1 // SPDX-License-Identifier: GPL-2.0-or-later !! 1 /* -*- mode: c; c-basic-offset: 8; -*-
2 /* !! 2 * vim: noexpandtab sw=8 ts=8 sts=0:
>> 3 *
3 * file.c 4 * file.c
4 * 5 *
5 * File open, close, extend, truncate 6 * File open, close, extend, truncate
6 * 7 *
7 * Copyright (C) 2002, 2004 Oracle. All right 8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
>> 9 *
>> 10 * This program is free software; you can redistribute it and/or
>> 11 * modify it under the terms of the GNU General Public
>> 12 * License as published by the Free Software Foundation; either
>> 13 * version 2 of the License, or (at your option) any later version.
>> 14 *
>> 15 * This program is distributed in the hope that it will be useful,
>> 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
>> 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
>> 18 * General Public License for more details.
>> 19 *
>> 20 * You should have received a copy of the GNU General Public
>> 21 * License along with this program; if not, write to the
>> 22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
>> 23 * Boston, MA 021110-1307, USA.
8 */ 24 */
9 25
10 #include <linux/capability.h> 26 #include <linux/capability.h>
11 #include <linux/fs.h> 27 #include <linux/fs.h>
12 #include <linux/types.h> 28 #include <linux/types.h>
13 #include <linux/slab.h> 29 #include <linux/slab.h>
14 #include <linux/highmem.h> 30 #include <linux/highmem.h>
15 #include <linux/pagemap.h> 31 #include <linux/pagemap.h>
16 #include <linux/uio.h> 32 #include <linux/uio.h>
17 #include <linux/sched.h> 33 #include <linux/sched.h>
18 #include <linux/splice.h> 34 #include <linux/splice.h>
19 #include <linux/mount.h> 35 #include <linux/mount.h>
20 #include <linux/writeback.h> 36 #include <linux/writeback.h>
21 #include <linux/falloc.h> 37 #include <linux/falloc.h>
22 #include <linux/quotaops.h> 38 #include <linux/quotaops.h>
23 #include <linux/blkdev.h> 39 #include <linux/blkdev.h>
24 #include <linux/backing-dev.h> 40 #include <linux/backing-dev.h>
25 41
26 #include <cluster/masklog.h> 42 #include <cluster/masklog.h>
27 43
28 #include "ocfs2.h" 44 #include "ocfs2.h"
29 45
30 #include "alloc.h" 46 #include "alloc.h"
31 #include "aops.h" 47 #include "aops.h"
32 #include "dir.h" 48 #include "dir.h"
33 #include "dlmglue.h" 49 #include "dlmglue.h"
34 #include "extent_map.h" 50 #include "extent_map.h"
35 #include "file.h" 51 #include "file.h"
36 #include "sysfile.h" 52 #include "sysfile.h"
37 #include "inode.h" 53 #include "inode.h"
38 #include "ioctl.h" 54 #include "ioctl.h"
39 #include "journal.h" 55 #include "journal.h"
40 #include "locks.h" 56 #include "locks.h"
41 #include "mmap.h" 57 #include "mmap.h"
42 #include "suballoc.h" 58 #include "suballoc.h"
43 #include "super.h" 59 #include "super.h"
44 #include "xattr.h" 60 #include "xattr.h"
45 #include "acl.h" 61 #include "acl.h"
46 #include "quota.h" 62 #include "quota.h"
47 #include "refcounttree.h" 63 #include "refcounttree.h"
48 #include "ocfs2_trace.h" 64 #include "ocfs2_trace.h"
49 65
50 #include "buffer_head_io.h" 66 #include "buffer_head_io.h"
51 67
52 static int ocfs2_init_file_private(struct inod 68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
53 { 69 {
54 struct ocfs2_file_private *fp; 70 struct ocfs2_file_private *fp;
55 71
56 fp = kzalloc(sizeof(struct ocfs2_file_ 72 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
57 if (!fp) 73 if (!fp)
58 return -ENOMEM; 74 return -ENOMEM;
59 75
60 fp->fp_file = file; 76 fp->fp_file = file;
61 mutex_init(&fp->fp_mutex); 77 mutex_init(&fp->fp_mutex);
62 ocfs2_file_lock_res_init(&fp->fp_flock 78 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
63 file->private_data = fp; 79 file->private_data = fp;
64 80
65 return 0; 81 return 0;
66 } 82 }
67 83
68 static void ocfs2_free_file_private(struct ino 84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
69 { 85 {
70 struct ocfs2_file_private *fp = file-> 86 struct ocfs2_file_private *fp = file->private_data;
71 struct ocfs2_super *osb = OCFS2_SB(ino 87 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
72 88
73 if (fp) { 89 if (fp) {
74 ocfs2_simple_drop_lockres(osb, 90 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
75 ocfs2_lock_res_free(&fp->fp_fl 91 ocfs2_lock_res_free(&fp->fp_flock);
76 kfree(fp); 92 kfree(fp);
77 file->private_data = NULL; 93 file->private_data = NULL;
78 } 94 }
79 } 95 }
80 96
81 static int ocfs2_file_open(struct inode *inode 97 static int ocfs2_file_open(struct inode *inode, struct file *file)
82 { 98 {
83 int status; 99 int status;
84 int mode = file->f_flags; 100 int mode = file->f_flags;
85 struct ocfs2_inode_info *oi = OCFS2_I( 101 struct ocfs2_inode_info *oi = OCFS2_I(inode);
86 102
87 trace_ocfs2_file_open(inode, file, fil 103 trace_ocfs2_file_open(inode, file, file->f_path.dentry,
88 (unsigned long l 104 (unsigned long long)oi->ip_blkno,
89 file->f_path.den 105 file->f_path.dentry->d_name.len,
90 file->f_path.den 106 file->f_path.dentry->d_name.name, mode);
91 107
92 if (file->f_mode & FMODE_WRITE) { 108 if (file->f_mode & FMODE_WRITE) {
93 status = dquot_initialize(inod 109 status = dquot_initialize(inode);
94 if (status) 110 if (status)
95 goto leave; 111 goto leave;
96 } 112 }
97 113
98 spin_lock(&oi->ip_lock); 114 spin_lock(&oi->ip_lock);
99 115
100 /* Check that the inode hasn't been wi 116 /* Check that the inode hasn't been wiped from disk by another
101 * node. If it hasn't then we're safe 117 * node. If it hasn't then we're safe as long as we hold the
102 * spin lock until our increment of op 118 * spin lock until our increment of open count. */
103 if (oi->ip_flags & OCFS2_INODE_DELETED 119 if (oi->ip_flags & OCFS2_INODE_DELETED) {
104 spin_unlock(&oi->ip_lock); 120 spin_unlock(&oi->ip_lock);
105 121
106 status = -ENOENT; 122 status = -ENOENT;
107 goto leave; 123 goto leave;
108 } 124 }
109 125
110 if (mode & O_DIRECT) 126 if (mode & O_DIRECT)
111 oi->ip_flags |= OCFS2_INODE_OP 127 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
112 128
113 oi->ip_open_count++; 129 oi->ip_open_count++;
114 spin_unlock(&oi->ip_lock); 130 spin_unlock(&oi->ip_lock);
115 131
116 status = ocfs2_init_file_private(inode 132 status = ocfs2_init_file_private(inode, file);
117 if (status) { 133 if (status) {
118 /* 134 /*
119 * We want to set open count b 135 * We want to set open count back if we're failing the
120 * open. 136 * open.
121 */ 137 */
122 spin_lock(&oi->ip_lock); 138 spin_lock(&oi->ip_lock);
123 oi->ip_open_count--; 139 oi->ip_open_count--;
124 spin_unlock(&oi->ip_lock); 140 spin_unlock(&oi->ip_lock);
125 } 141 }
126 142
127 file->f_mode |= FMODE_NOWAIT; 143 file->f_mode |= FMODE_NOWAIT;
128 144
129 leave: 145 leave:
130 return status; 146 return status;
131 } 147 }
132 148
133 static int ocfs2_file_release(struct inode *in 149 static int ocfs2_file_release(struct inode *inode, struct file *file)
134 { 150 {
135 struct ocfs2_inode_info *oi = OCFS2_I( 151 struct ocfs2_inode_info *oi = OCFS2_I(inode);
136 152
137 spin_lock(&oi->ip_lock); 153 spin_lock(&oi->ip_lock);
138 if (!--oi->ip_open_count) 154 if (!--oi->ip_open_count)
139 oi->ip_flags &= ~OCFS2_INODE_O 155 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
140 156
141 trace_ocfs2_file_release(inode, file, 157 trace_ocfs2_file_release(inode, file, file->f_path.dentry,
142 oi->ip_blkno, 158 oi->ip_blkno,
143 file->f_path. 159 file->f_path.dentry->d_name.len,
144 file->f_path. 160 file->f_path.dentry->d_name.name,
145 oi->ip_open_c 161 oi->ip_open_count);
146 spin_unlock(&oi->ip_lock); 162 spin_unlock(&oi->ip_lock);
147 163
148 ocfs2_free_file_private(inode, file); 164 ocfs2_free_file_private(inode, file);
149 165
150 return 0; 166 return 0;
151 } 167 }
152 168
153 static int ocfs2_dir_open(struct inode *inode, 169 static int ocfs2_dir_open(struct inode *inode, struct file *file)
154 { 170 {
155 return ocfs2_init_file_private(inode, 171 return ocfs2_init_file_private(inode, file);
156 } 172 }
157 173
158 static int ocfs2_dir_release(struct inode *ino 174 static int ocfs2_dir_release(struct inode *inode, struct file *file)
159 { 175 {
160 ocfs2_free_file_private(inode, file); 176 ocfs2_free_file_private(inode, file);
161 return 0; 177 return 0;
162 } 178 }
163 179
164 static int ocfs2_sync_file(struct file *file, 180 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
165 int datasync) 181 int datasync)
166 { 182 {
167 int err = 0; 183 int err = 0;
168 struct inode *inode = file->f_mapping- 184 struct inode *inode = file->f_mapping->host;
169 struct ocfs2_super *osb = OCFS2_SB(ino 185 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
170 struct ocfs2_inode_info *oi = OCFS2_I( 186 struct ocfs2_inode_info *oi = OCFS2_I(inode);
171 journal_t *journal = osb->journal->j_j 187 journal_t *journal = osb->journal->j_journal;
172 int ret; 188 int ret;
173 tid_t commit_tid; 189 tid_t commit_tid;
174 bool needs_barrier = false; 190 bool needs_barrier = false;
175 191
176 trace_ocfs2_sync_file(inode, file, fil 192 trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
177 oi->ip_blkno, 193 oi->ip_blkno,
178 file->f_path.den 194 file->f_path.dentry->d_name.len,
179 file->f_path.den 195 file->f_path.dentry->d_name.name,
180 (unsigned long l 196 (unsigned long long)datasync);
181 197
182 if (ocfs2_is_hard_readonly(osb) || ocf 198 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
183 return -EROFS; 199 return -EROFS;
184 200
185 err = file_write_and_wait_range(file, 201 err = file_write_and_wait_range(file, start, end);
186 if (err) 202 if (err)
187 return err; 203 return err;
188 204
189 commit_tid = datasync ? oi->i_datasync 205 commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
190 if (journal->j_flags & JBD2_BARRIER && 206 if (journal->j_flags & JBD2_BARRIER &&
191 !jbd2_trans_will_send_data_barrier 207 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
192 needs_barrier = true; 208 needs_barrier = true;
193 err = jbd2_complete_transaction(journa 209 err = jbd2_complete_transaction(journal, commit_tid);
194 if (needs_barrier) { 210 if (needs_barrier) {
195 ret = blkdev_issue_flush(inode !! 211 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
196 if (!err) 212 if (!err)
197 err = ret; 213 err = ret;
198 } 214 }
199 215
200 if (err) 216 if (err)
201 mlog_errno(err); 217 mlog_errno(err);
202 218
203 return (err < 0) ? -EIO : 0; 219 return (err < 0) ? -EIO : 0;
204 } 220 }
205 221
206 int ocfs2_should_update_atime(struct inode *in 222 int ocfs2_should_update_atime(struct inode *inode,
207 struct vfsmount 223 struct vfsmount *vfsmnt)
208 { 224 {
209 struct timespec64 now; 225 struct timespec64 now;
210 struct ocfs2_super *osb = OCFS2_SB(ino 226 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
211 227
212 if (ocfs2_is_hard_readonly(osb) || ocf 228 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
213 return 0; 229 return 0;
214 230
215 if ((inode->i_flags & S_NOATIME) || 231 if ((inode->i_flags & S_NOATIME) ||
216 ((inode->i_sb->s_flags & SB_NODIRA 232 ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
217 return 0; 233 return 0;
218 234
219 /* 235 /*
220 * We can be called with no vfsmnt str 236 * We can be called with no vfsmnt structure - NFSD will
221 * sometimes do this. 237 * sometimes do this.
222 * 238 *
223 * Note that our action here is differ 239 * Note that our action here is different than touch_atime() -
224 * if we can't tell whether this is a 240 * if we can't tell whether this is a noatime mount, then we
225 * don't know whether to trust the val 241 * don't know whether to trust the value of s_atime_quantum.
226 */ 242 */
227 if (vfsmnt == NULL) 243 if (vfsmnt == NULL)
228 return 0; 244 return 0;
229 245
230 if ((vfsmnt->mnt_flags & MNT_NOATIME) 246 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
231 ((vfsmnt->mnt_flags & MNT_NODIRATI 247 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
232 return 0; 248 return 0;
233 249
234 if (vfsmnt->mnt_flags & MNT_RELATIME) 250 if (vfsmnt->mnt_flags & MNT_RELATIME) {
235 struct timespec64 ctime = inod !! 251 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
236 struct timespec64 atime = inod !! 252 (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
237 struct timespec64 mtime = inod <<
238 <<
239 if ((timespec64_compare(&atime <<
240 (timespec64_compare(&atime <<
241 return 1; 253 return 1;
242 254
243 return 0; 255 return 0;
244 } 256 }
245 257
246 now = current_time(inode); 258 now = current_time(inode);
247 if ((now.tv_sec - inode_get_atime_sec( !! 259 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
248 return 0; 260 return 0;
249 else 261 else
250 return 1; 262 return 1;
251 } 263 }
252 264
253 int ocfs2_update_inode_atime(struct inode *ino 265 int ocfs2_update_inode_atime(struct inode *inode,
254 struct buffer_hea 266 struct buffer_head *bh)
255 { 267 {
256 int ret; 268 int ret;
257 struct ocfs2_super *osb = OCFS2_SB(ino 269 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
258 handle_t *handle; 270 handle_t *handle;
259 struct ocfs2_dinode *di = (struct ocfs 271 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
260 272
261 handle = ocfs2_start_trans(osb, OCFS2_ 273 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
262 if (IS_ERR(handle)) { 274 if (IS_ERR(handle)) {
263 ret = PTR_ERR(handle); 275 ret = PTR_ERR(handle);
264 mlog_errno(ret); 276 mlog_errno(ret);
265 goto out; 277 goto out;
266 } 278 }
267 279
268 ret = ocfs2_journal_access_di(handle, 280 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
269 OCFS2_JO 281 OCFS2_JOURNAL_ACCESS_WRITE);
270 if (ret) { 282 if (ret) {
271 mlog_errno(ret); 283 mlog_errno(ret);
272 goto out_commit; 284 goto out_commit;
273 } 285 }
274 286
275 /* 287 /*
276 * Don't use ocfs2_mark_inode_dirty() 288 * Don't use ocfs2_mark_inode_dirty() here as we don't always
277 * have i_rwsem to guard against concu !! 289 * have i_mutex to guard against concurrent changes to other
278 * inode fields. 290 * inode fields.
279 */ 291 */
280 inode_set_atime_to_ts(inode, current_t !! 292 inode->i_atime = current_time(inode);
281 di->i_atime = cpu_to_le64(inode_get_at !! 293 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
282 di->i_atime_nsec = cpu_to_le32(inode_g !! 294 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
283 ocfs2_update_inode_fsync_trans(handle, 295 ocfs2_update_inode_fsync_trans(handle, inode, 0);
284 ocfs2_journal_dirty(handle, bh); 296 ocfs2_journal_dirty(handle, bh);
285 297
286 out_commit: 298 out_commit:
287 ocfs2_commit_trans(osb, handle); 299 ocfs2_commit_trans(osb, handle);
288 out: 300 out:
289 return ret; 301 return ret;
290 } 302 }
291 303
292 int ocfs2_set_inode_size(handle_t *handle, 304 int ocfs2_set_inode_size(handle_t *handle,
293 struct inode * 305 struct inode *inode,
294 struct buffer_ 306 struct buffer_head *fe_bh,
295 u64 new_i_size 307 u64 new_i_size)
296 { 308 {
297 int status; 309 int status;
298 310
299 i_size_write(inode, new_i_size); 311 i_size_write(inode, new_i_size);
300 inode->i_blocks = ocfs2_inode_sector_c 312 inode->i_blocks = ocfs2_inode_sector_count(inode);
301 inode_set_mtime_to_ts(inode, inode_set !! 313 inode->i_ctime = inode->i_mtime = current_time(inode);
302 314
303 status = ocfs2_mark_inode_dirty(handle 315 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
304 if (status < 0) { 316 if (status < 0) {
305 mlog_errno(status); 317 mlog_errno(status);
306 goto bail; 318 goto bail;
307 } 319 }
308 320
309 bail: 321 bail:
310 return status; 322 return status;
311 } 323 }
312 324
313 int ocfs2_simple_size_update(struct inode *ino 325 int ocfs2_simple_size_update(struct inode *inode,
314 struct buffer_hea 326 struct buffer_head *di_bh,
315 u64 new_i_size) 327 u64 new_i_size)
316 { 328 {
317 int ret; 329 int ret;
318 struct ocfs2_super *osb = OCFS2_SB(ino 330 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
319 handle_t *handle = NULL; 331 handle_t *handle = NULL;
320 332
321 handle = ocfs2_start_trans(osb, OCFS2_ 333 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
322 if (IS_ERR(handle)) { 334 if (IS_ERR(handle)) {
323 ret = PTR_ERR(handle); 335 ret = PTR_ERR(handle);
324 mlog_errno(ret); 336 mlog_errno(ret);
325 goto out; 337 goto out;
326 } 338 }
327 339
328 ret = ocfs2_set_inode_size(handle, ino 340 ret = ocfs2_set_inode_size(handle, inode, di_bh,
329 new_i_size) 341 new_i_size);
330 if (ret < 0) 342 if (ret < 0)
331 mlog_errno(ret); 343 mlog_errno(ret);
332 344
333 ocfs2_update_inode_fsync_trans(handle, 345 ocfs2_update_inode_fsync_trans(handle, inode, 0);
334 ocfs2_commit_trans(osb, handle); 346 ocfs2_commit_trans(osb, handle);
335 out: 347 out:
336 return ret; 348 return ret;
337 } 349 }
338 350
339 static int ocfs2_cow_file_pos(struct inode *in 351 static int ocfs2_cow_file_pos(struct inode *inode,
340 struct buffer_he 352 struct buffer_head *fe_bh,
341 u64 offset) 353 u64 offset)
342 { 354 {
343 int status; 355 int status;
344 u32 phys, cpos = offset >> OCFS2_SB(in 356 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
345 unsigned int num_clusters = 0; 357 unsigned int num_clusters = 0;
346 unsigned int ext_flags = 0; 358 unsigned int ext_flags = 0;
347 359
348 /* 360 /*
349 * If the new offset is aligned to the 361 * If the new offset is aligned to the range of the cluster, there is
350 * no space for ocfs2_zero_range_for_t 362 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
351 * CoW either. 363 * CoW either.
