2_read_inode_block(inode, 65 status = ocfs2_read_inode_block(inode, &bh);
64 if (status < 0) { 66 if (status < 0) {
65 mlog_errno(status); 67 mlog_errno(status);
66 goto bail; 68 goto bail;
67 } 69 }
68 fe = (struct ocfs2_dinode *) bh->b_dat 70 fe = (struct ocfs2_dinode *) bh->b_data;
69 71
70 if ((u64)iblock >= ocfs2_clusters_to_b 72 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
71 73 le32_to_cpu(fe->i_clusters))) {
72 err = -ENOMEM; 74 err = -ENOMEM;
73 mlog(ML_ERROR, "block offset i 75 mlog(ML_ERROR, "block offset is outside the allocated size: "
74 "%llu\n", (unsigned long 76 "%llu\n", (unsigned long long)iblock);
75 goto bail; 77 goto bail;
76 } 78 }
77 79
78 /* We don't use the page cache to crea 80 /* We don't use the page cache to create symlink data, so if
79 * need be, copy it over from the buff 81 * need be, copy it over from the buffer cache. */
80 if (!buffer_uptodate(bh_result) && ocf 82 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
81 u64 blkno = le64_to_cpu(fe->id 83 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
82 iblock; 84 iblock;
83 buffer_cache_bh = sb_getblk(os 85 buffer_cache_bh = sb_getblk(osb->sb, blkno);
84 if (!buffer_cache_bh) { 86 if (!buffer_cache_bh) {
85 err = -ENOMEM; 87 err = -ENOMEM;
86 mlog(ML_ERROR, "couldn 88 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
87 goto bail; 89 goto bail;
88 } 90 }
89 91
90 /* we haven't locked out trans 92 /* we haven't locked out transactions, so a commit
91 * could've happened. Since we 93 * could've happened. Since we've got a reference on
92 * the bh, even if it commits 94 * the bh, even if it commits while we're doing the
93 * copy, the data is still goo 95 * copy, the data is still good. */
94 if (buffer_jbd(buffer_cache_bh 96 if (buffer_jbd(buffer_cache_bh)
95 && ocfs2_inode_is_new(inod 97 && ocfs2_inode_is_new(inode)) {
96 kaddr = kmap_atomic(bh 98 kaddr = kmap_atomic(bh_result->b_page);
97 if (!kaddr) { 99 if (!kaddr) {
98 mlog(ML_ERROR, 100 mlog(ML_ERROR, "couldn't kmap!\n");
99 goto bail; 101 goto bail;
100 } 102 }
101 memcpy(kaddr + (bh_res 103 memcpy(kaddr + (bh_result->b_size * iblock),
102 buffer_cache_bh 104 buffer_cache_bh->b_data,
103 bh_result->b_si 105 bh_result->b_size);
104 kunmap_atomic(kaddr); 106 kunmap_atomic(kaddr);
105 set_buffer_uptodate(bh 107 set_buffer_uptodate(bh_result);
106 } 108 }
107 brelse(buffer_cache_bh); 109 brelse(buffer_cache_bh);
108 } 110 }
109 111
110 map_bh(bh_result, inode->i_sb, 112 map_bh(bh_result, inode->i_sb,
111 le64_to_cpu(fe->id2.i_list.l_re 113 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
112 114
113 err = 0; 115 err = 0;
114 116
115 bail: 117 bail:
116 brelse(bh); 118 brelse(bh);
117 119
118 return err; 120 return err;
119 } 121 }
120 122
121 static int ocfs2_lock_get_block(struct inode * 123 static int ocfs2_lock_get_block(struct inode *inode, sector_t iblock,
122 struct buffer_head *bh_res 124 struct buffer_head *bh_result, int create)
123 { 125 {
124 int ret = 0; 126 int ret = 0;
125 struct ocfs2_inode_info *oi = OCFS2_I( 127 struct ocfs2_inode_info *oi = OCFS2_I(inode);
126 128
127 down_read(&oi->ip_alloc_sem); 129 down_read(&oi->ip_alloc_sem);
128 ret = ocfs2_get_block(inode, iblock, b 130 ret = ocfs2_get_block(inode, iblock, bh_result, create);
129 up_read(&oi->ip_alloc_sem); 131 up_read(&oi->ip_alloc_sem);
130 132
131 return ret; 133 return ret;
132 } 134 }
133 135
134 int ocfs2_get_block(struct inode *inode, secto 136 int ocfs2_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh_res 137 struct buffer_head *bh_result, int create)
136 { 138 {
137 int err = 0; 139 int err = 0;
138 unsigned int ext_flags; 140 unsigned int ext_flags;
139 u64 max_blocks = bh_result->b_size >> 141 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
140 u64 p_blkno, count, past_eof; 142 u64 p_blkno, count, past_eof;
141 struct ocfs2_super *osb = OCFS2_SB(ino 143 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
142 144
143 trace_ocfs2_get_block((unsigned long l 145 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
144 (unsigned long l 146 (unsigned long long)iblock, bh_result, create);
145 147
146 if (OCFS2_I(inode)->ip_flags & OCFS2_I 148 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
147 mlog(ML_NOTICE, "get_block on 149 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
148 inode, inode->i_ino); 150 inode, inode->i_ino);
149 151
150 if (S_ISLNK(inode->i_mode)) { 152 if (S_ISLNK(inode->i_mode)) {
151 /* this always does I/O for so 153 /* this always does I/O for some reason. */
152 err = ocfs2_symlink_get_block( 154 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
153 goto bail; 155 goto bail;
154 } 156 }
155 157
156 err = ocfs2_extent_map_get_blocks(inod 158 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
157 &ext 159 &ext_flags);
158 if (err) { 160 if (err) {
159 mlog(ML_ERROR, "get_blocks() f !! 161 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
160 "block: %llu\n", inode, ( !! 162 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
>> 163 (unsigned long long)p_blkno);
161 goto bail; 164 goto bail;
162 } 165 }
163 166
164 if (max_blocks < count) 167 if (max_blocks < count)
165 count = max_blocks; 168 count = max_blocks;
166 169
167 /* 170 /*
168 * ocfs2 never allocates in this funct 171 * ocfs2 never allocates in this function - the only time we
169 * need to use BH_New is when we're ex 172 * need to use BH_New is when we're extending i_size on a file
170 * system which doesn't support holes, 173 * system which doesn't support holes, in which case BH_New
171 * allows __block_write_begin() to zer 174 * allows __block_write_begin() to zero.
172 * 175 *
173 * If we see this on a sparse file sys 176 * If we see this on a sparse file system, then a truncate has
174 * raced us and removed the cluster. I 177 * raced us and removed the cluster. In this case, we clear
175 * the buffers dirty and uptodate bits 178 * the buffers dirty and uptodate bits and let the buffer code
176 * ignore it as a hole. 179 * ignore it as a hole.
177 */ 180 */
178 if (create && p_blkno == 0 && ocfs2_sp 181 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
179 clear_buffer_dirty(bh_result); 182 clear_buffer_dirty(bh_result);
180 clear_buffer_uptodate(bh_resul 183 clear_buffer_uptodate(bh_result);
181 goto bail; 184 goto bail;
182 } 185 }
183 186
184 /* Treat the unwritten extent as a hol 187 /* Treat the unwritten extent as a hole for zeroing purposes. */
185 if (p_blkno && !(ext_flags & OCFS2_EXT 188 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
186 map_bh(bh_result, inode->i_sb, 189 map_bh(bh_result, inode->i_sb, p_blkno);
187 190
188 bh_result->b_size = count << inode->i_ 191 bh_result->b_size = count << inode->i_blkbits;
189 192
190 if (!ocfs2_sparse_alloc(osb)) { 193 if (!ocfs2_sparse_alloc(osb)) {
191 if (p_blkno == 0) { 194 if (p_blkno == 0) {
192 err = -EIO; 195 err = -EIO;
193 mlog(ML_ERROR, 196 mlog(ML_ERROR,
194 "iblock = %llu p_ 197 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
195 (unsigned long lo 198 (unsigned long long)iblock,
196 (unsigned long lo 199 (unsigned long long)p_blkno,
197 (unsigned long lo 200 (unsigned long long)OCFS2_I(inode)->ip_blkno);
198 mlog(ML_ERROR, "Size % 201 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
199 dump_stack(); 202 dump_stack();
200 goto bail; 203 goto bail;
201 } 204 }
202 } 205 }
203 206
204 past_eof = ocfs2_blocks_for_bytes(inod 207 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
205 208
206 trace_ocfs2_get_block_end((unsigned lo 209 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
207 (unsigned lo 210 (unsigned long long)past_eof);
208 if (create && (iblock >= past_eof)) 211 if (create && (iblock >= past_eof))
209 set_buffer_new(bh_result); 212 set_buffer_new(bh_result);
210 213
211 bail: 214 bail:
212 if (err < 0) 215 if (err < 0)
213 err = -EIO; 216 err = -EIO;
214 217
215 return err; 218 return err;
216 } 219 }
217 220
218 int ocfs2_read_inline_data(struct inode *inode 221 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
219 struct buffer_head 222 struct buffer_head *di_bh)
220 { 223 {
221 void *kaddr; 224 void *kaddr;
222 loff_t size; 225 loff_t size;
223 struct ocfs2_dinode *di = (struct ocfs 226 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
224 227
225 if (!(le16_to_cpu(di->i_dyn_features) 228 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
226 ocfs2_error(inode->i_sb, "Inod 229 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n",
227 (unsigned long lon 230 (unsigned long long)OCFS2_I(inode)->ip_blkno);
228 return -EROFS; 231 return -EROFS;
229 } 232 }
230 233
231 size = i_size_read(inode); 234 size = i_size_read(inode);
232 235
233 if (size > PAGE_SIZE || 236 if (size > PAGE_SIZE ||
234 size > ocfs2_max_inline_data_with_ 237 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
235 ocfs2_error(inode->i_sb, 238 ocfs2_error(inode->i_sb,
236 "Inode %llu has wi 239 "Inode %llu has with inline data has bad size: %Lu\n",
237 (unsigned long lon 240 (unsigned long long)OCFS2_I(inode)->ip_blkno,
238 (unsigned long lon 241 (unsigned long long)size);
239 return -EROFS; 242 return -EROFS;
240 } 243 }
241 244
242 kaddr = kmap_atomic(page); 245 kaddr = kmap_atomic(page);
243 if (size) 246 if (size)
244 memcpy(kaddr, di->id2.i_data.i 247 memcpy(kaddr, di->id2.i_data.id_data, size);
245 /* Clear the remaining part of the pag 248 /* Clear the remaining part of the page */
246 memset(kaddr + size, 0, PAGE_SIZE - si 249 memset(kaddr + size, 0, PAGE_SIZE - size);
247 flush_dcache_page(page); 250 flush_dcache_page(page);
248 kunmap_atomic(kaddr); 251 kunmap_atomic(kaddr);
249 252
250 SetPageUptodate(page); 253 SetPageUptodate(page);
251 254
252 return 0; 255 return 0;
253 } 256 }
254 257
255 static int ocfs2_readpage_inline(struct inode 258 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
256 { 259 {
257 int ret; 260 int ret;
258 struct buffer_head *di_bh = NULL; 261 struct buffer_head *di_bh = NULL;
259 262
260 BUG_ON(!PageLocked(page)); 263 BUG_ON(!PageLocked(page));
261 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 264 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
262 265
263 ret = ocfs2_read_inode_block(inode, &d 266 ret = ocfs2_read_inode_block(inode, &di_bh);
264 if (ret) { 267 if (ret) {
265 mlog_errno(ret); 268 mlog_errno(ret);
266 goto out; 269 goto out;
267 } 270 }
268 271
269 ret = ocfs2_read_inline_data(inode, pa 272 ret = ocfs2_read_inline_data(inode, page, di_bh);
270 out: 273 out:
271 unlock_page(page); 274 unlock_page(page);
272 275
273 brelse(di_bh); 276 brelse(di_bh);
274 return ret; 277 return ret;
275 } 278 }
276 279
277 static int ocfs2_read_folio(struct file *file, !! 280 static int ocfs2_readpage(struct file *file, struct page *page)
278 { 281 {
279 struct inode *inode = folio->mapping-> !! 282 struct inode *inode = page->mapping->host;
280 struct ocfs2_inode_info *oi = OCFS2_I( 283 struct ocfs2_inode_info *oi = OCFS2_I(inode);
281 loff_t start = folio_pos(folio); !! 284 loff_t start = (loff_t)page->index << PAGE_SHIFT;
282 int ret, unlock = 1; 285 int ret, unlock = 1;
283 286
284 trace_ocfs2_readpage((unsigned long lo !! 287 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
>> 288 (page ? page->index : 0));
285 289
286 ret = ocfs2_inode_lock_with_page(inode !! 290 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
287 if (ret != 0) { 291 if (ret != 0) {
288 if (ret == AOP_TRUNCATED_PAGE) 292 if (ret == AOP_TRUNCATED_PAGE)
289 unlock = 0; 293 unlock = 0;
290 mlog_errno(ret); 294 mlog_errno(ret);
291 goto out; 295 goto out;
292 } 296 }
293 297
294 if (down_read_trylock(&oi->ip_alloc_se 298 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
295 /* 299 /*
296 * Unlock the folio and cycle !! 300 * Unlock the page and cycle ip_alloc_sem so that we don't
297 * busyloop waiting for ip_all 301 * busyloop waiting for ip_alloc_sem to unlock
298 */ 302 */
299 ret = AOP_TRUNCATED_PAGE; 303 ret = AOP_TRUNCATED_PAGE;
300 folio_unlock(folio); !! 304 unlock_page(page);
301 unlock = 0; 305 unlock = 0;
302 down_read(&oi->ip_alloc_sem); 306 down_read(&oi->ip_alloc_sem);
303 up_read(&oi->ip_alloc_sem); 307 up_read(&oi->ip_alloc_sem);
304 goto out_inode_unlock; 308 goto out_inode_unlock;
305 } 309 }
306 310
307 /* 311 /*
308 * i_size might have just been updated 312 * i_size might have just been updated as we grabed the meta lock. We
309 * might now be discovering a truncate 313 * might now be discovering a truncate that hit on another node.
310 * block_read_full_folio->get_block fr !! 314 * block_read_full_page->get_block freaks out if it is asked to read
311 * beyond the end of a file, so we che 315 * beyond the end of a file, so we check here. Callers
312 * (generic_file_read, vm_ops->fault) 316 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
313 * and notice that the folio they just !! 317 * and notice that the page they just read isn't needed.
314 * 318 *
315 * XXX sys_readahead() seems to get th 319 * XXX sys_readahead() seems to get that wrong?
316 */ 320 */
317 if (start >= i_size_read(inode)) { 321 if (start >= i_size_read(inode)) {
318 folio_zero_segment(folio, 0, f !! 322 zero_user(page, 0, PAGE_SIZE);
319 folio_mark_uptodate(folio); !! 323 SetPageUptodate(page);
320 ret = 0; 324 ret = 0;
321 goto out_alloc; 325 goto out_alloc;
322 } 326 }
323 327
324 if (oi->ip_dyn_features & OCFS2_INLINE 328 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
325 ret = ocfs2_readpage_inline(in !! 329 ret = ocfs2_readpage_inline(inode, page);
326 else 330 else
327 ret = block_read_full_folio(fo !! 331 ret = block_read_full_page(page, ocfs2_get_block);
328 unlock = 0; 332 unlock = 0;
329 333
330 out_alloc: 334 out_alloc:
331 up_read(&oi->ip_alloc_sem); 335 up_read(&oi->ip_alloc_sem);
332 out_inode_unlock: 336 out_inode_unlock:
333 ocfs2_inode_unlock(inode, 0); 337 ocfs2_inode_unlock(inode, 0);
334 out: 338 out:
335 if (unlock) 339 if (unlock)
336 folio_unlock(folio); !! 340 unlock_page(page);
337 return ret; 341 return ret;
338 } 342 }
339 343
340 /* 344 /*
341 * This is used only for read-ahead. Failures 345 * This is used only for read-ahead. Failures or difficult to handle
342 * situations are safe to ignore. 346 * situations are safe to ignore.
343 * 347 *
344 * Right now, we don't bother with BH_Boundary 348 * Right now, we don't bother with BH_Boundary - in-inode extent lists
345 * are quite large (243 extents on 4k blocks), 349 * are quite large (243 extents on 4k blocks), so most inodes don't
346 * grow out to a tree. If need be, detecting b 350 * grow out to a tree. If need be, detecting boundary extents could
347 * trivially be added in a future version of o 351 * trivially be added in a future version of ocfs2_get_block().
348 */ 352 */
349 static void ocfs2_readahead(struct readahead_c !! 353 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
>> 354 struct list_head *pages, unsigned nr_pages)
350 { 355 {
351 int ret; !! 356 int ret, err = -EIO;
352 struct inode *inode = rac->mapping->ho !! 357 struct inode *inode = mapping->host;
353 struct ocfs2_inode_info *oi = OCFS2_I( 358 struct ocfs2_inode_info *oi = OCFS2_I(inode);
>> 359 loff_t start;
>> 360 struct page *last;
354 361
355 /* 362 /*
356 * Use the nonblocking flag for the dl 363 * Use the nonblocking flag for the dlm code to avoid page
357 * lock inversion, but don't bother wi 364 * lock inversion, but don't bother with retrying.
