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