352 */ 364 */
353 if ((offset & (OCFS2_SB(inode->i_sb)-> 365 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
354 return 0; 366 return 0;
355 367
356 status = ocfs2_get_clusters(inode, cpo 368 status = ocfs2_get_clusters(inode, cpos, &phys,
357 &num_clust 369 &num_clusters, &ext_flags);
358 if (status) { 370 if (status) {
359 mlog_errno(status); 371 mlog_errno(status);
360 goto out; 372 goto out;
361 } 373 }
362 374
363 if (!(ext_flags & OCFS2_EXT_REFCOUNTED 375 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
364 goto out; 376 goto out;
365 377
366 return ocfs2_refcount_cow(inode, fe_bh 378 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
367 379
368 out: 380 out:
369 return status; 381 return status;
370 } 382 }
371 383
372 static int ocfs2_orphan_for_truncate(struct oc 384 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
373 struct in 385 struct inode *inode,
374 struct bu 386 struct buffer_head *fe_bh,
375 u64 new_i 387 u64 new_i_size)
376 { 388 {
377 int status; 389 int status;
378 handle_t *handle; 390 handle_t *handle;
379 struct ocfs2_dinode *di; 391 struct ocfs2_dinode *di;
380 u64 cluster_bytes; 392 u64 cluster_bytes;
381 393
382 /* 394 /*
383 * We need to CoW the cluster contains 395 * We need to CoW the cluster contains the offset if it is reflinked
384 * since we will call ocfs2_zero_range 396 * since we will call ocfs2_zero_range_for_truncate later which will
385 * write "" from offset to the end of 397 * write "" from offset to the end of the cluster.
386 */ 398 */
387 status = ocfs2_cow_file_pos(inode, fe_ 399 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
388 if (status) { 400 if (status) {
389 mlog_errno(status); 401 mlog_errno(status);
390 return status; 402 return status;
391 } 403 }
392 404
393 /* TODO: This needs to actually orphan 405 /* TODO: This needs to actually orphan the inode in this
394 * transaction. */ 406 * transaction. */
395 407
396 handle = ocfs2_start_trans(osb, OCFS2_ 408 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
397 if (IS_ERR(handle)) { 409 if (IS_ERR(handle)) {
398 status = PTR_ERR(handle); 410 status = PTR_ERR(handle);
399 mlog_errno(status); 411 mlog_errno(status);
400 goto out; 412 goto out;
401 } 413 }
402 414
403 status = ocfs2_journal_access_di(handl 415 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
404 OCFS2 416 OCFS2_JOURNAL_ACCESS_WRITE);
405 if (status < 0) { 417 if (status < 0) {
406 mlog_errno(status); 418 mlog_errno(status);
407 goto out_commit; 419 goto out_commit;
408 } 420 }
409 421
410 /* 422 /*
411 * Do this before setting i_size. 423 * Do this before setting i_size.
412 */ 424 */
413 cluster_bytes = ocfs2_align_bytes_to_c 425 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
414 status = ocfs2_zero_range_for_truncate 426 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
415 427 cluster_bytes);
416 if (status) { 428 if (status) {
417 mlog_errno(status); 429 mlog_errno(status);
418 goto out_commit; 430 goto out_commit;
419 } 431 }
420 432
421 i_size_write(inode, new_i_size); 433 i_size_write(inode, new_i_size);
422 inode_set_mtime_to_ts(inode, inode_set !! 434 inode->i_ctime = inode->i_mtime = current_time(inode);
423 435
424 di = (struct ocfs2_dinode *) fe_bh->b_ 436 di = (struct ocfs2_dinode *) fe_bh->b_data;
425 di->i_size = cpu_to_le64(new_i_size); 437 di->i_size = cpu_to_le64(new_i_size);
426 di->i_ctime = di->i_mtime = cpu_to_le6 !! 438 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
427 di->i_ctime_nsec = di->i_mtime_nsec = !! 439 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
428 ocfs2_update_inode_fsync_trans(handle, 440 ocfs2_update_inode_fsync_trans(handle, inode, 0);
429 441
430 ocfs2_journal_dirty(handle, fe_bh); 442 ocfs2_journal_dirty(handle, fe_bh);
431 443
432 out_commit: 444 out_commit:
433 ocfs2_commit_trans(osb, handle); 445 ocfs2_commit_trans(osb, handle);
434 out: 446 out:
435 return status; 447 return status;
436 } 448 }
437 449
438 int ocfs2_truncate_file(struct inode *inode, 450 int ocfs2_truncate_file(struct inode *inode,
439 struct buffer_h 451 struct buffer_head *di_bh,
440 u64 new_i_size) 452 u64 new_i_size)
441 { 453 {
442 int status = 0; 454 int status = 0;
443 struct ocfs2_dinode *fe = NULL; 455 struct ocfs2_dinode *fe = NULL;
444 struct ocfs2_super *osb = OCFS2_SB(ino 456 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
445 457
446 /* We trust di_bh because it comes fro 458 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
447 * already validated it */ 459 * already validated it */
448 fe = (struct ocfs2_dinode *) di_bh->b_ 460 fe = (struct ocfs2_dinode *) di_bh->b_data;
449 461
450 trace_ocfs2_truncate_file((unsigned lo 462 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
451 (unsigned lo 463 (unsigned long long)le64_to_cpu(fe->i_size),
452 (unsigned lo 464 (unsigned long long)new_i_size);
453 465
454 mlog_bug_on_msg(le64_to_cpu(fe->i_size 466 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
455 "Inode %llu, inode i_s 467 "Inode %llu, inode i_size = %lld != di "
456 "i_size = %llu, i_flag 468 "i_size = %llu, i_flags = 0x%x\n",
457 (unsigned long long)OC 469 (unsigned long long)OCFS2_I(inode)->ip_blkno,
458 i_size_read(inode), 470 i_size_read(inode),
459 (unsigned long long)le 471 (unsigned long long)le64_to_cpu(fe->i_size),
460 le32_to_cpu(fe->i_flag 472 le32_to_cpu(fe->i_flags));
461 473
462 if (new_i_size > le64_to_cpu(fe->i_siz 474 if (new_i_size > le64_to_cpu(fe->i_size)) {
463 trace_ocfs2_truncate_file_erro 475 trace_ocfs2_truncate_file_error(
464 (unsigned long long)le 476 (unsigned long long)le64_to_cpu(fe->i_size),
465 (unsigned long long)ne 477 (unsigned long long)new_i_size);
466 status = -EINVAL; 478 status = -EINVAL;
467 mlog_errno(status); 479 mlog_errno(status);
468 goto bail; 480 goto bail;
469 } 481 }
470 482
471 down_write(&OCFS2_I(inode)->ip_alloc_s 483 down_write(&OCFS2_I(inode)->ip_alloc_sem);
472 484
473 ocfs2_resv_discard(&osb->osb_la_resmap 485 ocfs2_resv_discard(&osb->osb_la_resmap,
474 &OCFS2_I(inode)->ip 486 &OCFS2_I(inode)->ip_la_data_resv);
475 487
476 /* 488 /*
477 * The inode lock forced other nodes t 489 * The inode lock forced other nodes to sync and drop their
478 * pages, which (correctly) happens ev 490 * pages, which (correctly) happens even if we have a truncate
479 * without allocation change - ocfs2 c 491 * without allocation change - ocfs2 cluster sizes can be much
480 * greater than page size, so we have 492 * greater than page size, so we have to truncate them
481 * anyway. 493 * anyway.
482 */ 494 */
>> 495 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
>> 496 truncate_inode_pages(inode->i_mapping, new_i_size);
483 497
484 if (OCFS2_I(inode)->ip_dyn_features & 498 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
485 unmap_mapping_range(inode->i_m <<
486 new_i_size <<
487 truncate_inode_pages(inode->i_ <<
488 status = ocfs2_truncate_inline 499 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
489 500 i_size_read(inode), 1);
490 if (status) 501 if (status)
491 mlog_errno(status); 502 mlog_errno(status);
492 503
493 goto bail_unlock_sem; 504 goto bail_unlock_sem;
494 } 505 }
495 506
496 /* alright, we're going to need to do 507 /* alright, we're going to need to do a full blown alloc size
497 * change. Orphan the inode so that re 508 * change. Orphan the inode so that recovery can complete the
498 * truncate if necessary. This does th 509 * truncate if necessary. This does the task of marking
499 * i_size. */ 510 * i_size. */
500 status = ocfs2_orphan_for_truncate(osb 511 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
501 if (status < 0) { 512 if (status < 0) {
502 mlog_errno(status); 513 mlog_errno(status);
503 goto bail_unlock_sem; 514 goto bail_unlock_sem;
504 } 515 }
505 516
506 unmap_mapping_range(inode->i_mapping, <<
507 truncate_inode_pages(inode->i_mapping, <<
508 <<
509 status = ocfs2_commit_truncate(osb, in 517 status = ocfs2_commit_truncate(osb, inode, di_bh);
510 if (status < 0) { 518 if (status < 0) {
511 mlog_errno(status); 519 mlog_errno(status);
512 goto bail_unlock_sem; 520 goto bail_unlock_sem;
513 } 521 }
514 522
515 /* TODO: orphan dir cleanup here. */ 523 /* TODO: orphan dir cleanup here. */
516 bail_unlock_sem: 524 bail_unlock_sem:
517 up_write(&OCFS2_I(inode)->ip_alloc_sem 525 up_write(&OCFS2_I(inode)->ip_alloc_sem);
518 526
519 bail: 527 bail:
520 if (!status && OCFS2_I(inode)->ip_clus 528 if (!status && OCFS2_I(inode)->ip_clusters == 0)
521 status = ocfs2_try_remove_refc 529 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
522 530
523 return status; 531 return status;
524 } 532 }
525 533
526 /* 534 /*
527 * extend file allocation only here. 535 * extend file allocation only here.
528 * we'll update all the disk stuff, and oip->a 536 * we'll update all the disk stuff, and oip->alloc_size
529 * 537 *
530 * expect stuff to be locked, a transaction st 538 * expect stuff to be locked, a transaction started and enough data /
531 * metadata reservations in the contexts. 539 * metadata reservations in the contexts.
532 * 540 *
533 * Will return -EAGAIN, and a reason if a rest 541 * Will return -EAGAIN, and a reason if a restart is needed.
534 * If passed in, *reason will always be set, e 542 * If passed in, *reason will always be set, even in error.
535 */ 543 */
536 int ocfs2_add_inode_data(struct ocfs2_super *o 544 int ocfs2_add_inode_data(struct ocfs2_super *osb,
537 struct inode *inode, 545 struct inode *inode,
538 u32 *logical_offset, 546 u32 *logical_offset,
539 u32 clusters_to_add, 547 u32 clusters_to_add,
540 int mark_unwritten, 548 int mark_unwritten,
541 struct buffer_head *f 549 struct buffer_head *fe_bh,
542 handle_t *handle, 550 handle_t *handle,
543 struct ocfs2_alloc_co 551 struct ocfs2_alloc_context *data_ac,
544 struct ocfs2_alloc_co 552 struct ocfs2_alloc_context *meta_ac,
545 enum ocfs2_alloc_rest 553 enum ocfs2_alloc_restarted *reason_ret)
546 { 554 {
>> 555 int ret;
547 struct ocfs2_extent_tree et; 556 struct ocfs2_extent_tree et;
548 557
549 ocfs2_init_dinode_extent_tree(&et, INO 558 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
550 return ocfs2_add_clusters_in_btree(han !! 559 ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
551 clu !! 560 clusters_to_add, mark_unwritten,
552 dat !! 561 data_ac, meta_ac, reason_ret);
>> 562
>> 563 return ret;
553 } 564 }
554 565
555 static int ocfs2_extend_allocation(struct inod 566 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
556 u32 cluster 567 u32 clusters_to_add, int mark_unwritten)
557 { 568 {
558 int status = 0; 569 int status = 0;
559 int restart_func = 0; 570 int restart_func = 0;
560 int credits; 571 int credits;
561 u32 prev_clusters; 572 u32 prev_clusters;
562 struct buffer_head *bh = NULL; 573 struct buffer_head *bh = NULL;
563 struct ocfs2_dinode *fe = NULL; 574 struct ocfs2_dinode *fe = NULL;
564 handle_t *handle = NULL; 575 handle_t *handle = NULL;
565 struct ocfs2_alloc_context *data_ac = 576 struct ocfs2_alloc_context *data_ac = NULL;
566 struct ocfs2_alloc_context *meta_ac = 577 struct ocfs2_alloc_context *meta_ac = NULL;
567 enum ocfs2_alloc_restarted why = RESTA 578 enum ocfs2_alloc_restarted why = RESTART_NONE;
568 struct ocfs2_super *osb = OCFS2_SB(ino 579 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
569 struct ocfs2_extent_tree et; 580 struct ocfs2_extent_tree et;
570 int did_quota = 0; 581 int did_quota = 0;
571 582
572 /* 583 /*
573 * Unwritten extent only exists for fi 584 * Unwritten extent only exists for file systems which
574 * support holes. 585 * support holes.
575 */ 586 */
576 BUG_ON(mark_unwritten && !ocfs2_sparse 587 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
577 588
578 status = ocfs2_read_inode_block(inode, 589 status = ocfs2_read_inode_block(inode, &bh);
579 if (status < 0) { 590 if (status < 0) {
580 mlog_errno(status); 591 mlog_errno(status);
581 goto leave; 592 goto leave;
582 } 593 }
583 fe = (struct ocfs2_dinode *) bh->b_dat 594 fe = (struct ocfs2_dinode *) bh->b_data;
584 595
585 restart_all: 596 restart_all:
586 BUG_ON(le32_to_cpu(fe->i_clusters) != 597 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
587 598
588 ocfs2_init_dinode_extent_tree(&et, INO 599 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
589 status = ocfs2_lock_allocators(inode, 600 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
590 &data_a 601 &data_ac, &meta_ac);
591 if (status) { 602 if (status) {
592 mlog_errno(status); 603 mlog_errno(status);
593 goto leave; 604 goto leave;
594 } 605 }
595 606
596 credits = ocfs2_calc_extend_credits(os 607 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
597 handle = ocfs2_start_trans(osb, credit 608 handle = ocfs2_start_trans(osb, credits);
598 if (IS_ERR(handle)) { 609 if (IS_ERR(handle)) {
599 status = PTR_ERR(handle); 610 status = PTR_ERR(handle);
600 handle = NULL; 611 handle = NULL;
601 mlog_errno(status); 612 mlog_errno(status);
602 goto leave; 613 goto leave;
603 } 614 }
604 615
605 restarted_transaction: 616 restarted_transaction:
606 trace_ocfs2_extend_allocation( 617 trace_ocfs2_extend_allocation(
607 (unsigned long long)OCFS2_I(in 618 (unsigned long long)OCFS2_I(inode)->ip_blkno,
608 (unsigned long long)i_size_rea 619 (unsigned long long)i_size_read(inode),
609 le32_to_cpu(fe->i_clusters), c 620 le32_to_cpu(fe->i_clusters), clusters_to_add,
610 why, restart_func); 621 why, restart_func);
611 622
612 status = dquot_alloc_space_nodirty(ino 623 status = dquot_alloc_space_nodirty(inode,
613 ocfs2_clusters_to_byte 624 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
614 if (status) 625 if (status)
615 goto leave; 626 goto leave;
616 did_quota = 1; 627 did_quota = 1;
617 628
618 /* reserve a write to the file entry e 629 /* reserve a write to the file entry early on - that we if we
619 * run out of credits in the allocatio 630 * run out of credits in the allocation path, we can still
620 * update i_size. */ 631 * update i_size. */
621 status = ocfs2_journal_access_di(handl 632 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
622 OCFS2 633 OCFS2_JOURNAL_ACCESS_WRITE);
623 if (status < 0) { 634 if (status < 0) {
624 mlog_errno(status); 635 mlog_errno(status);
625 goto leave; 636 goto leave;
626 } 637 }
627 638
628 prev_clusters = OCFS2_I(inode)->ip_clu 639 prev_clusters = OCFS2_I(inode)->ip_clusters;
629 640
630 status = ocfs2_add_inode_data(osb, 641 status = ocfs2_add_inode_data(osb,
631 inode, 642 inode,
632 &logical 643 &logical_start,
633 clusters 644 clusters_to_add,
634 mark_unw 645 mark_unwritten,
635 bh, 646 bh,
636 handle, 647 handle,
637 data_ac, 648 data_ac,
638 meta_ac, 649 meta_ac,
639 &why); 650 &why);
640 if ((status < 0) && (status != -EAGAIN 651 if ((status < 0) && (status != -EAGAIN)) {
641 if (status != -ENOSPC) 652 if (status != -ENOSPC)
642 mlog_errno(status); 653 mlog_errno(status);
643 goto leave; 654 goto leave;
644 } 655 }
645 ocfs2_update_inode_fsync_trans(handle, 656 ocfs2_update_inode_fsync_trans(handle, inode, 1);
646 ocfs2_journal_dirty(handle, bh); 657 ocfs2_journal_dirty(handle, bh);
647 658
648 spin_lock(&OCFS2_I(inode)->ip_lock); 659 spin_lock(&OCFS2_I(inode)->ip_lock);
649 clusters_to_add -= (OCFS2_I(inode)->ip 660 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
650 spin_unlock(&OCFS2_I(inode)->ip_lock); 661 spin_unlock(&OCFS2_I(inode)->ip_lock);
651 /* Release unused quota reservation */ 662 /* Release unused quota reservation */
652 dquot_free_space(inode, 663 dquot_free_space(inode,
653 ocfs2_clusters_to_byte 664 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
654 did_quota = 0; 665 did_quota = 0;
655 666
656 if (why != RESTART_NONE && clusters_to 667 if (why != RESTART_NONE && clusters_to_add) {
657 if (why == RESTART_META) { 668 if (why == RESTART_META) {
658 restart_func = 1; 669 restart_func = 1;
659 status = 0; 670 status = 0;
660 } else { 671 } else {
661 BUG_ON(why != RESTART_ 672 BUG_ON(why != RESTART_TRANS);
662 673
663 status = ocfs2_allocat 674 status = ocfs2_allocate_extend_trans(handle, 1);
664 if (status < 0) { 675 if (status < 0) {
665 /* handle stil 676 /* handle still has to be committed at
666 * this point. 677 * this point. */
667 status = -ENOM 678 status = -ENOMEM;
668 mlog_errno(sta 679 mlog_errno(status);
669 goto leave; 680 goto leave;
670 } 681 }
671 goto restarted_transac 682 goto restarted_transaction;
672 } 683 }
673 } 684 }
674 685
675 trace_ocfs2_extend_allocation_end(OCFS 686 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
676 le32_to_cpu(fe->i_clusters), 687 le32_to_cpu(fe->i_clusters),
677 (unsigned long long)le64_to_cpu(f 688 (unsigned long long)le64_to_cpu(fe->i_size),
678 OCFS2_I(inode)->ip_clusters, 689 OCFS2_I(inode)->ip_clusters,
679 (unsigned long long)i_size_read(i 690 (unsigned long long)i_size_read(inode));
680 691
681 leave: 692 leave:
682 if (status < 0 && did_quota) 693 if (status < 0 && did_quota)
683 dquot_free_space(inode, 694 dquot_free_space(inode,
684 ocfs2_clusters_to_byte 695 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
685 if (handle) { 696 if (handle) {
686 ocfs2_commit_trans(osb, handle 697 ocfs2_commit_trans(osb, handle);
687 handle = NULL; 698 handle = NULL;
688 } 699 }
689 if (data_ac) { 700 if (data_ac) {
690 ocfs2_free_alloc_context(data_ 701 ocfs2_free_alloc_context(data_ac);
691 data_ac = NULL; 702 data_ac = NULL;
692 } 703 }
693 if (meta_ac) { 704 if (meta_ac) {
694 ocfs2_free_alloc_context(meta_ 705 ocfs2_free_alloc_context(meta_ac);
695 meta_ac = NULL; 706 meta_ac = NULL;
696 } 707 }
697 if ((!status) && restart_func) { 708 if ((!status) && restart_func) {
698 restart_func = 0; 709 restart_func = 0;
699 goto restart_all; 710 goto restart_all;
700 } 711 }
701 brelse(bh); 712 brelse(bh);
702 bh = NULL; 713 bh = NULL;
703 714
704 return status; 715 return status;
705 } 716 }
706 717
707 /* 718 /*
708 * While a write will already be ordering the 719 * While a write will already be ordering the data, a truncate will not.
709 * Thus, we need to explicitly order the zeroe 720 * Thus, we need to explicitly order the zeroed pages.