358 */ 365 */
359 ret = ocfs2_inode_lock_full(inode, NUL 366 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
360 if (ret) 367 if (ret)
361 return; !! 368 return err;
362 369
363 if (down_read_trylock(&oi->ip_alloc_se !! 370 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
364 goto out_unlock; !! 371 ocfs2_inode_unlock(inode, 0);
>> 372 return err;
>> 373 }
365 374
366 /* 375 /*
367 * Don't bother with inline-data. Ther 376 * Don't bother with inline-data. There isn't anything
368 * to read-ahead in that case anyway.. 377 * to read-ahead in that case anyway...
369 */ 378 */
370 if (oi->ip_dyn_features & OCFS2_INLINE 379 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
371 goto out_up; !! 380 goto out_unlock;
372 381
373 /* 382 /*
374 * Check whether a remote node truncat 383 * Check whether a remote node truncated this file - we just
375 * drop out in that case as it's not w 384 * drop out in that case as it's not worth handling here.
376 */ 385 */
377 if (readahead_pos(rac) >= i_size_read( !! 386 last = lru_to_page(pages);
378 goto out_up; !! 387 start = (loff_t)last->index << PAGE_SHIFT;
>> 388 if (start >= i_size_read(inode))
>> 389 goto out_unlock;
379 390
380 mpage_readahead(rac, ocfs2_get_block); !! 391 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
381 392
382 out_up: <<
383 up_read(&oi->ip_alloc_sem); <<
384 out_unlock: 393 out_unlock:
>> 394 up_read(&oi->ip_alloc_sem);
385 ocfs2_inode_unlock(inode, 0); 395 ocfs2_inode_unlock(inode, 0);
>> 396
>> 397 return err;
386 } 398 }
387 399
388 /* Note: Because we don't support holes, our a 400 /* Note: Because we don't support holes, our allocation has
389 * already happened (allocation writes zeros t 401 * already happened (allocation writes zeros to the file data)
390 * so we don't have to worry about ordered wri 402 * so we don't have to worry about ordered writes in
391 * ocfs2_writepages. !! 403 * ocfs2_writepage.
392 * 404 *
393 * ->writepages is called during the process o !! 405 * ->writepage is called during the process of invalidating the page cache
394 * during blocked lock processing. It can't b 406 * during blocked lock processing. It can't block on any cluster locks
395 * to during block mapping. It's relying on t 407 * to during block mapping. It's relying on the fact that the block
396 * mapping can't have disappeared under the di 408 * mapping can't have disappeared under the dirty pages that it is
397 * being asked to write back. 409 * being asked to write back.
398 */ 410 */
399 static int ocfs2_writepages(struct address_spa !! 411 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
400 struct writeback_control *wbc) <<
401 { 412 {
402 return mpage_writepages(mapping, wbc, !! 413 trace_ocfs2_writepage(
>> 414 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
>> 415 page->index);
>> 416
>> 417 return block_write_full_page(page, ocfs2_get_block, wbc);
403 } 418 }
404 419
405 /* Taken from ext3. We don't necessarily need 420 /* Taken from ext3. We don't necessarily need the full blown
406 * functionality yet, but IMHO it's better to 421 * functionality yet, but IMHO it's better to cut and paste the whole
407 * thing so we can avoid introducing our own b 422 * thing so we can avoid introducing our own bugs (and easily pick up
408 * their fixes when they happen) --Mark */ 423 * their fixes when they happen) --Mark */
409 int walk_page_buffers( handle_t *handle, 424 int walk_page_buffers( handle_t *handle,
410 struct buffer_head *he 425 struct buffer_head *head,
411 unsigned from, 426 unsigned from,
412 unsigned to, 427 unsigned to,
413 int *partial, 428 int *partial,
414 int (*fn)( handle 429 int (*fn)( handle_t *handle,
415 struct 430 struct buffer_head *bh))
416 { 431 {
417 struct buffer_head *bh; 432 struct buffer_head *bh;
418 unsigned block_start, block_end; 433 unsigned block_start, block_end;
419 unsigned blocksize = head->b_size; 434 unsigned blocksize = head->b_size;
420 int err, ret = 0; 435 int err, ret = 0;
421 struct buffer_head *next; 436 struct buffer_head *next;
422 437
423 for ( bh = head, block_start = 0; 438 for ( bh = head, block_start = 0;
424 ret == 0 && (bh != head || !bl 439 ret == 0 && (bh != head || !block_start);
425 block_start = block_end, bh = 440 block_start = block_end, bh = next)
426 { 441 {
427 next = bh->b_this_page; 442 next = bh->b_this_page;
428 block_end = block_start + bloc 443 block_end = block_start + blocksize;
429 if (block_end <= from || block 444 if (block_end <= from || block_start >= to) {
430 if (partial && !buffer 445 if (partial && !buffer_uptodate(bh))
431 *partial = 1; 446 *partial = 1;
432 continue; 447 continue;
433 } 448 }
434 err = (*fn)(handle, bh); 449 err = (*fn)(handle, bh);
435 if (!ret) 450 if (!ret)
436 ret = err; 451 ret = err;
437 } 452 }
438 return ret; 453 return ret;
439 } 454 }
440 455
441 static sector_t ocfs2_bmap(struct address_spac 456 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
442 { 457 {
443 sector_t status; 458 sector_t status;
444 u64 p_blkno = 0; 459 u64 p_blkno = 0;
445 int err = 0; 460 int err = 0;
446 struct inode *inode = mapping->host; 461 struct inode *inode = mapping->host;
447 462
448 trace_ocfs2_bmap((unsigned long long)O 463 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
449 (unsigned long long)b 464 (unsigned long long)block);
450 465
451 /* 466 /*
452 * The swap code (ab-)uses ->bmap to g 467 * The swap code (ab-)uses ->bmap to get a block mapping and then
453 * bypasseѕ the file system for actua 468 * bypasseѕ the file system for actual I/O. We really can't allow
454 * that on refcounted inodes, so we ha 469 * that on refcounted inodes, so we have to skip out here. And yes,
455 * 0 is the magic code for a bmap erro 470 * 0 is the magic code for a bmap error..
456 */ 471 */
457 if (ocfs2_is_refcount_inode(inode)) 472 if (ocfs2_is_refcount_inode(inode))
458 return 0; 473 return 0;
459 474
460 /* We don't need to lock journal syste 475 /* We don't need to lock journal system files, since they aren't
461 * accessed concurrently from multiple 476 * accessed concurrently from multiple nodes.
462 */ 477 */
463 if (!INODE_JOURNAL(inode)) { 478 if (!INODE_JOURNAL(inode)) {
464 err = ocfs2_inode_lock(inode, 479 err = ocfs2_inode_lock(inode, NULL, 0);
465 if (err) { 480 if (err) {
466 if (err != -ENOENT) 481 if (err != -ENOENT)
467 mlog_errno(err 482 mlog_errno(err);
468 goto bail; 483 goto bail;
469 } 484 }
470 down_read(&OCFS2_I(inode)->ip_ 485 down_read(&OCFS2_I(inode)->ip_alloc_sem);
471 } 486 }
472 487
473 if (!(OCFS2_I(inode)->ip_dyn_features 488 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
474 err = ocfs2_extent_map_get_blo 489 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
475 490 NULL);
476 491
477 if (!INODE_JOURNAL(inode)) { 492 if (!INODE_JOURNAL(inode)) {
478 up_read(&OCFS2_I(inode)->ip_al 493 up_read(&OCFS2_I(inode)->ip_alloc_sem);
479 ocfs2_inode_unlock(inode, 0); 494 ocfs2_inode_unlock(inode, 0);
480 } 495 }
481 496
482 if (err) { 497 if (err) {
483 mlog(ML_ERROR, "get_blocks() f 498 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
484 (unsigned long long)block 499 (unsigned long long)block);
485 mlog_errno(err); 500 mlog_errno(err);
486 goto bail; 501 goto bail;
487 } 502 }
488 503
489 bail: 504 bail:
490 status = err ? 0 : p_blkno; 505 status = err ? 0 : p_blkno;
491 506
492 return status; 507 return status;
493 } 508 }
494 509
495 static bool ocfs2_release_folio(struct folio * !! 510 static int ocfs2_releasepage(struct page *page, gfp_t wait)
496 { 511 {
497 if (!folio_buffers(folio)) !! 512 if (!page_has_buffers(page))
498 return false; !! 513 return 0;
499 return try_to_free_buffers(folio); !! 514 return try_to_free_buffers(page);
500 } 515 }
501 516
502 static void ocfs2_figure_cluster_boundaries(st 517 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
503 u3 518 u32 cpos,
504 un 519 unsigned int *start,
505 un 520 unsigned int *end)
506 { 521 {
507 unsigned int cluster_start = 0, cluste 522 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE;
508 523
509 if (unlikely(PAGE_SHIFT > osb->s_clust 524 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) {
510 unsigned int cpp; 525 unsigned int cpp;
511 526
512 cpp = 1 << (PAGE_SHIFT - osb-> 527 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits);
513 528
514 cluster_start = cpos % cpp; 529 cluster_start = cpos % cpp;
515 cluster_start = cluster_start 530 cluster_start = cluster_start << osb->s_clustersize_bits;
516 531
517 cluster_end = cluster_start + 532 cluster_end = cluster_start + osb->s_clustersize;
518 } 533 }
519 534
520 BUG_ON(cluster_start > PAGE_SIZE); 535 BUG_ON(cluster_start > PAGE_SIZE);
521 BUG_ON(cluster_end > PAGE_SIZE); 536 BUG_ON(cluster_end > PAGE_SIZE);
522 537
523 if (start) 538 if (start)
524 *start = cluster_start; 539 *start = cluster_start;
525 if (end) 540 if (end)
526 *end = cluster_end; 541 *end = cluster_end;
527 } 542 }
528 543
529 /* 544 /*
530 * 'from' and 'to' are the region in the page 545 * 'from' and 'to' are the region in the page to avoid zeroing.
531 * 546 *
532 * If pagesize > clustersize, this function wi 547 * If pagesize > clustersize, this function will avoid zeroing outside
533 * of the cluster boundary. 548 * of the cluster boundary.
534 * 549 *
535 * from == to == 0 is code for "zero the entir 550 * from == to == 0 is code for "zero the entire cluster region"
536 */ 551 */
537 static void ocfs2_clear_page_regions(struct pa 552 static void ocfs2_clear_page_regions(struct page *page,
538 struct oc 553 struct ocfs2_super *osb, u32 cpos,
539 unsigned 554 unsigned from, unsigned to)
540 { 555 {
541 void *kaddr; 556 void *kaddr;
542 unsigned int cluster_start, cluster_en 557 unsigned int cluster_start, cluster_end;
543 558
544 ocfs2_figure_cluster_boundaries(osb, c 559 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
545 560
546 kaddr = kmap_atomic(page); 561 kaddr = kmap_atomic(page);
547 562
548 if (from || to) { 563 if (from || to) {
549 if (from > cluster_start) 564 if (from > cluster_start)
550 memset(kaddr + cluster 565 memset(kaddr + cluster_start, 0, from - cluster_start);
551 if (to < cluster_end) 566 if (to < cluster_end)
552 memset(kaddr + to, 0, 567 memset(kaddr + to, 0, cluster_end - to);
553 } else { 568 } else {
554 memset(kaddr + cluster_start, 569 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
555 } 570 }
556 571
557 kunmap_atomic(kaddr); 572 kunmap_atomic(kaddr);
558 } 573 }
559 574
560 /* 575 /*
561 * Nonsparse file systems fully allocate befor 576 * Nonsparse file systems fully allocate before we get to the write
562 * code. This prevents ocfs2_write() from tagg 577 * code. This prevents ocfs2_write() from tagging the write as an
563 * allocating one, which means ocfs2_map_page_ 578 * allocating one, which means ocfs2_map_page_blocks() might try to
564 * read-in the blocks at the tail of our file. 579 * read-in the blocks at the tail of our file. Avoid reading them by
565 * testing i_size against each block offset. 580 * testing i_size against each block offset.
566 */ 581 */
567 static int ocfs2_should_read_blk(struct inode !! 582 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
568 unsigned int 583 unsigned int block_start)
569 { 584 {
570 u64 offset = folio_pos(folio) + block_ !! 585 u64 offset = page_offset(page) + block_start;
571 586
572 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 587 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
573 return 1; 588 return 1;
574 589
575 if (i_size_read(inode) > offset) 590 if (i_size_read(inode) > offset)
576 return 1; 591 return 1;
577 592
578 return 0; 593 return 0;
579 } 594 }
580 595
581 /* 596 /*
582 * Some of this taken from __block_write_begin 597 * Some of this taken from __block_write_begin(). We already have our
583 * mapping by now though, and the entire write 598 * mapping by now though, and the entire write will be allocating or
584 * it won't, so not much need to use BH_New. 599 * it won't, so not much need to use BH_New.
585 * 600 *
586 * This will also skip zeroing, which is handl 601 * This will also skip zeroing, which is handled externally.
587 */ 602 */
588 int ocfs2_map_page_blocks(struct page *page, u 603 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
589 struct inode *inode, 604 struct inode *inode, unsigned int from,
590 unsigned int to, int 605 unsigned int to, int new)
591 { 606 {
592 struct folio *folio = page_folio(page) <<
593 int ret = 0; 607 int ret = 0;
594 struct buffer_head *head, *bh, *wait[2 608 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
595 unsigned int block_end, block_start; 609 unsigned int block_end, block_start;
596 unsigned int bsize = i_blocksize(inode 610 unsigned int bsize = i_blocksize(inode);
597 611
598 head = folio_buffers(folio); !! 612 if (!page_has_buffers(page))
599 if (!head) !! 613 create_empty_buffers(page, bsize, 0);
600 head = create_empty_buffers(fo <<
601 614
>> 615 head = page_buffers(page);
602 for (bh = head, block_start = 0; bh != 616 for (bh = head, block_start = 0; bh != head || !block_start;
603 bh = bh->b_this_page, block_start 617 bh = bh->b_this_page, block_start += bsize) {
604 block_end = block_start + bsiz 618 block_end = block_start + bsize;
605 619
606 clear_buffer_new(bh); 620 clear_buffer_new(bh);
607 621
608 /* 622 /*
609 * Ignore blocks outside of ou 623 * Ignore blocks outside of our i/o range -
610 * they may belong to unalloca 624 * they may belong to unallocated clusters.
611 */ 625 */
612 if (block_start >= to || block 626 if (block_start >= to || block_end <= from) {
613 if (folio_test_uptodat !! 627 if (PageUptodate(page))
614 set_buffer_upt 628 set_buffer_uptodate(bh);
615 continue; 629 continue;
616 } 630 }
617 631
618 /* 632 /*
619 * For an allocating write wit 633 * For an allocating write with cluster size >= page
620 * size, we always write the e 634 * size, we always write the entire page.
621 */ 635 */
622 if (new) 636 if (new)
623 set_buffer_new(bh); 637 set_buffer_new(bh);
624 638
625 if (!buffer_mapped(bh)) { 639 if (!buffer_mapped(bh)) {
626 map_bh(bh, inode->i_sb 640 map_bh(bh, inode->i_sb, *p_blkno);
627 clean_bdev_bh_alias(bh 641 clean_bdev_bh_alias(bh);
628 } 642 }
629 643
630 if (folio_test_uptodate(folio) !! 644 if (PageUptodate(page)) {
631 set_buffer_uptodate(bh !! 645 if (!buffer_uptodate(bh))
>> 646 set_buffer_uptodate(bh);
632 } else if (!buffer_uptodate(bh 647 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
633 !buffer_new(bh) && 648 !buffer_new(bh) &&
634 ocfs2_should_read_b !! 649 ocfs2_should_read_blk(inode, page, block_start) &&
635 (block_start < from 650 (block_start < from || block_end > to)) {
636 bh_read_nowait(bh, 0); !! 651 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
637 *wait_bh++=bh; 652 *wait_bh++=bh;
638 } 653 }
639 654
640 *p_blkno = *p_blkno + 1; 655 *p_blkno = *p_blkno + 1;
641 } 656 }
642 657
643 /* 658 /*
644 * If we issued read requests - let th 659 * If we issued read requests - let them complete.
645 */ 660 */
646 while(wait_bh > wait) { 661 while(wait_bh > wait) {
647 wait_on_buffer(*--wait_bh); 662 wait_on_buffer(*--wait_bh);
648 if (!buffer_uptodate(*wait_bh) 663 if (!buffer_uptodate(*wait_bh))
649 ret = -EIO; 664 ret = -EIO;
650 } 665 }
651 666
652 if (ret == 0 || !new) 667 if (ret == 0 || !new)
653 return ret; 668 return ret;
654 669
655 /* 670 /*
656 * If we get -EIO above, zero out any 671 * If we get -EIO above, zero out any newly allocated blocks
657 * to avoid exposing stale data. 672 * to avoid exposing stale data.
658 */ 673 */
659 bh = head; 674 bh = head;
660 block_start = 0; 675 block_start = 0;
661 do { 676 do {
662 block_end = block_start + bsiz 677 block_end = block_start + bsize;
663 if (block_end <= from) 678 if (block_end <= from)
664 goto next_bh; 679 goto next_bh;
665 if (block_start >= to) 680 if (block_start >= to)
666 break; 681 break;
667 682
668 folio_zero_range(folio, block_ !! 683 zero_user(page, block_start, bh->b_size);
669 set_buffer_uptodate(bh); 684 set_buffer_uptodate(bh);
670 mark_buffer_dirty(bh); 685 mark_buffer_dirty(bh);
671 686
672 next_bh: 687 next_bh:
673 block_start = block_end; 688 block_start = block_end;
674 bh = bh->b_this_page; 689 bh = bh->b_this_page;
675 } while (bh != head); 690 } while (bh != head);
676 691
677 return ret; 692 return ret;
678 } 693 }
679 694
680 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 695 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
681 #define OCFS2_MAX_CTXT_PAGES 1 696 #define OCFS2_MAX_CTXT_PAGES 1
682 #else 697 #else
683 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 698 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE)
684 #endif 699 #endif
685 700
686 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 701 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE)
687 702
688 struct ocfs2_unwritten_extent { 703 struct ocfs2_unwritten_extent {
689 struct list_head ue_node; 704 struct list_head ue_node;
690 struct list_head ue_ip_node; 705 struct list_head ue_ip_node;
691 u32 ue_cpos; 706 u32 ue_cpos;
692 u32 ue_phys; 707 u32 ue_phys;
693 }; 708 };
694 709
695 /* 710 /*
696 * Describe the state of a single cluster to b 711 * Describe the state of a single cluster to be written to.
697 */ 712 */
698 struct ocfs2_write_cluster_desc { 713 struct ocfs2_write_cluster_desc {
699 u32 c_cpos; 714 u32 c_cpos;
700 u32 c_phys; 715 u32 c_phys;
701 /* 716 /*
702 * Give this a unique field because c_ 717 * Give this a unique field because c_phys eventually gets
703 * filled. 718 * filled.
704 */ 719 */
705 unsigned c_new; 720 unsigned c_new;
706 unsigned c_clear_unwritten; 721 unsigned c_clear_unwritten;
707 unsigned c_needs_zero; 722 unsigned c_needs_zero;
708 }; 723 };
709 724
710 struct ocfs2_write_ctxt { 725 struct ocfs2_write_ctxt {
711 /* Logical cluster position / len of w 726 /* Logical cluster position / len of write */
712 u32 w_cpos 727 u32 w_cpos;
713 u32 w_clen 728 u32 w_clen;
714 729
715 /* First cluster allocated in a nonspa 730 /* First cluster allocated in a nonsparse extend */
716 u32 w_firs 731 u32 w_first_new_cpos;
717 732
718 /* Type of caller. Must be one of buff 733 /* Type of caller. Must be one of buffer, mmap, direct. */
719 ocfs2_write_type_t w_type 734 ocfs2_write_type_t w_type;
720 735
721 struct ocfs2_write_cluster_desc w_desc 736 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
722 737
723 /* 738 /*
724 * This is true if page_size > cluster 739 * This is true if page_size > cluster_size.
725 * 740 *
726 * It triggers a set of special cases 741 * It triggers a set of special cases during write which might
727 * have to deal with allocating writes 742 * have to deal with allocating writes to partial pages.
728 */ 743 */
729 unsigned int w_larg 744 unsigned int w_large_pages;
730 745
731 /* 746 /*
732 * Pages involved in this write. 747 * Pages involved in this write.
733 * 748 *
734 * w_target_page is the page being wri 749 * w_target_page is the page being written to by the user.
735 * 750 *
736 * w_pages is an array of pages which 751 * w_pages is an array of pages which always contains
737 * w_target_page, and in the case of a 752 * w_target_page, and in the case of an allocating write with
738 * page_size < cluster size, it will c 753 * page_size < cluster size, it will contain zero'd and mapped
739 * pages adjacent to w_target_page whi 754 * pages adjacent to w_target_page which need to be written
740 * out in so that future reads from th 755 * out in so that future reads from that region will get
741 * zero's. 756 * zero's.
742 */ 757 */
743 unsigned int w_num_ 758 unsigned int w_num_pages;
744 struct page *w_pag 759 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
745 struct page *w_tar 760 struct page *w_target_page;
746 761
747 /* 762 /*
748 * w_target_locked is used for page_mk 763 * w_target_locked is used for page_mkwrite path indicating no unlocking
749 * against w_target_page in ocfs2_writ 764 * against w_target_page in ocfs2_write_end_nolock.
750 */ 765 */
751 unsigned int w_targ 766 unsigned int w_target_locked:1;
752 767
753 /* 768 /*
754 * ocfs2_write_end() uses this to know 769 * ocfs2_write_end() uses this to know what the real range to
755 * write in the target should be. 770 * write in the target should be.
756 */ 771 */
757 unsigned int w_targ 772 unsigned int w_target_from;
758 unsigned int w_targ 773 unsigned int w_target_to;
759 774
760 /* 775 /*
761 * We could use journal_current_handle 776 * We could use journal_current_handle() but this is cleaner,
762 * IMHO -Mark 777 * IMHO -Mark
763 */ 778 */
764 handle_t *w_han 779 handle_t *w_handle;
765 780
766 struct buffer_head *w_di_ 781 struct buffer_head *w_di_bh;
767 782
768 struct ocfs2_cached_dealloc_ctxt w_dea 783 struct ocfs2_cached_dealloc_ctxt w_dealloc;
769 784
770 struct list_head w_unwr 785 struct list_head w_unwritten_list;
771 unsigned int w_unwr 786 unsigned int w_unwritten_count;
772 }; 787 };
773 788
774 void ocfs2_unlock_and_free_pages(struct page * 789 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
775 { 790 {
776 int i; 791 int i;
777 792
778 for(i = 0; i < num_pages; i++) { 793 for(i = 0; i < num_pages; i++) {
779 if (pages[i]) { 794 if (pages[i]) {
780 unlock_page(pages[i]); 795 unlock_page(pages[i]);
781 mark_page_accessed(pag 796 mark_page_accessed(pages[i]);
782 put_page(pages[i]); 797 put_page(pages[i]);
783 } 798 }
784 } 799 }
785 } 800 }
786 801
787 static void ocfs2_unlock_pages(struct ocfs2_wr 802 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
788 { 803 {
789 int i; 804 int i;
790 805
791 /* 806 /*
792 * w_target_locked is only set to true 807 * w_target_locked is only set to true in the page_mkwrite() case.
793 * The intent is to allow us to lock t 808 * The intent is to allow us to lock the target page from write_begin()
794 * to write_end(). The caller must hol 809 * to write_end(). The caller must hold a ref on w_target_page.
795 */ 810 */
796 if (wc->w_target_locked) { 811 if (wc->w_target_locked) {
797 BUG_ON(!wc->w_target_page); 812 BUG_ON(!wc->w_target_page);
798 for (i = 0; i < wc->w_num_page 813 for (i = 0; i < wc->w_num_pages; i++) {
799 if (wc->w_target_page 814 if (wc->w_target_page == wc->w_pages[i]) {
800 wc->w_pages[i] 815 wc->w_pages[i] = NULL;
801 break; 816 break;
802 } 817 }
803 } 818 }
804 mark_page_accessed(wc->w_targe 819 mark_page_accessed(wc->w_target_page);
805 put_page(wc->w_target_page); 820 put_page(wc->w_target_page);
806 } 821 }
807 ocfs2_unlock_and_free_pages(wc->w_page 822 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
808 } 823 }
809 824
810 static void ocfs2_free_unwritten_list(struct i 825 static void ocfs2_free_unwritten_list(struct inode *inode,
811 struct list_h 826 struct list_head *head)
812 { 827 {
813 struct ocfs2_inode_info *oi = OCFS2_I( 828 struct ocfs2_inode_info *oi = OCFS2_I(inode);
814 struct ocfs2_unwritten_extent *ue = NU 829 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL;
815 830
816 list_for_each_entry_safe(ue, tmp, head 831 list_for_each_entry_safe(ue, tmp, head, ue_node) {
817 list_del(&ue->ue_node); 832 list_del(&ue->ue_node);
818 spin_lock(&oi->ip_lock); 833 spin_lock(&oi->ip_lock);
819 list_del(&ue->ue_ip_node); 834 list_del(&ue->ue_ip_node);
820 spin_unlock(&oi->ip_lock); 835 spin_unlock(&oi->ip_lock);
821 kfree(ue); 836 kfree(ue);
822 } 837 }
823 } 838 }
824 839
825 static void ocfs2_free_write_ctxt(struct inode 840 static void ocfs2_free_write_ctxt(struct inode *inode,
826 struct ocfs2 841 struct ocfs2_write_ctxt *wc)
827 { 842 {
828 ocfs2_free_unwritten_list(inode, &wc-> 843 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list);
829 ocfs2_unlock_pages(wc); 844 ocfs2_unlock_pages(wc);
830 brelse(wc->w_di_bh); 845 brelse(wc->w_di_bh);
831 kfree(wc); 846 kfree(wc);
832 } 847 }
833 848
834 static int ocfs2_alloc_write_ctxt(struct ocfs2 849 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
835 struct ocfs2 850 struct ocfs2_super *osb, loff_t pos,
836 unsigned len 851 unsigned len, ocfs2_write_type_t type,
837 struct buffe 852 struct buffer_head *di_bh)
838 { 853 {
839 u32 cend; 854 u32 cend;
840 struct ocfs2_write_ctxt *wc; 855 struct ocfs2_write_ctxt *wc;
841 856
842 wc = kzalloc(sizeof(struct ocfs2_write 857 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
843 if (!wc) 858 if (!wc)
844 return -ENOMEM; 859 return -ENOMEM;
845 860
846 wc->w_cpos = pos >> osb->s_clustersize 861 wc->w_cpos = pos >> osb->s_clustersize_bits;
847 wc->w_first_new_cpos = UINT_MAX; 862 wc->w_first_new_cpos = UINT_MAX;
848 cend = (pos + len - 1) >> osb->s_clust 863 cend = (pos + len - 1) >> osb->s_clustersize_bits;
849 wc->w_clen = cend - wc->w_cpos + 1; 864 wc->w_clen = cend - wc->w_cpos + 1;
850 get_bh(di_bh); 865 get_bh(di_bh);
851 wc->w_di_bh = di_bh; 866 wc->w_di_bh = di_bh;
852 wc->w_type = type; 867 wc->w_type = type;
853 868
854 if (unlikely(PAGE_SHIFT > osb->s_clust 869 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits))
855 wc->w_large_pages = 1; 870 wc->w_large_pages = 1;
856 else 871 else
857 wc->w_large_pages = 0; 872 wc->w_large_pages = 0;
858 873
859 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 874 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
860 INIT_LIST_HEAD(&wc->w_unwritten_list); 875 INIT_LIST_HEAD(&wc->w_unwritten_list);
861 876
862 *wcp = wc; 877 *wcp = wc;
863 878
864 return 0; 879 return 0;
865 } 880 }
866 881
867 /* 882 /*
868 * If a page has any new buffers, zero them ou 883 * If a page has any new buffers, zero them out here, and mark them uptodate
869 * and dirty so they'll be written out (in ord 884 * and dirty so they'll be written out (in order to prevent uninitialised
870 * block data from leaking). And clear the new 885 * block data from leaking). And clear the new bit.
871 */ 886 */
872 static void ocfs2_zero_new_buffers(struct page 887 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
873 { 888 {
874 unsigned int block_start, block_end; 889 unsigned int block_start, block_end;
875 struct buffer_head *head, *bh; 890 struct buffer_head *head, *bh;
876 891
877 BUG_ON(!PageLocked(page)); 892 BUG_ON(!PageLocked(page));
878 if (!page_has_buffers(page)) 893 if (!page_has_buffers(page))
879 return; 894 return;
880 895
881 bh = head = page_buffers(page); 896 bh = head = page_buffers(page);
882 block_start = 0; 897 block_start = 0;
883 do { 898 do {
884 block_end = block_start + bh-> 899 block_end = block_start + bh->b_size;
885 900
886 if (buffer_new(bh)) { 901 if (buffer_new(bh)) {
887 if (block_end > from & 902 if (block_end > from && block_start < to) {
888 if (!PageUptod 903 if (!PageUptodate(page)) {
889 unsign 904 unsigned start, end;
890 905
891 start 906 start = max(from, block_start);
892 end = 907 end = min(to, block_end);
893 908
894 zero_u 909 zero_user_segment(page, start, end);
895 set_bu 910 set_buffer_uptodate(bh);
896 } 911 }
897 912
898 clear_buffer_n 913 clear_buffer_new(bh);
899 mark_buffer_di 914 mark_buffer_dirty(bh);
900 } 915 }
901 } 916 }
902 917
903 block_start = block_end; 918 block_start = block_end;
904 bh = bh->b_this_page; 919 bh = bh->b_this_page;
905 } while (bh != head); 920 } while (bh != head);
906 } 921 }
907 922
908 /* 923 /*
909 * Only called when we have a failure during a 924 * Only called when we have a failure during allocating write to write
910 * zero's to the newly allocated region. 925 * zero's to the newly allocated region.
911 */ 926 */
912 static void ocfs2_write_failure(struct inode * 927 static void ocfs2_write_failure(struct inode *inode,
913 struct ocfs2_w 928 struct ocfs2_write_ctxt *wc,
914 loff_t user_po 929 loff_t user_pos, unsigned user_len)
915 { 930 {
916 int i; 931 int i;
917 unsigned from = user_pos & (PAGE_SIZE 932 unsigned from = user_pos & (PAGE_SIZE - 1),
918 to = user_pos + user_len; 933 to = user_pos + user_len;
919 struct page *tmppage; 934 struct page *tmppage;
920 935
921 if (wc->w_target_page) 936 if (wc->w_target_page)
922 ocfs2_zero_new_buffers(wc->w_t 937 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
923 938
924 for(i = 0; i < wc->w_num_pages; i++) { 939 for(i = 0; i < wc->w_num_pages; i++) {
925 tmppage = wc->w_pages[i]; 940 tmppage = wc->w_pages[i];
926 941
927 if (tmppage && page_has_buffer 942 if (tmppage && page_has_buffers(tmppage)) {
928 if (ocfs2_should_order 943 if (ocfs2_should_order_data(inode))
929 ocfs2_jbd2_ino 944 ocfs2_jbd2_inode_add_write(wc->w_handle, inode,
930 945 user_pos, user_len);
931 946
932 block_commit_write(tmp 947 block_commit_write(tmppage, from, to);
933 } 948 }
934 } 949 }
935 } 950 }
936 951
937 static int ocfs2_prepare_page_for_write(struct 952 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
938 struct 953 struct ocfs2_write_ctxt *wc,
939 struct 954 struct page *page, u32 cpos,
940 loff_t 955 loff_t user_pos, unsigned user_len,
941 int ne 956 int new)
942 { 957 {
943 int ret; 958 int ret;
944 unsigned int map_from = 0, map_to = 0; 959 unsigned int map_from = 0, map_to = 0;
945 unsigned int cluster_start, cluster_en 960 unsigned int cluster_start, cluster_end;
946 unsigned int user_data_from = 0, user_ 961 unsigned int user_data_from = 0, user_data_to = 0;
947 962
948 ocfs2_figure_cluster_boundaries(OCFS2_ 963 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
949 &clust 964 &cluster_start, &cluster_end);
950 965
951 /* treat the write as new if the a hol 966 /* treat the write as new if the a hole/lseek spanned across
952 * the page boundary. 967 * the page boundary.
953 */ 968 */
954 new = new | ((i_size_read(inode) <= pa 969 new = new | ((i_size_read(inode) <= page_offset(page)) &&
955 (page_offset(page) <= 970 (page_offset(page) <= user_pos));
956 971
957 if (page == wc->w_target_page) { 972 if (page == wc->w_target_page) {
958 map_from = user_pos & (PAGE_SI 973 map_from = user_pos & (PAGE_SIZE - 1);
959 map_to = map_from + user_len; 974 map_to = map_from + user_len;
960 975
961 if (new) 976 if (new)
962 ret = ocfs2_map_page_b 977 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
963 978 cluster_start, cluster_end,
964 979 new);
965 else 980 else
966 ret = ocfs2_map_page_b 981 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
967 982 map_from, map_to, new);
968 if (ret) { 983 if (ret) {
969 mlog_errno(ret); 984 mlog_errno(ret);
970 goto out; 985 goto out;
971 } 986 }
972 987
973 user_data_from = map_from; 988 user_data_from = map_from;
974 user_data_to = map_to; 989 user_data_to = map_to;
975 if (new) { 990 if (new) {
976 map_from = cluster_sta 991 map_from = cluster_start;
977 map_to = cluster_end; 992 map_to = cluster_end;
978 } 993 }
979 } else { 994 } else {
980 /* 995 /*
981 * If we haven't allocated the 996 * If we haven't allocated the new page yet, we
982 * shouldn't be writing it out 997 * shouldn't be writing it out without copying user
983 * data. This is likely a math 998 * data. This is likely a math error from the caller.
984 */ 999 */
985 BUG_ON(!new); 1000 BUG_ON(!new);
986 1001
987 map_from = cluster_start; 1002 map_from = cluster_start;
988 map_to = cluster_end; 1003 map_to = cluster_end;
989 1004
990 ret = ocfs2_map_page_blocks(pa 1005 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
991 cl 1006 cluster_start, cluster_end, new);
992 if (ret) { 1007 if (ret) {
993 mlog_errno(ret); 1008 mlog_errno(ret);
994 goto out; 1009 goto out;
995 } 1010 }
996 } 1011 }
997 1012
998 /* 1013 /*
999 * Parts of newly allocated pages need 1014 * Parts of newly allocated pages need to be zero'd.
1000 * 1015 *
1001 * Above, we have also rewritten 'to' 1016 * Above, we have also rewritten 'to' and 'from' - as far as
1002 * the rest of the function is concer 1017 * the rest of the function is concerned, the entire cluster
1003 * range inside of a page needs to be 1018 * range inside of a page needs to be written.
1004 * 1019 *
1005 * We can skip this if the page is up 1020 * We can skip this if the page is up to date - it's already
1006 * been zero'd from being read in as 1021 * been zero'd from being read in as a hole.