710 */ 721 */
711 static handle_t *ocfs2_zero_start_ordered_tran 722 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
712 !! 723 struct buffer_head *di_bh)
713 <<
714 <<
715 { 724 {
716 struct ocfs2_super *osb = OCFS2_SB(ino 725 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
717 handle_t *handle = NULL; 726 handle_t *handle = NULL;
718 int ret = 0; 727 int ret = 0;
719 728
720 if (!ocfs2_should_order_data(inode)) 729 if (!ocfs2_should_order_data(inode))
721 goto out; 730 goto out;
722 731
723 handle = ocfs2_start_trans(osb, OCFS2_ 732 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
724 if (IS_ERR(handle)) { 733 if (IS_ERR(handle)) {
725 ret = -ENOMEM; 734 ret = -ENOMEM;
726 mlog_errno(ret); 735 mlog_errno(ret);
727 goto out; 736 goto out;
728 } 737 }
729 738
730 ret = ocfs2_jbd2_inode_add_write(handl !! 739 ret = ocfs2_jbd2_file_inode(handle, inode);
731 if (ret < 0) { 740 if (ret < 0) {
732 mlog_errno(ret); 741 mlog_errno(ret);
733 goto out; 742 goto out;
734 } 743 }
735 744
736 ret = ocfs2_journal_access_di(handle, 745 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
737 OCFS2_JO 746 OCFS2_JOURNAL_ACCESS_WRITE);
738 if (ret) 747 if (ret)
739 mlog_errno(ret); 748 mlog_errno(ret);
740 ocfs2_update_inode_fsync_trans(handle, 749 ocfs2_update_inode_fsync_trans(handle, inode, 1);
741 750
742 out: 751 out:
743 if (ret) { 752 if (ret) {
744 if (!IS_ERR(handle)) 753 if (!IS_ERR(handle))
745 ocfs2_commit_trans(osb 754 ocfs2_commit_trans(osb, handle);
746 handle = ERR_PTR(ret); 755 handle = ERR_PTR(ret);
747 } 756 }
748 return handle; 757 return handle;
749 } 758 }
750 759
751 /* Some parts of this taken from generic_cont_ 760 /* Some parts of this taken from generic_cont_expand, which turned out
752 * to be too fragile to do exactly what we nee 761 * to be too fragile to do exactly what we need without us having to
753 * worry about recursive locking in ->write_be 762 * worry about recursive locking in ->write_begin() and ->write_end(). */
754 static int ocfs2_write_zero_page(struct inode 763 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
755 u64 abs_to, s 764 u64 abs_to, struct buffer_head *di_bh)
756 { 765 {
757 struct address_space *mapping = inode- 766 struct address_space *mapping = inode->i_mapping;
758 struct folio *folio; !! 767 struct page *page;
759 unsigned long index = abs_from >> PAGE 768 unsigned long index = abs_from >> PAGE_SHIFT;
760 handle_t *handle; 769 handle_t *handle;
761 int ret = 0; 770 int ret = 0;
762 unsigned zero_from, zero_to, block_sta 771 unsigned zero_from, zero_to, block_start, block_end;
763 struct ocfs2_dinode *di = (struct ocfs 772 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
764 773
765 BUG_ON(abs_from >= abs_to); 774 BUG_ON(abs_from >= abs_to);
766 BUG_ON(abs_to > (((u64)index + 1) << P 775 BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
767 BUG_ON(abs_from & (inode->i_blkbits - 776 BUG_ON(abs_from & (inode->i_blkbits - 1));
768 777
769 handle = ocfs2_zero_start_ordered_tran !! 778 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
770 <<
771 <<
772 if (IS_ERR(handle)) { 779 if (IS_ERR(handle)) {
773 ret = PTR_ERR(handle); 780 ret = PTR_ERR(handle);
774 goto out; 781 goto out;
775 } 782 }
776 783
777 folio = __filemap_get_folio(mapping, i !! 784 page = find_or_create_page(mapping, index, GFP_NOFS);
778 FGP_LOCK | FGP_ACCESSE !! 785 if (!page) {
779 if (IS_ERR(folio)) { !! 786 ret = -ENOMEM;
780 ret = PTR_ERR(folio); <<
781 mlog_errno(ret); 787 mlog_errno(ret);
782 goto out_commit_trans; 788 goto out_commit_trans;
783 } 789 }
784 790
785 /* Get the offsets within the page tha 791 /* Get the offsets within the page that we want to zero */
786 zero_from = abs_from & (PAGE_SIZE - 1) 792 zero_from = abs_from & (PAGE_SIZE - 1);
787 zero_to = abs_to & (PAGE_SIZE - 1); 793 zero_to = abs_to & (PAGE_SIZE - 1);
788 if (!zero_to) 794 if (!zero_to)
789 zero_to = PAGE_SIZE; 795 zero_to = PAGE_SIZE;
790 796
791 trace_ocfs2_write_zero_page( 797 trace_ocfs2_write_zero_page(
792 (unsigned long long)OC 798 (unsigned long long)OCFS2_I(inode)->ip_blkno,
793 (unsigned long long)ab 799 (unsigned long long)abs_from,
794 (unsigned long long)ab 800 (unsigned long long)abs_to,
795 index, zero_from, zero 801 index, zero_from, zero_to);
796 802
797 /* We know that zero_from is block ali 803 /* We know that zero_from is block aligned */
798 for (block_start = zero_from; block_st 804 for (block_start = zero_from; block_start < zero_to;
799 block_start = block_end) { 805 block_start = block_end) {
800 block_end = block_start + i_bl 806 block_end = block_start + i_blocksize(inode);
801 807
802 /* 808 /*
803 * block_start is block-aligne 809 * block_start is block-aligned. Bump it by one to force
804 * __block_write_begin and blo 810 * __block_write_begin and block_commit_write to zero the
805 * whole block. 811 * whole block.
806 */ 812 */
807 ret = __block_write_begin(foli !! 813 ret = __block_write_begin(page, block_start + 1, 0,
808 ocfs 814 ocfs2_get_block);
809 if (ret < 0) { 815 if (ret < 0) {
810 mlog_errno(ret); 816 mlog_errno(ret);
811 goto out_unlock; 817 goto out_unlock;
812 } 818 }
813 819
814 820
815 /* must not update i_size! */ 821 /* must not update i_size! */
816 block_commit_write(&folio->pag !! 822 ret = block_commit_write(page, block_start + 1,
>> 823 block_start + 1);
>> 824 if (ret < 0)
>> 825 mlog_errno(ret);
>> 826 else
>> 827 ret = 0;
817 } 828 }
818 829
819 /* 830 /*
820 * fs-writeback will release the dirty 831 * fs-writeback will release the dirty pages without page lock
821 * whose offset are over inode size, t 832 * whose offset are over inode size, the release happens at
822 * block_write_full_folio(). !! 833 * block_write_full_page().
823 */ 834 */
824 i_size_write(inode, abs_to); 835 i_size_write(inode, abs_to);
825 inode->i_blocks = ocfs2_inode_sector_c 836 inode->i_blocks = ocfs2_inode_sector_count(inode);
826 di->i_size = cpu_to_le64((u64)i_size_r 837 di->i_size = cpu_to_le64((u64)i_size_read(inode));
827 inode_set_mtime_to_ts(inode, inode_set !! 838 inode->i_mtime = inode->i_ctime = current_time(inode);
828 di->i_mtime = di->i_ctime = cpu_to_le6 !! 839 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
829 di->i_ctime_nsec = cpu_to_le32(inode_g !! 840 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
830 di->i_mtime_nsec = di->i_ctime_nsec; 841 di->i_mtime_nsec = di->i_ctime_nsec;
831 if (handle) { 842 if (handle) {
832 ocfs2_journal_dirty(handle, di 843 ocfs2_journal_dirty(handle, di_bh);
833 ocfs2_update_inode_fsync_trans 844 ocfs2_update_inode_fsync_trans(handle, inode, 1);
834 } 845 }
835 846
836 out_unlock: 847 out_unlock:
837 folio_unlock(folio); !! 848 unlock_page(page);
838 folio_put(folio); !! 849 put_page(page);
839 out_commit_trans: 850 out_commit_trans:
840 if (handle) 851 if (handle)
841 ocfs2_commit_trans(OCFS2_SB(in 852 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
842 out: 853 out:
843 return ret; 854 return ret;
844 } 855 }
845 856
846 /* 857 /*
847 * Find the next range to zero. We do this in 858 * Find the next range to zero. We do this in terms of bytes because
848 * that's what ocfs2_zero_extend() wants, and 859 * that's what ocfs2_zero_extend() wants, and it is dealing with the
849 * pagecache. We may return multiple extents. 860 * pagecache. We may return multiple extents.
850 * 861 *
851 * zero_start and zero_end are ocfs2_zero_exte 862 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
852 * needs to be zeroed. range_start and range_ 863 * needs to be zeroed. range_start and range_end return the next zeroing
853 * range. A subsequent call should pass the p 864 * range. A subsequent call should pass the previous range_end as its
854 * zero_start. If range_end is 0, there's not 865 * zero_start. If range_end is 0, there's nothing to do.
855 * 866 *
856 * Unwritten extents are skipped over. Refcou 867 * Unwritten extents are skipped over. Refcounted extents are CoWd.
857 */ 868 */
858 static int ocfs2_zero_extend_get_range(struct 869 static int ocfs2_zero_extend_get_range(struct inode *inode,
859 struct 870 struct buffer_head *di_bh,
860 u64 zer 871 u64 zero_start, u64 zero_end,
861 u64 *ra 872 u64 *range_start, u64 *range_end)
862 { 873 {
863 int rc = 0, needs_cow = 0; 874 int rc = 0, needs_cow = 0;
864 u32 p_cpos, zero_clusters = 0; 875 u32 p_cpos, zero_clusters = 0;
865 u32 zero_cpos = 876 u32 zero_cpos =
866 zero_start >> OCFS2_SB(inode-> 877 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
867 u32 last_cpos = ocfs2_clusters_for_byt 878 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
868 unsigned int num_clusters = 0; 879 unsigned int num_clusters = 0;
869 unsigned int ext_flags = 0; 880 unsigned int ext_flags = 0;
870 881
871 while (zero_cpos < last_cpos) { 882 while (zero_cpos < last_cpos) {
872 rc = ocfs2_get_clusters(inode, 883 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
873 &num_c 884 &num_clusters, &ext_flags);
874 if (rc) { 885 if (rc) {
875 mlog_errno(rc); 886 mlog_errno(rc);
876 goto out; 887 goto out;
877 } 888 }
878 889
879 if (p_cpos && !(ext_flags & OC 890 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
880 zero_clusters = num_cl 891 zero_clusters = num_clusters;
881 if (ext_flags & OCFS2_ 892 if (ext_flags & OCFS2_EXT_REFCOUNTED)
882 needs_cow = 1; 893 needs_cow = 1;
883 break; 894 break;
884 } 895 }
885 896
886 zero_cpos += num_clusters; 897 zero_cpos += num_clusters;
887 } 898 }
888 if (!zero_clusters) { 899 if (!zero_clusters) {
889 *range_end = 0; 900 *range_end = 0;
890 goto out; 901 goto out;
891 } 902 }
892 903
893 while ((zero_cpos + zero_clusters) < l 904 while ((zero_cpos + zero_clusters) < last_cpos) {
894 rc = ocfs2_get_clusters(inode, 905 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
895 &p_cpo 906 &p_cpos, &num_clusters,
896 &ext_f 907 &ext_flags);
897 if (rc) { 908 if (rc) {
898 mlog_errno(rc); 909 mlog_errno(rc);
899 goto out; 910 goto out;
900 } 911 }
901 912
902 if (!p_cpos || (ext_flags & OC 913 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
903 break; 914 break;
904 if (ext_flags & OCFS2_EXT_REFC 915 if (ext_flags & OCFS2_EXT_REFCOUNTED)
905 needs_cow = 1; 916 needs_cow = 1;
906 zero_clusters += num_clusters; 917 zero_clusters += num_clusters;
907 } 918 }
908 if ((zero_cpos + zero_clusters) > last 919 if ((zero_cpos + zero_clusters) > last_cpos)
909 zero_clusters = last_cpos - ze 920 zero_clusters = last_cpos - zero_cpos;
910 921
911 if (needs_cow) { 922 if (needs_cow) {
912 rc = ocfs2_refcount_cow(inode, 923 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
913 zero_c 924 zero_clusters, UINT_MAX);
914 if (rc) { 925 if (rc) {
915 mlog_errno(rc); 926 mlog_errno(rc);
916 goto out; 927 goto out;
917 } 928 }
918 } 929 }
919 930
920 *range_start = ocfs2_clusters_to_bytes 931 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
921 *range_end = ocfs2_clusters_to_bytes(i 932 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
922 z 933 zero_cpos + zero_clusters);
923 934
924 out: 935 out:
925 return rc; 936 return rc;
926 } 937 }
927 938
928 /* 939 /*
929 * Zero one range returned from ocfs2_zero_ext 940 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
930 * has made sure that the entire range needs z 941 * has made sure that the entire range needs zeroing.
931 */ 942 */
932 static int ocfs2_zero_extend_range(struct inod 943 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
933 u64 range_e 944 u64 range_end, struct buffer_head *di_bh)
934 { 945 {
935 int rc = 0; 946 int rc = 0;
936 u64 next_pos; 947 u64 next_pos;
937 u64 zero_pos = range_start; 948 u64 zero_pos = range_start;
938 949
939 trace_ocfs2_zero_extend_range( 950 trace_ocfs2_zero_extend_range(
940 (unsigned long long)OC 951 (unsigned long long)OCFS2_I(inode)->ip_blkno,
941 (unsigned long long)ra 952 (unsigned long long)range_start,
942 (unsigned long long)ra 953 (unsigned long long)range_end);
943 BUG_ON(range_start >= range_end); 954 BUG_ON(range_start >= range_end);
944 955
945 while (zero_pos < range_end) { 956 while (zero_pos < range_end) {
946 next_pos = (zero_pos & PAGE_MA 957 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
947 if (next_pos > range_end) 958 if (next_pos > range_end)
948 next_pos = range_end; 959 next_pos = range_end;
949 rc = ocfs2_write_zero_page(ino 960 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
950 if (rc < 0) { 961 if (rc < 0) {
951 mlog_errno(rc); 962 mlog_errno(rc);
952 break; 963 break;
953 } 964 }
954 zero_pos = next_pos; 965 zero_pos = next_pos;
955 966
956 /* 967 /*
957 * Very large extends have the 968 * Very large extends have the potential to lock up
958 * the cpu for extended period 969 * the cpu for extended periods of time.
959 */ 970 */
960 cond_resched(); 971 cond_resched();
961 } 972 }
962 973
963 return rc; 974 return rc;
964 } 975 }
965 976
966 int ocfs2_zero_extend(struct inode *inode, str 977 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
967 loff_t zero_to_size) 978 loff_t zero_to_size)
968 { 979 {
969 int ret = 0; 980 int ret = 0;
970 u64 zero_start, range_start = 0, range 981 u64 zero_start, range_start = 0, range_end = 0;
971 struct super_block *sb = inode->i_sb; 982 struct super_block *sb = inode->i_sb;
972 983
973 zero_start = ocfs2_align_bytes_to_bloc 984 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
974 trace_ocfs2_zero_extend((unsigned long 985 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
975 (unsigned long 986 (unsigned long long)zero_start,
976 (unsigned long 987 (unsigned long long)i_size_read(inode));
977 while (zero_start < zero_to_size) { 988 while (zero_start < zero_to_size) {
978 ret = ocfs2_zero_extend_get_ra 989 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
979 990 zero_to_size,
980 991 &range_start,
981 992 &range_end);
982 if (ret) { 993 if (ret) {
983 mlog_errno(ret); 994 mlog_errno(ret);
984 break; 995 break;
985 } 996 }
986 if (!range_end) 997 if (!range_end)
987 break; 998 break;
988 /* Trim the ends */ 999 /* Trim the ends */
989 if (range_start < zero_start) 1000 if (range_start < zero_start)
990 range_start = zero_sta 1001 range_start = zero_start;
991 if (range_end > zero_to_size) 1002 if (range_end > zero_to_size)
992 range_end = zero_to_si 1003 range_end = zero_to_size;
993 1004
994 ret = ocfs2_zero_extend_range( 1005 ret = ocfs2_zero_extend_range(inode, range_start,
995 1006 range_end, di_bh);
996 if (ret) { 1007 if (ret) {
997 mlog_errno(ret); 1008 mlog_errno(ret);
998 break; 1009 break;
999 } 1010 }
1000 zero_start = range_end; 1011 zero_start = range_end;
1001 } 1012 }
1002 1013
1003 return ret; 1014 return ret;
1004 } 1015 }
1005 1016
1006 int ocfs2_extend_no_holes(struct inode *inode 1017 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1007 u64 new_i_size, u64 1018 u64 new_i_size, u64 zero_to)
1008 { 1019 {
1009 int ret; 1020 int ret;
1010 u32 clusters_to_add; 1021 u32 clusters_to_add;
1011 struct ocfs2_inode_info *oi = OCFS2_I 1022 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1012 1023
1013 /* 1024 /*
1014 * Only quota files call this without 1025 * Only quota files call this without a bh, and they can't be
1015 * refcounted. 1026 * refcounted.
1016 */ 1027 */
1017 BUG_ON(!di_bh && ocfs2_is_refcount_in 1028 BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1018 BUG_ON(!di_bh && !(oi->ip_flags & OCF 1029 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1019 1030
1020 clusters_to_add = ocfs2_clusters_for_ 1031 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1021 if (clusters_to_add < oi->ip_clusters 1032 if (clusters_to_add < oi->ip_clusters)
1022 clusters_to_add = 0; 1033 clusters_to_add = 0;
1023 else 1034 else
1024 clusters_to_add -= oi->ip_clu 1035 clusters_to_add -= oi->ip_clusters;
1025 1036
1026 if (clusters_to_add) { 1037 if (clusters_to_add) {
1027 ret = ocfs2_extend_allocation 1038 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1028 1039 clusters_to_add, 0);
1029 if (ret) { 1040 if (ret) {
1030 mlog_errno(ret); 1041 mlog_errno(ret);
1031 goto out; 1042 goto out;
1032 } 1043 }
1033 } 1044 }
1034 1045
1035 /* 1046 /*
1036 * Call this even if we don't add any 1047 * Call this even if we don't add any clusters to the tree. We
1037 * still need to zero the area betwee 1048 * still need to zero the area between the old i_size and the
1038 * new i_size. 1049 * new i_size.
1039 */ 1050 */
1040 ret = ocfs2_zero_extend(inode, di_bh, 1051 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1041 if (ret < 0) 1052 if (ret < 0)
1042 mlog_errno(ret); 1053 mlog_errno(ret);
1043 1054
1044 out: 1055 out:
1045 return ret; 1056 return ret;
1046 } 1057 }
1047 1058
1048 static int ocfs2_extend_file(struct inode *in 1059 static int ocfs2_extend_file(struct inode *inode,
1049 struct buffer_he 1060 struct buffer_head *di_bh,
1050 u64 new_i_size) 1061 u64 new_i_size)
1051 { 1062 {
1052 int ret = 0; 1063 int ret = 0;
1053 struct ocfs2_inode_info *oi = OCFS2_I 1064 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1054 1065
1055 BUG_ON(!di_bh); 1066 BUG_ON(!di_bh);
1056 1067
1057 /* setattr sometimes calls us like th 1068 /* setattr sometimes calls us like this. */
1058 if (new_i_size == 0) 1069 if (new_i_size == 0)
1059 goto out; 1070 goto out;
1060 1071
1061 if (i_size_read(inode) == new_i_size) 1072 if (i_size_read(inode) == new_i_size)
1062 goto out; 1073 goto out;
1063 BUG_ON(new_i_size < i_size_read(inode 1074 BUG_ON(new_i_size < i_size_read(inode));
1064 1075
1065 /* 1076 /*
1066 * The alloc sem blocks people in rea 1077 * The alloc sem blocks people in read/write from reading our
1067 * allocation until we're done changi 1078 * allocation until we're done changing it. We depend on
1068 * i_rwsem to block other extend/trun !! 1079 * i_mutex to block other extend/truncate calls while we're
1069 * here. We even have to hold it for 1080 * here. We even have to hold it for sparse files because there
1070 * might be some tail zeroing. 1081 * might be some tail zeroing.
1071 */ 1082 */
1072 down_write(&oi->ip_alloc_sem); 1083 down_write(&oi->ip_alloc_sem);
1073 1084
1074 if (oi->ip_dyn_features & OCFS2_INLIN 1085 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1075 /* 1086 /*
1076 * We can optimize small exte 1087 * We can optimize small extends by keeping the inodes
1077 * inline data. 1088 * inline data.
1078 */ 1089 */
1079 if (ocfs2_size_fits_inline_da 1090 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1080 up_write(&oi->ip_allo 1091 up_write(&oi->ip_alloc_sem);
1081 goto out_update_size; 1092 goto out_update_size;
1082 } 1093 }
1083 1094
1084 ret = ocfs2_convert_inline_da 1095 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1085 if (ret) { 1096 if (ret) {
1086 up_write(&oi->ip_allo 1097 up_write(&oi->ip_alloc_sem);
1087 mlog_errno(ret); 1098 mlog_errno(ret);
1088 goto out; 1099 goto out;
1089 } 1100 }
1090 } 1101 }
1091 1102
1092 if (ocfs2_sparse_alloc(OCFS2_SB(inode 1103 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1093 ret = ocfs2_zero_extend(inode 1104 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1094 else 1105 else
1095 ret = ocfs2_extend_no_holes(i 1106 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1096 n 1107 new_i_size);
1097 1108
1098 up_write(&oi->ip_alloc_sem); 1109 up_write(&oi->ip_alloc_sem);
1099 1110
1100 if (ret < 0) { 1111 if (ret < 0) {
1101 mlog_errno(ret); 1112 mlog_errno(ret);
1102 goto out; 1113 goto out;
1103 } 1114 }
1104 1115
1105 out_update_size: 1116 out_update_size:
1106 ret = ocfs2_simple_size_update(inode, 1117 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1107 if (ret < 0) 1118 if (ret < 0)
1108 mlog_errno(ret); 1119 mlog_errno(ret);
1109 1120
1110 out: 1121 out:
1111 return ret; 1122 return ret;
1112 } 1123 }
1113 1124
1114 int ocfs2_setattr(struct mnt_idmap *idmap, st !! 1125 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1115 struct iattr *attr) <<
1116 { 1126 {
1117 int status = 0, size_change; 1127 int status = 0, size_change;
1118 int inode_locked = 0; 1128 int inode_locked = 0;
1119 struct inode *inode = d_inode(dentry) 1129 struct inode *inode = d_inode(dentry);
1120 struct super_block *sb = inode->i_sb; 1130 struct super_block *sb = inode->i_sb;
1121 struct ocfs2_super *osb = OCFS2_SB(sb 1131 struct ocfs2_super *osb = OCFS2_SB(sb);
1122 struct buffer_head *bh = NULL; 1132 struct buffer_head *bh = NULL;
1123 handle_t *handle = NULL; 1133 handle_t *handle = NULL;
1124 struct dquot *transfer_to[MAXQUOTAS] 1134 struct dquot *transfer_to[MAXQUOTAS] = { };
1125 int qtype; 1135 int qtype;
1126 int had_lock; 1136 int had_lock;
1127 struct ocfs2_lock_holder oh; 1137 struct ocfs2_lock_holder oh;
1128 1138
1129 trace_ocfs2_setattr(inode, dentry, 1139 trace_ocfs2_setattr(inode, dentry,
1130 (unsigned long lo 1140 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1131 dentry->d_name.le 1141 dentry->d_name.len, dentry->d_name.name,
1132 attr->ia_valid, !! 1142 attr->ia_valid, attr->ia_mode,
1133 attr->ia_vali !! 1143 from_kuid(&init_user_ns, attr->ia_uid),
1134 attr->ia_vali !! 1144 from_kgid(&init_user_ns, attr->ia_gid));
1135 from_ <<
1136 attr->ia_vali <<
1137 from_ <<
1138 1145
1139 /* ensuring we don't even attempt to 1146 /* ensuring we don't even attempt to truncate a symlink */
1140 if (S_ISLNK(inode->i_mode)) 1147 if (S_ISLNK(inode->i_mode))
1141 attr->ia_valid &= ~ATTR_SIZE; 1148 attr->ia_valid &= ~ATTR_SIZE;
1142 1149
1143 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_ 1150 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1144 | ATTR_GID | ATTR_ 1151 | ATTR_GID | ATTR_UID | ATTR_MODE)
1145 if (!(attr->ia_valid & OCFS2_VALID_AT 1152 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1146 return 0; 1153 return 0;
1147 1154
1148 status = setattr_prepare(&nop_mnt_idm !! 1155 status = setattr_prepare(dentry, attr);
1149 if (status) 1156 if (status)
1150 return status; 1157 return status;
1151 1158
1152 if (is_quota_modification(&nop_mnt_id !! 1159 if (is_quota_modification(inode, attr)) {
1153 status = dquot_initialize(ino 1160 status = dquot_initialize(inode);
1154 if (status) 1161 if (status)
1155 return status; 1162 return status;
1156 } 1163 }
1157 size_change = S_ISREG(inode->i_mode) 1164 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1158 if (size_change) { 1165 if (size_change) {
1159 /* 1166 /*
1160 * Here we should wait dio to 1167 * Here we should wait dio to finish before inode lock
1161 * to avoid a deadlock betwee 1168 * to avoid a deadlock between ocfs2_setattr() and
1162 * ocfs2_dio_end_io_write() 1169 * ocfs2_dio_end_io_write()
1163 */ 1170 */
1164 inode_dio_wait(inode); 1171 inode_dio_wait(inode);
1165 1172
1166 status = ocfs2_rw_lock(inode, 1173 status = ocfs2_rw_lock(inode, 1);
1167 if (status < 0) { 1174 if (status < 0) {
1168 mlog_errno(status); 1175 mlog_errno(status);
1169 goto bail; 1176 goto bail;
1170 } 1177 }
1171 } 1178 }
1172 1179
1173 had_lock = ocfs2_inode_lock_tracker(i 1180 had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1174 if (had_lock < 0) { 1181 if (had_lock < 0) {
1175 status = had_lock; 1182 status = had_lock;
1176 goto bail_unlock_rw; 1183 goto bail_unlock_rw;
1177 } else if (had_lock) { 1184 } else if (had_lock) {
1178 /* 1185 /*
1179 * As far as we know, ocfs2_s 1186 * As far as we know, ocfs2_setattr() could only be the first
1180 * VFS entry point in the cal 1187 * VFS entry point in the call chain of recursive cluster
1181 * locking issue. 1188 * locking issue.
1182 * 1189 *
1183 * For instance: 1190 * For instance:
1184 * chmod_common() 1191 * chmod_common()
1185 * notify_change() 1192 * notify_change()
1186 * ocfs2_setattr() 1193 * ocfs2_setattr()
1187 * posix_acl_chmod() 1194 * posix_acl_chmod()
1188 * ocfs2_iop_get_acl() 1195 * ocfs2_iop_get_acl()
1189 * 1196 *
1190 * But, we're not 100% sure i 1197 * But, we're not 100% sure if it's always true, because the
1191 * ordering of the VFS entry 1198 * ordering of the VFS entry points in the call chain is out
1192 * of our control. So, we'd b 1199 * of our control. So, we'd better dump the stack here to
1193 * catch the other cases of r 1200 * catch the other cases of recursive locking.
1194 */ 1201 */
1195 mlog(ML_ERROR, "Another case 1202 mlog(ML_ERROR, "Another case of recursive locking:\n");
1196 dump_stack(); 1203 dump_stack();
1197 } 1204 }
1198 inode_locked = 1; 1205 inode_locked = 1;
1199 1206
1200 if (size_change) { 1207 if (size_change) {
1201 status = inode_newsize_ok(ino 1208 status = inode_newsize_ok(inode, attr->ia_size);
1202 if (status) 1209 if (status)
1203 goto bail_unlock; 1210 goto bail_unlock;
1204 1211
1205 if (i_size_read(inode) >= att 1212 if (i_size_read(inode) >= attr->ia_size) {
1206 if (ocfs2_should_orde 1213 if (ocfs2_should_order_data(inode)) {
1207 status = ocfs 1214 status = ocfs2_begin_ordered_truncate(inode,
1208 1215 attr->ia_size);
1209 if (status) 1216 if (status)
1210 goto 1217 goto bail_unlock;
1211 } 1218 }
1212 status = ocfs2_trunca 1219 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1213 } else 1220 } else
1214 status = ocfs2_extend 1221 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1215 if (status < 0) { 1222 if (status < 0) {
1216 if (status != -ENOSPC 1223 if (status != -ENOSPC)
1217 mlog_errno(st 1224 mlog_errno(status);
1218 status = -ENOSPC; 1225 status = -ENOSPC;
1219 goto bail_unlock; 1226 goto bail_unlock;
1220 } 1227 }
1221 } 1228 }
1222 1229
1223 if ((attr->ia_valid & ATTR_UID && !ui 1230 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1224 (attr->ia_valid & ATTR_GID && !gi 1231 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1225 /* 1232 /*
1226 * Gather pointers to quota s 1233 * Gather pointers to quota structures so that allocation /
1227 * freeing of quota structure 1234 * freeing of quota structures happens here and not inside
1228 * dquot_transfer() where we 1235 * dquot_transfer() where we have problems with lock ordering
1229 */ 1236 */
1230 if (attr->ia_valid & ATTR_UID 1237 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1231 && OCFS2_HAS_RO_COMPAT_FE 1238 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1232 OCFS2_FEATURE_RO_COMPAT_U 1239 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1233 transfer_to[USRQUOTA] 1240 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1234 if (IS_ERR(transfer_t 1241 if (IS_ERR(transfer_to[USRQUOTA])) {
1235 status = PTR_ 1242 status = PTR_ERR(transfer_to[USRQUOTA]);
1236 transfer_to[U <<
1237 goto bail_unl 1243 goto bail_unlock;
1238 } 1244 }
1239 } 1245 }
1240 if (attr->ia_valid & ATTR_GID 1246 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1241 && OCFS2_HAS_RO_COMPAT_FE 1247 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1242 OCFS2_FEATURE_RO_COMPAT_G 1248 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1243 transfer_to[GRPQUOTA] 1249 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1244 if (IS_ERR(transfer_t 1250 if (IS_ERR(transfer_to[GRPQUOTA])) {
1245 status = PTR_ 1251 status = PTR_ERR(transfer_to[GRPQUOTA]);
1246 transfer_to[G <<
1247 goto bail_unl 1252 goto bail_unlock;
1248 } 1253 }
1249 } 1254 }
1250 down_write(&OCFS2_I(inode)->i <<
1251 handle = ocfs2_start_trans(os 1255 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1252 2 1256 2 * ocfs2_quota_trans_credits(sb));
1253 if (IS_ERR(handle)) { 1257 if (IS_ERR(handle)) {
1254 status = PTR_ERR(hand 1258 status = PTR_ERR(handle);
1255 mlog_errno(status); 1259 mlog_errno(status);
1256 goto bail_unlock_allo !! 1260 goto bail_unlock;
1257 } 1261 }
1258 status = __dquot_transfer(ino 1262 status = __dquot_transfer(inode, transfer_to);
1259 if (status < 0) 1263 if (status < 0)
1260 goto bail_commit; 1264 goto bail_commit;
1261 } else { 1265 } else {
1262 down_write(&OCFS2_I(inode)->i <<
1263 handle = ocfs2_start_trans(os 1266 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1264 if (IS_ERR(handle)) { 1267 if (IS_ERR(handle)) {
1265 status = PTR_ERR(hand 1268 status = PTR_ERR(handle);
1266 mlog_errno(status); 1269 mlog_errno(status);
1267 goto bail_unlock_allo !! 1270 goto bail_unlock;
1268 } 1271 }
1269 } 1272 }
1270 1273
1271 setattr_copy(&nop_mnt_idmap, inode, a !! 1274 setattr_copy(inode, attr);
1272 mark_inode_dirty(inode); 1275 mark_inode_dirty(inode);
1273 1276
1274 status = ocfs2_mark_inode_dirty(handl 1277 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1275 if (status < 0) 1278 if (status < 0)
1276 mlog_errno(status); 1279 mlog_errno(status);
1277 1280
1278 bail_commit: 1281 bail_commit:
1279 ocfs2_commit_trans(osb, handle); 1282 ocfs2_commit_trans(osb, handle);
1280 bail_unlock_alloc: <<
1281 up_write(&OCFS2_I(inode)->ip_alloc_se <<
1282 bail_unlock: 1283 bail_unlock:
1283 if (status && inode_locked) { 1284 if (status && inode_locked) {
1284 ocfs2_inode_unlock_tracker(in 1285 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1285 inode_locked = 0; 1286 inode_locked = 0;
1286 } 1287 }
1287 bail_unlock_rw: 1288 bail_unlock_rw:
1288 if (size_change) 1289 if (size_change)
1289 ocfs2_rw_unlock(inode, 1); 1290 ocfs2_rw_unlock(inode, 1);
1290 bail: 1291 bail:
1291 1292
1292 /* Release quota pointers in case we 1293 /* Release quota pointers in case we acquired them */
1293 for (qtype = 0; qtype < OCFS2_MAXQUOT 1294 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1294 dqput(transfer_to[qtype]); 1295 dqput(transfer_to[qtype]);
1295 1296
1296 if (!status && attr->ia_valid & ATTR_ 1297 if (!status && attr->ia_valid & ATTR_MODE) {
1297 status = ocfs2_acl_chmod(inod 1298 status = ocfs2_acl_chmod(inode, bh);
1298 if (status < 0) 1299 if (status < 0)
1299 mlog_errno(status); 1300 mlog_errno(status);
1300 } 1301 }
1301 if (inode_locked) 1302 if (inode_locked)
1302 ocfs2_inode_unlock_tracker(in 1303 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1303 1304
1304 brelse(bh); 1305 brelse(bh);
1305 return status; 1306 return status;
1306 } 1307 }
1307 1308
1308 int ocfs2_getattr(struct mnt_idmap *idmap, co !! 1309 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1309 struct kstat *stat, u32 req !! 1310 u32 request_mask, unsigned int flags)
1310 { 1311 {
1311 struct inode *inode = d_inode(path->d 1312 struct inode *inode = d_inode(path->dentry);
1312 struct super_block *sb = path->dentry 1313 struct super_block *sb = path->dentry->d_sb;
1313 struct ocfs2_super *osb = sb->s_fs_in 1314 struct ocfs2_super *osb = sb->s_fs_info;
1314 int err; 1315 int err;
1315 1316
1316 err = ocfs2_inode_revalidate(path->de 1317 err = ocfs2_inode_revalidate(path->dentry);
1317 if (err) { 1318 if (err) {
1318 if (err != -ENOENT) 1319 if (err != -ENOENT)
1319 mlog_errno(err); 1320 mlog_errno(err);
1320 goto bail; 1321 goto bail;
1321 } 1322 }
1322 1323
1323 generic_fillattr(&nop_mnt_idmap, requ !! 1324 generic_fillattr(inode, stat);
1324 /* 1325 /*
1325 * If there is inline data in the ino 1326 * If there is inline data in the inode, the inode will normally not
1326 * have data blocks allocated (it may 1327 * have data blocks allocated (it may have an external xattr block).
1327 * Report at least one sector for suc 1328 * Report at least one sector for such files, so tools like tar, rsync,
1328 * others don't incorrectly think the 1329 * others don't incorrectly think the file is completely sparse.
1329 */ 1330 */
1330 if (unlikely(OCFS2_I(inode)->ip_dyn_f 1331 if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1331 stat->blocks += (stat->size + 1332 stat->blocks += (stat->size + 511)>>9;
1332 1333
1333 /* We set the blksize from the cluste 1334 /* We set the blksize from the cluster size for performance */
1334 stat->blksize = osb->s_clustersize; 1335 stat->blksize = osb->s_clustersize;
1335 1336
1336 bail: 1337 bail:
1337 return err; 1338 return err;
1338 } 1339 }
1339 1340
1340 int ocfs2_permission(struct mnt_idmap *idmap, !! 1341 int ocfs2_permission(struct inode *inode, int mask)
1341 int mask) <<
1342 { 1342 {
1343 int ret, had_lock; 1343 int ret, had_lock;
1344 struct ocfs2_lock_holder oh; 1344 struct ocfs2_lock_holder oh;
1345 1345
1346 if (mask & MAY_NOT_BLOCK) 1346 if (mask & MAY_NOT_BLOCK)
1347 return -ECHILD; 1347 return -ECHILD;
1348 1348
1349 had_lock = ocfs2_inode_lock_tracker(i 1349 had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1350 if (had_lock < 0) { 1350 if (had_lock < 0) {
1351 ret = had_lock; 1351 ret = had_lock;
1352 goto out; 1352 goto out;
1353 } else if (had_lock) { 1353 } else if (had_lock) {
1354 /* See comments in ocfs2_seta 1354 /* See comments in ocfs2_setattr() for details.
1355 * The call chain of this cas 1355 * The call chain of this case could be:
1356 * do_sys_open() 1356 * do_sys_open()
1357 * may_open() 1357 * may_open()
1358 * inode_permission() 1358 * inode_permission()
1359 * ocfs2_permission() 1359 * ocfs2_permission()
1360 * ocfs2_iop_get_acl() 1360 * ocfs2_iop_get_acl()
1361 */ 1361 */
1362 mlog(ML_ERROR, "Another case 1362 mlog(ML_ERROR, "Another case of recursive locking:\n");
1363 dump_stack(); 1363 dump_stack();
1364 } 1364 }
1365 1365
1366 ret = generic_permission(&nop_mnt_idm !! 1366 ret = generic_permission(inode, mask);
1367 1367
1368 ocfs2_inode_unlock_tracker(inode, 0, 1368 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1369 out: 1369 out:
1370 return ret; 1370 return ret;
1371 } 1371 }
1372 1372
1373 static int __ocfs2_write_remove_suid(struct i 1373 static int __ocfs2_write_remove_suid(struct inode *inode,
1374 struct b 1374 struct buffer_head *bh)
1375 { 1375 {
1376 int ret; 1376 int ret;
1377 handle_t *handle; 1377 handle_t *handle;
1378 struct ocfs2_super *osb = OCFS2_SB(in 1378 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1379 struct ocfs2_dinode *di; 1379 struct ocfs2_dinode *di;
1380 1380
1381 trace_ocfs2_write_remove_suid( 1381 trace_ocfs2_write_remove_suid(
1382 (unsigned long long)O 1382 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1383 inode->i_mode); 1383 inode->i_mode);
1384 1384
1385 handle = ocfs2_start_trans(osb, OCFS2 1385 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1386 if (IS_ERR(handle)) { 1386 if (IS_ERR(handle)) {
1387 ret = PTR_ERR(handle); 1387 ret = PTR_ERR(handle);
1388 mlog_errno(ret); 1388 mlog_errno(ret);
1389 goto out; 1389 goto out;
1390 } 1390 }
1391 1391
1392 ret = ocfs2_journal_access_di(handle, 1392 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1393 OCFS2_J 1393 OCFS2_JOURNAL_ACCESS_WRITE);
1394 if (ret < 0) { 1394 if (ret < 0) {
1395 mlog_errno(ret); 1395 mlog_errno(ret);
1396 goto out_trans; 1396 goto out_trans;
1397 } 1397 }
1398 1398
1399 inode->i_mode &= ~S_ISUID; 1399 inode->i_mode &= ~S_ISUID;
1400 if ((inode->i_mode & S_ISGID) && (ino 1400 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1401 inode->i_mode &= ~S_ISGID; 1401 inode->i_mode &= ~S_ISGID;
1402 1402
1403 di = (struct ocfs2_dinode *) bh->b_da 1403 di = (struct ocfs2_dinode *) bh->b_data;
1404 di->i_mode = cpu_to_le16(inode->i_mod 1404 di->i_mode = cpu_to_le16(inode->i_mode);
1405 ocfs2_update_inode_fsync_trans(handle 1405 ocfs2_update_inode_fsync_trans(handle, inode, 0);
1406 1406
1407 ocfs2_journal_dirty(handle, bh); 1407 ocfs2_journal_dirty(handle, bh);
1408 1408
1409 out_trans: 1409 out_trans:
1410 ocfs2_commit_trans(osb, handle); 1410 ocfs2_commit_trans(osb, handle);
1411 out: 1411 out:
1412 return ret; 1412 return ret;
1413 } 1413 }
1414 1414
1415 static int ocfs2_write_remove_suid(struct ino 1415 static int ocfs2_write_remove_suid(struct inode *inode)
1416 { 1416 {
1417 int ret; 1417 int ret;
1418 struct buffer_head *bh = NULL; 1418 struct buffer_head *bh = NULL;
1419 1419
1420 ret = ocfs2_read_inode_block(inode, & 1420 ret = ocfs2_read_inode_block(inode, &bh);
1421 if (ret < 0) { 1421 if (ret < 0) {
1422 mlog_errno(ret); 1422 mlog_errno(ret);
1423 goto out; 1423 goto out;
1424 } 1424 }
1425 1425
1426 ret = __ocfs2_write_remove_suid(inod 1426 ret = __ocfs2_write_remove_suid(inode, bh);
1427 out: 1427 out:
1428 brelse(bh); 1428 brelse(bh);
1429 return ret; 1429 return ret;
1430 } 1430 }
1431 1431
1432 /* 1432 /*
1433 * Allocate enough extents to cover the regio 1433 * Allocate enough extents to cover the region starting at byte offset
1434 * start for len bytes. Existing extents are 1434 * start for len bytes. Existing extents are skipped, any extents
1435 * added are marked as "unwritten". 1435 * added are marked as "unwritten".