1007 */ 1022 */
1008 if (new && !PageUptodate(page)) 1023 if (new && !PageUptodate(page))
1009 ocfs2_clear_page_regions(page 1024 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1010 cpos 1025 cpos, user_data_from, user_data_to);
1011 1026
1012 flush_dcache_page(page); 1027 flush_dcache_page(page);
1013 1028
1014 out: 1029 out:
1015 return ret; 1030 return ret;
1016 } 1031 }
1017 1032
1018 /* 1033 /*
1019 * This function will only grab one clusters 1034 * This function will only grab one clusters worth of pages.
1020 */ 1035 */
1021 static int ocfs2_grab_pages_for_write(struct 1036 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1022 struct 1037 struct ocfs2_write_ctxt *wc,
1023 u32 cpo 1038 u32 cpos, loff_t user_pos,
1024 unsigne 1039 unsigned user_len, int new,
1025 struct 1040 struct page *mmap_page)
1026 { 1041 {
1027 int ret = 0, i; 1042 int ret = 0, i;
1028 unsigned long start, target_index, en 1043 unsigned long start, target_index, end_index, index;
1029 struct inode *inode = mapping->host; 1044 struct inode *inode = mapping->host;
1030 loff_t last_byte; 1045 loff_t last_byte;
1031 1046
1032 target_index = user_pos >> PAGE_SHIFT 1047 target_index = user_pos >> PAGE_SHIFT;
1033 1048
1034 /* 1049 /*
1035 * Figure out how many pages we'll be 1050 * Figure out how many pages we'll be manipulating here. For
1036 * non allocating write, we just chan 1051 * non allocating write, we just change the one
1037 * page. Otherwise, we'll need a whol 1052 * page. Otherwise, we'll need a whole clusters worth. If we're
1038 * writing past i_size, we only need 1053 * writing past i_size, we only need enough pages to cover the
1039 * last page of the write. 1054 * last page of the write.
1040 */ 1055 */
1041 if (new) { 1056 if (new) {
1042 wc->w_num_pages = ocfs2_pages 1057 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1043 start = ocfs2_align_clusters_ 1058 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1044 /* 1059 /*
1045 * We need the index *past* t 1060 * We need the index *past* the last page we could possibly
1046 * touch. This is the page p 1061 * touch. This is the page past the end of the write or
1047 * i_size, whichever is great 1062 * i_size, whichever is greater.
1048 */ 1063 */
1049 last_byte = max(user_pos + us 1064 last_byte = max(user_pos + user_len, i_size_read(inode));
1050 BUG_ON(last_byte < 1); 1065 BUG_ON(last_byte < 1);
1051 end_index = ((last_byte - 1) 1066 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1;
1052 if ((start + wc->w_num_pages) 1067 if ((start + wc->w_num_pages) > end_index)
1053 wc->w_num_pages = end 1068 wc->w_num_pages = end_index - start;
1054 } else { 1069 } else {
1055 wc->w_num_pages = 1; 1070 wc->w_num_pages = 1;
1056 start = target_index; 1071 start = target_index;
1057 } 1072 }
1058 end_index = (user_pos + user_len - 1) 1073 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT;
1059 1074
1060 for(i = 0; i < wc->w_num_pages; i++) 1075 for(i = 0; i < wc->w_num_pages; i++) {
1061 index = start + i; 1076 index = start + i;
1062 1077
1063 if (index >= target_index && 1078 if (index >= target_index && index <= end_index &&
1064 wc->w_type == OCFS2_WRITE 1079 wc->w_type == OCFS2_WRITE_MMAP) {
1065 /* 1080 /*
1066 * ocfs2_pagemkwrite( 1081 * ocfs2_pagemkwrite() is a little different
1067 * and wants us to di 1082 * and wants us to directly use the page
1068 * passed in. 1083 * passed in.
1069 */ 1084 */
1070 lock_page(mmap_page); 1085 lock_page(mmap_page);
1071 1086
1072 /* Exit and let the c 1087 /* Exit and let the caller retry */
1073 if (mmap_page->mappin 1088 if (mmap_page->mapping != mapping) {
1074 WARN_ON(mmap_ 1089 WARN_ON(mmap_page->mapping);
1075 unlock_page(m 1090 unlock_page(mmap_page);
1076 ret = -EAGAIN 1091 ret = -EAGAIN;
1077 goto out; 1092 goto out;
1078 } 1093 }
1079 1094
1080 get_page(mmap_page); 1095 get_page(mmap_page);
1081 wc->w_pages[i] = mmap 1096 wc->w_pages[i] = mmap_page;
1082 wc->w_target_locked = 1097 wc->w_target_locked = true;
1083 } else if (index >= target_in 1098 } else if (index >= target_index && index <= end_index &&
1084 wc->w_type == OCFS 1099 wc->w_type == OCFS2_WRITE_DIRECT) {
1085 /* Direct write has n 1100 /* Direct write has no mapping page. */
1086 wc->w_pages[i] = NULL 1101 wc->w_pages[i] = NULL;
1087 continue; 1102 continue;
1088 } else { 1103 } else {
1089 wc->w_pages[i] = find 1104 wc->w_pages[i] = find_or_create_page(mapping, index,
1090 1105 GFP_NOFS);
1091 if (!wc->w_pages[i]) 1106 if (!wc->w_pages[i]) {
1092 ret = -ENOMEM 1107 ret = -ENOMEM;
1093 mlog_errno(re 1108 mlog_errno(ret);
1094 goto out; 1109 goto out;
1095 } 1110 }
1096 } 1111 }
1097 wait_for_stable_page(wc->w_pa 1112 wait_for_stable_page(wc->w_pages[i]);
1098 1113
1099 if (index == target_index) 1114 if (index == target_index)
1100 wc->w_target_page = w 1115 wc->w_target_page = wc->w_pages[i];
1101 } 1116 }
1102 out: 1117 out:
1103 if (ret) 1118 if (ret)
1104 wc->w_target_locked = false; 1119 wc->w_target_locked = false;
1105 return ret; 1120 return ret;
1106 } 1121 }
1107 1122
1108 /* 1123 /*
1109 * Prepare a single cluster for write one clu 1124 * Prepare a single cluster for write one cluster into the file.
1110 */ 1125 */
1111 static int ocfs2_write_cluster(struct address 1126 static int ocfs2_write_cluster(struct address_space *mapping,
1112 u32 *phys, uns 1127 u32 *phys, unsigned int new,
1113 unsigned int c 1128 unsigned int clear_unwritten,
1114 unsigned int s 1129 unsigned int should_zero,
1115 struct ocfs2_a 1130 struct ocfs2_alloc_context *data_ac,
1116 struct ocfs2_a 1131 struct ocfs2_alloc_context *meta_ac,
1117 struct ocfs2_w 1132 struct ocfs2_write_ctxt *wc, u32 cpos,
1118 loff_t user_po 1133 loff_t user_pos, unsigned user_len)
1119 { 1134 {
1120 int ret, i; 1135 int ret, i;
1121 u64 p_blkno; 1136 u64 p_blkno;
1122 struct inode *inode = mapping->host; 1137 struct inode *inode = mapping->host;
1123 struct ocfs2_extent_tree et; 1138 struct ocfs2_extent_tree et;
1124 int bpc = ocfs2_clusters_to_blocks(in 1139 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1125 1140
1126 if (new) { 1141 if (new) {
1127 u32 tmp_pos; 1142 u32 tmp_pos;
1128 1143
1129 /* 1144 /*
1130 * This is safe to call with 1145 * This is safe to call with the page locks - it won't take
1131 * any additional semaphores 1146 * any additional semaphores or cluster locks.
1132 */ 1147 */
1133 tmp_pos = cpos; 1148 tmp_pos = cpos;
1134 ret = ocfs2_add_inode_data(OC 1149 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1135 &t 1150 &tmp_pos, 1, !clear_unwritten,
1136 wc 1151 wc->w_di_bh, wc->w_handle,
1137 da 1152 data_ac, meta_ac, NULL);
1138 /* 1153 /*
1139 * This shouldn't happen beca 1154 * This shouldn't happen because we must have already
1140 * calculated the correct met 1155 * calculated the correct meta data allocation required. The
1141 * internal tree allocation c 1156 * internal tree allocation code should know how to increase
1142 * transaction credits itself 1157 * transaction credits itself.
1143 * 1158 *
1144 * If need be, we could handl 1159 * If need be, we could handle -EAGAIN for a
1145 * RESTART_TRANS here. 1160 * RESTART_TRANS here.
1146 */ 1161 */
1147 mlog_bug_on_msg(ret == -EAGAI 1162 mlog_bug_on_msg(ret == -EAGAIN,
1148 "Inode %llu: 1163 "Inode %llu: EAGAIN return during allocation.\n",
1149 (unsigned lon 1164 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1150 if (ret < 0) { 1165 if (ret < 0) {
1151 mlog_errno(ret); 1166 mlog_errno(ret);
1152 goto out; 1167 goto out;
1153 } 1168 }
1154 } else if (clear_unwritten) { 1169 } else if (clear_unwritten) {
1155 ocfs2_init_dinode_extent_tree 1170 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1156 1171 wc->w_di_bh);
1157 ret = ocfs2_mark_extent_writt 1172 ret = ocfs2_mark_extent_written(inode, &et,
1158 1173 wc->w_handle, cpos, 1, *phys,
1159 1174 meta_ac, &wc->w_dealloc);
1160 if (ret < 0) { 1175 if (ret < 0) {
1161 mlog_errno(ret); 1176 mlog_errno(ret);
1162 goto out; 1177 goto out;
1163 } 1178 }
1164 } 1179 }
1165 1180
1166 /* 1181 /*
1167 * The only reason this should fail i 1182 * The only reason this should fail is due to an inability to
1168 * find the extent added. 1183 * find the extent added.
1169 */ 1184 */
1170 ret = ocfs2_get_clusters(inode, cpos, 1185 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL);
1171 if (ret < 0) { 1186 if (ret < 0) {
1172 mlog(ML_ERROR, "Get physical 1187 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1173 "at logical clust 1188 "at logical cluster %u",
1174 (unsigned long lo 1189 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
1175 goto out; 1190 goto out;
1176 } 1191 }
1177 1192
1178 BUG_ON(*phys == 0); 1193 BUG_ON(*phys == 0);
1179 1194
1180 p_blkno = ocfs2_clusters_to_blocks(in 1195 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys);
1181 if (!should_zero) 1196 if (!should_zero)
1182 p_blkno += (user_pos >> inode 1197 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1);
1183 1198
1184 for(i = 0; i < wc->w_num_pages; i++) 1199 for(i = 0; i < wc->w_num_pages; i++) {
1185 int tmpret; 1200 int tmpret;
1186 1201
1187 /* This is the direct io targ 1202 /* This is the direct io target page. */
1188 if (wc->w_pages[i] == NULL) { 1203 if (wc->w_pages[i] == NULL) {
1189 p_blkno++; 1204 p_blkno++;
1190 continue; 1205 continue;
1191 } 1206 }
1192 1207
1193 tmpret = ocfs2_prepare_page_f 1208 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1194 1209 wc->w_pages[i], cpos,
1195 1210 user_pos, user_len,
1196 1211 should_zero);
1197 if (tmpret) { 1212 if (tmpret) {
1198 mlog_errno(tmpret); 1213 mlog_errno(tmpret);
1199 if (ret == 0) 1214 if (ret == 0)
1200 ret = tmpret; 1215 ret = tmpret;
1201 } 1216 }
1202 } 1217 }
1203 1218
1204 /* 1219 /*
1205 * We only have cleanup to do in case 1220 * We only have cleanup to do in case of allocating write.
1206 */ 1221 */
1207 if (ret && new) 1222 if (ret && new)
1208 ocfs2_write_failure(inode, wc 1223 ocfs2_write_failure(inode, wc, user_pos, user_len);
1209 1224
1210 out: 1225 out:
1211 1226
1212 return ret; 1227 return ret;
1213 } 1228 }
1214 1229
1215 static int ocfs2_write_cluster_by_desc(struct 1230 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1216 struct 1231 struct ocfs2_alloc_context *data_ac,
1217 struct 1232 struct ocfs2_alloc_context *meta_ac,
1218 struct 1233 struct ocfs2_write_ctxt *wc,
1219 loff_t 1234 loff_t pos, unsigned len)
1220 { 1235 {
1221 int ret, i; 1236 int ret, i;
1222 loff_t cluster_off; 1237 loff_t cluster_off;
1223 unsigned int local_len = len; 1238 unsigned int local_len = len;
1224 struct ocfs2_write_cluster_desc *desc 1239 struct ocfs2_write_cluster_desc *desc;
1225 struct ocfs2_super *osb = OCFS2_SB(ma 1240 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1226 1241
1227 for (i = 0; i < wc->w_clen; i++) { 1242 for (i = 0; i < wc->w_clen; i++) {
1228 desc = &wc->w_desc[i]; 1243 desc = &wc->w_desc[i];
1229 1244
1230 /* 1245 /*
1231 * We have to make sure that 1246 * We have to make sure that the total write passed in
1232 * doesn't extend past a sing 1247 * doesn't extend past a single cluster.
1233 */ 1248 */
1234 local_len = len; 1249 local_len = len;
1235 cluster_off = pos & (osb->s_c 1250 cluster_off = pos & (osb->s_clustersize - 1);
1236 if ((cluster_off + local_len) 1251 if ((cluster_off + local_len) > osb->s_clustersize)
1237 local_len = osb->s_cl 1252 local_len = osb->s_clustersize - cluster_off;
1238 1253
1239 ret = ocfs2_write_cluster(map 1254 ret = ocfs2_write_cluster(mapping, &desc->c_phys,
1240 des 1255 desc->c_new,
1241 des 1256 desc->c_clear_unwritten,
1242 des 1257 desc->c_needs_zero,
1243 dat 1258 data_ac, meta_ac,
1244 wc, 1259 wc, desc->c_cpos, pos, local_len);
1245 if (ret) { 1260 if (ret) {
1246 mlog_errno(ret); 1261 mlog_errno(ret);
1247 goto out; 1262 goto out;
1248 } 1263 }
1249 1264
1250 len -= local_len; 1265 len -= local_len;
1251 pos += local_len; 1266 pos += local_len;
1252 } 1267 }
1253 1268
1254 ret = 0; 1269 ret = 0;
1255 out: 1270 out:
1256 return ret; 1271 return ret;
1257 } 1272 }
1258 1273
1259 /* 1274 /*
1260 * ocfs2_write_end() wants to know which part 1275 * ocfs2_write_end() wants to know which parts of the target page it
1261 * should complete the write on. It's easiest 1276 * should complete the write on. It's easiest to compute them ahead of
1262 * time when a more complete view of the writ 1277 * time when a more complete view of the write is available.
1263 */ 1278 */
1264 static void ocfs2_set_target_boundaries(struc 1279 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1265 struc 1280 struct ocfs2_write_ctxt *wc,
1266 loff_ 1281 loff_t pos, unsigned len, int alloc)
1267 { 1282 {
1268 struct ocfs2_write_cluster_desc *desc 1283 struct ocfs2_write_cluster_desc *desc;
1269 1284
1270 wc->w_target_from = pos & (PAGE_SIZE 1285 wc->w_target_from = pos & (PAGE_SIZE - 1);
1271 wc->w_target_to = wc->w_target_from + 1286 wc->w_target_to = wc->w_target_from + len;
1272 1287
1273 if (alloc == 0) 1288 if (alloc == 0)
1274 return; 1289 return;
1275 1290
1276 /* 1291 /*
1277 * Allocating write - we may have dif 1292 * Allocating write - we may have different boundaries based
1278 * on page size and cluster size. 1293 * on page size and cluster size.
1279 * 1294 *
1280 * NOTE: We can no longer compute one 1295 * NOTE: We can no longer compute one value from the other as
1281 * the actual write length and user p 1296 * the actual write length and user provided length may be
1282 * different. 1297 * different.
1283 */ 1298 */
1284 1299
1285 if (wc->w_large_pages) { 1300 if (wc->w_large_pages) {
1286 /* 1301 /*
1287 * We only care about the 1st 1302 * We only care about the 1st and last cluster within
1288 * our range and whether they 1303 * our range and whether they should be zero'd or not. Either
1289 * value may be extended out 1304 * value may be extended out to the start/end of a
1290 * newly allocated cluster. 1305 * newly allocated cluster.
1291 */ 1306 */
1292 desc = &wc->w_desc[0]; 1307 desc = &wc->w_desc[0];
1293 if (desc->c_needs_zero) 1308 if (desc->c_needs_zero)
1294 ocfs2_figure_cluster_ 1309 ocfs2_figure_cluster_boundaries(osb,
1295 1310 desc->c_cpos,
1296 1311 &wc->w_target_from,
1297 1312 NULL);
1298 1313
1299 desc = &wc->w_desc[wc->w_clen 1314 desc = &wc->w_desc[wc->w_clen - 1];
1300 if (desc->c_needs_zero) 1315 if (desc->c_needs_zero)
1301 ocfs2_figure_cluster_ 1316 ocfs2_figure_cluster_boundaries(osb,
1302 1317 desc->c_cpos,
1303 1318 NULL,
1304 1319 &wc->w_target_to);
1305 } else { 1320 } else {
1306 wc->w_target_from = 0; 1321 wc->w_target_from = 0;
1307 wc->w_target_to = PAGE_SIZE; 1322 wc->w_target_to = PAGE_SIZE;
1308 } 1323 }
1309 } 1324 }
1310 1325
1311 /* 1326 /*
1312 * Check if this extent is marked UNWRITTEN b 1327 * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to
1313 * do the zero work. And should not to clear 1328 * do the zero work. And should not to clear UNWRITTEN since it will be cleared
1314 * by the direct io procedure. 1329 * by the direct io procedure.