1436 */ 1436 */
1437 static int ocfs2_allocate_unwritten_extents(s 1437 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1438 u 1438 u64 start, u64 len)
1439 { 1439 {
1440 int ret; 1440 int ret;
1441 u32 cpos, phys_cpos, clusters, alloc_ 1441 u32 cpos, phys_cpos, clusters, alloc_size;
1442 u64 end = start + len; 1442 u64 end = start + len;
1443 struct buffer_head *di_bh = NULL; 1443 struct buffer_head *di_bh = NULL;
1444 1444
1445 if (OCFS2_I(inode)->ip_dyn_features & 1445 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1446 ret = ocfs2_read_inode_block( 1446 ret = ocfs2_read_inode_block(inode, &di_bh);
1447 if (ret) { 1447 if (ret) {
1448 mlog_errno(ret); 1448 mlog_errno(ret);
1449 goto out; 1449 goto out;
1450 } 1450 }
1451 1451
1452 /* 1452 /*
1453 * Nothing to do if the reque 1453 * Nothing to do if the requested reservation range
1454 * fits within the inode. 1454 * fits within the inode.
1455 */ 1455 */
1456 if (ocfs2_size_fits_inline_da 1456 if (ocfs2_size_fits_inline_data(di_bh, end))
1457 goto out; 1457 goto out;
1458 1458
1459 ret = ocfs2_convert_inline_da 1459 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1460 if (ret) { 1460 if (ret) {
1461 mlog_errno(ret); 1461 mlog_errno(ret);
1462 goto out; 1462 goto out;
1463 } 1463 }
1464 } 1464 }
1465 1465
1466 /* 1466 /*
1467 * We consider both start and len to 1467 * We consider both start and len to be inclusive.
1468 */ 1468 */
1469 cpos = start >> OCFS2_SB(inode->i_sb) 1469 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1470 clusters = ocfs2_clusters_for_bytes(i 1470 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1471 clusters -= cpos; 1471 clusters -= cpos;
1472 1472
1473 while (clusters) { 1473 while (clusters) {
1474 ret = ocfs2_get_clusters(inod 1474 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1475 &all 1475 &alloc_size, NULL);
1476 if (ret) { 1476 if (ret) {
1477 mlog_errno(ret); 1477 mlog_errno(ret);
1478 goto out; 1478 goto out;
1479 } 1479 }
1480 1480
1481 /* 1481 /*
1482 * Hole or existing extent le 1482 * Hole or existing extent len can be arbitrary, so
1483 * cap it to our own allocati 1483 * cap it to our own allocation request.
1484 */ 1484 */
1485 if (alloc_size > clusters) 1485 if (alloc_size > clusters)
1486 alloc_size = clusters 1486 alloc_size = clusters;
1487 1487
1488 if (phys_cpos) { 1488 if (phys_cpos) {
1489 /* 1489 /*
1490 * We already have an 1490 * We already have an allocation at this
1491 * region so we can s 1491 * region so we can safely skip it.
1492 */ 1492 */
1493 goto next; 1493 goto next;
1494 } 1494 }
1495 1495
1496 ret = ocfs2_extend_allocation 1496 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1497 if (ret) { 1497 if (ret) {
1498 if (ret != -ENOSPC) 1498 if (ret != -ENOSPC)
1499 mlog_errno(re 1499 mlog_errno(ret);
1500 goto out; 1500 goto out;
1501 } 1501 }
1502 1502
1503 next: 1503 next:
1504 cpos += alloc_size; 1504 cpos += alloc_size;
1505 clusters -= alloc_size; 1505 clusters -= alloc_size;
1506 } 1506 }
1507 1507
1508 ret = 0; 1508 ret = 0;
1509 out: 1509 out:
1510 1510
1511 brelse(di_bh); 1511 brelse(di_bh);
1512 return ret; 1512 return ret;
1513 } 1513 }
1514 1514
1515 /* 1515 /*
1516 * Truncate a byte range, avoiding pages with 1516 * Truncate a byte range, avoiding pages within partial clusters. This
1517 * preserves those pages for the zeroing code 1517 * preserves those pages for the zeroing code to write to.
1518 */ 1518 */
1519 static void ocfs2_truncate_cluster_pages(stru 1519 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1520 u64 1520 u64 byte_len)
1521 { 1521 {
1522 struct ocfs2_super *osb = OCFS2_SB(in 1522 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1523 loff_t start, end; 1523 loff_t start, end;
1524 struct address_space *mapping = inode 1524 struct address_space *mapping = inode->i_mapping;
1525 1525
1526 start = (loff_t)ocfs2_align_bytes_to_ 1526 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1527 end = byte_start + byte_len; 1527 end = byte_start + byte_len;
1528 end = end & ~(osb->s_clustersize - 1) 1528 end = end & ~(osb->s_clustersize - 1);
1529 1529
1530 if (start < end) { 1530 if (start < end) {
1531 unmap_mapping_range(mapping, 1531 unmap_mapping_range(mapping, start, end - start, 0);
1532 truncate_inode_pages_range(ma 1532 truncate_inode_pages_range(mapping, start, end - 1);
1533 } 1533 }
1534 } 1534 }
1535 1535
1536 /* <<
1537 * zero out partial blocks of one cluster. <<
1538 * <<
1539 * start: file offset where zero starts, will <<
1540 * len: it will be trimmed to the end of curr <<
1541 * is bigger than it. <<
1542 */ <<
1543 static int ocfs2_zeroout_partial_cluster(stru <<
1544 u64 s <<
1545 { <<
1546 int ret; <<
1547 u64 start_block, end_block, nr_blocks <<
1548 u64 p_block, offset; <<
1549 u32 cluster, p_cluster, nr_clusters; <<
1550 struct super_block *sb = inode->i_sb; <<
1551 u64 end = ocfs2_align_bytes_to_cluste <<
1552 <<
1553 if (start + len < end) <<
1554 end = start + len; <<
1555 <<
1556 start_block = ocfs2_blocks_for_bytes( <<
1557 end_block = ocfs2_blocks_for_bytes(sb <<
1558 nr_blocks = end_block - start_block; <<
1559 if (!nr_blocks) <<
1560 return 0; <<
1561 <<
1562 cluster = ocfs2_bytes_to_clusters(sb, <<
1563 ret = ocfs2_get_clusters(inode, clust <<
1564 &nr_clusters, <<
1565 if (ret) <<
1566 return ret; <<
1567 if (!p_cluster) <<
1568 return 0; <<
1569 <<
1570 offset = start_block - ocfs2_clusters <<
1571 p_block = ocfs2_clusters_to_blocks(sb <<
1572 return sb_issue_zeroout(sb, p_block, <<
1573 } <<
1574 <<
1575 static int ocfs2_zero_partial_clusters(struct 1536 static int ocfs2_zero_partial_clusters(struct inode *inode,
1576 u64 st 1537 u64 start, u64 len)
1577 { 1538 {
1578 int ret = 0; 1539 int ret = 0;
1579 u64 tmpend = 0; 1540 u64 tmpend = 0;
1580 u64 end = start + len; 1541 u64 end = start + len;
1581 struct ocfs2_super *osb = OCFS2_SB(in 1542 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1582 unsigned int csize = osb->s_clustersi 1543 unsigned int csize = osb->s_clustersize;
1583 handle_t *handle; 1544 handle_t *handle;
1584 loff_t isize = i_size_read(inode); <<
1585 1545
1586 /* 1546 /*
1587 * The "start" and "end" values are N 1547 * The "start" and "end" values are NOT necessarily part of
1588 * the range whose allocation is bein 1548 * the range whose allocation is being deleted. Rather, this
1589 * is what the user passed in with th 1549 * is what the user passed in with the request. We must zero
1590 * partial clusters here. There's no 1550 * partial clusters here. There's no need to worry about
1591 * physical allocation - the zeroing 1551 * physical allocation - the zeroing code knows to skip holes.
1592 */ 1552 */
1593 trace_ocfs2_zero_partial_clusters( 1553 trace_ocfs2_zero_partial_clusters(
1594 (unsigned long long)OCFS2_I(i 1554 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1595 (unsigned long long)start, (u 1555 (unsigned long long)start, (unsigned long long)end);
1596 1556
1597 /* 1557 /*
1598 * If both edges are on a cluster bou 1558 * If both edges are on a cluster boundary then there's no
1599 * zeroing required as the region is 1559 * zeroing required as the region is part of the allocation to
1600 * be truncated. 1560 * be truncated.
1601 */ 1561 */
1602 if ((start & (csize - 1)) == 0 && (en 1562 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1603 goto out; 1563 goto out;
1604 1564
1605 /* No page cache for EOF blocks, issu <<
1606 if (end > isize) { <<
1607 /* <<
1608 * zeroout eof blocks in last <<
1609 * "isize" even "start" > "is <<
1610 * complicated to zeroout jus <<
1611 * may be not aligned with bl <<
1612 * would be required to do th <<
1613 * write is not supported. <<
1614 */ <<
1615 ret = ocfs2_zeroout_partial_c <<
1616 end - <<
1617 if (ret) { <<
1618 mlog_errno(ret); <<
1619 goto out; <<
1620 } <<
1621 if (start >= isize) <<
1622 goto out; <<
1623 end = isize; <<
1624 } <<
1625 handle = ocfs2_start_trans(osb, OCFS2 1565 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1626 if (IS_ERR(handle)) { 1566 if (IS_ERR(handle)) {
1627 ret = PTR_ERR(handle); 1567 ret = PTR_ERR(handle);
1628 mlog_errno(ret); 1568 mlog_errno(ret);
1629 goto out; 1569 goto out;
1630 } 1570 }
1631 1571
1632 /* 1572 /*
1633 * If start is on a cluster boundary 1573 * If start is on a cluster boundary and end is somewhere in another
1634 * cluster, we have not COWed the clu 1574 * cluster, we have not COWed the cluster starting at start, unless
1635 * end is also within the same cluste 1575 * end is also within the same cluster. So, in this case, we skip this
1636 * first call to ocfs2_zero_range_for 1576 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1637 * to the next one. 1577 * to the next one.
1638 */ 1578 */
1639 if ((start & (csize - 1)) != 0) { 1579 if ((start & (csize - 1)) != 0) {
1640 /* 1580 /*
1641 * We want to get the byte of 1581 * We want to get the byte offset of the end of the 1st
1642 * cluster. 1582 * cluster.
1643 */ 1583 */
1644 tmpend = (u64)osb->s_clusters 1584 tmpend = (u64)osb->s_clustersize +
1645 (start & ~(osb->s_clu 1585 (start & ~(osb->s_clustersize - 1));
1646 if (tmpend > end) 1586 if (tmpend > end)
1647 tmpend = end; 1587 tmpend = end;
1648 1588
1649 trace_ocfs2_zero_partial_clus 1589 trace_ocfs2_zero_partial_clusters_range1(
1650 (unsigned long long)s 1590 (unsigned long long)start,
1651 (unsigned long long)t 1591 (unsigned long long)tmpend);
1652 1592
1653 ret = ocfs2_zero_range_for_tr 1593 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1654 1594 tmpend);
1655 if (ret) 1595 if (ret)
1656 mlog_errno(ret); 1596 mlog_errno(ret);
1657 } 1597 }
1658 1598
1659 if (tmpend < end) { 1599 if (tmpend < end) {
1660 /* 1600 /*
1661 * This may make start and en 1601 * This may make start and end equal, but the zeroing
1662 * code will skip any work in 1602 * code will skip any work in that case so there's no
1663 * need to catch it up here. 1603 * need to catch it up here.
1664 */ 1604 */
1665 start = end & ~(osb->s_cluste 1605 start = end & ~(osb->s_clustersize - 1);
1666 1606
1667 trace_ocfs2_zero_partial_clus 1607 trace_ocfs2_zero_partial_clusters_range2(
1668 (unsigned long long)s 1608 (unsigned long long)start, (unsigned long long)end);
1669 1609
1670 ret = ocfs2_zero_range_for_tr 1610 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1671 if (ret) 1611 if (ret)
1672 mlog_errno(ret); 1612 mlog_errno(ret);
1673 } 1613 }
1674 ocfs2_update_inode_fsync_trans(handle 1614 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1675 1615
1676 ocfs2_commit_trans(osb, handle); 1616 ocfs2_commit_trans(osb, handle);
1677 out: 1617 out:
1678 return ret; 1618 return ret;
1679 } 1619 }
1680 1620
1681 static int ocfs2_find_rec(struct ocfs2_extent 1621 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1682 { 1622 {
1683 int i; 1623 int i;
1684 struct ocfs2_extent_rec *rec = NULL; 1624 struct ocfs2_extent_rec *rec = NULL;
1685 1625
1686 for (i = le16_to_cpu(el->l_next_free_ 1626 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1687 1627
1688 rec = &el->l_recs[i]; 1628 rec = &el->l_recs[i];
1689 1629
1690 if (le32_to_cpu(rec->e_cpos) 1630 if (le32_to_cpu(rec->e_cpos) < pos)
1691 break; 1631 break;
1692 } 1632 }
1693 1633
1694 return i; 1634 return i;
1695 } 1635 }
1696 1636
1697 /* 1637 /*
1698 * Helper to calculate the punching pos and l 1638 * Helper to calculate the punching pos and length in one run, we handle the
1699 * following three cases in order: 1639 * following three cases in order:
1700 * 1640 *
1701 * - remove the entire record 1641 * - remove the entire record
1702 * - remove a partial record 1642 * - remove a partial record
1703 * - no record needs to be removed (hole-punc 1643 * - no record needs to be removed (hole-punching completed)
1704 */ 1644 */
1705 static void ocfs2_calc_trunc_pos(struct inode 1645 static void ocfs2_calc_trunc_pos(struct inode *inode,
1706 struct ocfs2 1646 struct ocfs2_extent_list *el,
1707 struct ocfs2 1647 struct ocfs2_extent_rec *rec,
1708 u32 trunc_st 1648 u32 trunc_start, u32 *trunc_cpos,
1709 u32 *trunc_l 1649 u32 *trunc_len, u32 *trunc_end,
1710 u64 *blkno, 1650 u64 *blkno, int *done)
1711 { 1651 {
1712 int ret = 0; 1652 int ret = 0;
1713 u32 coff, range; 1653 u32 coff, range;
1714 1654
1715 range = le32_to_cpu(rec->e_cpos) + oc 1655 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1716 1656
1717 if (le32_to_cpu(rec->e_cpos) >= trunc 1657 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1718 /* 1658 /*
1719 * remove an entire extent re 1659 * remove an entire extent record.
1720 */ 1660 */
1721 *trunc_cpos = le32_to_cpu(rec 1661 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1722 /* 1662 /*
1723 * Skip holes if any. 1663 * Skip holes if any.
1724 */ 1664 */
1725 if (range < *trunc_end) 1665 if (range < *trunc_end)
1726 *trunc_end = range; 1666 *trunc_end = range;
1727 *trunc_len = *trunc_end - le3 1667 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1728 *blkno = le64_to_cpu(rec->e_b 1668 *blkno = le64_to_cpu(rec->e_blkno);
1729 *trunc_end = le32_to_cpu(rec- 1669 *trunc_end = le32_to_cpu(rec->e_cpos);
1730 } else if (range > trunc_start) { 1670 } else if (range > trunc_start) {
1731 /* 1671 /*
1732 * remove a partial extent re 1672 * remove a partial extent record, which means we're
1733 * removing the last extent r 1673 * removing the last extent record.
1734 */ 1674 */
1735 *trunc_cpos = trunc_start; 1675 *trunc_cpos = trunc_start;
1736 /* 1676 /*
1737 * skip hole if any. 1677 * skip hole if any.
1738 */ 1678 */
1739 if (range < *trunc_end) 1679 if (range < *trunc_end)
1740 *trunc_end = range; 1680 *trunc_end = range;
1741 *trunc_len = *trunc_end - tru 1681 *trunc_len = *trunc_end - trunc_start;
1742 coff = trunc_start - le32_to_ 1682 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1743 *blkno = le64_to_cpu(rec->e_b 1683 *blkno = le64_to_cpu(rec->e_blkno) +
1744 ocfs2_cluster 1684 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1745 *trunc_end = trunc_start; 1685 *trunc_end = trunc_start;
1746 } else { 1686 } else {
1747 /* 1687 /*
1748 * It may have two following 1688 * It may have two following possibilities:
1749 * 1689 *
1750 * - last record has been rem 1690 * - last record has been removed
1751 * - trunc_start was within a 1691 * - trunc_start was within a hole
1752 * 1692 *
1753 * both two cases mean the co 1693 * both two cases mean the completion of hole punching.
1754 */ 1694 */
1755 ret = 1; 1695 ret = 1;
1756 } 1696 }
1757 1697
1758 *done = ret; 1698 *done = ret;
1759 } 1699 }
1760 1700
1761 int ocfs2_remove_inode_range(struct inode *in 1701 int ocfs2_remove_inode_range(struct inode *inode,
1762 struct buffer_he 1702 struct buffer_head *di_bh, u64 byte_start,
1763 u64 byte_len) 1703 u64 byte_len)
1764 { 1704 {
1765 int ret = 0, flags = 0, done = 0, i; 1705 int ret = 0, flags = 0, done = 0, i;
1766 u32 trunc_start, trunc_len, trunc_end 1706 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1767 u32 cluster_in_el; 1707 u32 cluster_in_el;
1768 struct ocfs2_super *osb = OCFS2_SB(in 1708 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1769 struct ocfs2_cached_dealloc_ctxt deal 1709 struct ocfs2_cached_dealloc_ctxt dealloc;
1770 struct address_space *mapping = inode 1710 struct address_space *mapping = inode->i_mapping;
1771 struct ocfs2_extent_tree et; 1711 struct ocfs2_extent_tree et;
1772 struct ocfs2_path *path = NULL; 1712 struct ocfs2_path *path = NULL;
1773 struct ocfs2_extent_list *el = NULL; 1713 struct ocfs2_extent_list *el = NULL;
1774 struct ocfs2_extent_rec *rec = NULL; 1714 struct ocfs2_extent_rec *rec = NULL;
1775 struct ocfs2_dinode *di = (struct ocf 1715 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1776 u64 blkno, refcount_loc = le64_to_cpu 1716 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1777 1717
1778 ocfs2_init_dinode_extent_tree(&et, IN 1718 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1779 ocfs2_init_dealloc_ctxt(&dealloc); 1719 ocfs2_init_dealloc_ctxt(&dealloc);
1780 1720
1781 trace_ocfs2_remove_inode_range( 1721 trace_ocfs2_remove_inode_range(
1782 (unsigned long long)O 1722 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1783 (unsigned long long)b 1723 (unsigned long long)byte_start,
1784 (unsigned long long)b 1724 (unsigned long long)byte_len);
1785 1725
1786 if (byte_len == 0) 1726 if (byte_len == 0)
1787 return 0; 1727 return 0;
1788 1728
1789 if (OCFS2_I(inode)->ip_dyn_features & 1729 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1790 int id_count = ocfs2_max_inli <<
1791 <<
1792 if (byte_start > id_count || <<
1793 ret = -EINVAL; <<
1794 mlog_errno(ret); <<
1795 goto out; <<
1796 } <<
1797 <<
1798 ret = ocfs2_truncate_inline(i 1730 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1799 b 1731 byte_start + byte_len, 0);
1800 if (ret) { 1732 if (ret) {
1801 mlog_errno(ret); 1733 mlog_errno(ret);
1802 goto out; 1734 goto out;
1803 } 1735 }
1804 /* 1736 /*
1805 * There's no need to get fan 1737 * There's no need to get fancy with the page cache
1806 * truncate of an inline-data 1738 * truncate of an inline-data inode. We're talking
1807 * about less than a page her 1739 * about less than a page here, which will be cached
1808 * in the dinode buffer anywa 1740 * in the dinode buffer anyway.