1315 * If this is a new extent that allocated by 1330 * If this is a new extent that allocated by direct io, we should mark it in
1316 * the ip_unwritten_list. 1331 * the ip_unwritten_list.
1317 */ 1332 */
1318 static int ocfs2_unwritten_check(struct inode 1333 static int ocfs2_unwritten_check(struct inode *inode,
1319 struct ocfs2 1334 struct ocfs2_write_ctxt *wc,
1320 struct ocfs2 1335 struct ocfs2_write_cluster_desc *desc)
1321 { 1336 {
1322 struct ocfs2_inode_info *oi = OCFS2_I 1337 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1323 struct ocfs2_unwritten_extent *ue = N 1338 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL;
1324 int ret = 0; 1339 int ret = 0;
1325 1340
1326 if (!desc->c_needs_zero) 1341 if (!desc->c_needs_zero)
1327 return 0; 1342 return 0;
1328 1343
1329 retry: 1344 retry:
1330 spin_lock(&oi->ip_lock); 1345 spin_lock(&oi->ip_lock);
1331 /* Needs not to zero no metter buffer 1346 /* Needs not to zero no metter buffer or direct. The one who is zero
1332 * the cluster is doing zero. And he 1347 * the cluster is doing zero. And he will clear unwritten after all
1333 * cluster io finished. */ 1348 * cluster io finished. */
1334 list_for_each_entry(ue, &oi->ip_unwri 1349 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) {
1335 if (desc->c_cpos == ue->ue_cp 1350 if (desc->c_cpos == ue->ue_cpos) {
1336 BUG_ON(desc->c_new); 1351 BUG_ON(desc->c_new);
1337 desc->c_needs_zero = 1352 desc->c_needs_zero = 0;
1338 desc->c_clear_unwritt 1353 desc->c_clear_unwritten = 0;
1339 goto unlock; 1354 goto unlock;
1340 } 1355 }
1341 } 1356 }
1342 1357
1343 if (wc->w_type != OCFS2_WRITE_DIRECT) 1358 if (wc->w_type != OCFS2_WRITE_DIRECT)
1344 goto unlock; 1359 goto unlock;
1345 1360
1346 if (new == NULL) { 1361 if (new == NULL) {
1347 spin_unlock(&oi->ip_lock); 1362 spin_unlock(&oi->ip_lock);
1348 new = kmalloc(sizeof(struct o 1363 new = kmalloc(sizeof(struct ocfs2_unwritten_extent),
1349 GFP_NOFS); 1364 GFP_NOFS);
1350 if (new == NULL) { 1365 if (new == NULL) {
1351 ret = -ENOMEM; 1366 ret = -ENOMEM;
1352 goto out; 1367 goto out;
1353 } 1368 }
1354 goto retry; 1369 goto retry;
1355 } 1370 }
1356 /* This direct write will doing zero. 1371 /* This direct write will doing zero. */
1357 new->ue_cpos = desc->c_cpos; 1372 new->ue_cpos = desc->c_cpos;
1358 new->ue_phys = desc->c_phys; 1373 new->ue_phys = desc->c_phys;
1359 desc->c_clear_unwritten = 0; 1374 desc->c_clear_unwritten = 0;
1360 list_add_tail(&new->ue_ip_node, &oi-> 1375 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list);
1361 list_add_tail(&new->ue_node, &wc->w_u 1376 list_add_tail(&new->ue_node, &wc->w_unwritten_list);
1362 wc->w_unwritten_count++; 1377 wc->w_unwritten_count++;
1363 new = NULL; 1378 new = NULL;
1364 unlock: 1379 unlock:
1365 spin_unlock(&oi->ip_lock); 1380 spin_unlock(&oi->ip_lock);
1366 out: 1381 out:
1367 kfree(new); 1382 kfree(new);
1368 return ret; 1383 return ret;
1369 } 1384 }
1370 1385
1371 /* 1386 /*
1372 * Populate each single-cluster write descrip 1387 * Populate each single-cluster write descriptor in the write context
1373 * with information about the i/o to be done. 1388 * with information about the i/o to be done.
1374 * 1389 *
1375 * Returns the number of clusters that will h 1390 * Returns the number of clusters that will have to be allocated, as
1376 * well as a worst case estimate of the numbe 1391 * well as a worst case estimate of the number of extent records that
1377 * would have to be created during a write to 1392 * would have to be created during a write to an unwritten region.
1378 */ 1393 */
1379 static int ocfs2_populate_write_desc(struct i 1394 static int ocfs2_populate_write_desc(struct inode *inode,
1380 struct o 1395 struct ocfs2_write_ctxt *wc,
1381 unsigned 1396 unsigned int *clusters_to_alloc,
1382 unsigned 1397 unsigned int *extents_to_split)
1383 { 1398 {
1384 int ret; 1399 int ret;
1385 struct ocfs2_write_cluster_desc *desc 1400 struct ocfs2_write_cluster_desc *desc;
1386 unsigned int num_clusters = 0; 1401 unsigned int num_clusters = 0;
1387 unsigned int ext_flags = 0; 1402 unsigned int ext_flags = 0;
1388 u32 phys = 0; 1403 u32 phys = 0;
1389 int i; 1404 int i;
1390 1405
1391 *clusters_to_alloc = 0; 1406 *clusters_to_alloc = 0;
1392 *extents_to_split = 0; 1407 *extents_to_split = 0;
1393 1408
1394 for (i = 0; i < wc->w_clen; i++) { 1409 for (i = 0; i < wc->w_clen; i++) {
1395 desc = &wc->w_desc[i]; 1410 desc = &wc->w_desc[i];
1396 desc->c_cpos = wc->w_cpos + i 1411 desc->c_cpos = wc->w_cpos + i;
1397 1412
1398 if (num_clusters == 0) { 1413 if (num_clusters == 0) {
1399 /* 1414 /*
1400 * Need to look up th 1415 * Need to look up the next extent record.
1401 */ 1416 */
1402 ret = ocfs2_get_clust 1417 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1403 1418 &num_clusters, &ext_flags);
1404 if (ret) { 1419 if (ret) {
1405 mlog_errno(re 1420 mlog_errno(ret);
1406 goto out; 1421 goto out;
1407 } 1422 }
1408 1423
1409 /* We should already 1424 /* We should already CoW the refcountd extent. */
1410 BUG_ON(ext_flags & OC 1425 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1411 1426
1412 /* 1427 /*
1413 * Assume worst case 1428 * Assume worst case - that we're writing in
1414 * the middle of the 1429 * the middle of the extent.
1415 * 1430 *
1416 * We can assume that 1431 * We can assume that the write proceeds from
1417 * left to right, in 1432 * left to right, in which case the extent
1418 * insert code is sma 1433 * insert code is smart enough to coalesce the
1419 * next splits into t 1434 * next splits into the previous records created.
1420 */ 1435 */
1421 if (ext_flags & OCFS2 1436 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1422 *extents_to_s 1437 *extents_to_split = *extents_to_split + 2;
1423 } else if (phys) { 1438 } else if (phys) {
1424 /* 1439 /*
1425 * Only increment phy 1440 * Only increment phys if it doesn't describe
1426 * a hole. 1441 * a hole.
1427 */ 1442 */
1428 phys++; 1443 phys++;
1429 } 1444 }
1430 1445
1431 /* 1446 /*
1432 * If w_first_new_cpos is < U 1447 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1433 * file that got extended. w 1448 * file that got extended. w_first_new_cpos tells us
1434 * where the newly allocated 1449 * where the newly allocated clusters are so we can
1435 * zero them. 1450 * zero them.
1436 */ 1451 */
1437 if (desc->c_cpos >= wc->w_fir 1452 if (desc->c_cpos >= wc->w_first_new_cpos) {
1438 BUG_ON(phys == 0); 1453 BUG_ON(phys == 0);
1439 desc->c_needs_zero = 1454 desc->c_needs_zero = 1;
1440 } 1455 }
1441 1456
1442 desc->c_phys = phys; 1457 desc->c_phys = phys;
1443 if (phys == 0) { 1458 if (phys == 0) {
1444 desc->c_new = 1; 1459 desc->c_new = 1;
1445 desc->c_needs_zero = 1460 desc->c_needs_zero = 1;
1446 desc->c_clear_unwritt 1461 desc->c_clear_unwritten = 1;
1447 *clusters_to_alloc = 1462 *clusters_to_alloc = *clusters_to_alloc + 1;
1448 } 1463 }
1449 1464
1450 if (ext_flags & OCFS2_EXT_UNW 1465 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1451 desc->c_clear_unwritt 1466 desc->c_clear_unwritten = 1;
1452 desc->c_needs_zero = 1467 desc->c_needs_zero = 1;
1453 } 1468 }
1454 1469
1455 ret = ocfs2_unwritten_check(i 1470 ret = ocfs2_unwritten_check(inode, wc, desc);
1456 if (ret) { 1471 if (ret) {
1457 mlog_errno(ret); 1472 mlog_errno(ret);
1458 goto out; 1473 goto out;
1459 } 1474 }
1460 1475
1461 num_clusters--; 1476 num_clusters--;
1462 } 1477 }
1463 1478
1464 ret = 0; 1479 ret = 0;
1465 out: 1480 out:
1466 return ret; 1481 return ret;
1467 } 1482 }
1468 1483
1469 static int ocfs2_write_begin_inline(struct ad 1484 static int ocfs2_write_begin_inline(struct address_space *mapping,
1470 struct in 1485 struct inode *inode,
1471 struct oc 1486 struct ocfs2_write_ctxt *wc)
1472 { 1487 {
1473 int ret; 1488 int ret;
1474 struct ocfs2_super *osb = OCFS2_SB(in 1489 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1475 struct page *page; 1490 struct page *page;
1476 handle_t *handle; 1491 handle_t *handle;
1477 struct ocfs2_dinode *di = (struct ocf 1492 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1478 1493
1479 handle = ocfs2_start_trans(osb, OCFS2 1494 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1480 if (IS_ERR(handle)) { 1495 if (IS_ERR(handle)) {
1481 ret = PTR_ERR(handle); 1496 ret = PTR_ERR(handle);
1482 mlog_errno(ret); 1497 mlog_errno(ret);
1483 goto out; 1498 goto out;
1484 } 1499 }
1485 1500
1486 page = find_or_create_page(mapping, 0 1501 page = find_or_create_page(mapping, 0, GFP_NOFS);
1487 if (!page) { 1502 if (!page) {
1488 ocfs2_commit_trans(osb, handl 1503 ocfs2_commit_trans(osb, handle);
1489 ret = -ENOMEM; 1504 ret = -ENOMEM;
1490 mlog_errno(ret); 1505 mlog_errno(ret);
1491 goto out; 1506 goto out;
1492 } 1507 }
1493 /* 1508 /*
1494 * If we don't set w_num_pages then t 1509 * If we don't set w_num_pages then this page won't get unlocked
1495 * and freed on cleanup of the write 1510 * and freed on cleanup of the write context.
1496 */ 1511 */
1497 wc->w_pages[0] = wc->w_target_page = 1512 wc->w_pages[0] = wc->w_target_page = page;
1498 wc->w_num_pages = 1; 1513 wc->w_num_pages = 1;
1499 1514
1500 ret = ocfs2_journal_access_di(handle, 1515 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1501 OCFS2_J 1516 OCFS2_JOURNAL_ACCESS_WRITE);
1502 if (ret) { 1517 if (ret) {
1503 ocfs2_commit_trans(osb, handl 1518 ocfs2_commit_trans(osb, handle);
1504 1519
1505 mlog_errno(ret); 1520 mlog_errno(ret);
1506 goto out; 1521 goto out;
1507 } 1522 }
1508 1523
1509 if (!(OCFS2_I(inode)->ip_dyn_features 1524 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1510 ocfs2_set_inode_data_inline(i 1525 ocfs2_set_inode_data_inline(inode, di);
1511 1526
1512 if (!PageUptodate(page)) { 1527 if (!PageUptodate(page)) {
1513 ret = ocfs2_read_inline_data( 1528 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1514 if (ret) { 1529 if (ret) {
1515 ocfs2_commit_trans(os 1530 ocfs2_commit_trans(osb, handle);
1516 1531
1517 goto out; 1532 goto out;
1518 } 1533 }
1519 } 1534 }
1520 1535
1521 wc->w_handle = handle; 1536 wc->w_handle = handle;
1522 out: 1537 out:
1523 return ret; 1538 return ret;
1524 } 1539 }
1525 1540
1526 int ocfs2_size_fits_inline_data(struct buffer 1541 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1527 { 1542 {
1528 struct ocfs2_dinode *di = (struct ocf 1543 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1529 1544
1530 if (new_size <= le16_to_cpu(di->id2.i 1545 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1531 return 1; 1546 return 1;
1532 return 0; 1547 return 0;
1533 } 1548 }
1534 1549
1535 static int ocfs2_try_to_write_inline_data(str 1550 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1536 str 1551 struct inode *inode, loff_t pos,
1537 uns 1552 unsigned len, struct page *mmap_page,
1538 str 1553 struct ocfs2_write_ctxt *wc)
1539 { 1554 {
1540 int ret, written = 0; 1555 int ret, written = 0;
1541 loff_t end = pos + len; 1556 loff_t end = pos + len;
1542 struct ocfs2_inode_info *oi = OCFS2_I 1557 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1543 struct ocfs2_dinode *di = NULL; 1558 struct ocfs2_dinode *di = NULL;
1544 1559
1545 trace_ocfs2_try_to_write_inline_data( 1560 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1546 1561 len, (unsigned long long)pos,
1547 1562 oi->ip_dyn_features);
1548 1563
1549 /* 1564 /*
1550 * Handle inodes which already have i 1565 * Handle inodes which already have inline data 1st.
1551 */ 1566 */
1552 if (oi->ip_dyn_features & OCFS2_INLIN 1567 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1553 if (mmap_page == NULL && 1568 if (mmap_page == NULL &&
1554 ocfs2_size_fits_inline_da 1569 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1555 goto do_inline_write; 1570 goto do_inline_write;
1556 1571
1557 /* 1572 /*
1558 * The write won't fit - we h 1573 * The write won't fit - we have to give this inode an
1559 * inline extent list now. 1574 * inline extent list now.
1560 */ 1575 */
1561 ret = ocfs2_convert_inline_da 1576 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1562 if (ret) 1577 if (ret)
1563 mlog_errno(ret); 1578 mlog_errno(ret);
1564 goto out; 1579 goto out;
1565 } 1580 }
1566 1581
1567 /* 1582 /*
1568 * Check whether the inode can accept 1583 * Check whether the inode can accept inline data.
1569 */ 1584 */
1570 if (oi->ip_clusters != 0 || i_size_re 1585 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1571 return 0; 1586 return 0;
1572 1587
1573 /* 1588 /*
1574 * Check whether the write can fit. 1589 * Check whether the write can fit.
1575 */ 1590 */
1576 di = (struct ocfs2_dinode *)wc->w_di_ 1591 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1577 if (mmap_page || 1592 if (mmap_page ||
1578 end > ocfs2_max_inline_data_with_ 1593 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1579 return 0; 1594 return 0;
1580 1595
1581 do_inline_write: 1596 do_inline_write:
1582 ret = ocfs2_write_begin_inline(mappin 1597 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1583 if (ret) { 1598 if (ret) {
1584 mlog_errno(ret); 1599 mlog_errno(ret);
1585 goto out; 1600 goto out;
1586 } 1601 }
1587 1602
1588 /* 1603 /*
1589 * This signals to the caller that th 1604 * This signals to the caller that the data can be written
1590 * inline. 1605 * inline.
1591 */ 1606 */
1592 written = 1; 1607 written = 1;
1593 out: 1608 out:
1594 return written ? written : ret; 1609 return written ? written : ret;
1595 } 1610 }
1596 1611
1597 /* 1612 /*
1598 * This function only does anything for file 1613 * This function only does anything for file systems which can't
1599 * handle sparse files. 1614 * handle sparse files.
1600 * 1615 *
1601 * What we want to do here is fill in any hol 1616 * What we want to do here is fill in any hole between the current end
1602 * of allocation and the end of our write. Th 1617 * of allocation and the end of our write. That way the rest of the
1603 * write path can treat it as an non-allocati 1618 * write path can treat it as an non-allocating write, which has no
1604 * special case code for sparse/nonsparse fil 1619 * special case code for sparse/nonsparse files.