1809 */ 1741 */
1810 unmap_mapping_range(mapping, 1742 unmap_mapping_range(mapping, 0, 0, 0);
1811 truncate_inode_pages(mapping, 1743 truncate_inode_pages(mapping, 0);
1812 goto out; 1744 goto out;
1813 } 1745 }
1814 1746
1815 /* 1747 /*
1816 * For reflinks, we may need to CoW 2 1748 * For reflinks, we may need to CoW 2 clusters which might be
1817 * partially zero'd later, if hole's 1749 * partially zero'd later, if hole's start and end offset were
1818 * within one cluster(means is not ex 1750 * within one cluster(means is not exactly aligned to clustersize).
1819 */ 1751 */
1820 1752
1821 if (ocfs2_is_refcount_inode(inode)) { 1753 if (ocfs2_is_refcount_inode(inode)) {
1822 ret = ocfs2_cow_file_pos(inod 1754 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1823 if (ret) { 1755 if (ret) {
1824 mlog_errno(ret); 1756 mlog_errno(ret);
1825 goto out; 1757 goto out;
1826 } 1758 }
1827 1759
1828 ret = ocfs2_cow_file_pos(inod 1760 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1829 if (ret) { 1761 if (ret) {
1830 mlog_errno(ret); 1762 mlog_errno(ret);
1831 goto out; 1763 goto out;
1832 } 1764 }
1833 } 1765 }
1834 1766
1835 trunc_start = ocfs2_clusters_for_byte 1767 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1836 trunc_end = (byte_start + byte_len) > 1768 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1837 cluster_in_el = trunc_end; 1769 cluster_in_el = trunc_end;
1838 1770
1839 ret = ocfs2_zero_partial_clusters(ino 1771 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1840 if (ret) { 1772 if (ret) {
1841 mlog_errno(ret); 1773 mlog_errno(ret);
1842 goto out; 1774 goto out;
1843 } 1775 }
1844 1776
1845 path = ocfs2_new_path_from_et(&et); 1777 path = ocfs2_new_path_from_et(&et);
1846 if (!path) { 1778 if (!path) {
1847 ret = -ENOMEM; 1779 ret = -ENOMEM;
1848 mlog_errno(ret); 1780 mlog_errno(ret);
1849 goto out; 1781 goto out;
1850 } 1782 }
1851 1783
1852 while (trunc_end > trunc_start) { 1784 while (trunc_end > trunc_start) {
1853 1785
1854 ret = ocfs2_find_path(INODE_C 1786 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1855 cluster 1787 cluster_in_el);
1856 if (ret) { 1788 if (ret) {
1857 mlog_errno(ret); 1789 mlog_errno(ret);
1858 goto out; 1790 goto out;
1859 } 1791 }
1860 1792
1861 el = path_leaf_el(path); 1793 el = path_leaf_el(path);
1862 1794
1863 i = ocfs2_find_rec(el, trunc_ 1795 i = ocfs2_find_rec(el, trunc_end);
1864 /* 1796 /*
1865 * Need to go to previous ext 1797 * Need to go to previous extent block.
1866 */ 1798 */
1867 if (i < 0) { 1799 if (i < 0) {
1868 if (path->p_tree_dept 1800 if (path->p_tree_depth == 0)
1869 break; 1801 break;
1870 1802
1871 ret = ocfs2_find_cpos 1803 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1872 1804 path,
1873 1805 &cluster_in_el);
1874 if (ret) { 1806 if (ret) {
1875 mlog_errno(re 1807 mlog_errno(ret);
1876 goto out; 1808 goto out;
1877 } 1809 }
1878 1810
1879 /* 1811 /*
1880 * We've reached the 1812 * We've reached the leftmost extent block,
1881 * it's safe to leave 1813 * it's safe to leave.
1882 */ 1814 */
1883 if (cluster_in_el == 1815 if (cluster_in_el == 0)
1884 break; 1816 break;
1885 1817
1886 /* 1818 /*
1887 * The 'pos' searched 1819 * The 'pos' searched for previous extent block is
1888 * always one cluster 1820 * always one cluster less than actual trunc_end.
1889 */ 1821 */
1890 trunc_end = cluster_i 1822 trunc_end = cluster_in_el + 1;
1891 1823
1892 ocfs2_reinit_path(pat 1824 ocfs2_reinit_path(path, 1);
1893 1825
1894 continue; 1826 continue;
1895 1827
1896 } else 1828 } else
1897 rec = &el->l_recs[i]; 1829 rec = &el->l_recs[i];
1898 1830
1899 ocfs2_calc_trunc_pos(inode, e 1831 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1900 &trunc_l 1832 &trunc_len, &trunc_end, &blkno, &done);
1901 if (done) 1833 if (done)
1902 break; 1834 break;
1903 1835
1904 flags = rec->e_flags; 1836 flags = rec->e_flags;
1905 phys_cpos = ocfs2_blocks_to_c 1837 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1906 1838
1907 ret = ocfs2_remove_btree_rang 1839 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1908 1840 phys_cpos, trunc_len, flags,
1909 1841 &dealloc, refcount_loc, false);
1910 if (ret < 0) { 1842 if (ret < 0) {
1911 mlog_errno(ret); 1843 mlog_errno(ret);
1912 goto out; 1844 goto out;
1913 } 1845 }
1914 1846
1915 cluster_in_el = trunc_end; 1847 cluster_in_el = trunc_end;
1916 1848
1917 ocfs2_reinit_path(path, 1); 1849 ocfs2_reinit_path(path, 1);
1918 } 1850 }
1919 1851
1920 ocfs2_truncate_cluster_pages(inode, b 1852 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1921 1853
1922 out: 1854 out:
1923 ocfs2_free_path(path); 1855 ocfs2_free_path(path);
1924 ocfs2_schedule_truncate_log_flush(osb 1856 ocfs2_schedule_truncate_log_flush(osb, 1);
1925 ocfs2_run_deallocs(osb, &dealloc); 1857 ocfs2_run_deallocs(osb, &dealloc);
1926 1858
1927 return ret; 1859 return ret;
1928 } 1860 }
1929 1861
1930 /* 1862 /*
1931 * Parts of this function taken from xfs_chan 1863 * Parts of this function taken from xfs_change_file_space()
1932 */ 1864 */
1933 static int __ocfs2_change_file_space(struct f 1865 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1934 loff_t f 1866 loff_t f_pos, unsigned int cmd,
1935 struct o 1867 struct ocfs2_space_resv *sr,
1936 int chan 1868 int change_size)
1937 { 1869 {
1938 int ret; 1870 int ret;
1939 s64 llen; 1871 s64 llen;
1940 loff_t size, orig_isize; !! 1872 loff_t size;
1941 struct ocfs2_super *osb = OCFS2_SB(in 1873 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1942 struct buffer_head *di_bh = NULL; 1874 struct buffer_head *di_bh = NULL;
1943 handle_t *handle; 1875 handle_t *handle;
1944 unsigned long long max_off = inode->i 1876 unsigned long long max_off = inode->i_sb->s_maxbytes;
1945 1877
1946 if (ocfs2_is_hard_readonly(osb) || oc 1878 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1947 return -EROFS; 1879 return -EROFS;
1948 1880
1949 inode_lock(inode); 1881 inode_lock(inode);
1950 1882
1951 /* Wait all existing dio workers, new <<
1952 inode_dio_wait(inode); <<
1953 /* 1883 /*
1954 * This prevents concurrent writes on 1884 * This prevents concurrent writes on other nodes
1955 */ 1885 */
1956 ret = ocfs2_rw_lock(inode, 1); 1886 ret = ocfs2_rw_lock(inode, 1);
1957 if (ret) { 1887 if (ret) {
1958 mlog_errno(ret); 1888 mlog_errno(ret);
1959 goto out; 1889 goto out;
1960 } 1890 }
1961 1891
1962 ret = ocfs2_inode_lock(inode, &di_bh, 1892 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1963 if (ret) { 1893 if (ret) {
1964 mlog_errno(ret); 1894 mlog_errno(ret);
1965 goto out_rw_unlock; 1895 goto out_rw_unlock;
1966 } 1896 }
1967 1897
1968 if (inode->i_flags & (S_IMMUTABLE|S_A 1898 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1969 ret = -EPERM; 1899 ret = -EPERM;
1970 goto out_inode_unlock; 1900 goto out_inode_unlock;
1971 } 1901 }
1972 1902
1973 switch (sr->l_whence) { 1903 switch (sr->l_whence) {
1974 case 0: /*SEEK_SET*/ 1904 case 0: /*SEEK_SET*/
1975 break; 1905 break;
1976 case 1: /*SEEK_CUR*/ 1906 case 1: /*SEEK_CUR*/
1977 sr->l_start += f_pos; 1907 sr->l_start += f_pos;
1978 break; 1908 break;
1979 case 2: /*SEEK_END*/ 1909 case 2: /*SEEK_END*/
1980 sr->l_start += i_size_read(in 1910 sr->l_start += i_size_read(inode);
1981 break; 1911 break;
1982 default: 1912 default:
1983 ret = -EINVAL; 1913 ret = -EINVAL;
1984 goto out_inode_unlock; 1914 goto out_inode_unlock;
1985 } 1915 }
1986 sr->l_whence = 0; 1916 sr->l_whence = 0;
1987 1917
1988 llen = sr->l_len > 0 ? sr->l_len - 1 1918 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1989 1919
1990 if (sr->l_start < 0 1920 if (sr->l_start < 0
1991 || sr->l_start > max_off 1921 || sr->l_start > max_off
1992 || (sr->l_start + llen) < 0 1922 || (sr->l_start + llen) < 0
1993 || (sr->l_start + llen) > max_off 1923 || (sr->l_start + llen) > max_off) {
1994 ret = -EINVAL; 1924 ret = -EINVAL;
1995 goto out_inode_unlock; 1925 goto out_inode_unlock;
1996 } 1926 }
1997 size = sr->l_start + sr->l_len; 1927 size = sr->l_start + sr->l_len;
1998 1928
1999 if (cmd == OCFS2_IOC_RESVSP || cmd == 1929 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
2000 cmd == OCFS2_IOC_UNRESVSP || cmd 1930 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
2001 if (sr->l_len <= 0) { 1931 if (sr->l_len <= 0) {
2002 ret = -EINVAL; 1932 ret = -EINVAL;
2003 goto out_inode_unlock 1933 goto out_inode_unlock;
2004 } 1934 }
2005 } 1935 }
2006 1936
2007 if (file && setattr_should_drop_suidg !! 1937 if (file && should_remove_suid(file->f_path.dentry)) {
2008 ret = __ocfs2_write_remove_su 1938 ret = __ocfs2_write_remove_suid(inode, di_bh);
2009 if (ret) { 1939 if (ret) {
2010 mlog_errno(ret); 1940 mlog_errno(ret);
2011 goto out_inode_unlock 1941 goto out_inode_unlock;
2012 } 1942 }
2013 } 1943 }
2014 1944
2015 down_write(&OCFS2_I(inode)->ip_alloc_ 1945 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2016 switch (cmd) { 1946 switch (cmd) {
2017 case OCFS2_IOC_RESVSP: 1947 case OCFS2_IOC_RESVSP:
2018 case OCFS2_IOC_RESVSP64: 1948 case OCFS2_IOC_RESVSP64:
2019 /* 1949 /*
2020 * This takes unsigned offset 1950 * This takes unsigned offsets, but the signed ones we
2021 * pass have been checked aga 1951 * pass have been checked against overflow above.
2022 */ 1952 */
2023 ret = ocfs2_allocate_unwritte 1953 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2024 1954 sr->l_len);
2025 break; 1955 break;
2026 case OCFS2_IOC_UNRESVSP: 1956 case OCFS2_IOC_UNRESVSP:
2027 case OCFS2_IOC_UNRESVSP64: 1957 case OCFS2_IOC_UNRESVSP64:
2028 ret = ocfs2_remove_inode_rang 1958 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2029 1959 sr->l_len);
2030 break; 1960 break;
2031 default: 1961 default:
2032 ret = -EINVAL; 1962 ret = -EINVAL;
2033 } 1963 }
2034 <<
2035 orig_isize = i_size_read(inode); <<
2036 /* zeroout eof blocks in the cluster. <<
2037 if (!ret && change_size && orig_isize <<
2038 ret = ocfs2_zeroout_partial_c <<
2039 size <<
2040 if (!ret) <<
2041 i_size_write(inode, s <<
2042 } <<
2043 up_write(&OCFS2_I(inode)->ip_alloc_se 1964 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2044 if (ret) { 1965 if (ret) {
2045 mlog_errno(ret); 1966 mlog_errno(ret);
2046 goto out_inode_unlock; 1967 goto out_inode_unlock;
2047 } 1968 }
2048 1969
2049 /* 1970 /*
2050 * We update c/mtime for these change 1971 * We update c/mtime for these changes
2051 */ 1972 */
2052 handle = ocfs2_start_trans(osb, OCFS2 1973 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2053 if (IS_ERR(handle)) { 1974 if (IS_ERR(handle)) {
2054 ret = PTR_ERR(handle); 1975 ret = PTR_ERR(handle);
2055 mlog_errno(ret); 1976 mlog_errno(ret);
2056 goto out_inode_unlock; 1977 goto out_inode_unlock;
2057 } 1978 }
2058 1979
2059 inode_set_mtime_to_ts(inode, inode_se !! 1980 if (change_size && i_size_read(inode) < size)
>> 1981 i_size_write(inode, size);
>> 1982
>> 1983 inode->i_ctime = inode->i_mtime = current_time(inode);
2060 ret = ocfs2_mark_inode_dirty(handle, 1984 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2061 if (ret < 0) 1985 if (ret < 0)
2062 mlog_errno(ret); 1986 mlog_errno(ret);
2063 1987
2064 if (file && (file->f_flags & O_SYNC)) 1988 if (file && (file->f_flags & O_SYNC))
2065 handle->h_sync = 1; 1989 handle->h_sync = 1;
2066 1990
2067 ocfs2_commit_trans(osb, handle); 1991 ocfs2_commit_trans(osb, handle);
2068 1992
2069 out_inode_unlock: 1993 out_inode_unlock:
2070 brelse(di_bh); 1994 brelse(di_bh);
2071 ocfs2_inode_unlock(inode, 1); 1995 ocfs2_inode_unlock(inode, 1);
2072 out_rw_unlock: 1996 out_rw_unlock:
2073 ocfs2_rw_unlock(inode, 1); 1997 ocfs2_rw_unlock(inode, 1);
2074 1998
2075 out: 1999 out:
2076 inode_unlock(inode); 2000 inode_unlock(inode);
2077 return ret; 2001 return ret;
2078 } 2002 }
2079 2003
2080 int ocfs2_change_file_space(struct file *file 2004 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2081 struct ocfs2_spac 2005 struct ocfs2_space_resv *sr)
2082 { 2006 {
2083 struct inode *inode = file_inode(file 2007 struct inode *inode = file_inode(file);
2084 struct ocfs2_super *osb = OCFS2_SB(in 2008 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2085 int ret; 2009 int ret;
2086 2010
2087 if ((cmd == OCFS2_IOC_RESVSP || cmd = 2011 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2088 !ocfs2_writes_unwritten_extents(o 2012 !ocfs2_writes_unwritten_extents(osb))
2089 return -ENOTTY; 2013 return -ENOTTY;
2090 else if ((cmd == OCFS2_IOC_UNRESVSP | 2014 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2091 !ocfs2_sparse_alloc(osb)) 2015 !ocfs2_sparse_alloc(osb))
2092 return -ENOTTY; 2016 return -ENOTTY;
2093 2017
2094 if (!S_ISREG(inode->i_mode)) 2018 if (!S_ISREG(inode->i_mode))
2095 return -EINVAL; 2019 return -EINVAL;
2096 2020
2097 if (!(file->f_mode & FMODE_WRITE)) 2021 if (!(file->f_mode & FMODE_WRITE))
2098 return -EBADF; 2022 return -EBADF;
2099 2023
2100 ret = mnt_want_write_file(file); 2024 ret = mnt_want_write_file(file);
2101 if (ret) 2025 if (ret)
2102 return ret; 2026 return ret;
2103 ret = __ocfs2_change_file_space(file, 2027 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2104 mnt_drop_write_file(file); 2028 mnt_drop_write_file(file);
2105 return ret; 2029 return ret;
2106 } 2030 }
2107 2031
2108 static long ocfs2_fallocate(struct file *file 2032 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2109 loff_t len) 2033 loff_t len)
2110 { 2034 {
2111 struct inode *inode = file_inode(file 2035 struct inode *inode = file_inode(file);
2112 struct ocfs2_super *osb = OCFS2_SB(in 2036 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2113 struct ocfs2_space_resv sr; 2037 struct ocfs2_space_resv sr;
2114 int change_size = 1; 2038 int change_size = 1;
2115 int cmd = OCFS2_IOC_RESVSP64; 2039 int cmd = OCFS2_IOC_RESVSP64;
2116 int ret = 0; <<
2117 2040
2118 if (mode & ~(FALLOC_FL_KEEP_SIZE | FA 2041 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2119 return -EOPNOTSUPP; 2042 return -EOPNOTSUPP;
2120 if (!ocfs2_writes_unwritten_extents(o 2043 if (!ocfs2_writes_unwritten_extents(osb))
2121 return -EOPNOTSUPP; 2044 return -EOPNOTSUPP;
2122 2045
2123 if (mode & FALLOC_FL_KEEP_SIZE) { !! 2046 if (mode & FALLOC_FL_KEEP_SIZE)
2124 change_size = 0; 2047 change_size = 0;
2125 } else { <<
2126 ret = inode_newsize_ok(inode, <<
2127 if (ret) <<
2128 return ret; <<
2129 } <<
2130 2048
2131 if (mode & FALLOC_FL_PUNCH_HOLE) 2049 if (mode & FALLOC_FL_PUNCH_HOLE)
2132 cmd = OCFS2_IOC_UNRESVSP64; 2050 cmd = OCFS2_IOC_UNRESVSP64;
2133 2051
2134 sr.l_whence = 0; 2052 sr.l_whence = 0;
2135 sr.l_start = (s64)offset; 2053 sr.l_start = (s64)offset;
2136 sr.l_len = (s64)len; 2054 sr.l_len = (s64)len;
2137 2055
2138 return __ocfs2_change_file_space(NULL 2056 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2139 chan 2057 change_size);
2140 } 2058 }
2141 2059
2142 int ocfs2_check_range_for_refcount(struct ino 2060 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2143 size_t cou 2061 size_t count)
2144 { 2062 {
2145 int ret = 0; 2063 int ret = 0;
2146 unsigned int extent_flags; 2064 unsigned int extent_flags;
2147 u32 cpos, clusters, extent_len, phys_ 2065 u32 cpos, clusters, extent_len, phys_cpos;
2148 struct super_block *sb = inode->i_sb; 2066 struct super_block *sb = inode->i_sb;
2149 2067
2150 if (!ocfs2_refcount_tree(OCFS2_SB(ino 2068 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2151 !ocfs2_is_refcount_inode(inode) | 2069 !ocfs2_is_refcount_inode(inode) ||
2152 OCFS2_I(inode)->ip_dyn_features & 2070 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2153 return 0; 2071 return 0;
2154 2072
2155 cpos = pos >> OCFS2_SB(sb)->s_cluster 2073 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2156 clusters = ocfs2_clusters_for_bytes(s 2074 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2157 2075
2158 while (clusters) { 2076 while (clusters) {
2159 ret = ocfs2_get_clusters(inod 2077 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2160 &ext 2078 &extent_flags);
2161 if (ret < 0) { 2079 if (ret < 0) {
2162 mlog_errno(ret); 2080 mlog_errno(ret);
2163 goto out; 2081 goto out;
2164 } 2082 }
2165 2083
2166 if (phys_cpos && (extent_flag 2084 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2167 ret = 1; 2085 ret = 1;
2168 break; 2086 break;
2169 } 2087 }
2170 2088
2171 if (extent_len > clusters) 2089 if (extent_len > clusters)
2172 extent_len = clusters 2090 extent_len = clusters;
2173 2091
2174 clusters -= extent_len; 2092 clusters -= extent_len;
2175 cpos += extent_len; 2093 cpos += extent_len;
2176 } 2094 }
2177 out: 2095 out:
2178 return ret; 2096 return ret;
2179 } 2097 }
2180 2098
2181 static int ocfs2_is_io_unaligned(struct inode 2099 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2182 { 2100 {
2183 int blockmask = inode->i_sb->s_blocks 2101 int blockmask = inode->i_sb->s_blocksize - 1;
2184 loff_t final_size = pos + count; 2102 loff_t final_size = pos + count;
2185 2103
2186 if ((pos & blockmask) || (final_size 2104 if ((pos & blockmask) || (final_size & blockmask))
2187 return 1; 2105 return 1;
2188 return 0; 2106 return 0;
2189 } 2107 }
2190 2108
2191 static int ocfs2_inode_lock_for_extent_tree(s !! 2109 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2192 s !! 2110 struct file *file,
2193 i !! 2111 loff_t pos, size_t count,
2194 i !! 2112 int *meta_level)
2195 i <<
2196 { 2113 {
2197 int ret = 0; !! 2114 int ret;
>> 2115 struct buffer_head *di_bh = NULL;
>> 2116 u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
>> 2117 u32 clusters =
>> 2118 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2198 2119
2199 if (wait) !! 2120 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2200 ret = ocfs2_inode_lock(inode, !! 2121 if (ret) {
2201 else !! 2122 mlog_errno(ret);
2202 ret = ocfs2_try_inode_lock(in <<
2203 if (ret < 0) <<
2204 goto out; 2123 goto out;
2205 <<
2206 if (wait) { <<
2207 if (write_sem) <<
2208 down_write(&OCFS2_I(i <<
2209 else <<
2210 down_read(&OCFS2_I(in <<
2211 } else { <<
2212 if (write_sem) <<
2213 ret = down_write_tryl <<
2214 else <<
2215 ret = down_read_trylo <<
2216 <<
2217 if (!ret) { <<
2218 ret = -EAGAIN; <<
2219 goto out_unlock; <<
2220 } <<
2221 } 2124 }
2222 2125
2223 return ret; !! 2126 *meta_level = 1;
2224 2127
2225 out_unlock: !! 2128 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2226 brelse(*di_bh); !! 2129 if (ret)
2227 *di_bh = NULL; !! 2130 mlog_errno(ret);
2228 ocfs2_inode_unlock(inode, meta_level) <<
2229 out: 2131 out:
>> 2132 brelse(di_bh);
2230 return ret; 2133 return ret;
2231 } 2134 }
2232 2135
2233 static void ocfs2_inode_unlock_for_extent_tre <<
2234 <<
2235 <<
2236 <<
2237 { <<
2238 if (write_sem) <<
2239 up_write(&OCFS2_I(inode)->ip_ <<
2240 else <<
2241 up_read(&OCFS2_I(inode)->ip_a <<
2242 <<
2243 brelse(*di_bh); <<
2244 *di_bh = NULL; <<
2245 <<
2246 if (meta_level >= 0) <<
2247 ocfs2_inode_unlock(inode, met <<
2248 } <<
2249 <<
2250 static int ocfs2_prepare_inode_for_write(stru 2136 static int ocfs2_prepare_inode_for_write(struct file *file,
2251 loff 2137 loff_t pos, size_t count, int wait)
2252 { 2138 {
2253 int ret = 0, meta_level = 0, overwrit 2139 int ret = 0, meta_level = 0, overwrite_io = 0;
2254 int write_sem = 0; <<
2255 struct dentry *dentry = file->f_path. 2140 struct dentry *dentry = file->f_path.dentry;
2256 struct inode *inode = d_inode(dentry) 2141 struct inode *inode = d_inode(dentry);
2257 struct buffer_head *di_bh = NULL; 2142 struct buffer_head *di_bh = NULL;
2258 u32 cpos; !! 2143 loff_t end;
2259 u32 clusters; <<
2260 2144
2261 /* 2145 /*
2262 * We start with a read level meta lo 2146 * We start with a read level meta lock and only jump to an ex
2263 * if we need to make modifications h 2147 * if we need to make modifications here.