1605 */ 1620 */
1606 static int ocfs2_expand_nonsparse_inode(struc 1621 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1607 struc 1622 struct buffer_head *di_bh,
1608 loff_ 1623 loff_t pos, unsigned len,
1609 struc 1624 struct ocfs2_write_ctxt *wc)
1610 { 1625 {
1611 int ret; 1626 int ret;
1612 loff_t newsize = pos + len; 1627 loff_t newsize = pos + len;
1613 1628
1614 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1629 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1615 1630
1616 if (newsize <= i_size_read(inode)) 1631 if (newsize <= i_size_read(inode))
1617 return 0; 1632 return 0;
1618 1633
1619 ret = ocfs2_extend_no_holes(inode, di 1634 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1620 if (ret) 1635 if (ret)
1621 mlog_errno(ret); 1636 mlog_errno(ret);
1622 1637
1623 /* There is no wc if this is call fro 1638 /* There is no wc if this is call from direct. */
1624 if (wc) 1639 if (wc)
1625 wc->w_first_new_cpos = 1640 wc->w_first_new_cpos =
1626 ocfs2_clusters_for_by 1641 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1627 1642
1628 return ret; 1643 return ret;
1629 } 1644 }
1630 1645
1631 static int ocfs2_zero_tail(struct inode *inod 1646 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1632 loff_t pos) 1647 loff_t pos)
1633 { 1648 {
1634 int ret = 0; 1649 int ret = 0;
1635 1650
1636 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1651 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1637 if (pos > i_size_read(inode)) 1652 if (pos > i_size_read(inode))
1638 ret = ocfs2_zero_extend(inode 1653 ret = ocfs2_zero_extend(inode, di_bh, pos);
1639 1654
1640 return ret; 1655 return ret;
1641 } 1656 }
1642 1657
1643 int ocfs2_write_begin_nolock(struct address_s 1658 int ocfs2_write_begin_nolock(struct address_space *mapping,
1644 loff_t pos, unsi 1659 loff_t pos, unsigned len, ocfs2_write_type_t type,
1645 struct page **pa 1660 struct page **pagep, void **fsdata,
1646 struct buffer_he 1661 struct buffer_head *di_bh, struct page *mmap_page)
1647 { 1662 {
1648 int ret, cluster_of_pages, credits = 1663 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1649 unsigned int clusters_to_alloc, exten 1664 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1650 struct ocfs2_write_ctxt *wc; 1665 struct ocfs2_write_ctxt *wc;
1651 struct inode *inode = mapping->host; 1666 struct inode *inode = mapping->host;
1652 struct ocfs2_super *osb = OCFS2_SB(in 1667 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1653 struct ocfs2_dinode *di; 1668 struct ocfs2_dinode *di;
1654 struct ocfs2_alloc_context *data_ac = 1669 struct ocfs2_alloc_context *data_ac = NULL;
1655 struct ocfs2_alloc_context *meta_ac = 1670 struct ocfs2_alloc_context *meta_ac = NULL;
1656 handle_t *handle; 1671 handle_t *handle;
1657 struct ocfs2_extent_tree et; 1672 struct ocfs2_extent_tree et;
1658 int try_free = 1, ret1; 1673 int try_free = 1, ret1;
1659 1674
1660 try_again: 1675 try_again:
1661 ret = ocfs2_alloc_write_ctxt(&wc, osb 1676 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh);
1662 if (ret) { 1677 if (ret) {
1663 mlog_errno(ret); 1678 mlog_errno(ret);
1664 return ret; 1679 return ret;
1665 } 1680 }
1666 1681
1667 if (ocfs2_supports_inline_data(osb)) 1682 if (ocfs2_supports_inline_data(osb)) {
1668 ret = ocfs2_try_to_write_inli 1683 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1669 1684 mmap_page, wc);
1670 if (ret == 1) { 1685 if (ret == 1) {
1671 ret = 0; 1686 ret = 0;
1672 goto success; 1687 goto success;
1673 } 1688 }
1674 if (ret < 0) { 1689 if (ret < 0) {
1675 mlog_errno(ret); 1690 mlog_errno(ret);
1676 goto out; 1691 goto out;
1677 } 1692 }
1678 } 1693 }
1679 1694
1680 /* Direct io change i_size late, shou 1695 /* Direct io change i_size late, should not zero tail here. */
1681 if (type != OCFS2_WRITE_DIRECT) { 1696 if (type != OCFS2_WRITE_DIRECT) {
1682 if (ocfs2_sparse_alloc(osb)) 1697 if (ocfs2_sparse_alloc(osb))
1683 ret = ocfs2_zero_tail 1698 ret = ocfs2_zero_tail(inode, di_bh, pos);
1684 else 1699 else
1685 ret = ocfs2_expand_no 1700 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
1686 1701 len, wc);
1687 if (ret) { 1702 if (ret) {
1688 mlog_errno(ret); 1703 mlog_errno(ret);
1689 goto out; 1704 goto out;
1690 } 1705 }
1691 } 1706 }
1692 1707
1693 ret = ocfs2_check_range_for_refcount( 1708 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1694 if (ret < 0) { 1709 if (ret < 0) {
1695 mlog_errno(ret); 1710 mlog_errno(ret);
1696 goto out; 1711 goto out;
1697 } else if (ret == 1) { 1712 } else if (ret == 1) {
1698 clusters_need = wc->w_clen; 1713 clusters_need = wc->w_clen;
1699 ret = ocfs2_refcount_cow(inod 1714 ret = ocfs2_refcount_cow(inode, di_bh,
1700 wc-> 1715 wc->w_cpos, wc->w_clen, UINT_MAX);
1701 if (ret) { 1716 if (ret) {
1702 mlog_errno(ret); 1717 mlog_errno(ret);
1703 goto out; 1718 goto out;
1704 } 1719 }
1705 } 1720 }
1706 1721
1707 ret = ocfs2_populate_write_desc(inode 1722 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1708 &exte 1723 &extents_to_split);
1709 if (ret) { 1724 if (ret) {
1710 mlog_errno(ret); 1725 mlog_errno(ret);
1711 goto out; 1726 goto out;
1712 } 1727 }
1713 clusters_need += clusters_to_alloc; 1728 clusters_need += clusters_to_alloc;
1714 1729
1715 di = (struct ocfs2_dinode *)wc->w_di_ 1730 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1716 1731
1717 trace_ocfs2_write_begin_nolock( 1732 trace_ocfs2_write_begin_nolock(
1718 (unsigned long long)O 1733 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1719 (long long)i_size_rea 1734 (long long)i_size_read(inode),
1720 le32_to_cpu(di->i_clu 1735 le32_to_cpu(di->i_clusters),
1721 pos, len, type, mmap_ 1736 pos, len, type, mmap_page,
1722 clusters_to_alloc, ex 1737 clusters_to_alloc, extents_to_split);
1723 1738
1724 /* 1739 /*
1725 * We set w_target_from, w_target_to 1740 * We set w_target_from, w_target_to here so that
1726 * ocfs2_write_end() knows which rang 1741 * ocfs2_write_end() knows which range in the target page to
1727 * write out. An allocation requires 1742 * write out. An allocation requires that we write the entire
1728 * cluster range. 1743 * cluster range.
1729 */ 1744 */
1730 if (clusters_to_alloc || extents_to_s 1745 if (clusters_to_alloc || extents_to_split) {
1731 /* 1746 /*
1732 * XXX: We are stretching the 1747 * XXX: We are stretching the limits of
1733 * ocfs2_lock_allocators(). I 1748 * ocfs2_lock_allocators(). It greatly over-estimates
1734 * the work to be done. 1749 * the work to be done.
1735 */ 1750 */
1736 ocfs2_init_dinode_extent_tree 1751 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1737 1752 wc->w_di_bh);
1738 ret = ocfs2_lock_allocators(i 1753 ret = ocfs2_lock_allocators(inode, &et,
1739 c 1754 clusters_to_alloc, extents_to_split,
1740 & 1755 &data_ac, &meta_ac);
1741 if (ret) { 1756 if (ret) {
1742 mlog_errno(ret); 1757 mlog_errno(ret);
1743 goto out; 1758 goto out;
1744 } 1759 }
1745 1760
1746 if (data_ac) 1761 if (data_ac)
1747 data_ac->ac_resv = &O 1762 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1748 1763
1749 credits = ocfs2_calc_extend_c 1764 credits = ocfs2_calc_extend_credits(inode->i_sb,
1750 1765 &di->id2.i_list);
1751 } else if (type == OCFS2_WRITE_DIRECT 1766 } else if (type == OCFS2_WRITE_DIRECT)
1752 /* direct write needs not to 1767 /* direct write needs not to start trans if no extents alloc. */
1753 goto success; 1768 goto success;
1754 1769
1755 /* 1770 /*
1756 * We have to zero sparse allocated c 1771 * We have to zero sparse allocated clusters, unwritten extent clusters,
1757 * and non-sparse clusters we just ex 1772 * and non-sparse clusters we just extended. For non-sparse writes,
1758 * we know zeros will only be needed 1773 * we know zeros will only be needed in the first and/or last cluster.
1759 */ 1774 */
1760 if (wc->w_clen && (wc->w_desc[0].c_ne 1775 if (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1761 wc->w_desc[wc->w_c 1776 wc->w_desc[wc->w_clen - 1].c_needs_zero))
1762 cluster_of_pages = 1; 1777 cluster_of_pages = 1;
1763 else 1778 else
1764 cluster_of_pages = 0; 1779 cluster_of_pages = 0;
1765 1780
1766 ocfs2_set_target_boundaries(osb, wc, 1781 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1767 1782
1768 handle = ocfs2_start_trans(osb, credi 1783 handle = ocfs2_start_trans(osb, credits);
1769 if (IS_ERR(handle)) { 1784 if (IS_ERR(handle)) {
1770 ret = PTR_ERR(handle); 1785 ret = PTR_ERR(handle);
1771 mlog_errno(ret); 1786 mlog_errno(ret);
1772 goto out; 1787 goto out;
1773 } 1788 }
1774 1789
1775 wc->w_handle = handle; 1790 wc->w_handle = handle;
1776 1791
1777 if (clusters_to_alloc) { 1792 if (clusters_to_alloc) {
1778 ret = dquot_alloc_space_nodir 1793 ret = dquot_alloc_space_nodirty(inode,
1779 ocfs2_clusters_to_byt 1794 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1780 if (ret) 1795 if (ret)
1781 goto out_commit; 1796 goto out_commit;
1782 } 1797 }
1783 1798
1784 ret = ocfs2_journal_access_di(handle, 1799 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1785 OCFS2_J 1800 OCFS2_JOURNAL_ACCESS_WRITE);
1786 if (ret) { 1801 if (ret) {
1787 mlog_errno(ret); 1802 mlog_errno(ret);
1788 goto out_quota; 1803 goto out_quota;
1789 } 1804 }
1790 1805
1791 /* 1806 /*
1792 * Fill our page array first. That wa 1807 * Fill our page array first. That way we've grabbed enough so
1793 * that we can zero and flush if we e 1808 * that we can zero and flush if we error after adding the
1794 * extent. 1809 * extent.
1795 */ 1810 */
1796 ret = ocfs2_grab_pages_for_write(mapp 1811 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
1797 clus 1812 cluster_of_pages, mmap_page);
1798 if (ret) { !! 1813 if (ret && ret != -EAGAIN) {
1799 /* <<
1800 * ocfs2_grab_pages_for_write <<
1801 * the target page. In this c <<
1802 * page. This will trigger th <<
1803 * the operation. <<
1804 */ <<
1805 if (type == OCFS2_WRITE_MMAP <<
1806 BUG_ON(wc->w_target_p <<
1807 ret = 0; <<
1808 goto out_quota; <<
1809 } <<
1810 <<
1811 mlog_errno(ret); 1814 mlog_errno(ret);
1812 goto out_quota; 1815 goto out_quota;
1813 } 1816 }
1814 1817
>> 1818 /*
>> 1819 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
>> 1820 * the target page. In this case, we exit with no error and no target
>> 1821 * page. This will trigger the caller, page_mkwrite(), to re-try
>> 1822 * the operation.
>> 1823 */
>> 1824 if (ret == -EAGAIN) {
>> 1825 BUG_ON(wc->w_target_page);
>> 1826 ret = 0;
>> 1827 goto out_quota;
>> 1828 }
>> 1829
1815 ret = ocfs2_write_cluster_by_desc(map 1830 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1816 len 1831 len);
1817 if (ret) { 1832 if (ret) {
1818 mlog_errno(ret); 1833 mlog_errno(ret);
1819 goto out_quota; 1834 goto out_quota;
1820 } 1835 }
1821 1836
1822 if (data_ac) 1837 if (data_ac)
1823 ocfs2_free_alloc_context(data 1838 ocfs2_free_alloc_context(data_ac);
1824 if (meta_ac) 1839 if (meta_ac)
1825 ocfs2_free_alloc_context(meta 1840 ocfs2_free_alloc_context(meta_ac);
1826 1841
1827 success: 1842 success:
1828 if (pagep) 1843 if (pagep)
1829 *pagep = wc->w_target_page; 1844 *pagep = wc->w_target_page;
1830 *fsdata = wc; 1845 *fsdata = wc;
1831 return 0; 1846 return 0;
1832 out_quota: 1847 out_quota:
1833 if (clusters_to_alloc) 1848 if (clusters_to_alloc)
1834 dquot_free_space(inode, 1849 dquot_free_space(inode,
1835 ocfs2_clusters_to_b 1850 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1836 out_commit: 1851 out_commit:
1837 ocfs2_commit_trans(osb, handle); 1852 ocfs2_commit_trans(osb, handle);
1838 1853
1839 out: 1854 out:
1840 /* 1855 /*
1841 * The mmapped page won't be unlocked 1856 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(),
1842 * even in case of error here like EN 1857 * even in case of error here like ENOSPC and ENOMEM. So, we need
1843 * to unlock the target page manually 1858 * to unlock the target page manually to prevent deadlocks when
1844 * retrying again on ENOSPC, or when 1859 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED
1845 * to VM code. 1860 * to VM code.
1846 */ 1861 */
1847 if (wc->w_target_locked) 1862 if (wc->w_target_locked)
1848 unlock_page(mmap_page); 1863 unlock_page(mmap_page);
1849 1864
1850 ocfs2_free_write_ctxt(inode, wc); 1865 ocfs2_free_write_ctxt(inode, wc);
1851 1866
1852 if (data_ac) { 1867 if (data_ac) {
1853 ocfs2_free_alloc_context(data 1868 ocfs2_free_alloc_context(data_ac);
1854 data_ac = NULL; 1869 data_ac = NULL;
1855 } 1870 }
1856 if (meta_ac) { 1871 if (meta_ac) {
1857 ocfs2_free_alloc_context(meta 1872 ocfs2_free_alloc_context(meta_ac);
1858 meta_ac = NULL; 1873 meta_ac = NULL;
1859 } 1874 }
1860 1875
1861 if (ret == -ENOSPC && try_free) { 1876 if (ret == -ENOSPC && try_free) {
1862 /* 1877 /*
1863 * Try to free some truncate 1878 * Try to free some truncate log so that we can have enough
1864 * clusters to allocate. 1879 * clusters to allocate.
1865 */ 1880 */
1866 try_free = 0; 1881 try_free = 0;
1867 1882
1868 ret1 = ocfs2_try_to_free_trun 1883 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
1869 if (ret1 == 1) 1884 if (ret1 == 1)
1870 goto try_again; 1885 goto try_again;
1871 1886
1872 if (ret1 < 0) 1887 if (ret1 < 0)
1873 mlog_errno(ret1); 1888 mlog_errno(ret1);
1874 } 1889 }
1875 1890
1876 return ret; 1891 return ret;
1877 } 1892 }
1878 1893
1879 static int ocfs2_write_begin(struct file *fil 1894 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1880 loff_t pos, unsi !! 1895 loff_t pos, unsigned len, unsigned flags,
1881 struct page **pa 1896 struct page **pagep, void **fsdata)
1882 { 1897 {
1883 int ret; 1898 int ret;
1884 struct buffer_head *di_bh = NULL; 1899 struct buffer_head *di_bh = NULL;
1885 struct inode *inode = mapping->host; 1900 struct inode *inode = mapping->host;
1886 1901
1887 ret = ocfs2_inode_lock(inode, &di_bh, 1902 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1888 if (ret) { 1903 if (ret) {
1889 mlog_errno(ret); 1904 mlog_errno(ret);
1890 return ret; 1905 return ret;
1891 } 1906 }
1892 1907
1893 /* 1908 /*
1894 * Take alloc sem here to prevent con 1909 * Take alloc sem here to prevent concurrent lookups. That way
1895 * the mapping, zeroing and tree mani 1910 * the mapping, zeroing and tree manipulation within
1896 * ocfs2_write() will be safe against !! 1911 * ocfs2_write() will be safe against ->readpage(). This
1897 * should also serve to lock out allo 1912 * should also serve to lock out allocation from a shared
1898 * writeable region. 1913 * writeable region.