2264 */ 2148 */
2265 for(;;) { 2149 for(;;) {
2266 ret = ocfs2_inode_lock_for_ex !! 2150 if (wait)
2267 !! 2151 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2268 !! 2152 else
2269 !! 2153 ret = ocfs2_try_inode_lock(inode,
2270 !! 2154 overwrite_io ? NULL : &di_bh, meta_level);
2271 if (ret < 0) { 2155 if (ret < 0) {
>> 2156 meta_level = -1;
2272 if (ret != -EAGAIN) 2157 if (ret != -EAGAIN)
2273 mlog_errno(re 2158 mlog_errno(ret);
2274 goto out; 2159 goto out;
2275 } 2160 }
2276 2161
2277 /* 2162 /*
2278 * Check if IO will overwrite 2163 * Check if IO will overwrite allocated blocks in case
2279 * IOCB_NOWAIT flag is set. 2164 * IOCB_NOWAIT flag is set.
2280 */ 2165 */
2281 if (!wait && !overwrite_io) { 2166 if (!wait && !overwrite_io) {
2282 overwrite_io = 1; 2167 overwrite_io = 1;
>> 2168 if (!down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem)) {
>> 2169 ret = -EAGAIN;
>> 2170 goto out_unlock;
>> 2171 }
2283 2172
2284 ret = ocfs2_overwrite 2173 ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
>> 2174 brelse(di_bh);
>> 2175 di_bh = NULL;
>> 2176 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2285 if (ret < 0) { 2177 if (ret < 0) {
2286 if (ret != -E 2178 if (ret != -EAGAIN)
2287 mlog_ 2179 mlog_errno(ret);
2288 goto out_unlo 2180 goto out_unlock;
2289 } 2181 }
2290 } 2182 }
2291 2183
2292 /* Clear suid / sgid if neces 2184 /* Clear suid / sgid if necessary. We do this here
2293 * instead of later in the wr 2185 * instead of later in the write path because
2294 * remove_suid() calls ->seta 2186 * remove_suid() calls ->setattr without any hint that
2295 * we may have already done o 2187 * we may have already done our cluster locking. Since
2296 * ocfs2_setattr() *must* tak 2188 * ocfs2_setattr() *must* take cluster locks to
2297 * proceed, this will lead us 2189 * proceed, this will lead us to recursively lock the
2298 * inode. There's also the di 2190 * inode. There's also the dinode i_size state which
2299 * can be lost via setattr du 2191 * can be lost via setattr during extending writes (we
2300 * set inode->i_size at the e 2192 * set inode->i_size at the end of a write. */
2301 if (setattr_should_drop_suidg !! 2193 if (should_remove_suid(dentry)) {
2302 if (meta_level == 0) 2194 if (meta_level == 0) {
2303 ocfs2_inode_u !! 2195 ocfs2_inode_unlock(inode, meta_level);
2304 <<
2305 <<
2306 <<
2307 meta_level = 2196 meta_level = 1;
2308 continue; 2197 continue;
2309 } 2198 }
2310 2199
2311 ret = ocfs2_write_rem 2200 ret = ocfs2_write_remove_suid(inode);
2312 if (ret < 0) { 2201 if (ret < 0) {
2313 mlog_errno(re 2202 mlog_errno(ret);
2314 goto out_unlo 2203 goto out_unlock;
2315 } 2204 }
2316 } 2205 }
2317 2206
>> 2207 end = pos + count;
>> 2208
2318 ret = ocfs2_check_range_for_r 2209 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2319 if (ret == 1) { 2210 if (ret == 1) {
2320 ocfs2_inode_unlock_fo !! 2211 ocfs2_inode_unlock(inode, meta_level);
2321 !! 2212 meta_level = -1;
2322 <<
2323 <<
2324 meta_level = 1; <<
2325 write_sem = 1; <<
2326 ret = ocfs2_inode_loc <<
2327 <<
2328 <<
2329 <<
2330 <<
2331 if (ret < 0) { <<
2332 if (ret != -E <<
2333 mlog_ <<
2334 goto out; <<
2335 } <<
2336 2213
2337 cpos = pos >> OCFS2_S !! 2214 ret = ocfs2_prepare_inode_for_refcount(inode,
2338 clusters = !! 2215 file,
2339 ocfs2_cluster !! 2216 pos,
2340 ret = ocfs2_refcount_ !! 2217 count,
>> 2218 &meta_level);
2341 } 2219 }
2342 2220
2343 if (ret < 0) { 2221 if (ret < 0) {
2344 if (ret != -EAGAIN) !! 2222 mlog_errno(ret);
2345 mlog_errno(re <<
2346 goto out_unlock; 2223 goto out_unlock;
2347 } 2224 }
2348 2225
2349 break; 2226 break;
2350 } 2227 }
2351 2228
2352 out_unlock: 2229 out_unlock:
2353 trace_ocfs2_prepare_inode_for_write(O 2230 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2354 p 2231 pos, count, wait);
2355 2232
2356 ocfs2_inode_unlock_for_extent_tree(in !! 2233 brelse(di_bh);
2357 &d !! 2234
2358 me !! 2235 if (meta_level >= 0)
2359 wr !! 2236 ocfs2_inode_unlock(inode, meta_level);
2360 2237
2361 out: 2238 out:
2362 return ret; 2239 return ret;
2363 } 2240 }
2364 2241
2365 static ssize_t ocfs2_file_write_iter(struct k 2242 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2366 struct io 2243 struct iov_iter *from)
2367 { 2244 {
2368 int rw_level; 2245 int rw_level;
2369 ssize_t written = 0; 2246 ssize_t written = 0;
2370 ssize_t ret; 2247 ssize_t ret;
2371 size_t count = iov_iter_count(from); 2248 size_t count = iov_iter_count(from);
2372 struct file *file = iocb->ki_filp; 2249 struct file *file = iocb->ki_filp;
2373 struct inode *inode = file_inode(file 2250 struct inode *inode = file_inode(file);
2374 struct ocfs2_super *osb = OCFS2_SB(in 2251 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2375 int full_coherency = !(osb->s_mount_o 2252 int full_coherency = !(osb->s_mount_opt &
2376 OCFS2_MOUNT_CO 2253 OCFS2_MOUNT_COHERENCY_BUFFERED);
2377 void *saved_ki_complete = NULL; 2254 void *saved_ki_complete = NULL;
2378 int append_write = ((iocb->ki_pos + c 2255 int append_write = ((iocb->ki_pos + count) >=
2379 i_size_read(inode) ? 2256 i_size_read(inode) ? 1 : 0);
2380 int direct_io = iocb->ki_flags & IOCB 2257 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2381 int nowait = iocb->ki_flags & IOCB_NO 2258 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2382 2259
2383 trace_ocfs2_file_write_iter(inode, fi 2260 trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2384 (unsigned long long)OCFS2_I(i 2261 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2385 file->f_path.dentry->d_name.l 2262 file->f_path.dentry->d_name.len,
2386 file->f_path.dentry->d_name.n 2263 file->f_path.dentry->d_name.name,
2387 (unsigned int)from->nr_segs); 2264 (unsigned int)from->nr_segs); /* GRRRRR */
2388 2265
2389 if (!direct_io && nowait) 2266 if (!direct_io && nowait)
2390 return -EOPNOTSUPP; 2267 return -EOPNOTSUPP;
2391 2268
2392 if (count == 0) 2269 if (count == 0)
2393 return 0; 2270 return 0;
2394 2271
2395 if (nowait) { 2272 if (nowait) {
2396 if (!inode_trylock(inode)) 2273 if (!inode_trylock(inode))
2397 return -EAGAIN; 2274 return -EAGAIN;
2398 } else 2275 } else
2399 inode_lock(inode); 2276 inode_lock(inode);
2400 2277
2401 /* 2278 /*
2402 * Concurrent O_DIRECT writes are all 2279 * Concurrent O_DIRECT writes are allowed with
2403 * mount_option "coherency=buffered". 2280 * mount_option "coherency=buffered".
2404 * For append write, we must take rw 2281 * For append write, we must take rw EX.
2405 */ 2282 */
2406 rw_level = (!direct_io || full_cohere 2283 rw_level = (!direct_io || full_coherency || append_write);
2407 2284
2408 if (nowait) 2285 if (nowait)
2409 ret = ocfs2_try_rw_lock(inode 2286 ret = ocfs2_try_rw_lock(inode, rw_level);
2410 else 2287 else
2411 ret = ocfs2_rw_lock(inode, rw 2288 ret = ocfs2_rw_lock(inode, rw_level);
2412 if (ret < 0) { 2289 if (ret < 0) {
2413 if (ret != -EAGAIN) 2290 if (ret != -EAGAIN)
2414 mlog_errno(ret); 2291 mlog_errno(ret);
2415 goto out_mutex; 2292 goto out_mutex;
2416 } 2293 }
2417 2294
2418 /* 2295 /*
2419 * O_DIRECT writes with "coherency=fu 2296 * O_DIRECT writes with "coherency=full" need to take EX cluster
2420 * inode_lock to guarantee coherency. 2297 * inode_lock to guarantee coherency.
2421 */ 2298 */
2422 if (direct_io && full_coherency) { 2299 if (direct_io && full_coherency) {
2423 /* 2300 /*
2424 * We need to take and drop t 2301 * We need to take and drop the inode lock to force
2425 * other nodes to drop their 2302 * other nodes to drop their caches. Buffered I/O
2426 * already does this in write 2303 * already does this in write_begin().
2427 */ 2304 */
2428 if (nowait) 2305 if (nowait)
2429 ret = ocfs2_try_inode 2306 ret = ocfs2_try_inode_lock(inode, NULL, 1);
2430 else 2307 else
2431 ret = ocfs2_inode_loc 2308 ret = ocfs2_inode_lock(inode, NULL, 1);
2432 if (ret < 0) { 2309 if (ret < 0) {
2433 if (ret != -EAGAIN) 2310 if (ret != -EAGAIN)
2434 mlog_errno(re 2311 mlog_errno(ret);
2435 goto out; 2312 goto out;
2436 } 2313 }
2437 2314
2438 ocfs2_inode_unlock(inode, 1); 2315 ocfs2_inode_unlock(inode, 1);
2439 } 2316 }
2440 2317
2441 ret = generic_write_checks(iocb, from 2318 ret = generic_write_checks(iocb, from);
2442 if (ret <= 0) { 2319 if (ret <= 0) {
2443 if (ret) 2320 if (ret)
2444 mlog_errno(ret); 2321 mlog_errno(ret);
2445 goto out; 2322 goto out;
2446 } 2323 }
2447 count = ret; 2324 count = ret;
2448 2325
2449 ret = ocfs2_prepare_inode_for_write(f 2326 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2450 if (ret < 0) { 2327 if (ret < 0) {
2451 if (ret != -EAGAIN) 2328 if (ret != -EAGAIN)
2452 mlog_errno(ret); 2329 mlog_errno(ret);
2453 goto out; 2330 goto out;
2454 } 2331 }
2455 2332
2456 if (direct_io && !is_sync_kiocb(iocb) 2333 if (direct_io && !is_sync_kiocb(iocb) &&
2457 ocfs2_is_io_unaligned(inode, coun 2334 ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2458 /* 2335 /*
2459 * Make it a sync io if it's 2336 * Make it a sync io if it's an unaligned aio.
2460 */ 2337 */
2461 saved_ki_complete = xchg(&ioc 2338 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2462 } 2339 }
2463 2340
2464 /* communicate with ocfs2_dio_end_io 2341 /* communicate with ocfs2_dio_end_io */
2465 ocfs2_iocb_set_rw_locked(iocb, rw_lev 2342 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2466 2343
2467 written = __generic_file_write_iter(i 2344 written = __generic_file_write_iter(iocb, from);
2468 /* buffered aio wouldn't have proper 2345 /* buffered aio wouldn't have proper lock coverage today */
2469 BUG_ON(written == -EIOCBQUEUED && !di 2346 BUG_ON(written == -EIOCBQUEUED && !direct_io);
2470 2347
2471 /* 2348 /*
2472 * deep in g_f_a_w_n()->ocfs2_direct_ 2349 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2473 * function pointer which is called w 2350 * function pointer which is called when o_direct io completes so that
2474 * it can unlock our rw lock. 2351 * it can unlock our rw lock.
2475 * Unfortunately there are error case 2352 * Unfortunately there are error cases which call end_io and others
2476 * that don't. so we don't have to u 2353 * that don't. so we don't have to unlock the rw_lock if either an
2477 * async dio is going to do it in the 2354 * async dio is going to do it in the future or an end_io after an
2478 * error has already done it. 2355 * error has already done it.
2479 */ 2356 */
2480 if ((written == -EIOCBQUEUED) || (!oc 2357 if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2481 rw_level = -1; 2358 rw_level = -1;
2482 } 2359 }
2483 2360
2484 if (unlikely(written <= 0)) 2361 if (unlikely(written <= 0))
2485 goto out; 2362 goto out;
2486 2363
2487 if (((file->f_flags & O_DSYNC) && !di 2364 if (((file->f_flags & O_DSYNC) && !direct_io) ||
2488 IS_SYNC(inode)) { 2365 IS_SYNC(inode)) {
2489 ret = filemap_fdatawrite_rang 2366 ret = filemap_fdatawrite_range(file->f_mapping,
2490 2367 iocb->ki_pos - written,
2491 2368 iocb->ki_pos - 1);
2492 if (ret < 0) 2369 if (ret < 0)
2493 written = ret; 2370 written = ret;
2494 2371
2495 if (!ret) { 2372 if (!ret) {
2496 ret = jbd2_journal_fo 2373 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2497 if (ret < 0) 2374 if (ret < 0)
2498 written = ret 2375 written = ret;
2499 } 2376 }
2500 2377
2501 if (!ret) 2378 if (!ret)
2502 ret = filemap_fdatawa 2379 ret = filemap_fdatawait_range(file->f_mapping,
2503 2380 iocb->ki_pos - written,
2504 2381 iocb->ki_pos - 1);
2505 } 2382 }
2506 2383
2507 out: 2384 out:
2508 if (saved_ki_complete) 2385 if (saved_ki_complete)
2509 xchg(&iocb->ki_complete, save 2386 xchg(&iocb->ki_complete, saved_ki_complete);
2510 2387
2511 if (rw_level != -1) 2388 if (rw_level != -1)
2512 ocfs2_rw_unlock(inode, rw_lev 2389 ocfs2_rw_unlock(inode, rw_level);
2513 2390
2514 out_mutex: 2391 out_mutex:
2515 inode_unlock(inode); 2392 inode_unlock(inode);
2516 2393
2517 if (written) 2394 if (written)
2518 ret = written; 2395 ret = written;
2519 return ret; 2396 return ret;
2520 } 2397 }
2521 2398
2522 static ssize_t ocfs2_file_read_iter(struct ki 2399 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2523 struct iov 2400 struct iov_iter *to)
2524 { 2401 {
2525 int ret = 0, rw_level = -1, lock_leve 2402 int ret = 0, rw_level = -1, lock_level = 0;
2526 struct file *filp = iocb->ki_filp; 2403 struct file *filp = iocb->ki_filp;
2527 struct inode *inode = file_inode(filp 2404 struct inode *inode = file_inode(filp);
2528 int direct_io = iocb->ki_flags & IOCB 2405 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2529 int nowait = iocb->ki_flags & IOCB_NO 2406 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2530 2407
2531 trace_ocfs2_file_read_iter(inode, fil 2408 trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2532 (unsigned long long)O 2409 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2533 filp->f_path.dentry-> 2410 filp->f_path.dentry->d_name.len,
2534 filp->f_path.dentry-> 2411 filp->f_path.dentry->d_name.name,
2535 to->nr_segs); /* GR 2412 to->nr_segs); /* GRRRRR */
2536 2413
2537 2414
2538 if (!inode) { 2415 if (!inode) {
2539 ret = -EINVAL; 2416 ret = -EINVAL;
2540 mlog_errno(ret); 2417 mlog_errno(ret);
2541 goto bail; 2418 goto bail;
2542 } 2419 }
2543 2420
2544 if (!direct_io && nowait) 2421 if (!direct_io && nowait)
2545 return -EOPNOTSUPP; 2422 return -EOPNOTSUPP;
2546 2423
2547 /* 2424 /*
2548 * buffered reads protect themselves !! 2425 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2549 * need locks to protect pending read 2426 * need locks to protect pending reads from racing with truncate.
2550 */ 2427 */
2551 if (direct_io) { 2428 if (direct_io) {
2552 if (nowait) 2429 if (nowait)
2553 ret = ocfs2_try_rw_lo 2430 ret = ocfs2_try_rw_lock(inode, 0);
2554 else 2431 else
2555 ret = ocfs2_rw_lock(i 2432 ret = ocfs2_rw_lock(inode, 0);
2556 2433
2557 if (ret < 0) { 2434 if (ret < 0) {
2558 if (ret != -EAGAIN) 2435 if (ret != -EAGAIN)
2559 mlog_errno(re 2436 mlog_errno(ret);
2560 goto bail; 2437 goto bail;
2561 } 2438 }
2562 rw_level = 0; 2439 rw_level = 0;
2563 /* communicate with ocfs2_dio 2440 /* communicate with ocfs2_dio_end_io */
2564 ocfs2_iocb_set_rw_locked(iocb 2441 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2565 } 2442 }
2566 2443
2567 /* 2444 /*
2568 * We're fine letting folks race trun 2445 * We're fine letting folks race truncates and extending
2569 * writes with read across the cluste 2446 * writes with read across the cluster, just like they can
2570 * locally. Hence no rw_lock during r 2447 * locally. Hence no rw_lock during read.
2571 * 2448 *
2572 * Take and drop the meta data lock t 2449 * Take and drop the meta data lock to update inode fields
2573 * like i_size. This allows the check 2450 * like i_size. This allows the checks down below
2574 * copy_splice_read() a chance of act !! 2451 * generic_file_read_iter() a chance of actually working.
2575 */ 2452 */
2576 ret = ocfs2_inode_lock_atime(inode, f 2453 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2577 !nowait) 2454 !nowait);
2578 if (ret < 0) { 2455 if (ret < 0) {
2579 if (ret != -EAGAIN) 2456 if (ret != -EAGAIN)
2580 mlog_errno(ret); 2457 mlog_errno(ret);
2581 goto bail; 2458 goto bail;
2582 } 2459 }
2583 ocfs2_inode_unlock(inode, lock_level) 2460 ocfs2_inode_unlock(inode, lock_level);
2584 2461
2585 ret = generic_file_read_iter(iocb, to 2462 ret = generic_file_read_iter(iocb, to);
2586 trace_generic_file_read_iter_ret(ret) 2463 trace_generic_file_read_iter_ret(ret);
2587 2464
2588 /* buffered aio wouldn't have proper 2465 /* buffered aio wouldn't have proper lock coverage today */
2589 BUG_ON(ret == -EIOCBQUEUED && !direct 2466 BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2590 2467
2591 /* see ocfs2_file_write_iter */ 2468 /* see ocfs2_file_write_iter */
2592 if (ret == -EIOCBQUEUED || !ocfs2_ioc 2469 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2593 rw_level = -1; 2470 rw_level = -1;
2594 } 2471 }
2595 2472
2596 bail: 2473 bail:
2597 if (rw_level != -1) 2474 if (rw_level != -1)
2598 ocfs2_rw_unlock(inode, rw_lev 2475 ocfs2_rw_unlock(inode, rw_level);
2599 2476
2600 return ret; 2477 return ret;
2601 } 2478 }
2602 2479
2603 static ssize_t ocfs2_file_splice_read(struct <<
2604 struct <<
2605 size_t <<
2606 { <<
2607 struct inode *inode = file_inode(in); <<
2608 ssize_t ret = 0; <<
2609 int lock_level = 0; <<
2610 <<
2611 trace_ocfs2_file_splice_read(inode, i <<
2612 (unsigne <<
2613 in->f_pa <<
2614 in->f_pa <<
2615 flags); <<
2616 <<
2617 /* <<
2618 * We're fine letting folks race trun <<
2619 * read across the cluster, just like <<
2620 * rw_lock during read. <<
2621 * <<
2622 * Take and drop the meta data lock t <<
2623 * This allows the checks down below <<
2624 * actually working. <<
2625 */ <<
2626 ret = ocfs2_inode_lock_atime(inode, i <<
2627 if (ret < 0) { <<
2628 if (ret != -EAGAIN) <<
2629 mlog_errno(ret); <<
2630 goto bail; <<
2631 } <<
2632 ocfs2_inode_unlock(inode, lock_level) <<
2633 <<
2634 ret = filemap_splice_read(in, ppos, p <<
2635 trace_filemap_splice_read_ret(ret); <<
2636 bail: <<
2637 return ret; <<
2638 } <<
2639 <<
2640 /* Refer generic_file_llseek_unlocked() */ 2480 /* Refer generic_file_llseek_unlocked() */
2641 static loff_t ocfs2_file_llseek(struct file * 2481 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2642 { 2482 {
2643 struct inode *inode = file->f_mapping 2483 struct inode *inode = file->f_mapping->host;
2644 int ret = 0; 2484 int ret = 0;
2645 2485
2646 inode_lock(inode); 2486 inode_lock(inode);
2647 2487
2648 switch (whence) { 2488 switch (whence) {
2649 case SEEK_SET: 2489 case SEEK_SET:
2650 break; 2490 break;
2651 case SEEK_END: 2491 case SEEK_END:
2652 /* SEEK_END requires the OCFS 2492 /* SEEK_END requires the OCFS2 inode lock for the file
2653 * because it references the 2493 * because it references the file's size.
2654 */ 2494 */
2655 ret = ocfs2_inode_lock(inode, 2495 ret = ocfs2_inode_lock(inode, NULL, 0);
2656 if (ret < 0) { 2496 if (ret < 0) {
2657 mlog_errno(ret); 2497 mlog_errno(ret);
2658 goto out; 2498 goto out;
2659 } 2499 }
2660 offset += i_size_read(inode); 2500 offset += i_size_read(inode);
2661 ocfs2_inode_unlock(inode, 0); 2501 ocfs2_inode_unlock(inode, 0);
2662 break; 2502 break;
2663 case SEEK_CUR: 2503 case SEEK_CUR:
2664 if (offset == 0) { 2504 if (offset == 0) {
2665 offset = file->f_pos; 2505 offset = file->f_pos;
2666 goto out; 2506 goto out;
2667 } 2507 }
2668 offset += file->f_pos; 2508 offset += file->f_pos;
2669 break; 2509 break;
2670 case SEEK_DATA: 2510 case SEEK_DATA:
2671 case SEEK_HOLE: 2511 case SEEK_HOLE:
2672 ret = ocfs2_seek_data_hole_of 2512 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2673 if (ret) 2513 if (ret)
2674 goto out; 2514 goto out;
2675 break; 2515 break;
2676 default: 2516 default:
2677 ret = -EINVAL; 2517 ret = -EINVAL;
2678 goto out; 2518 goto out;
2679 } 2519 }
2680 2520
2681 offset = vfs_setpos(file, offset, ino 2521 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2682 2522
2683 out: 2523 out:
2684 inode_unlock(inode); 2524 inode_unlock(inode);
2685 if (ret) 2525 if (ret)
2686 return ret; 2526 return ret;
2687 return offset; 2527 return offset;
2688 } 2528 }
2689 2529
2690 static loff_t ocfs2_remap_file_range(struct f 2530 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2691 struct f 2531 struct file *file_out, loff_t pos_out,
2692 loff_t l 2532 loff_t len, unsigned int remap_flags)
2693 { 2533 {
2694 struct inode *inode_in = file_inode(f 2534 struct inode *inode_in = file_inode(file_in);
2695 struct inode *inode_out = file_inode( 2535 struct inode *inode_out = file_inode(file_out);
2696 struct ocfs2_super *osb = OCFS2_SB(in 2536 struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2697 struct buffer_head *in_bh = NULL, *ou 2537 struct buffer_head *in_bh = NULL, *out_bh = NULL;
2698 bool same_inode = (inode_in == inode_ 2538 bool same_inode = (inode_in == inode_out);
2699 loff_t remapped = 0; 2539 loff_t remapped = 0;
2700 ssize_t ret; 2540 ssize_t ret;
2701 2541
2702 if (remap_flags & ~(REMAP_FILE_DEDUP 2542 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2703 return -EINVAL; 2543 return -EINVAL;
2704 if (!ocfs2_refcount_tree(osb)) 2544 if (!ocfs2_refcount_tree(osb))
2705 return -EOPNOTSUPP; 2545 return -EOPNOTSUPP;
2706 if (ocfs2_is_hard_readonly(osb) || oc 2546 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2707 return -EROFS; 2547 return -EROFS;
2708 2548
2709 /* Lock both files against IO */ 2549 /* Lock both files against IO */
2710 ret = ocfs2_reflink_inodes_lock(inode 2550 ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2711 if (ret) 2551 if (ret)
2712 return ret; 2552 return ret;
2713 2553
2714 /* Check file eligibility and prepare 2554 /* Check file eligibility and prepare for block sharing. */
2715 ret = -EINVAL; 2555 ret = -EINVAL;
2716 if ((OCFS2_I(inode_in)->ip_flags & OC 2556 if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2717 (OCFS2_I(inode_out)->ip_flags & O 2557 (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2718 goto out_unlock; 2558 goto out_unlock;
2719 2559
2720 ret = generic_remap_file_range_prep(f 2560 ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2721 &len, remap_flags); 2561 &len, remap_flags);
2722 if (ret < 0 || len == 0) 2562 if (ret < 0 || len == 0)
2723 goto out_unlock; 2563 goto out_unlock;
2724 2564
2725 /* Lock out changes to the allocation 2565 /* Lock out changes to the allocation maps and remap. */
2726 down_write(&OCFS2_I(inode_in)->ip_all 2566 down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2727 if (!same_inode) 2567 if (!same_inode)
2728 down_write_nested(&OCFS2_I(in 2568 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2729 SINGLE_DEPT 2569 SINGLE_DEPTH_NESTING);
2730 2570
2731 /* Zap any page cache for the destina 2571 /* Zap any page cache for the destination file's range. */
2732 truncate_inode_pages_range(&inode_out 2572 truncate_inode_pages_range(&inode_out->i_data,
2733 round_down 2573 round_down(pos_out, PAGE_SIZE),
2734 round_up(p 2574 round_up(pos_out + len, PAGE_SIZE) - 1);
2735 2575
2736 remapped = ocfs2_reflink_remap_blocks 2576 remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2737 inode_out, out_bh, po 2577 inode_out, out_bh, pos_out, len);
2738 up_write(&OCFS2_I(inode_in)->ip_alloc 2578 up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2739 if (!same_inode) 2579 if (!same_inode)
2740 up_write(&OCFS2_I(inode_out)- 2580 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2741 if (remapped < 0) { 2581 if (remapped < 0) {
2742 ret = remapped; 2582 ret = remapped;
2743 mlog_errno(ret); 2583 mlog_errno(ret);
2744 goto out_unlock; 2584 goto out_unlock;
2745 } 2585 }
2746 2586
2747 /* 2587 /*
2748 * Empty the extent map so that we ma 2588 * Empty the extent map so that we may get the right extent
2749 * record from the disk. 2589 * record from the disk.
2750 */ 2590 */
2751 ocfs2_extent_map_trunc(inode_in, 0); 2591 ocfs2_extent_map_trunc(inode_in, 0);
2752 ocfs2_extent_map_trunc(inode_out, 0); 2592 ocfs2_extent_map_trunc(inode_out, 0);
2753 2593
2754 ret = ocfs2_reflink_update_dest(inode 2594 ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2755 if (ret) { 2595 if (ret) {
2756 mlog_errno(ret); 2596 mlog_errno(ret);
2757 goto out_unlock; 2597 goto out_unlock;
2758 } 2598 }
2759 2599
2760 out_unlock: 2600 out_unlock:
2761 ocfs2_reflink_inodes_unlock(inode_in, 2601 ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2762 return remapped > 0 ? remapped : ret; 2602 return remapped > 0 ? remapped : ret;
2763 } 2603 }
2764 2604
2765 static loff_t ocfs2_dir_llseek(struct file *f <<
2766 { <<
2767 struct ocfs2_file_private *fp = file- <<
2768 <<
2769 return generic_llseek_cookie(file, of <<
2770 } <<
2771 <<
2772 const struct inode_operations ocfs2_file_iops 2605 const struct inode_operations ocfs2_file_iops = {
2773 .setattr = ocfs2_setattr, 2606 .setattr = ocfs2_setattr,
2774 .getattr = ocfs2_getattr, 2607 .getattr = ocfs2_getattr,
2775 .permission = ocfs2_permission, 2608 .permission = ocfs2_permission,
2776 .listxattr = ocfs2_listxattr, 2609 .listxattr = ocfs2_listxattr,
2777 .fiemap = ocfs2_fiemap, 2610 .fiemap = ocfs2_fiemap,
2778 .get_inode_acl = ocfs2_iop_get_acl, !! 2611 .get_acl = ocfs2_iop_get_acl,
2779 .set_acl = ocfs2_iop_set_acl, 2612 .set_acl = ocfs2_iop_set_acl,
2780 .fileattr_get = ocfs2_fileattr_get, <<
2781 .fileattr_set = ocfs2_fileattr_set, <<
2782 }; 2613 };
2783 2614
2784 const struct inode_operations ocfs2_special_f 2615 const struct inode_operations ocfs2_special_file_iops = {
2785 .setattr = ocfs2_setattr, 2616 .setattr = ocfs2_setattr,
2786 .getattr = ocfs2_getattr, 2617 .getattr = ocfs2_getattr,
2787 .listxattr = ocfs2_listxattr, <<
2788 .permission = ocfs2_permission, 2618 .permission = ocfs2_permission,
2789 .get_inode_acl = ocfs2_iop_get_acl, !! 2619 .get_acl = ocfs2_iop_get_acl,
2790 .set_acl = ocfs2_iop_set_acl, 2620 .set_acl = ocfs2_iop_set_acl,
2791 }; 2621 };
2792 2622
2793 /* 2623 /*
2794 * Other than ->lock, keep ocfs2_fops and ocf 2624 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2795 * ocfs2_fops_no_plocks and ocfs2_dops_no_plo 2625 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2796 */ 2626 */
2797 const struct file_operations ocfs2_fops = { 2627 const struct file_operations ocfs2_fops = {
2798 .llseek = ocfs2_file_llseek, 2628 .llseek = ocfs2_file_llseek,
2799 .mmap = ocfs2_mmap, 2629 .mmap = ocfs2_mmap,
2800 .fsync = ocfs2_sync_file, 2630 .fsync = ocfs2_sync_file,
2801 .release = ocfs2_file_release, 2631 .release = ocfs2_file_release,
2802 .open = ocfs2_file_open, 2632 .open = ocfs2_file_open,
2803 .read_iter = ocfs2_file_read_ite 2633 .read_iter = ocfs2_file_read_iter,
2804 .write_iter = ocfs2_file_write_it 2634 .write_iter = ocfs2_file_write_iter,
2805 .unlocked_ioctl = ocfs2_ioctl, 2635 .unlocked_ioctl = ocfs2_ioctl,
2806 #ifdef CONFIG_COMPAT 2636 #ifdef CONFIG_COMPAT
2807 .compat_ioctl = ocfs2_compat_ioctl, 2637 .compat_ioctl = ocfs2_compat_ioctl,
2808 #endif 2638 #endif
2809 .lock = ocfs2_lock, 2639 .lock = ocfs2_lock,
2810 .flock = ocfs2_flock, 2640 .flock = ocfs2_flock,
2811 .splice_read = ocfs2_file_splice_r !! 2641 .splice_read = generic_file_splice_read,
2812 .splice_write = iter_file_splice_wr 2642 .splice_write = iter_file_splice_write,
2813 .fallocate = ocfs2_fallocate, 2643 .fallocate = ocfs2_fallocate,
2814 .remap_file_range = ocfs2_remap_file_ 2644 .remap_file_range = ocfs2_remap_file_range,
2815 }; 2645 };
2816 2646
2817 WRAP_DIR_ITER(ocfs2_readdir) // FIXME! <<
2818 const struct file_operations ocfs2_dops = { 2647 const struct file_operations ocfs2_dops = {
2819 .llseek = ocfs2_dir_llseek, !! 2648 .llseek = generic_file_llseek,
2820 .read = generic_read_dir, 2649 .read = generic_read_dir,
2821 .iterate_shared = shared_ocfs2_readdi !! 2650 .iterate = ocfs2_readdir,
2822 .fsync = ocfs2_sync_file, 2651 .fsync = ocfs2_sync_file,
2823 .release = ocfs2_dir_release, 2652 .release = ocfs2_dir_release,
2824 .open = ocfs2_dir_open, 2653 .open = ocfs2_dir_open,
2825 .unlocked_ioctl = ocfs2_ioctl, 2654 .unlocked_ioctl = ocfs2_ioctl,
2826 #ifdef CONFIG_COMPAT 2655 #ifdef CONFIG_COMPAT
2827 .compat_ioctl = ocfs2_compat_ioctl, 2656 .compat_ioctl = ocfs2_compat_ioctl,
2828 #endif 2657 #endif
2829 .lock = ocfs2_lock, 2658 .lock = ocfs2_lock,
2830 .flock = ocfs2_flock, 2659 .flock = ocfs2_flock,
2831 }; 2660 };
2832 2661
2833 /* 2662 /*
2834 * POSIX-lockless variants of our file_operat 2663 * POSIX-lockless variants of our file_operations.
2835 * 2664 *
2836 * These will be used if the underlying clust 2665 * These will be used if the underlying cluster stack does not support
2837 * posix file locking, if the user passes the 2666 * posix file locking, if the user passes the "localflocks" mount
2838 * option, or if we have a local-only fs. 2667 * option, or if we have a local-only fs.
2839 * 2668 *
2840 * ocfs2_flock is in here because all stacks 2669 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2841 * so we still want it in the case of no stac 2670 * so we still want it in the case of no stack support for
2842 * plocks. Internally, it will do the right t 2671 * plocks. Internally, it will do the right thing when asked to ignore
2843 * the cluster. 2672 * the cluster.
2844 */ 2673 */
2845 const struct file_operations ocfs2_fops_no_pl 2674 const struct file_operations ocfs2_fops_no_plocks = {
2846 .llseek = ocfs2_file_llseek, 2675 .llseek = ocfs2_file_llseek,
2847 .mmap = ocfs2_mmap, 2676 .mmap = ocfs2_mmap,
2848 .fsync = ocfs2_sync_file, 2677 .fsync = ocfs2_sync_file,
2849 .release = ocfs2_file_release, 2678 .release = ocfs2_file_release,
2850 .open = ocfs2_file_open, 2679 .open = ocfs2_file_open,
2851 .read_iter = ocfs2_file_read_ite 2680 .read_iter = ocfs2_file_read_iter,
2852 .write_iter = ocfs2_file_write_it 2681 .write_iter = ocfs2_file_write_iter,
2853 .unlocked_ioctl = ocfs2_ioctl, 2682 .unlocked_ioctl = ocfs2_ioctl,
2854 #ifdef CONFIG_COMPAT 2683 #ifdef CONFIG_COMPAT
2855 .compat_ioctl = ocfs2_compat_ioctl, 2684 .compat_ioctl = ocfs2_compat_ioctl,
2856 #endif 2685 #endif
2857 .flock = ocfs2_flock, 2686 .flock = ocfs2_flock,
2858 .splice_read = filemap_splice_read !! 2687 .splice_read = generic_file_splice_read,
2859 .splice_write = iter_file_splice_wr 2688 .splice_write = iter_file_splice_write,
2860 .fallocate = ocfs2_fallocate, 2689 .fallocate = ocfs2_fallocate,
2861 .remap_file_range = ocfs2_remap_file_ 2690 .remap_file_range = ocfs2_remap_file_range,
2862 }; 2691 };
2863 2692
2864 const struct file_operations ocfs2_dops_no_pl 2693 const struct file_operations ocfs2_dops_no_plocks = {
2865 .llseek = ocfs2_dir_llseek, !! 2694 .llseek = generic_file_llseek,
2866 .read = generic_read_dir, 2695 .read = generic_read_dir,
2867 .iterate_shared = shared_ocfs2_readdi !! 2696 .iterate = ocfs2_readdir,
2868 .fsync = ocfs2_sync_file, 2697 .fsync = ocfs2_sync_file,
2869 .release = ocfs2_dir_release, 2698 .release = ocfs2_dir_release,
2870 .open = ocfs2_dir_open, 2699 .open = ocfs2_dir_open,
2871 .unlocked_ioctl = ocfs2_ioctl, 2700 .unlocked_ioctl = ocfs2_ioctl,
2872 #ifdef CONFIG_COMPAT 2701 #ifdef CONFIG_COMPAT
2873 .compat_ioctl = ocfs2_compat_ioctl, 2702 .compat_ioctl = ocfs2_compat_ioctl,
2874 #endif 2703 #endif
2875 .flock = ocfs2_flock, 2704 .flock = ocfs2_flock,
2876 }; 2705 };
2877 2706
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.