1899 */ 1914 */
1900 down_write(&OCFS2_I(inode)->ip_alloc_ 1915 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1901 1916
1902 ret = ocfs2_write_begin_nolock(mappin 1917 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER,
1903 pagep, 1918 pagep, fsdata, di_bh, NULL);
1904 if (ret) { 1919 if (ret) {
1905 mlog_errno(ret); 1920 mlog_errno(ret);
1906 goto out_fail; 1921 goto out_fail;
1907 } 1922 }
1908 1923
1909 brelse(di_bh); 1924 brelse(di_bh);
1910 1925
1911 return 0; 1926 return 0;
1912 1927
1913 out_fail: 1928 out_fail:
1914 up_write(&OCFS2_I(inode)->ip_alloc_se 1929 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1915 1930
1916 brelse(di_bh); 1931 brelse(di_bh);
1917 ocfs2_inode_unlock(inode, 1); 1932 ocfs2_inode_unlock(inode, 1);
1918 1933
1919 return ret; 1934 return ret;
1920 } 1935 }
1921 1936
1922 static void ocfs2_write_end_inline(struct ino 1937 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1923 unsigned l 1938 unsigned len, unsigned *copied,
1924 struct ocf 1939 struct ocfs2_dinode *di,
1925 struct ocf 1940 struct ocfs2_write_ctxt *wc)
1926 { 1941 {
1927 void *kaddr; 1942 void *kaddr;
1928 1943
1929 if (unlikely(*copied < len)) { 1944 if (unlikely(*copied < len)) {
1930 if (!PageUptodate(wc->w_targe 1945 if (!PageUptodate(wc->w_target_page)) {
1931 *copied = 0; 1946 *copied = 0;
1932 return; 1947 return;
1933 } 1948 }
1934 } 1949 }
1935 1950
1936 kaddr = kmap_atomic(wc->w_target_page 1951 kaddr = kmap_atomic(wc->w_target_page);
1937 memcpy(di->id2.i_data.id_data + pos, 1952 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1938 kunmap_atomic(kaddr); 1953 kunmap_atomic(kaddr);
1939 1954
1940 trace_ocfs2_write_end_inline( 1955 trace_ocfs2_write_end_inline(
1941 (unsigned long long)OCFS2_I(inod 1956 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1942 (unsigned long long)pos, *copied 1957 (unsigned long long)pos, *copied,
1943 le16_to_cpu(di->id2.i_data.id_co 1958 le16_to_cpu(di->id2.i_data.id_count),
1944 le16_to_cpu(di->i_dyn_features)) 1959 le16_to_cpu(di->i_dyn_features));
1945 } 1960 }
1946 1961
1947 int ocfs2_write_end_nolock(struct address_spa 1962 int ocfs2_write_end_nolock(struct address_space *mapping,
1948 loff_t pos, unsign 1963 loff_t pos, unsigned len, unsigned copied, void *fsdata)
1949 { 1964 {
1950 int i, ret; 1965 int i, ret;
1951 unsigned from, to, start = pos & (PAG 1966 unsigned from, to, start = pos & (PAGE_SIZE - 1);
1952 struct inode *inode = mapping->host; 1967 struct inode *inode = mapping->host;
1953 struct ocfs2_super *osb = OCFS2_SB(in 1968 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1954 struct ocfs2_write_ctxt *wc = fsdata; 1969 struct ocfs2_write_ctxt *wc = fsdata;
1955 struct ocfs2_dinode *di = (struct ocf 1970 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1956 handle_t *handle = wc->w_handle; 1971 handle_t *handle = wc->w_handle;
1957 struct page *tmppage; 1972 struct page *tmppage;
1958 1973
1959 BUG_ON(!list_empty(&wc->w_unwritten_l 1974 BUG_ON(!list_empty(&wc->w_unwritten_list));
1960 1975
1961 if (handle) { 1976 if (handle) {
1962 ret = ocfs2_journal_access_di 1977 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
1963 wc->w_di_bh, 1978 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1964 if (ret) { 1979 if (ret) {
1965 copied = ret; 1980 copied = ret;
1966 mlog_errno(ret); 1981 mlog_errno(ret);
1967 goto out; 1982 goto out;
1968 } 1983 }
1969 } 1984 }
1970 1985
1971 if (OCFS2_I(inode)->ip_dyn_features & 1986 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1972 ocfs2_write_end_inline(inode, 1987 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1973 goto out_write_size; 1988 goto out_write_size;
1974 } 1989 }
1975 1990
1976 if (unlikely(copied < len) && wc->w_t 1991 if (unlikely(copied < len) && wc->w_target_page) {
1977 loff_t new_isize; <<
1978 <<
1979 if (!PageUptodate(wc->w_targe 1992 if (!PageUptodate(wc->w_target_page))
1980 copied = 0; 1993 copied = 0;
1981 1994
1982 new_isize = max_t(loff_t, i_s !! 1995 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1983 if (new_isize > page_offset(w !! 1996 start+len);
1984 ocfs2_zero_new_buffer <<
1985 <<
1986 else { <<
1987 /* <<
1988 * When page is fully <<
1989 * failed), do not bo <<
1990 * it instead so that <<
1991 * put page & buffer <<
1992 * state. <<
1993 */ <<
1994 block_invalidate_foli <<
1995 <<
1996 } <<
1997 } 1997 }
1998 if (wc->w_target_page) 1998 if (wc->w_target_page)
1999 flush_dcache_page(wc->w_targe 1999 flush_dcache_page(wc->w_target_page);
2000 2000
2001 for(i = 0; i < wc->w_num_pages; i++) 2001 for(i = 0; i < wc->w_num_pages; i++) {
2002 tmppage = wc->w_pages[i]; 2002 tmppage = wc->w_pages[i];
2003 2003
2004 /* This is the direct io targ 2004 /* This is the direct io target page. */
2005 if (tmppage == NULL) 2005 if (tmppage == NULL)
2006 continue; 2006 continue;
2007 2007
2008 if (tmppage == wc->w_target_p 2008 if (tmppage == wc->w_target_page) {
2009 from = wc->w_target_f 2009 from = wc->w_target_from;
2010 to = wc->w_target_to; 2010 to = wc->w_target_to;
2011 2011
2012 BUG_ON(from > PAGE_SI 2012 BUG_ON(from > PAGE_SIZE ||
2013 to > PAGE_SIZE 2013 to > PAGE_SIZE ||
2014 to < from); 2014 to < from);
2015 } else { 2015 } else {
2016 /* 2016 /*
2017 * Pages adjacent to 2017 * Pages adjacent to the target (if any) imply
2018 * a hole-filling wri 2018 * a hole-filling write in which case we want
2019 * to flush their ent 2019 * to flush their entire range.
2020 */ 2020 */
2021 from = 0; 2021 from = 0;
2022 to = PAGE_SIZE; 2022 to = PAGE_SIZE;
2023 } 2023 }
2024 2024
2025 if (page_has_buffers(tmppage) 2025 if (page_has_buffers(tmppage)) {
2026 if (handle && ocfs2_s 2026 if (handle && ocfs2_should_order_data(inode)) {
2027 loff_t start_ 2027 loff_t start_byte =
2028 ((lof 2028 ((loff_t)tmppage->index << PAGE_SHIFT) +
2029 from; 2029 from;
2030 loff_t length 2030 loff_t length = to - from;
2031 ocfs2_jbd2_in 2031 ocfs2_jbd2_inode_add_write(handle, inode,
2032 2032 start_byte, length);
2033 } 2033 }
2034 block_commit_write(tm 2034 block_commit_write(tmppage, from, to);
2035 } 2035 }
2036 } 2036 }
2037 2037
2038 out_write_size: 2038 out_write_size:
2039 /* Direct io do not update i_size her 2039 /* Direct io do not update i_size here. */
2040 if (wc->w_type != OCFS2_WRITE_DIRECT) 2040 if (wc->w_type != OCFS2_WRITE_DIRECT) {
2041 pos += copied; 2041 pos += copied;
2042 if (pos > i_size_read(inode)) 2042 if (pos > i_size_read(inode)) {
2043 i_size_write(inode, p 2043 i_size_write(inode, pos);
2044 mark_inode_dirty(inod 2044 mark_inode_dirty(inode);
2045 } 2045 }
2046 inode->i_blocks = ocfs2_inode 2046 inode->i_blocks = ocfs2_inode_sector_count(inode);
2047 di->i_size = cpu_to_le64((u64 2047 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2048 inode_set_mtime_to_ts(inode, !! 2048 inode->i_mtime = inode->i_ctime = current_time(inode);
2049 di->i_mtime = di->i_ctime = c !! 2049 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2050 di->i_mtime_nsec = di->i_ctim !! 2050 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2051 if (handle) 2051 if (handle)
2052 ocfs2_update_inode_fs 2052 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2053 } 2053 }
2054 if (handle) 2054 if (handle)
2055 ocfs2_journal_dirty(handle, w 2055 ocfs2_journal_dirty(handle, wc->w_di_bh);
2056 2056
2057 out: 2057 out:
2058 /* unlock pages before dealloc since 2058 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2059 * lock, or it will cause a deadlock 2059 * lock, or it will cause a deadlock since journal commit threads holds
2060 * this lock and will ask for the pag 2060 * this lock and will ask for the page lock when flushing the data.
2061 * put it here to preserve the unlock 2061 * put it here to preserve the unlock order.
2062 */ 2062 */
2063 ocfs2_unlock_pages(wc); 2063 ocfs2_unlock_pages(wc);
2064 2064
2065 if (handle) 2065 if (handle)
2066 ocfs2_commit_trans(osb, handl 2066 ocfs2_commit_trans(osb, handle);
2067 2067
2068 ocfs2_run_deallocs(osb, &wc->w_deallo 2068 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2069 2069
2070 brelse(wc->w_di_bh); 2070 brelse(wc->w_di_bh);
2071 kfree(wc); 2071 kfree(wc);
2072 2072
2073 return copied; 2073 return copied;
2074 } 2074 }
2075 2075
2076 static int ocfs2_write_end(struct file *file, 2076 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2077 loff_t pos, unsign 2077 loff_t pos, unsigned len, unsigned copied,
2078 struct page *page, 2078 struct page *page, void *fsdata)
2079 { 2079 {
2080 int ret; 2080 int ret;
2081 struct inode *inode = mapping->host; 2081 struct inode *inode = mapping->host;
2082 2082
2083 ret = ocfs2_write_end_nolock(mapping, 2083 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata);
2084 2084
2085 up_write(&OCFS2_I(inode)->ip_alloc_se 2085 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2086 ocfs2_inode_unlock(inode, 1); 2086 ocfs2_inode_unlock(inode, 1);
2087 2087
2088 return ret; 2088 return ret;
2089 } 2089 }
2090 2090
2091 struct ocfs2_dio_write_ctxt { 2091 struct ocfs2_dio_write_ctxt {
2092 struct list_head dw_zero_list; 2092 struct list_head dw_zero_list;
2093 unsigned dw_zero_count 2093 unsigned dw_zero_count;
2094 int dw_orphaned; 2094 int dw_orphaned;
2095 pid_t dw_writer_pid 2095 pid_t dw_writer_pid;
2096 }; 2096 };
2097 2097
2098 static struct ocfs2_dio_write_ctxt * 2098 static struct ocfs2_dio_write_ctxt *
2099 ocfs2_dio_alloc_write_ctx(struct buffer_head 2099 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc)
2100 { 2100 {
2101 struct ocfs2_dio_write_ctxt *dwc = NU 2101 struct ocfs2_dio_write_ctxt *dwc = NULL;
2102 2102
2103 if (bh->b_private) 2103 if (bh->b_private)
2104 return bh->b_private; 2104 return bh->b_private;
2105 2105
2106 dwc = kmalloc(sizeof(struct ocfs2_dio 2106 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS);
2107 if (dwc == NULL) 2107 if (dwc == NULL)
2108 return NULL; 2108 return NULL;
2109 INIT_LIST_HEAD(&dwc->dw_zero_list); 2109 INIT_LIST_HEAD(&dwc->dw_zero_list);
2110 dwc->dw_zero_count = 0; 2110 dwc->dw_zero_count = 0;
2111 dwc->dw_orphaned = 0; 2111 dwc->dw_orphaned = 0;
2112 dwc->dw_writer_pid = task_pid_nr(curr 2112 dwc->dw_writer_pid = task_pid_nr(current);
2113 bh->b_private = dwc; 2113 bh->b_private = dwc;
2114 *alloc = 1; 2114 *alloc = 1;
2115 2115
2116 return dwc; 2116 return dwc;
2117 } 2117 }
2118 2118
2119 static void ocfs2_dio_free_write_ctx(struct i 2119 static void ocfs2_dio_free_write_ctx(struct inode *inode,
2120 struct o 2120 struct ocfs2_dio_write_ctxt *dwc)
2121 { 2121 {
2122 ocfs2_free_unwritten_list(inode, &dwc 2122 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list);
2123 kfree(dwc); 2123 kfree(dwc);
2124 } 2124 }
2125 2125
2126 /* 2126 /*
2127 * TODO: Make this into a generic get_blocks 2127 * TODO: Make this into a generic get_blocks function.
2128 * 2128 *
2129 * From do_direct_io in direct-io.c: 2129 * From do_direct_io in direct-io.c:
2130 * "So what we do is to permit the ->get_blo 2130 * "So what we do is to permit the ->get_blocks function to populate
2131 * bh.b_size with the size of IO which is p 2131 * bh.b_size with the size of IO which is permitted at this offset and
2132 * this i_blkbits." 2132 * this i_blkbits."
2133 * 2133 *
2134 * This function is called directly from get_ 2134 * This function is called directly from get_more_blocks in direct-io.c.
2135 * 2135 *
2136 * called like this: dio->get_blocks(dio->ino 2136 * called like this: dio->get_blocks(dio->inode, fs_startblk,
2137 * fs_co 2137 * fs_count, map_bh, dio->rw == WRITE);
2138 */ 2138 */
2139 static int ocfs2_dio_wr_get_block(struct inod 2139 static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock,
2140 struct buffer_ 2140 struct buffer_head *bh_result, int create)
2141 { 2141 {
2142 struct ocfs2_super *osb = OCFS2_SB(in 2142 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2143 struct ocfs2_inode_info *oi = OCFS2_I 2143 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2144 struct ocfs2_write_ctxt *wc; 2144 struct ocfs2_write_ctxt *wc;
2145 struct ocfs2_write_cluster_desc *desc 2145 struct ocfs2_write_cluster_desc *desc = NULL;
2146 struct ocfs2_dio_write_ctxt *dwc = NU 2146 struct ocfs2_dio_write_ctxt *dwc = NULL;
2147 struct buffer_head *di_bh = NULL; 2147 struct buffer_head *di_bh = NULL;
2148 u64 p_blkno; 2148 u64 p_blkno;
2149 unsigned int i_blkbits = inode->i_sb- 2149 unsigned int i_blkbits = inode->i_sb->s_blocksize_bits;
2150 loff_t pos = iblock << i_blkbits; 2150 loff_t pos = iblock << i_blkbits;
2151 sector_t endblk = (i_size_read(inode) 2151 sector_t endblk = (i_size_read(inode) - 1) >> i_blkbits;
2152 unsigned len, total_len = bh_result-> 2152 unsigned len, total_len = bh_result->b_size;
2153 int ret = 0, first_get_block = 0; 2153 int ret = 0, first_get_block = 0;
2154 2154
2155 len = osb->s_clustersize - (pos & (os 2155 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1));
2156 len = min(total_len, len); 2156 len = min(total_len, len);
2157 2157
2158 /* 2158 /*
2159 * bh_result->b_size is count in get_ 2159 * bh_result->b_size is count in get_more_blocks according to write
2160 * "pos" and "end", we need map twice 2160 * "pos" and "end", we need map twice to return different buffer state:
2161 * 1. area in file size, not set NEW; 2161 * 1. area in file size, not set NEW;
2162 * 2. area out file size, set NEW. 2162 * 2. area out file size, set NEW.
2163 * 2163 *
2164 * iblock endblk 2164 * iblock endblk
2165 * |--------|---------|---------|---- 2165 * |--------|---------|---------|---------
2166 * |<-------area in file------->| 2166 * |<-------area in file------->|
2167 */ 2167 */
2168 2168
2169 if ((iblock <= endblk) && 2169 if ((iblock <= endblk) &&
2170 ((iblock + ((len - 1) >> i_blkbit 2170 ((iblock + ((len - 1) >> i_blkbits)) > endblk))
2171 len = (endblk - iblock + 1) < 2171 len = (endblk - iblock + 1) << i_blkbits;
2172 2172
2173 mlog(0, "get block of %lu at %llu:%u 2173 mlog(0, "get block of %lu at %llu:%u req %u\n",
2174 inode->i_ino, pos, le 2174 inode->i_ino, pos, len, total_len);
2175 2175
2176 /* 2176 /*
2177 * Because we need to change file siz 2177 * Because we need to change file size in ocfs2_dio_end_io_write(), or
2178 * we may need to add it to orphan di 2178 * we may need to add it to orphan dir. So can not fall to fast path
2179 * while file size will be changed. 2179 * while file size will be changed.
2180 */ 2180 */
2181 if (pos + total_len <= i_size_read(in 2181 if (pos + total_len <= i_size_read(inode)) {
2182 2182
2183 /* This is the fast path for 2183 /* This is the fast path for re-write. */
2184 ret = ocfs2_lock_get_block(in 2184 ret = ocfs2_lock_get_block(inode, iblock, bh_result, create);
2185 if (buffer_mapped(bh_result) 2185 if (buffer_mapped(bh_result) &&
2186 !buffer_new(bh_result) && 2186 !buffer_new(bh_result) &&
2187 ret == 0) 2187 ret == 0)
2188 goto out; 2188 goto out;
2189 2189
2190 /* Clear state set by ocfs2_g 2190 /* Clear state set by ocfs2_get_block. */
2191 bh_result->b_state = 0; 2191 bh_result->b_state = 0;
2192 } 2192 }
2193 2193
2194 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2194 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block);
2195 if (unlikely(dwc == NULL)) { 2195 if (unlikely(dwc == NULL)) {
2196 ret = -ENOMEM; 2196 ret = -ENOMEM;
2197 mlog_errno(ret); 2197 mlog_errno(ret);
2198 goto out; 2198 goto out;
2199 } 2199 }
2200 2200
2201 if (ocfs2_clusters_for_bytes(inode->i 2201 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) >
2202 ocfs2_clusters_for_bytes(inode->i 2202 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) &&
2203 !dwc->dw_orphaned) { 2203 !dwc->dw_orphaned) {
2204 /* 2204 /*
2205 * when we are going to alloc 2205 * when we are going to alloc extents beyond file size, add the
2206 * inode to orphan dir, so we 2206 * inode to orphan dir, so we can recall those spaces when
2207 * system crashed during writ 2207 * system crashed during write.
2208 */ 2208 */
2209 ret = ocfs2_add_inode_to_orph 2209 ret = ocfs2_add_inode_to_orphan(osb, inode);
2210 if (ret < 0) { 2210 if (ret < 0) {
2211 mlog_errno(ret); 2211 mlog_errno(ret);
2212 goto out; 2212 goto out;
2213 } 2213 }
2214 dwc->dw_orphaned = 1; 2214 dwc->dw_orphaned = 1;
2215 } 2215 }
2216 2216
2217 ret = ocfs2_inode_lock(inode, &di_bh, 2217 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2218 if (ret) { 2218 if (ret) {
2219 mlog_errno(ret); 2219 mlog_errno(ret);
2220 goto out; 2220 goto out;
2221 } 2221 }
2222 2222
2223 down_write(&oi->ip_alloc_sem); 2223 down_write(&oi->ip_alloc_sem);
2224 2224
2225 if (first_get_block) { 2225 if (first_get_block) {
2226 if (ocfs2_sparse_alloc(osb)) 2226 if (ocfs2_sparse_alloc(osb))
2227 ret = ocfs2_zero_tail 2227 ret = ocfs2_zero_tail(inode, di_bh, pos);
2228 else 2228 else
2229 ret = ocfs2_expand_no 2229 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos,
2230 2230 total_len, NULL);
2231 if (ret < 0) { 2231 if (ret < 0) {
2232 mlog_errno(ret); 2232 mlog_errno(ret);
2233 goto unlock; 2233 goto unlock;
2234 } 2234 }
2235 } 2235 }
2236 2236
2237 ret = ocfs2_write_begin_nolock(inode- 2237 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len,
2238 OCFS2_ 2238 OCFS2_WRITE_DIRECT, NULL,
2239 (void 2239 (void **)&wc, di_bh, NULL);
2240 if (ret) { 2240 if (ret) {
2241 mlog_errno(ret); 2241 mlog_errno(ret);
2242 goto unlock; 2242 goto unlock;
2243 } 2243 }
2244 2244
2245 desc = &wc->w_desc[0]; 2245 desc = &wc->w_desc[0];
2246 2246
2247 p_blkno = ocfs2_clusters_to_blocks(in 2247 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys);
2248 BUG_ON(p_blkno == 0); 2248 BUG_ON(p_blkno == 0);
2249 p_blkno += iblock & (u64)(ocfs2_clust 2249 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1);
2250 2250
2251 map_bh(bh_result, inode->i_sb, p_blkn 2251 map_bh(bh_result, inode->i_sb, p_blkno);
2252 bh_result->b_size = len; 2252 bh_result->b_size = len;
2253 if (desc->c_needs_zero) 2253 if (desc->c_needs_zero)
2254 set_buffer_new(bh_result); 2254 set_buffer_new(bh_result);
2255 2255
2256 if (iblock > endblk) 2256 if (iblock > endblk)
2257 set_buffer_new(bh_result); 2257 set_buffer_new(bh_result);
2258 2258
2259 /* May sleep in end_io. It should not 2259 /* May sleep in end_io. It should not happen in a irq context. So defer
2260 * it to dio work queue. */ 2260 * it to dio work queue. */
2261 set_buffer_defer_completion(bh_result 2261 set_buffer_defer_completion(bh_result);
2262 2262
2263 if (!list_empty(&wc->w_unwritten_list 2263 if (!list_empty(&wc->w_unwritten_list)) {
2264 struct ocfs2_unwritten_extent 2264 struct ocfs2_unwritten_extent *ue = NULL;
2265 2265
2266 ue = list_first_entry(&wc->w_ 2266 ue = list_first_entry(&wc->w_unwritten_list,
2267 struct 2267 struct ocfs2_unwritten_extent,
2268 ue_node 2268 ue_node);
2269 BUG_ON(ue->ue_cpos != desc->c 2269 BUG_ON(ue->ue_cpos != desc->c_cpos);
2270 /* The physical address may b 2270 /* The physical address may be 0, fill it. */
2271 ue->ue_phys = desc->c_phys; 2271 ue->ue_phys = desc->c_phys;
2272 2272
2273 list_splice_tail_init(&wc->w_ 2273 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list);
2274 dwc->dw_zero_count += wc->w_u 2274 dwc->dw_zero_count += wc->w_unwritten_count;
2275 } 2275 }
2276 2276
2277 ret = ocfs2_write_end_nolock(inode->i 2277 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc);
2278 BUG_ON(ret != len); 2278 BUG_ON(ret != len);
2279 ret = 0; 2279 ret = 0;
2280 unlock: 2280 unlock:
2281 up_write(&oi->ip_alloc_sem); 2281 up_write(&oi->ip_alloc_sem);
2282 ocfs2_inode_unlock(inode, 1); 2282 ocfs2_inode_unlock(inode, 1);
2283 brelse(di_bh); 2283 brelse(di_bh);
2284 out: 2284 out:
>> 2285 if (ret < 0)
>> 2286 ret = -EIO;
2285 return ret; 2287 return ret;
2286 } 2288 }
2287 2289
2288 static int ocfs2_dio_end_io_write(struct inod 2290 static int ocfs2_dio_end_io_write(struct inode *inode,
2289 struct ocfs 2291 struct ocfs2_dio_write_ctxt *dwc,
2290 loff_t offs 2292 loff_t offset,
2291 ssize_t byt 2293 ssize_t bytes)
2292 { 2294 {
2293 struct ocfs2_cached_dealloc_ctxt deal 2295 struct ocfs2_cached_dealloc_ctxt dealloc;
2294 struct ocfs2_extent_tree et; 2296 struct ocfs2_extent_tree et;
2295 struct ocfs2_super *osb = OCFS2_SB(in 2297 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2296 struct ocfs2_inode_info *oi = OCFS2_I 2298 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2297 struct ocfs2_unwritten_extent *ue = N 2299 struct ocfs2_unwritten_extent *ue = NULL;
2298 struct buffer_head *di_bh = NULL; 2300 struct buffer_head *di_bh = NULL;
2299 struct ocfs2_dinode *di; 2301 struct ocfs2_dinode *di;
2300 struct ocfs2_alloc_context *data_ac = 2302 struct ocfs2_alloc_context *data_ac = NULL;
2301 struct ocfs2_alloc_context *meta_ac = 2303 struct ocfs2_alloc_context *meta_ac = NULL;
2302 handle_t *handle = NULL; 2304 handle_t *handle = NULL;
2303 loff_t end = offset + bytes; 2305 loff_t end = offset + bytes;
2304 int ret = 0, credits = 0; !! 2306 int ret = 0, credits = 0, locked = 0;
2305 2307
2306 ocfs2_init_dealloc_ctxt(&dealloc); 2308 ocfs2_init_dealloc_ctxt(&dealloc);
2307 2309
2308 /* We do clear unwritten, delete orph 2310 /* We do clear unwritten, delete orphan, change i_size here. If neither
2309 * of these happen, we can skip all t 2311 * of these happen, we can skip all this. */
2310 if (list_empty(&dwc->dw_zero_list) && 2312 if (list_empty(&dwc->dw_zero_list) &&
2311 end <= i_size_read(inode) && 2313 end <= i_size_read(inode) &&
2312 !dwc->dw_orphaned) 2314 !dwc->dw_orphaned)
2313 goto out; 2315 goto out;
2314 2316
>> 2317 /* ocfs2_file_write_iter will get i_mutex, so we need not lock if we
>> 2318 * are in that context. */
>> 2319 if (dwc->dw_writer_pid != task_pid_nr(current)) {
>> 2320 inode_lock(inode);
>> 2321 locked = 1;
>> 2322 }
>> 2323
2315 ret = ocfs2_inode_lock(inode, &di_bh, 2324 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2316 if (ret < 0) { 2325 if (ret < 0) {
2317 mlog_errno(ret); 2326 mlog_errno(ret);
2318 goto out; 2327 goto out;
2319 } 2328 }
2320 2329
2321 down_write(&oi->ip_alloc_sem); 2330 down_write(&oi->ip_alloc_sem);
2322 2331
2323 /* Delete orphan before acquire i_rws !! 2332 /* Delete orphan before acquire i_mutex. */
2324 if (dwc->dw_orphaned) { 2333 if (dwc->dw_orphaned) {
2325 BUG_ON(dwc->dw_writer_pid != 2334 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current));
2326 2335
2327 end = end > i_size_read(inode 2336 end = end > i_size_read(inode) ? end : 0;
2328 2337
2329 ret = ocfs2_del_inode_from_or 2338 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh,
2330 !!end, end); 2339 !!end, end);
2331 if (ret < 0) 2340 if (ret < 0)
2332 mlog_errno(ret); 2341 mlog_errno(ret);
2333 } 2342 }
2334 2343
2335 di = (struct ocfs2_dinode *)di_bh->b_ 2344 di = (struct ocfs2_dinode *)di_bh->b_data;
2336 2345
2337 ocfs2_init_dinode_extent_tree(&et, IN 2346 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
2338 2347
2339 /* Attach dealloc with extent tree in 2348 /* Attach dealloc with extent tree in case that we may reuse extents
2340 * which are already unlinked from cu 2349 * which are already unlinked from current extent tree due to extent
2341 * rotation and merging. 2350 * rotation and merging.
2342 */ 2351 */
2343 et.et_dealloc = &dealloc; 2352 et.et_dealloc = &dealloc;
2344 2353
2345 ret = ocfs2_lock_allocators(inode, &e 2354 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2,
2346 &data_ac, 2355 &data_ac, &meta_ac);
2347 if (ret) { 2356 if (ret) {
2348 mlog_errno(ret); 2357 mlog_errno(ret);
2349 goto unlock; 2358 goto unlock;
2350 } 2359 }
2351 2360
2352 credits = ocfs2_calc_extend_credits(i 2361 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list);
2353 2362
2354 handle = ocfs2_start_trans(osb, credi 2363 handle = ocfs2_start_trans(osb, credits);
2355 if (IS_ERR(handle)) { 2364 if (IS_ERR(handle)) {
2356 ret = PTR_ERR(handle); 2365 ret = PTR_ERR(handle);
2357 mlog_errno(ret); 2366 mlog_errno(ret);
2358 goto unlock; 2367 goto unlock;
2359 } 2368 }
2360 ret = ocfs2_journal_access_di(handle, 2369 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2361 OCFS2_J 2370 OCFS2_JOURNAL_ACCESS_WRITE);
2362 if (ret) { 2371 if (ret) {
2363 mlog_errno(ret); 2372 mlog_errno(ret);
2364 goto commit; 2373 goto commit;
2365 } 2374 }
2366 2375
2367 list_for_each_entry(ue, &dwc->dw_zero 2376 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) {
2368 ret = ocfs2_assure_trans_cred <<
2369 if (ret < 0) { <<
2370 mlog_errno(ret); <<
2371 break; <<
2372 } <<
2373 ret = ocfs2_mark_extent_writt 2377 ret = ocfs2_mark_extent_written(inode, &et, handle,
2374 2378 ue->ue_cpos, 1,
2375 2379 ue->ue_phys,
2376 2380 meta_ac, &dealloc);
2377 if (ret < 0) { 2381 if (ret < 0) {
2378 mlog_errno(ret); 2382 mlog_errno(ret);
2379 break; 2383 break;
2380 } 2384 }
2381 } 2385 }
2382 2386
2383 if (end > i_size_read(inode)) { 2387 if (end > i_size_read(inode)) {
2384 ret = ocfs2_set_inode_size(ha 2388 ret = ocfs2_set_inode_size(handle, inode, di_bh, end);
2385 if (ret < 0) 2389 if (ret < 0)
2386 mlog_errno(ret); 2390 mlog_errno(ret);
2387 } 2391 }
2388 commit: 2392 commit:
2389 ocfs2_commit_trans(osb, handle); 2393 ocfs2_commit_trans(osb, handle);
2390 unlock: 2394 unlock:
2391 up_write(&oi->ip_alloc_sem); 2395 up_write(&oi->ip_alloc_sem);
2392 ocfs2_inode_unlock(inode, 1); 2396 ocfs2_inode_unlock(inode, 1);
2393 brelse(di_bh); 2397 brelse(di_bh);
2394 out: 2398 out:
2395 if (data_ac) 2399 if (data_ac)
2396 ocfs2_free_alloc_context(data 2400 ocfs2_free_alloc_context(data_ac);
2397 if (meta_ac) 2401 if (meta_ac)
2398 ocfs2_free_alloc_context(meta 2402 ocfs2_free_alloc_context(meta_ac);
2399 ocfs2_run_deallocs(osb, &dealloc); 2403 ocfs2_run_deallocs(osb, &dealloc);
>> 2404 if (locked)
>> 2405 inode_unlock(inode);
2400 ocfs2_dio_free_write_ctx(inode, dwc); 2406 ocfs2_dio_free_write_ctx(inode, dwc);
2401 2407
2402 return ret; 2408 return ret;
2403 } 2409 }
2404 2410
2405 /* 2411 /*
2406 * ocfs2_dio_end_io is called by the dio core 2412 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
2407 * particularly interested in the aio/dio cas 2413 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
2408 * to protect io on one node from truncation 2414 * to protect io on one node from truncation on another.
2409 */ 2415 */
2410 static int ocfs2_dio_end_io(struct kiocb *ioc 2416 static int ocfs2_dio_end_io(struct kiocb *iocb,
2411 loff_t offset, 2417 loff_t offset,
2412 ssize_t bytes, 2418 ssize_t bytes,
2413 void *private) 2419 void *private)
2414 { 2420 {
2415 struct inode *inode = file_inode(iocb 2421 struct inode *inode = file_inode(iocb->ki_filp);
2416 int level; 2422 int level;
2417 int ret = 0; 2423 int ret = 0;
2418 2424
2419 /* this io's submitter should not hav 2425 /* this io's submitter should not have unlocked this before we could */
2420 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2426 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
2421 2427
2422 if (bytes <= 0) 2428 if (bytes <= 0)
2423 mlog_ratelimited(ML_ERROR, "D 2429 mlog_ratelimited(ML_ERROR, "Direct IO failed, bytes = %lld",
2424 (long long)b 2430 (long long)bytes);
2425 if (private) { 2431 if (private) {
2426 if (bytes > 0) 2432 if (bytes > 0)
2427 ret = ocfs2_dio_end_i 2433 ret = ocfs2_dio_end_io_write(inode, private, offset,
2428 2434 bytes);
2429 else 2435 else
2430 ocfs2_dio_free_write_ 2436 ocfs2_dio_free_write_ctx(inode, private);
2431 } 2437 }
2432 2438
2433 ocfs2_iocb_clear_rw_locked(iocb); 2439 ocfs2_iocb_clear_rw_locked(iocb);
2434 2440
2435 level = ocfs2_iocb_rw_locked_level(io 2441 level = ocfs2_iocb_rw_locked_level(iocb);
2436 ocfs2_rw_unlock(inode, level); 2442 ocfs2_rw_unlock(inode, level);
2437 return ret; 2443 return ret;
2438 } 2444 }
2439 2445
2440 static ssize_t ocfs2_direct_IO(struct kiocb * 2446 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2441 { 2447 {
2442 struct file *file = iocb->ki_filp; 2448 struct file *file = iocb->ki_filp;
2443 struct inode *inode = file->f_mapping 2449 struct inode *inode = file->f_mapping->host;
2444 struct ocfs2_super *osb = OCFS2_SB(in 2450 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2445 get_block_t *get_block; 2451 get_block_t *get_block;
2446 2452
2447 /* 2453 /*
2448 * Fallback to buffered I/O if we see 2454 * Fallback to buffered I/O if we see an inode without
2449 * extents. 2455 * extents.
2450 */ 2456 */
2451 if (OCFS2_I(inode)->ip_dyn_features & 2457 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2452 return 0; 2458 return 0;
2453 2459
2454 /* Fallback to buffered I/O if we do 2460 /* Fallback to buffered I/O if we do not support append dio. */
2455 if (iocb->ki_pos + iter->count > i_si 2461 if (iocb->ki_pos + iter->count > i_size_read(inode) &&
2456 !ocfs2_supports_append_dio(osb)) 2462 !ocfs2_supports_append_dio(osb))
2457 return 0; 2463 return 0;
2458 2464
2459 if (iov_iter_rw(iter) == READ) 2465 if (iov_iter_rw(iter) == READ)
2460 get_block = ocfs2_lock_get_bl 2466 get_block = ocfs2_lock_get_block;
2461 else 2467 else
2462 get_block = ocfs2_dio_wr_get_ 2468 get_block = ocfs2_dio_wr_get_block;
2463 2469
2464 return __blockdev_direct_IO(iocb, ino 2470 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2465 iter, get 2471 iter, get_block,
2466 ocfs2_dio !! 2472 ocfs2_dio_end_io, NULL, 0);
2467 } 2473 }
2468 2474
2469 const struct address_space_operations ocfs2_a 2475 const struct address_space_operations ocfs2_aops = {
2470 .dirty_folio = block_dirty !! 2476 .readpage = ocfs2_readpage,
2471 .read_folio = ocfs2_read_ !! 2477 .readpages = ocfs2_readpages,
2472 .readahead = ocfs2_reada !! 2478 .writepage = ocfs2_writepage,
2473 .writepages = ocfs2_write <<
2474 .write_begin = ocfs2_write 2479 .write_begin = ocfs2_write_begin,
2475 .write_end = ocfs2_write 2480 .write_end = ocfs2_write_end,
2476 .bmap = ocfs2_bmap, 2481 .bmap = ocfs2_bmap,
2477 .direct_IO = ocfs2_direc 2482 .direct_IO = ocfs2_direct_IO,
2478 .invalidate_folio = block_inval !! 2483 .invalidatepage = block_invalidatepage,
2479 .release_folio = ocfs2_relea !! 2484 .releasepage = ocfs2_releasepage,
2480 .migrate_folio = buffer_migr !! 2485 .migratepage = buffer_migrate_page,
2481 .is_partially_uptodate = block_is_pa 2486 .is_partially_uptodate = block_is_partially_uptodate,
2482 .error_remove_folio = generic_err !! 2487 .error_remove_page = generic_error_remove_page,
2483 }; 2488 };
2484 2489
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.